1 /* SPDX-License-Identifier: BSD-3-Clause 2 * Copyright (C) 2018 Intel Corporation. All rights reserved. 3 * Copyright (c) 2019, 2020 Mellanox Technologies LTD. All rights reserved. 4 * Copyright (c) 2022-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved. 5 */ 6 7 #include "spdk/accel.h" 8 #include "spdk/stdinc.h" 9 #include "spdk/crc32.h" 10 #include "spdk/endian.h" 11 #include "spdk/assert.h" 12 #include "spdk/thread.h" 13 #include "spdk/nvmf_transport.h" 14 #include "spdk/string.h" 15 #include "spdk/trace.h" 16 #include "spdk/util.h" 17 #include "spdk/log.h" 18 #include "spdk/keyring.h" 19 20 #include "spdk_internal/assert.h" 21 #include "spdk_internal/nvme_tcp.h" 22 #include "spdk_internal/sock.h" 23 24 #include "nvmf_internal.h" 25 #include "transport.h" 26 27 #include "spdk_internal/trace_defs.h" 28 29 #define NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME 16 30 #define SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY 16 31 #define SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY 0 32 #define SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM 32 33 #define SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION true 34 35 #define SPDK_NVMF_TCP_MIN_IO_QUEUE_DEPTH 2 36 #define SPDK_NVMF_TCP_MAX_IO_QUEUE_DEPTH 65535 37 #define SPDK_NVMF_TCP_MIN_ADMIN_QUEUE_DEPTH 2 38 #define SPDK_NVMF_TCP_MAX_ADMIN_QUEUE_DEPTH 4096 39 40 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH 128 41 #define SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH 128 42 #define SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR 128 43 #define SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE 4096 44 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE 131072 45 #define SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE 131072 46 #define SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS 511 47 #define SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE UINT32_MAX 48 #define SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP false 49 #define SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC 1 50 51 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp; 52 static bool g_tls_log = false; 53 54 /* spdk nvmf related structure */ 55 enum spdk_nvmf_tcp_req_state { 56 57 /* The request is not currently in use */ 58 TCP_REQUEST_STATE_FREE = 0, 59 60 /* Initial state when request first received */ 61 TCP_REQUEST_STATE_NEW = 1, 62 63 /* The request is queued until a data buffer is available. */ 64 TCP_REQUEST_STATE_NEED_BUFFER = 2, 65 66 /* The request has the data buffer available */ 67 TCP_REQUEST_STATE_HAVE_BUFFER = 3, 68 69 /* The request is waiting for zcopy_start to finish */ 70 TCP_REQUEST_STATE_AWAITING_ZCOPY_START = 4, 71 72 /* The request has received a zero-copy buffer */ 73 TCP_REQUEST_STATE_ZCOPY_START_COMPLETED = 5, 74 75 /* The request is currently transferring data from the host to the controller. */ 76 TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER = 6, 77 78 /* The request is waiting for the R2T send acknowledgement. */ 79 TCP_REQUEST_STATE_AWAITING_R2T_ACK = 7, 80 81 /* The request is ready to execute at the block device */ 82 TCP_REQUEST_STATE_READY_TO_EXECUTE = 8, 83 84 /* The request is currently executing at the block device */ 85 TCP_REQUEST_STATE_EXECUTING = 9, 86 87 /* The request is waiting for zcopy buffers to be committed */ 88 TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT = 10, 89 90 /* The request finished executing at the block device */ 91 TCP_REQUEST_STATE_EXECUTED = 11, 92 93 /* The request is ready to send a completion */ 94 TCP_REQUEST_STATE_READY_TO_COMPLETE = 12, 95 96 /* The request is currently transferring final pdus from the controller to the host. */ 97 TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST = 13, 98 99 /* The request is waiting for zcopy buffers to be released (without committing) */ 100 TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE = 14, 101 102 /* The request completed and can be marked free. */ 103 TCP_REQUEST_STATE_COMPLETED = 15, 104 105 /* Terminator */ 106 TCP_REQUEST_NUM_STATES, 107 }; 108 109 enum nvmf_tcp_qpair_state { 110 NVMF_TCP_QPAIR_STATE_INVALID = 0, 111 NVMF_TCP_QPAIR_STATE_INITIALIZING = 1, 112 NVMF_TCP_QPAIR_STATE_RUNNING = 2, 113 NVMF_TCP_QPAIR_STATE_EXITING = 3, 114 NVMF_TCP_QPAIR_STATE_EXITED = 4, 115 }; 116 117 static const char *spdk_nvmf_tcp_term_req_fes_str[] = { 118 "Invalid PDU Header Field", 119 "PDU Sequence Error", 120 "Header Digiest Error", 121 "Data Transfer Out of Range", 122 "R2T Limit Exceeded", 123 "Unsupported parameter", 124 }; 125 126 static void 127 nvmf_tcp_trace(void) 128 { 129 spdk_trace_register_owner_type(OWNER_TYPE_NVMF_TCP, 't'); 130 spdk_trace_register_object(OBJECT_NVMF_TCP_IO, 'r'); 131 spdk_trace_register_description("TCP_REQ_NEW", 132 TRACE_TCP_REQUEST_STATE_NEW, 133 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 1, 134 SPDK_TRACE_ARG_TYPE_INT, "qd"); 135 spdk_trace_register_description("TCP_REQ_NEED_BUFFER", 136 TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 137 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 138 SPDK_TRACE_ARG_TYPE_INT, ""); 139 spdk_trace_register_description("TCP_REQ_HAVE_BUFFER", 140 TRACE_TCP_REQUEST_STATE_HAVE_BUFFER, 141 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 142 SPDK_TRACE_ARG_TYPE_INT, ""); 143 spdk_trace_register_description("TCP_REQ_WAIT_ZCPY_START", 144 TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_START, 145 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 146 SPDK_TRACE_ARG_TYPE_INT, ""); 147 spdk_trace_register_description("TCP_REQ_ZCPY_START_CPL", 148 TRACE_TCP_REQUEST_STATE_ZCOPY_START_COMPLETED, 149 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 150 SPDK_TRACE_ARG_TYPE_INT, ""); 151 spdk_trace_register_description("TCP_REQ_TX_H_TO_C", 152 TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 153 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 154 SPDK_TRACE_ARG_TYPE_INT, ""); 155 spdk_trace_register_description("TCP_REQ_RDY_TO_EXECUTE", 156 TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 157 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 158 SPDK_TRACE_ARG_TYPE_INT, ""); 159 spdk_trace_register_description("TCP_REQ_EXECUTING", 160 TRACE_TCP_REQUEST_STATE_EXECUTING, 161 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 162 SPDK_TRACE_ARG_TYPE_INT, ""); 163 spdk_trace_register_description("TCP_REQ_WAIT_ZCPY_CMT", 164 TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_COMMIT, 165 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 166 SPDK_TRACE_ARG_TYPE_INT, ""); 167 spdk_trace_register_description("TCP_REQ_EXECUTED", 168 TRACE_TCP_REQUEST_STATE_EXECUTED, 169 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 170 SPDK_TRACE_ARG_TYPE_INT, ""); 171 spdk_trace_register_description("TCP_REQ_RDY_TO_COMPLETE", 172 TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 173 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 174 SPDK_TRACE_ARG_TYPE_INT, ""); 175 spdk_trace_register_description("TCP_REQ_TRANSFER_C2H", 176 TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 177 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 178 SPDK_TRACE_ARG_TYPE_INT, ""); 179 spdk_trace_register_description("TCP_REQ_AWAIT_ZCPY_RLS", 180 TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_RELEASE, 181 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 182 SPDK_TRACE_ARG_TYPE_INT, ""); 183 spdk_trace_register_description("TCP_REQ_COMPLETED", 184 TRACE_TCP_REQUEST_STATE_COMPLETED, 185 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 186 SPDK_TRACE_ARG_TYPE_INT, "qd"); 187 spdk_trace_register_description("TCP_READ_DONE", 188 TRACE_TCP_READ_FROM_SOCKET_DONE, 189 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 190 SPDK_TRACE_ARG_TYPE_INT, ""); 191 spdk_trace_register_description("TCP_REQ_AWAIT_R2T_ACK", 192 TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 193 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 194 SPDK_TRACE_ARG_TYPE_INT, ""); 195 196 spdk_trace_register_description("TCP_QP_CREATE", TRACE_TCP_QP_CREATE, 197 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 198 SPDK_TRACE_ARG_TYPE_INT, ""); 199 spdk_trace_register_description("TCP_QP_SOCK_INIT", TRACE_TCP_QP_SOCK_INIT, 200 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 201 SPDK_TRACE_ARG_TYPE_INT, ""); 202 spdk_trace_register_description("TCP_QP_STATE_CHANGE", TRACE_TCP_QP_STATE_CHANGE, 203 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 204 SPDK_TRACE_ARG_TYPE_INT, "state"); 205 spdk_trace_register_description("TCP_QP_DISCONNECT", TRACE_TCP_QP_DISCONNECT, 206 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 207 SPDK_TRACE_ARG_TYPE_INT, ""); 208 spdk_trace_register_description("TCP_QP_DESTROY", TRACE_TCP_QP_DESTROY, 209 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 210 SPDK_TRACE_ARG_TYPE_INT, ""); 211 spdk_trace_register_description("TCP_QP_ABORT_REQ", TRACE_TCP_QP_ABORT_REQ, 212 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 213 SPDK_TRACE_ARG_TYPE_INT, ""); 214 spdk_trace_register_description("TCP_QP_RCV_STATE_CHANGE", TRACE_TCP_QP_RCV_STATE_CHANGE, 215 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 216 SPDK_TRACE_ARG_TYPE_INT, "state"); 217 218 spdk_trace_tpoint_register_relation(TRACE_BDEV_IO_START, OBJECT_NVMF_TCP_IO, 1); 219 spdk_trace_tpoint_register_relation(TRACE_BDEV_IO_DONE, OBJECT_NVMF_TCP_IO, 0); 220 spdk_trace_tpoint_register_relation(TRACE_SOCK_REQ_QUEUE, OBJECT_NVMF_TCP_IO, 0); 221 spdk_trace_tpoint_register_relation(TRACE_SOCK_REQ_PEND, OBJECT_NVMF_TCP_IO, 0); 222 spdk_trace_tpoint_register_relation(TRACE_SOCK_REQ_COMPLETE, OBJECT_NVMF_TCP_IO, 0); 223 } 224 SPDK_TRACE_REGISTER_FN(nvmf_tcp_trace, "nvmf_tcp", TRACE_GROUP_NVMF_TCP) 225 226 struct spdk_nvmf_tcp_req { 227 struct spdk_nvmf_request req; 228 struct spdk_nvme_cpl rsp; 229 struct spdk_nvme_cmd cmd; 230 231 /* A PDU that can be used for sending responses. This is 232 * not the incoming PDU! */ 233 struct nvme_tcp_pdu *pdu; 234 235 /* In-capsule data buffer */ 236 uint8_t *buf; 237 238 struct spdk_nvmf_tcp_req *fused_pair; 239 240 /* 241 * The PDU for a request may be used multiple times in serial over 242 * the request's lifetime. For example, first to send an R2T, then 243 * to send a completion. To catch mistakes where the PDU is used 244 * twice at the same time, add a debug flag here for init/fini. 245 */ 246 bool pdu_in_use; 247 bool has_in_capsule_data; 248 bool fused_failed; 249 250 /* transfer_tag */ 251 uint16_t ttag; 252 253 enum spdk_nvmf_tcp_req_state state; 254 255 /* 256 * h2c_offset is used when we receive the h2c_data PDU. 257 */ 258 uint32_t h2c_offset; 259 260 STAILQ_ENTRY(spdk_nvmf_tcp_req) link; 261 TAILQ_ENTRY(spdk_nvmf_tcp_req) state_link; 262 STAILQ_ENTRY(spdk_nvmf_tcp_req) control_msg_link; 263 }; 264 265 struct spdk_nvmf_tcp_qpair { 266 struct spdk_nvmf_qpair qpair; 267 struct spdk_nvmf_tcp_poll_group *group; 268 struct spdk_sock *sock; 269 270 enum nvme_tcp_pdu_recv_state recv_state; 271 enum nvmf_tcp_qpair_state state; 272 273 /* PDU being actively received */ 274 struct nvme_tcp_pdu *pdu_in_progress; 275 276 struct spdk_nvmf_tcp_req *fused_first; 277 278 /* Queues to track the requests in all states */ 279 TAILQ_HEAD(, spdk_nvmf_tcp_req) tcp_req_working_queue; 280 TAILQ_HEAD(, spdk_nvmf_tcp_req) tcp_req_free_queue; 281 SLIST_HEAD(, nvme_tcp_pdu) tcp_pdu_free_queue; 282 /* Number of working pdus */ 283 uint32_t tcp_pdu_working_count; 284 285 /* Number of requests in each state */ 286 uint32_t state_cntr[TCP_REQUEST_NUM_STATES]; 287 288 uint8_t cpda; 289 290 bool host_hdgst_enable; 291 bool host_ddgst_enable; 292 293 bool await_req_msg_pending; 294 295 /* This is a spare PDU used for sending special management 296 * operations. Primarily, this is used for the initial 297 * connection response and c2h termination request. */ 298 struct nvme_tcp_pdu *mgmt_pdu; 299 300 /* Arrays of in-capsule buffers, requests, and pdus. 301 * Each array is 'resource_count' number of elements */ 302 void *bufs; 303 struct spdk_nvmf_tcp_req *reqs; 304 struct nvme_tcp_pdu *pdus; 305 uint32_t resource_count; 306 uint32_t recv_buf_size; 307 308 struct spdk_nvmf_tcp_port *port; 309 310 /* IP address */ 311 char initiator_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 312 char target_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 313 314 /* IP port */ 315 uint16_t initiator_port; 316 uint16_t target_port; 317 318 /* Wait until the host terminates the connection (e.g. after sending C2HTermReq) */ 319 bool wait_terminate; 320 321 /* Timer used to destroy qpair after detecting transport error issue if initiator does 322 * not close the connection. 323 */ 324 struct spdk_poller *timeout_poller; 325 326 spdk_nvmf_transport_qpair_fini_cb fini_cb_fn; 327 void *fini_cb_arg; 328 329 TAILQ_ENTRY(spdk_nvmf_tcp_qpair) link; 330 bool pending_flush; 331 }; 332 333 struct spdk_nvmf_tcp_control_msg { 334 STAILQ_ENTRY(spdk_nvmf_tcp_control_msg) link; 335 }; 336 337 struct spdk_nvmf_tcp_control_msg_list { 338 void *msg_buf; 339 STAILQ_HEAD(, spdk_nvmf_tcp_control_msg) free_msgs; 340 STAILQ_HEAD(, spdk_nvmf_tcp_req) waiting_for_msg_reqs; 341 }; 342 343 struct spdk_nvmf_tcp_poll_group { 344 struct spdk_nvmf_transport_poll_group group; 345 struct spdk_sock_group *sock_group; 346 347 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) qpairs; 348 349 struct spdk_io_channel *accel_channel; 350 struct spdk_nvmf_tcp_control_msg_list *control_msg_list; 351 352 TAILQ_ENTRY(spdk_nvmf_tcp_poll_group) link; 353 }; 354 355 struct spdk_nvmf_tcp_port { 356 const struct spdk_nvme_transport_id *trid; 357 struct spdk_sock *listen_sock; 358 struct spdk_nvmf_transport *transport; 359 TAILQ_ENTRY(spdk_nvmf_tcp_port) link; 360 }; 361 362 struct tcp_transport_opts { 363 bool c2h_success; 364 uint16_t control_msg_num; 365 uint32_t sock_priority; 366 }; 367 368 struct tcp_psk_entry { 369 char hostnqn[SPDK_NVMF_NQN_MAX_LEN + 1]; 370 char subnqn[SPDK_NVMF_NQN_MAX_LEN + 1]; 371 char pskid[NVMF_PSK_IDENTITY_LEN]; 372 uint8_t psk[SPDK_TLS_PSK_MAX_LEN]; 373 struct spdk_key *key; 374 uint32_t psk_size; 375 enum nvme_tcp_cipher_suite tls_cipher_suite; 376 TAILQ_ENTRY(tcp_psk_entry) link; 377 }; 378 379 struct spdk_nvmf_tcp_transport { 380 struct spdk_nvmf_transport transport; 381 struct tcp_transport_opts tcp_opts; 382 uint32_t ack_timeout; 383 384 struct spdk_nvmf_tcp_poll_group *next_pg; 385 386 struct spdk_poller *accept_poller; 387 struct spdk_sock_group *listen_sock_group; 388 389 TAILQ_HEAD(, spdk_nvmf_tcp_port) ports; 390 TAILQ_HEAD(, spdk_nvmf_tcp_poll_group) poll_groups; 391 392 TAILQ_HEAD(, tcp_psk_entry) psks; 393 }; 394 395 static const struct spdk_json_object_decoder tcp_transport_opts_decoder[] = { 396 { 397 "c2h_success", offsetof(struct tcp_transport_opts, c2h_success), 398 spdk_json_decode_bool, true 399 }, 400 { 401 "control_msg_num", offsetof(struct tcp_transport_opts, control_msg_num), 402 spdk_json_decode_uint16, true 403 }, 404 { 405 "sock_priority", offsetof(struct tcp_transport_opts, sock_priority), 406 spdk_json_decode_uint32, true 407 }, 408 }; 409 410 static bool nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 411 struct spdk_nvmf_tcp_req *tcp_req); 412 static void nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group); 413 414 static void _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 415 struct spdk_nvmf_tcp_req *tcp_req); 416 static void nvmf_tcp_qpair_process(struct spdk_nvmf_tcp_qpair *tqpair); 417 418 static inline void 419 nvmf_tcp_req_set_state(struct spdk_nvmf_tcp_req *tcp_req, 420 enum spdk_nvmf_tcp_req_state state) 421 { 422 struct spdk_nvmf_qpair *qpair; 423 struct spdk_nvmf_tcp_qpair *tqpair; 424 425 qpair = tcp_req->req.qpair; 426 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 427 428 assert(tqpair->state_cntr[tcp_req->state] > 0); 429 tqpair->state_cntr[tcp_req->state]--; 430 tqpair->state_cntr[state]++; 431 432 tcp_req->state = state; 433 } 434 435 static inline struct nvme_tcp_pdu * 436 nvmf_tcp_req_pdu_init(struct spdk_nvmf_tcp_req *tcp_req) 437 { 438 assert(tcp_req->pdu_in_use == false); 439 440 memset(tcp_req->pdu, 0, sizeof(*tcp_req->pdu)); 441 tcp_req->pdu->qpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 442 443 return tcp_req->pdu; 444 } 445 446 static struct spdk_nvmf_tcp_req * 447 nvmf_tcp_req_get(struct spdk_nvmf_tcp_qpair *tqpair) 448 { 449 struct spdk_nvmf_tcp_req *tcp_req; 450 451 tcp_req = TAILQ_FIRST(&tqpair->tcp_req_free_queue); 452 if (spdk_unlikely(!tcp_req)) { 453 return NULL; 454 } 455 456 memset(&tcp_req->rsp, 0, sizeof(tcp_req->rsp)); 457 tcp_req->h2c_offset = 0; 458 tcp_req->has_in_capsule_data = false; 459 tcp_req->req.dif_enabled = false; 460 tcp_req->req.zcopy_phase = NVMF_ZCOPY_PHASE_NONE; 461 462 TAILQ_REMOVE(&tqpair->tcp_req_free_queue, tcp_req, state_link); 463 TAILQ_INSERT_TAIL(&tqpair->tcp_req_working_queue, tcp_req, state_link); 464 tqpair->qpair.queue_depth++; 465 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW); 466 return tcp_req; 467 } 468 469 static void 470 handle_await_req(void *arg) 471 { 472 struct spdk_nvmf_tcp_qpair *tqpair = arg; 473 474 tqpair->await_req_msg_pending = false; 475 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 476 nvmf_tcp_qpair_process(tqpair); 477 } 478 } 479 480 static inline void 481 nvmf_tcp_req_put(struct spdk_nvmf_tcp_qpair *tqpair, struct spdk_nvmf_tcp_req *tcp_req) 482 { 483 assert(!tcp_req->pdu_in_use); 484 485 TAILQ_REMOVE(&tqpair->tcp_req_working_queue, tcp_req, state_link); 486 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 487 tqpair->qpair.queue_depth--; 488 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_FREE); 489 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ && 490 !tqpair->await_req_msg_pending) { 491 tqpair->await_req_msg_pending = true; 492 spdk_thread_send_msg(spdk_get_thread(), handle_await_req, tqpair); 493 } 494 } 495 496 static void 497 nvmf_tcp_req_get_buffers_done(struct spdk_nvmf_request *req) 498 { 499 struct spdk_nvmf_tcp_req *tcp_req; 500 struct spdk_nvmf_transport *transport; 501 struct spdk_nvmf_tcp_transport *ttransport; 502 503 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 504 transport = req->qpair->transport; 505 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 506 507 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_HAVE_BUFFER); 508 nvmf_tcp_req_process(ttransport, tcp_req); 509 } 510 511 static void 512 nvmf_tcp_request_free(void *cb_arg) 513 { 514 struct spdk_nvmf_tcp_transport *ttransport; 515 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 516 517 assert(tcp_req != NULL); 518 519 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req=%p will be freed\n", tcp_req); 520 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 521 struct spdk_nvmf_tcp_transport, transport); 522 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 523 nvmf_tcp_req_process(ttransport, tcp_req); 524 } 525 526 static int 527 nvmf_tcp_req_free(struct spdk_nvmf_request *req) 528 { 529 struct spdk_nvmf_tcp_req *tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 530 531 nvmf_tcp_request_free(tcp_req); 532 533 return 0; 534 } 535 536 static void 537 nvmf_tcp_drain_state_queue(struct spdk_nvmf_tcp_qpair *tqpair, 538 enum spdk_nvmf_tcp_req_state state) 539 { 540 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 541 542 assert(state != TCP_REQUEST_STATE_FREE); 543 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 544 if (state == tcp_req->state) { 545 nvmf_tcp_request_free(tcp_req); 546 } 547 } 548 } 549 550 static inline void 551 nvmf_tcp_request_get_buffers_abort(struct spdk_nvmf_tcp_req *tcp_req) 552 { 553 /* Request can wait either for the iobuf or control_msg */ 554 struct spdk_nvmf_poll_group *group = tcp_req->req.qpair->group; 555 struct spdk_nvmf_transport *transport = tcp_req->req.qpair->transport; 556 struct spdk_nvmf_transport_poll_group *tgroup = nvmf_get_transport_poll_group(group, transport); 557 struct spdk_nvmf_tcp_poll_group *tcp_group = SPDK_CONTAINEROF(tgroup, 558 struct spdk_nvmf_tcp_poll_group, group); 559 struct spdk_nvmf_tcp_req *tmp_req, *abort_req; 560 561 assert(tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER); 562 563 STAILQ_FOREACH_SAFE(abort_req, &tcp_group->control_msg_list->waiting_for_msg_reqs, control_msg_link, 564 tmp_req) { 565 if (abort_req == tcp_req) { 566 STAILQ_REMOVE(&tcp_group->control_msg_list->waiting_for_msg_reqs, abort_req, spdk_nvmf_tcp_req, 567 control_msg_link); 568 return; 569 } 570 } 571 572 if (!nvmf_request_get_buffers_abort(&tcp_req->req)) { 573 SPDK_ERRLOG("Failed to abort tcp_req=%p\n", tcp_req); 574 assert(0 && "Should never happen"); 575 } 576 } 577 578 static void 579 nvmf_tcp_cleanup_all_states(struct spdk_nvmf_tcp_qpair *tqpair) 580 { 581 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 582 583 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 584 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEW); 585 586 /* Wipe the requests waiting for buffer from the waiting list */ 587 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 588 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 589 nvmf_tcp_request_get_buffers_abort(tcp_req); 590 } 591 } 592 593 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEED_BUFFER); 594 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_EXECUTING); 595 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 596 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 597 } 598 599 static void 600 nvmf_tcp_dump_qpair_req_contents(struct spdk_nvmf_tcp_qpair *tqpair) 601 { 602 int i; 603 struct spdk_nvmf_tcp_req *tcp_req; 604 605 SPDK_ERRLOG("Dumping contents of queue pair (QID %d)\n", tqpair->qpair.qid); 606 for (i = 1; i < TCP_REQUEST_NUM_STATES; i++) { 607 SPDK_ERRLOG("\tNum of requests in state[%d] = %u\n", i, tqpair->state_cntr[i]); 608 TAILQ_FOREACH(tcp_req, &tqpair->tcp_req_working_queue, state_link) { 609 if ((int)tcp_req->state == i) { 610 SPDK_ERRLOG("\t\tRequest Data From Pool: %d\n", tcp_req->req.data_from_pool); 611 SPDK_ERRLOG("\t\tRequest opcode: %d\n", tcp_req->req.cmd->nvmf_cmd.opcode); 612 } 613 } 614 } 615 } 616 617 static void 618 _nvmf_tcp_qpair_destroy(void *_tqpair) 619 { 620 struct spdk_nvmf_tcp_qpair *tqpair = _tqpair; 621 spdk_nvmf_transport_qpair_fini_cb cb_fn = tqpair->fini_cb_fn; 622 void *cb_arg = tqpair->fini_cb_arg; 623 int err = 0; 624 625 spdk_trace_record(TRACE_TCP_QP_DESTROY, tqpair->qpair.trace_id, 0, 0); 626 627 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 628 629 err = spdk_sock_close(&tqpair->sock); 630 assert(err == 0); 631 nvmf_tcp_cleanup_all_states(tqpair); 632 633 if (tqpair->state_cntr[TCP_REQUEST_STATE_FREE] != tqpair->resource_count) { 634 SPDK_ERRLOG("tqpair(%p) free tcp request num is %u but should be %u\n", tqpair, 635 tqpair->state_cntr[TCP_REQUEST_STATE_FREE], 636 tqpair->resource_count); 637 err++; 638 } 639 640 if (err > 0) { 641 nvmf_tcp_dump_qpair_req_contents(tqpair); 642 } 643 644 /* The timeout poller might still be registered here if we close the qpair before host 645 * terminates the connection. 646 */ 647 spdk_poller_unregister(&tqpair->timeout_poller); 648 spdk_dma_free(tqpair->pdus); 649 free(tqpair->reqs); 650 spdk_free(tqpair->bufs); 651 spdk_trace_unregister_owner(tqpair->qpair.trace_id); 652 free(tqpair); 653 654 if (cb_fn != NULL) { 655 cb_fn(cb_arg); 656 } 657 658 SPDK_DEBUGLOG(nvmf_tcp, "Leave\n"); 659 } 660 661 static void 662 nvmf_tcp_qpair_destroy(struct spdk_nvmf_tcp_qpair *tqpair) 663 { 664 /* Delay the destruction to make sure it isn't performed from the context of a sock 665 * callback. Otherwise, spdk_sock_close() might not abort pending requests, causing their 666 * completions to be executed after the qpair is freed. (Note: this fixed issue #2471.) 667 */ 668 spdk_thread_send_msg(spdk_get_thread(), _nvmf_tcp_qpair_destroy, tqpair); 669 } 670 671 static void 672 nvmf_tcp_dump_opts(struct spdk_nvmf_transport *transport, struct spdk_json_write_ctx *w) 673 { 674 struct spdk_nvmf_tcp_transport *ttransport; 675 assert(w != NULL); 676 677 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 678 spdk_json_write_named_bool(w, "c2h_success", ttransport->tcp_opts.c2h_success); 679 spdk_json_write_named_uint32(w, "sock_priority", ttransport->tcp_opts.sock_priority); 680 } 681 682 static void 683 nvmf_tcp_free_psk_entry(struct tcp_psk_entry *entry) 684 { 685 if (entry == NULL) { 686 return; 687 } 688 689 spdk_memset_s(entry->psk, sizeof(entry->psk), 0, sizeof(entry->psk)); 690 spdk_keyring_put_key(entry->key); 691 free(entry); 692 } 693 694 static int 695 nvmf_tcp_destroy(struct spdk_nvmf_transport *transport, 696 spdk_nvmf_transport_destroy_done_cb cb_fn, void *cb_arg) 697 { 698 struct spdk_nvmf_tcp_transport *ttransport; 699 struct tcp_psk_entry *entry, *tmp; 700 701 assert(transport != NULL); 702 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 703 704 TAILQ_FOREACH_SAFE(entry, &ttransport->psks, link, tmp) { 705 TAILQ_REMOVE(&ttransport->psks, entry, link); 706 nvmf_tcp_free_psk_entry(entry); 707 } 708 709 spdk_poller_unregister(&ttransport->accept_poller); 710 spdk_sock_group_unregister_interrupt(ttransport->listen_sock_group); 711 spdk_sock_group_close(&ttransport->listen_sock_group); 712 free(ttransport); 713 714 if (cb_fn) { 715 cb_fn(cb_arg); 716 } 717 return 0; 718 } 719 720 static int nvmf_tcp_accept(void *ctx); 721 722 static void nvmf_tcp_accept_cb(void *ctx, struct spdk_sock_group *group, struct spdk_sock *sock); 723 724 static struct spdk_nvmf_transport * 725 nvmf_tcp_create(struct spdk_nvmf_transport_opts *opts) 726 { 727 struct spdk_nvmf_tcp_transport *ttransport; 728 uint32_t sge_count; 729 uint32_t min_shared_buffers; 730 int rc; 731 uint64_t period; 732 733 ttransport = calloc(1, sizeof(*ttransport)); 734 if (!ttransport) { 735 return NULL; 736 } 737 738 TAILQ_INIT(&ttransport->ports); 739 TAILQ_INIT(&ttransport->poll_groups); 740 TAILQ_INIT(&ttransport->psks); 741 742 ttransport->transport.ops = &spdk_nvmf_transport_tcp; 743 744 ttransport->tcp_opts.c2h_success = SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION; 745 ttransport->tcp_opts.sock_priority = SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY; 746 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 747 if (opts->transport_specific != NULL && 748 spdk_json_decode_object_relaxed(opts->transport_specific, tcp_transport_opts_decoder, 749 SPDK_COUNTOF(tcp_transport_opts_decoder), 750 &ttransport->tcp_opts)) { 751 SPDK_ERRLOG("spdk_json_decode_object_relaxed failed\n"); 752 free(ttransport); 753 return NULL; 754 } 755 756 SPDK_NOTICELOG("*** TCP Transport Init ***\n"); 757 758 SPDK_INFOLOG(nvmf_tcp, "*** TCP Transport Init ***\n" 759 " Transport opts: max_ioq_depth=%d, max_io_size=%d,\n" 760 " max_io_qpairs_per_ctrlr=%d, io_unit_size=%d,\n" 761 " in_capsule_data_size=%d, max_aq_depth=%d\n" 762 " num_shared_buffers=%d, c2h_success=%d,\n" 763 " dif_insert_or_strip=%d, sock_priority=%d\n" 764 " abort_timeout_sec=%d, control_msg_num=%hu\n" 765 " ack_timeout=%d\n", 766 opts->max_queue_depth, 767 opts->max_io_size, 768 opts->max_qpairs_per_ctrlr - 1, 769 opts->io_unit_size, 770 opts->in_capsule_data_size, 771 opts->max_aq_depth, 772 opts->num_shared_buffers, 773 ttransport->tcp_opts.c2h_success, 774 opts->dif_insert_or_strip, 775 ttransport->tcp_opts.sock_priority, 776 opts->abort_timeout_sec, 777 ttransport->tcp_opts.control_msg_num, 778 opts->ack_timeout); 779 780 if (ttransport->tcp_opts.sock_priority > SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY) { 781 SPDK_ERRLOG("Unsupported socket_priority=%d, the current range is: 0 to %d\n" 782 "you can use man 7 socket to view the range of priority under SO_PRIORITY item\n", 783 ttransport->tcp_opts.sock_priority, SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY); 784 free(ttransport); 785 return NULL; 786 } 787 788 if (ttransport->tcp_opts.control_msg_num == 0 && 789 opts->in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 790 SPDK_WARNLOG("TCP param control_msg_num can't be 0 if ICD is less than %u bytes. Using default value %u\n", 791 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM); 792 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 793 } 794 795 /* I/O unit size cannot be larger than max I/O size */ 796 if (opts->io_unit_size > opts->max_io_size) { 797 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", 798 opts->io_unit_size, opts->max_io_size); 799 opts->io_unit_size = opts->max_io_size; 800 } 801 802 /* In capsule data size cannot be larger than max I/O size */ 803 if (opts->in_capsule_data_size > opts->max_io_size) { 804 SPDK_WARNLOG("TCP param ICD size %u can't be larger than max_io_size %u. Using max_io_size as ICD size\n", 805 opts->io_unit_size, opts->max_io_size); 806 opts->in_capsule_data_size = opts->max_io_size; 807 } 808 809 /* max IO queue depth cannot be smaller than 2 or larger than 65535. 810 * We will not check SPDK_NVMF_TCP_MAX_IO_QUEUE_DEPTH, because max_queue_depth is 16bits and always not larger than 64k. */ 811 if (opts->max_queue_depth < SPDK_NVMF_TCP_MIN_IO_QUEUE_DEPTH) { 812 SPDK_WARNLOG("TCP param max_queue_depth %u can't be smaller than %u or larger than %u. Using default value %u\n", 813 opts->max_queue_depth, SPDK_NVMF_TCP_MIN_IO_QUEUE_DEPTH, 814 SPDK_NVMF_TCP_MAX_IO_QUEUE_DEPTH, SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH); 815 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH; 816 } 817 818 /* max admin queue depth cannot be smaller than 2 or larger than 4096 */ 819 if (opts->max_aq_depth < SPDK_NVMF_TCP_MIN_ADMIN_QUEUE_DEPTH || 820 opts->max_aq_depth > SPDK_NVMF_TCP_MAX_ADMIN_QUEUE_DEPTH) { 821 SPDK_WARNLOG("TCP param max_aq_depth %u can't be smaller than %u or larger than %u. Using default value %u\n", 822 opts->max_aq_depth, SPDK_NVMF_TCP_MIN_ADMIN_QUEUE_DEPTH, 823 SPDK_NVMF_TCP_MAX_ADMIN_QUEUE_DEPTH, SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH); 824 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH; 825 } 826 827 sge_count = opts->max_io_size / opts->io_unit_size; 828 if (sge_count > SPDK_NVMF_MAX_SGL_ENTRIES) { 829 SPDK_ERRLOG("Unsupported IO Unit size specified, %d bytes\n", opts->io_unit_size); 830 free(ttransport); 831 return NULL; 832 } 833 834 /* If buf_cache_size == UINT32_MAX, we will dynamically pick a cache size later that we know will fit. */ 835 if (opts->buf_cache_size < UINT32_MAX) { 836 min_shared_buffers = spdk_env_get_core_count() * opts->buf_cache_size; 837 if (min_shared_buffers > opts->num_shared_buffers) { 838 SPDK_ERRLOG("There are not enough buffers to satisfy " 839 "per-poll group caches for each thread. (%" PRIu32 ") " 840 "supplied. (%" PRIu32 ") required\n", opts->num_shared_buffers, min_shared_buffers); 841 SPDK_ERRLOG("Please specify a larger number of shared buffers\n"); 842 free(ttransport); 843 return NULL; 844 } 845 } 846 847 period = spdk_interrupt_mode_is_enabled() ? 0 : opts->acceptor_poll_rate; 848 ttransport->accept_poller = SPDK_POLLER_REGISTER(nvmf_tcp_accept, &ttransport->transport, period); 849 if (!ttransport->accept_poller) { 850 free(ttransport); 851 return NULL; 852 } 853 854 spdk_poller_register_interrupt(ttransport->accept_poller, NULL, NULL); 855 856 ttransport->listen_sock_group = spdk_sock_group_create(NULL); 857 if (ttransport->listen_sock_group == NULL) { 858 SPDK_ERRLOG("Failed to create socket group for listen sockets\n"); 859 spdk_poller_unregister(&ttransport->accept_poller); 860 free(ttransport); 861 return NULL; 862 } 863 864 if (spdk_interrupt_mode_is_enabled()) { 865 rc = SPDK_SOCK_GROUP_REGISTER_INTERRUPT(ttransport->listen_sock_group, 866 SPDK_INTERRUPT_EVENT_IN | SPDK_INTERRUPT_EVENT_OUT, nvmf_tcp_accept, &ttransport->transport); 867 if (rc != 0) { 868 SPDK_ERRLOG("Failed to register interrupt for listen socker sock group\n"); 869 spdk_sock_group_close(&ttransport->listen_sock_group); 870 spdk_poller_unregister(&ttransport->accept_poller); 871 free(ttransport); 872 return NULL; 873 } 874 } 875 876 return &ttransport->transport; 877 } 878 879 static int 880 nvmf_tcp_trsvcid_to_int(const char *trsvcid) 881 { 882 unsigned long long ull; 883 char *end = NULL; 884 885 ull = strtoull(trsvcid, &end, 10); 886 if (end == NULL || end == trsvcid || *end != '\0') { 887 return -1; 888 } 889 890 /* Valid TCP/IP port numbers are in [1, 65535] */ 891 if (ull == 0 || ull > 65535) { 892 return -1; 893 } 894 895 return (int)ull; 896 } 897 898 /** 899 * Canonicalize a listen address trid. 900 */ 901 static int 902 nvmf_tcp_canon_listen_trid(struct spdk_nvme_transport_id *canon_trid, 903 const struct spdk_nvme_transport_id *trid) 904 { 905 int trsvcid_int; 906 907 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 908 if (trsvcid_int < 0) { 909 return -EINVAL; 910 } 911 912 memset(canon_trid, 0, sizeof(*canon_trid)); 913 spdk_nvme_trid_populate_transport(canon_trid, SPDK_NVME_TRANSPORT_TCP); 914 canon_trid->adrfam = trid->adrfam; 915 snprintf(canon_trid->traddr, sizeof(canon_trid->traddr), "%s", trid->traddr); 916 snprintf(canon_trid->trsvcid, sizeof(canon_trid->trsvcid), "%d", trsvcid_int); 917 918 return 0; 919 } 920 921 /** 922 * Find an existing listening port. 923 */ 924 static struct spdk_nvmf_tcp_port * 925 nvmf_tcp_find_port(struct spdk_nvmf_tcp_transport *ttransport, 926 const struct spdk_nvme_transport_id *trid) 927 { 928 struct spdk_nvme_transport_id canon_trid; 929 struct spdk_nvmf_tcp_port *port; 930 931 if (nvmf_tcp_canon_listen_trid(&canon_trid, trid) != 0) { 932 return NULL; 933 } 934 935 TAILQ_FOREACH(port, &ttransport->ports, link) { 936 if (spdk_nvme_transport_id_compare(&canon_trid, port->trid) == 0) { 937 return port; 938 } 939 } 940 941 return NULL; 942 } 943 944 static int 945 tcp_sock_get_key(uint8_t *out, int out_len, const char **cipher, const char *pskid, 946 void *get_key_ctx) 947 { 948 struct tcp_psk_entry *entry; 949 struct spdk_nvmf_tcp_transport *ttransport = get_key_ctx; 950 size_t psk_len; 951 int rc; 952 953 TAILQ_FOREACH(entry, &ttransport->psks, link) { 954 if (strcmp(pskid, entry->pskid) != 0) { 955 continue; 956 } 957 958 psk_len = entry->psk_size; 959 if ((size_t)out_len < psk_len) { 960 SPDK_ERRLOG("Out buffer of size: %" PRIu32 " cannot fit PSK of len: %lu\n", 961 out_len, psk_len); 962 return -ENOBUFS; 963 } 964 965 /* Convert PSK to the TLS PSK format. */ 966 rc = nvme_tcp_derive_tls_psk(entry->psk, psk_len, pskid, out, out_len, 967 entry->tls_cipher_suite); 968 if (rc < 0) { 969 SPDK_ERRLOG("Could not generate TLS PSK\n"); 970 } 971 972 switch (entry->tls_cipher_suite) { 973 case NVME_TCP_CIPHER_AES_128_GCM_SHA256: 974 *cipher = "TLS_AES_128_GCM_SHA256"; 975 break; 976 case NVME_TCP_CIPHER_AES_256_GCM_SHA384: 977 *cipher = "TLS_AES_256_GCM_SHA384"; 978 break; 979 default: 980 *cipher = NULL; 981 return -ENOTSUP; 982 } 983 984 return rc; 985 } 986 987 SPDK_ERRLOG("Could not find PSK for identity: %s\n", pskid); 988 989 return -EINVAL; 990 } 991 992 static int 993 nvmf_tcp_listen(struct spdk_nvmf_transport *transport, const struct spdk_nvme_transport_id *trid, 994 struct spdk_nvmf_listen_opts *listen_opts) 995 { 996 struct spdk_nvmf_tcp_transport *ttransport; 997 struct spdk_nvmf_tcp_port *port; 998 int trsvcid_int; 999 uint8_t adrfam; 1000 const char *sock_impl_name; 1001 struct spdk_sock_impl_opts impl_opts; 1002 size_t impl_opts_size = sizeof(impl_opts); 1003 struct spdk_sock_opts opts; 1004 int rc; 1005 1006 if (!strlen(trid->trsvcid)) { 1007 SPDK_ERRLOG("Service id is required\n"); 1008 return -EINVAL; 1009 } 1010 1011 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1012 1013 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 1014 if (trsvcid_int < 0) { 1015 SPDK_ERRLOG("Invalid trsvcid '%s'\n", trid->trsvcid); 1016 return -EINVAL; 1017 } 1018 1019 port = calloc(1, sizeof(*port)); 1020 if (!port) { 1021 SPDK_ERRLOG("Port allocation failed\n"); 1022 return -ENOMEM; 1023 } 1024 1025 port->trid = trid; 1026 1027 sock_impl_name = NULL; 1028 1029 opts.opts_size = sizeof(opts); 1030 spdk_sock_get_default_opts(&opts); 1031 opts.priority = ttransport->tcp_opts.sock_priority; 1032 opts.ack_timeout = transport->opts.ack_timeout; 1033 if (listen_opts->secure_channel) { 1034 if (listen_opts->sock_impl && 1035 strncmp("ssl", listen_opts->sock_impl, strlen(listen_opts->sock_impl))) { 1036 SPDK_ERRLOG("Enabling secure_channel while specifying a sock_impl different from 'ssl' is unsupported"); 1037 free(port); 1038 return -EINVAL; 1039 } 1040 listen_opts->sock_impl = "ssl"; 1041 } 1042 1043 if (listen_opts->sock_impl) { 1044 sock_impl_name = listen_opts->sock_impl; 1045 spdk_sock_impl_get_opts(sock_impl_name, &impl_opts, &impl_opts_size); 1046 1047 if (!strncmp("ssl", sock_impl_name, strlen(sock_impl_name))) { 1048 if (!g_tls_log) { 1049 SPDK_NOTICELOG("TLS support is considered experimental\n"); 1050 g_tls_log = true; 1051 } 1052 impl_opts.tls_version = SPDK_TLS_VERSION_1_3; 1053 impl_opts.get_key = tcp_sock_get_key; 1054 impl_opts.get_key_ctx = ttransport; 1055 impl_opts.tls_cipher_suites = "TLS_AES_256_GCM_SHA384:TLS_AES_128_GCM_SHA256"; 1056 } 1057 1058 opts.impl_opts = &impl_opts; 1059 opts.impl_opts_size = sizeof(impl_opts); 1060 } 1061 1062 port->listen_sock = spdk_sock_listen_ext(trid->traddr, trsvcid_int, 1063 sock_impl_name, &opts); 1064 if (port->listen_sock == NULL) { 1065 SPDK_ERRLOG("spdk_sock_listen(%s, %d) failed: %s (%d)\n", 1066 trid->traddr, trsvcid_int, 1067 spdk_strerror(errno), errno); 1068 free(port); 1069 return -errno; 1070 } 1071 1072 if (spdk_sock_is_ipv4(port->listen_sock)) { 1073 adrfam = SPDK_NVMF_ADRFAM_IPV4; 1074 } else if (spdk_sock_is_ipv6(port->listen_sock)) { 1075 adrfam = SPDK_NVMF_ADRFAM_IPV6; 1076 } else { 1077 SPDK_ERRLOG("Unhandled socket type\n"); 1078 adrfam = 0; 1079 } 1080 1081 if (adrfam != trid->adrfam) { 1082 SPDK_ERRLOG("Socket address family mismatch\n"); 1083 spdk_sock_close(&port->listen_sock); 1084 free(port); 1085 return -EINVAL; 1086 } 1087 1088 rc = spdk_sock_group_add_sock(ttransport->listen_sock_group, port->listen_sock, nvmf_tcp_accept_cb, 1089 port); 1090 if (rc < 0) { 1091 SPDK_ERRLOG("Failed to add socket to the listen socket group\n"); 1092 spdk_sock_close(&port->listen_sock); 1093 free(port); 1094 return -errno; 1095 } 1096 1097 port->transport = transport; 1098 1099 SPDK_NOTICELOG("*** NVMe/TCP Target Listening on %s port %s ***\n", 1100 trid->traddr, trid->trsvcid); 1101 1102 TAILQ_INSERT_TAIL(&ttransport->ports, port, link); 1103 return 0; 1104 } 1105 1106 static void 1107 nvmf_tcp_stop_listen(struct spdk_nvmf_transport *transport, 1108 const struct spdk_nvme_transport_id *trid) 1109 { 1110 struct spdk_nvmf_tcp_transport *ttransport; 1111 struct spdk_nvmf_tcp_port *port; 1112 1113 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1114 1115 SPDK_DEBUGLOG(nvmf_tcp, "Removing listen address %s port %s\n", 1116 trid->traddr, trid->trsvcid); 1117 1118 port = nvmf_tcp_find_port(ttransport, trid); 1119 if (port) { 1120 spdk_sock_group_remove_sock(ttransport->listen_sock_group, port->listen_sock); 1121 TAILQ_REMOVE(&ttransport->ports, port, link); 1122 spdk_sock_close(&port->listen_sock); 1123 free(port); 1124 } 1125 } 1126 1127 static void nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1128 enum nvme_tcp_pdu_recv_state state); 1129 1130 static void 1131 nvmf_tcp_qpair_set_state(struct spdk_nvmf_tcp_qpair *tqpair, enum nvmf_tcp_qpair_state state) 1132 { 1133 tqpair->state = state; 1134 spdk_trace_record(TRACE_TCP_QP_STATE_CHANGE, tqpair->qpair.trace_id, 0, 0, 1135 (uint64_t)tqpair->state); 1136 } 1137 1138 static void 1139 nvmf_tcp_qpair_disconnect(struct spdk_nvmf_tcp_qpair *tqpair) 1140 { 1141 SPDK_DEBUGLOG(nvmf_tcp, "Disconnecting qpair %p\n", tqpair); 1142 1143 spdk_trace_record(TRACE_TCP_QP_DISCONNECT, tqpair->qpair.trace_id, 0, 0); 1144 1145 if (tqpair->state <= NVMF_TCP_QPAIR_STATE_RUNNING) { 1146 nvmf_tcp_qpair_set_state(tqpair, NVMF_TCP_QPAIR_STATE_EXITING); 1147 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1148 spdk_poller_unregister(&tqpair->timeout_poller); 1149 1150 /* This will end up calling nvmf_tcp_close_qpair */ 1151 spdk_nvmf_qpair_disconnect(&tqpair->qpair); 1152 } 1153 } 1154 1155 static void 1156 _mgmt_pdu_write_done(void *_tqpair, int err) 1157 { 1158 struct spdk_nvmf_tcp_qpair *tqpair = _tqpair; 1159 struct nvme_tcp_pdu *pdu = tqpair->mgmt_pdu; 1160 1161 if (spdk_unlikely(err != 0)) { 1162 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1163 return; 1164 } 1165 1166 assert(pdu->cb_fn != NULL); 1167 pdu->cb_fn(pdu->cb_arg); 1168 } 1169 1170 static void 1171 _req_pdu_write_done(void *req, int err) 1172 { 1173 struct spdk_nvmf_tcp_req *tcp_req = req; 1174 struct nvme_tcp_pdu *pdu = tcp_req->pdu; 1175 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 1176 1177 assert(tcp_req->pdu_in_use); 1178 tcp_req->pdu_in_use = false; 1179 1180 /* If the request is in a completed state, we're waiting for write completion to free it */ 1181 if (spdk_unlikely(tcp_req->state == TCP_REQUEST_STATE_COMPLETED)) { 1182 nvmf_tcp_request_free(tcp_req); 1183 return; 1184 } 1185 1186 if (spdk_unlikely(err != 0)) { 1187 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1188 return; 1189 } 1190 1191 assert(pdu->cb_fn != NULL); 1192 pdu->cb_fn(pdu->cb_arg); 1193 } 1194 1195 static void 1196 _pdu_write_done(struct nvme_tcp_pdu *pdu, int err) 1197 { 1198 pdu->sock_req.cb_fn(pdu->sock_req.cb_arg, err); 1199 } 1200 1201 static void 1202 tcp_sock_flush_cb(void *arg) 1203 { 1204 struct spdk_nvmf_tcp_qpair *tqpair = arg; 1205 int rc = spdk_sock_flush(tqpair->sock); 1206 1207 if (rc < 0 && errno == EAGAIN) { 1208 spdk_thread_send_msg(spdk_get_thread(), tcp_sock_flush_cb, tqpair); 1209 return; 1210 } 1211 1212 tqpair->pending_flush = false; 1213 if (rc < 0) { 1214 SPDK_ERRLOG("Could not write to socket: rc=%d, errno=%d\n", rc, errno); 1215 } 1216 } 1217 1218 static void 1219 _tcp_write_pdu(struct nvme_tcp_pdu *pdu) 1220 { 1221 int rc; 1222 uint32_t mapped_length; 1223 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 1224 1225 pdu->sock_req.iovcnt = nvme_tcp_build_iovs(pdu->iov, SPDK_COUNTOF(pdu->iov), pdu, 1226 tqpair->host_hdgst_enable, tqpair->host_ddgst_enable, &mapped_length); 1227 spdk_sock_writev_async(tqpair->sock, &pdu->sock_req); 1228 1229 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP || 1230 pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ) { 1231 /* Try to force the send immediately. */ 1232 rc = spdk_sock_flush(tqpair->sock); 1233 if (rc > 0 && (uint32_t)rc == mapped_length) { 1234 _pdu_write_done(pdu, 0); 1235 } else { 1236 SPDK_ERRLOG("Could not write %s to socket: rc=%d, errno=%d\n", 1237 pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP ? 1238 "IC_RESP" : "TERM_REQ", rc, errno); 1239 _pdu_write_done(pdu, rc >= 0 ? -EAGAIN : -errno); 1240 } 1241 } else if (spdk_interrupt_mode_is_enabled()) { 1242 /* Async writes must be flushed */ 1243 if (!tqpair->pending_flush) { 1244 tqpair->pending_flush = true; 1245 spdk_thread_send_msg(spdk_get_thread(), tcp_sock_flush_cb, tqpair); 1246 } 1247 } 1248 } 1249 1250 static void 1251 data_crc32_accel_done(void *cb_arg, int status) 1252 { 1253 struct nvme_tcp_pdu *pdu = cb_arg; 1254 1255 if (spdk_unlikely(status)) { 1256 SPDK_ERRLOG("Failed to compute the data digest for pdu =%p\n", pdu); 1257 _pdu_write_done(pdu, status); 1258 return; 1259 } 1260 1261 pdu->data_digest_crc32 ^= SPDK_CRC32C_XOR; 1262 MAKE_DIGEST_WORD(pdu->data_digest, pdu->data_digest_crc32); 1263 1264 _tcp_write_pdu(pdu); 1265 } 1266 1267 static void 1268 pdu_data_crc32_compute(struct nvme_tcp_pdu *pdu) 1269 { 1270 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 1271 int rc = 0; 1272 1273 /* Data Digest */ 1274 if (pdu->data_len > 0 && g_nvme_tcp_ddgst[pdu->hdr.common.pdu_type] && tqpair->host_ddgst_enable) { 1275 /* Only support this limitated case for the first step */ 1276 if (spdk_likely(!pdu->dif_ctx && (pdu->data_len % SPDK_NVME_TCP_DIGEST_ALIGNMENT == 0) 1277 && tqpair->group)) { 1278 rc = spdk_accel_submit_crc32cv(tqpair->group->accel_channel, &pdu->data_digest_crc32, pdu->data_iov, 1279 pdu->data_iovcnt, 0, data_crc32_accel_done, pdu); 1280 if (spdk_likely(rc == 0)) { 1281 return; 1282 } 1283 } else { 1284 pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu); 1285 } 1286 data_crc32_accel_done(pdu, rc); 1287 } else { 1288 _tcp_write_pdu(pdu); 1289 } 1290 } 1291 1292 static void 1293 nvmf_tcp_qpair_write_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1294 struct nvme_tcp_pdu *pdu, 1295 nvme_tcp_qpair_xfer_complete_cb cb_fn, 1296 void *cb_arg) 1297 { 1298 int hlen; 1299 uint32_t crc32c; 1300 1301 assert(tqpair->pdu_in_progress != pdu); 1302 1303 hlen = pdu->hdr.common.hlen; 1304 pdu->cb_fn = cb_fn; 1305 pdu->cb_arg = cb_arg; 1306 1307 pdu->iov[0].iov_base = &pdu->hdr.raw; 1308 pdu->iov[0].iov_len = hlen; 1309 1310 /* Header Digest */ 1311 if (g_nvme_tcp_hdgst[pdu->hdr.common.pdu_type] && tqpair->host_hdgst_enable) { 1312 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 1313 MAKE_DIGEST_WORD((uint8_t *)pdu->hdr.raw + hlen, crc32c); 1314 } 1315 1316 /* Data Digest */ 1317 pdu_data_crc32_compute(pdu); 1318 } 1319 1320 static void 1321 nvmf_tcp_qpair_write_mgmt_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1322 nvme_tcp_qpair_xfer_complete_cb cb_fn, 1323 void *cb_arg) 1324 { 1325 struct nvme_tcp_pdu *pdu = tqpair->mgmt_pdu; 1326 1327 pdu->sock_req.cb_fn = _mgmt_pdu_write_done; 1328 pdu->sock_req.cb_arg = tqpair; 1329 1330 nvmf_tcp_qpair_write_pdu(tqpair, pdu, cb_fn, cb_arg); 1331 } 1332 1333 static void 1334 nvmf_tcp_qpair_write_req_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1335 struct spdk_nvmf_tcp_req *tcp_req, 1336 nvme_tcp_qpair_xfer_complete_cb cb_fn, 1337 void *cb_arg) 1338 { 1339 struct nvme_tcp_pdu *pdu = tcp_req->pdu; 1340 1341 pdu->sock_req.cb_fn = _req_pdu_write_done; 1342 pdu->sock_req.cb_arg = tcp_req; 1343 1344 assert(!tcp_req->pdu_in_use); 1345 tcp_req->pdu_in_use = true; 1346 1347 nvmf_tcp_qpair_write_pdu(tqpair, pdu, cb_fn, cb_arg); 1348 } 1349 1350 static int 1351 nvmf_tcp_qpair_init_mem_resource(struct spdk_nvmf_tcp_qpair *tqpair) 1352 { 1353 uint32_t i; 1354 struct spdk_nvmf_transport_opts *opts; 1355 uint32_t in_capsule_data_size; 1356 1357 opts = &tqpair->qpair.transport->opts; 1358 1359 in_capsule_data_size = opts->in_capsule_data_size; 1360 if (opts->dif_insert_or_strip) { 1361 in_capsule_data_size = SPDK_BDEV_BUF_SIZE_WITH_MD(in_capsule_data_size); 1362 } 1363 1364 tqpair->resource_count = opts->max_queue_depth; 1365 1366 tqpair->reqs = calloc(tqpair->resource_count, sizeof(*tqpair->reqs)); 1367 if (!tqpair->reqs) { 1368 SPDK_ERRLOG("Unable to allocate reqs on tqpair=%p\n", tqpair); 1369 return -1; 1370 } 1371 1372 if (in_capsule_data_size) { 1373 tqpair->bufs = spdk_zmalloc(tqpair->resource_count * in_capsule_data_size, 0x1000, 1374 NULL, SPDK_ENV_LCORE_ID_ANY, 1375 SPDK_MALLOC_DMA); 1376 if (!tqpair->bufs) { 1377 SPDK_ERRLOG("Unable to allocate bufs on tqpair=%p.\n", tqpair); 1378 return -1; 1379 } 1380 } 1381 /* prepare memory space for receiving pdus and tcp_req */ 1382 /* Add additional 1 member, which will be used for mgmt_pdu owned by the tqpair */ 1383 tqpair->pdus = spdk_dma_zmalloc((2 * tqpair->resource_count + 1) * sizeof(*tqpair->pdus), 0x1000, 1384 NULL); 1385 if (!tqpair->pdus) { 1386 SPDK_ERRLOG("Unable to allocate pdu pool on tqpair =%p.\n", tqpair); 1387 return -1; 1388 } 1389 1390 for (i = 0; i < tqpair->resource_count; i++) { 1391 struct spdk_nvmf_tcp_req *tcp_req = &tqpair->reqs[i]; 1392 1393 tcp_req->ttag = i + 1; 1394 tcp_req->req.qpair = &tqpair->qpair; 1395 1396 tcp_req->pdu = &tqpair->pdus[i]; 1397 tcp_req->pdu->qpair = tqpair; 1398 1399 /* Set up memory to receive commands */ 1400 if (tqpair->bufs) { 1401 tcp_req->buf = (void *)((uintptr_t)tqpair->bufs + (i * in_capsule_data_size)); 1402 } 1403 1404 /* Set the cmdn and rsp */ 1405 tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp; 1406 tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd; 1407 1408 tcp_req->req.stripped_data = NULL; 1409 1410 /* Initialize request state to FREE */ 1411 tcp_req->state = TCP_REQUEST_STATE_FREE; 1412 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 1413 tqpair->state_cntr[TCP_REQUEST_STATE_FREE]++; 1414 } 1415 1416 for (; i < 2 * tqpair->resource_count; i++) { 1417 struct nvme_tcp_pdu *pdu = &tqpair->pdus[i]; 1418 1419 pdu->qpair = tqpair; 1420 SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, pdu, slist); 1421 } 1422 1423 tqpair->mgmt_pdu = &tqpair->pdus[i]; 1424 tqpair->mgmt_pdu->qpair = tqpair; 1425 tqpair->pdu_in_progress = SLIST_FIRST(&tqpair->tcp_pdu_free_queue); 1426 SLIST_REMOVE_HEAD(&tqpair->tcp_pdu_free_queue, slist); 1427 tqpair->tcp_pdu_working_count = 1; 1428 1429 tqpair->recv_buf_size = (in_capsule_data_size + sizeof(struct spdk_nvme_tcp_cmd) + 2 * 1430 SPDK_NVME_TCP_DIGEST_LEN) * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1431 1432 return 0; 1433 } 1434 1435 static int 1436 nvmf_tcp_qpair_init(struct spdk_nvmf_qpair *qpair) 1437 { 1438 struct spdk_nvmf_tcp_qpair *tqpair; 1439 1440 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1441 1442 SPDK_DEBUGLOG(nvmf_tcp, "New TCP Connection: %p\n", qpair); 1443 1444 spdk_trace_record(TRACE_TCP_QP_CREATE, tqpair->qpair.trace_id, 0, 0); 1445 1446 /* Initialise request state queues of the qpair */ 1447 TAILQ_INIT(&tqpair->tcp_req_free_queue); 1448 TAILQ_INIT(&tqpair->tcp_req_working_queue); 1449 SLIST_INIT(&tqpair->tcp_pdu_free_queue); 1450 tqpair->qpair.queue_depth = 0; 1451 1452 tqpair->host_hdgst_enable = true; 1453 tqpair->host_ddgst_enable = true; 1454 1455 return 0; 1456 } 1457 1458 static int 1459 nvmf_tcp_qpair_sock_init(struct spdk_nvmf_tcp_qpair *tqpair) 1460 { 1461 char saddr[32], caddr[32]; 1462 uint16_t sport, cport; 1463 char owner[256]; 1464 int rc; 1465 1466 rc = spdk_sock_getaddr(tqpair->sock, saddr, sizeof(saddr), &sport, 1467 caddr, sizeof(caddr), &cport); 1468 if (rc != 0) { 1469 SPDK_ERRLOG("spdk_sock_getaddr() failed\n"); 1470 return rc; 1471 } 1472 snprintf(owner, sizeof(owner), "%s:%d", caddr, cport); 1473 tqpair->qpair.trace_id = spdk_trace_register_owner(OWNER_TYPE_NVMF_TCP, owner); 1474 spdk_trace_record(TRACE_TCP_QP_SOCK_INIT, tqpair->qpair.trace_id, 0, 0); 1475 1476 /* set low water mark */ 1477 rc = spdk_sock_set_recvlowat(tqpair->sock, 1); 1478 if (rc != 0) { 1479 SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n"); 1480 return rc; 1481 } 1482 1483 return 0; 1484 } 1485 1486 static void 1487 nvmf_tcp_handle_connect(struct spdk_nvmf_tcp_port *port, struct spdk_sock *sock) 1488 { 1489 struct spdk_nvmf_tcp_qpair *tqpair; 1490 int rc; 1491 1492 SPDK_DEBUGLOG(nvmf_tcp, "New connection accepted on %s port %s\n", 1493 port->trid->traddr, port->trid->trsvcid); 1494 1495 tqpair = calloc(1, sizeof(struct spdk_nvmf_tcp_qpair)); 1496 if (tqpair == NULL) { 1497 SPDK_ERRLOG("Could not allocate new connection.\n"); 1498 spdk_sock_close(&sock); 1499 return; 1500 } 1501 1502 tqpair->sock = sock; 1503 tqpair->state_cntr[TCP_REQUEST_STATE_FREE] = 0; 1504 tqpair->port = port; 1505 tqpair->qpair.transport = port->transport; 1506 tqpair->qpair.numa.id_valid = 1; 1507 tqpair->qpair.numa.id = spdk_sock_get_numa_id(sock); 1508 1509 rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr, 1510 sizeof(tqpair->target_addr), &tqpair->target_port, 1511 tqpair->initiator_addr, sizeof(tqpair->initiator_addr), 1512 &tqpair->initiator_port); 1513 if (rc < 0) { 1514 SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair); 1515 nvmf_tcp_qpair_destroy(tqpair); 1516 return; 1517 } 1518 1519 spdk_nvmf_tgt_new_qpair(port->transport->tgt, &tqpair->qpair); 1520 } 1521 1522 static uint32_t 1523 nvmf_tcp_port_accept(struct spdk_nvmf_tcp_port *port) 1524 { 1525 struct spdk_sock *sock; 1526 uint32_t count = 0; 1527 int i; 1528 1529 for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) { 1530 sock = spdk_sock_accept(port->listen_sock); 1531 if (sock == NULL) { 1532 break; 1533 } 1534 count++; 1535 nvmf_tcp_handle_connect(port, sock); 1536 } 1537 1538 return count; 1539 } 1540 1541 static int 1542 nvmf_tcp_accept(void *ctx) 1543 { 1544 struct spdk_nvmf_transport *transport = ctx; 1545 struct spdk_nvmf_tcp_transport *ttransport; 1546 int count; 1547 1548 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1549 1550 count = spdk_sock_group_poll(ttransport->listen_sock_group); 1551 if (count < 0) { 1552 SPDK_ERRLOG("Fail in TCP listen socket group poll\n"); 1553 } 1554 1555 return count != 0 ? SPDK_POLLER_BUSY : SPDK_POLLER_IDLE; 1556 } 1557 1558 static void 1559 nvmf_tcp_accept_cb(void *ctx, struct spdk_sock_group *group, struct spdk_sock *sock) 1560 { 1561 struct spdk_nvmf_tcp_port *port = ctx; 1562 1563 nvmf_tcp_port_accept(port); 1564 } 1565 1566 static void 1567 nvmf_tcp_discover(struct spdk_nvmf_transport *transport, 1568 struct spdk_nvme_transport_id *trid, 1569 struct spdk_nvmf_discovery_log_page_entry *entry) 1570 { 1571 struct spdk_nvmf_tcp_port *port; 1572 struct spdk_nvmf_tcp_transport *ttransport; 1573 1574 entry->trtype = SPDK_NVMF_TRTYPE_TCP; 1575 entry->adrfam = trid->adrfam; 1576 1577 spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' '); 1578 spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' '); 1579 1580 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1581 port = nvmf_tcp_find_port(ttransport, trid); 1582 1583 assert(port != NULL); 1584 1585 if (strcmp(spdk_sock_get_impl_name(port->listen_sock), "ssl") == 0) { 1586 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_REQUIRED; 1587 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_TLS_1_3; 1588 } else { 1589 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_REQUIRED; 1590 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE; 1591 } 1592 } 1593 1594 static struct spdk_nvmf_tcp_control_msg_list * 1595 nvmf_tcp_control_msg_list_create(uint16_t num_messages) 1596 { 1597 struct spdk_nvmf_tcp_control_msg_list *list; 1598 struct spdk_nvmf_tcp_control_msg *msg; 1599 uint16_t i; 1600 1601 list = calloc(1, sizeof(*list)); 1602 if (!list) { 1603 SPDK_ERRLOG("Failed to allocate memory for list structure\n"); 1604 return NULL; 1605 } 1606 1607 list->msg_buf = spdk_zmalloc(num_messages * SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, 1608 NVMF_DATA_BUFFER_ALIGNMENT, NULL, SPDK_ENV_NUMA_ID_ANY, SPDK_MALLOC_DMA); 1609 if (!list->msg_buf) { 1610 SPDK_ERRLOG("Failed to allocate memory for control message buffers\n"); 1611 free(list); 1612 return NULL; 1613 } 1614 1615 STAILQ_INIT(&list->free_msgs); 1616 STAILQ_INIT(&list->waiting_for_msg_reqs); 1617 1618 for (i = 0; i < num_messages; i++) { 1619 msg = (struct spdk_nvmf_tcp_control_msg *)((char *)list->msg_buf + i * 1620 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1621 STAILQ_INSERT_TAIL(&list->free_msgs, msg, link); 1622 } 1623 1624 return list; 1625 } 1626 1627 static void 1628 nvmf_tcp_control_msg_list_free(struct spdk_nvmf_tcp_control_msg_list *list) 1629 { 1630 if (!list) { 1631 return; 1632 } 1633 1634 spdk_free(list->msg_buf); 1635 free(list); 1636 } 1637 1638 static int nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group); 1639 1640 static int 1641 nvmf_tcp_poll_group_intr(void *ctx) 1642 { 1643 struct spdk_nvmf_transport_poll_group *group = ctx; 1644 int ret = 0; 1645 1646 ret = nvmf_tcp_poll_group_poll(group); 1647 1648 return ret != 0 ? SPDK_POLLER_BUSY : SPDK_POLLER_IDLE; 1649 } 1650 1651 static struct spdk_nvmf_transport_poll_group * 1652 nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport, 1653 struct spdk_nvmf_poll_group *group) 1654 { 1655 struct spdk_nvmf_tcp_transport *ttransport; 1656 struct spdk_nvmf_tcp_poll_group *tgroup; 1657 int rc; 1658 1659 tgroup = calloc(1, sizeof(*tgroup)); 1660 if (!tgroup) { 1661 return NULL; 1662 } 1663 1664 tgroup->sock_group = spdk_sock_group_create(&tgroup->group); 1665 if (!tgroup->sock_group) { 1666 goto cleanup; 1667 } 1668 1669 TAILQ_INIT(&tgroup->qpairs); 1670 1671 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1672 1673 if (transport->opts.in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 1674 SPDK_DEBUGLOG(nvmf_tcp, "ICD %u is less than min required for admin/fabric commands (%u). " 1675 "Creating control messages list\n", transport->opts.in_capsule_data_size, 1676 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1677 tgroup->control_msg_list = nvmf_tcp_control_msg_list_create(ttransport->tcp_opts.control_msg_num); 1678 if (!tgroup->control_msg_list) { 1679 goto cleanup; 1680 } 1681 } 1682 1683 tgroup->accel_channel = spdk_accel_get_io_channel(); 1684 if (spdk_unlikely(!tgroup->accel_channel)) { 1685 SPDK_ERRLOG("Cannot create accel_channel for tgroup=%p\n", tgroup); 1686 goto cleanup; 1687 } 1688 1689 TAILQ_INSERT_TAIL(&ttransport->poll_groups, tgroup, link); 1690 if (ttransport->next_pg == NULL) { 1691 ttransport->next_pg = tgroup; 1692 } 1693 1694 if (spdk_interrupt_mode_is_enabled()) { 1695 rc = SPDK_SOCK_GROUP_REGISTER_INTERRUPT(tgroup->sock_group, 1696 SPDK_INTERRUPT_EVENT_IN | SPDK_INTERRUPT_EVENT_OUT, nvmf_tcp_poll_group_intr, &tgroup->group); 1697 if (rc != 0) { 1698 SPDK_ERRLOG("Failed to register interrupt for sock group\n"); 1699 goto cleanup; 1700 } 1701 } 1702 1703 return &tgroup->group; 1704 1705 cleanup: 1706 nvmf_tcp_poll_group_destroy(&tgroup->group); 1707 return NULL; 1708 } 1709 1710 static struct spdk_nvmf_transport_poll_group * 1711 nvmf_tcp_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair) 1712 { 1713 struct spdk_nvmf_tcp_transport *ttransport; 1714 struct spdk_nvmf_tcp_poll_group **pg; 1715 struct spdk_nvmf_tcp_qpair *tqpair; 1716 struct spdk_sock_group *group = NULL, *hint = NULL; 1717 int rc; 1718 1719 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 1720 1721 if (TAILQ_EMPTY(&ttransport->poll_groups)) { 1722 return NULL; 1723 } 1724 1725 pg = &ttransport->next_pg; 1726 assert(*pg != NULL); 1727 hint = (*pg)->sock_group; 1728 1729 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1730 rc = spdk_sock_get_optimal_sock_group(tqpair->sock, &group, hint); 1731 if (rc != 0) { 1732 return NULL; 1733 } else if (group != NULL) { 1734 /* Optimal poll group was found */ 1735 return spdk_sock_group_get_ctx(group); 1736 } 1737 1738 /* The hint was used for optimal poll group, advance next_pg. */ 1739 *pg = TAILQ_NEXT(*pg, link); 1740 if (*pg == NULL) { 1741 *pg = TAILQ_FIRST(&ttransport->poll_groups); 1742 } 1743 1744 return spdk_sock_group_get_ctx(hint); 1745 } 1746 1747 static void 1748 nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group) 1749 { 1750 struct spdk_nvmf_tcp_poll_group *tgroup, *next_tgroup; 1751 struct spdk_nvmf_tcp_transport *ttransport; 1752 1753 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 1754 spdk_sock_group_unregister_interrupt(tgroup->sock_group); 1755 spdk_sock_group_close(&tgroup->sock_group); 1756 if (tgroup->control_msg_list) { 1757 nvmf_tcp_control_msg_list_free(tgroup->control_msg_list); 1758 } 1759 1760 if (tgroup->accel_channel) { 1761 spdk_put_io_channel(tgroup->accel_channel); 1762 } 1763 1764 if (tgroup->group.transport == NULL) { 1765 /* Transport can be NULL when nvmf_tcp_poll_group_create() 1766 * calls this function directly in a failure path. */ 1767 free(tgroup); 1768 return; 1769 } 1770 1771 ttransport = SPDK_CONTAINEROF(tgroup->group.transport, struct spdk_nvmf_tcp_transport, transport); 1772 1773 next_tgroup = TAILQ_NEXT(tgroup, link); 1774 TAILQ_REMOVE(&ttransport->poll_groups, tgroup, link); 1775 if (next_tgroup == NULL) { 1776 next_tgroup = TAILQ_FIRST(&ttransport->poll_groups); 1777 } 1778 if (ttransport->next_pg == tgroup) { 1779 ttransport->next_pg = next_tgroup; 1780 } 1781 1782 free(tgroup); 1783 } 1784 1785 static void 1786 nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1787 enum nvme_tcp_pdu_recv_state state) 1788 { 1789 if (tqpair->recv_state == state) { 1790 SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n", 1791 tqpair, state); 1792 return; 1793 } 1794 1795 if (spdk_unlikely(state == NVME_TCP_PDU_RECV_STATE_QUIESCING)) { 1796 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH && tqpair->pdu_in_progress) { 1797 SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, tqpair->pdu_in_progress, slist); 1798 tqpair->tcp_pdu_working_count--; 1799 } 1800 } 1801 1802 if (spdk_unlikely(state == NVME_TCP_PDU_RECV_STATE_ERROR)) { 1803 assert(tqpair->tcp_pdu_working_count == 0); 1804 } 1805 1806 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv state=%d\n", tqpair, state); 1807 tqpair->recv_state = state; 1808 1809 spdk_trace_record(TRACE_TCP_QP_RCV_STATE_CHANGE, tqpair->qpair.trace_id, 0, 0, 1810 (uint64_t)tqpair->recv_state); 1811 } 1812 1813 static int 1814 nvmf_tcp_qpair_handle_timeout(void *ctx) 1815 { 1816 struct spdk_nvmf_tcp_qpair *tqpair = ctx; 1817 1818 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1819 1820 SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair, 1821 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT); 1822 1823 nvmf_tcp_qpair_disconnect(tqpair); 1824 return SPDK_POLLER_BUSY; 1825 } 1826 1827 static void 1828 nvmf_tcp_send_c2h_term_req_complete(void *cb_arg) 1829 { 1830 struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg; 1831 1832 if (!tqpair->timeout_poller) { 1833 tqpair->timeout_poller = SPDK_POLLER_REGISTER(nvmf_tcp_qpair_handle_timeout, tqpair, 1834 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000); 1835 } 1836 } 1837 1838 static void 1839 nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu, 1840 enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset) 1841 { 1842 struct nvme_tcp_pdu *rsp_pdu; 1843 struct spdk_nvme_tcp_term_req_hdr *c2h_term_req; 1844 uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req); 1845 uint32_t copy_len; 1846 1847 rsp_pdu = tqpair->mgmt_pdu; 1848 1849 c2h_term_req = &rsp_pdu->hdr.term_req; 1850 c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ; 1851 c2h_term_req->common.hlen = c2h_term_req_hdr_len; 1852 c2h_term_req->fes = fes; 1853 1854 if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1855 (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1856 DSET32(&c2h_term_req->fei, error_offset); 1857 } 1858 1859 copy_len = spdk_min(pdu->hdr.common.hlen, SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE); 1860 1861 /* Copy the error info into the buffer */ 1862 memcpy((uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, pdu->hdr.raw, copy_len); 1863 nvme_tcp_pdu_set_data(rsp_pdu, (uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, copy_len); 1864 1865 /* Contain the header of the wrong received pdu */ 1866 c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len; 1867 tqpair->wait_terminate = true; 1868 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1869 nvmf_tcp_qpair_write_mgmt_pdu(tqpair, nvmf_tcp_send_c2h_term_req_complete, tqpair); 1870 } 1871 1872 static void 1873 nvmf_tcp_capsule_cmd_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1874 struct spdk_nvmf_tcp_qpair *tqpair, 1875 struct nvme_tcp_pdu *pdu) 1876 { 1877 struct spdk_nvmf_tcp_req *tcp_req; 1878 1879 assert(pdu->psh_valid_bytes == pdu->psh_len); 1880 assert(pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD); 1881 1882 tcp_req = nvmf_tcp_req_get(tqpair); 1883 if (!tcp_req) { 1884 /* Directly return and make the allocation retry again. This can happen if we're 1885 * using asynchronous writes to send the response to the host or when releasing 1886 * zero-copy buffers after a response has been sent. In both cases, the host might 1887 * receive the response before we've finished processing the request and is free to 1888 * send another one. 1889 */ 1890 if (tqpair->state_cntr[TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST] > 0 || 1891 tqpair->state_cntr[TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE] > 0) { 1892 return; 1893 } 1894 1895 /* The host sent more commands than the maximum queue depth. */ 1896 SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair); 1897 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1898 return; 1899 } 1900 1901 pdu->req = tcp_req; 1902 assert(tcp_req->state == TCP_REQUEST_STATE_NEW); 1903 nvmf_tcp_req_process(ttransport, tcp_req); 1904 } 1905 1906 static void 1907 nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1908 struct spdk_nvmf_tcp_qpair *tqpair, 1909 struct nvme_tcp_pdu *pdu) 1910 { 1911 struct spdk_nvmf_tcp_req *tcp_req; 1912 struct spdk_nvme_tcp_cmd *capsule_cmd; 1913 uint32_t error_offset = 0; 1914 enum spdk_nvme_tcp_term_req_fes fes; 1915 struct spdk_nvme_cpl *rsp; 1916 1917 capsule_cmd = &pdu->hdr.capsule_cmd; 1918 tcp_req = pdu->req; 1919 assert(tcp_req != NULL); 1920 1921 /* Zero-copy requests don't support ICD */ 1922 assert(!spdk_nvmf_request_using_zcopy(&tcp_req->req)); 1923 1924 if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) { 1925 SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n", 1926 SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo); 1927 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1928 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1929 goto err; 1930 } 1931 1932 rsp = &tcp_req->req.rsp->nvme_cpl; 1933 if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) { 1934 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 1935 } else { 1936 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1937 } 1938 1939 nvmf_tcp_req_process(ttransport, tcp_req); 1940 1941 return; 1942 err: 1943 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1944 } 1945 1946 static void 1947 nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1948 struct spdk_nvmf_tcp_qpair *tqpair, 1949 struct nvme_tcp_pdu *pdu) 1950 { 1951 struct spdk_nvmf_tcp_req *tcp_req; 1952 uint32_t error_offset = 0; 1953 enum spdk_nvme_tcp_term_req_fes fes = 0; 1954 struct spdk_nvme_tcp_h2c_data_hdr *h2c_data; 1955 1956 h2c_data = &pdu->hdr.h2c_data; 1957 1958 SPDK_DEBUGLOG(nvmf_tcp, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n", 1959 tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag); 1960 1961 if (h2c_data->ttag > tqpair->resource_count) { 1962 SPDK_DEBUGLOG(nvmf_tcp, "ttag %u is larger than allowed %u.\n", h2c_data->ttag, 1963 tqpair->resource_count); 1964 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1965 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1966 goto err; 1967 } 1968 1969 tcp_req = &tqpair->reqs[h2c_data->ttag - 1]; 1970 1971 if (spdk_unlikely(tcp_req->state != TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER && 1972 tcp_req->state != TCP_REQUEST_STATE_AWAITING_R2T_ACK)) { 1973 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req(%p), tqpair=%p, has error state in %d\n", tcp_req, tqpair, 1974 tcp_req->state); 1975 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1976 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1977 goto err; 1978 } 1979 1980 if (spdk_unlikely(tcp_req->req.cmd->nvme_cmd.cid != h2c_data->cccid)) { 1981 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req(%p), tqpair=%p, expected %u but %u for cccid.\n", tcp_req, tqpair, 1982 tcp_req->req.cmd->nvme_cmd.cid, h2c_data->cccid); 1983 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1984 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid); 1985 goto err; 1986 } 1987 1988 if (tcp_req->h2c_offset != h2c_data->datao) { 1989 SPDK_DEBUGLOG(nvmf_tcp, 1990 "tcp_req(%p), tqpair=%p, expected data offset %u, but data offset is %u\n", 1991 tcp_req, tqpair, tcp_req->h2c_offset, h2c_data->datao); 1992 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1993 goto err; 1994 } 1995 1996 if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) { 1997 SPDK_DEBUGLOG(nvmf_tcp, 1998 "tcp_req(%p), tqpair=%p, (datao=%u + datal=%u) exceeds requested length=%u\n", 1999 tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length); 2000 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 2001 goto err; 2002 } 2003 2004 pdu->req = tcp_req; 2005 2006 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2007 pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 2008 } 2009 2010 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2011 h2c_data->datao, h2c_data->datal); 2012 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2013 return; 2014 2015 err: 2016 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2017 } 2018 2019 static void 2020 nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req, 2021 struct spdk_nvmf_tcp_qpair *tqpair) 2022 { 2023 struct nvme_tcp_pdu *rsp_pdu; 2024 struct spdk_nvme_tcp_rsp *capsule_resp; 2025 2026 SPDK_DEBUGLOG(nvmf_tcp, "enter, tqpair=%p\n", tqpair); 2027 2028 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 2029 assert(rsp_pdu != NULL); 2030 2031 capsule_resp = &rsp_pdu->hdr.capsule_resp; 2032 capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP; 2033 capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp); 2034 capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl; 2035 if (tqpair->host_hdgst_enable) { 2036 capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 2037 capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 2038 } 2039 2040 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_request_free, tcp_req); 2041 } 2042 2043 static void 2044 nvmf_tcp_pdu_c2h_data_complete(void *cb_arg) 2045 { 2046 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 2047 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, 2048 struct spdk_nvmf_tcp_qpair, qpair); 2049 2050 assert(tqpair != NULL); 2051 2052 if (spdk_unlikely(tcp_req->pdu->rw_offset < tcp_req->req.length)) { 2053 SPDK_DEBUGLOG(nvmf_tcp, "sending another C2H part, offset %u length %u\n", tcp_req->pdu->rw_offset, 2054 tcp_req->req.length); 2055 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2056 return; 2057 } 2058 2059 if (tcp_req->pdu->hdr.c2h_data.common.flags & SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS) { 2060 nvmf_tcp_request_free(tcp_req); 2061 } else { 2062 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2063 } 2064 } 2065 2066 static void 2067 nvmf_tcp_r2t_complete(void *cb_arg) 2068 { 2069 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 2070 struct spdk_nvmf_tcp_transport *ttransport; 2071 2072 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 2073 struct spdk_nvmf_tcp_transport, transport); 2074 2075 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 2076 2077 if (tcp_req->h2c_offset == tcp_req->req.length) { 2078 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2079 nvmf_tcp_req_process(ttransport, tcp_req); 2080 } 2081 } 2082 2083 static void 2084 nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 2085 struct spdk_nvmf_tcp_req *tcp_req) 2086 { 2087 struct nvme_tcp_pdu *rsp_pdu; 2088 struct spdk_nvme_tcp_r2t_hdr *r2t; 2089 2090 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 2091 assert(rsp_pdu != NULL); 2092 2093 r2t = &rsp_pdu->hdr.r2t; 2094 r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T; 2095 r2t->common.plen = r2t->common.hlen = sizeof(*r2t); 2096 2097 if (tqpair->host_hdgst_enable) { 2098 r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 2099 r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 2100 } 2101 2102 r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid; 2103 r2t->ttag = tcp_req->ttag; 2104 r2t->r2to = tcp_req->h2c_offset; 2105 r2t->r2tl = tcp_req->req.length; 2106 2107 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 2108 2109 SPDK_DEBUGLOG(nvmf_tcp, 2110 "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n", 2111 tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl); 2112 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_r2t_complete, tcp_req); 2113 } 2114 2115 static void 2116 nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 2117 struct spdk_nvmf_tcp_qpair *tqpair, 2118 struct nvme_tcp_pdu *pdu) 2119 { 2120 struct spdk_nvmf_tcp_req *tcp_req; 2121 struct spdk_nvme_cpl *rsp; 2122 2123 tcp_req = pdu->req; 2124 assert(tcp_req != NULL); 2125 2126 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2127 2128 tcp_req->h2c_offset += pdu->data_len; 2129 2130 /* Wait for all of the data to arrive AND for the initial R2T PDU send to be 2131 * acknowledged before moving on. */ 2132 if (tcp_req->h2c_offset == tcp_req->req.length && 2133 tcp_req->state == TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER) { 2134 /* After receiving all the h2c data, we need to check whether there is 2135 * transient transport error */ 2136 rsp = &tcp_req->req.rsp->nvme_cpl; 2137 if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) { 2138 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2139 } else { 2140 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2141 } 2142 nvmf_tcp_req_process(ttransport, tcp_req); 2143 } 2144 } 2145 2146 static void 2147 nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req) 2148 { 2149 SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req, 2150 spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]); 2151 if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 2152 (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 2153 SPDK_DEBUGLOG(nvmf_tcp, "The offset from the start of the PDU header is %u\n", 2154 DGET32(h2c_term_req->fei)); 2155 } 2156 } 2157 2158 static void 2159 nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair, 2160 struct nvme_tcp_pdu *pdu) 2161 { 2162 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 2163 uint32_t error_offset = 0; 2164 enum spdk_nvme_tcp_term_req_fes fes; 2165 2166 if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) { 2167 SPDK_ERRLOG("Fatal Error Status(FES) is unknown for h2c_term_req pdu=%p\n", pdu); 2168 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2169 error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes); 2170 goto end; 2171 } 2172 2173 /* set the data buffer */ 2174 nvme_tcp_pdu_set_data(pdu, (uint8_t *)pdu->hdr.raw + h2c_term_req->common.hlen, 2175 h2c_term_req->common.plen - h2c_term_req->common.hlen); 2176 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2177 return; 2178 end: 2179 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2180 } 2181 2182 static void 2183 nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 2184 struct nvme_tcp_pdu *pdu) 2185 { 2186 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 2187 2188 nvmf_tcp_h2c_term_req_dump(h2c_term_req); 2189 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2190 } 2191 2192 static void 2193 _nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu) 2194 { 2195 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 2196 struct spdk_nvmf_tcp_transport, transport); 2197 2198 switch (pdu->hdr.common.pdu_type) { 2199 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2200 nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu); 2201 break; 2202 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2203 nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu); 2204 break; 2205 2206 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2207 nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu); 2208 break; 2209 2210 default: 2211 /* The code should not go to here */ 2212 SPDK_ERRLOG("ERROR pdu type %d\n", pdu->hdr.common.pdu_type); 2213 break; 2214 } 2215 SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, pdu, slist); 2216 tqpair->tcp_pdu_working_count--; 2217 } 2218 2219 static inline void 2220 nvmf_tcp_req_set_cpl(struct spdk_nvmf_tcp_req *treq, int sct, int sc) 2221 { 2222 treq->req.rsp->nvme_cpl.status.sct = sct; 2223 treq->req.rsp->nvme_cpl.status.sc = sc; 2224 treq->req.rsp->nvme_cpl.cid = treq->req.cmd->nvme_cmd.cid; 2225 } 2226 2227 static void 2228 data_crc32_calc_done(void *cb_arg, int status) 2229 { 2230 struct nvme_tcp_pdu *pdu = cb_arg; 2231 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 2232 2233 /* async crc32 calculation is failed and use direct calculation to check */ 2234 if (spdk_unlikely(status)) { 2235 SPDK_ERRLOG("Data digest on tqpair=(%p) with pdu=%p failed to be calculated asynchronously\n", 2236 tqpair, pdu); 2237 pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu); 2238 } 2239 pdu->data_digest_crc32 ^= SPDK_CRC32C_XOR; 2240 if (!MATCH_DIGEST_WORD(pdu->data_digest, pdu->data_digest_crc32)) { 2241 SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 2242 assert(pdu->req != NULL); 2243 nvmf_tcp_req_set_cpl(pdu->req, SPDK_NVME_SCT_GENERIC, 2244 SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR); 2245 } 2246 _nvmf_tcp_pdu_payload_handle(tqpair, pdu); 2247 } 2248 2249 static void 2250 nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu) 2251 { 2252 int rc = 0; 2253 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2254 tqpair->pdu_in_progress = NULL; 2255 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2256 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2257 /* check data digest if need */ 2258 if (pdu->ddgst_enable) { 2259 if (tqpair->qpair.qid != 0 && !pdu->dif_ctx && tqpair->group && 2260 (pdu->data_len % SPDK_NVME_TCP_DIGEST_ALIGNMENT == 0)) { 2261 rc = spdk_accel_submit_crc32cv(tqpair->group->accel_channel, &pdu->data_digest_crc32, pdu->data_iov, 2262 pdu->data_iovcnt, 0, data_crc32_calc_done, pdu); 2263 if (spdk_likely(rc == 0)) { 2264 return; 2265 } 2266 } else { 2267 pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu); 2268 } 2269 data_crc32_calc_done(pdu, rc); 2270 } else { 2271 _nvmf_tcp_pdu_payload_handle(tqpair, pdu); 2272 } 2273 } 2274 2275 static void 2276 nvmf_tcp_send_icresp_complete(void *cb_arg) 2277 { 2278 struct spdk_nvmf_tcp_qpair *tqpair = cb_arg; 2279 2280 nvmf_tcp_qpair_set_state(tqpair, NVMF_TCP_QPAIR_STATE_RUNNING); 2281 } 2282 2283 static void 2284 nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport, 2285 struct spdk_nvmf_tcp_qpair *tqpair, 2286 struct nvme_tcp_pdu *pdu) 2287 { 2288 struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr.ic_req; 2289 struct nvme_tcp_pdu *rsp_pdu; 2290 struct spdk_nvme_tcp_ic_resp *ic_resp; 2291 uint32_t error_offset = 0; 2292 enum spdk_nvme_tcp_term_req_fes fes; 2293 2294 /* Only PFV 0 is defined currently */ 2295 if (ic_req->pfv != 0) { 2296 SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv); 2297 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2298 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv); 2299 goto end; 2300 } 2301 2302 /* This value is 0’s based value in units of dwords should not be larger than SPDK_NVME_TCP_HPDA_MAX */ 2303 if (ic_req->hpda > SPDK_NVME_TCP_HPDA_MAX) { 2304 SPDK_ERRLOG("ICReq HPDA out of range 0 to 31, got %u\n", ic_req->hpda); 2305 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2306 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, hpda); 2307 goto end; 2308 } 2309 2310 /* MAXR2T is 0's based */ 2311 SPDK_DEBUGLOG(nvmf_tcp, "maxr2t =%u\n", (ic_req->maxr2t + 1u)); 2312 2313 tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false; 2314 if (!tqpair->host_hdgst_enable) { 2315 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 2316 } 2317 2318 tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false; 2319 if (!tqpair->host_ddgst_enable) { 2320 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 2321 } 2322 2323 tqpair->recv_buf_size = spdk_max(tqpair->recv_buf_size, MIN_SOCK_PIPE_SIZE); 2324 /* Now that we know whether digests are enabled, properly size the receive buffer */ 2325 if (spdk_sock_set_recvbuf(tqpair->sock, tqpair->recv_buf_size) < 0) { 2326 SPDK_WARNLOG("Unable to allocate enough memory for receive buffer on tqpair=%p with size=%d\n", 2327 tqpair, 2328 tqpair->recv_buf_size); 2329 /* Not fatal. */ 2330 } 2331 2332 tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX); 2333 SPDK_DEBUGLOG(nvmf_tcp, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda); 2334 2335 rsp_pdu = tqpair->mgmt_pdu; 2336 2337 ic_resp = &rsp_pdu->hdr.ic_resp; 2338 ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP; 2339 ic_resp->common.hlen = ic_resp->common.plen = sizeof(*ic_resp); 2340 ic_resp->pfv = 0; 2341 ic_resp->cpda = tqpair->cpda; 2342 ic_resp->maxh2cdata = ttransport->transport.opts.max_io_size; 2343 ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0; 2344 ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0; 2345 2346 SPDK_DEBUGLOG(nvmf_tcp, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable); 2347 SPDK_DEBUGLOG(nvmf_tcp, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable); 2348 2349 nvmf_tcp_qpair_set_state(tqpair, NVMF_TCP_QPAIR_STATE_INITIALIZING); 2350 nvmf_tcp_qpair_write_mgmt_pdu(tqpair, nvmf_tcp_send_icresp_complete, tqpair); 2351 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2352 return; 2353 end: 2354 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2355 } 2356 2357 static void 2358 nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair, 2359 struct spdk_nvmf_tcp_transport *ttransport) 2360 { 2361 struct nvme_tcp_pdu *pdu; 2362 int rc; 2363 uint32_t crc32c, error_offset = 0; 2364 enum spdk_nvme_tcp_term_req_fes fes; 2365 2366 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 2367 pdu = tqpair->pdu_in_progress; 2368 2369 SPDK_DEBUGLOG(nvmf_tcp, "pdu type of tqpair(%p) is %d\n", tqpair, 2370 pdu->hdr.common.pdu_type); 2371 /* check header digest if needed */ 2372 if (pdu->has_hdgst) { 2373 SPDK_DEBUGLOG(nvmf_tcp, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair); 2374 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 2375 rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, crc32c); 2376 if (rc == 0) { 2377 SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 2378 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 2379 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2380 return; 2381 2382 } 2383 } 2384 2385 switch (pdu->hdr.common.pdu_type) { 2386 case SPDK_NVME_TCP_PDU_TYPE_IC_REQ: 2387 nvmf_tcp_icreq_handle(ttransport, tqpair, pdu); 2388 break; 2389 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2390 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_REQ); 2391 break; 2392 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2393 nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu); 2394 break; 2395 2396 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2397 nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu); 2398 break; 2399 2400 default: 2401 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress->hdr.common.pdu_type); 2402 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2403 error_offset = 1; 2404 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2405 break; 2406 } 2407 } 2408 2409 static void 2410 nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair) 2411 { 2412 struct nvme_tcp_pdu *pdu; 2413 uint32_t error_offset = 0; 2414 enum spdk_nvme_tcp_term_req_fes fes; 2415 uint8_t expected_hlen, pdo; 2416 bool plen_error = false, pdo_error = false; 2417 2418 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 2419 pdu = tqpair->pdu_in_progress; 2420 assert(pdu); 2421 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) { 2422 if (tqpair->state != NVMF_TCP_QPAIR_STATE_INVALID) { 2423 SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu); 2424 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 2425 goto err; 2426 } 2427 expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req); 2428 if (pdu->hdr.common.plen != expected_hlen) { 2429 plen_error = true; 2430 } 2431 } else { 2432 if (tqpair->state != NVMF_TCP_QPAIR_STATE_RUNNING) { 2433 SPDK_ERRLOG("The TCP/IP connection is not negotiated\n"); 2434 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 2435 goto err; 2436 } 2437 2438 switch (pdu->hdr.common.pdu_type) { 2439 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2440 expected_hlen = sizeof(struct spdk_nvme_tcp_cmd); 2441 pdo = pdu->hdr.common.pdo; 2442 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 2443 pdo_error = true; 2444 break; 2445 } 2446 2447 if (pdu->hdr.common.plen < expected_hlen) { 2448 plen_error = true; 2449 } 2450 break; 2451 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2452 expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr); 2453 pdo = pdu->hdr.common.pdo; 2454 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 2455 pdo_error = true; 2456 break; 2457 } 2458 if (pdu->hdr.common.plen < expected_hlen) { 2459 plen_error = true; 2460 } 2461 break; 2462 2463 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2464 expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr); 2465 if ((pdu->hdr.common.plen <= expected_hlen) || 2466 (pdu->hdr.common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) { 2467 plen_error = true; 2468 } 2469 break; 2470 2471 default: 2472 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr.common.pdu_type); 2473 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2474 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type); 2475 goto err; 2476 } 2477 } 2478 2479 if (pdu->hdr.common.hlen != expected_hlen) { 2480 SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n", 2481 pdu->hdr.common.pdu_type, 2482 expected_hlen, pdu->hdr.common.hlen, tqpair); 2483 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2484 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen); 2485 goto err; 2486 } else if (pdo_error) { 2487 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2488 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 2489 } else if (plen_error) { 2490 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2491 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 2492 goto err; 2493 } else { 2494 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 2495 nvme_tcp_pdu_calc_psh_len(tqpair->pdu_in_progress, tqpair->host_hdgst_enable); 2496 return; 2497 } 2498 err: 2499 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2500 } 2501 2502 static int 2503 nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair) 2504 { 2505 int rc = 0; 2506 struct nvme_tcp_pdu *pdu; 2507 enum nvme_tcp_pdu_recv_state prev_state; 2508 uint32_t data_len; 2509 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 2510 struct spdk_nvmf_tcp_transport, transport); 2511 2512 /* The loop here is to allow for several back-to-back state changes. */ 2513 do { 2514 prev_state = tqpair->recv_state; 2515 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state); 2516 2517 pdu = tqpair->pdu_in_progress; 2518 assert(pdu != NULL || 2519 tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY || 2520 tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_QUIESCING || 2521 tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 2522 2523 switch (tqpair->recv_state) { 2524 /* Wait for the common header */ 2525 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 2526 if (!pdu) { 2527 pdu = SLIST_FIRST(&tqpair->tcp_pdu_free_queue); 2528 if (spdk_unlikely(!pdu)) { 2529 return NVME_TCP_PDU_IN_PROGRESS; 2530 } 2531 SLIST_REMOVE_HEAD(&tqpair->tcp_pdu_free_queue, slist); 2532 tqpair->pdu_in_progress = pdu; 2533 tqpair->tcp_pdu_working_count++; 2534 } 2535 memset(pdu, 0, offsetof(struct nvme_tcp_pdu, qpair)); 2536 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 2537 /* FALLTHROUGH */ 2538 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 2539 if (spdk_unlikely(tqpair->state == NVMF_TCP_QPAIR_STATE_INITIALIZING)) { 2540 return rc; 2541 } 2542 2543 rc = nvme_tcp_read_data(tqpair->sock, 2544 sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes, 2545 (void *)&pdu->hdr.common + pdu->ch_valid_bytes); 2546 if (rc < 0) { 2547 SPDK_DEBUGLOG(nvmf_tcp, "will disconnect tqpair=%p\n", tqpair); 2548 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2549 break; 2550 } else if (rc > 0) { 2551 pdu->ch_valid_bytes += rc; 2552 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, tqpair->qpair.trace_id, rc, 0); 2553 } 2554 2555 if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) { 2556 return NVME_TCP_PDU_IN_PROGRESS; 2557 } 2558 2559 /* The command header of this PDU has now been read from the socket. */ 2560 nvmf_tcp_pdu_ch_handle(tqpair); 2561 break; 2562 /* Wait for the pdu specific header */ 2563 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 2564 rc = nvme_tcp_read_data(tqpair->sock, 2565 pdu->psh_len - pdu->psh_valid_bytes, 2566 (void *)&pdu->hdr.raw + sizeof(struct spdk_nvme_tcp_common_pdu_hdr) + pdu->psh_valid_bytes); 2567 if (rc < 0) { 2568 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2569 break; 2570 } else if (rc > 0) { 2571 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, tqpair->qpair.trace_id, rc, 0); 2572 pdu->psh_valid_bytes += rc; 2573 } 2574 2575 if (pdu->psh_valid_bytes < pdu->psh_len) { 2576 return NVME_TCP_PDU_IN_PROGRESS; 2577 } 2578 2579 /* All header(ch, psh, head digits) of this PDU has now been read from the socket. */ 2580 nvmf_tcp_pdu_psh_handle(tqpair, ttransport); 2581 break; 2582 /* Wait for the req slot */ 2583 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 2584 nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu); 2585 break; 2586 /* Wait for the request processing loop to acquire a buffer for the PDU */ 2587 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_BUF: 2588 break; 2589 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 2590 /* check whether the data is valid, if not we just return */ 2591 if (!pdu->data_len) { 2592 return NVME_TCP_PDU_IN_PROGRESS; 2593 } 2594 2595 data_len = pdu->data_len; 2596 /* data digest */ 2597 if (spdk_unlikely((pdu->hdr.common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) && 2598 tqpair->host_ddgst_enable)) { 2599 data_len += SPDK_NVME_TCP_DIGEST_LEN; 2600 pdu->ddgst_enable = true; 2601 } 2602 2603 rc = nvme_tcp_read_payload_data(tqpair->sock, pdu); 2604 if (rc < 0) { 2605 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2606 break; 2607 } 2608 pdu->rw_offset += rc; 2609 2610 if (pdu->rw_offset < data_len) { 2611 return NVME_TCP_PDU_IN_PROGRESS; 2612 } 2613 2614 /* Generate and insert DIF to whole data block received if DIF is enabled */ 2615 if (spdk_unlikely(pdu->dif_ctx != NULL) && 2616 spdk_dif_generate_stream(pdu->data_iov, pdu->data_iovcnt, 0, data_len, 2617 pdu->dif_ctx) != 0) { 2618 SPDK_ERRLOG("DIF generate failed\n"); 2619 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2620 break; 2621 } 2622 2623 /* All of this PDU has now been read from the socket. */ 2624 nvmf_tcp_pdu_payload_handle(tqpair, pdu); 2625 break; 2626 case NVME_TCP_PDU_RECV_STATE_QUIESCING: 2627 if (tqpair->tcp_pdu_working_count != 0) { 2628 return NVME_TCP_PDU_IN_PROGRESS; 2629 } 2630 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 2631 break; 2632 case NVME_TCP_PDU_RECV_STATE_ERROR: 2633 if (spdk_sock_is_connected(tqpair->sock) && tqpair->wait_terminate) { 2634 return NVME_TCP_PDU_IN_PROGRESS; 2635 } 2636 return NVME_TCP_PDU_FATAL; 2637 default: 2638 SPDK_ERRLOG("The state(%d) is invalid\n", tqpair->recv_state); 2639 abort(); 2640 break; 2641 } 2642 } while (tqpair->recv_state != prev_state); 2643 2644 return rc; 2645 } 2646 2647 static inline void * 2648 nvmf_tcp_control_msg_get(struct spdk_nvmf_tcp_control_msg_list *list, 2649 struct spdk_nvmf_tcp_req *tcp_req) 2650 { 2651 struct spdk_nvmf_tcp_control_msg *msg; 2652 2653 assert(list); 2654 2655 msg = STAILQ_FIRST(&list->free_msgs); 2656 if (!msg) { 2657 SPDK_DEBUGLOG(nvmf_tcp, "Out of control messages\n"); 2658 STAILQ_INSERT_TAIL(&list->waiting_for_msg_reqs, tcp_req, control_msg_link); 2659 return NULL; 2660 } 2661 STAILQ_REMOVE_HEAD(&list->free_msgs, link); 2662 return msg; 2663 } 2664 2665 static inline void 2666 nvmf_tcp_control_msg_put(struct spdk_nvmf_tcp_control_msg_list *list, void *_msg) 2667 { 2668 struct spdk_nvmf_tcp_control_msg *msg = _msg; 2669 struct spdk_nvmf_tcp_req *tcp_req; 2670 struct spdk_nvmf_tcp_transport *ttransport; 2671 2672 assert(list); 2673 STAILQ_INSERT_HEAD(&list->free_msgs, msg, link); 2674 if (!STAILQ_EMPTY(&list->waiting_for_msg_reqs)) { 2675 tcp_req = STAILQ_FIRST(&list->waiting_for_msg_reqs); 2676 STAILQ_REMOVE_HEAD(&list->waiting_for_msg_reqs, control_msg_link); 2677 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 2678 struct spdk_nvmf_tcp_transport, transport); 2679 nvmf_tcp_req_process(ttransport, tcp_req); 2680 } 2681 } 2682 2683 static void 2684 nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_req *tcp_req, 2685 struct spdk_nvmf_transport *transport, 2686 struct spdk_nvmf_transport_poll_group *group) 2687 { 2688 struct spdk_nvmf_request *req = &tcp_req->req; 2689 struct spdk_nvme_cmd *cmd; 2690 struct spdk_nvme_sgl_descriptor *sgl; 2691 struct spdk_nvmf_tcp_poll_group *tgroup; 2692 enum spdk_nvme_tcp_term_req_fes fes; 2693 struct nvme_tcp_pdu *pdu; 2694 struct spdk_nvmf_tcp_qpair *tqpair; 2695 uint32_t length, error_offset = 0; 2696 2697 cmd = &req->cmd->nvme_cmd; 2698 sgl = &cmd->dptr.sgl1; 2699 2700 if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK && 2701 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) { 2702 /* get request length from sgl */ 2703 length = sgl->unkeyed.length; 2704 if (spdk_unlikely(length > transport->opts.max_io_size)) { 2705 SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n", 2706 length, transport->opts.max_io_size); 2707 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_LIMIT_EXCEEDED; 2708 goto fatal_err; 2709 } 2710 2711 /* fill request length and populate iovs */ 2712 req->length = length; 2713 2714 SPDK_DEBUGLOG(nvmf_tcp, "Data requested length= 0x%x\n", length); 2715 2716 if (spdk_unlikely(req->dif_enabled)) { 2717 req->dif.orig_length = length; 2718 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2719 req->dif.elba_length = length; 2720 } 2721 2722 if (nvmf_ctrlr_use_zcopy(req)) { 2723 SPDK_DEBUGLOG(nvmf_tcp, "Using zero-copy to execute request %p\n", tcp_req); 2724 req->data_from_pool = false; 2725 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_HAVE_BUFFER); 2726 return; 2727 } 2728 2729 if (spdk_nvmf_request_get_buffers(req, group, transport, length)) { 2730 /* No available buffers. Queue this request up. */ 2731 SPDK_DEBUGLOG(nvmf_tcp, "No available large data buffers. Queueing request %p\n", 2732 tcp_req); 2733 return; 2734 } 2735 2736 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_HAVE_BUFFER); 2737 SPDK_DEBUGLOG(nvmf_tcp, "Request %p took %d buffer/s from central pool, and data=%p\n", 2738 tcp_req, req->iovcnt, req->iov[0].iov_base); 2739 2740 return; 2741 } else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK && 2742 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) { 2743 uint64_t offset = sgl->address; 2744 uint32_t max_len = transport->opts.in_capsule_data_size; 2745 2746 assert(tcp_req->has_in_capsule_data); 2747 /* Capsule Cmd with In-capsule Data should get data length from pdu header */ 2748 tqpair = tcp_req->pdu->qpair; 2749 /* receiving pdu is not same with the pdu in tcp_req */ 2750 pdu = tqpair->pdu_in_progress; 2751 length = pdu->hdr.common.plen - pdu->psh_len - sizeof(struct spdk_nvme_tcp_common_pdu_hdr); 2752 if (tqpair->host_ddgst_enable) { 2753 length -= SPDK_NVME_TCP_DIGEST_LEN; 2754 } 2755 /* This error is not defined in NVMe/TCP spec, take this error as fatal error */ 2756 if (spdk_unlikely(length != sgl->unkeyed.length)) { 2757 SPDK_ERRLOG("In-Capsule Data length 0x%x is not equal to SGL data length 0x%x\n", 2758 length, sgl->unkeyed.length); 2759 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2760 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 2761 goto fatal_err; 2762 } 2763 2764 SPDK_DEBUGLOG(nvmf_tcp, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n", 2765 offset, length); 2766 2767 /* The NVMe/TCP transport does not use ICDOFF to control the in-capsule data offset. ICDOFF should be '0' */ 2768 if (spdk_unlikely(offset != 0)) { 2769 /* Not defined fatal error in NVMe/TCP spec, handle this error as a fatal error */ 2770 SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " should be ZERO in NVMe/TCP\n", offset); 2771 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 2772 error_offset = offsetof(struct spdk_nvme_tcp_cmd, ccsqe.dptr.sgl1.address); 2773 goto fatal_err; 2774 } 2775 2776 if (spdk_unlikely(length > max_len)) { 2777 /* According to the SPEC we should support ICD up to 8192 bytes for admin and fabric commands */ 2778 if (length <= SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE && 2779 (cmd->opc == SPDK_NVME_OPC_FABRIC || req->qpair->qid == 0)) { 2780 2781 /* Get a buffer from dedicated list */ 2782 SPDK_DEBUGLOG(nvmf_tcp, "Getting a buffer from control msg list\n"); 2783 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2784 assert(tgroup->control_msg_list); 2785 req->iov[0].iov_base = nvmf_tcp_control_msg_get(tgroup->control_msg_list, tcp_req); 2786 if (!req->iov[0].iov_base) { 2787 /* No available buffers. Queue this request up. */ 2788 SPDK_DEBUGLOG(nvmf_tcp, "No available ICD buffers. Queueing request %p\n", tcp_req); 2789 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_BUF); 2790 return; 2791 } 2792 } else { 2793 SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n", 2794 length, max_len); 2795 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_LIMIT_EXCEEDED; 2796 goto fatal_err; 2797 } 2798 } else { 2799 req->iov[0].iov_base = tcp_req->buf; 2800 } 2801 2802 req->length = length; 2803 req->data_from_pool = false; 2804 2805 if (spdk_unlikely(req->dif_enabled)) { 2806 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2807 req->dif.elba_length = length; 2808 } 2809 2810 req->iov[0].iov_len = length; 2811 req->iovcnt = 1; 2812 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_HAVE_BUFFER); 2813 2814 return; 2815 } 2816 /* If we want to handle the problem here, then we can't skip the following data segment. 2817 * Because this function runs before reading data part, now handle all errors as fatal errors. */ 2818 SPDK_ERRLOG("Invalid NVMf I/O Command SGL: Type 0x%x, Subtype 0x%x\n", 2819 sgl->generic.type, sgl->generic.subtype); 2820 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 2821 error_offset = offsetof(struct spdk_nvme_tcp_cmd, ccsqe.dptr.sgl1.generic); 2822 fatal_err: 2823 nvmf_tcp_send_c2h_term_req(tcp_req->pdu->qpair, tcp_req->pdu, fes, error_offset); 2824 } 2825 2826 static inline enum spdk_nvme_media_error_status_code 2827 nvmf_tcp_dif_error_to_compl_status(uint8_t err_type) { 2828 enum spdk_nvme_media_error_status_code result; 2829 2830 switch (err_type) 2831 { 2832 case SPDK_DIF_REFTAG_ERROR: 2833 result = SPDK_NVME_SC_REFERENCE_TAG_CHECK_ERROR; 2834 break; 2835 case SPDK_DIF_APPTAG_ERROR: 2836 result = SPDK_NVME_SC_APPLICATION_TAG_CHECK_ERROR; 2837 break; 2838 case SPDK_DIF_GUARD_ERROR: 2839 result = SPDK_NVME_SC_GUARD_CHECK_ERROR; 2840 break; 2841 default: 2842 SPDK_UNREACHABLE(); 2843 break; 2844 } 2845 2846 return result; 2847 } 2848 2849 static void 2850 _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2851 struct spdk_nvmf_tcp_req *tcp_req) 2852 { 2853 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF( 2854 tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 2855 struct nvme_tcp_pdu *rsp_pdu; 2856 struct spdk_nvme_tcp_c2h_data_hdr *c2h_data; 2857 uint32_t plen, pdo, alignment; 2858 int rc; 2859 2860 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2861 2862 rsp_pdu = tcp_req->pdu; 2863 assert(rsp_pdu != NULL); 2864 2865 c2h_data = &rsp_pdu->hdr.c2h_data; 2866 c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA; 2867 plen = c2h_data->common.hlen = sizeof(*c2h_data); 2868 2869 if (tqpair->host_hdgst_enable) { 2870 plen += SPDK_NVME_TCP_DIGEST_LEN; 2871 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 2872 } 2873 2874 /* set the psh */ 2875 c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid; 2876 c2h_data->datal = tcp_req->req.length - tcp_req->pdu->rw_offset; 2877 c2h_data->datao = tcp_req->pdu->rw_offset; 2878 2879 /* set the padding */ 2880 rsp_pdu->padding_len = 0; 2881 pdo = plen; 2882 if (tqpair->cpda) { 2883 alignment = (tqpair->cpda + 1) << 2; 2884 if (plen % alignment != 0) { 2885 pdo = (plen + alignment) / alignment * alignment; 2886 rsp_pdu->padding_len = pdo - plen; 2887 plen = pdo; 2888 } 2889 } 2890 2891 c2h_data->common.pdo = pdo; 2892 plen += c2h_data->datal; 2893 if (tqpair->host_ddgst_enable) { 2894 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF; 2895 plen += SPDK_NVME_TCP_DIGEST_LEN; 2896 } 2897 2898 c2h_data->common.plen = plen; 2899 2900 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2901 rsp_pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 2902 } 2903 2904 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2905 c2h_data->datao, c2h_data->datal); 2906 2907 2908 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU; 2909 /* Need to send the capsule response if response is not all 0 */ 2910 if (ttransport->tcp_opts.c2h_success && 2911 tcp_req->rsp.cdw0 == 0 && tcp_req->rsp.cdw1 == 0) { 2912 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS; 2913 } 2914 2915 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2916 struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl; 2917 struct spdk_dif_error err_blk = {}; 2918 uint32_t mapped_length = 0; 2919 uint32_t available_iovs = SPDK_COUNTOF(rsp_pdu->iov); 2920 uint32_t ddgst_len = 0; 2921 2922 if (tqpair->host_ddgst_enable) { 2923 /* Data digest consumes additional iov entry */ 2924 available_iovs--; 2925 /* plen needs to be updated since nvme_tcp_build_iovs compares expected and actual plen */ 2926 ddgst_len = SPDK_NVME_TCP_DIGEST_LEN; 2927 c2h_data->common.plen -= ddgst_len; 2928 } 2929 /* Temp call to estimate if data can be described by limited number of iovs. 2930 * iov vector will be rebuilt in nvmf_tcp_qpair_write_pdu */ 2931 nvme_tcp_build_iovs(rsp_pdu->iov, available_iovs, rsp_pdu, tqpair->host_hdgst_enable, 2932 false, &mapped_length); 2933 2934 if (mapped_length != c2h_data->common.plen) { 2935 c2h_data->datal = mapped_length - (c2h_data->common.plen - c2h_data->datal); 2936 SPDK_DEBUGLOG(nvmf_tcp, 2937 "Part C2H, data_len %u (of %u), PDU len %u, updated PDU len %u, offset %u\n", 2938 c2h_data->datal, tcp_req->req.length, c2h_data->common.plen, mapped_length, rsp_pdu->rw_offset); 2939 c2h_data->common.plen = mapped_length; 2940 2941 /* Rebuild pdu->data_iov since data length is changed */ 2942 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, c2h_data->datao, 2943 c2h_data->datal); 2944 2945 c2h_data->common.flags &= ~(SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU | 2946 SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS); 2947 } 2948 2949 c2h_data->common.plen += ddgst_len; 2950 2951 assert(rsp_pdu->rw_offset <= tcp_req->req.length); 2952 2953 rc = spdk_dif_verify_stream(rsp_pdu->data_iov, rsp_pdu->data_iovcnt, 2954 0, rsp_pdu->data_len, rsp_pdu->dif_ctx, &err_blk); 2955 if (rc != 0) { 2956 SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n", 2957 err_blk.err_type, err_blk.err_offset); 2958 rsp->status.sct = SPDK_NVME_SCT_MEDIA_ERROR; 2959 rsp->status.sc = nvmf_tcp_dif_error_to_compl_status(err_blk.err_type); 2960 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2961 return; 2962 } 2963 } 2964 2965 rsp_pdu->rw_offset += c2h_data->datal; 2966 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_pdu_c2h_data_complete, tcp_req); 2967 } 2968 2969 static void 2970 nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2971 struct spdk_nvmf_tcp_req *tcp_req) 2972 { 2973 nvmf_tcp_req_pdu_init(tcp_req); 2974 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2975 } 2976 2977 static int 2978 request_transfer_out(struct spdk_nvmf_request *req) 2979 { 2980 struct spdk_nvmf_tcp_req *tcp_req; 2981 struct spdk_nvmf_qpair *qpair; 2982 struct spdk_nvmf_tcp_qpair *tqpair; 2983 struct spdk_nvme_cpl *rsp; 2984 2985 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2986 2987 qpair = req->qpair; 2988 rsp = &req->rsp->nvme_cpl; 2989 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2990 2991 /* Advance our sq_head pointer */ 2992 if (qpair->sq_head == qpair->sq_head_max) { 2993 qpair->sq_head = 0; 2994 } else { 2995 qpair->sq_head++; 2996 } 2997 rsp->sqhd = qpair->sq_head; 2998 2999 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 3000 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 3001 if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) { 3002 nvmf_tcp_send_c2h_data(tqpair, tcp_req); 3003 } else { 3004 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 3005 } 3006 3007 return 0; 3008 } 3009 3010 static void 3011 nvmf_tcp_check_fused_ordering(struct spdk_nvmf_tcp_transport *ttransport, 3012 struct spdk_nvmf_tcp_qpair *tqpair, 3013 struct spdk_nvmf_tcp_req *tcp_req) 3014 { 3015 enum spdk_nvme_cmd_fuse last, next; 3016 3017 last = tqpair->fused_first ? tqpair->fused_first->cmd.fuse : SPDK_NVME_CMD_FUSE_NONE; 3018 next = tcp_req->cmd.fuse; 3019 3020 assert(last != SPDK_NVME_CMD_FUSE_SECOND); 3021 3022 if (spdk_likely(last == SPDK_NVME_CMD_FUSE_NONE && next == SPDK_NVME_CMD_FUSE_NONE)) { 3023 return; 3024 } 3025 3026 if (last == SPDK_NVME_CMD_FUSE_FIRST) { 3027 if (next == SPDK_NVME_CMD_FUSE_SECOND) { 3028 /* This is a valid pair of fused commands. Point them at each other 3029 * so they can be submitted consecutively once ready to be executed. 3030 */ 3031 tqpair->fused_first->fused_pair = tcp_req; 3032 tcp_req->fused_pair = tqpair->fused_first; 3033 tqpair->fused_first = NULL; 3034 return; 3035 } else { 3036 /* Mark the last req as failed since it wasn't followed by a SECOND. */ 3037 tqpair->fused_first->fused_failed = true; 3038 3039 /* 3040 * If the last req is in READY_TO_EXECUTE state, then call 3041 * nvmf_tcp_req_process(), otherwise nothing else will kick it. 3042 */ 3043 if (tqpair->fused_first->state == TCP_REQUEST_STATE_READY_TO_EXECUTE) { 3044 nvmf_tcp_req_process(ttransport, tqpair->fused_first); 3045 } 3046 3047 tqpair->fused_first = NULL; 3048 } 3049 } 3050 3051 if (next == SPDK_NVME_CMD_FUSE_FIRST) { 3052 /* Set tqpair->fused_first here so that we know to check that the next request 3053 * is a SECOND (and to fail this one if it isn't). 3054 */ 3055 tqpair->fused_first = tcp_req; 3056 } else if (next == SPDK_NVME_CMD_FUSE_SECOND) { 3057 /* Mark this req failed since it is a SECOND and the last one was not a FIRST. */ 3058 tcp_req->fused_failed = true; 3059 } 3060 } 3061 3062 static bool 3063 nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 3064 struct spdk_nvmf_tcp_req *tcp_req) 3065 { 3066 struct spdk_nvmf_tcp_qpair *tqpair; 3067 uint32_t plen; 3068 struct nvme_tcp_pdu *pdu; 3069 enum spdk_nvmf_tcp_req_state prev_state; 3070 bool progress = false; 3071 struct spdk_nvmf_transport *transport = &ttransport->transport; 3072 struct spdk_nvmf_transport_poll_group *group; 3073 struct spdk_nvmf_tcp_poll_group *tgroup; 3074 3075 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 3076 group = &tqpair->group->group; 3077 assert(tcp_req->state != TCP_REQUEST_STATE_FREE); 3078 3079 /* If the qpair is not active, we need to abort the outstanding requests. */ 3080 if (!spdk_nvmf_qpair_is_active(&tqpair->qpair)) { 3081 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 3082 nvmf_tcp_request_get_buffers_abort(tcp_req); 3083 } 3084 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 3085 } 3086 3087 /* The loop here is to allow for several back-to-back state changes. */ 3088 do { 3089 prev_state = tcp_req->state; 3090 3091 SPDK_DEBUGLOG(nvmf_tcp, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state, 3092 tqpair); 3093 3094 switch (tcp_req->state) { 3095 case TCP_REQUEST_STATE_FREE: 3096 /* Some external code must kick a request into TCP_REQUEST_STATE_NEW 3097 * to escape this state. */ 3098 break; 3099 case TCP_REQUEST_STATE_NEW: 3100 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req, 3101 tqpair->qpair.queue_depth); 3102 3103 /* copy the cmd from the receive pdu */ 3104 tcp_req->cmd = tqpair->pdu_in_progress->hdr.capsule_cmd.ccsqe; 3105 3106 if (spdk_unlikely(spdk_nvmf_request_get_dif_ctx(&tcp_req->req, &tcp_req->req.dif.dif_ctx))) { 3107 tcp_req->req.dif_enabled = true; 3108 tqpair->pdu_in_progress->dif_ctx = &tcp_req->req.dif.dif_ctx; 3109 } 3110 3111 nvmf_tcp_check_fused_ordering(ttransport, tqpair, tcp_req); 3112 3113 /* The next state transition depends on the data transfer needs of this request. */ 3114 tcp_req->req.xfer = spdk_nvmf_req_get_xfer(&tcp_req->req); 3115 3116 if (spdk_unlikely(tcp_req->req.xfer == SPDK_NVME_DATA_BIDIRECTIONAL)) { 3117 nvmf_tcp_req_set_cpl(tcp_req, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_INVALID_OPCODE); 3118 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 3119 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3120 SPDK_DEBUGLOG(nvmf_tcp, "Request %p: invalid xfer type (BIDIRECTIONAL)\n", tcp_req); 3121 break; 3122 } 3123 3124 /* If no data to transfer, ready to execute. */ 3125 if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) { 3126 /* Reset the tqpair receiving pdu state */ 3127 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 3128 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 3129 break; 3130 } 3131 3132 pdu = tqpair->pdu_in_progress; 3133 plen = pdu->hdr.common.hlen; 3134 if (tqpair->host_hdgst_enable) { 3135 plen += SPDK_NVME_TCP_DIGEST_LEN; 3136 } 3137 if (pdu->hdr.common.plen != plen) { 3138 tcp_req->has_in_capsule_data = true; 3139 } else { 3140 /* Data is transmitted by C2H PDUs */ 3141 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 3142 } 3143 3144 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER); 3145 break; 3146 case TCP_REQUEST_STATE_NEED_BUFFER: 3147 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, tqpair->qpair.trace_id, 0, 3148 (uintptr_t)tcp_req); 3149 3150 assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE); 3151 3152 /* Try to get a data buffer */ 3153 nvmf_tcp_req_parse_sgl(tcp_req, transport, group); 3154 break; 3155 case TCP_REQUEST_STATE_HAVE_BUFFER: 3156 spdk_trace_record(TRACE_TCP_REQUEST_STATE_HAVE_BUFFER, tqpair->qpair.trace_id, 0, 3157 (uintptr_t)tcp_req); 3158 /* Get a zcopy buffer if the request can be serviced through zcopy */ 3159 if (spdk_nvmf_request_using_zcopy(&tcp_req->req)) { 3160 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 3161 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 3162 tcp_req->req.length = tcp_req->req.dif.elba_length; 3163 } 3164 3165 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_START); 3166 spdk_nvmf_request_zcopy_start(&tcp_req->req); 3167 break; 3168 } 3169 3170 assert(tcp_req->req.iovcnt > 0); 3171 3172 /* If data is transferring from host to controller, we need to do a transfer from the host. */ 3173 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 3174 if (tcp_req->req.data_from_pool) { 3175 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 3176 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 3177 } else { 3178 struct nvme_tcp_pdu *pdu; 3179 3180 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 3181 3182 pdu = tqpair->pdu_in_progress; 3183 SPDK_DEBUGLOG(nvmf_tcp, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, 3184 tqpair); 3185 /* No need to send r2t, contained in the capsuled data */ 3186 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 3187 0, tcp_req->req.length); 3188 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 3189 } 3190 break; 3191 } 3192 3193 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 3194 break; 3195 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3196 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_START, tqpair->qpair.trace_id, 0, 3197 (uintptr_t)tcp_req); 3198 /* Some external code must kick a request into TCP_REQUEST_STATE_ZCOPY_START_COMPLETED 3199 * to escape this state. */ 3200 break; 3201 case TCP_REQUEST_STATE_ZCOPY_START_COMPLETED: 3202 spdk_trace_record(TRACE_TCP_REQUEST_STATE_ZCOPY_START_COMPLETED, tqpair->qpair.trace_id, 0, 3203 (uintptr_t)tcp_req); 3204 if (spdk_unlikely(spdk_nvme_cpl_is_error(&tcp_req->req.rsp->nvme_cpl))) { 3205 SPDK_DEBUGLOG(nvmf_tcp, "Zero-copy start failed for tcp_req(%p) on tqpair=%p\n", 3206 tcp_req, tqpair); 3207 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3208 break; 3209 } 3210 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 3211 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 3212 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 3213 } else { 3214 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 3215 } 3216 break; 3217 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 3218 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, tqpair->qpair.trace_id, 0, 3219 (uintptr_t)tcp_req); 3220 /* The R2T completion or the h2c data incoming will kick it out of this state. */ 3221 break; 3222 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 3223 3224 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, tqpair->qpair.trace_id, 3225 0, (uintptr_t)tcp_req); 3226 /* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE 3227 * to escape this state. */ 3228 break; 3229 case TCP_REQUEST_STATE_READY_TO_EXECUTE: 3230 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, tqpair->qpair.trace_id, 0, 3231 (uintptr_t)tcp_req); 3232 3233 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 3234 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 3235 tcp_req->req.length = tcp_req->req.dif.elba_length; 3236 } 3237 3238 if (tcp_req->cmd.fuse != SPDK_NVME_CMD_FUSE_NONE) { 3239 if (tcp_req->fused_failed) { 3240 /* This request failed FUSED semantics. Fail it immediately, without 3241 * even sending it to the target layer. 3242 */ 3243 nvmf_tcp_req_set_cpl(tcp_req, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_ABORTED_MISSING_FUSED); 3244 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3245 break; 3246 } 3247 3248 if (tcp_req->fused_pair == NULL || 3249 tcp_req->fused_pair->state != TCP_REQUEST_STATE_READY_TO_EXECUTE) { 3250 /* This request is ready to execute, but either we don't know yet if it's 3251 * valid - i.e. this is a FIRST but we haven't received the next request yet), 3252 * or the other request of this fused pair isn't ready to execute. So 3253 * break here and this request will get processed later either when the 3254 * other request is ready or we find that this request isn't valid. 3255 */ 3256 break; 3257 } 3258 } 3259 3260 if (!spdk_nvmf_request_using_zcopy(&tcp_req->req)) { 3261 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING); 3262 /* If we get to this point, and this request is a fused command, we know that 3263 * it is part of a valid sequence (FIRST followed by a SECOND) and that both 3264 * requests are READY_TO_EXECUTE. So call spdk_nvmf_request_exec() both on this 3265 * request, and the other request of the fused pair, in the correct order. 3266 * Also clear the ->fused_pair pointers on both requests, since after this point 3267 * we no longer need to maintain the relationship between these two requests. 3268 */ 3269 if (tcp_req->cmd.fuse == SPDK_NVME_CMD_FUSE_SECOND) { 3270 assert(tcp_req->fused_pair != NULL); 3271 assert(tcp_req->fused_pair->fused_pair == tcp_req); 3272 nvmf_tcp_req_set_state(tcp_req->fused_pair, TCP_REQUEST_STATE_EXECUTING); 3273 spdk_nvmf_request_exec(&tcp_req->fused_pair->req); 3274 tcp_req->fused_pair->fused_pair = NULL; 3275 tcp_req->fused_pair = NULL; 3276 } 3277 spdk_nvmf_request_exec(&tcp_req->req); 3278 if (tcp_req->cmd.fuse == SPDK_NVME_CMD_FUSE_FIRST) { 3279 assert(tcp_req->fused_pair != NULL); 3280 assert(tcp_req->fused_pair->fused_pair == tcp_req); 3281 nvmf_tcp_req_set_state(tcp_req->fused_pair, TCP_REQUEST_STATE_EXECUTING); 3282 spdk_nvmf_request_exec(&tcp_req->fused_pair->req); 3283 tcp_req->fused_pair->fused_pair = NULL; 3284 tcp_req->fused_pair = NULL; 3285 } 3286 } else { 3287 /* For zero-copy, only requests with data coming from host to the 3288 * controller can end up here. */ 3289 assert(tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER); 3290 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT); 3291 spdk_nvmf_request_zcopy_end(&tcp_req->req, true); 3292 } 3293 3294 break; 3295 case TCP_REQUEST_STATE_EXECUTING: 3296 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req); 3297 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 3298 * to escape this state. */ 3299 break; 3300 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3301 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_COMMIT, tqpair->qpair.trace_id, 0, 3302 (uintptr_t)tcp_req); 3303 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 3304 * to escape this state. */ 3305 break; 3306 case TCP_REQUEST_STATE_EXECUTED: 3307 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req); 3308 3309 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 3310 tcp_req->req.length = tcp_req->req.dif.orig_length; 3311 } 3312 3313 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3314 break; 3315 case TCP_REQUEST_STATE_READY_TO_COMPLETE: 3316 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, tqpair->qpair.trace_id, 0, 3317 (uintptr_t)tcp_req); 3318 if (request_transfer_out(&tcp_req->req) != 0) { 3319 assert(0); /* No good way to handle this currently */ 3320 } 3321 break; 3322 case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST: 3323 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, tqpair->qpair.trace_id, 3324 0, (uintptr_t)tcp_req); 3325 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 3326 * to escape this state. */ 3327 break; 3328 case TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE: 3329 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_RELEASE, tqpair->qpair.trace_id, 0, 3330 (uintptr_t)tcp_req); 3331 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 3332 * to escape this state. */ 3333 break; 3334 case TCP_REQUEST_STATE_COMPLETED: 3335 spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req, 3336 tqpair->qpair.queue_depth); 3337 /* If there's an outstanding PDU sent to the host, the request is completed 3338 * due to the qpair being disconnected. We must delay the completion until 3339 * that write is done to avoid freeing the request twice. */ 3340 if (spdk_unlikely(tcp_req->pdu_in_use)) { 3341 SPDK_DEBUGLOG(nvmf_tcp, "Delaying completion due to outstanding " 3342 "write on req=%p\n", tcp_req); 3343 /* This can only happen for zcopy requests */ 3344 assert(spdk_nvmf_request_using_zcopy(&tcp_req->req)); 3345 assert(!spdk_nvmf_qpair_is_active(&tqpair->qpair)); 3346 break; 3347 } 3348 3349 if (tcp_req->req.data_from_pool) { 3350 spdk_nvmf_request_free_buffers(&tcp_req->req, group, transport); 3351 } else if (spdk_unlikely(tcp_req->has_in_capsule_data && 3352 (tcp_req->cmd.opc == SPDK_NVME_OPC_FABRIC || 3353 tqpair->qpair.qid == 0) && tcp_req->req.length > transport->opts.in_capsule_data_size)) { 3354 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3355 assert(tgroup->control_msg_list); 3356 SPDK_DEBUGLOG(nvmf_tcp, "Put buf to control msg list\n"); 3357 nvmf_tcp_control_msg_put(tgroup->control_msg_list, 3358 tcp_req->req.iov[0].iov_base); 3359 } else if (tcp_req->req.zcopy_bdev_io != NULL) { 3360 /* If the request has an unreleased zcopy bdev_io, it's either a 3361 * read, a failed write, or the qpair is being disconnected */ 3362 assert(spdk_nvmf_request_using_zcopy(&tcp_req->req)); 3363 assert(tcp_req->req.xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST || 3364 spdk_nvme_cpl_is_error(&tcp_req->req.rsp->nvme_cpl) || 3365 !spdk_nvmf_qpair_is_active(&tqpair->qpair)); 3366 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE); 3367 spdk_nvmf_request_zcopy_end(&tcp_req->req, false); 3368 break; 3369 } 3370 tcp_req->req.length = 0; 3371 tcp_req->req.iovcnt = 0; 3372 tcp_req->fused_failed = false; 3373 if (tcp_req->fused_pair) { 3374 /* This req was part of a valid fused pair, but failed before it got to 3375 * READ_TO_EXECUTE state. This means we need to fail the other request 3376 * in the pair, because it is no longer part of a valid pair. If the pair 3377 * already reached READY_TO_EXECUTE state, we need to kick it. 3378 */ 3379 tcp_req->fused_pair->fused_failed = true; 3380 if (tcp_req->fused_pair->state == TCP_REQUEST_STATE_READY_TO_EXECUTE) { 3381 nvmf_tcp_req_process(ttransport, tcp_req->fused_pair); 3382 } 3383 tcp_req->fused_pair = NULL; 3384 } 3385 3386 nvmf_tcp_req_put(tqpair, tcp_req); 3387 break; 3388 case TCP_REQUEST_NUM_STATES: 3389 default: 3390 assert(0); 3391 break; 3392 } 3393 3394 if (tcp_req->state != prev_state) { 3395 progress = true; 3396 } 3397 } while (tcp_req->state != prev_state); 3398 3399 return progress; 3400 } 3401 3402 static void 3403 nvmf_tcp_qpair_process(struct spdk_nvmf_tcp_qpair *tqpair) 3404 { 3405 int rc; 3406 3407 assert(tqpair != NULL); 3408 rc = nvmf_tcp_sock_process(tqpair); 3409 3410 /* If there was a new socket error, disconnect */ 3411 if (rc < 0) { 3412 nvmf_tcp_qpair_disconnect(tqpair); 3413 } 3414 } 3415 3416 static void 3417 nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock) 3418 { 3419 struct spdk_nvmf_tcp_qpair *tqpair = arg; 3420 3421 nvmf_tcp_qpair_process(tqpair); 3422 } 3423 3424 static int 3425 nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group, 3426 struct spdk_nvmf_qpair *qpair) 3427 { 3428 struct spdk_nvmf_tcp_poll_group *tgroup; 3429 struct spdk_nvmf_tcp_qpair *tqpair; 3430 int rc; 3431 3432 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3433 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3434 3435 rc = nvmf_tcp_qpair_sock_init(tqpair); 3436 if (rc != 0) { 3437 SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair); 3438 return -1; 3439 } 3440 3441 rc = nvmf_tcp_qpair_init(&tqpair->qpair); 3442 if (rc < 0) { 3443 SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair); 3444 return -1; 3445 } 3446 3447 rc = nvmf_tcp_qpair_init_mem_resource(tqpair); 3448 if (rc < 0) { 3449 SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair); 3450 return -1; 3451 } 3452 3453 rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock, 3454 nvmf_tcp_sock_cb, tqpair); 3455 if (rc != 0) { 3456 SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n", 3457 spdk_strerror(errno), errno); 3458 return -1; 3459 } 3460 3461 tqpair->group = tgroup; 3462 nvmf_tcp_qpair_set_state(tqpair, NVMF_TCP_QPAIR_STATE_INVALID); 3463 TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link); 3464 3465 return 0; 3466 } 3467 3468 static int 3469 nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group, 3470 struct spdk_nvmf_qpair *qpair) 3471 { 3472 struct spdk_nvmf_tcp_poll_group *tgroup; 3473 struct spdk_nvmf_tcp_qpair *tqpair; 3474 int rc; 3475 3476 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3477 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3478 3479 assert(tqpair->group == tgroup); 3480 3481 SPDK_DEBUGLOG(nvmf_tcp, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup); 3482 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 3483 /* Change the state to move the qpair from the await_req list to the main list 3484 * and prevent adding it again later by nvmf_tcp_qpair_set_recv_state() */ 3485 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 3486 } 3487 TAILQ_REMOVE(&tgroup->qpairs, tqpair, link); 3488 3489 /* Try to force out any pending writes */ 3490 spdk_sock_flush(tqpair->sock); 3491 3492 rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock); 3493 if (rc != 0) { 3494 SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n", 3495 spdk_strerror(errno), errno); 3496 } 3497 3498 return rc; 3499 } 3500 3501 static int 3502 nvmf_tcp_req_complete(struct spdk_nvmf_request *req) 3503 { 3504 struct spdk_nvmf_tcp_transport *ttransport; 3505 struct spdk_nvmf_tcp_req *tcp_req; 3506 3507 ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport); 3508 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 3509 3510 switch (tcp_req->state) { 3511 case TCP_REQUEST_STATE_EXECUTING: 3512 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3513 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 3514 break; 3515 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3516 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_ZCOPY_START_COMPLETED); 3517 break; 3518 case TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE: 3519 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 3520 break; 3521 default: 3522 SPDK_ERRLOG("Unexpected request state %d (cntlid:%d, qid:%d)\n", 3523 tcp_req->state, req->qpair->ctrlr->cntlid, req->qpair->qid); 3524 assert(0 && "Unexpected request state"); 3525 break; 3526 } 3527 3528 nvmf_tcp_req_process(ttransport, tcp_req); 3529 3530 return 0; 3531 } 3532 3533 static void 3534 nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair, 3535 spdk_nvmf_transport_qpair_fini_cb cb_fn, void *cb_arg) 3536 { 3537 struct spdk_nvmf_tcp_qpair *tqpair; 3538 3539 SPDK_DEBUGLOG(nvmf_tcp, "Qpair: %p\n", qpair); 3540 3541 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3542 3543 assert(tqpair->fini_cb_fn == NULL); 3544 tqpair->fini_cb_fn = cb_fn; 3545 tqpair->fini_cb_arg = cb_arg; 3546 3547 nvmf_tcp_qpair_set_state(tqpair, NVMF_TCP_QPAIR_STATE_EXITED); 3548 nvmf_tcp_qpair_destroy(tqpair); 3549 } 3550 3551 static int 3552 nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group) 3553 { 3554 struct spdk_nvmf_tcp_poll_group *tgroup; 3555 int num_events; 3556 3557 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3558 3559 if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs))) { 3560 return 0; 3561 } 3562 3563 num_events = spdk_sock_group_poll(tgroup->sock_group); 3564 if (spdk_unlikely(num_events < 0)) { 3565 SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group); 3566 } 3567 3568 return num_events; 3569 } 3570 3571 static int 3572 nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair, 3573 struct spdk_nvme_transport_id *trid, bool peer) 3574 { 3575 struct spdk_nvmf_tcp_qpair *tqpair; 3576 uint16_t port; 3577 3578 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3579 spdk_nvme_trid_populate_transport(trid, SPDK_NVME_TRANSPORT_TCP); 3580 3581 if (peer) { 3582 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr); 3583 port = tqpair->initiator_port; 3584 } else { 3585 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr); 3586 port = tqpair->target_port; 3587 } 3588 3589 if (spdk_sock_is_ipv4(tqpair->sock)) { 3590 trid->adrfam = SPDK_NVMF_ADRFAM_IPV4; 3591 } else if (spdk_sock_is_ipv6(tqpair->sock)) { 3592 trid->adrfam = SPDK_NVMF_ADRFAM_IPV6; 3593 } else { 3594 return -1; 3595 } 3596 3597 snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port); 3598 return 0; 3599 } 3600 3601 static int 3602 nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair, 3603 struct spdk_nvme_transport_id *trid) 3604 { 3605 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 3606 } 3607 3608 static int 3609 nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair, 3610 struct spdk_nvme_transport_id *trid) 3611 { 3612 return nvmf_tcp_qpair_get_trid(qpair, trid, 1); 3613 } 3614 3615 static int 3616 nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair, 3617 struct spdk_nvme_transport_id *trid) 3618 { 3619 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 3620 } 3621 3622 static void 3623 nvmf_tcp_req_set_abort_status(struct spdk_nvmf_request *req, 3624 struct spdk_nvmf_tcp_req *tcp_req_to_abort) 3625 { 3626 nvmf_tcp_req_set_cpl(tcp_req_to_abort, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_ABORTED_BY_REQUEST); 3627 nvmf_tcp_req_set_state(tcp_req_to_abort, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3628 3629 req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command was successfully aborted. */ 3630 } 3631 3632 static int 3633 _nvmf_tcp_qpair_abort_request(void *ctx) 3634 { 3635 struct spdk_nvmf_request *req = ctx; 3636 struct spdk_nvmf_tcp_req *tcp_req_to_abort = SPDK_CONTAINEROF(req->req_to_abort, 3637 struct spdk_nvmf_tcp_req, req); 3638 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(req->req_to_abort->qpair, 3639 struct spdk_nvmf_tcp_qpair, qpair); 3640 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 3641 struct spdk_nvmf_tcp_transport, transport); 3642 int rc; 3643 3644 spdk_poller_unregister(&req->poller); 3645 3646 switch (tcp_req_to_abort->state) { 3647 case TCP_REQUEST_STATE_EXECUTING: 3648 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3649 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3650 rc = nvmf_ctrlr_abort_request(req); 3651 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS) { 3652 return SPDK_POLLER_BUSY; 3653 } 3654 break; 3655 3656 case TCP_REQUEST_STATE_NEED_BUFFER: 3657 nvmf_tcp_request_get_buffers_abort(tcp_req_to_abort); 3658 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 3659 nvmf_tcp_req_process(ttransport, tcp_req_to_abort); 3660 break; 3661 3662 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 3663 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 3664 if (spdk_get_ticks() < req->timeout_tsc) { 3665 req->poller = SPDK_POLLER_REGISTER(_nvmf_tcp_qpair_abort_request, req, 0); 3666 return SPDK_POLLER_BUSY; 3667 } 3668 break; 3669 3670 default: 3671 /* Requests in other states are either un-abortable (e.g. 3672 * TRANSFERRING_CONTROLLER_TO_HOST) or should never end up here, as they're 3673 * immediately transitioned to other states in nvmf_tcp_req_process() (e.g. 3674 * READY_TO_EXECUTE). But it is fine to end up here, as we'll simply complete the 3675 * abort request with the bit0 of dword0 set (command not aborted). 3676 */ 3677 break; 3678 } 3679 3680 spdk_nvmf_request_complete(req); 3681 return SPDK_POLLER_BUSY; 3682 } 3683 3684 static void 3685 nvmf_tcp_qpair_abort_request(struct spdk_nvmf_qpair *qpair, 3686 struct spdk_nvmf_request *req) 3687 { 3688 struct spdk_nvmf_tcp_qpair *tqpair; 3689 struct spdk_nvmf_tcp_transport *ttransport; 3690 struct spdk_nvmf_transport *transport; 3691 uint16_t cid; 3692 uint32_t i; 3693 struct spdk_nvmf_tcp_req *tcp_req_to_abort = NULL; 3694 3695 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3696 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 3697 transport = &ttransport->transport; 3698 3699 cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid; 3700 3701 for (i = 0; i < tqpair->resource_count; i++) { 3702 if (tqpair->reqs[i].state != TCP_REQUEST_STATE_FREE && 3703 tqpair->reqs[i].req.cmd->nvme_cmd.cid == cid) { 3704 tcp_req_to_abort = &tqpair->reqs[i]; 3705 break; 3706 } 3707 } 3708 3709 spdk_trace_record(TRACE_TCP_QP_ABORT_REQ, tqpair->qpair.trace_id, 0, (uintptr_t)req); 3710 3711 if (tcp_req_to_abort == NULL) { 3712 spdk_nvmf_request_complete(req); 3713 return; 3714 } 3715 3716 req->req_to_abort = &tcp_req_to_abort->req; 3717 req->timeout_tsc = spdk_get_ticks() + 3718 transport->opts.abort_timeout_sec * spdk_get_ticks_hz(); 3719 req->poller = NULL; 3720 3721 _nvmf_tcp_qpair_abort_request(req); 3722 } 3723 3724 struct tcp_subsystem_add_host_opts { 3725 char *psk; 3726 }; 3727 3728 static const struct spdk_json_object_decoder tcp_subsystem_add_host_opts_decoder[] = { 3729 {"psk", offsetof(struct tcp_subsystem_add_host_opts, psk), spdk_json_decode_string, true}, 3730 }; 3731 3732 static int 3733 nvmf_tcp_subsystem_add_host(struct spdk_nvmf_transport *transport, 3734 const struct spdk_nvmf_subsystem *subsystem, 3735 const char *hostnqn, 3736 const struct spdk_json_val *transport_specific) 3737 { 3738 struct tcp_subsystem_add_host_opts opts; 3739 struct spdk_nvmf_tcp_transport *ttransport; 3740 struct tcp_psk_entry *tmp, *entry = NULL; 3741 uint8_t psk_configured[SPDK_TLS_PSK_MAX_LEN] = {}; 3742 char psk_interchange[SPDK_TLS_PSK_MAX_LEN + 1] = {}; 3743 uint8_t tls_cipher_suite; 3744 int rc = 0; 3745 uint8_t psk_retained_hash; 3746 uint64_t psk_configured_size; 3747 3748 if (transport_specific == NULL) { 3749 return 0; 3750 } 3751 3752 assert(transport != NULL); 3753 assert(subsystem != NULL); 3754 3755 memset(&opts, 0, sizeof(opts)); 3756 3757 /* Decode PSK (either name of a key or file path) */ 3758 if (spdk_json_decode_object_relaxed(transport_specific, tcp_subsystem_add_host_opts_decoder, 3759 SPDK_COUNTOF(tcp_subsystem_add_host_opts_decoder), &opts)) { 3760 SPDK_ERRLOG("spdk_json_decode_object failed\n"); 3761 return -EINVAL; 3762 } 3763 3764 if (opts.psk == NULL) { 3765 return 0; 3766 } 3767 3768 entry = calloc(1, sizeof(struct tcp_psk_entry)); 3769 if (entry == NULL) { 3770 SPDK_ERRLOG("Unable to allocate memory for PSK entry!\n"); 3771 rc = -ENOMEM; 3772 goto end; 3773 } 3774 3775 entry->key = spdk_keyring_get_key(opts.psk); 3776 if (entry->key == NULL) { 3777 SPDK_ERRLOG("Key '%s' does not exist\n", opts.psk); 3778 rc = -EINVAL; 3779 goto end; 3780 } 3781 3782 rc = spdk_key_get_key(entry->key, psk_interchange, SPDK_TLS_PSK_MAX_LEN); 3783 if (rc < 0) { 3784 SPDK_ERRLOG("Failed to retrieve PSK '%s'\n", opts.psk); 3785 rc = -EINVAL; 3786 goto end; 3787 } 3788 3789 /* Parse PSK interchange to get length of base64 encoded data. 3790 * This is then used to decide which cipher suite should be used 3791 * to generate PSK identity and TLS PSK later on. */ 3792 rc = nvme_tcp_parse_interchange_psk(psk_interchange, psk_configured, sizeof(psk_configured), 3793 &psk_configured_size, &psk_retained_hash); 3794 if (rc < 0) { 3795 SPDK_ERRLOG("Failed to parse PSK interchange!\n"); 3796 goto end; 3797 } 3798 3799 /* The Base64 string encodes the configured PSK (32 or 48 bytes binary). 3800 * This check also ensures that psk_configured_size is smaller than 3801 * psk_retained buffer size. */ 3802 if (psk_configured_size == SHA256_DIGEST_LENGTH) { 3803 tls_cipher_suite = NVME_TCP_CIPHER_AES_128_GCM_SHA256; 3804 } else if (psk_configured_size == SHA384_DIGEST_LENGTH) { 3805 tls_cipher_suite = NVME_TCP_CIPHER_AES_256_GCM_SHA384; 3806 } else { 3807 SPDK_ERRLOG("Unrecognized cipher suite!\n"); 3808 rc = -EINVAL; 3809 goto end; 3810 } 3811 3812 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 3813 /* Generate PSK identity. */ 3814 rc = nvme_tcp_generate_psk_identity(entry->pskid, sizeof(entry->pskid), hostnqn, 3815 subsystem->subnqn, tls_cipher_suite); 3816 if (rc) { 3817 rc = -EINVAL; 3818 goto end; 3819 } 3820 /* Check if PSK identity entry already exists. */ 3821 TAILQ_FOREACH(tmp, &ttransport->psks, link) { 3822 if (strncmp(tmp->pskid, entry->pskid, NVMF_PSK_IDENTITY_LEN) == 0) { 3823 SPDK_ERRLOG("Given PSK identity: %s entry already exists!\n", entry->pskid); 3824 rc = -EEXIST; 3825 goto end; 3826 } 3827 } 3828 3829 if (snprintf(entry->hostnqn, sizeof(entry->hostnqn), "%s", hostnqn) < 0) { 3830 SPDK_ERRLOG("Could not write hostnqn string!\n"); 3831 rc = -EINVAL; 3832 goto end; 3833 } 3834 if (snprintf(entry->subnqn, sizeof(entry->subnqn), "%s", subsystem->subnqn) < 0) { 3835 SPDK_ERRLOG("Could not write subnqn string!\n"); 3836 rc = -EINVAL; 3837 goto end; 3838 } 3839 3840 entry->tls_cipher_suite = tls_cipher_suite; 3841 3842 /* No hash indicates that Configured PSK must be used as Retained PSK. */ 3843 if (psk_retained_hash == NVME_TCP_HASH_ALGORITHM_NONE) { 3844 /* Psk configured is either 32 or 48 bytes long. */ 3845 memcpy(entry->psk, psk_configured, psk_configured_size); 3846 entry->psk_size = psk_configured_size; 3847 } else { 3848 /* Derive retained PSK. */ 3849 rc = nvme_tcp_derive_retained_psk(psk_configured, psk_configured_size, hostnqn, entry->psk, 3850 SPDK_TLS_PSK_MAX_LEN, psk_retained_hash); 3851 if (rc < 0) { 3852 SPDK_ERRLOG("Unable to derive retained PSK!\n"); 3853 goto end; 3854 } 3855 entry->psk_size = rc; 3856 } 3857 3858 TAILQ_INSERT_TAIL(&ttransport->psks, entry, link); 3859 rc = 0; 3860 3861 end: 3862 spdk_memset_s(psk_configured, sizeof(psk_configured), 0, sizeof(psk_configured)); 3863 spdk_memset_s(psk_interchange, sizeof(psk_interchange), 0, sizeof(psk_interchange)); 3864 3865 free(opts.psk); 3866 if (rc != 0) { 3867 nvmf_tcp_free_psk_entry(entry); 3868 } 3869 3870 return rc; 3871 } 3872 3873 static void 3874 nvmf_tcp_subsystem_remove_host(struct spdk_nvmf_transport *transport, 3875 const struct spdk_nvmf_subsystem *subsystem, 3876 const char *hostnqn) 3877 { 3878 struct spdk_nvmf_tcp_transport *ttransport; 3879 struct tcp_psk_entry *entry, *tmp; 3880 3881 assert(transport != NULL); 3882 assert(subsystem != NULL); 3883 3884 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 3885 TAILQ_FOREACH_SAFE(entry, &ttransport->psks, link, tmp) { 3886 if ((strncmp(entry->hostnqn, hostnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0 && 3887 (strncmp(entry->subnqn, subsystem->subnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0) { 3888 TAILQ_REMOVE(&ttransport->psks, entry, link); 3889 nvmf_tcp_free_psk_entry(entry); 3890 break; 3891 } 3892 } 3893 } 3894 3895 static void 3896 nvmf_tcp_subsystem_dump_host(struct spdk_nvmf_transport *transport, 3897 const struct spdk_nvmf_subsystem *subsystem, const char *hostnqn, 3898 struct spdk_json_write_ctx *w) 3899 { 3900 struct spdk_nvmf_tcp_transport *ttransport; 3901 struct tcp_psk_entry *entry; 3902 3903 assert(transport != NULL); 3904 assert(subsystem != NULL); 3905 3906 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 3907 TAILQ_FOREACH(entry, &ttransport->psks, link) { 3908 if ((strncmp(entry->hostnqn, hostnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0 && 3909 (strncmp(entry->subnqn, subsystem->subnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0) { 3910 spdk_json_write_named_string(w, "psk", spdk_key_get_name(entry->key)); 3911 break; 3912 } 3913 } 3914 } 3915 3916 static void 3917 nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts) 3918 { 3919 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH; 3920 opts->max_qpairs_per_ctrlr = SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR; 3921 opts->in_capsule_data_size = SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE; 3922 opts->max_io_size = SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE; 3923 opts->io_unit_size = SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE; 3924 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH; 3925 opts->num_shared_buffers = SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS; 3926 opts->buf_cache_size = SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE; 3927 opts->dif_insert_or_strip = SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP; 3928 opts->abort_timeout_sec = SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC; 3929 opts->transport_specific = NULL; 3930 } 3931 3932 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = { 3933 .name = "TCP", 3934 .type = SPDK_NVME_TRANSPORT_TCP, 3935 .opts_init = nvmf_tcp_opts_init, 3936 .create = nvmf_tcp_create, 3937 .dump_opts = nvmf_tcp_dump_opts, 3938 .destroy = nvmf_tcp_destroy, 3939 3940 .listen = nvmf_tcp_listen, 3941 .stop_listen = nvmf_tcp_stop_listen, 3942 3943 .listener_discover = nvmf_tcp_discover, 3944 3945 .poll_group_create = nvmf_tcp_poll_group_create, 3946 .get_optimal_poll_group = nvmf_tcp_get_optimal_poll_group, 3947 .poll_group_destroy = nvmf_tcp_poll_group_destroy, 3948 .poll_group_add = nvmf_tcp_poll_group_add, 3949 .poll_group_remove = nvmf_tcp_poll_group_remove, 3950 .poll_group_poll = nvmf_tcp_poll_group_poll, 3951 3952 .req_free = nvmf_tcp_req_free, 3953 .req_complete = nvmf_tcp_req_complete, 3954 .req_get_buffers_done = nvmf_tcp_req_get_buffers_done, 3955 3956 .qpair_fini = nvmf_tcp_close_qpair, 3957 .qpair_get_local_trid = nvmf_tcp_qpair_get_local_trid, 3958 .qpair_get_peer_trid = nvmf_tcp_qpair_get_peer_trid, 3959 .qpair_get_listen_trid = nvmf_tcp_qpair_get_listen_trid, 3960 .qpair_abort_request = nvmf_tcp_qpair_abort_request, 3961 .subsystem_add_host = nvmf_tcp_subsystem_add_host, 3962 .subsystem_remove_host = nvmf_tcp_subsystem_remove_host, 3963 .subsystem_dump_host = nvmf_tcp_subsystem_dump_host, 3964 }; 3965 3966 SPDK_NVMF_TRANSPORT_REGISTER(tcp, &spdk_nvmf_transport_tcp); 3967 SPDK_LOG_REGISTER_COMPONENT(nvmf_tcp) 3968