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 #define TCP_PSK_INVALID_PERMISSIONS 0177 52 53 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp; 54 static bool g_tls_log = false; 55 56 /* spdk nvmf related structure */ 57 enum spdk_nvmf_tcp_req_state { 58 59 /* The request is not currently in use */ 60 TCP_REQUEST_STATE_FREE = 0, 61 62 /* Initial state when request first received */ 63 TCP_REQUEST_STATE_NEW = 1, 64 65 /* The request is queued until a data buffer is available. */ 66 TCP_REQUEST_STATE_NEED_BUFFER = 2, 67 68 /* The request has the data buffer available */ 69 TCP_REQUEST_STATE_HAVE_BUFFER = 3, 70 71 /* The request is waiting for zcopy_start to finish */ 72 TCP_REQUEST_STATE_AWAITING_ZCOPY_START = 4, 73 74 /* The request has received a zero-copy buffer */ 75 TCP_REQUEST_STATE_ZCOPY_START_COMPLETED = 5, 76 77 /* The request is currently transferring data from the host to the controller. */ 78 TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER = 6, 79 80 /* The request is waiting for the R2T send acknowledgement. */ 81 TCP_REQUEST_STATE_AWAITING_R2T_ACK = 7, 82 83 /* The request is ready to execute at the block device */ 84 TCP_REQUEST_STATE_READY_TO_EXECUTE = 8, 85 86 /* The request is currently executing at the block device */ 87 TCP_REQUEST_STATE_EXECUTING = 9, 88 89 /* The request is waiting for zcopy buffers to be committed */ 90 TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT = 10, 91 92 /* The request finished executing at the block device */ 93 TCP_REQUEST_STATE_EXECUTED = 11, 94 95 /* The request is ready to send a completion */ 96 TCP_REQUEST_STATE_READY_TO_COMPLETE = 12, 97 98 /* The request is currently transferring final pdus from the controller to the host. */ 99 TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST = 13, 100 101 /* The request is waiting for zcopy buffers to be released (without committing) */ 102 TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE = 14, 103 104 /* The request completed and can be marked free. */ 105 TCP_REQUEST_STATE_COMPLETED = 15, 106 107 /* Terminator */ 108 TCP_REQUEST_NUM_STATES, 109 }; 110 111 static const char *spdk_nvmf_tcp_term_req_fes_str[] = { 112 "Invalid PDU Header Field", 113 "PDU Sequence Error", 114 "Header Digiest Error", 115 "Data Transfer Out of Range", 116 "R2T Limit Exceeded", 117 "Unsupported parameter", 118 }; 119 120 SPDK_TRACE_REGISTER_FN(nvmf_tcp_trace, "nvmf_tcp", TRACE_GROUP_NVMF_TCP) 121 { 122 spdk_trace_register_owner_type(OWNER_TYPE_NVMF_TCP, 't'); 123 spdk_trace_register_object(OBJECT_NVMF_TCP_IO, 'r'); 124 spdk_trace_register_description("TCP_REQ_NEW", 125 TRACE_TCP_REQUEST_STATE_NEW, 126 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 1, 127 SPDK_TRACE_ARG_TYPE_INT, "qd"); 128 spdk_trace_register_description("TCP_REQ_NEED_BUFFER", 129 TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 130 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 131 SPDK_TRACE_ARG_TYPE_INT, ""); 132 spdk_trace_register_description("TCP_REQ_HAVE_BUFFER", 133 TRACE_TCP_REQUEST_STATE_HAVE_BUFFER, 134 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 135 SPDK_TRACE_ARG_TYPE_INT, ""); 136 spdk_trace_register_description("TCP_REQ_WAIT_ZCPY_START", 137 TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_START, 138 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 139 SPDK_TRACE_ARG_TYPE_INT, ""); 140 spdk_trace_register_description("TCP_REQ_ZCPY_START_CPL", 141 TRACE_TCP_REQUEST_STATE_ZCOPY_START_COMPLETED, 142 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 143 SPDK_TRACE_ARG_TYPE_INT, ""); 144 spdk_trace_register_description("TCP_REQ_TX_H_TO_C", 145 TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 146 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 147 SPDK_TRACE_ARG_TYPE_INT, ""); 148 spdk_trace_register_description("TCP_REQ_RDY_TO_EXECUTE", 149 TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 150 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 151 SPDK_TRACE_ARG_TYPE_INT, ""); 152 spdk_trace_register_description("TCP_REQ_EXECUTING", 153 TRACE_TCP_REQUEST_STATE_EXECUTING, 154 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 155 SPDK_TRACE_ARG_TYPE_INT, ""); 156 spdk_trace_register_description("TCP_REQ_WAIT_ZCPY_CMT", 157 TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_COMMIT, 158 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 159 SPDK_TRACE_ARG_TYPE_INT, ""); 160 spdk_trace_register_description("TCP_REQ_EXECUTED", 161 TRACE_TCP_REQUEST_STATE_EXECUTED, 162 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 163 SPDK_TRACE_ARG_TYPE_INT, ""); 164 spdk_trace_register_description("TCP_REQ_RDY_TO_COMPLETE", 165 TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 166 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 167 SPDK_TRACE_ARG_TYPE_INT, ""); 168 spdk_trace_register_description("TCP_REQ_TRANSFER_C2H", 169 TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 170 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 171 SPDK_TRACE_ARG_TYPE_INT, ""); 172 spdk_trace_register_description("TCP_REQ_AWAIT_ZCPY_RLS", 173 TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_RELEASE, 174 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 175 SPDK_TRACE_ARG_TYPE_INT, ""); 176 spdk_trace_register_description("TCP_REQ_COMPLETED", 177 TRACE_TCP_REQUEST_STATE_COMPLETED, 178 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 179 SPDK_TRACE_ARG_TYPE_INT, "qd"); 180 spdk_trace_register_description("TCP_READ_DONE", 181 TRACE_TCP_READ_FROM_SOCKET_DONE, 182 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 183 SPDK_TRACE_ARG_TYPE_INT, ""); 184 spdk_trace_register_description("TCP_REQ_AWAIT_R2T_ACK", 185 TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 186 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 187 SPDK_TRACE_ARG_TYPE_INT, ""); 188 189 spdk_trace_register_description("TCP_QP_CREATE", TRACE_TCP_QP_CREATE, 190 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 191 SPDK_TRACE_ARG_TYPE_INT, ""); 192 spdk_trace_register_description("TCP_QP_SOCK_INIT", TRACE_TCP_QP_SOCK_INIT, 193 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 194 SPDK_TRACE_ARG_TYPE_INT, ""); 195 spdk_trace_register_description("TCP_QP_STATE_CHANGE", TRACE_TCP_QP_STATE_CHANGE, 196 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 197 SPDK_TRACE_ARG_TYPE_INT, "state"); 198 spdk_trace_register_description("TCP_QP_DISCONNECT", TRACE_TCP_QP_DISCONNECT, 199 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 200 SPDK_TRACE_ARG_TYPE_INT, ""); 201 spdk_trace_register_description("TCP_QP_DESTROY", TRACE_TCP_QP_DESTROY, 202 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 203 SPDK_TRACE_ARG_TYPE_INT, ""); 204 spdk_trace_register_description("TCP_QP_ABORT_REQ", TRACE_TCP_QP_ABORT_REQ, 205 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 206 SPDK_TRACE_ARG_TYPE_INT, ""); 207 spdk_trace_register_description("TCP_QP_RCV_STATE_CHANGE", TRACE_TCP_QP_RCV_STATE_CHANGE, 208 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 209 SPDK_TRACE_ARG_TYPE_INT, "state"); 210 211 spdk_trace_tpoint_register_relation(TRACE_BDEV_IO_START, OBJECT_NVMF_TCP_IO, 1); 212 spdk_trace_tpoint_register_relation(TRACE_BDEV_IO_DONE, OBJECT_NVMF_TCP_IO, 0); 213 spdk_trace_tpoint_register_relation(TRACE_SOCK_REQ_QUEUE, OBJECT_NVMF_TCP_IO, 0); 214 spdk_trace_tpoint_register_relation(TRACE_SOCK_REQ_PEND, OBJECT_NVMF_TCP_IO, 0); 215 spdk_trace_tpoint_register_relation(TRACE_SOCK_REQ_COMPLETE, OBJECT_NVMF_TCP_IO, 0); 216 } 217 218 struct spdk_nvmf_tcp_req { 219 struct spdk_nvmf_request req; 220 struct spdk_nvme_cpl rsp; 221 struct spdk_nvme_cmd cmd; 222 223 /* A PDU that can be used for sending responses. This is 224 * not the incoming PDU! */ 225 struct nvme_tcp_pdu *pdu; 226 227 /* In-capsule data buffer */ 228 uint8_t *buf; 229 230 struct spdk_nvmf_tcp_req *fused_pair; 231 232 /* 233 * The PDU for a request may be used multiple times in serial over 234 * the request's lifetime. For example, first to send an R2T, then 235 * to send a completion. To catch mistakes where the PDU is used 236 * twice at the same time, add a debug flag here for init/fini. 237 */ 238 bool pdu_in_use; 239 bool has_in_capsule_data; 240 bool fused_failed; 241 242 /* transfer_tag */ 243 uint16_t ttag; 244 245 enum spdk_nvmf_tcp_req_state state; 246 247 /* 248 * h2c_offset is used when we receive the h2c_data PDU. 249 */ 250 uint32_t h2c_offset; 251 252 STAILQ_ENTRY(spdk_nvmf_tcp_req) link; 253 TAILQ_ENTRY(spdk_nvmf_tcp_req) state_link; 254 STAILQ_ENTRY(spdk_nvmf_tcp_req) control_msg_link; 255 }; 256 257 struct spdk_nvmf_tcp_qpair { 258 struct spdk_nvmf_qpair qpair; 259 struct spdk_nvmf_tcp_poll_group *group; 260 struct spdk_sock *sock; 261 262 enum nvme_tcp_pdu_recv_state recv_state; 263 enum nvme_tcp_qpair_state state; 264 265 /* PDU being actively received */ 266 struct nvme_tcp_pdu *pdu_in_progress; 267 268 struct spdk_nvmf_tcp_req *fused_first; 269 270 /* Queues to track the requests in all states */ 271 TAILQ_HEAD(, spdk_nvmf_tcp_req) tcp_req_working_queue; 272 TAILQ_HEAD(, spdk_nvmf_tcp_req) tcp_req_free_queue; 273 SLIST_HEAD(, nvme_tcp_pdu) tcp_pdu_free_queue; 274 /* Number of working pdus */ 275 uint32_t tcp_pdu_working_count; 276 277 /* Number of requests in each state */ 278 uint32_t state_cntr[TCP_REQUEST_NUM_STATES]; 279 280 uint8_t cpda; 281 282 bool host_hdgst_enable; 283 bool host_ddgst_enable; 284 285 /* This is a spare PDU used for sending special management 286 * operations. Primarily, this is used for the initial 287 * connection response and c2h termination request. */ 288 struct nvme_tcp_pdu *mgmt_pdu; 289 290 /* Arrays of in-capsule buffers, requests, and pdus. 291 * Each array is 'resource_count' number of elements */ 292 void *bufs; 293 struct spdk_nvmf_tcp_req *reqs; 294 struct nvme_tcp_pdu *pdus; 295 uint32_t resource_count; 296 uint32_t recv_buf_size; 297 298 struct spdk_nvmf_tcp_port *port; 299 300 /* IP address */ 301 char initiator_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 302 char target_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 303 304 /* IP port */ 305 uint16_t initiator_port; 306 uint16_t target_port; 307 308 /* Wait until the host terminates the connection (e.g. after sending C2HTermReq) */ 309 bool wait_terminate; 310 311 /* Timer used to destroy qpair after detecting transport error issue if initiator does 312 * not close the connection. 313 */ 314 struct spdk_poller *timeout_poller; 315 316 spdk_nvmf_transport_qpair_fini_cb fini_cb_fn; 317 void *fini_cb_arg; 318 319 TAILQ_ENTRY(spdk_nvmf_tcp_qpair) link; 320 }; 321 322 struct spdk_nvmf_tcp_control_msg { 323 STAILQ_ENTRY(spdk_nvmf_tcp_control_msg) link; 324 }; 325 326 struct spdk_nvmf_tcp_control_msg_list { 327 void *msg_buf; 328 STAILQ_HEAD(, spdk_nvmf_tcp_control_msg) free_msgs; 329 STAILQ_HEAD(, spdk_nvmf_tcp_req) waiting_for_msg_reqs; 330 }; 331 332 struct spdk_nvmf_tcp_poll_group { 333 struct spdk_nvmf_transport_poll_group group; 334 struct spdk_sock_group *sock_group; 335 336 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) qpairs; 337 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) await_req; 338 339 struct spdk_io_channel *accel_channel; 340 struct spdk_nvmf_tcp_control_msg_list *control_msg_list; 341 342 TAILQ_ENTRY(spdk_nvmf_tcp_poll_group) link; 343 }; 344 345 struct spdk_nvmf_tcp_port { 346 const struct spdk_nvme_transport_id *trid; 347 struct spdk_sock *listen_sock; 348 struct spdk_nvmf_transport *transport; 349 TAILQ_ENTRY(spdk_nvmf_tcp_port) link; 350 }; 351 352 struct tcp_transport_opts { 353 bool c2h_success; 354 uint16_t control_msg_num; 355 uint32_t sock_priority; 356 }; 357 358 struct tcp_psk_entry { 359 char hostnqn[SPDK_NVMF_NQN_MAX_LEN + 1]; 360 char subnqn[SPDK_NVMF_NQN_MAX_LEN + 1]; 361 char pskid[NVMF_PSK_IDENTITY_LEN]; 362 uint8_t psk[SPDK_TLS_PSK_MAX_LEN]; 363 struct spdk_key *key; 364 365 /* Original path saved to emit SPDK configuration via "save_config". */ 366 char psk_path[PATH_MAX]; 367 uint32_t psk_size; 368 enum nvme_tcp_cipher_suite tls_cipher_suite; 369 TAILQ_ENTRY(tcp_psk_entry) link; 370 }; 371 372 struct spdk_nvmf_tcp_transport { 373 struct spdk_nvmf_transport transport; 374 struct tcp_transport_opts tcp_opts; 375 uint32_t ack_timeout; 376 377 struct spdk_nvmf_tcp_poll_group *next_pg; 378 379 struct spdk_poller *accept_poller; 380 struct spdk_sock_group *listen_sock_group; 381 382 TAILQ_HEAD(, spdk_nvmf_tcp_port) ports; 383 TAILQ_HEAD(, spdk_nvmf_tcp_poll_group) poll_groups; 384 385 TAILQ_HEAD(, tcp_psk_entry) psks; 386 }; 387 388 static const struct spdk_json_object_decoder tcp_transport_opts_decoder[] = { 389 { 390 "c2h_success", offsetof(struct tcp_transport_opts, c2h_success), 391 spdk_json_decode_bool, true 392 }, 393 { 394 "control_msg_num", offsetof(struct tcp_transport_opts, control_msg_num), 395 spdk_json_decode_uint16, true 396 }, 397 { 398 "sock_priority", offsetof(struct tcp_transport_opts, sock_priority), 399 spdk_json_decode_uint32, true 400 }, 401 }; 402 403 static bool nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 404 struct spdk_nvmf_tcp_req *tcp_req); 405 static void nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group); 406 407 static void _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 408 struct spdk_nvmf_tcp_req *tcp_req); 409 410 static inline void 411 nvmf_tcp_req_set_state(struct spdk_nvmf_tcp_req *tcp_req, 412 enum spdk_nvmf_tcp_req_state state) 413 { 414 struct spdk_nvmf_qpair *qpair; 415 struct spdk_nvmf_tcp_qpair *tqpair; 416 417 qpair = tcp_req->req.qpair; 418 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 419 420 assert(tqpair->state_cntr[tcp_req->state] > 0); 421 tqpair->state_cntr[tcp_req->state]--; 422 tqpair->state_cntr[state]++; 423 424 tcp_req->state = state; 425 } 426 427 static inline struct nvme_tcp_pdu * 428 nvmf_tcp_req_pdu_init(struct spdk_nvmf_tcp_req *tcp_req) 429 { 430 assert(tcp_req->pdu_in_use == false); 431 432 memset(tcp_req->pdu, 0, sizeof(*tcp_req->pdu)); 433 tcp_req->pdu->qpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 434 435 return tcp_req->pdu; 436 } 437 438 static struct spdk_nvmf_tcp_req * 439 nvmf_tcp_req_get(struct spdk_nvmf_tcp_qpair *tqpair) 440 { 441 struct spdk_nvmf_tcp_req *tcp_req; 442 443 tcp_req = TAILQ_FIRST(&tqpair->tcp_req_free_queue); 444 if (spdk_unlikely(!tcp_req)) { 445 return NULL; 446 } 447 448 memset(&tcp_req->rsp, 0, sizeof(tcp_req->rsp)); 449 tcp_req->h2c_offset = 0; 450 tcp_req->has_in_capsule_data = false; 451 tcp_req->req.dif_enabled = false; 452 tcp_req->req.zcopy_phase = NVMF_ZCOPY_PHASE_NONE; 453 454 TAILQ_REMOVE(&tqpair->tcp_req_free_queue, tcp_req, state_link); 455 TAILQ_INSERT_TAIL(&tqpair->tcp_req_working_queue, tcp_req, state_link); 456 tqpair->qpair.queue_depth++; 457 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW); 458 return tcp_req; 459 } 460 461 static inline void 462 nvmf_tcp_req_put(struct spdk_nvmf_tcp_qpair *tqpair, struct spdk_nvmf_tcp_req *tcp_req) 463 { 464 assert(!tcp_req->pdu_in_use); 465 466 TAILQ_REMOVE(&tqpair->tcp_req_working_queue, tcp_req, state_link); 467 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 468 tqpair->qpair.queue_depth--; 469 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_FREE); 470 } 471 472 static void 473 nvmf_tcp_req_get_buffers_done(struct spdk_nvmf_request *req) 474 { 475 struct spdk_nvmf_tcp_req *tcp_req; 476 struct spdk_nvmf_transport *transport; 477 struct spdk_nvmf_tcp_transport *ttransport; 478 479 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 480 transport = req->qpair->transport; 481 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 482 483 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_HAVE_BUFFER); 484 nvmf_tcp_req_process(ttransport, tcp_req); 485 } 486 487 static void 488 nvmf_tcp_request_free(void *cb_arg) 489 { 490 struct spdk_nvmf_tcp_transport *ttransport; 491 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 492 493 assert(tcp_req != NULL); 494 495 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req=%p will be freed\n", tcp_req); 496 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 497 struct spdk_nvmf_tcp_transport, transport); 498 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 499 nvmf_tcp_req_process(ttransport, tcp_req); 500 } 501 502 static int 503 nvmf_tcp_req_free(struct spdk_nvmf_request *req) 504 { 505 struct spdk_nvmf_tcp_req *tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 506 507 nvmf_tcp_request_free(tcp_req); 508 509 return 0; 510 } 511 512 static void 513 nvmf_tcp_drain_state_queue(struct spdk_nvmf_tcp_qpair *tqpair, 514 enum spdk_nvmf_tcp_req_state state) 515 { 516 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 517 518 assert(state != TCP_REQUEST_STATE_FREE); 519 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 520 if (state == tcp_req->state) { 521 nvmf_tcp_request_free(tcp_req); 522 } 523 } 524 } 525 526 static inline void 527 nvmf_tcp_request_get_buffers_abort(struct spdk_nvmf_tcp_req *tcp_req) 528 { 529 /* Request can wait either for the iobuf or control_msg */ 530 struct spdk_nvmf_poll_group *group = tcp_req->req.qpair->group; 531 struct spdk_nvmf_transport *transport = tcp_req->req.qpair->transport; 532 struct spdk_nvmf_transport_poll_group *tgroup = nvmf_get_transport_poll_group(group, transport); 533 struct spdk_nvmf_tcp_poll_group *tcp_group = SPDK_CONTAINEROF(tgroup, 534 struct spdk_nvmf_tcp_poll_group, group); 535 struct spdk_nvmf_tcp_req *tmp_req, *abort_req; 536 537 assert(tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER); 538 539 STAILQ_FOREACH_SAFE(abort_req, &tcp_group->control_msg_list->waiting_for_msg_reqs, control_msg_link, 540 tmp_req) { 541 if (abort_req == tcp_req) { 542 STAILQ_REMOVE(&tcp_group->control_msg_list->waiting_for_msg_reqs, abort_req, spdk_nvmf_tcp_req, 543 control_msg_link); 544 return; 545 } 546 } 547 548 if (!nvmf_request_get_buffers_abort(&tcp_req->req)) { 549 SPDK_ERRLOG("Failed to abort tcp_req=%p\n", tcp_req); 550 assert(0 && "Should never happen"); 551 } 552 } 553 554 static void 555 nvmf_tcp_cleanup_all_states(struct spdk_nvmf_tcp_qpair *tqpair) 556 { 557 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 558 559 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 560 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEW); 561 562 /* Wipe the requests waiting for buffer from the waiting list */ 563 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 564 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 565 nvmf_tcp_request_get_buffers_abort(tcp_req); 566 } 567 } 568 569 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEED_BUFFER); 570 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_EXECUTING); 571 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 572 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 573 } 574 575 static void 576 nvmf_tcp_dump_qpair_req_contents(struct spdk_nvmf_tcp_qpair *tqpair) 577 { 578 int i; 579 struct spdk_nvmf_tcp_req *tcp_req; 580 581 SPDK_ERRLOG("Dumping contents of queue pair (QID %d)\n", tqpair->qpair.qid); 582 for (i = 1; i < TCP_REQUEST_NUM_STATES; i++) { 583 SPDK_ERRLOG("\tNum of requests in state[%d] = %u\n", i, tqpair->state_cntr[i]); 584 TAILQ_FOREACH(tcp_req, &tqpair->tcp_req_working_queue, state_link) { 585 if ((int)tcp_req->state == i) { 586 SPDK_ERRLOG("\t\tRequest Data From Pool: %d\n", tcp_req->req.data_from_pool); 587 SPDK_ERRLOG("\t\tRequest opcode: %d\n", tcp_req->req.cmd->nvmf_cmd.opcode); 588 } 589 } 590 } 591 } 592 593 static void 594 _nvmf_tcp_qpair_destroy(void *_tqpair) 595 { 596 struct spdk_nvmf_tcp_qpair *tqpair = _tqpair; 597 spdk_nvmf_transport_qpair_fini_cb cb_fn = tqpair->fini_cb_fn; 598 void *cb_arg = tqpair->fini_cb_arg; 599 int err = 0; 600 601 spdk_trace_record(TRACE_TCP_QP_DESTROY, tqpair->qpair.trace_id, 0, 0); 602 603 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 604 605 err = spdk_sock_close(&tqpair->sock); 606 assert(err == 0); 607 nvmf_tcp_cleanup_all_states(tqpair); 608 609 if (tqpair->state_cntr[TCP_REQUEST_STATE_FREE] != tqpair->resource_count) { 610 SPDK_ERRLOG("tqpair(%p) free tcp request num is %u but should be %u\n", tqpair, 611 tqpair->state_cntr[TCP_REQUEST_STATE_FREE], 612 tqpair->resource_count); 613 err++; 614 } 615 616 if (err > 0) { 617 nvmf_tcp_dump_qpair_req_contents(tqpair); 618 } 619 620 /* The timeout poller might still be registered here if we close the qpair before host 621 * terminates the connection. 622 */ 623 spdk_poller_unregister(&tqpair->timeout_poller); 624 spdk_dma_free(tqpair->pdus); 625 free(tqpair->reqs); 626 spdk_free(tqpair->bufs); 627 spdk_trace_unregister_owner(tqpair->qpair.trace_id); 628 free(tqpair); 629 630 if (cb_fn != NULL) { 631 cb_fn(cb_arg); 632 } 633 634 SPDK_DEBUGLOG(nvmf_tcp, "Leave\n"); 635 } 636 637 static void 638 nvmf_tcp_qpair_destroy(struct spdk_nvmf_tcp_qpair *tqpair) 639 { 640 /* Delay the destruction to make sure it isn't performed from the context of a sock 641 * callback. Otherwise, spdk_sock_close() might not abort pending requests, causing their 642 * completions to be executed after the qpair is freed. (Note: this fixed issue #2471.) 643 */ 644 spdk_thread_send_msg(spdk_get_thread(), _nvmf_tcp_qpair_destroy, tqpair); 645 } 646 647 static void 648 nvmf_tcp_dump_opts(struct spdk_nvmf_transport *transport, struct spdk_json_write_ctx *w) 649 { 650 struct spdk_nvmf_tcp_transport *ttransport; 651 assert(w != NULL); 652 653 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 654 spdk_json_write_named_bool(w, "c2h_success", ttransport->tcp_opts.c2h_success); 655 spdk_json_write_named_uint32(w, "sock_priority", ttransport->tcp_opts.sock_priority); 656 } 657 658 static void 659 nvmf_tcp_free_psk_entry(struct tcp_psk_entry *entry) 660 { 661 if (entry == NULL) { 662 return; 663 } 664 665 spdk_memset_s(entry->psk, sizeof(entry->psk), 0, sizeof(entry->psk)); 666 spdk_keyring_put_key(entry->key); 667 free(entry); 668 } 669 670 static int 671 nvmf_tcp_destroy(struct spdk_nvmf_transport *transport, 672 spdk_nvmf_transport_destroy_done_cb cb_fn, void *cb_arg) 673 { 674 struct spdk_nvmf_tcp_transport *ttransport; 675 struct tcp_psk_entry *entry, *tmp; 676 677 assert(transport != NULL); 678 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 679 680 TAILQ_FOREACH_SAFE(entry, &ttransport->psks, link, tmp) { 681 TAILQ_REMOVE(&ttransport->psks, entry, link); 682 nvmf_tcp_free_psk_entry(entry); 683 } 684 685 spdk_poller_unregister(&ttransport->accept_poller); 686 spdk_sock_group_close(&ttransport->listen_sock_group); 687 free(ttransport); 688 689 if (cb_fn) { 690 cb_fn(cb_arg); 691 } 692 return 0; 693 } 694 695 static int nvmf_tcp_accept(void *ctx); 696 697 static void nvmf_tcp_accept_cb(void *ctx, struct spdk_sock_group *group, struct spdk_sock *sock); 698 699 static struct spdk_nvmf_transport * 700 nvmf_tcp_create(struct spdk_nvmf_transport_opts *opts) 701 { 702 struct spdk_nvmf_tcp_transport *ttransport; 703 uint32_t sge_count; 704 uint32_t min_shared_buffers; 705 706 ttransport = calloc(1, sizeof(*ttransport)); 707 if (!ttransport) { 708 return NULL; 709 } 710 711 TAILQ_INIT(&ttransport->ports); 712 TAILQ_INIT(&ttransport->poll_groups); 713 TAILQ_INIT(&ttransport->psks); 714 715 ttransport->transport.ops = &spdk_nvmf_transport_tcp; 716 717 ttransport->tcp_opts.c2h_success = SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION; 718 ttransport->tcp_opts.sock_priority = SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY; 719 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 720 if (opts->transport_specific != NULL && 721 spdk_json_decode_object_relaxed(opts->transport_specific, tcp_transport_opts_decoder, 722 SPDK_COUNTOF(tcp_transport_opts_decoder), 723 &ttransport->tcp_opts)) { 724 SPDK_ERRLOG("spdk_json_decode_object_relaxed failed\n"); 725 free(ttransport); 726 return NULL; 727 } 728 729 SPDK_NOTICELOG("*** TCP Transport Init ***\n"); 730 731 SPDK_INFOLOG(nvmf_tcp, "*** TCP Transport Init ***\n" 732 " Transport opts: max_ioq_depth=%d, max_io_size=%d,\n" 733 " max_io_qpairs_per_ctrlr=%d, io_unit_size=%d,\n" 734 " in_capsule_data_size=%d, max_aq_depth=%d\n" 735 " num_shared_buffers=%d, c2h_success=%d,\n" 736 " dif_insert_or_strip=%d, sock_priority=%d\n" 737 " abort_timeout_sec=%d, control_msg_num=%hu\n" 738 " ack_timeout=%d\n", 739 opts->max_queue_depth, 740 opts->max_io_size, 741 opts->max_qpairs_per_ctrlr - 1, 742 opts->io_unit_size, 743 opts->in_capsule_data_size, 744 opts->max_aq_depth, 745 opts->num_shared_buffers, 746 ttransport->tcp_opts.c2h_success, 747 opts->dif_insert_or_strip, 748 ttransport->tcp_opts.sock_priority, 749 opts->abort_timeout_sec, 750 ttransport->tcp_opts.control_msg_num, 751 opts->ack_timeout); 752 753 if (ttransport->tcp_opts.sock_priority > SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY) { 754 SPDK_ERRLOG("Unsupported socket_priority=%d, the current range is: 0 to %d\n" 755 "you can use man 7 socket to view the range of priority under SO_PRIORITY item\n", 756 ttransport->tcp_opts.sock_priority, SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY); 757 free(ttransport); 758 return NULL; 759 } 760 761 if (ttransport->tcp_opts.control_msg_num == 0 && 762 opts->in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 763 SPDK_WARNLOG("TCP param control_msg_num can't be 0 if ICD is less than %u bytes. Using default value %u\n", 764 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM); 765 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 766 } 767 768 /* I/O unit size cannot be larger than max I/O size */ 769 if (opts->io_unit_size > opts->max_io_size) { 770 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", 771 opts->io_unit_size, opts->max_io_size); 772 opts->io_unit_size = opts->max_io_size; 773 } 774 775 /* In capsule data size cannot be larger than max I/O size */ 776 if (opts->in_capsule_data_size > opts->max_io_size) { 777 SPDK_WARNLOG("TCP param ICD size %u can't be larger than max_io_size %u. Using max_io_size as ICD size\n", 778 opts->io_unit_size, opts->max_io_size); 779 opts->in_capsule_data_size = opts->max_io_size; 780 } 781 782 /* max IO queue depth cannot be smaller than 2 or larger than 65535. 783 * We will not check SPDK_NVMF_TCP_MAX_IO_QUEUE_DEPTH, because max_queue_depth is 16bits and always not larger than 64k. */ 784 if (opts->max_queue_depth < SPDK_NVMF_TCP_MIN_IO_QUEUE_DEPTH) { 785 SPDK_WARNLOG("TCP param max_queue_depth %u can't be smaller than %u or larger than %u. Using default value %u\n", 786 opts->max_queue_depth, SPDK_NVMF_TCP_MIN_IO_QUEUE_DEPTH, 787 SPDK_NVMF_TCP_MAX_IO_QUEUE_DEPTH, SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH); 788 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH; 789 } 790 791 /* max admin queue depth cannot be smaller than 2 or larger than 4096 */ 792 if (opts->max_aq_depth < SPDK_NVMF_TCP_MIN_ADMIN_QUEUE_DEPTH || 793 opts->max_aq_depth > SPDK_NVMF_TCP_MAX_ADMIN_QUEUE_DEPTH) { 794 SPDK_WARNLOG("TCP param max_aq_depth %u can't be smaller than %u or larger than %u. Using default value %u\n", 795 opts->max_aq_depth, SPDK_NVMF_TCP_MIN_ADMIN_QUEUE_DEPTH, 796 SPDK_NVMF_TCP_MAX_ADMIN_QUEUE_DEPTH, SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH); 797 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH; 798 } 799 800 sge_count = opts->max_io_size / opts->io_unit_size; 801 if (sge_count > SPDK_NVMF_MAX_SGL_ENTRIES) { 802 SPDK_ERRLOG("Unsupported IO Unit size specified, %d bytes\n", opts->io_unit_size); 803 free(ttransport); 804 return NULL; 805 } 806 807 /* If buf_cache_size == UINT32_MAX, we will dynamically pick a cache size later that we know will fit. */ 808 if (opts->buf_cache_size < UINT32_MAX) { 809 min_shared_buffers = spdk_env_get_core_count() * opts->buf_cache_size; 810 if (min_shared_buffers > opts->num_shared_buffers) { 811 SPDK_ERRLOG("There are not enough buffers to satisfy " 812 "per-poll group caches for each thread. (%" PRIu32 ") " 813 "supplied. (%" PRIu32 ") required\n", opts->num_shared_buffers, min_shared_buffers); 814 SPDK_ERRLOG("Please specify a larger number of shared buffers\n"); 815 free(ttransport); 816 return NULL; 817 } 818 } 819 820 ttransport->accept_poller = SPDK_POLLER_REGISTER(nvmf_tcp_accept, &ttransport->transport, 821 opts->acceptor_poll_rate); 822 if (!ttransport->accept_poller) { 823 free(ttransport); 824 return NULL; 825 } 826 827 ttransport->listen_sock_group = spdk_sock_group_create(NULL); 828 if (ttransport->listen_sock_group == NULL) { 829 SPDK_ERRLOG("Failed to create socket group for listen sockets\n"); 830 spdk_poller_unregister(&ttransport->accept_poller); 831 free(ttransport); 832 return NULL; 833 } 834 835 return &ttransport->transport; 836 } 837 838 static int 839 nvmf_tcp_trsvcid_to_int(const char *trsvcid) 840 { 841 unsigned long long ull; 842 char *end = NULL; 843 844 ull = strtoull(trsvcid, &end, 10); 845 if (end == NULL || end == trsvcid || *end != '\0') { 846 return -1; 847 } 848 849 /* Valid TCP/IP port numbers are in [1, 65535] */ 850 if (ull == 0 || ull > 65535) { 851 return -1; 852 } 853 854 return (int)ull; 855 } 856 857 /** 858 * Canonicalize a listen address trid. 859 */ 860 static int 861 nvmf_tcp_canon_listen_trid(struct spdk_nvme_transport_id *canon_trid, 862 const struct spdk_nvme_transport_id *trid) 863 { 864 int trsvcid_int; 865 866 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 867 if (trsvcid_int < 0) { 868 return -EINVAL; 869 } 870 871 memset(canon_trid, 0, sizeof(*canon_trid)); 872 spdk_nvme_trid_populate_transport(canon_trid, SPDK_NVME_TRANSPORT_TCP); 873 canon_trid->adrfam = trid->adrfam; 874 snprintf(canon_trid->traddr, sizeof(canon_trid->traddr), "%s", trid->traddr); 875 snprintf(canon_trid->trsvcid, sizeof(canon_trid->trsvcid), "%d", trsvcid_int); 876 877 return 0; 878 } 879 880 /** 881 * Find an existing listening port. 882 */ 883 static struct spdk_nvmf_tcp_port * 884 nvmf_tcp_find_port(struct spdk_nvmf_tcp_transport *ttransport, 885 const struct spdk_nvme_transport_id *trid) 886 { 887 struct spdk_nvme_transport_id canon_trid; 888 struct spdk_nvmf_tcp_port *port; 889 890 if (nvmf_tcp_canon_listen_trid(&canon_trid, trid) != 0) { 891 return NULL; 892 } 893 894 TAILQ_FOREACH(port, &ttransport->ports, link) { 895 if (spdk_nvme_transport_id_compare(&canon_trid, port->trid) == 0) { 896 return port; 897 } 898 } 899 900 return NULL; 901 } 902 903 static int 904 tcp_sock_get_key(uint8_t *out, int out_len, const char **cipher, const char *pskid, 905 void *get_key_ctx) 906 { 907 struct tcp_psk_entry *entry; 908 struct spdk_nvmf_tcp_transport *ttransport = get_key_ctx; 909 size_t psk_len; 910 int rc; 911 912 TAILQ_FOREACH(entry, &ttransport->psks, link) { 913 if (strcmp(pskid, entry->pskid) != 0) { 914 continue; 915 } 916 917 psk_len = entry->psk_size; 918 if ((size_t)out_len < psk_len) { 919 SPDK_ERRLOG("Out buffer of size: %" PRIu32 " cannot fit PSK of len: %lu\n", 920 out_len, psk_len); 921 return -ENOBUFS; 922 } 923 924 /* Convert PSK to the TLS PSK format. */ 925 rc = nvme_tcp_derive_tls_psk(entry->psk, psk_len, pskid, out, out_len, 926 entry->tls_cipher_suite); 927 if (rc < 0) { 928 SPDK_ERRLOG("Could not generate TLS PSK\n"); 929 } 930 931 switch (entry->tls_cipher_suite) { 932 case NVME_TCP_CIPHER_AES_128_GCM_SHA256: 933 *cipher = "TLS_AES_128_GCM_SHA256"; 934 break; 935 case NVME_TCP_CIPHER_AES_256_GCM_SHA384: 936 *cipher = "TLS_AES_256_GCM_SHA384"; 937 break; 938 default: 939 *cipher = NULL; 940 return -ENOTSUP; 941 } 942 943 return rc; 944 } 945 946 SPDK_ERRLOG("Could not find PSK for identity: %s\n", pskid); 947 948 return -EINVAL; 949 } 950 951 static int 952 nvmf_tcp_listen(struct spdk_nvmf_transport *transport, const struct spdk_nvme_transport_id *trid, 953 struct spdk_nvmf_listen_opts *listen_opts) 954 { 955 struct spdk_nvmf_tcp_transport *ttransport; 956 struct spdk_nvmf_tcp_port *port; 957 int trsvcid_int; 958 uint8_t adrfam; 959 const char *sock_impl_name; 960 struct spdk_sock_impl_opts impl_opts; 961 size_t impl_opts_size = sizeof(impl_opts); 962 struct spdk_sock_opts opts; 963 int rc; 964 965 if (!strlen(trid->trsvcid)) { 966 SPDK_ERRLOG("Service id is required\n"); 967 return -EINVAL; 968 } 969 970 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 971 972 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 973 if (trsvcid_int < 0) { 974 SPDK_ERRLOG("Invalid trsvcid '%s'\n", trid->trsvcid); 975 return -EINVAL; 976 } 977 978 port = calloc(1, sizeof(*port)); 979 if (!port) { 980 SPDK_ERRLOG("Port allocation failed\n"); 981 return -ENOMEM; 982 } 983 984 port->trid = trid; 985 986 sock_impl_name = NULL; 987 988 opts.opts_size = sizeof(opts); 989 spdk_sock_get_default_opts(&opts); 990 opts.priority = ttransport->tcp_opts.sock_priority; 991 opts.ack_timeout = transport->opts.ack_timeout; 992 if (listen_opts->secure_channel) { 993 if (listen_opts->sock_impl && 994 strncmp("ssl", listen_opts->sock_impl, strlen(listen_opts->sock_impl))) { 995 SPDK_ERRLOG("Enabling secure_channel while specifying a sock_impl different from 'ssl' is unsupported"); 996 free(port); 997 return -EINVAL; 998 } 999 listen_opts->sock_impl = "ssl"; 1000 } 1001 1002 if (listen_opts->sock_impl) { 1003 sock_impl_name = listen_opts->sock_impl; 1004 spdk_sock_impl_get_opts(sock_impl_name, &impl_opts, &impl_opts_size); 1005 1006 if (!strncmp("ssl", sock_impl_name, strlen(sock_impl_name))) { 1007 if (!g_tls_log) { 1008 SPDK_NOTICELOG("TLS support is considered experimental\n"); 1009 g_tls_log = true; 1010 } 1011 impl_opts.tls_version = SPDK_TLS_VERSION_1_3; 1012 impl_opts.get_key = tcp_sock_get_key; 1013 impl_opts.get_key_ctx = ttransport; 1014 impl_opts.tls_cipher_suites = "TLS_AES_256_GCM_SHA384:TLS_AES_128_GCM_SHA256"; 1015 } 1016 1017 opts.impl_opts = &impl_opts; 1018 opts.impl_opts_size = sizeof(impl_opts); 1019 } 1020 1021 port->listen_sock = spdk_sock_listen_ext(trid->traddr, trsvcid_int, 1022 sock_impl_name, &opts); 1023 if (port->listen_sock == NULL) { 1024 SPDK_ERRLOG("spdk_sock_listen(%s, %d) failed: %s (%d)\n", 1025 trid->traddr, trsvcid_int, 1026 spdk_strerror(errno), errno); 1027 free(port); 1028 return -errno; 1029 } 1030 1031 if (spdk_sock_is_ipv4(port->listen_sock)) { 1032 adrfam = SPDK_NVMF_ADRFAM_IPV4; 1033 } else if (spdk_sock_is_ipv6(port->listen_sock)) { 1034 adrfam = SPDK_NVMF_ADRFAM_IPV6; 1035 } else { 1036 SPDK_ERRLOG("Unhandled socket type\n"); 1037 adrfam = 0; 1038 } 1039 1040 if (adrfam != trid->adrfam) { 1041 SPDK_ERRLOG("Socket address family mismatch\n"); 1042 spdk_sock_close(&port->listen_sock); 1043 free(port); 1044 return -EINVAL; 1045 } 1046 1047 rc = spdk_sock_group_add_sock(ttransport->listen_sock_group, port->listen_sock, nvmf_tcp_accept_cb, 1048 port); 1049 if (rc < 0) { 1050 SPDK_ERRLOG("Failed to add socket to the listen socket group\n"); 1051 spdk_sock_close(&port->listen_sock); 1052 free(port); 1053 return -errno; 1054 } 1055 1056 port->transport = transport; 1057 1058 SPDK_NOTICELOG("*** NVMe/TCP Target Listening on %s port %s ***\n", 1059 trid->traddr, trid->trsvcid); 1060 1061 TAILQ_INSERT_TAIL(&ttransport->ports, port, link); 1062 return 0; 1063 } 1064 1065 static void 1066 nvmf_tcp_stop_listen(struct spdk_nvmf_transport *transport, 1067 const struct spdk_nvme_transport_id *trid) 1068 { 1069 struct spdk_nvmf_tcp_transport *ttransport; 1070 struct spdk_nvmf_tcp_port *port; 1071 1072 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1073 1074 SPDK_DEBUGLOG(nvmf_tcp, "Removing listen address %s port %s\n", 1075 trid->traddr, trid->trsvcid); 1076 1077 port = nvmf_tcp_find_port(ttransport, trid); 1078 if (port) { 1079 spdk_sock_group_remove_sock(ttransport->listen_sock_group, port->listen_sock); 1080 TAILQ_REMOVE(&ttransport->ports, port, link); 1081 spdk_sock_close(&port->listen_sock); 1082 free(port); 1083 } 1084 } 1085 1086 static void nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1087 enum nvme_tcp_pdu_recv_state state); 1088 1089 static void 1090 nvmf_tcp_qpair_set_state(struct spdk_nvmf_tcp_qpair *tqpair, enum nvme_tcp_qpair_state state) 1091 { 1092 tqpair->state = state; 1093 spdk_trace_record(TRACE_TCP_QP_STATE_CHANGE, tqpair->qpair.trace_id, 0, 0, 1094 (uint64_t)tqpair->state); 1095 } 1096 1097 static void 1098 nvmf_tcp_qpair_disconnect(struct spdk_nvmf_tcp_qpair *tqpair) 1099 { 1100 SPDK_DEBUGLOG(nvmf_tcp, "Disconnecting qpair %p\n", tqpair); 1101 1102 spdk_trace_record(TRACE_TCP_QP_DISCONNECT, tqpair->qpair.trace_id, 0, 0); 1103 1104 if (tqpair->state <= NVME_TCP_QPAIR_STATE_RUNNING) { 1105 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_EXITING); 1106 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1107 spdk_poller_unregister(&tqpair->timeout_poller); 1108 1109 /* This will end up calling nvmf_tcp_close_qpair */ 1110 spdk_nvmf_qpair_disconnect(&tqpair->qpair); 1111 } 1112 } 1113 1114 static void 1115 _mgmt_pdu_write_done(void *_tqpair, int err) 1116 { 1117 struct spdk_nvmf_tcp_qpair *tqpair = _tqpair; 1118 struct nvme_tcp_pdu *pdu = tqpair->mgmt_pdu; 1119 1120 if (spdk_unlikely(err != 0)) { 1121 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1122 return; 1123 } 1124 1125 assert(pdu->cb_fn != NULL); 1126 pdu->cb_fn(pdu->cb_arg); 1127 } 1128 1129 static void 1130 _req_pdu_write_done(void *req, int err) 1131 { 1132 struct spdk_nvmf_tcp_req *tcp_req = req; 1133 struct nvme_tcp_pdu *pdu = tcp_req->pdu; 1134 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 1135 1136 assert(tcp_req->pdu_in_use); 1137 tcp_req->pdu_in_use = false; 1138 1139 /* If the request is in a completed state, we're waiting for write completion to free it */ 1140 if (spdk_unlikely(tcp_req->state == TCP_REQUEST_STATE_COMPLETED)) { 1141 nvmf_tcp_request_free(tcp_req); 1142 return; 1143 } 1144 1145 if (spdk_unlikely(err != 0)) { 1146 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1147 return; 1148 } 1149 1150 assert(pdu->cb_fn != NULL); 1151 pdu->cb_fn(pdu->cb_arg); 1152 } 1153 1154 static void 1155 _pdu_write_done(struct nvme_tcp_pdu *pdu, int err) 1156 { 1157 pdu->sock_req.cb_fn(pdu->sock_req.cb_arg, err); 1158 } 1159 1160 static void 1161 _tcp_write_pdu(struct nvme_tcp_pdu *pdu) 1162 { 1163 int rc; 1164 uint32_t mapped_length; 1165 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 1166 1167 pdu->sock_req.iovcnt = nvme_tcp_build_iovs(pdu->iov, SPDK_COUNTOF(pdu->iov), pdu, 1168 tqpair->host_hdgst_enable, tqpair->host_ddgst_enable, &mapped_length); 1169 spdk_sock_writev_async(tqpair->sock, &pdu->sock_req); 1170 1171 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP || 1172 pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ) { 1173 /* Try to force the send immediately. */ 1174 rc = spdk_sock_flush(tqpair->sock); 1175 if (rc > 0 && (uint32_t)rc == mapped_length) { 1176 _pdu_write_done(pdu, 0); 1177 } else { 1178 SPDK_ERRLOG("Could not write %s to socket: rc=%d, errno=%d\n", 1179 pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP ? 1180 "IC_RESP" : "TERM_REQ", rc, errno); 1181 _pdu_write_done(pdu, rc >= 0 ? -EAGAIN : -errno); 1182 } 1183 } 1184 } 1185 1186 static void 1187 data_crc32_accel_done(void *cb_arg, int status) 1188 { 1189 struct nvme_tcp_pdu *pdu = cb_arg; 1190 1191 if (spdk_unlikely(status)) { 1192 SPDK_ERRLOG("Failed to compute the data digest for pdu =%p\n", pdu); 1193 _pdu_write_done(pdu, status); 1194 return; 1195 } 1196 1197 pdu->data_digest_crc32 ^= SPDK_CRC32C_XOR; 1198 MAKE_DIGEST_WORD(pdu->data_digest, pdu->data_digest_crc32); 1199 1200 _tcp_write_pdu(pdu); 1201 } 1202 1203 static void 1204 pdu_data_crc32_compute(struct nvme_tcp_pdu *pdu) 1205 { 1206 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 1207 int rc = 0; 1208 1209 /* Data Digest */ 1210 if (pdu->data_len > 0 && g_nvme_tcp_ddgst[pdu->hdr.common.pdu_type] && tqpair->host_ddgst_enable) { 1211 /* Only support this limitated case for the first step */ 1212 if (spdk_likely(!pdu->dif_ctx && (pdu->data_len % SPDK_NVME_TCP_DIGEST_ALIGNMENT == 0) 1213 && tqpair->group)) { 1214 rc = spdk_accel_submit_crc32cv(tqpair->group->accel_channel, &pdu->data_digest_crc32, pdu->data_iov, 1215 pdu->data_iovcnt, 0, data_crc32_accel_done, pdu); 1216 if (spdk_likely(rc == 0)) { 1217 return; 1218 } 1219 } else { 1220 pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu); 1221 } 1222 data_crc32_accel_done(pdu, rc); 1223 } else { 1224 _tcp_write_pdu(pdu); 1225 } 1226 } 1227 1228 static void 1229 nvmf_tcp_qpair_write_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1230 struct nvme_tcp_pdu *pdu, 1231 nvme_tcp_qpair_xfer_complete_cb cb_fn, 1232 void *cb_arg) 1233 { 1234 int hlen; 1235 uint32_t crc32c; 1236 1237 assert(tqpair->pdu_in_progress != pdu); 1238 1239 hlen = pdu->hdr.common.hlen; 1240 pdu->cb_fn = cb_fn; 1241 pdu->cb_arg = cb_arg; 1242 1243 pdu->iov[0].iov_base = &pdu->hdr.raw; 1244 pdu->iov[0].iov_len = hlen; 1245 1246 /* Header Digest */ 1247 if (g_nvme_tcp_hdgst[pdu->hdr.common.pdu_type] && tqpair->host_hdgst_enable) { 1248 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 1249 MAKE_DIGEST_WORD((uint8_t *)pdu->hdr.raw + hlen, crc32c); 1250 } 1251 1252 /* Data Digest */ 1253 pdu_data_crc32_compute(pdu); 1254 } 1255 1256 static void 1257 nvmf_tcp_qpair_write_mgmt_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1258 nvme_tcp_qpair_xfer_complete_cb cb_fn, 1259 void *cb_arg) 1260 { 1261 struct nvme_tcp_pdu *pdu = tqpair->mgmt_pdu; 1262 1263 pdu->sock_req.cb_fn = _mgmt_pdu_write_done; 1264 pdu->sock_req.cb_arg = tqpair; 1265 1266 nvmf_tcp_qpair_write_pdu(tqpair, pdu, cb_fn, cb_arg); 1267 } 1268 1269 static void 1270 nvmf_tcp_qpair_write_req_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1271 struct spdk_nvmf_tcp_req *tcp_req, 1272 nvme_tcp_qpair_xfer_complete_cb cb_fn, 1273 void *cb_arg) 1274 { 1275 struct nvme_tcp_pdu *pdu = tcp_req->pdu; 1276 1277 pdu->sock_req.cb_fn = _req_pdu_write_done; 1278 pdu->sock_req.cb_arg = tcp_req; 1279 1280 assert(!tcp_req->pdu_in_use); 1281 tcp_req->pdu_in_use = true; 1282 1283 nvmf_tcp_qpair_write_pdu(tqpair, pdu, cb_fn, cb_arg); 1284 } 1285 1286 static int 1287 nvmf_tcp_qpair_init_mem_resource(struct spdk_nvmf_tcp_qpair *tqpair) 1288 { 1289 uint32_t i; 1290 struct spdk_nvmf_transport_opts *opts; 1291 uint32_t in_capsule_data_size; 1292 1293 opts = &tqpair->qpair.transport->opts; 1294 1295 in_capsule_data_size = opts->in_capsule_data_size; 1296 if (opts->dif_insert_or_strip) { 1297 in_capsule_data_size = SPDK_BDEV_BUF_SIZE_WITH_MD(in_capsule_data_size); 1298 } 1299 1300 tqpair->resource_count = opts->max_queue_depth; 1301 1302 tqpair->reqs = calloc(tqpair->resource_count, sizeof(*tqpair->reqs)); 1303 if (!tqpair->reqs) { 1304 SPDK_ERRLOG("Unable to allocate reqs on tqpair=%p\n", tqpair); 1305 return -1; 1306 } 1307 1308 if (in_capsule_data_size) { 1309 tqpair->bufs = spdk_zmalloc(tqpair->resource_count * in_capsule_data_size, 0x1000, 1310 NULL, SPDK_ENV_LCORE_ID_ANY, 1311 SPDK_MALLOC_DMA); 1312 if (!tqpair->bufs) { 1313 SPDK_ERRLOG("Unable to allocate bufs on tqpair=%p.\n", tqpair); 1314 return -1; 1315 } 1316 } 1317 /* prepare memory space for receiving pdus and tcp_req */ 1318 /* Add additional 1 member, which will be used for mgmt_pdu owned by the tqpair */ 1319 tqpair->pdus = spdk_dma_zmalloc((2 * tqpair->resource_count + 1) * sizeof(*tqpair->pdus), 0x1000, 1320 NULL); 1321 if (!tqpair->pdus) { 1322 SPDK_ERRLOG("Unable to allocate pdu pool on tqpair =%p.\n", tqpair); 1323 return -1; 1324 } 1325 1326 for (i = 0; i < tqpair->resource_count; i++) { 1327 struct spdk_nvmf_tcp_req *tcp_req = &tqpair->reqs[i]; 1328 1329 tcp_req->ttag = i + 1; 1330 tcp_req->req.qpair = &tqpair->qpair; 1331 1332 tcp_req->pdu = &tqpair->pdus[i]; 1333 tcp_req->pdu->qpair = tqpair; 1334 1335 /* Set up memory to receive commands */ 1336 if (tqpair->bufs) { 1337 tcp_req->buf = (void *)((uintptr_t)tqpair->bufs + (i * in_capsule_data_size)); 1338 } 1339 1340 /* Set the cmdn and rsp */ 1341 tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp; 1342 tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd; 1343 1344 tcp_req->req.stripped_data = NULL; 1345 1346 /* Initialize request state to FREE */ 1347 tcp_req->state = TCP_REQUEST_STATE_FREE; 1348 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 1349 tqpair->state_cntr[TCP_REQUEST_STATE_FREE]++; 1350 } 1351 1352 for (; i < 2 * tqpair->resource_count; i++) { 1353 struct nvme_tcp_pdu *pdu = &tqpair->pdus[i]; 1354 1355 pdu->qpair = tqpair; 1356 SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, pdu, slist); 1357 } 1358 1359 tqpair->mgmt_pdu = &tqpair->pdus[i]; 1360 tqpair->mgmt_pdu->qpair = tqpair; 1361 tqpair->pdu_in_progress = SLIST_FIRST(&tqpair->tcp_pdu_free_queue); 1362 SLIST_REMOVE_HEAD(&tqpair->tcp_pdu_free_queue, slist); 1363 tqpair->tcp_pdu_working_count = 1; 1364 1365 tqpair->recv_buf_size = (in_capsule_data_size + sizeof(struct spdk_nvme_tcp_cmd) + 2 * 1366 SPDK_NVME_TCP_DIGEST_LEN) * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1367 1368 return 0; 1369 } 1370 1371 static int 1372 nvmf_tcp_qpair_init(struct spdk_nvmf_qpair *qpair) 1373 { 1374 struct spdk_nvmf_tcp_qpair *tqpair; 1375 1376 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1377 1378 SPDK_DEBUGLOG(nvmf_tcp, "New TCP Connection: %p\n", qpair); 1379 1380 spdk_trace_record(TRACE_TCP_QP_CREATE, tqpair->qpair.trace_id, 0, 0); 1381 1382 /* Initialise request state queues of the qpair */ 1383 TAILQ_INIT(&tqpair->tcp_req_free_queue); 1384 TAILQ_INIT(&tqpair->tcp_req_working_queue); 1385 SLIST_INIT(&tqpair->tcp_pdu_free_queue); 1386 tqpair->qpair.queue_depth = 0; 1387 1388 tqpair->host_hdgst_enable = true; 1389 tqpair->host_ddgst_enable = true; 1390 1391 return 0; 1392 } 1393 1394 static int 1395 nvmf_tcp_qpair_sock_init(struct spdk_nvmf_tcp_qpair *tqpair) 1396 { 1397 char saddr[32], caddr[32]; 1398 uint16_t sport, cport; 1399 char owner[256]; 1400 int rc; 1401 1402 rc = spdk_sock_getaddr(tqpair->sock, saddr, sizeof(saddr), &sport, 1403 caddr, sizeof(caddr), &cport); 1404 if (rc != 0) { 1405 SPDK_ERRLOG("spdk_sock_getaddr() failed\n"); 1406 return rc; 1407 } 1408 snprintf(owner, sizeof(owner), "%s:%d", caddr, cport); 1409 tqpair->qpair.trace_id = spdk_trace_register_owner(OWNER_TYPE_NVMF_TCP, owner); 1410 spdk_trace_record(TRACE_TCP_QP_SOCK_INIT, tqpair->qpair.trace_id, 0, 0); 1411 1412 /* set low water mark */ 1413 rc = spdk_sock_set_recvlowat(tqpair->sock, 1); 1414 if (rc != 0) { 1415 SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n"); 1416 return rc; 1417 } 1418 1419 return 0; 1420 } 1421 1422 static void 1423 nvmf_tcp_handle_connect(struct spdk_nvmf_tcp_port *port, struct spdk_sock *sock) 1424 { 1425 struct spdk_nvmf_tcp_qpair *tqpair; 1426 int rc; 1427 1428 SPDK_DEBUGLOG(nvmf_tcp, "New connection accepted on %s port %s\n", 1429 port->trid->traddr, port->trid->trsvcid); 1430 1431 tqpair = calloc(1, sizeof(struct spdk_nvmf_tcp_qpair)); 1432 if (tqpair == NULL) { 1433 SPDK_ERRLOG("Could not allocate new connection.\n"); 1434 spdk_sock_close(&sock); 1435 return; 1436 } 1437 1438 tqpair->sock = sock; 1439 tqpair->state_cntr[TCP_REQUEST_STATE_FREE] = 0; 1440 tqpair->port = port; 1441 tqpair->qpair.transport = port->transport; 1442 1443 rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr, 1444 sizeof(tqpair->target_addr), &tqpair->target_port, 1445 tqpair->initiator_addr, sizeof(tqpair->initiator_addr), 1446 &tqpair->initiator_port); 1447 if (rc < 0) { 1448 SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair); 1449 nvmf_tcp_qpair_destroy(tqpair); 1450 return; 1451 } 1452 1453 spdk_nvmf_tgt_new_qpair(port->transport->tgt, &tqpair->qpair); 1454 } 1455 1456 static uint32_t 1457 nvmf_tcp_port_accept(struct spdk_nvmf_tcp_port *port) 1458 { 1459 struct spdk_sock *sock; 1460 uint32_t count = 0; 1461 int i; 1462 1463 for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) { 1464 sock = spdk_sock_accept(port->listen_sock); 1465 if (sock == NULL) { 1466 break; 1467 } 1468 count++; 1469 nvmf_tcp_handle_connect(port, sock); 1470 } 1471 1472 return count; 1473 } 1474 1475 static int 1476 nvmf_tcp_accept(void *ctx) 1477 { 1478 struct spdk_nvmf_transport *transport = ctx; 1479 struct spdk_nvmf_tcp_transport *ttransport; 1480 int count; 1481 1482 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1483 1484 count = spdk_sock_group_poll(ttransport->listen_sock_group); 1485 if (count < 0) { 1486 SPDK_ERRLOG("Fail in TCP listen socket group poll\n"); 1487 } 1488 1489 return count != 0 ? SPDK_POLLER_BUSY : SPDK_POLLER_IDLE; 1490 } 1491 1492 static void 1493 nvmf_tcp_accept_cb(void *ctx, struct spdk_sock_group *group, struct spdk_sock *sock) 1494 { 1495 struct spdk_nvmf_tcp_port *port = ctx; 1496 1497 nvmf_tcp_port_accept(port); 1498 } 1499 1500 static void 1501 nvmf_tcp_discover(struct spdk_nvmf_transport *transport, 1502 struct spdk_nvme_transport_id *trid, 1503 struct spdk_nvmf_discovery_log_page_entry *entry) 1504 { 1505 struct spdk_nvmf_tcp_port *port; 1506 struct spdk_nvmf_tcp_transport *ttransport; 1507 1508 entry->trtype = SPDK_NVMF_TRTYPE_TCP; 1509 entry->adrfam = trid->adrfam; 1510 1511 spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' '); 1512 spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' '); 1513 1514 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1515 port = nvmf_tcp_find_port(ttransport, trid); 1516 1517 assert(port != NULL); 1518 1519 if (strcmp(spdk_sock_get_impl_name(port->listen_sock), "ssl") == 0) { 1520 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_REQUIRED; 1521 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_TLS_1_3; 1522 } else { 1523 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_REQUIRED; 1524 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE; 1525 } 1526 } 1527 1528 static struct spdk_nvmf_tcp_control_msg_list * 1529 nvmf_tcp_control_msg_list_create(uint16_t num_messages) 1530 { 1531 struct spdk_nvmf_tcp_control_msg_list *list; 1532 struct spdk_nvmf_tcp_control_msg *msg; 1533 uint16_t i; 1534 1535 list = calloc(1, sizeof(*list)); 1536 if (!list) { 1537 SPDK_ERRLOG("Failed to allocate memory for list structure\n"); 1538 return NULL; 1539 } 1540 1541 list->msg_buf = spdk_zmalloc(num_messages * SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, 1542 NVMF_DATA_BUFFER_ALIGNMENT, NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 1543 if (!list->msg_buf) { 1544 SPDK_ERRLOG("Failed to allocate memory for control message buffers\n"); 1545 free(list); 1546 return NULL; 1547 } 1548 1549 STAILQ_INIT(&list->free_msgs); 1550 STAILQ_INIT(&list->waiting_for_msg_reqs); 1551 1552 for (i = 0; i < num_messages; i++) { 1553 msg = (struct spdk_nvmf_tcp_control_msg *)((char *)list->msg_buf + i * 1554 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1555 STAILQ_INSERT_TAIL(&list->free_msgs, msg, link); 1556 } 1557 1558 return list; 1559 } 1560 1561 static void 1562 nvmf_tcp_control_msg_list_free(struct spdk_nvmf_tcp_control_msg_list *list) 1563 { 1564 if (!list) { 1565 return; 1566 } 1567 1568 spdk_free(list->msg_buf); 1569 free(list); 1570 } 1571 1572 static struct spdk_nvmf_transport_poll_group * 1573 nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport, 1574 struct spdk_nvmf_poll_group *group) 1575 { 1576 struct spdk_nvmf_tcp_transport *ttransport; 1577 struct spdk_nvmf_tcp_poll_group *tgroup; 1578 1579 if (spdk_interrupt_mode_is_enabled()) { 1580 SPDK_ERRLOG("TCP transport does not support interrupt mode\n"); 1581 return NULL; 1582 } 1583 1584 tgroup = calloc(1, sizeof(*tgroup)); 1585 if (!tgroup) { 1586 return NULL; 1587 } 1588 1589 tgroup->sock_group = spdk_sock_group_create(&tgroup->group); 1590 if (!tgroup->sock_group) { 1591 goto cleanup; 1592 } 1593 1594 TAILQ_INIT(&tgroup->qpairs); 1595 TAILQ_INIT(&tgroup->await_req); 1596 1597 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1598 1599 if (transport->opts.in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 1600 SPDK_DEBUGLOG(nvmf_tcp, "ICD %u is less than min required for admin/fabric commands (%u). " 1601 "Creating control messages list\n", transport->opts.in_capsule_data_size, 1602 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1603 tgroup->control_msg_list = nvmf_tcp_control_msg_list_create(ttransport->tcp_opts.control_msg_num); 1604 if (!tgroup->control_msg_list) { 1605 goto cleanup; 1606 } 1607 } 1608 1609 tgroup->accel_channel = spdk_accel_get_io_channel(); 1610 if (spdk_unlikely(!tgroup->accel_channel)) { 1611 SPDK_ERRLOG("Cannot create accel_channel for tgroup=%p\n", tgroup); 1612 goto cleanup; 1613 } 1614 1615 TAILQ_INSERT_TAIL(&ttransport->poll_groups, tgroup, link); 1616 if (ttransport->next_pg == NULL) { 1617 ttransport->next_pg = tgroup; 1618 } 1619 1620 return &tgroup->group; 1621 1622 cleanup: 1623 nvmf_tcp_poll_group_destroy(&tgroup->group); 1624 return NULL; 1625 } 1626 1627 static struct spdk_nvmf_transport_poll_group * 1628 nvmf_tcp_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair) 1629 { 1630 struct spdk_nvmf_tcp_transport *ttransport; 1631 struct spdk_nvmf_tcp_poll_group **pg; 1632 struct spdk_nvmf_tcp_qpair *tqpair; 1633 struct spdk_sock_group *group = NULL, *hint = NULL; 1634 int rc; 1635 1636 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 1637 1638 if (TAILQ_EMPTY(&ttransport->poll_groups)) { 1639 return NULL; 1640 } 1641 1642 pg = &ttransport->next_pg; 1643 assert(*pg != NULL); 1644 hint = (*pg)->sock_group; 1645 1646 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1647 rc = spdk_sock_get_optimal_sock_group(tqpair->sock, &group, hint); 1648 if (rc != 0) { 1649 return NULL; 1650 } else if (group != NULL) { 1651 /* Optimal poll group was found */ 1652 return spdk_sock_group_get_ctx(group); 1653 } 1654 1655 /* The hint was used for optimal poll group, advance next_pg. */ 1656 *pg = TAILQ_NEXT(*pg, link); 1657 if (*pg == NULL) { 1658 *pg = TAILQ_FIRST(&ttransport->poll_groups); 1659 } 1660 1661 return spdk_sock_group_get_ctx(hint); 1662 } 1663 1664 static void 1665 nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group) 1666 { 1667 struct spdk_nvmf_tcp_poll_group *tgroup, *next_tgroup; 1668 struct spdk_nvmf_tcp_transport *ttransport; 1669 1670 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 1671 spdk_sock_group_close(&tgroup->sock_group); 1672 if (tgroup->control_msg_list) { 1673 nvmf_tcp_control_msg_list_free(tgroup->control_msg_list); 1674 } 1675 1676 if (tgroup->accel_channel) { 1677 spdk_put_io_channel(tgroup->accel_channel); 1678 } 1679 1680 if (tgroup->group.transport == NULL) { 1681 /* Transport can be NULL when nvmf_tcp_poll_group_create() 1682 * calls this function directly in a failure path. */ 1683 free(tgroup); 1684 return; 1685 } 1686 1687 ttransport = SPDK_CONTAINEROF(tgroup->group.transport, struct spdk_nvmf_tcp_transport, transport); 1688 1689 next_tgroup = TAILQ_NEXT(tgroup, link); 1690 TAILQ_REMOVE(&ttransport->poll_groups, tgroup, link); 1691 if (next_tgroup == NULL) { 1692 next_tgroup = TAILQ_FIRST(&ttransport->poll_groups); 1693 } 1694 if (ttransport->next_pg == tgroup) { 1695 ttransport->next_pg = next_tgroup; 1696 } 1697 1698 free(tgroup); 1699 } 1700 1701 static void 1702 nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1703 enum nvme_tcp_pdu_recv_state state) 1704 { 1705 if (tqpair->recv_state == state) { 1706 SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n", 1707 tqpair, state); 1708 return; 1709 } 1710 1711 if (spdk_unlikely(state == NVME_TCP_PDU_RECV_STATE_QUIESCING)) { 1712 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH && tqpair->pdu_in_progress) { 1713 SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, tqpair->pdu_in_progress, slist); 1714 tqpair->tcp_pdu_working_count--; 1715 } 1716 } 1717 1718 if (spdk_unlikely(state == NVME_TCP_PDU_RECV_STATE_ERROR)) { 1719 assert(tqpair->tcp_pdu_working_count == 0); 1720 } 1721 1722 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1723 /* When leaving the await req state, move the qpair to the main list */ 1724 TAILQ_REMOVE(&tqpair->group->await_req, tqpair, link); 1725 TAILQ_INSERT_TAIL(&tqpair->group->qpairs, tqpair, link); 1726 } else if (state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1727 TAILQ_REMOVE(&tqpair->group->qpairs, tqpair, link); 1728 TAILQ_INSERT_TAIL(&tqpair->group->await_req, tqpair, link); 1729 } 1730 1731 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv state=%d\n", tqpair, state); 1732 tqpair->recv_state = state; 1733 1734 spdk_trace_record(TRACE_TCP_QP_RCV_STATE_CHANGE, tqpair->qpair.trace_id, 0, 0, 1735 (uint64_t)tqpair->recv_state); 1736 } 1737 1738 static int 1739 nvmf_tcp_qpair_handle_timeout(void *ctx) 1740 { 1741 struct spdk_nvmf_tcp_qpair *tqpair = ctx; 1742 1743 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1744 1745 SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair, 1746 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT); 1747 1748 nvmf_tcp_qpair_disconnect(tqpair); 1749 return SPDK_POLLER_BUSY; 1750 } 1751 1752 static void 1753 nvmf_tcp_send_c2h_term_req_complete(void *cb_arg) 1754 { 1755 struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg; 1756 1757 if (!tqpair->timeout_poller) { 1758 tqpair->timeout_poller = SPDK_POLLER_REGISTER(nvmf_tcp_qpair_handle_timeout, tqpair, 1759 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000); 1760 } 1761 } 1762 1763 static void 1764 nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu, 1765 enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset) 1766 { 1767 struct nvme_tcp_pdu *rsp_pdu; 1768 struct spdk_nvme_tcp_term_req_hdr *c2h_term_req; 1769 uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req); 1770 uint32_t copy_len; 1771 1772 rsp_pdu = tqpair->mgmt_pdu; 1773 1774 c2h_term_req = &rsp_pdu->hdr.term_req; 1775 c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ; 1776 c2h_term_req->common.hlen = c2h_term_req_hdr_len; 1777 c2h_term_req->fes = fes; 1778 1779 if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1780 (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1781 DSET32(&c2h_term_req->fei, error_offset); 1782 } 1783 1784 copy_len = spdk_min(pdu->hdr.common.hlen, SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE); 1785 1786 /* Copy the error info into the buffer */ 1787 memcpy((uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, pdu->hdr.raw, copy_len); 1788 nvme_tcp_pdu_set_data(rsp_pdu, (uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, copy_len); 1789 1790 /* Contain the header of the wrong received pdu */ 1791 c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len; 1792 tqpair->wait_terminate = true; 1793 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1794 nvmf_tcp_qpair_write_mgmt_pdu(tqpair, nvmf_tcp_send_c2h_term_req_complete, tqpair); 1795 } 1796 1797 static void 1798 nvmf_tcp_capsule_cmd_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1799 struct spdk_nvmf_tcp_qpair *tqpair, 1800 struct nvme_tcp_pdu *pdu) 1801 { 1802 struct spdk_nvmf_tcp_req *tcp_req; 1803 1804 assert(pdu->psh_valid_bytes == pdu->psh_len); 1805 assert(pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD); 1806 1807 tcp_req = nvmf_tcp_req_get(tqpair); 1808 if (!tcp_req) { 1809 /* Directly return and make the allocation retry again. This can happen if we're 1810 * using asynchronous writes to send the response to the host or when releasing 1811 * zero-copy buffers after a response has been sent. In both cases, the host might 1812 * receive the response before we've finished processing the request and is free to 1813 * send another one. 1814 */ 1815 if (tqpair->state_cntr[TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST] > 0 || 1816 tqpair->state_cntr[TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE] > 0) { 1817 return; 1818 } 1819 1820 /* The host sent more commands than the maximum queue depth. */ 1821 SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair); 1822 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1823 return; 1824 } 1825 1826 pdu->req = tcp_req; 1827 assert(tcp_req->state == TCP_REQUEST_STATE_NEW); 1828 nvmf_tcp_req_process(ttransport, tcp_req); 1829 } 1830 1831 static void 1832 nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1833 struct spdk_nvmf_tcp_qpair *tqpair, 1834 struct nvme_tcp_pdu *pdu) 1835 { 1836 struct spdk_nvmf_tcp_req *tcp_req; 1837 struct spdk_nvme_tcp_cmd *capsule_cmd; 1838 uint32_t error_offset = 0; 1839 enum spdk_nvme_tcp_term_req_fes fes; 1840 struct spdk_nvme_cpl *rsp; 1841 1842 capsule_cmd = &pdu->hdr.capsule_cmd; 1843 tcp_req = pdu->req; 1844 assert(tcp_req != NULL); 1845 1846 /* Zero-copy requests don't support ICD */ 1847 assert(!spdk_nvmf_request_using_zcopy(&tcp_req->req)); 1848 1849 if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) { 1850 SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n", 1851 SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo); 1852 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1853 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1854 goto err; 1855 } 1856 1857 rsp = &tcp_req->req.rsp->nvme_cpl; 1858 if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) { 1859 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 1860 } else { 1861 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1862 } 1863 1864 nvmf_tcp_req_process(ttransport, tcp_req); 1865 1866 return; 1867 err: 1868 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1869 } 1870 1871 static void 1872 nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1873 struct spdk_nvmf_tcp_qpair *tqpair, 1874 struct nvme_tcp_pdu *pdu) 1875 { 1876 struct spdk_nvmf_tcp_req *tcp_req; 1877 uint32_t error_offset = 0; 1878 enum spdk_nvme_tcp_term_req_fes fes = 0; 1879 struct spdk_nvme_tcp_h2c_data_hdr *h2c_data; 1880 1881 h2c_data = &pdu->hdr.h2c_data; 1882 1883 SPDK_DEBUGLOG(nvmf_tcp, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n", 1884 tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag); 1885 1886 if (h2c_data->ttag > tqpair->resource_count) { 1887 SPDK_DEBUGLOG(nvmf_tcp, "ttag %u is larger than allowed %u.\n", h2c_data->ttag, 1888 tqpair->resource_count); 1889 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1890 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1891 goto err; 1892 } 1893 1894 tcp_req = &tqpair->reqs[h2c_data->ttag - 1]; 1895 1896 if (spdk_unlikely(tcp_req->state != TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER && 1897 tcp_req->state != TCP_REQUEST_STATE_AWAITING_R2T_ACK)) { 1898 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req(%p), tqpair=%p, has error state in %d\n", tcp_req, tqpair, 1899 tcp_req->state); 1900 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1901 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1902 goto err; 1903 } 1904 1905 if (spdk_unlikely(tcp_req->req.cmd->nvme_cmd.cid != h2c_data->cccid)) { 1906 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req(%p), tqpair=%p, expected %u but %u for cccid.\n", tcp_req, tqpair, 1907 tcp_req->req.cmd->nvme_cmd.cid, h2c_data->cccid); 1908 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1909 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid); 1910 goto err; 1911 } 1912 1913 if (tcp_req->h2c_offset != h2c_data->datao) { 1914 SPDK_DEBUGLOG(nvmf_tcp, 1915 "tcp_req(%p), tqpair=%p, expected data offset %u, but data offset is %u\n", 1916 tcp_req, tqpair, tcp_req->h2c_offset, h2c_data->datao); 1917 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1918 goto err; 1919 } 1920 1921 if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) { 1922 SPDK_DEBUGLOG(nvmf_tcp, 1923 "tcp_req(%p), tqpair=%p, (datao=%u + datal=%u) exceeds requested length=%u\n", 1924 tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length); 1925 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1926 goto err; 1927 } 1928 1929 pdu->req = tcp_req; 1930 1931 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 1932 pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 1933 } 1934 1935 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 1936 h2c_data->datao, h2c_data->datal); 1937 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1938 return; 1939 1940 err: 1941 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1942 } 1943 1944 static void 1945 nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req, 1946 struct spdk_nvmf_tcp_qpair *tqpair) 1947 { 1948 struct nvme_tcp_pdu *rsp_pdu; 1949 struct spdk_nvme_tcp_rsp *capsule_resp; 1950 1951 SPDK_DEBUGLOG(nvmf_tcp, "enter, tqpair=%p\n", tqpair); 1952 1953 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1954 assert(rsp_pdu != NULL); 1955 1956 capsule_resp = &rsp_pdu->hdr.capsule_resp; 1957 capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP; 1958 capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp); 1959 capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl; 1960 if (tqpair->host_hdgst_enable) { 1961 capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1962 capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1963 } 1964 1965 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_request_free, tcp_req); 1966 } 1967 1968 static void 1969 nvmf_tcp_pdu_c2h_data_complete(void *cb_arg) 1970 { 1971 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1972 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, 1973 struct spdk_nvmf_tcp_qpair, qpair); 1974 1975 assert(tqpair != NULL); 1976 1977 if (spdk_unlikely(tcp_req->pdu->rw_offset < tcp_req->req.length)) { 1978 SPDK_DEBUGLOG(nvmf_tcp, "sending another C2H part, offset %u length %u\n", tcp_req->pdu->rw_offset, 1979 tcp_req->req.length); 1980 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 1981 return; 1982 } 1983 1984 if (tcp_req->pdu->hdr.c2h_data.common.flags & SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS) { 1985 nvmf_tcp_request_free(tcp_req); 1986 } else { 1987 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 1988 } 1989 } 1990 1991 static void 1992 nvmf_tcp_r2t_complete(void *cb_arg) 1993 { 1994 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1995 struct spdk_nvmf_tcp_transport *ttransport; 1996 1997 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 1998 struct spdk_nvmf_tcp_transport, transport); 1999 2000 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 2001 2002 if (tcp_req->h2c_offset == tcp_req->req.length) { 2003 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2004 nvmf_tcp_req_process(ttransport, tcp_req); 2005 } 2006 } 2007 2008 static void 2009 nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 2010 struct spdk_nvmf_tcp_req *tcp_req) 2011 { 2012 struct nvme_tcp_pdu *rsp_pdu; 2013 struct spdk_nvme_tcp_r2t_hdr *r2t; 2014 2015 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 2016 assert(rsp_pdu != NULL); 2017 2018 r2t = &rsp_pdu->hdr.r2t; 2019 r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T; 2020 r2t->common.plen = r2t->common.hlen = sizeof(*r2t); 2021 2022 if (tqpair->host_hdgst_enable) { 2023 r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 2024 r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 2025 } 2026 2027 r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid; 2028 r2t->ttag = tcp_req->ttag; 2029 r2t->r2to = tcp_req->h2c_offset; 2030 r2t->r2tl = tcp_req->req.length; 2031 2032 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 2033 2034 SPDK_DEBUGLOG(nvmf_tcp, 2035 "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n", 2036 tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl); 2037 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_r2t_complete, tcp_req); 2038 } 2039 2040 static void 2041 nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 2042 struct spdk_nvmf_tcp_qpair *tqpair, 2043 struct nvme_tcp_pdu *pdu) 2044 { 2045 struct spdk_nvmf_tcp_req *tcp_req; 2046 struct spdk_nvme_cpl *rsp; 2047 2048 tcp_req = pdu->req; 2049 assert(tcp_req != NULL); 2050 2051 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2052 2053 tcp_req->h2c_offset += pdu->data_len; 2054 2055 /* Wait for all of the data to arrive AND for the initial R2T PDU send to be 2056 * acknowledged before moving on. */ 2057 if (tcp_req->h2c_offset == tcp_req->req.length && 2058 tcp_req->state == TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER) { 2059 /* After receiving all the h2c data, we need to check whether there is 2060 * transient transport error */ 2061 rsp = &tcp_req->req.rsp->nvme_cpl; 2062 if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) { 2063 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2064 } else { 2065 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2066 } 2067 nvmf_tcp_req_process(ttransport, tcp_req); 2068 } 2069 } 2070 2071 static void 2072 nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req) 2073 { 2074 SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req, 2075 spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]); 2076 if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 2077 (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 2078 SPDK_DEBUGLOG(nvmf_tcp, "The offset from the start of the PDU header is %u\n", 2079 DGET32(h2c_term_req->fei)); 2080 } 2081 } 2082 2083 static void 2084 nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair, 2085 struct nvme_tcp_pdu *pdu) 2086 { 2087 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 2088 uint32_t error_offset = 0; 2089 enum spdk_nvme_tcp_term_req_fes fes; 2090 2091 if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) { 2092 SPDK_ERRLOG("Fatal Error Status(FES) is unknown for h2c_term_req pdu=%p\n", pdu); 2093 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2094 error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes); 2095 goto end; 2096 } 2097 2098 /* set the data buffer */ 2099 nvme_tcp_pdu_set_data(pdu, (uint8_t *)pdu->hdr.raw + h2c_term_req->common.hlen, 2100 h2c_term_req->common.plen - h2c_term_req->common.hlen); 2101 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2102 return; 2103 end: 2104 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2105 } 2106 2107 static void 2108 nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 2109 struct nvme_tcp_pdu *pdu) 2110 { 2111 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 2112 2113 nvmf_tcp_h2c_term_req_dump(h2c_term_req); 2114 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2115 } 2116 2117 static void 2118 _nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu) 2119 { 2120 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 2121 struct spdk_nvmf_tcp_transport, transport); 2122 2123 switch (pdu->hdr.common.pdu_type) { 2124 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2125 nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu); 2126 break; 2127 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2128 nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu); 2129 break; 2130 2131 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2132 nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu); 2133 break; 2134 2135 default: 2136 /* The code should not go to here */ 2137 SPDK_ERRLOG("ERROR pdu type %d\n", pdu->hdr.common.pdu_type); 2138 break; 2139 } 2140 SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, pdu, slist); 2141 tqpair->tcp_pdu_working_count--; 2142 } 2143 2144 static inline void 2145 nvmf_tcp_req_set_cpl(struct spdk_nvmf_tcp_req *treq, int sct, int sc) 2146 { 2147 treq->req.rsp->nvme_cpl.status.sct = sct; 2148 treq->req.rsp->nvme_cpl.status.sc = sc; 2149 treq->req.rsp->nvme_cpl.cid = treq->req.cmd->nvme_cmd.cid; 2150 } 2151 2152 static void 2153 data_crc32_calc_done(void *cb_arg, int status) 2154 { 2155 struct nvme_tcp_pdu *pdu = cb_arg; 2156 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 2157 2158 /* async crc32 calculation is failed and use direct calculation to check */ 2159 if (spdk_unlikely(status)) { 2160 SPDK_ERRLOG("Data digest on tqpair=(%p) with pdu=%p failed to be calculated asynchronously\n", 2161 tqpair, pdu); 2162 pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu); 2163 } 2164 pdu->data_digest_crc32 ^= SPDK_CRC32C_XOR; 2165 if (!MATCH_DIGEST_WORD(pdu->data_digest, pdu->data_digest_crc32)) { 2166 SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 2167 assert(pdu->req != NULL); 2168 nvmf_tcp_req_set_cpl(pdu->req, SPDK_NVME_SCT_GENERIC, 2169 SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR); 2170 } 2171 _nvmf_tcp_pdu_payload_handle(tqpair, pdu); 2172 } 2173 2174 static void 2175 nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu) 2176 { 2177 int rc = 0; 2178 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2179 tqpair->pdu_in_progress = NULL; 2180 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2181 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2182 /* check data digest if need */ 2183 if (pdu->ddgst_enable) { 2184 if (tqpair->qpair.qid != 0 && !pdu->dif_ctx && tqpair->group && 2185 (pdu->data_len % SPDK_NVME_TCP_DIGEST_ALIGNMENT == 0)) { 2186 rc = spdk_accel_submit_crc32cv(tqpair->group->accel_channel, &pdu->data_digest_crc32, pdu->data_iov, 2187 pdu->data_iovcnt, 0, data_crc32_calc_done, pdu); 2188 if (spdk_likely(rc == 0)) { 2189 return; 2190 } 2191 } else { 2192 pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu); 2193 } 2194 data_crc32_calc_done(pdu, rc); 2195 } else { 2196 _nvmf_tcp_pdu_payload_handle(tqpair, pdu); 2197 } 2198 } 2199 2200 static void 2201 nvmf_tcp_send_icresp_complete(void *cb_arg) 2202 { 2203 struct spdk_nvmf_tcp_qpair *tqpair = cb_arg; 2204 2205 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_RUNNING); 2206 } 2207 2208 static void 2209 nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport, 2210 struct spdk_nvmf_tcp_qpair *tqpair, 2211 struct nvme_tcp_pdu *pdu) 2212 { 2213 struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr.ic_req; 2214 struct nvme_tcp_pdu *rsp_pdu; 2215 struct spdk_nvme_tcp_ic_resp *ic_resp; 2216 uint32_t error_offset = 0; 2217 enum spdk_nvme_tcp_term_req_fes fes; 2218 2219 /* Only PFV 0 is defined currently */ 2220 if (ic_req->pfv != 0) { 2221 SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv); 2222 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2223 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv); 2224 goto end; 2225 } 2226 2227 /* This value is 0’s based value in units of dwords should not be larger than SPDK_NVME_TCP_HPDA_MAX */ 2228 if (ic_req->hpda > SPDK_NVME_TCP_HPDA_MAX) { 2229 SPDK_ERRLOG("ICReq HPDA out of range 0 to 31, got %u\n", ic_req->hpda); 2230 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2231 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, hpda); 2232 goto end; 2233 } 2234 2235 /* MAXR2T is 0's based */ 2236 SPDK_DEBUGLOG(nvmf_tcp, "maxr2t =%u\n", (ic_req->maxr2t + 1u)); 2237 2238 tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false; 2239 if (!tqpair->host_hdgst_enable) { 2240 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 2241 } 2242 2243 tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false; 2244 if (!tqpair->host_ddgst_enable) { 2245 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 2246 } 2247 2248 tqpair->recv_buf_size = spdk_max(tqpair->recv_buf_size, MIN_SOCK_PIPE_SIZE); 2249 /* Now that we know whether digests are enabled, properly size the receive buffer */ 2250 if (spdk_sock_set_recvbuf(tqpair->sock, tqpair->recv_buf_size) < 0) { 2251 SPDK_WARNLOG("Unable to allocate enough memory for receive buffer on tqpair=%p with size=%d\n", 2252 tqpair, 2253 tqpair->recv_buf_size); 2254 /* Not fatal. */ 2255 } 2256 2257 tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX); 2258 SPDK_DEBUGLOG(nvmf_tcp, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda); 2259 2260 rsp_pdu = tqpair->mgmt_pdu; 2261 2262 ic_resp = &rsp_pdu->hdr.ic_resp; 2263 ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP; 2264 ic_resp->common.hlen = ic_resp->common.plen = sizeof(*ic_resp); 2265 ic_resp->pfv = 0; 2266 ic_resp->cpda = tqpair->cpda; 2267 ic_resp->maxh2cdata = ttransport->transport.opts.max_io_size; 2268 ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0; 2269 ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0; 2270 2271 SPDK_DEBUGLOG(nvmf_tcp, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable); 2272 SPDK_DEBUGLOG(nvmf_tcp, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable); 2273 2274 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_INITIALIZING); 2275 nvmf_tcp_qpair_write_mgmt_pdu(tqpair, nvmf_tcp_send_icresp_complete, tqpair); 2276 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2277 return; 2278 end: 2279 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2280 } 2281 2282 static void 2283 nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair, 2284 struct spdk_nvmf_tcp_transport *ttransport) 2285 { 2286 struct nvme_tcp_pdu *pdu; 2287 int rc; 2288 uint32_t crc32c, error_offset = 0; 2289 enum spdk_nvme_tcp_term_req_fes fes; 2290 2291 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 2292 pdu = tqpair->pdu_in_progress; 2293 2294 SPDK_DEBUGLOG(nvmf_tcp, "pdu type of tqpair(%p) is %d\n", tqpair, 2295 pdu->hdr.common.pdu_type); 2296 /* check header digest if needed */ 2297 if (pdu->has_hdgst) { 2298 SPDK_DEBUGLOG(nvmf_tcp, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair); 2299 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 2300 rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, crc32c); 2301 if (rc == 0) { 2302 SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 2303 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 2304 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2305 return; 2306 2307 } 2308 } 2309 2310 switch (pdu->hdr.common.pdu_type) { 2311 case SPDK_NVME_TCP_PDU_TYPE_IC_REQ: 2312 nvmf_tcp_icreq_handle(ttransport, tqpair, pdu); 2313 break; 2314 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2315 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_REQ); 2316 break; 2317 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2318 nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu); 2319 break; 2320 2321 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2322 nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu); 2323 break; 2324 2325 default: 2326 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress->hdr.common.pdu_type); 2327 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2328 error_offset = 1; 2329 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2330 break; 2331 } 2332 } 2333 2334 static void 2335 nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair) 2336 { 2337 struct nvme_tcp_pdu *pdu; 2338 uint32_t error_offset = 0; 2339 enum spdk_nvme_tcp_term_req_fes fes; 2340 uint8_t expected_hlen, pdo; 2341 bool plen_error = false, pdo_error = false; 2342 2343 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 2344 pdu = tqpair->pdu_in_progress; 2345 assert(pdu); 2346 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) { 2347 if (tqpair->state != NVME_TCP_QPAIR_STATE_INVALID) { 2348 SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu); 2349 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 2350 goto err; 2351 } 2352 expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req); 2353 if (pdu->hdr.common.plen != expected_hlen) { 2354 plen_error = true; 2355 } 2356 } else { 2357 if (tqpair->state != NVME_TCP_QPAIR_STATE_RUNNING) { 2358 SPDK_ERRLOG("The TCP/IP connection is not negotiated\n"); 2359 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 2360 goto err; 2361 } 2362 2363 switch (pdu->hdr.common.pdu_type) { 2364 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2365 expected_hlen = sizeof(struct spdk_nvme_tcp_cmd); 2366 pdo = pdu->hdr.common.pdo; 2367 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 2368 pdo_error = true; 2369 break; 2370 } 2371 2372 if (pdu->hdr.common.plen < expected_hlen) { 2373 plen_error = true; 2374 } 2375 break; 2376 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2377 expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr); 2378 pdo = pdu->hdr.common.pdo; 2379 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 2380 pdo_error = true; 2381 break; 2382 } 2383 if (pdu->hdr.common.plen < expected_hlen) { 2384 plen_error = true; 2385 } 2386 break; 2387 2388 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2389 expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr); 2390 if ((pdu->hdr.common.plen <= expected_hlen) || 2391 (pdu->hdr.common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) { 2392 plen_error = true; 2393 } 2394 break; 2395 2396 default: 2397 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr.common.pdu_type); 2398 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2399 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type); 2400 goto err; 2401 } 2402 } 2403 2404 if (pdu->hdr.common.hlen != expected_hlen) { 2405 SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n", 2406 pdu->hdr.common.pdu_type, 2407 expected_hlen, pdu->hdr.common.hlen, tqpair); 2408 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2409 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen); 2410 goto err; 2411 } else if (pdo_error) { 2412 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2413 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 2414 } else if (plen_error) { 2415 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2416 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 2417 goto err; 2418 } else { 2419 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 2420 nvme_tcp_pdu_calc_psh_len(tqpair->pdu_in_progress, tqpair->host_hdgst_enable); 2421 return; 2422 } 2423 err: 2424 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2425 } 2426 2427 static int 2428 nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair) 2429 { 2430 int rc = 0; 2431 struct nvme_tcp_pdu *pdu; 2432 enum nvme_tcp_pdu_recv_state prev_state; 2433 uint32_t data_len; 2434 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 2435 struct spdk_nvmf_tcp_transport, transport); 2436 2437 /* The loop here is to allow for several back-to-back state changes. */ 2438 do { 2439 prev_state = tqpair->recv_state; 2440 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state); 2441 2442 pdu = tqpair->pdu_in_progress; 2443 assert(pdu != NULL || 2444 tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY || 2445 tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_QUIESCING || 2446 tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 2447 2448 switch (tqpair->recv_state) { 2449 /* Wait for the common header */ 2450 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 2451 if (!pdu) { 2452 pdu = SLIST_FIRST(&tqpair->tcp_pdu_free_queue); 2453 if (spdk_unlikely(!pdu)) { 2454 return NVME_TCP_PDU_IN_PROGRESS; 2455 } 2456 SLIST_REMOVE_HEAD(&tqpair->tcp_pdu_free_queue, slist); 2457 tqpair->pdu_in_progress = pdu; 2458 tqpair->tcp_pdu_working_count++; 2459 } 2460 memset(pdu, 0, offsetof(struct nvme_tcp_pdu, qpair)); 2461 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 2462 /* FALLTHROUGH */ 2463 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 2464 if (spdk_unlikely(tqpair->state == NVME_TCP_QPAIR_STATE_INITIALIZING)) { 2465 return rc; 2466 } 2467 2468 rc = nvme_tcp_read_data(tqpair->sock, 2469 sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes, 2470 (void *)&pdu->hdr.common + pdu->ch_valid_bytes); 2471 if (rc < 0) { 2472 SPDK_DEBUGLOG(nvmf_tcp, "will disconnect tqpair=%p\n", tqpair); 2473 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2474 break; 2475 } else if (rc > 0) { 2476 pdu->ch_valid_bytes += rc; 2477 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, tqpair->qpair.trace_id, rc, 0); 2478 } 2479 2480 if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) { 2481 return NVME_TCP_PDU_IN_PROGRESS; 2482 } 2483 2484 /* The command header of this PDU has now been read from the socket. */ 2485 nvmf_tcp_pdu_ch_handle(tqpair); 2486 break; 2487 /* Wait for the pdu specific header */ 2488 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 2489 rc = nvme_tcp_read_data(tqpair->sock, 2490 pdu->psh_len - pdu->psh_valid_bytes, 2491 (void *)&pdu->hdr.raw + sizeof(struct spdk_nvme_tcp_common_pdu_hdr) + pdu->psh_valid_bytes); 2492 if (rc < 0) { 2493 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2494 break; 2495 } else if (rc > 0) { 2496 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, tqpair->qpair.trace_id, rc, 0); 2497 pdu->psh_valid_bytes += rc; 2498 } 2499 2500 if (pdu->psh_valid_bytes < pdu->psh_len) { 2501 return NVME_TCP_PDU_IN_PROGRESS; 2502 } 2503 2504 /* All header(ch, psh, head digits) of this PDU has now been read from the socket. */ 2505 nvmf_tcp_pdu_psh_handle(tqpair, ttransport); 2506 break; 2507 /* Wait for the req slot */ 2508 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 2509 nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu); 2510 break; 2511 /* Wait for the request processing loop to acquire a buffer for the PDU */ 2512 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_BUF: 2513 break; 2514 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 2515 /* check whether the data is valid, if not we just return */ 2516 if (!pdu->data_len) { 2517 return NVME_TCP_PDU_IN_PROGRESS; 2518 } 2519 2520 data_len = pdu->data_len; 2521 /* data digest */ 2522 if (spdk_unlikely((pdu->hdr.common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) && 2523 tqpair->host_ddgst_enable)) { 2524 data_len += SPDK_NVME_TCP_DIGEST_LEN; 2525 pdu->ddgst_enable = true; 2526 } 2527 2528 rc = nvme_tcp_read_payload_data(tqpair->sock, pdu); 2529 if (rc < 0) { 2530 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2531 break; 2532 } 2533 pdu->rw_offset += rc; 2534 2535 if (pdu->rw_offset < data_len) { 2536 return NVME_TCP_PDU_IN_PROGRESS; 2537 } 2538 2539 /* Generate and insert DIF to whole data block received if DIF is enabled */ 2540 if (spdk_unlikely(pdu->dif_ctx != NULL) && 2541 spdk_dif_generate_stream(pdu->data_iov, pdu->data_iovcnt, 0, data_len, 2542 pdu->dif_ctx) != 0) { 2543 SPDK_ERRLOG("DIF generate failed\n"); 2544 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2545 break; 2546 } 2547 2548 /* All of this PDU has now been read from the socket. */ 2549 nvmf_tcp_pdu_payload_handle(tqpair, pdu); 2550 break; 2551 case NVME_TCP_PDU_RECV_STATE_QUIESCING: 2552 if (tqpair->tcp_pdu_working_count != 0) { 2553 return NVME_TCP_PDU_IN_PROGRESS; 2554 } 2555 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 2556 break; 2557 case NVME_TCP_PDU_RECV_STATE_ERROR: 2558 if (spdk_sock_is_connected(tqpair->sock) && tqpair->wait_terminate) { 2559 return NVME_TCP_PDU_IN_PROGRESS; 2560 } 2561 return NVME_TCP_PDU_FATAL; 2562 default: 2563 SPDK_ERRLOG("The state(%d) is invalid\n", tqpair->recv_state); 2564 abort(); 2565 break; 2566 } 2567 } while (tqpair->recv_state != prev_state); 2568 2569 return rc; 2570 } 2571 2572 static inline void * 2573 nvmf_tcp_control_msg_get(struct spdk_nvmf_tcp_control_msg_list *list, 2574 struct spdk_nvmf_tcp_req *tcp_req) 2575 { 2576 struct spdk_nvmf_tcp_control_msg *msg; 2577 2578 assert(list); 2579 2580 msg = STAILQ_FIRST(&list->free_msgs); 2581 if (!msg) { 2582 SPDK_DEBUGLOG(nvmf_tcp, "Out of control messages\n"); 2583 STAILQ_INSERT_TAIL(&list->waiting_for_msg_reqs, tcp_req, control_msg_link); 2584 return NULL; 2585 } 2586 STAILQ_REMOVE_HEAD(&list->free_msgs, link); 2587 return msg; 2588 } 2589 2590 static inline void 2591 nvmf_tcp_control_msg_put(struct spdk_nvmf_tcp_control_msg_list *list, void *_msg) 2592 { 2593 struct spdk_nvmf_tcp_control_msg *msg = _msg; 2594 struct spdk_nvmf_tcp_req *tcp_req; 2595 struct spdk_nvmf_tcp_transport *ttransport; 2596 2597 assert(list); 2598 STAILQ_INSERT_HEAD(&list->free_msgs, msg, link); 2599 if (!STAILQ_EMPTY(&list->waiting_for_msg_reqs)) { 2600 tcp_req = STAILQ_FIRST(&list->waiting_for_msg_reqs); 2601 STAILQ_REMOVE_HEAD(&list->waiting_for_msg_reqs, control_msg_link); 2602 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 2603 struct spdk_nvmf_tcp_transport, transport); 2604 nvmf_tcp_req_process(ttransport, tcp_req); 2605 } 2606 } 2607 2608 static void 2609 nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_req *tcp_req, 2610 struct spdk_nvmf_transport *transport, 2611 struct spdk_nvmf_transport_poll_group *group) 2612 { 2613 struct spdk_nvmf_request *req = &tcp_req->req; 2614 struct spdk_nvme_cmd *cmd; 2615 struct spdk_nvme_sgl_descriptor *sgl; 2616 struct spdk_nvmf_tcp_poll_group *tgroup; 2617 enum spdk_nvme_tcp_term_req_fes fes; 2618 struct nvme_tcp_pdu *pdu; 2619 struct spdk_nvmf_tcp_qpair *tqpair; 2620 uint32_t length, error_offset = 0; 2621 2622 cmd = &req->cmd->nvme_cmd; 2623 sgl = &cmd->dptr.sgl1; 2624 2625 if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK && 2626 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) { 2627 /* get request length from sgl */ 2628 length = sgl->unkeyed.length; 2629 if (spdk_unlikely(length > transport->opts.max_io_size)) { 2630 SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n", 2631 length, transport->opts.max_io_size); 2632 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_LIMIT_EXCEEDED; 2633 goto fatal_err; 2634 } 2635 2636 /* fill request length and populate iovs */ 2637 req->length = length; 2638 2639 SPDK_DEBUGLOG(nvmf_tcp, "Data requested length= 0x%x\n", length); 2640 2641 if (spdk_unlikely(req->dif_enabled)) { 2642 req->dif.orig_length = length; 2643 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2644 req->dif.elba_length = length; 2645 } 2646 2647 if (nvmf_ctrlr_use_zcopy(req)) { 2648 SPDK_DEBUGLOG(nvmf_tcp, "Using zero-copy to execute request %p\n", tcp_req); 2649 req->data_from_pool = false; 2650 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_HAVE_BUFFER); 2651 return; 2652 } 2653 2654 if (spdk_nvmf_request_get_buffers(req, group, transport, length)) { 2655 /* No available buffers. Queue this request up. */ 2656 SPDK_DEBUGLOG(nvmf_tcp, "No available large data buffers. Queueing request %p\n", 2657 tcp_req); 2658 return; 2659 } 2660 2661 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_HAVE_BUFFER); 2662 SPDK_DEBUGLOG(nvmf_tcp, "Request %p took %d buffer/s from central pool, and data=%p\n", 2663 tcp_req, req->iovcnt, req->iov[0].iov_base); 2664 2665 return; 2666 } else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK && 2667 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) { 2668 uint64_t offset = sgl->address; 2669 uint32_t max_len = transport->opts.in_capsule_data_size; 2670 2671 assert(tcp_req->has_in_capsule_data); 2672 /* Capsule Cmd with In-capsule Data should get data length from pdu header */ 2673 tqpair = tcp_req->pdu->qpair; 2674 /* receiving pdu is not same with the pdu in tcp_req */ 2675 pdu = tqpair->pdu_in_progress; 2676 length = pdu->hdr.common.plen - pdu->psh_len - sizeof(struct spdk_nvme_tcp_common_pdu_hdr); 2677 if (tqpair->host_ddgst_enable) { 2678 length -= SPDK_NVME_TCP_DIGEST_LEN; 2679 } 2680 /* This error is not defined in NVMe/TCP spec, take this error as fatal error */ 2681 if (spdk_unlikely(length != sgl->unkeyed.length)) { 2682 SPDK_ERRLOG("In-Capsule Data length 0x%x is not equal to SGL data length 0x%x\n", 2683 length, sgl->unkeyed.length); 2684 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2685 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 2686 goto fatal_err; 2687 } 2688 2689 SPDK_DEBUGLOG(nvmf_tcp, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n", 2690 offset, length); 2691 2692 /* The NVMe/TCP transport does not use ICDOFF to control the in-capsule data offset. ICDOFF should be '0' */ 2693 if (spdk_unlikely(offset != 0)) { 2694 /* Not defined fatal error in NVMe/TCP spec, handle this error as a fatal error */ 2695 SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " should be ZERO in NVMe/TCP\n", offset); 2696 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 2697 error_offset = offsetof(struct spdk_nvme_tcp_cmd, ccsqe.dptr.sgl1.address); 2698 goto fatal_err; 2699 } 2700 2701 if (spdk_unlikely(length > max_len)) { 2702 /* According to the SPEC we should support ICD up to 8192 bytes for admin and fabric commands */ 2703 if (length <= SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE && 2704 (cmd->opc == SPDK_NVME_OPC_FABRIC || req->qpair->qid == 0)) { 2705 2706 /* Get a buffer from dedicated list */ 2707 SPDK_DEBUGLOG(nvmf_tcp, "Getting a buffer from control msg list\n"); 2708 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2709 assert(tgroup->control_msg_list); 2710 req->iov[0].iov_base = nvmf_tcp_control_msg_get(tgroup->control_msg_list, tcp_req); 2711 if (!req->iov[0].iov_base) { 2712 /* No available buffers. Queue this request up. */ 2713 SPDK_DEBUGLOG(nvmf_tcp, "No available ICD buffers. Queueing request %p\n", tcp_req); 2714 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_BUF); 2715 return; 2716 } 2717 } else { 2718 SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n", 2719 length, max_len); 2720 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_LIMIT_EXCEEDED; 2721 goto fatal_err; 2722 } 2723 } else { 2724 req->iov[0].iov_base = tcp_req->buf; 2725 } 2726 2727 req->length = length; 2728 req->data_from_pool = false; 2729 2730 if (spdk_unlikely(req->dif_enabled)) { 2731 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2732 req->dif.elba_length = length; 2733 } 2734 2735 req->iov[0].iov_len = length; 2736 req->iovcnt = 1; 2737 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_HAVE_BUFFER); 2738 2739 return; 2740 } 2741 /* If we want to handle the problem here, then we can't skip the following data segment. 2742 * Because this function runs before reading data part, now handle all errors as fatal errors. */ 2743 SPDK_ERRLOG("Invalid NVMf I/O Command SGL: Type 0x%x, Subtype 0x%x\n", 2744 sgl->generic.type, sgl->generic.subtype); 2745 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 2746 error_offset = offsetof(struct spdk_nvme_tcp_cmd, ccsqe.dptr.sgl1.generic); 2747 fatal_err: 2748 nvmf_tcp_send_c2h_term_req(tcp_req->pdu->qpair, tcp_req->pdu, fes, error_offset); 2749 } 2750 2751 static inline enum spdk_nvme_media_error_status_code 2752 nvmf_tcp_dif_error_to_compl_status(uint8_t err_type) { 2753 enum spdk_nvme_media_error_status_code result; 2754 2755 switch (err_type) 2756 { 2757 case SPDK_DIF_REFTAG_ERROR: 2758 result = SPDK_NVME_SC_REFERENCE_TAG_CHECK_ERROR; 2759 break; 2760 case SPDK_DIF_APPTAG_ERROR: 2761 result = SPDK_NVME_SC_APPLICATION_TAG_CHECK_ERROR; 2762 break; 2763 case SPDK_DIF_GUARD_ERROR: 2764 result = SPDK_NVME_SC_GUARD_CHECK_ERROR; 2765 break; 2766 default: 2767 SPDK_UNREACHABLE(); 2768 break; 2769 } 2770 2771 return result; 2772 } 2773 2774 static void 2775 _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2776 struct spdk_nvmf_tcp_req *tcp_req) 2777 { 2778 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF( 2779 tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 2780 struct nvme_tcp_pdu *rsp_pdu; 2781 struct spdk_nvme_tcp_c2h_data_hdr *c2h_data; 2782 uint32_t plen, pdo, alignment; 2783 int rc; 2784 2785 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2786 2787 rsp_pdu = tcp_req->pdu; 2788 assert(rsp_pdu != NULL); 2789 2790 c2h_data = &rsp_pdu->hdr.c2h_data; 2791 c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA; 2792 plen = c2h_data->common.hlen = sizeof(*c2h_data); 2793 2794 if (tqpair->host_hdgst_enable) { 2795 plen += SPDK_NVME_TCP_DIGEST_LEN; 2796 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 2797 } 2798 2799 /* set the psh */ 2800 c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid; 2801 c2h_data->datal = tcp_req->req.length - tcp_req->pdu->rw_offset; 2802 c2h_data->datao = tcp_req->pdu->rw_offset; 2803 2804 /* set the padding */ 2805 rsp_pdu->padding_len = 0; 2806 pdo = plen; 2807 if (tqpair->cpda) { 2808 alignment = (tqpair->cpda + 1) << 2; 2809 if (plen % alignment != 0) { 2810 pdo = (plen + alignment) / alignment * alignment; 2811 rsp_pdu->padding_len = pdo - plen; 2812 plen = pdo; 2813 } 2814 } 2815 2816 c2h_data->common.pdo = pdo; 2817 plen += c2h_data->datal; 2818 if (tqpair->host_ddgst_enable) { 2819 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF; 2820 plen += SPDK_NVME_TCP_DIGEST_LEN; 2821 } 2822 2823 c2h_data->common.plen = plen; 2824 2825 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2826 rsp_pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 2827 } 2828 2829 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2830 c2h_data->datao, c2h_data->datal); 2831 2832 2833 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU; 2834 /* Need to send the capsule response if response is not all 0 */ 2835 if (ttransport->tcp_opts.c2h_success && 2836 tcp_req->rsp.cdw0 == 0 && tcp_req->rsp.cdw1 == 0) { 2837 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS; 2838 } 2839 2840 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2841 struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl; 2842 struct spdk_dif_error err_blk = {}; 2843 uint32_t mapped_length = 0; 2844 uint32_t available_iovs = SPDK_COUNTOF(rsp_pdu->iov); 2845 uint32_t ddgst_len = 0; 2846 2847 if (tqpair->host_ddgst_enable) { 2848 /* Data digest consumes additional iov entry */ 2849 available_iovs--; 2850 /* plen needs to be updated since nvme_tcp_build_iovs compares expected and actual plen */ 2851 ddgst_len = SPDK_NVME_TCP_DIGEST_LEN; 2852 c2h_data->common.plen -= ddgst_len; 2853 } 2854 /* Temp call to estimate if data can be described by limited number of iovs. 2855 * iov vector will be rebuilt in nvmf_tcp_qpair_write_pdu */ 2856 nvme_tcp_build_iovs(rsp_pdu->iov, available_iovs, rsp_pdu, tqpair->host_hdgst_enable, 2857 false, &mapped_length); 2858 2859 if (mapped_length != c2h_data->common.plen) { 2860 c2h_data->datal = mapped_length - (c2h_data->common.plen - c2h_data->datal); 2861 SPDK_DEBUGLOG(nvmf_tcp, 2862 "Part C2H, data_len %u (of %u), PDU len %u, updated PDU len %u, offset %u\n", 2863 c2h_data->datal, tcp_req->req.length, c2h_data->common.plen, mapped_length, rsp_pdu->rw_offset); 2864 c2h_data->common.plen = mapped_length; 2865 2866 /* Rebuild pdu->data_iov since data length is changed */ 2867 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, c2h_data->datao, 2868 c2h_data->datal); 2869 2870 c2h_data->common.flags &= ~(SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU | 2871 SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS); 2872 } 2873 2874 c2h_data->common.plen += ddgst_len; 2875 2876 assert(rsp_pdu->rw_offset <= tcp_req->req.length); 2877 2878 rc = spdk_dif_verify_stream(rsp_pdu->data_iov, rsp_pdu->data_iovcnt, 2879 0, rsp_pdu->data_len, rsp_pdu->dif_ctx, &err_blk); 2880 if (rc != 0) { 2881 SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n", 2882 err_blk.err_type, err_blk.err_offset); 2883 rsp->status.sct = SPDK_NVME_SCT_MEDIA_ERROR; 2884 rsp->status.sc = nvmf_tcp_dif_error_to_compl_status(err_blk.err_type); 2885 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2886 return; 2887 } 2888 } 2889 2890 rsp_pdu->rw_offset += c2h_data->datal; 2891 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_pdu_c2h_data_complete, tcp_req); 2892 } 2893 2894 static void 2895 nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2896 struct spdk_nvmf_tcp_req *tcp_req) 2897 { 2898 nvmf_tcp_req_pdu_init(tcp_req); 2899 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2900 } 2901 2902 static int 2903 request_transfer_out(struct spdk_nvmf_request *req) 2904 { 2905 struct spdk_nvmf_tcp_req *tcp_req; 2906 struct spdk_nvmf_qpair *qpair; 2907 struct spdk_nvmf_tcp_qpair *tqpair; 2908 struct spdk_nvme_cpl *rsp; 2909 2910 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2911 2912 qpair = req->qpair; 2913 rsp = &req->rsp->nvme_cpl; 2914 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2915 2916 /* Advance our sq_head pointer */ 2917 if (qpair->sq_head == qpair->sq_head_max) { 2918 qpair->sq_head = 0; 2919 } else { 2920 qpair->sq_head++; 2921 } 2922 rsp->sqhd = qpair->sq_head; 2923 2924 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2925 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 2926 if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) { 2927 nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2928 } else { 2929 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2930 } 2931 2932 return 0; 2933 } 2934 2935 static void 2936 nvmf_tcp_check_fused_ordering(struct spdk_nvmf_tcp_transport *ttransport, 2937 struct spdk_nvmf_tcp_qpair *tqpair, 2938 struct spdk_nvmf_tcp_req *tcp_req) 2939 { 2940 enum spdk_nvme_cmd_fuse last, next; 2941 2942 last = tqpair->fused_first ? tqpair->fused_first->cmd.fuse : SPDK_NVME_CMD_FUSE_NONE; 2943 next = tcp_req->cmd.fuse; 2944 2945 assert(last != SPDK_NVME_CMD_FUSE_SECOND); 2946 2947 if (spdk_likely(last == SPDK_NVME_CMD_FUSE_NONE && next == SPDK_NVME_CMD_FUSE_NONE)) { 2948 return; 2949 } 2950 2951 if (last == SPDK_NVME_CMD_FUSE_FIRST) { 2952 if (next == SPDK_NVME_CMD_FUSE_SECOND) { 2953 /* This is a valid pair of fused commands. Point them at each other 2954 * so they can be submitted consecutively once ready to be executed. 2955 */ 2956 tqpair->fused_first->fused_pair = tcp_req; 2957 tcp_req->fused_pair = tqpair->fused_first; 2958 tqpair->fused_first = NULL; 2959 return; 2960 } else { 2961 /* Mark the last req as failed since it wasn't followed by a SECOND. */ 2962 tqpair->fused_first->fused_failed = true; 2963 2964 /* 2965 * If the last req is in READY_TO_EXECUTE state, then call 2966 * nvmf_tcp_req_process(), otherwise nothing else will kick it. 2967 */ 2968 if (tqpair->fused_first->state == TCP_REQUEST_STATE_READY_TO_EXECUTE) { 2969 nvmf_tcp_req_process(ttransport, tqpair->fused_first); 2970 } 2971 2972 tqpair->fused_first = NULL; 2973 } 2974 } 2975 2976 if (next == SPDK_NVME_CMD_FUSE_FIRST) { 2977 /* Set tqpair->fused_first here so that we know to check that the next request 2978 * is a SECOND (and to fail this one if it isn't). 2979 */ 2980 tqpair->fused_first = tcp_req; 2981 } else if (next == SPDK_NVME_CMD_FUSE_SECOND) { 2982 /* Mark this req failed since it is a SECOND and the last one was not a FIRST. */ 2983 tcp_req->fused_failed = true; 2984 } 2985 } 2986 2987 static bool 2988 nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 2989 struct spdk_nvmf_tcp_req *tcp_req) 2990 { 2991 struct spdk_nvmf_tcp_qpair *tqpair; 2992 uint32_t plen; 2993 struct nvme_tcp_pdu *pdu; 2994 enum spdk_nvmf_tcp_req_state prev_state; 2995 bool progress = false; 2996 struct spdk_nvmf_transport *transport = &ttransport->transport; 2997 struct spdk_nvmf_transport_poll_group *group; 2998 struct spdk_nvmf_tcp_poll_group *tgroup; 2999 3000 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 3001 group = &tqpair->group->group; 3002 assert(tcp_req->state != TCP_REQUEST_STATE_FREE); 3003 3004 /* If the qpair is not active, we need to abort the outstanding requests. */ 3005 if (!spdk_nvmf_qpair_is_active(&tqpair->qpair)) { 3006 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 3007 nvmf_tcp_request_get_buffers_abort(tcp_req); 3008 } 3009 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 3010 } 3011 3012 /* The loop here is to allow for several back-to-back state changes. */ 3013 do { 3014 prev_state = tcp_req->state; 3015 3016 SPDK_DEBUGLOG(nvmf_tcp, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state, 3017 tqpair); 3018 3019 switch (tcp_req->state) { 3020 case TCP_REQUEST_STATE_FREE: 3021 /* Some external code must kick a request into TCP_REQUEST_STATE_NEW 3022 * to escape this state. */ 3023 break; 3024 case TCP_REQUEST_STATE_NEW: 3025 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req, 3026 tqpair->qpair.queue_depth); 3027 3028 /* copy the cmd from the receive pdu */ 3029 tcp_req->cmd = tqpair->pdu_in_progress->hdr.capsule_cmd.ccsqe; 3030 3031 if (spdk_unlikely(spdk_nvmf_request_get_dif_ctx(&tcp_req->req, &tcp_req->req.dif.dif_ctx))) { 3032 tcp_req->req.dif_enabled = true; 3033 tqpair->pdu_in_progress->dif_ctx = &tcp_req->req.dif.dif_ctx; 3034 } 3035 3036 nvmf_tcp_check_fused_ordering(ttransport, tqpair, tcp_req); 3037 3038 /* The next state transition depends on the data transfer needs of this request. */ 3039 tcp_req->req.xfer = spdk_nvmf_req_get_xfer(&tcp_req->req); 3040 3041 if (spdk_unlikely(tcp_req->req.xfer == SPDK_NVME_DATA_BIDIRECTIONAL)) { 3042 nvmf_tcp_req_set_cpl(tcp_req, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_INVALID_OPCODE); 3043 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 3044 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3045 SPDK_DEBUGLOG(nvmf_tcp, "Request %p: invalid xfer type (BIDIRECTIONAL)\n", tcp_req); 3046 break; 3047 } 3048 3049 /* If no data to transfer, ready to execute. */ 3050 if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) { 3051 /* Reset the tqpair receiving pdu state */ 3052 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 3053 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 3054 break; 3055 } 3056 3057 pdu = tqpair->pdu_in_progress; 3058 plen = pdu->hdr.common.hlen; 3059 if (tqpair->host_hdgst_enable) { 3060 plen += SPDK_NVME_TCP_DIGEST_LEN; 3061 } 3062 if (pdu->hdr.common.plen != plen) { 3063 tcp_req->has_in_capsule_data = true; 3064 } else { 3065 /* Data is transmitted by C2H PDUs */ 3066 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 3067 } 3068 3069 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER); 3070 break; 3071 case TCP_REQUEST_STATE_NEED_BUFFER: 3072 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, tqpair->qpair.trace_id, 0, 3073 (uintptr_t)tcp_req); 3074 3075 assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE); 3076 3077 /* Try to get a data buffer */ 3078 nvmf_tcp_req_parse_sgl(tcp_req, transport, group); 3079 break; 3080 case TCP_REQUEST_STATE_HAVE_BUFFER: 3081 spdk_trace_record(TRACE_TCP_REQUEST_STATE_HAVE_BUFFER, tqpair->qpair.trace_id, 0, 3082 (uintptr_t)tcp_req); 3083 /* Get a zcopy buffer if the request can be serviced through zcopy */ 3084 if (spdk_nvmf_request_using_zcopy(&tcp_req->req)) { 3085 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 3086 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 3087 tcp_req->req.length = tcp_req->req.dif.elba_length; 3088 } 3089 3090 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_START); 3091 spdk_nvmf_request_zcopy_start(&tcp_req->req); 3092 break; 3093 } 3094 3095 assert(tcp_req->req.iovcnt > 0); 3096 3097 /* If data is transferring from host to controller, we need to do a transfer from the host. */ 3098 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 3099 if (tcp_req->req.data_from_pool) { 3100 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 3101 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 3102 } else { 3103 struct nvme_tcp_pdu *pdu; 3104 3105 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 3106 3107 pdu = tqpair->pdu_in_progress; 3108 SPDK_DEBUGLOG(nvmf_tcp, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, 3109 tqpair); 3110 /* No need to send r2t, contained in the capsuled data */ 3111 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 3112 0, tcp_req->req.length); 3113 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 3114 } 3115 break; 3116 } 3117 3118 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 3119 break; 3120 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3121 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_START, tqpair->qpair.trace_id, 0, 3122 (uintptr_t)tcp_req); 3123 /* Some external code must kick a request into TCP_REQUEST_STATE_ZCOPY_START_COMPLETED 3124 * to escape this state. */ 3125 break; 3126 case TCP_REQUEST_STATE_ZCOPY_START_COMPLETED: 3127 spdk_trace_record(TRACE_TCP_REQUEST_STATE_ZCOPY_START_COMPLETED, tqpair->qpair.trace_id, 0, 3128 (uintptr_t)tcp_req); 3129 if (spdk_unlikely(spdk_nvme_cpl_is_error(&tcp_req->req.rsp->nvme_cpl))) { 3130 SPDK_DEBUGLOG(nvmf_tcp, "Zero-copy start failed for tcp_req(%p) on tqpair=%p\n", 3131 tcp_req, tqpair); 3132 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3133 break; 3134 } 3135 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 3136 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 3137 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 3138 } else { 3139 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 3140 } 3141 break; 3142 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 3143 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, tqpair->qpair.trace_id, 0, 3144 (uintptr_t)tcp_req); 3145 /* The R2T completion or the h2c data incoming will kick it out of this state. */ 3146 break; 3147 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 3148 3149 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, tqpair->qpair.trace_id, 3150 0, (uintptr_t)tcp_req); 3151 /* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE 3152 * to escape this state. */ 3153 break; 3154 case TCP_REQUEST_STATE_READY_TO_EXECUTE: 3155 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, tqpair->qpair.trace_id, 0, 3156 (uintptr_t)tcp_req); 3157 3158 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 3159 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 3160 tcp_req->req.length = tcp_req->req.dif.elba_length; 3161 } 3162 3163 if (tcp_req->cmd.fuse != SPDK_NVME_CMD_FUSE_NONE) { 3164 if (tcp_req->fused_failed) { 3165 /* This request failed FUSED semantics. Fail it immediately, without 3166 * even sending it to the target layer. 3167 */ 3168 nvmf_tcp_req_set_cpl(tcp_req, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_ABORTED_MISSING_FUSED); 3169 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3170 break; 3171 } 3172 3173 if (tcp_req->fused_pair == NULL || 3174 tcp_req->fused_pair->state != TCP_REQUEST_STATE_READY_TO_EXECUTE) { 3175 /* This request is ready to execute, but either we don't know yet if it's 3176 * valid - i.e. this is a FIRST but we haven't received the next request yet), 3177 * or the other request of this fused pair isn't ready to execute. So 3178 * break here and this request will get processed later either when the 3179 * other request is ready or we find that this request isn't valid. 3180 */ 3181 break; 3182 } 3183 } 3184 3185 if (!spdk_nvmf_request_using_zcopy(&tcp_req->req)) { 3186 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING); 3187 /* If we get to this point, and this request is a fused command, we know that 3188 * it is part of a valid sequence (FIRST followed by a SECOND) and that both 3189 * requests are READY_TO_EXECUTE. So call spdk_nvmf_request_exec() both on this 3190 * request, and the other request of the fused pair, in the correct order. 3191 * Also clear the ->fused_pair pointers on both requests, since after this point 3192 * we no longer need to maintain the relationship between these two requests. 3193 */ 3194 if (tcp_req->cmd.fuse == SPDK_NVME_CMD_FUSE_SECOND) { 3195 assert(tcp_req->fused_pair != NULL); 3196 assert(tcp_req->fused_pair->fused_pair == tcp_req); 3197 nvmf_tcp_req_set_state(tcp_req->fused_pair, TCP_REQUEST_STATE_EXECUTING); 3198 spdk_nvmf_request_exec(&tcp_req->fused_pair->req); 3199 tcp_req->fused_pair->fused_pair = NULL; 3200 tcp_req->fused_pair = NULL; 3201 } 3202 spdk_nvmf_request_exec(&tcp_req->req); 3203 if (tcp_req->cmd.fuse == SPDK_NVME_CMD_FUSE_FIRST) { 3204 assert(tcp_req->fused_pair != NULL); 3205 assert(tcp_req->fused_pair->fused_pair == tcp_req); 3206 nvmf_tcp_req_set_state(tcp_req->fused_pair, TCP_REQUEST_STATE_EXECUTING); 3207 spdk_nvmf_request_exec(&tcp_req->fused_pair->req); 3208 tcp_req->fused_pair->fused_pair = NULL; 3209 tcp_req->fused_pair = NULL; 3210 } 3211 } else { 3212 /* For zero-copy, only requests with data coming from host to the 3213 * controller can end up here. */ 3214 assert(tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER); 3215 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT); 3216 spdk_nvmf_request_zcopy_end(&tcp_req->req, true); 3217 } 3218 3219 break; 3220 case TCP_REQUEST_STATE_EXECUTING: 3221 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req); 3222 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 3223 * to escape this state. */ 3224 break; 3225 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3226 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_COMMIT, tqpair->qpair.trace_id, 0, 3227 (uintptr_t)tcp_req); 3228 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 3229 * to escape this state. */ 3230 break; 3231 case TCP_REQUEST_STATE_EXECUTED: 3232 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req); 3233 3234 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 3235 tcp_req->req.length = tcp_req->req.dif.orig_length; 3236 } 3237 3238 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3239 break; 3240 case TCP_REQUEST_STATE_READY_TO_COMPLETE: 3241 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, tqpair->qpair.trace_id, 0, 3242 (uintptr_t)tcp_req); 3243 if (request_transfer_out(&tcp_req->req) != 0) { 3244 assert(0); /* No good way to handle this currently */ 3245 } 3246 break; 3247 case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST: 3248 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, tqpair->qpair.trace_id, 3249 0, (uintptr_t)tcp_req); 3250 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 3251 * to escape this state. */ 3252 break; 3253 case TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE: 3254 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_RELEASE, tqpair->qpair.trace_id, 0, 3255 (uintptr_t)tcp_req); 3256 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 3257 * to escape this state. */ 3258 break; 3259 case TCP_REQUEST_STATE_COMPLETED: 3260 spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req, 3261 tqpair->qpair.queue_depth); 3262 /* If there's an outstanding PDU sent to the host, the request is completed 3263 * due to the qpair being disconnected. We must delay the completion until 3264 * that write is done to avoid freeing the request twice. */ 3265 if (spdk_unlikely(tcp_req->pdu_in_use)) { 3266 SPDK_DEBUGLOG(nvmf_tcp, "Delaying completion due to outstanding " 3267 "write on req=%p\n", tcp_req); 3268 /* This can only happen for zcopy requests */ 3269 assert(spdk_nvmf_request_using_zcopy(&tcp_req->req)); 3270 assert(!spdk_nvmf_qpair_is_active(&tqpair->qpair)); 3271 break; 3272 } 3273 3274 if (tcp_req->req.data_from_pool) { 3275 spdk_nvmf_request_free_buffers(&tcp_req->req, group, transport); 3276 } else if (spdk_unlikely(tcp_req->has_in_capsule_data && 3277 (tcp_req->cmd.opc == SPDK_NVME_OPC_FABRIC || 3278 tqpair->qpair.qid == 0) && tcp_req->req.length > transport->opts.in_capsule_data_size)) { 3279 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3280 assert(tgroup->control_msg_list); 3281 SPDK_DEBUGLOG(nvmf_tcp, "Put buf to control msg list\n"); 3282 nvmf_tcp_control_msg_put(tgroup->control_msg_list, 3283 tcp_req->req.iov[0].iov_base); 3284 } else if (tcp_req->req.zcopy_bdev_io != NULL) { 3285 /* If the request has an unreleased zcopy bdev_io, it's either a 3286 * read, a failed write, or the qpair is being disconnected */ 3287 assert(spdk_nvmf_request_using_zcopy(&tcp_req->req)); 3288 assert(tcp_req->req.xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST || 3289 spdk_nvme_cpl_is_error(&tcp_req->req.rsp->nvme_cpl) || 3290 !spdk_nvmf_qpair_is_active(&tqpair->qpair)); 3291 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE); 3292 spdk_nvmf_request_zcopy_end(&tcp_req->req, false); 3293 break; 3294 } 3295 tcp_req->req.length = 0; 3296 tcp_req->req.iovcnt = 0; 3297 tcp_req->fused_failed = false; 3298 if (tcp_req->fused_pair) { 3299 /* This req was part of a valid fused pair, but failed before it got to 3300 * READ_TO_EXECUTE state. This means we need to fail the other request 3301 * in the pair, because it is no longer part of a valid pair. If the pair 3302 * already reached READY_TO_EXECUTE state, we need to kick it. 3303 */ 3304 tcp_req->fused_pair->fused_failed = true; 3305 if (tcp_req->fused_pair->state == TCP_REQUEST_STATE_READY_TO_EXECUTE) { 3306 nvmf_tcp_req_process(ttransport, tcp_req->fused_pair); 3307 } 3308 tcp_req->fused_pair = NULL; 3309 } 3310 3311 nvmf_tcp_req_put(tqpair, tcp_req); 3312 break; 3313 case TCP_REQUEST_NUM_STATES: 3314 default: 3315 assert(0); 3316 break; 3317 } 3318 3319 if (tcp_req->state != prev_state) { 3320 progress = true; 3321 } 3322 } while (tcp_req->state != prev_state); 3323 3324 return progress; 3325 } 3326 3327 static void 3328 nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock) 3329 { 3330 struct spdk_nvmf_tcp_qpair *tqpair = arg; 3331 int rc; 3332 3333 assert(tqpair != NULL); 3334 rc = nvmf_tcp_sock_process(tqpair); 3335 3336 /* If there was a new socket error, disconnect */ 3337 if (rc < 0) { 3338 nvmf_tcp_qpair_disconnect(tqpair); 3339 } 3340 } 3341 3342 static int 3343 nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group, 3344 struct spdk_nvmf_qpair *qpair) 3345 { 3346 struct spdk_nvmf_tcp_poll_group *tgroup; 3347 struct spdk_nvmf_tcp_qpair *tqpair; 3348 int rc; 3349 3350 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3351 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3352 3353 rc = nvmf_tcp_qpair_sock_init(tqpair); 3354 if (rc != 0) { 3355 SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair); 3356 return -1; 3357 } 3358 3359 rc = nvmf_tcp_qpair_init(&tqpair->qpair); 3360 if (rc < 0) { 3361 SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair); 3362 return -1; 3363 } 3364 3365 rc = nvmf_tcp_qpair_init_mem_resource(tqpair); 3366 if (rc < 0) { 3367 SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair); 3368 return -1; 3369 } 3370 3371 rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock, 3372 nvmf_tcp_sock_cb, tqpair); 3373 if (rc != 0) { 3374 SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n", 3375 spdk_strerror(errno), errno); 3376 return -1; 3377 } 3378 3379 tqpair->group = tgroup; 3380 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_INVALID); 3381 TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link); 3382 3383 return 0; 3384 } 3385 3386 static int 3387 nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group, 3388 struct spdk_nvmf_qpair *qpair) 3389 { 3390 struct spdk_nvmf_tcp_poll_group *tgroup; 3391 struct spdk_nvmf_tcp_qpair *tqpair; 3392 int rc; 3393 3394 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3395 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3396 3397 assert(tqpair->group == tgroup); 3398 3399 SPDK_DEBUGLOG(nvmf_tcp, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup); 3400 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 3401 /* Change the state to move the qpair from the await_req list to the main list 3402 * and prevent adding it again later by nvmf_tcp_qpair_set_recv_state() */ 3403 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 3404 } 3405 TAILQ_REMOVE(&tgroup->qpairs, tqpair, link); 3406 3407 /* Try to force out any pending writes */ 3408 spdk_sock_flush(tqpair->sock); 3409 3410 rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock); 3411 if (rc != 0) { 3412 SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n", 3413 spdk_strerror(errno), errno); 3414 } 3415 3416 return rc; 3417 } 3418 3419 static int 3420 nvmf_tcp_req_complete(struct spdk_nvmf_request *req) 3421 { 3422 struct spdk_nvmf_tcp_transport *ttransport; 3423 struct spdk_nvmf_tcp_req *tcp_req; 3424 3425 ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport); 3426 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 3427 3428 switch (tcp_req->state) { 3429 case TCP_REQUEST_STATE_EXECUTING: 3430 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3431 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 3432 break; 3433 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3434 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_ZCOPY_START_COMPLETED); 3435 break; 3436 case TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE: 3437 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 3438 break; 3439 default: 3440 SPDK_ERRLOG("Unexpected request state %d (cntlid:%d, qid:%d)\n", 3441 tcp_req->state, req->qpair->ctrlr->cntlid, req->qpair->qid); 3442 assert(0 && "Unexpected request state"); 3443 break; 3444 } 3445 3446 nvmf_tcp_req_process(ttransport, tcp_req); 3447 3448 return 0; 3449 } 3450 3451 static void 3452 nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair, 3453 spdk_nvmf_transport_qpair_fini_cb cb_fn, void *cb_arg) 3454 { 3455 struct spdk_nvmf_tcp_qpair *tqpair; 3456 3457 SPDK_DEBUGLOG(nvmf_tcp, "Qpair: %p\n", qpair); 3458 3459 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3460 3461 assert(tqpair->fini_cb_fn == NULL); 3462 tqpair->fini_cb_fn = cb_fn; 3463 tqpair->fini_cb_arg = cb_arg; 3464 3465 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_EXITED); 3466 nvmf_tcp_qpair_destroy(tqpair); 3467 } 3468 3469 static int 3470 nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group) 3471 { 3472 struct spdk_nvmf_tcp_poll_group *tgroup; 3473 int num_events, rc = 0, rc2; 3474 struct spdk_nvmf_tcp_qpair *tqpair, *tqpair_tmp; 3475 3476 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3477 3478 if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs) && TAILQ_EMPTY(&tgroup->await_req))) { 3479 return 0; 3480 } 3481 3482 num_events = spdk_sock_group_poll(tgroup->sock_group); 3483 if (spdk_unlikely(num_events < 0)) { 3484 SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group); 3485 } 3486 3487 TAILQ_FOREACH_SAFE(tqpair, &tgroup->await_req, link, tqpair_tmp) { 3488 rc2 = nvmf_tcp_sock_process(tqpair); 3489 3490 /* If there was a new socket error, disconnect */ 3491 if (spdk_unlikely(rc2 < 0)) { 3492 nvmf_tcp_qpair_disconnect(tqpair); 3493 if (rc == 0) { 3494 rc = rc2; 3495 } 3496 } 3497 } 3498 3499 return rc == 0 ? num_events : rc; 3500 } 3501 3502 static int 3503 nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair, 3504 struct spdk_nvme_transport_id *trid, bool peer) 3505 { 3506 struct spdk_nvmf_tcp_qpair *tqpair; 3507 uint16_t port; 3508 3509 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3510 spdk_nvme_trid_populate_transport(trid, SPDK_NVME_TRANSPORT_TCP); 3511 3512 if (peer) { 3513 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr); 3514 port = tqpair->initiator_port; 3515 } else { 3516 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr); 3517 port = tqpair->target_port; 3518 } 3519 3520 if (spdk_sock_is_ipv4(tqpair->sock)) { 3521 trid->adrfam = SPDK_NVMF_ADRFAM_IPV4; 3522 } else if (spdk_sock_is_ipv6(tqpair->sock)) { 3523 trid->adrfam = SPDK_NVMF_ADRFAM_IPV6; 3524 } else { 3525 return -1; 3526 } 3527 3528 snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port); 3529 return 0; 3530 } 3531 3532 static int 3533 nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair, 3534 struct spdk_nvme_transport_id *trid) 3535 { 3536 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 3537 } 3538 3539 static int 3540 nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair, 3541 struct spdk_nvme_transport_id *trid) 3542 { 3543 return nvmf_tcp_qpair_get_trid(qpair, trid, 1); 3544 } 3545 3546 static int 3547 nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair, 3548 struct spdk_nvme_transport_id *trid) 3549 { 3550 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 3551 } 3552 3553 static void 3554 nvmf_tcp_req_set_abort_status(struct spdk_nvmf_request *req, 3555 struct spdk_nvmf_tcp_req *tcp_req_to_abort) 3556 { 3557 nvmf_tcp_req_set_cpl(tcp_req_to_abort, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_ABORTED_BY_REQUEST); 3558 nvmf_tcp_req_set_state(tcp_req_to_abort, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3559 3560 req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command was successfully aborted. */ 3561 } 3562 3563 static int 3564 _nvmf_tcp_qpair_abort_request(void *ctx) 3565 { 3566 struct spdk_nvmf_request *req = ctx; 3567 struct spdk_nvmf_tcp_req *tcp_req_to_abort = SPDK_CONTAINEROF(req->req_to_abort, 3568 struct spdk_nvmf_tcp_req, req); 3569 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(req->req_to_abort->qpair, 3570 struct spdk_nvmf_tcp_qpair, qpair); 3571 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 3572 struct spdk_nvmf_tcp_transport, transport); 3573 int rc; 3574 3575 spdk_poller_unregister(&req->poller); 3576 3577 switch (tcp_req_to_abort->state) { 3578 case TCP_REQUEST_STATE_EXECUTING: 3579 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3580 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3581 rc = nvmf_ctrlr_abort_request(req); 3582 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS) { 3583 return SPDK_POLLER_BUSY; 3584 } 3585 break; 3586 3587 case TCP_REQUEST_STATE_NEED_BUFFER: 3588 nvmf_tcp_request_get_buffers_abort(tcp_req_to_abort); 3589 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 3590 nvmf_tcp_req_process(ttransport, tcp_req_to_abort); 3591 break; 3592 3593 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 3594 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 3595 if (spdk_get_ticks() < req->timeout_tsc) { 3596 req->poller = SPDK_POLLER_REGISTER(_nvmf_tcp_qpair_abort_request, req, 0); 3597 return SPDK_POLLER_BUSY; 3598 } 3599 break; 3600 3601 default: 3602 /* Requests in other states are either un-abortable (e.g. 3603 * TRANSFERRING_CONTROLLER_TO_HOST) or should never end up here, as they're 3604 * immediately transitioned to other states in nvmf_tcp_req_process() (e.g. 3605 * READY_TO_EXECUTE). But it is fine to end up here, as we'll simply complete the 3606 * abort request with the bit0 of dword0 set (command not aborted). 3607 */ 3608 break; 3609 } 3610 3611 spdk_nvmf_request_complete(req); 3612 return SPDK_POLLER_BUSY; 3613 } 3614 3615 static void 3616 nvmf_tcp_qpair_abort_request(struct spdk_nvmf_qpair *qpair, 3617 struct spdk_nvmf_request *req) 3618 { 3619 struct spdk_nvmf_tcp_qpair *tqpair; 3620 struct spdk_nvmf_tcp_transport *ttransport; 3621 struct spdk_nvmf_transport *transport; 3622 uint16_t cid; 3623 uint32_t i; 3624 struct spdk_nvmf_tcp_req *tcp_req_to_abort = NULL; 3625 3626 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3627 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 3628 transport = &ttransport->transport; 3629 3630 cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid; 3631 3632 for (i = 0; i < tqpair->resource_count; i++) { 3633 if (tqpair->reqs[i].state != TCP_REQUEST_STATE_FREE && 3634 tqpair->reqs[i].req.cmd->nvme_cmd.cid == cid) { 3635 tcp_req_to_abort = &tqpair->reqs[i]; 3636 break; 3637 } 3638 } 3639 3640 spdk_trace_record(TRACE_TCP_QP_ABORT_REQ, tqpair->qpair.trace_id, 0, (uintptr_t)req); 3641 3642 if (tcp_req_to_abort == NULL) { 3643 spdk_nvmf_request_complete(req); 3644 return; 3645 } 3646 3647 req->req_to_abort = &tcp_req_to_abort->req; 3648 req->timeout_tsc = spdk_get_ticks() + 3649 transport->opts.abort_timeout_sec * spdk_get_ticks_hz(); 3650 req->poller = NULL; 3651 3652 _nvmf_tcp_qpair_abort_request(req); 3653 } 3654 3655 struct tcp_subsystem_add_host_opts { 3656 char *psk; 3657 }; 3658 3659 static const struct spdk_json_object_decoder tcp_subsystem_add_host_opts_decoder[] = { 3660 {"psk", offsetof(struct tcp_subsystem_add_host_opts, psk), spdk_json_decode_string, true}, 3661 }; 3662 3663 static int 3664 tcp_load_psk(const char *fname, char *buf, size_t bufsz) 3665 { 3666 FILE *psk_file; 3667 struct stat statbuf; 3668 int rc; 3669 3670 if (stat(fname, &statbuf) != 0) { 3671 SPDK_ERRLOG("Could not read permissions for PSK file\n"); 3672 return -EACCES; 3673 } 3674 3675 if ((statbuf.st_mode & TCP_PSK_INVALID_PERMISSIONS) != 0) { 3676 SPDK_ERRLOG("Incorrect permissions for PSK file\n"); 3677 return -EPERM; 3678 } 3679 if ((size_t)statbuf.st_size > bufsz) { 3680 SPDK_ERRLOG("Invalid PSK: too long\n"); 3681 return -EINVAL; 3682 } 3683 psk_file = fopen(fname, "r"); 3684 if (psk_file == NULL) { 3685 SPDK_ERRLOG("Could not open PSK file\n"); 3686 return -EINVAL; 3687 } 3688 3689 rc = fread(buf, 1, statbuf.st_size, psk_file); 3690 if (rc != statbuf.st_size) { 3691 SPDK_ERRLOG("Failed to read PSK\n"); 3692 fclose(psk_file); 3693 return -EINVAL; 3694 } 3695 3696 fclose(psk_file); 3697 return 0; 3698 } 3699 3700 SPDK_LOG_DEPRECATION_REGISTER(nvmf_tcp_psk_path, "PSK path", "v24.09", 0); 3701 3702 static int 3703 nvmf_tcp_subsystem_add_host(struct spdk_nvmf_transport *transport, 3704 const struct spdk_nvmf_subsystem *subsystem, 3705 const char *hostnqn, 3706 const struct spdk_json_val *transport_specific) 3707 { 3708 struct tcp_subsystem_add_host_opts opts; 3709 struct spdk_nvmf_tcp_transport *ttransport; 3710 struct tcp_psk_entry *tmp, *entry = NULL; 3711 uint8_t psk_configured[SPDK_TLS_PSK_MAX_LEN] = {}; 3712 char psk_interchange[SPDK_TLS_PSK_MAX_LEN + 1] = {}; 3713 uint8_t tls_cipher_suite; 3714 int rc = 0; 3715 uint8_t psk_retained_hash; 3716 uint64_t psk_configured_size; 3717 3718 if (transport_specific == NULL) { 3719 return 0; 3720 } 3721 3722 assert(transport != NULL); 3723 assert(subsystem != NULL); 3724 3725 memset(&opts, 0, sizeof(opts)); 3726 3727 /* Decode PSK (either name of a key or file path) */ 3728 if (spdk_json_decode_object_relaxed(transport_specific, tcp_subsystem_add_host_opts_decoder, 3729 SPDK_COUNTOF(tcp_subsystem_add_host_opts_decoder), &opts)) { 3730 SPDK_ERRLOG("spdk_json_decode_object failed\n"); 3731 return -EINVAL; 3732 } 3733 3734 if (opts.psk == NULL) { 3735 return 0; 3736 } 3737 3738 entry = calloc(1, sizeof(struct tcp_psk_entry)); 3739 if (entry == NULL) { 3740 SPDK_ERRLOG("Unable to allocate memory for PSK entry!\n"); 3741 rc = -ENOMEM; 3742 goto end; 3743 } 3744 3745 entry->key = spdk_keyring_get_key(opts.psk); 3746 if (entry->key != NULL) { 3747 rc = spdk_key_get_key(entry->key, psk_interchange, SPDK_TLS_PSK_MAX_LEN); 3748 if (rc < 0) { 3749 SPDK_ERRLOG("Failed to retrieve PSK '%s'\n", opts.psk); 3750 rc = -EINVAL; 3751 goto end; 3752 } 3753 } else { 3754 if (strlen(opts.psk) >= sizeof(entry->psk)) { 3755 SPDK_ERRLOG("PSK path too long\n"); 3756 rc = -EINVAL; 3757 goto end; 3758 } 3759 3760 rc = tcp_load_psk(opts.psk, psk_interchange, SPDK_TLS_PSK_MAX_LEN); 3761 if (rc) { 3762 SPDK_ERRLOG("Could not retrieve PSK from file\n"); 3763 goto end; 3764 } 3765 3766 SPDK_LOG_DEPRECATED(nvmf_tcp_psk_path); 3767 } 3768 3769 /* Parse PSK interchange to get length of base64 encoded data. 3770 * This is then used to decide which cipher suite should be used 3771 * to generate PSK identity and TLS PSK later on. */ 3772 rc = nvme_tcp_parse_interchange_psk(psk_interchange, psk_configured, sizeof(psk_configured), 3773 &psk_configured_size, &psk_retained_hash); 3774 if (rc < 0) { 3775 SPDK_ERRLOG("Failed to parse PSK interchange!\n"); 3776 goto end; 3777 } 3778 3779 /* The Base64 string encodes the configured PSK (32 or 48 bytes binary). 3780 * This check also ensures that psk_configured_size is smaller than 3781 * psk_retained buffer size. */ 3782 if (psk_configured_size == SHA256_DIGEST_LENGTH) { 3783 tls_cipher_suite = NVME_TCP_CIPHER_AES_128_GCM_SHA256; 3784 } else if (psk_configured_size == SHA384_DIGEST_LENGTH) { 3785 tls_cipher_suite = NVME_TCP_CIPHER_AES_256_GCM_SHA384; 3786 } else { 3787 SPDK_ERRLOG("Unrecognized cipher suite!\n"); 3788 rc = -EINVAL; 3789 goto end; 3790 } 3791 3792 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 3793 /* Generate PSK identity. */ 3794 rc = nvme_tcp_generate_psk_identity(entry->pskid, sizeof(entry->pskid), hostnqn, 3795 subsystem->subnqn, tls_cipher_suite); 3796 if (rc) { 3797 rc = -EINVAL; 3798 goto end; 3799 } 3800 /* Check if PSK identity entry already exists. */ 3801 TAILQ_FOREACH(tmp, &ttransport->psks, link) { 3802 if (strncmp(tmp->pskid, entry->pskid, NVMF_PSK_IDENTITY_LEN) == 0) { 3803 SPDK_ERRLOG("Given PSK identity: %s entry already exists!\n", entry->pskid); 3804 rc = -EEXIST; 3805 goto end; 3806 } 3807 } 3808 3809 if (snprintf(entry->hostnqn, sizeof(entry->hostnqn), "%s", hostnqn) < 0) { 3810 SPDK_ERRLOG("Could not write hostnqn string!\n"); 3811 rc = -EINVAL; 3812 goto end; 3813 } 3814 if (snprintf(entry->subnqn, sizeof(entry->subnqn), "%s", subsystem->subnqn) < 0) { 3815 SPDK_ERRLOG("Could not write subnqn string!\n"); 3816 rc = -EINVAL; 3817 goto end; 3818 } 3819 3820 entry->tls_cipher_suite = tls_cipher_suite; 3821 3822 /* No hash indicates that Configured PSK must be used as Retained PSK. */ 3823 if (psk_retained_hash == NVME_TCP_HASH_ALGORITHM_NONE) { 3824 /* Psk configured is either 32 or 48 bytes long. */ 3825 memcpy(entry->psk, psk_configured, psk_configured_size); 3826 entry->psk_size = psk_configured_size; 3827 } else { 3828 /* Derive retained PSK. */ 3829 rc = nvme_tcp_derive_retained_psk(psk_configured, psk_configured_size, hostnqn, entry->psk, 3830 SPDK_TLS_PSK_MAX_LEN, psk_retained_hash); 3831 if (rc < 0) { 3832 SPDK_ERRLOG("Unable to derive retained PSK!\n"); 3833 goto end; 3834 } 3835 entry->psk_size = rc; 3836 } 3837 3838 if (entry->key == NULL) { 3839 rc = snprintf(entry->psk_path, sizeof(entry->psk_path), "%s", opts.psk); 3840 if (rc < 0 || (size_t)rc >= sizeof(entry->psk_path)) { 3841 SPDK_ERRLOG("Could not save PSK path!\n"); 3842 rc = -ENAMETOOLONG; 3843 goto end; 3844 } 3845 } 3846 3847 TAILQ_INSERT_TAIL(&ttransport->psks, entry, link); 3848 rc = 0; 3849 3850 end: 3851 spdk_memset_s(psk_configured, sizeof(psk_configured), 0, sizeof(psk_configured)); 3852 spdk_memset_s(psk_interchange, sizeof(psk_interchange), 0, sizeof(psk_interchange)); 3853 3854 free(opts.psk); 3855 if (rc != 0) { 3856 nvmf_tcp_free_psk_entry(entry); 3857 } 3858 3859 return rc; 3860 } 3861 3862 static void 3863 nvmf_tcp_subsystem_remove_host(struct spdk_nvmf_transport *transport, 3864 const struct spdk_nvmf_subsystem *subsystem, 3865 const char *hostnqn) 3866 { 3867 struct spdk_nvmf_tcp_transport *ttransport; 3868 struct tcp_psk_entry *entry, *tmp; 3869 3870 assert(transport != NULL); 3871 assert(subsystem != NULL); 3872 3873 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 3874 TAILQ_FOREACH_SAFE(entry, &ttransport->psks, link, tmp) { 3875 if ((strncmp(entry->hostnqn, hostnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0 && 3876 (strncmp(entry->subnqn, subsystem->subnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0) { 3877 TAILQ_REMOVE(&ttransport->psks, entry, link); 3878 nvmf_tcp_free_psk_entry(entry); 3879 break; 3880 } 3881 } 3882 } 3883 3884 static void 3885 nvmf_tcp_subsystem_dump_host(struct spdk_nvmf_transport *transport, 3886 const struct spdk_nvmf_subsystem *subsystem, const char *hostnqn, 3887 struct spdk_json_write_ctx *w) 3888 { 3889 struct spdk_nvmf_tcp_transport *ttransport; 3890 struct tcp_psk_entry *entry; 3891 3892 assert(transport != NULL); 3893 assert(subsystem != NULL); 3894 3895 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 3896 TAILQ_FOREACH(entry, &ttransport->psks, link) { 3897 if ((strncmp(entry->hostnqn, hostnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0 && 3898 (strncmp(entry->subnqn, subsystem->subnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0) { 3899 spdk_json_write_named_string(w, "psk", entry->key ? 3900 spdk_key_get_name(entry->key) : entry->psk_path); 3901 break; 3902 } 3903 } 3904 } 3905 3906 static void 3907 nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts) 3908 { 3909 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH; 3910 opts->max_qpairs_per_ctrlr = SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR; 3911 opts->in_capsule_data_size = SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE; 3912 opts->max_io_size = SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE; 3913 opts->io_unit_size = SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE; 3914 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH; 3915 opts->num_shared_buffers = SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS; 3916 opts->buf_cache_size = SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE; 3917 opts->dif_insert_or_strip = SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP; 3918 opts->abort_timeout_sec = SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC; 3919 opts->transport_specific = NULL; 3920 } 3921 3922 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = { 3923 .name = "TCP", 3924 .type = SPDK_NVME_TRANSPORT_TCP, 3925 .opts_init = nvmf_tcp_opts_init, 3926 .create = nvmf_tcp_create, 3927 .dump_opts = nvmf_tcp_dump_opts, 3928 .destroy = nvmf_tcp_destroy, 3929 3930 .listen = nvmf_tcp_listen, 3931 .stop_listen = nvmf_tcp_stop_listen, 3932 3933 .listener_discover = nvmf_tcp_discover, 3934 3935 .poll_group_create = nvmf_tcp_poll_group_create, 3936 .get_optimal_poll_group = nvmf_tcp_get_optimal_poll_group, 3937 .poll_group_destroy = nvmf_tcp_poll_group_destroy, 3938 .poll_group_add = nvmf_tcp_poll_group_add, 3939 .poll_group_remove = nvmf_tcp_poll_group_remove, 3940 .poll_group_poll = nvmf_tcp_poll_group_poll, 3941 3942 .req_free = nvmf_tcp_req_free, 3943 .req_complete = nvmf_tcp_req_complete, 3944 .req_get_buffers_done = nvmf_tcp_req_get_buffers_done, 3945 3946 .qpair_fini = nvmf_tcp_close_qpair, 3947 .qpair_get_local_trid = nvmf_tcp_qpair_get_local_trid, 3948 .qpair_get_peer_trid = nvmf_tcp_qpair_get_peer_trid, 3949 .qpair_get_listen_trid = nvmf_tcp_qpair_get_listen_trid, 3950 .qpair_abort_request = nvmf_tcp_qpair_abort_request, 3951 .subsystem_add_host = nvmf_tcp_subsystem_add_host, 3952 .subsystem_remove_host = nvmf_tcp_subsystem_remove_host, 3953 .subsystem_dump_host = nvmf_tcp_subsystem_dump_host, 3954 }; 3955 3956 SPDK_NVMF_TRANSPORT_REGISTER(tcp, &spdk_nvmf_transport_tcp); 3957 SPDK_LOG_REGISTER_COMPONENT(nvmf_tcp) 3958