1 /*- 2 * BSD LICENSE 3 * 4 * Copyright (c) Intel Corporation. All rights reserved. 5 * Copyright (c) 2019, 2020 Mellanox Technologies LTD. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * * Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * * Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * * Neither the name of Intel Corporation nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 #include "spdk/stdinc.h" 35 #include "spdk/crc32.h" 36 #include "spdk/endian.h" 37 #include "spdk/assert.h" 38 #include "spdk/thread.h" 39 #include "spdk/nvmf_transport.h" 40 #include "spdk/string.h" 41 #include "spdk/trace.h" 42 #include "spdk/util.h" 43 #include "spdk/log.h" 44 45 #include "spdk_internal/assert.h" 46 #include "spdk_internal/nvme_tcp.h" 47 #include "spdk_internal/sock.h" 48 49 #include "nvmf_internal.h" 50 51 #define NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME 16 52 #define SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY 16 53 #define SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY 0 54 #define SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM 32 55 #define SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION true 56 57 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp; 58 59 /* spdk nvmf related structure */ 60 enum spdk_nvmf_tcp_req_state { 61 62 /* The request is not currently in use */ 63 TCP_REQUEST_STATE_FREE = 0, 64 65 /* Initial state when request first received */ 66 TCP_REQUEST_STATE_NEW, 67 68 /* The request is queued until a data buffer is available. */ 69 TCP_REQUEST_STATE_NEED_BUFFER, 70 71 /* The request is currently transferring data from the host to the controller. */ 72 TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 73 74 /* The request is waiting for the R2T send acknowledgement. */ 75 TCP_REQUEST_STATE_AWAITING_R2T_ACK, 76 77 /* The request is ready to execute at the block device */ 78 TCP_REQUEST_STATE_READY_TO_EXECUTE, 79 80 /* The request is currently executing at the block device */ 81 TCP_REQUEST_STATE_EXECUTING, 82 83 /* The request finished executing at the block device */ 84 TCP_REQUEST_STATE_EXECUTED, 85 86 /* The request is ready to send a completion */ 87 TCP_REQUEST_STATE_READY_TO_COMPLETE, 88 89 /* The request is currently transferring final pdus from the controller to the host. */ 90 TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 91 92 /* The request completed and can be marked free. */ 93 TCP_REQUEST_STATE_COMPLETED, 94 95 /* Terminator */ 96 TCP_REQUEST_NUM_STATES, 97 }; 98 99 static const char *spdk_nvmf_tcp_term_req_fes_str[] = { 100 "Invalid PDU Header Field", 101 "PDU Sequence Error", 102 "Header Digiest Error", 103 "Data Transfer Out of Range", 104 "R2T Limit Exceeded", 105 "Unsupported parameter", 106 }; 107 108 #define OBJECT_NVMF_TCP_IO 0x80 109 110 #define TRACE_GROUP_NVMF_TCP 0x5 111 #define TRACE_TCP_REQUEST_STATE_NEW SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x0) 112 #define TRACE_TCP_REQUEST_STATE_NEED_BUFFER SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x1) 113 #define TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x2) 114 #define TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x3) 115 #define TRACE_TCP_REQUEST_STATE_EXECUTING SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x4) 116 #define TRACE_TCP_REQUEST_STATE_EXECUTED SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x5) 117 #define TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x6) 118 #define TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x7) 119 #define TRACE_TCP_REQUEST_STATE_COMPLETED SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x8) 120 #define TRACE_TCP_FLUSH_WRITEBUF_START SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x9) 121 #define TRACE_TCP_FLUSH_WRITEBUF_DONE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xA) 122 #define TRACE_TCP_READ_FROM_SOCKET_DONE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xB) 123 #define TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xC) 124 125 SPDK_TRACE_REGISTER_FN(nvmf_tcp_trace, "nvmf_tcp", TRACE_GROUP_NVMF_TCP) 126 { 127 spdk_trace_register_object(OBJECT_NVMF_TCP_IO, 'r'); 128 spdk_trace_register_description("TCP_REQ_NEW", 129 TRACE_TCP_REQUEST_STATE_NEW, 130 OWNER_NONE, OBJECT_NVMF_TCP_IO, 1, 1, ""); 131 spdk_trace_register_description("TCP_REQ_NEED_BUFFER", 132 TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 133 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 134 spdk_trace_register_description("TCP_REQ_TX_H_TO_C", 135 TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 136 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 137 spdk_trace_register_description("TCP_REQ_RDY_TO_EXECUTE", 138 TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 139 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 140 spdk_trace_register_description("TCP_REQ_EXECUTING", 141 TRACE_TCP_REQUEST_STATE_EXECUTING, 142 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 143 spdk_trace_register_description("TCP_REQ_EXECUTED", 144 TRACE_TCP_REQUEST_STATE_EXECUTED, 145 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 146 spdk_trace_register_description("TCP_REQ_RDY_TO_COMPLETE", 147 TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 148 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 149 spdk_trace_register_description("TCP_REQ_TRANSFER_C2H", 150 TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 151 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 152 spdk_trace_register_description("TCP_REQ_COMPLETED", 153 TRACE_TCP_REQUEST_STATE_COMPLETED, 154 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 155 spdk_trace_register_description("TCP_WRITE_START", 156 TRACE_TCP_FLUSH_WRITEBUF_START, 157 OWNER_NONE, OBJECT_NONE, 0, 0, ""); 158 spdk_trace_register_description("TCP_WRITE_DONE", 159 TRACE_TCP_FLUSH_WRITEBUF_DONE, 160 OWNER_NONE, OBJECT_NONE, 0, 0, ""); 161 spdk_trace_register_description("TCP_READ_DONE", 162 TRACE_TCP_READ_FROM_SOCKET_DONE, 163 OWNER_NONE, OBJECT_NONE, 0, 0, ""); 164 spdk_trace_register_description("TCP_REQ_AWAIT_R2T_ACK", 165 TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 166 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 167 } 168 169 struct spdk_nvmf_tcp_req { 170 struct spdk_nvmf_request req; 171 struct spdk_nvme_cpl rsp; 172 struct spdk_nvme_cmd cmd; 173 174 /* A PDU that can be used for sending responses. This is 175 * not the incoming PDU! */ 176 struct nvme_tcp_pdu *pdu; 177 178 /* In-capsule data buffer */ 179 uint8_t *buf; 180 /* 181 * The PDU for a request may be used multiple times in serial over 182 * the request's lifetime. For example, first to send an R2T, then 183 * to send a completion. To catch mistakes where the PDU is used 184 * twice at the same time, add a debug flag here for init/fini. 185 */ 186 bool pdu_in_use; 187 bool has_incapsule_data; 188 189 /* transfer_tag */ 190 uint16_t ttag; 191 192 enum spdk_nvmf_tcp_req_state state; 193 194 /* 195 * h2c_offset is used when we receive the h2c_data PDU. 196 */ 197 uint32_t h2c_offset; 198 199 STAILQ_ENTRY(spdk_nvmf_tcp_req) link; 200 TAILQ_ENTRY(spdk_nvmf_tcp_req) state_link; 201 }; 202 203 struct spdk_nvmf_tcp_qpair { 204 struct spdk_nvmf_qpair qpair; 205 struct spdk_nvmf_tcp_poll_group *group; 206 struct spdk_sock *sock; 207 208 enum nvme_tcp_pdu_recv_state recv_state; 209 enum nvme_tcp_qpair_state state; 210 211 /* PDU being actively received */ 212 struct nvme_tcp_pdu pdu_in_progress; 213 214 /* Queues to track the requests in all states */ 215 TAILQ_HEAD(, spdk_nvmf_tcp_req) state_queue[TCP_REQUEST_NUM_STATES]; 216 /* Number of requests in each state */ 217 uint32_t state_cntr[TCP_REQUEST_NUM_STATES]; 218 219 uint8_t cpda; 220 221 bool host_hdgst_enable; 222 bool host_ddgst_enable; 223 224 TAILQ_HEAD(, nvme_tcp_pdu) send_queue; 225 226 /* This is a spare PDU used for sending special management 227 * operations. Primarily, this is used for the initial 228 * connection response and c2h termination request. */ 229 struct nvme_tcp_pdu *mgmt_pdu; 230 231 /* Arrays of in-capsule buffers, requests, and pdus. 232 * Each array is 'resource_count' number of elements */ 233 void *bufs; 234 struct spdk_nvmf_tcp_req *reqs; 235 struct nvme_tcp_pdu *pdus; 236 uint32_t resource_count; 237 uint32_t recv_buf_size; 238 239 struct spdk_nvmf_tcp_port *port; 240 241 /* IP address */ 242 char initiator_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 243 char target_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 244 245 /* IP port */ 246 uint16_t initiator_port; 247 uint16_t target_port; 248 249 /* Timer used to destroy qpair after detecting transport error issue if initiator does 250 * not close the connection. 251 */ 252 struct spdk_poller *timeout_poller; 253 254 TAILQ_ENTRY(spdk_nvmf_tcp_qpair) link; 255 }; 256 257 struct spdk_nvmf_tcp_control_msg { 258 STAILQ_ENTRY(spdk_nvmf_tcp_control_msg) link; 259 }; 260 261 struct spdk_nvmf_tcp_control_msg_list { 262 void *msg_buf; 263 STAILQ_HEAD(, spdk_nvmf_tcp_control_msg) free_msgs; 264 }; 265 266 struct spdk_nvmf_tcp_poll_group { 267 struct spdk_nvmf_transport_poll_group group; 268 struct spdk_sock_group *sock_group; 269 270 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) qpairs; 271 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) await_req; 272 273 struct spdk_nvmf_tcp_control_msg_list *control_msg_list; 274 }; 275 276 struct spdk_nvmf_tcp_port { 277 const struct spdk_nvme_transport_id *trid; 278 struct spdk_sock *listen_sock; 279 TAILQ_ENTRY(spdk_nvmf_tcp_port) link; 280 }; 281 282 struct tcp_transport_opts { 283 bool c2h_success; 284 uint16_t control_msg_num; 285 uint32_t sock_priority; 286 }; 287 288 struct spdk_nvmf_tcp_transport { 289 struct spdk_nvmf_transport transport; 290 struct tcp_transport_opts tcp_opts; 291 292 pthread_mutex_t lock; 293 294 TAILQ_HEAD(, spdk_nvmf_tcp_port) ports; 295 }; 296 297 static const struct spdk_json_object_decoder tcp_transport_opts_decoder[] = { 298 { 299 "c2h_success", offsetof(struct tcp_transport_opts, c2h_success), 300 spdk_json_decode_bool, true 301 }, 302 { 303 "control_msg_num", offsetof(struct tcp_transport_opts, control_msg_num), 304 spdk_json_decode_uint16, true 305 }, 306 { 307 "sock_priority", offsetof(struct tcp_transport_opts, sock_priority), 308 spdk_json_decode_uint32, true 309 }, 310 }; 311 312 static bool nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 313 struct spdk_nvmf_tcp_req *tcp_req); 314 static void nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group); 315 316 static void 317 nvmf_tcp_req_set_state(struct spdk_nvmf_tcp_req *tcp_req, 318 enum spdk_nvmf_tcp_req_state state) 319 { 320 struct spdk_nvmf_qpair *qpair; 321 struct spdk_nvmf_tcp_qpair *tqpair; 322 323 qpair = tcp_req->req.qpair; 324 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 325 326 TAILQ_REMOVE(&tqpair->state_queue[tcp_req->state], tcp_req, state_link); 327 assert(tqpair->state_cntr[tcp_req->state] > 0); 328 tqpair->state_cntr[tcp_req->state]--; 329 330 TAILQ_INSERT_TAIL(&tqpair->state_queue[state], tcp_req, state_link); 331 tqpair->state_cntr[state]++; 332 333 tcp_req->state = state; 334 } 335 336 static inline struct nvme_tcp_pdu * 337 nvmf_tcp_req_pdu_init(struct spdk_nvmf_tcp_req *tcp_req) 338 { 339 assert(tcp_req->pdu_in_use == false); 340 tcp_req->pdu_in_use = true; 341 342 memset(tcp_req->pdu, 0, sizeof(*tcp_req->pdu)); 343 tcp_req->pdu->qpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 344 345 return tcp_req->pdu; 346 } 347 348 static inline void 349 nvmf_tcp_req_pdu_fini(struct spdk_nvmf_tcp_req *tcp_req) 350 { 351 tcp_req->pdu_in_use = false; 352 } 353 354 static struct spdk_nvmf_tcp_req * 355 nvmf_tcp_req_get(struct spdk_nvmf_tcp_qpair *tqpair) 356 { 357 struct spdk_nvmf_tcp_req *tcp_req; 358 359 tcp_req = TAILQ_FIRST(&tqpair->state_queue[TCP_REQUEST_STATE_FREE]); 360 if (!tcp_req) { 361 return NULL; 362 } 363 364 memset(&tcp_req->rsp, 0, sizeof(tcp_req->rsp)); 365 tcp_req->h2c_offset = 0; 366 tcp_req->has_incapsule_data = false; 367 tcp_req->req.dif.dif_insert_or_strip = false; 368 369 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW); 370 return tcp_req; 371 } 372 373 static void 374 nvmf_tcp_request_free(struct spdk_nvmf_tcp_req *tcp_req) 375 { 376 struct spdk_nvmf_tcp_transport *ttransport; 377 378 assert(tcp_req != NULL); 379 380 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req=%p will be freed\n", tcp_req); 381 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 382 struct spdk_nvmf_tcp_transport, transport); 383 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 384 nvmf_tcp_req_process(ttransport, tcp_req); 385 } 386 387 static int 388 nvmf_tcp_req_free(struct spdk_nvmf_request *req) 389 { 390 struct spdk_nvmf_tcp_req *tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 391 392 nvmf_tcp_request_free(tcp_req); 393 394 return 0; 395 } 396 397 static void 398 nvmf_tcp_drain_state_queue(struct spdk_nvmf_tcp_qpair *tqpair, 399 enum spdk_nvmf_tcp_req_state state) 400 { 401 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 402 403 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->state_queue[state], state_link, req_tmp) { 404 nvmf_tcp_request_free(tcp_req); 405 } 406 } 407 408 static void 409 nvmf_tcp_cleanup_all_states(struct spdk_nvmf_tcp_qpair *tqpair) 410 { 411 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 412 413 assert(TAILQ_EMPTY(&tqpair->send_queue)); 414 415 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 416 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEW); 417 418 /* Wipe the requests waiting for buffer from the global list */ 419 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->state_queue[TCP_REQUEST_STATE_NEED_BUFFER], state_link, 420 req_tmp) { 421 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, &tcp_req->req, 422 spdk_nvmf_request, buf_link); 423 } 424 425 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEED_BUFFER); 426 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_EXECUTING); 427 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 428 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 429 } 430 431 static void 432 nvmf_tcp_dump_qpair_req_contents(struct spdk_nvmf_tcp_qpair *tqpair) 433 { 434 int i; 435 struct spdk_nvmf_tcp_req *tcp_req; 436 437 SPDK_ERRLOG("Dumping contents of queue pair (QID %d)\n", tqpair->qpair.qid); 438 for (i = 1; i < TCP_REQUEST_NUM_STATES; i++) { 439 SPDK_ERRLOG("\tNum of requests in state[%d] = %u\n", i, tqpair->state_cntr[i]); 440 TAILQ_FOREACH(tcp_req, &tqpair->state_queue[i], state_link) { 441 SPDK_ERRLOG("\t\tRequest Data From Pool: %d\n", tcp_req->req.data_from_pool); 442 SPDK_ERRLOG("\t\tRequest opcode: %d\n", tcp_req->req.cmd->nvmf_cmd.opcode); 443 } 444 } 445 } 446 447 static void 448 nvmf_tcp_qpair_destroy(struct spdk_nvmf_tcp_qpair *tqpair) 449 { 450 int err = 0; 451 452 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 453 454 err = spdk_sock_close(&tqpair->sock); 455 assert(err == 0); 456 nvmf_tcp_cleanup_all_states(tqpair); 457 458 if (tqpair->state_cntr[TCP_REQUEST_STATE_FREE] != tqpair->resource_count) { 459 SPDK_ERRLOG("tqpair(%p) free tcp request num is %u but should be %u\n", tqpair, 460 tqpair->state_cntr[TCP_REQUEST_STATE_FREE], 461 tqpair->resource_count); 462 err++; 463 } 464 465 if (err > 0) { 466 nvmf_tcp_dump_qpair_req_contents(tqpair); 467 } 468 469 spdk_dma_free(tqpair->pdus); 470 free(tqpair->reqs); 471 spdk_free(tqpair->bufs); 472 free(tqpair); 473 SPDK_DEBUGLOG(nvmf_tcp, "Leave\n"); 474 } 475 476 static void 477 nvmf_tcp_dump_opts(struct spdk_nvmf_transport *transport, struct spdk_json_write_ctx *w) 478 { 479 struct spdk_nvmf_tcp_transport *ttransport; 480 assert(w != NULL); 481 482 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 483 spdk_json_write_named_bool(w, "c2h_success", ttransport->tcp_opts.c2h_success); 484 spdk_json_write_named_uint32(w, "sock_priority", ttransport->tcp_opts.sock_priority); 485 } 486 487 static int 488 nvmf_tcp_destroy(struct spdk_nvmf_transport *transport) 489 { 490 struct spdk_nvmf_tcp_transport *ttransport; 491 492 assert(transport != NULL); 493 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 494 495 pthread_mutex_destroy(&ttransport->lock); 496 free(ttransport); 497 return 0; 498 } 499 500 static struct spdk_nvmf_transport * 501 nvmf_tcp_create(struct spdk_nvmf_transport_opts *opts) 502 { 503 struct spdk_nvmf_tcp_transport *ttransport; 504 uint32_t sge_count; 505 uint32_t min_shared_buffers; 506 507 ttransport = calloc(1, sizeof(*ttransport)); 508 if (!ttransport) { 509 return NULL; 510 } 511 512 TAILQ_INIT(&ttransport->ports); 513 514 ttransport->transport.ops = &spdk_nvmf_transport_tcp; 515 516 ttransport->tcp_opts.c2h_success = SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION; 517 ttransport->tcp_opts.sock_priority = SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY; 518 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 519 if (opts->transport_specific != NULL && 520 spdk_json_decode_object_relaxed(opts->transport_specific, tcp_transport_opts_decoder, 521 SPDK_COUNTOF(tcp_transport_opts_decoder), 522 &ttransport->tcp_opts)) { 523 SPDK_ERRLOG("spdk_json_decode_object_relaxed failed\n"); 524 free(ttransport); 525 return NULL; 526 } 527 528 SPDK_NOTICELOG("*** TCP Transport Init ***\n"); 529 530 SPDK_INFOLOG(nvmf_tcp, "*** TCP Transport Init ***\n" 531 " Transport opts: max_ioq_depth=%d, max_io_size=%d,\n" 532 " max_io_qpairs_per_ctrlr=%d, io_unit_size=%d,\n" 533 " in_capsule_data_size=%d, max_aq_depth=%d\n" 534 " num_shared_buffers=%d, c2h_success=%d,\n" 535 " dif_insert_or_strip=%d, sock_priority=%d\n" 536 " abort_timeout_sec=%d, control_msg_num=%hu\n", 537 opts->max_queue_depth, 538 opts->max_io_size, 539 opts->max_qpairs_per_ctrlr - 1, 540 opts->io_unit_size, 541 opts->in_capsule_data_size, 542 opts->max_aq_depth, 543 opts->num_shared_buffers, 544 ttransport->tcp_opts.c2h_success, 545 opts->dif_insert_or_strip, 546 ttransport->tcp_opts.sock_priority, 547 opts->abort_timeout_sec, 548 ttransport->tcp_opts.control_msg_num); 549 550 if (ttransport->tcp_opts.sock_priority > SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY) { 551 SPDK_ERRLOG("Unsupported socket_priority=%d, the current range is: 0 to %d\n" 552 "you can use man 7 socket to view the range of priority under SO_PRIORITY item\n", 553 ttransport->tcp_opts.sock_priority, SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY); 554 free(ttransport); 555 return NULL; 556 } 557 558 if (ttransport->tcp_opts.control_msg_num == 0 && 559 opts->in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 560 SPDK_WARNLOG("TCP param control_msg_num can't be 0 if ICD is less than %u bytes. Using default value %u\n", 561 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM); 562 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 563 } 564 565 /* I/O unit size cannot be larger than max I/O size */ 566 if (opts->io_unit_size > opts->max_io_size) { 567 opts->io_unit_size = opts->max_io_size; 568 } 569 570 sge_count = opts->max_io_size / opts->io_unit_size; 571 if (sge_count > SPDK_NVMF_MAX_SGL_ENTRIES) { 572 SPDK_ERRLOG("Unsupported IO Unit size specified, %d bytes\n", opts->io_unit_size); 573 free(ttransport); 574 return NULL; 575 } 576 577 min_shared_buffers = spdk_env_get_core_count() * opts->buf_cache_size; 578 if (min_shared_buffers > opts->num_shared_buffers) { 579 SPDK_ERRLOG("There are not enough buffers to satisfy" 580 "per-poll group caches for each thread. (%" PRIu32 ")" 581 "supplied. (%" PRIu32 ") required\n", opts->num_shared_buffers, min_shared_buffers); 582 SPDK_ERRLOG("Please specify a larger number of shared buffers\n"); 583 nvmf_tcp_destroy(&ttransport->transport); 584 return NULL; 585 } 586 587 pthread_mutex_init(&ttransport->lock, NULL); 588 589 return &ttransport->transport; 590 } 591 592 static int 593 nvmf_tcp_trsvcid_to_int(const char *trsvcid) 594 { 595 unsigned long long ull; 596 char *end = NULL; 597 598 ull = strtoull(trsvcid, &end, 10); 599 if (end == NULL || end == trsvcid || *end != '\0') { 600 return -1; 601 } 602 603 /* Valid TCP/IP port numbers are in [0, 65535] */ 604 if (ull > 65535) { 605 return -1; 606 } 607 608 return (int)ull; 609 } 610 611 /** 612 * Canonicalize a listen address trid. 613 */ 614 static int 615 nvmf_tcp_canon_listen_trid(struct spdk_nvme_transport_id *canon_trid, 616 const struct spdk_nvme_transport_id *trid) 617 { 618 int trsvcid_int; 619 620 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 621 if (trsvcid_int < 0) { 622 return -EINVAL; 623 } 624 625 memset(canon_trid, 0, sizeof(*canon_trid)); 626 spdk_nvme_trid_populate_transport(canon_trid, SPDK_NVME_TRANSPORT_TCP); 627 canon_trid->adrfam = trid->adrfam; 628 snprintf(canon_trid->traddr, sizeof(canon_trid->traddr), "%s", trid->traddr); 629 snprintf(canon_trid->trsvcid, sizeof(canon_trid->trsvcid), "%d", trsvcid_int); 630 631 return 0; 632 } 633 634 /** 635 * Find an existing listening port. 636 * 637 * Caller must hold ttransport->lock. 638 */ 639 static struct spdk_nvmf_tcp_port * 640 nvmf_tcp_find_port(struct spdk_nvmf_tcp_transport *ttransport, 641 const struct spdk_nvme_transport_id *trid) 642 { 643 struct spdk_nvme_transport_id canon_trid; 644 struct spdk_nvmf_tcp_port *port; 645 646 if (nvmf_tcp_canon_listen_trid(&canon_trid, trid) != 0) { 647 return NULL; 648 } 649 650 TAILQ_FOREACH(port, &ttransport->ports, link) { 651 if (spdk_nvme_transport_id_compare(&canon_trid, port->trid) == 0) { 652 return port; 653 } 654 } 655 656 return NULL; 657 } 658 659 static int 660 nvmf_tcp_listen(struct spdk_nvmf_transport *transport, 661 const struct spdk_nvme_transport_id *trid) 662 { 663 struct spdk_nvmf_tcp_transport *ttransport; 664 struct spdk_nvmf_tcp_port *port; 665 int trsvcid_int; 666 uint8_t adrfam; 667 struct spdk_sock_opts opts; 668 669 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 670 671 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 672 if (trsvcid_int < 0) { 673 SPDK_ERRLOG("Invalid trsvcid '%s'\n", trid->trsvcid); 674 return -EINVAL; 675 } 676 677 pthread_mutex_lock(&ttransport->lock); 678 port = calloc(1, sizeof(*port)); 679 if (!port) { 680 SPDK_ERRLOG("Port allocation failed\n"); 681 pthread_mutex_unlock(&ttransport->lock); 682 return -ENOMEM; 683 } 684 685 port->trid = trid; 686 opts.opts_size = sizeof(opts); 687 spdk_sock_get_default_opts(&opts); 688 opts.priority = ttransport->tcp_opts.sock_priority; 689 port->listen_sock = spdk_sock_listen_ext(trid->traddr, trsvcid_int, 690 NULL, &opts); 691 if (port->listen_sock == NULL) { 692 SPDK_ERRLOG("spdk_sock_listen(%s, %d) failed: %s (%d)\n", 693 trid->traddr, trsvcid_int, 694 spdk_strerror(errno), errno); 695 free(port); 696 pthread_mutex_unlock(&ttransport->lock); 697 return -errno; 698 } 699 700 if (spdk_sock_is_ipv4(port->listen_sock)) { 701 adrfam = SPDK_NVMF_ADRFAM_IPV4; 702 } else if (spdk_sock_is_ipv6(port->listen_sock)) { 703 adrfam = SPDK_NVMF_ADRFAM_IPV6; 704 } else { 705 SPDK_ERRLOG("Unhandled socket type\n"); 706 adrfam = 0; 707 } 708 709 if (adrfam != trid->adrfam) { 710 SPDK_ERRLOG("Socket address family mismatch\n"); 711 spdk_sock_close(&port->listen_sock); 712 free(port); 713 pthread_mutex_unlock(&ttransport->lock); 714 return -EINVAL; 715 } 716 717 SPDK_NOTICELOG("*** NVMe/TCP Target Listening on %s port %s ***\n", 718 trid->traddr, trid->trsvcid); 719 720 TAILQ_INSERT_TAIL(&ttransport->ports, port, link); 721 pthread_mutex_unlock(&ttransport->lock); 722 return 0; 723 } 724 725 static void 726 nvmf_tcp_stop_listen(struct spdk_nvmf_transport *transport, 727 const struct spdk_nvme_transport_id *trid) 728 { 729 struct spdk_nvmf_tcp_transport *ttransport; 730 struct spdk_nvmf_tcp_port *port; 731 732 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 733 734 SPDK_DEBUGLOG(nvmf_tcp, "Removing listen address %s port %s\n", 735 trid->traddr, trid->trsvcid); 736 737 pthread_mutex_lock(&ttransport->lock); 738 port = nvmf_tcp_find_port(ttransport, trid); 739 if (port) { 740 TAILQ_REMOVE(&ttransport->ports, port, link); 741 spdk_sock_close(&port->listen_sock); 742 free(port); 743 } 744 745 pthread_mutex_unlock(&ttransport->lock); 746 } 747 748 static void nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 749 enum nvme_tcp_pdu_recv_state state); 750 751 static void 752 nvmf_tcp_qpair_disconnect(struct spdk_nvmf_tcp_qpair *tqpair) 753 { 754 SPDK_DEBUGLOG(nvmf_tcp, "Disconnecting qpair %p\n", tqpair); 755 756 if (tqpair->state <= NVME_TCP_QPAIR_STATE_RUNNING) { 757 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 758 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 759 spdk_poller_unregister(&tqpair->timeout_poller); 760 761 /* This will end up calling nvmf_tcp_close_qpair */ 762 spdk_nvmf_qpair_disconnect(&tqpair->qpair, NULL, NULL); 763 } 764 } 765 766 static void 767 _pdu_write_done(void *_pdu, int err) 768 { 769 struct nvme_tcp_pdu *pdu = _pdu; 770 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 771 772 TAILQ_REMOVE(&tqpair->send_queue, pdu, tailq); 773 774 if (err != 0) { 775 nvmf_tcp_qpair_disconnect(tqpair); 776 return; 777 } 778 779 assert(pdu->cb_fn != NULL); 780 pdu->cb_fn(pdu->cb_arg); 781 } 782 783 static void 784 nvmf_tcp_qpair_write_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 785 struct nvme_tcp_pdu *pdu, 786 nvme_tcp_qpair_xfer_complete_cb cb_fn, 787 void *cb_arg) 788 { 789 int hlen; 790 uint32_t crc32c; 791 uint32_t mapped_length = 0; 792 ssize_t rc; 793 794 assert(&tqpair->pdu_in_progress != pdu); 795 796 hlen = pdu->hdr.common.hlen; 797 798 /* Header Digest */ 799 if (g_nvme_tcp_hdgst[pdu->hdr.common.pdu_type] && tqpair->host_hdgst_enable) { 800 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 801 MAKE_DIGEST_WORD((uint8_t *)pdu->hdr.raw + hlen, crc32c); 802 } 803 804 /* Data Digest */ 805 if (pdu->data_len > 0 && g_nvme_tcp_ddgst[pdu->hdr.common.pdu_type] && tqpair->host_ddgst_enable) { 806 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 807 MAKE_DIGEST_WORD(pdu->data_digest, crc32c); 808 } 809 810 pdu->cb_fn = cb_fn; 811 pdu->cb_arg = cb_arg; 812 813 pdu->sock_req.iovcnt = nvme_tcp_build_iovs(pdu->iov, SPDK_COUNTOF(pdu->iov), pdu, 814 tqpair->host_hdgst_enable, tqpair->host_ddgst_enable, 815 &mapped_length); 816 pdu->sock_req.cb_fn = _pdu_write_done; 817 pdu->sock_req.cb_arg = pdu; 818 TAILQ_INSERT_TAIL(&tqpair->send_queue, pdu, tailq); 819 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP || 820 pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ) { 821 rc = spdk_sock_writev(tqpair->sock, pdu->iov, pdu->sock_req.iovcnt); 822 if (rc == mapped_length) { 823 _pdu_write_done(pdu, 0); 824 } else { 825 SPDK_ERRLOG("IC_RESP or TERM_REQ could not write to socket.\n"); 826 _pdu_write_done(pdu, -1); 827 } 828 } else { 829 spdk_sock_writev_async(tqpair->sock, &pdu->sock_req); 830 } 831 } 832 833 static int 834 nvmf_tcp_qpair_init_mem_resource(struct spdk_nvmf_tcp_qpair *tqpair) 835 { 836 uint32_t i; 837 struct spdk_nvmf_transport_opts *opts; 838 uint32_t in_capsule_data_size; 839 840 opts = &tqpair->qpair.transport->opts; 841 842 in_capsule_data_size = opts->in_capsule_data_size; 843 if (opts->dif_insert_or_strip) { 844 in_capsule_data_size = SPDK_BDEV_BUF_SIZE_WITH_MD(in_capsule_data_size); 845 } 846 847 tqpair->resource_count = opts->max_queue_depth; 848 849 tqpair->reqs = calloc(tqpair->resource_count, sizeof(*tqpair->reqs)); 850 if (!tqpair->reqs) { 851 SPDK_ERRLOG("Unable to allocate reqs on tqpair=%p\n", tqpair); 852 return -1; 853 } 854 855 if (in_capsule_data_size) { 856 tqpair->bufs = spdk_zmalloc(tqpair->resource_count * in_capsule_data_size, 0x1000, 857 NULL, SPDK_ENV_LCORE_ID_ANY, 858 SPDK_MALLOC_DMA); 859 if (!tqpair->bufs) { 860 SPDK_ERRLOG("Unable to allocate bufs on tqpair=%p.\n", tqpair); 861 return -1; 862 } 863 } 864 865 /* Add addtional one member, which will be used for mgmt_pdu owned by the tqpair */ 866 tqpair->pdus = spdk_dma_malloc((tqpair->resource_count + 1) * sizeof(*tqpair->pdus), 0x1000, NULL); 867 if (!tqpair->pdus) { 868 SPDK_ERRLOG("Unable to allocate pdu pool on tqpair =%p.\n", tqpair); 869 return -1; 870 } 871 872 for (i = 0; i < tqpair->resource_count; i++) { 873 struct spdk_nvmf_tcp_req *tcp_req = &tqpair->reqs[i]; 874 875 tcp_req->ttag = i + 1; 876 tcp_req->req.qpair = &tqpair->qpair; 877 878 tcp_req->pdu = &tqpair->pdus[i]; 879 tcp_req->pdu->qpair = tqpair; 880 881 /* Set up memory to receive commands */ 882 if (tqpair->bufs) { 883 tcp_req->buf = (void *)((uintptr_t)tqpair->bufs + (i * in_capsule_data_size)); 884 } 885 886 /* Set the cmdn and rsp */ 887 tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp; 888 tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd; 889 890 /* Initialize request state to FREE */ 891 tcp_req->state = TCP_REQUEST_STATE_FREE; 892 TAILQ_INSERT_TAIL(&tqpair->state_queue[tcp_req->state], tcp_req, state_link); 893 tqpair->state_cntr[TCP_REQUEST_STATE_FREE]++; 894 } 895 896 tqpair->mgmt_pdu = &tqpair->pdus[i]; 897 tqpair->mgmt_pdu->qpair = tqpair; 898 899 tqpair->recv_buf_size = (in_capsule_data_size + sizeof(struct spdk_nvme_tcp_cmd) + 2 * 900 SPDK_NVME_TCP_DIGEST_LEN) * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 901 902 return 0; 903 } 904 905 static int 906 nvmf_tcp_qpair_init(struct spdk_nvmf_qpair *qpair) 907 { 908 struct spdk_nvmf_tcp_qpair *tqpair; 909 int i; 910 911 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 912 913 SPDK_DEBUGLOG(nvmf_tcp, "New TCP Connection: %p\n", qpair); 914 915 TAILQ_INIT(&tqpair->send_queue); 916 917 /* Initialise request state queues of the qpair */ 918 for (i = TCP_REQUEST_STATE_FREE; i < TCP_REQUEST_NUM_STATES; i++) { 919 TAILQ_INIT(&tqpair->state_queue[i]); 920 } 921 922 tqpair->host_hdgst_enable = true; 923 tqpair->host_ddgst_enable = true; 924 925 return 0; 926 } 927 928 static int 929 nvmf_tcp_qpair_sock_init(struct spdk_nvmf_tcp_qpair *tqpair) 930 { 931 int rc; 932 933 /* set low water mark */ 934 rc = spdk_sock_set_recvlowat(tqpair->sock, sizeof(struct spdk_nvme_tcp_common_pdu_hdr)); 935 if (rc != 0) { 936 SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n"); 937 return rc; 938 } 939 940 return 0; 941 } 942 943 static void 944 nvmf_tcp_handle_connect(struct spdk_nvmf_transport *transport, 945 struct spdk_nvmf_tcp_port *port, 946 struct spdk_sock *sock) 947 { 948 struct spdk_nvmf_tcp_qpair *tqpair; 949 int rc; 950 951 SPDK_DEBUGLOG(nvmf_tcp, "New connection accepted on %s port %s\n", 952 port->trid->traddr, port->trid->trsvcid); 953 954 tqpair = calloc(1, sizeof(struct spdk_nvmf_tcp_qpair)); 955 if (tqpair == NULL) { 956 SPDK_ERRLOG("Could not allocate new connection.\n"); 957 spdk_sock_close(&sock); 958 return; 959 } 960 961 tqpair->sock = sock; 962 tqpair->state_cntr[TCP_REQUEST_STATE_FREE] = 0; 963 tqpair->port = port; 964 tqpair->qpair.transport = transport; 965 tqpair->qpair.trid = port->trid; 966 967 rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr, 968 sizeof(tqpair->target_addr), &tqpair->target_port, 969 tqpair->initiator_addr, sizeof(tqpair->initiator_addr), 970 &tqpair->initiator_port); 971 if (rc < 0) { 972 SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair); 973 nvmf_tcp_qpair_destroy(tqpair); 974 return; 975 } 976 977 spdk_nvmf_tgt_new_qpair(transport->tgt, &tqpair->qpair); 978 } 979 980 static uint32_t 981 nvmf_tcp_port_accept(struct spdk_nvmf_transport *transport, struct spdk_nvmf_tcp_port *port) 982 { 983 struct spdk_sock *sock; 984 uint32_t count = 0; 985 int i; 986 987 for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) { 988 sock = spdk_sock_accept(port->listen_sock); 989 if (sock == NULL) { 990 break; 991 } 992 count++; 993 nvmf_tcp_handle_connect(transport, port, sock); 994 } 995 996 return count; 997 } 998 999 static uint32_t 1000 nvmf_tcp_accept(struct spdk_nvmf_transport *transport) 1001 { 1002 struct spdk_nvmf_tcp_transport *ttransport; 1003 struct spdk_nvmf_tcp_port *port; 1004 uint32_t count = 0; 1005 1006 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1007 1008 TAILQ_FOREACH(port, &ttransport->ports, link) { 1009 count += nvmf_tcp_port_accept(transport, port); 1010 } 1011 1012 return count; 1013 } 1014 1015 static void 1016 nvmf_tcp_discover(struct spdk_nvmf_transport *transport, 1017 struct spdk_nvme_transport_id *trid, 1018 struct spdk_nvmf_discovery_log_page_entry *entry) 1019 { 1020 entry->trtype = SPDK_NVMF_TRTYPE_TCP; 1021 entry->adrfam = trid->adrfam; 1022 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_REQUIRED; 1023 1024 spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' '); 1025 spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' '); 1026 1027 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE; 1028 } 1029 1030 static struct spdk_nvmf_tcp_control_msg_list * 1031 nvmf_tcp_control_msg_list_create(uint16_t num_messages) 1032 { 1033 struct spdk_nvmf_tcp_control_msg_list *list; 1034 struct spdk_nvmf_tcp_control_msg *msg; 1035 uint16_t i; 1036 1037 list = calloc(1, sizeof(*list)); 1038 if (!list) { 1039 SPDK_ERRLOG("Failed to allocate memory for list structure\n"); 1040 return NULL; 1041 } 1042 1043 list->msg_buf = spdk_zmalloc(num_messages * SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, 1044 NVMF_DATA_BUFFER_ALIGNMENT, NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 1045 if (!list->msg_buf) { 1046 SPDK_ERRLOG("Failed to allocate memory for control message buffers\n"); 1047 free(list); 1048 return NULL; 1049 } 1050 1051 STAILQ_INIT(&list->free_msgs); 1052 1053 for (i = 0; i < num_messages; i++) { 1054 msg = (struct spdk_nvmf_tcp_control_msg *)((char *)list->msg_buf + i * 1055 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1056 STAILQ_INSERT_TAIL(&list->free_msgs, msg, link); 1057 } 1058 1059 return list; 1060 } 1061 1062 static void 1063 nvmf_tcp_control_msg_list_free(struct spdk_nvmf_tcp_control_msg_list *list) 1064 { 1065 if (!list) { 1066 return; 1067 } 1068 1069 spdk_free(list->msg_buf); 1070 free(list); 1071 } 1072 1073 static struct spdk_nvmf_transport_poll_group * 1074 nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport) 1075 { 1076 struct spdk_nvmf_tcp_transport *ttransport; 1077 struct spdk_nvmf_tcp_poll_group *tgroup; 1078 1079 tgroup = calloc(1, sizeof(*tgroup)); 1080 if (!tgroup) { 1081 return NULL; 1082 } 1083 1084 tgroup->sock_group = spdk_sock_group_create(&tgroup->group); 1085 if (!tgroup->sock_group) { 1086 goto cleanup; 1087 } 1088 1089 TAILQ_INIT(&tgroup->qpairs); 1090 TAILQ_INIT(&tgroup->await_req); 1091 1092 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1093 1094 if (transport->opts.in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 1095 SPDK_DEBUGLOG(nvmf_tcp, "ICD %u is less than min required for admin/fabric commands (%u). " 1096 "Creating control messages list\n", transport->opts.in_capsule_data_size, 1097 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1098 tgroup->control_msg_list = nvmf_tcp_control_msg_list_create(ttransport->tcp_opts.control_msg_num); 1099 if (!tgroup->control_msg_list) { 1100 goto cleanup; 1101 } 1102 } 1103 1104 return &tgroup->group; 1105 1106 cleanup: 1107 nvmf_tcp_poll_group_destroy(&tgroup->group); 1108 return NULL; 1109 } 1110 1111 static struct spdk_nvmf_transport_poll_group * 1112 nvmf_tcp_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair) 1113 { 1114 struct spdk_nvmf_tcp_qpair *tqpair; 1115 struct spdk_sock_group *group = NULL; 1116 int rc; 1117 1118 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1119 rc = spdk_sock_get_optimal_sock_group(tqpair->sock, &group); 1120 if (!rc && group != NULL) { 1121 return spdk_sock_group_get_ctx(group); 1122 } 1123 1124 return NULL; 1125 } 1126 1127 static void 1128 nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group) 1129 { 1130 struct spdk_nvmf_tcp_poll_group *tgroup; 1131 1132 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 1133 spdk_sock_group_close(&tgroup->sock_group); 1134 if (tgroup->control_msg_list) { 1135 nvmf_tcp_control_msg_list_free(tgroup->control_msg_list); 1136 } 1137 1138 free(tgroup); 1139 } 1140 1141 static void 1142 nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1143 enum nvme_tcp_pdu_recv_state state) 1144 { 1145 if (tqpair->recv_state == state) { 1146 SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n", 1147 tqpair, state); 1148 return; 1149 } 1150 1151 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1152 /* When leaving the await req state, move the qpair to the main list */ 1153 TAILQ_REMOVE(&tqpair->group->await_req, tqpair, link); 1154 TAILQ_INSERT_TAIL(&tqpair->group->qpairs, tqpair, link); 1155 } 1156 1157 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv state=%d\n", tqpair, state); 1158 tqpair->recv_state = state; 1159 1160 switch (state) { 1161 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1162 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 1163 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 1164 break; 1165 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 1166 TAILQ_REMOVE(&tqpair->group->qpairs, tqpair, link); 1167 TAILQ_INSERT_TAIL(&tqpair->group->await_req, tqpair, link); 1168 break; 1169 case NVME_TCP_PDU_RECV_STATE_ERROR: 1170 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1171 memset(&tqpair->pdu_in_progress, 0, sizeof(tqpair->pdu_in_progress)); 1172 break; 1173 default: 1174 SPDK_ERRLOG("The state(%d) is invalid\n", state); 1175 abort(); 1176 break; 1177 } 1178 } 1179 1180 static int 1181 nvmf_tcp_qpair_handle_timeout(void *ctx) 1182 { 1183 struct spdk_nvmf_tcp_qpair *tqpair = ctx; 1184 1185 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1186 1187 SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair, 1188 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT); 1189 1190 nvmf_tcp_qpair_disconnect(tqpair); 1191 return SPDK_POLLER_BUSY; 1192 } 1193 1194 static void 1195 nvmf_tcp_send_c2h_term_req_complete(void *cb_arg) 1196 { 1197 struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg; 1198 1199 if (!tqpair->timeout_poller) { 1200 tqpair->timeout_poller = SPDK_POLLER_REGISTER(nvmf_tcp_qpair_handle_timeout, tqpair, 1201 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000); 1202 } 1203 } 1204 1205 static void 1206 nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu, 1207 enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset) 1208 { 1209 struct nvme_tcp_pdu *rsp_pdu; 1210 struct spdk_nvme_tcp_term_req_hdr *c2h_term_req; 1211 uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req); 1212 uint32_t copy_len; 1213 1214 rsp_pdu = tqpair->mgmt_pdu; 1215 1216 c2h_term_req = &rsp_pdu->hdr.term_req; 1217 c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ; 1218 c2h_term_req->common.hlen = c2h_term_req_hdr_len; 1219 1220 if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1221 (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1222 DSET32(&c2h_term_req->fei, error_offset); 1223 } 1224 1225 copy_len = spdk_min(pdu->hdr.common.hlen, SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE); 1226 1227 /* Copy the error info into the buffer */ 1228 memcpy((uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, pdu->hdr.raw, copy_len); 1229 nvme_tcp_pdu_set_data(rsp_pdu, (uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, copy_len); 1230 1231 /* Contain the header of the wrong received pdu */ 1232 c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len; 1233 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1234 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_send_c2h_term_req_complete, tqpair); 1235 } 1236 1237 static void 1238 nvmf_tcp_capsule_cmd_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1239 struct spdk_nvmf_tcp_qpair *tqpair, 1240 struct nvme_tcp_pdu *pdu) 1241 { 1242 struct spdk_nvmf_tcp_req *tcp_req; 1243 1244 assert(pdu->psh_valid_bytes == pdu->psh_len); 1245 assert(pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD); 1246 1247 tcp_req = nvmf_tcp_req_get(tqpair); 1248 if (!tcp_req) { 1249 /* Directly return and make the allocation retry again */ 1250 if (tqpair->state_cntr[TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST] > 0) { 1251 return; 1252 } 1253 1254 /* The host sent more commands than the maximum queue depth. */ 1255 SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair); 1256 nvmf_tcp_qpair_disconnect(tqpair); 1257 return; 1258 } 1259 1260 pdu->req = tcp_req; 1261 assert(tcp_req->state == TCP_REQUEST_STATE_NEW); 1262 nvmf_tcp_req_process(ttransport, tcp_req); 1263 } 1264 1265 static void 1266 nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1267 struct spdk_nvmf_tcp_qpair *tqpair, 1268 struct nvme_tcp_pdu *pdu) 1269 { 1270 struct spdk_nvmf_tcp_req *tcp_req; 1271 struct spdk_nvme_tcp_cmd *capsule_cmd; 1272 uint32_t error_offset = 0; 1273 enum spdk_nvme_tcp_term_req_fes fes; 1274 1275 capsule_cmd = &pdu->hdr.capsule_cmd; 1276 tcp_req = pdu->req; 1277 assert(tcp_req != NULL); 1278 if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) { 1279 SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n", 1280 SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo); 1281 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1282 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1283 goto err; 1284 } 1285 1286 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1287 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1288 nvmf_tcp_req_process(ttransport, tcp_req); 1289 1290 return; 1291 err: 1292 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1293 } 1294 1295 static int 1296 nvmf_tcp_find_req_in_state(struct spdk_nvmf_tcp_qpair *tqpair, 1297 enum spdk_nvmf_tcp_req_state state, 1298 uint16_t cid, uint16_t tag, 1299 struct spdk_nvmf_tcp_req **req) 1300 { 1301 struct spdk_nvmf_tcp_req *tcp_req = NULL; 1302 1303 TAILQ_FOREACH(tcp_req, &tqpair->state_queue[state], state_link) { 1304 if (tcp_req->req.cmd->nvme_cmd.cid != cid) { 1305 continue; 1306 } 1307 1308 if (tcp_req->ttag == tag) { 1309 *req = tcp_req; 1310 return 0; 1311 } 1312 1313 *req = NULL; 1314 return -1; 1315 } 1316 1317 /* Didn't find it, but not an error */ 1318 *req = NULL; 1319 return 0; 1320 } 1321 1322 static void 1323 nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1324 struct spdk_nvmf_tcp_qpair *tqpair, 1325 struct nvme_tcp_pdu *pdu) 1326 { 1327 struct spdk_nvmf_tcp_req *tcp_req; 1328 uint32_t error_offset = 0; 1329 enum spdk_nvme_tcp_term_req_fes fes = 0; 1330 struct spdk_nvme_tcp_h2c_data_hdr *h2c_data; 1331 int rc; 1332 1333 h2c_data = &pdu->hdr.h2c_data; 1334 1335 SPDK_DEBUGLOG(nvmf_tcp, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n", 1336 tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag); 1337 1338 rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 1339 h2c_data->cccid, h2c_data->ttag, &tcp_req); 1340 if (rc == 0 && tcp_req == NULL) { 1341 rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK, h2c_data->cccid, 1342 h2c_data->ttag, &tcp_req); 1343 } 1344 1345 if (!tcp_req) { 1346 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req is not found for tqpair=%p\n", tqpair); 1347 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 1348 if (rc == 0) { 1349 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid); 1350 } else { 1351 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1352 } 1353 goto err; 1354 } 1355 1356 if (tcp_req->h2c_offset != h2c_data->datao) { 1357 SPDK_DEBUGLOG(nvmf_tcp, 1358 "tcp_req(%p), tqpair=%p, expected data offset %u, but data offset is %u\n", 1359 tcp_req, tqpair, tcp_req->h2c_offset, h2c_data->datao); 1360 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1361 goto err; 1362 } 1363 1364 if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) { 1365 SPDK_DEBUGLOG(nvmf_tcp, 1366 "tcp_req(%p), tqpair=%p, (datao=%u + datal=%u) execeeds requested length=%u\n", 1367 tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length); 1368 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1369 goto err; 1370 } 1371 1372 pdu->req = tcp_req; 1373 1374 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 1375 pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 1376 } 1377 1378 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 1379 h2c_data->datao, h2c_data->datal); 1380 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1381 return; 1382 1383 err: 1384 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1385 } 1386 1387 static void 1388 nvmf_tcp_pdu_cmd_complete(void *cb_arg) 1389 { 1390 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1391 nvmf_tcp_request_free(tcp_req); 1392 } 1393 1394 static void 1395 nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req, 1396 struct spdk_nvmf_tcp_qpair *tqpair) 1397 { 1398 struct nvme_tcp_pdu *rsp_pdu; 1399 struct spdk_nvme_tcp_rsp *capsule_resp; 1400 1401 SPDK_DEBUGLOG(nvmf_tcp, "enter, tqpair=%p\n", tqpair); 1402 1403 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1404 assert(rsp_pdu != NULL); 1405 1406 capsule_resp = &rsp_pdu->hdr.capsule_resp; 1407 capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP; 1408 capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp); 1409 capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl; 1410 if (tqpair->host_hdgst_enable) { 1411 capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1412 capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1413 } 1414 1415 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_pdu_cmd_complete, tcp_req); 1416 } 1417 1418 static void 1419 nvmf_tcp_pdu_c2h_data_complete(void *cb_arg) 1420 { 1421 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1422 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, 1423 struct spdk_nvmf_tcp_qpair, qpair); 1424 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF( 1425 tcp_req->req.qpair->transport, struct spdk_nvmf_tcp_transport, transport); 1426 1427 assert(tqpair != NULL); 1428 if (ttransport->tcp_opts.c2h_success) { 1429 nvmf_tcp_request_free(tcp_req); 1430 } else { 1431 nvmf_tcp_req_pdu_fini(tcp_req); 1432 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 1433 } 1434 } 1435 1436 static void 1437 nvmf_tcp_r2t_complete(void *cb_arg) 1438 { 1439 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1440 struct spdk_nvmf_tcp_transport *ttransport; 1441 1442 nvmf_tcp_req_pdu_fini(tcp_req); 1443 1444 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 1445 struct spdk_nvmf_tcp_transport, transport); 1446 1447 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 1448 1449 if (tcp_req->h2c_offset == tcp_req->req.length) { 1450 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1451 nvmf_tcp_req_process(ttransport, tcp_req); 1452 } 1453 } 1454 1455 static void 1456 nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1457 struct spdk_nvmf_tcp_req *tcp_req) 1458 { 1459 struct nvme_tcp_pdu *rsp_pdu; 1460 struct spdk_nvme_tcp_r2t_hdr *r2t; 1461 1462 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1463 assert(rsp_pdu != NULL); 1464 1465 r2t = &rsp_pdu->hdr.r2t; 1466 r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T; 1467 r2t->common.plen = r2t->common.hlen = sizeof(*r2t); 1468 1469 if (tqpair->host_hdgst_enable) { 1470 r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1471 r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1472 } 1473 1474 r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid; 1475 r2t->ttag = tcp_req->ttag; 1476 r2t->r2to = tcp_req->h2c_offset; 1477 r2t->r2tl = tcp_req->req.length; 1478 1479 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 1480 1481 SPDK_DEBUGLOG(nvmf_tcp, 1482 "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n", 1483 tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl); 1484 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_r2t_complete, tcp_req); 1485 } 1486 1487 static void 1488 nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1489 struct spdk_nvmf_tcp_qpair *tqpair, 1490 struct nvme_tcp_pdu *pdu) 1491 { 1492 struct spdk_nvmf_tcp_req *tcp_req; 1493 1494 tcp_req = pdu->req; 1495 assert(tcp_req != NULL); 1496 1497 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1498 1499 tcp_req->h2c_offset += pdu->data_len; 1500 1501 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1502 1503 /* Wait for all of the data to arrive AND for the initial R2T PDU send to be 1504 * acknowledged before moving on. */ 1505 if (tcp_req->h2c_offset == tcp_req->req.length && 1506 tcp_req->state == TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER) { 1507 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1508 nvmf_tcp_req_process(ttransport, tcp_req); 1509 } 1510 } 1511 1512 static void 1513 nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req) 1514 { 1515 SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req, 1516 spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]); 1517 if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1518 (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1519 SPDK_DEBUGLOG(nvmf_tcp, "The offset from the start of the PDU header is %u\n", 1520 DGET32(h2c_term_req->fei)); 1521 } 1522 } 1523 1524 static void 1525 nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1526 struct nvme_tcp_pdu *pdu) 1527 { 1528 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1529 uint32_t error_offset = 0; 1530 enum spdk_nvme_tcp_term_req_fes fes; 1531 1532 1533 if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) { 1534 SPDK_ERRLOG("Fatal Error Stauts(FES) is unknown for h2c_term_req pdu=%p\n", pdu); 1535 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1536 error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes); 1537 goto end; 1538 } 1539 1540 /* set the data buffer */ 1541 nvme_tcp_pdu_set_data(pdu, (uint8_t *)pdu->hdr.raw + h2c_term_req->common.hlen, 1542 h2c_term_req->common.plen - h2c_term_req->common.hlen); 1543 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1544 return; 1545 end: 1546 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1547 } 1548 1549 static void 1550 nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1551 struct nvme_tcp_pdu *pdu) 1552 { 1553 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1554 1555 nvmf_tcp_h2c_term_req_dump(h2c_term_req); 1556 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1557 } 1558 1559 static void 1560 nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1561 struct spdk_nvmf_tcp_transport *ttransport) 1562 { 1563 int rc = 0; 1564 struct nvme_tcp_pdu *pdu; 1565 uint32_t crc32c, error_offset = 0; 1566 enum spdk_nvme_tcp_term_req_fes fes; 1567 1568 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1569 pdu = &tqpair->pdu_in_progress; 1570 1571 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1572 /* check data digest if need */ 1573 if (pdu->ddgst_enable) { 1574 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 1575 rc = MATCH_DIGEST_WORD(pdu->data_digest, crc32c); 1576 if (rc == 0) { 1577 SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1578 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1579 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1580 return; 1581 1582 } 1583 } 1584 1585 switch (pdu->hdr.common.pdu_type) { 1586 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1587 nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu); 1588 break; 1589 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1590 nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu); 1591 break; 1592 1593 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1594 nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu); 1595 break; 1596 1597 default: 1598 /* The code should not go to here */ 1599 SPDK_ERRLOG("The code should not go to here\n"); 1600 break; 1601 } 1602 } 1603 1604 static void 1605 nvmf_tcp_send_icresp_complete(void *cb_arg) 1606 { 1607 struct spdk_nvmf_tcp_qpair *tqpair = cb_arg; 1608 1609 tqpair->state = NVME_TCP_QPAIR_STATE_RUNNING; 1610 } 1611 1612 static void 1613 nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport, 1614 struct spdk_nvmf_tcp_qpair *tqpair, 1615 struct nvme_tcp_pdu *pdu) 1616 { 1617 struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr.ic_req; 1618 struct nvme_tcp_pdu *rsp_pdu; 1619 struct spdk_nvme_tcp_ic_resp *ic_resp; 1620 uint32_t error_offset = 0; 1621 enum spdk_nvme_tcp_term_req_fes fes; 1622 1623 /* Only PFV 0 is defined currently */ 1624 if (ic_req->pfv != 0) { 1625 SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv); 1626 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1627 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv); 1628 goto end; 1629 } 1630 1631 /* MAXR2T is 0's based */ 1632 SPDK_DEBUGLOG(nvmf_tcp, "maxr2t =%u\n", (ic_req->maxr2t + 1u)); 1633 1634 tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false; 1635 if (!tqpair->host_hdgst_enable) { 1636 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1637 } 1638 1639 tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false; 1640 if (!tqpair->host_ddgst_enable) { 1641 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1642 } 1643 1644 tqpair->recv_buf_size = spdk_max(tqpair->recv_buf_size, MIN_SOCK_PIPE_SIZE); 1645 /* Now that we know whether digests are enabled, properly size the receive buffer */ 1646 if (spdk_sock_set_recvbuf(tqpair->sock, tqpair->recv_buf_size) < 0) { 1647 SPDK_WARNLOG("Unable to allocate enough memory for receive buffer on tqpair=%p with size=%d\n", 1648 tqpair, 1649 tqpair->recv_buf_size); 1650 /* Not fatal. */ 1651 } 1652 1653 tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX); 1654 SPDK_DEBUGLOG(nvmf_tcp, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda); 1655 1656 rsp_pdu = tqpair->mgmt_pdu; 1657 1658 ic_resp = &rsp_pdu->hdr.ic_resp; 1659 ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP; 1660 ic_resp->common.hlen = ic_resp->common.plen = sizeof(*ic_resp); 1661 ic_resp->pfv = 0; 1662 ic_resp->cpda = tqpair->cpda; 1663 ic_resp->maxh2cdata = ttransport->transport.opts.max_io_size; 1664 ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0; 1665 ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0; 1666 1667 SPDK_DEBUGLOG(nvmf_tcp, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable); 1668 SPDK_DEBUGLOG(nvmf_tcp, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable); 1669 1670 tqpair->state = NVME_TCP_QPAIR_STATE_INITIALIZING; 1671 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_send_icresp_complete, tqpair); 1672 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1673 return; 1674 end: 1675 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1676 } 1677 1678 static void 1679 nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1680 struct spdk_nvmf_tcp_transport *ttransport) 1681 { 1682 struct nvme_tcp_pdu *pdu; 1683 int rc; 1684 uint32_t crc32c, error_offset = 0; 1685 enum spdk_nvme_tcp_term_req_fes fes; 1686 1687 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1688 pdu = &tqpair->pdu_in_progress; 1689 1690 SPDK_DEBUGLOG(nvmf_tcp, "pdu type of tqpair(%p) is %d\n", tqpair, 1691 pdu->hdr.common.pdu_type); 1692 /* check header digest if needed */ 1693 if (pdu->has_hdgst) { 1694 SPDK_DEBUGLOG(nvmf_tcp, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair); 1695 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 1696 rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, crc32c); 1697 if (rc == 0) { 1698 SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1699 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1700 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1701 return; 1702 1703 } 1704 } 1705 1706 switch (pdu->hdr.common.pdu_type) { 1707 case SPDK_NVME_TCP_PDU_TYPE_IC_REQ: 1708 nvmf_tcp_icreq_handle(ttransport, tqpair, pdu); 1709 break; 1710 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1711 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_REQ); 1712 break; 1713 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1714 nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu); 1715 break; 1716 1717 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1718 nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu); 1719 break; 1720 1721 default: 1722 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress.hdr.common.pdu_type); 1723 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1724 error_offset = 1; 1725 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1726 break; 1727 } 1728 } 1729 1730 static void 1731 nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair) 1732 { 1733 struct nvme_tcp_pdu *pdu; 1734 uint32_t error_offset = 0; 1735 enum spdk_nvme_tcp_term_req_fes fes; 1736 uint8_t expected_hlen, pdo; 1737 bool plen_error = false, pdo_error = false; 1738 1739 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 1740 pdu = &tqpair->pdu_in_progress; 1741 1742 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) { 1743 if (tqpair->state != NVME_TCP_QPAIR_STATE_INVALID) { 1744 SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu); 1745 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1746 goto err; 1747 } 1748 expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req); 1749 if (pdu->hdr.common.plen != expected_hlen) { 1750 plen_error = true; 1751 } 1752 } else { 1753 if (tqpair->state != NVME_TCP_QPAIR_STATE_RUNNING) { 1754 SPDK_ERRLOG("The TCP/IP connection is not negotitated\n"); 1755 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1756 goto err; 1757 } 1758 1759 switch (pdu->hdr.common.pdu_type) { 1760 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1761 expected_hlen = sizeof(struct spdk_nvme_tcp_cmd); 1762 pdo = pdu->hdr.common.pdo; 1763 if ((tqpair->cpda != 0) && (pdo != ((tqpair->cpda + 1) << 2))) { 1764 pdo_error = true; 1765 break; 1766 } 1767 1768 if (pdu->hdr.common.plen < expected_hlen) { 1769 plen_error = true; 1770 } 1771 break; 1772 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1773 expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr); 1774 pdo = pdu->hdr.common.pdo; 1775 if ((tqpair->cpda != 0) && (pdo != ((tqpair->cpda + 1) << 2))) { 1776 pdo_error = true; 1777 break; 1778 } 1779 if (pdu->hdr.common.plen < expected_hlen) { 1780 plen_error = true; 1781 } 1782 break; 1783 1784 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1785 expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr); 1786 if ((pdu->hdr.common.plen <= expected_hlen) || 1787 (pdu->hdr.common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) { 1788 plen_error = true; 1789 } 1790 break; 1791 1792 default: 1793 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr.common.pdu_type); 1794 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1795 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type); 1796 goto err; 1797 } 1798 } 1799 1800 if (pdu->hdr.common.hlen != expected_hlen) { 1801 SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n", 1802 pdu->hdr.common.pdu_type, 1803 expected_hlen, pdu->hdr.common.hlen, tqpair); 1804 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1805 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen); 1806 goto err; 1807 } else if (pdo_error) { 1808 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1809 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1810 } else if (plen_error) { 1811 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1812 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 1813 goto err; 1814 } else { 1815 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1816 nvme_tcp_pdu_calc_psh_len(&tqpair->pdu_in_progress, tqpair->host_hdgst_enable); 1817 return; 1818 } 1819 err: 1820 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1821 } 1822 1823 static int 1824 nvmf_tcp_pdu_payload_insert_dif(struct nvme_tcp_pdu *pdu, uint32_t read_offset, 1825 int read_len) 1826 { 1827 int rc; 1828 1829 rc = spdk_dif_generate_stream(pdu->data_iov, pdu->data_iovcnt, 1830 read_offset, read_len, pdu->dif_ctx); 1831 if (rc != 0) { 1832 SPDK_ERRLOG("DIF generate failed\n"); 1833 } 1834 1835 return rc; 1836 } 1837 1838 static int 1839 nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair) 1840 { 1841 int rc = 0; 1842 struct nvme_tcp_pdu *pdu; 1843 enum nvme_tcp_pdu_recv_state prev_state; 1844 uint32_t data_len; 1845 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 1846 struct spdk_nvmf_tcp_transport, transport); 1847 1848 /* The loop here is to allow for several back-to-back state changes. */ 1849 do { 1850 prev_state = tqpair->recv_state; 1851 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state); 1852 1853 pdu = &tqpair->pdu_in_progress; 1854 switch (tqpair->recv_state) { 1855 /* Wait for the common header */ 1856 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1857 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1858 if (spdk_unlikely(tqpair->state == NVME_TCP_QPAIR_STATE_INITIALIZING)) { 1859 return rc; 1860 } 1861 1862 rc = nvme_tcp_read_data(tqpair->sock, 1863 sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes, 1864 (void *)&pdu->hdr.common + pdu->ch_valid_bytes); 1865 if (rc < 0) { 1866 SPDK_DEBUGLOG(nvmf_tcp, "will disconnect tqpair=%p\n", tqpair); 1867 return NVME_TCP_PDU_FATAL; 1868 } else if (rc > 0) { 1869 pdu->ch_valid_bytes += rc; 1870 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, 0, rc, 0, 0); 1871 if (spdk_likely(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY)) { 1872 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 1873 } 1874 } 1875 1876 if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) { 1877 return NVME_TCP_PDU_IN_PROGRESS; 1878 } 1879 1880 /* The command header of this PDU has now been read from the socket. */ 1881 nvmf_tcp_pdu_ch_handle(tqpair); 1882 break; 1883 /* Wait for the pdu specific header */ 1884 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 1885 rc = nvme_tcp_read_data(tqpair->sock, 1886 pdu->psh_len - pdu->psh_valid_bytes, 1887 (void *)&pdu->hdr.raw + sizeof(struct spdk_nvme_tcp_common_pdu_hdr) + pdu->psh_valid_bytes); 1888 if (rc < 0) { 1889 return NVME_TCP_PDU_FATAL; 1890 } else if (rc > 0) { 1891 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, 1892 0, rc, 0, 0); 1893 pdu->psh_valid_bytes += rc; 1894 } 1895 1896 if (pdu->psh_valid_bytes < pdu->psh_len) { 1897 return NVME_TCP_PDU_IN_PROGRESS; 1898 } 1899 1900 /* All header(ch, psh, head digist) of this PDU has now been read from the socket. */ 1901 nvmf_tcp_pdu_psh_handle(tqpair, ttransport); 1902 break; 1903 /* Wait for the req slot */ 1904 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 1905 nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu); 1906 break; 1907 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 1908 /* check whether the data is valid, if not we just return */ 1909 if (!pdu->data_len) { 1910 return NVME_TCP_PDU_IN_PROGRESS; 1911 } 1912 1913 data_len = pdu->data_len; 1914 /* data digest */ 1915 if (spdk_unlikely((pdu->hdr.common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) && 1916 tqpair->host_ddgst_enable)) { 1917 data_len += SPDK_NVME_TCP_DIGEST_LEN; 1918 pdu->ddgst_enable = true; 1919 } 1920 1921 rc = nvme_tcp_read_payload_data(tqpair->sock, pdu); 1922 if (rc < 0) { 1923 return NVME_TCP_PDU_FATAL; 1924 } 1925 pdu->readv_offset += rc; 1926 1927 if (spdk_unlikely(pdu->dif_ctx != NULL)) { 1928 rc = nvmf_tcp_pdu_payload_insert_dif(pdu, pdu->readv_offset - rc, rc); 1929 if (rc != 0) { 1930 return NVME_TCP_PDU_FATAL; 1931 } 1932 } 1933 1934 if (pdu->readv_offset < data_len) { 1935 return NVME_TCP_PDU_IN_PROGRESS; 1936 } 1937 1938 /* All of this PDU has now been read from the socket. */ 1939 nvmf_tcp_pdu_payload_handle(tqpair, ttransport); 1940 break; 1941 case NVME_TCP_PDU_RECV_STATE_ERROR: 1942 if (!spdk_sock_is_connected(tqpair->sock)) { 1943 return NVME_TCP_PDU_FATAL; 1944 } 1945 break; 1946 default: 1947 assert(0); 1948 SPDK_ERRLOG("code should not come to here"); 1949 break; 1950 } 1951 } while (tqpair->recv_state != prev_state); 1952 1953 return rc; 1954 } 1955 1956 static inline void * 1957 nvmf_tcp_control_msg_get(struct spdk_nvmf_tcp_control_msg_list *list) 1958 { 1959 struct spdk_nvmf_tcp_control_msg *msg; 1960 1961 assert(list); 1962 1963 msg = STAILQ_FIRST(&list->free_msgs); 1964 if (!msg) { 1965 SPDK_DEBUGLOG(nvmf_tcp, "Out of control messages\n"); 1966 return NULL; 1967 } 1968 STAILQ_REMOVE_HEAD(&list->free_msgs, link); 1969 return msg; 1970 } 1971 1972 static inline void 1973 nvmf_tcp_control_msg_put(struct spdk_nvmf_tcp_control_msg_list *list, void *_msg) 1974 { 1975 struct spdk_nvmf_tcp_control_msg *msg = _msg; 1976 1977 assert(list); 1978 STAILQ_INSERT_HEAD(&list->free_msgs, msg, link); 1979 } 1980 1981 static int 1982 nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_req *tcp_req, 1983 struct spdk_nvmf_transport *transport, 1984 struct spdk_nvmf_transport_poll_group *group) 1985 { 1986 struct spdk_nvmf_request *req = &tcp_req->req; 1987 struct spdk_nvme_cmd *cmd; 1988 struct spdk_nvme_cpl *rsp; 1989 struct spdk_nvme_sgl_descriptor *sgl; 1990 struct spdk_nvmf_tcp_poll_group *tgroup; 1991 uint32_t length; 1992 1993 cmd = &req->cmd->nvme_cmd; 1994 rsp = &req->rsp->nvme_cpl; 1995 sgl = &cmd->dptr.sgl1; 1996 1997 length = sgl->unkeyed.length; 1998 1999 if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK && 2000 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) { 2001 if (length > transport->opts.max_io_size) { 2002 SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n", 2003 length, transport->opts.max_io_size); 2004 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 2005 return -1; 2006 } 2007 2008 /* fill request length and populate iovs */ 2009 req->length = length; 2010 2011 SPDK_DEBUGLOG(nvmf_tcp, "Data requested length= 0x%x\n", length); 2012 2013 if (spdk_unlikely(req->dif.dif_insert_or_strip)) { 2014 req->dif.orig_length = length; 2015 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2016 req->dif.elba_length = length; 2017 } 2018 2019 if (spdk_nvmf_request_get_buffers(req, group, transport, length)) { 2020 /* No available buffers. Queue this request up. */ 2021 SPDK_DEBUGLOG(nvmf_tcp, "No available large data buffers. Queueing request %p\n", 2022 tcp_req); 2023 return 0; 2024 } 2025 2026 /* backward compatible */ 2027 req->data = req->iov[0].iov_base; 2028 2029 SPDK_DEBUGLOG(nvmf_tcp, "Request %p took %d buffer/s from central pool, and data=%p\n", 2030 tcp_req, req->iovcnt, req->data); 2031 2032 return 0; 2033 } else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK && 2034 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) { 2035 uint64_t offset = sgl->address; 2036 uint32_t max_len = transport->opts.in_capsule_data_size; 2037 assert(tcp_req->has_incapsule_data); 2038 2039 SPDK_DEBUGLOG(nvmf_tcp, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n", 2040 offset, length); 2041 2042 if (offset > max_len) { 2043 SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " exceeds capsule length 0x%x\n", 2044 offset, max_len); 2045 rsp->status.sc = SPDK_NVME_SC_INVALID_SGL_OFFSET; 2046 return -1; 2047 } 2048 max_len -= (uint32_t)offset; 2049 2050 if (spdk_unlikely(length > max_len)) { 2051 /* According to the SPEC we should support ICD up to 8192 bytes for admin and fabric commands */ 2052 if (length <= SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE && 2053 (cmd->opc == SPDK_NVME_OPC_FABRIC || req->qpair->qid == 0)) { 2054 2055 /* Get a buffer from dedicated list */ 2056 SPDK_DEBUGLOG(nvmf_tcp, "Getting a buffer from control msg list\n"); 2057 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2058 assert(tgroup->control_msg_list); 2059 req->data = nvmf_tcp_control_msg_get(tgroup->control_msg_list); 2060 if (!req->data) { 2061 /* No available buffers. Queue this request up. */ 2062 SPDK_DEBUGLOG(nvmf_tcp, "No available ICD buffers. Queueing request %p\n", tcp_req); 2063 return 0; 2064 } 2065 } else { 2066 SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n", 2067 length, max_len); 2068 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 2069 return -1; 2070 } 2071 } else { 2072 req->data = tcp_req->buf; 2073 } 2074 2075 req->length = length; 2076 req->data_from_pool = false; 2077 2078 if (spdk_unlikely(req->dif.dif_insert_or_strip)) { 2079 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2080 req->dif.elba_length = length; 2081 } 2082 2083 req->iov[0].iov_base = req->data; 2084 req->iov[0].iov_len = length; 2085 req->iovcnt = 1; 2086 2087 return 0; 2088 } 2089 2090 SPDK_ERRLOG("Invalid NVMf I/O Command SGL: Type 0x%x, Subtype 0x%x\n", 2091 sgl->generic.type, sgl->generic.subtype); 2092 rsp->status.sc = SPDK_NVME_SC_SGL_DESCRIPTOR_TYPE_INVALID; 2093 return -1; 2094 } 2095 2096 static inline enum spdk_nvme_media_error_status_code 2097 nvmf_tcp_dif_error_to_compl_status(uint8_t err_type) { 2098 enum spdk_nvme_media_error_status_code result; 2099 2100 switch (err_type) 2101 { 2102 case SPDK_DIF_REFTAG_ERROR: 2103 result = SPDK_NVME_SC_REFERENCE_TAG_CHECK_ERROR; 2104 break; 2105 case SPDK_DIF_APPTAG_ERROR: 2106 result = SPDK_NVME_SC_APPLICATION_TAG_CHECK_ERROR; 2107 break; 2108 case SPDK_DIF_GUARD_ERROR: 2109 result = SPDK_NVME_SC_GUARD_CHECK_ERROR; 2110 break; 2111 default: 2112 SPDK_UNREACHABLE(); 2113 break; 2114 } 2115 2116 return result; 2117 } 2118 2119 static void 2120 nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2121 struct spdk_nvmf_tcp_req *tcp_req) 2122 { 2123 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF( 2124 tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 2125 struct nvme_tcp_pdu *rsp_pdu; 2126 struct spdk_nvme_tcp_c2h_data_hdr *c2h_data; 2127 uint32_t plen, pdo, alignment; 2128 int rc; 2129 2130 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2131 2132 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 2133 assert(rsp_pdu != NULL); 2134 2135 c2h_data = &rsp_pdu->hdr.c2h_data; 2136 c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA; 2137 plen = c2h_data->common.hlen = sizeof(*c2h_data); 2138 2139 if (tqpair->host_hdgst_enable) { 2140 plen += SPDK_NVME_TCP_DIGEST_LEN; 2141 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 2142 } 2143 2144 /* set the psh */ 2145 c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid; 2146 c2h_data->datal = tcp_req->req.length; 2147 c2h_data->datao = 0; 2148 2149 /* set the padding */ 2150 rsp_pdu->padding_len = 0; 2151 pdo = plen; 2152 if (tqpair->cpda) { 2153 alignment = (tqpair->cpda + 1) << 2; 2154 if (alignment > plen) { 2155 rsp_pdu->padding_len = alignment - plen; 2156 pdo = plen = alignment; 2157 } 2158 } 2159 2160 c2h_data->common.pdo = pdo; 2161 plen += c2h_data->datal; 2162 if (tqpair->host_ddgst_enable) { 2163 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF; 2164 plen += SPDK_NVME_TCP_DIGEST_LEN; 2165 } 2166 2167 c2h_data->common.plen = plen; 2168 2169 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2170 rsp_pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 2171 } 2172 2173 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2174 c2h_data->datao, c2h_data->datal); 2175 2176 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2177 struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl; 2178 struct spdk_dif_error err_blk = {}; 2179 2180 rc = spdk_dif_verify_stream(rsp_pdu->data_iov, rsp_pdu->data_iovcnt, 2181 0, rsp_pdu->data_len, rsp_pdu->dif_ctx, &err_blk); 2182 if (rc != 0) { 2183 SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n", 2184 err_blk.err_type, err_blk.err_offset); 2185 rsp->status.sct = SPDK_NVME_SCT_MEDIA_ERROR; 2186 rsp->status.sc = nvmf_tcp_dif_error_to_compl_status(err_blk.err_type); 2187 nvmf_tcp_req_pdu_fini(tcp_req); 2188 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2189 return; 2190 } 2191 } 2192 2193 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU; 2194 if (ttransport->tcp_opts.c2h_success) { 2195 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS; 2196 } 2197 2198 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_pdu_c2h_data_complete, tcp_req); 2199 } 2200 2201 static int 2202 request_transfer_out(struct spdk_nvmf_request *req) 2203 { 2204 struct spdk_nvmf_tcp_req *tcp_req; 2205 struct spdk_nvmf_qpair *qpair; 2206 struct spdk_nvmf_tcp_qpair *tqpair; 2207 struct spdk_nvme_cpl *rsp; 2208 2209 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2210 2211 qpair = req->qpair; 2212 rsp = &req->rsp->nvme_cpl; 2213 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2214 2215 /* Advance our sq_head pointer */ 2216 if (qpair->sq_head == qpair->sq_head_max) { 2217 qpair->sq_head = 0; 2218 } else { 2219 qpair->sq_head++; 2220 } 2221 rsp->sqhd = qpair->sq_head; 2222 2223 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2224 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 2225 if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) { 2226 nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2227 } else { 2228 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2229 } 2230 2231 return 0; 2232 } 2233 2234 static void 2235 nvmf_tcp_set_incapsule_data(struct spdk_nvmf_tcp_qpair *tqpair, 2236 struct spdk_nvmf_tcp_req *tcp_req) 2237 { 2238 struct nvme_tcp_pdu *pdu; 2239 uint32_t plen = 0; 2240 2241 pdu = &tqpair->pdu_in_progress; 2242 plen = pdu->hdr.common.hlen; 2243 2244 if (tqpair->host_hdgst_enable) { 2245 plen += SPDK_NVME_TCP_DIGEST_LEN; 2246 } 2247 2248 if (pdu->hdr.common.plen != plen) { 2249 tcp_req->has_incapsule_data = true; 2250 } 2251 } 2252 2253 static bool 2254 nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 2255 struct spdk_nvmf_tcp_req *tcp_req) 2256 { 2257 struct spdk_nvmf_tcp_qpair *tqpair; 2258 int rc; 2259 enum spdk_nvmf_tcp_req_state prev_state; 2260 bool progress = false; 2261 struct spdk_nvmf_transport *transport = &ttransport->transport; 2262 struct spdk_nvmf_transport_poll_group *group; 2263 struct spdk_nvmf_tcp_poll_group *tgroup; 2264 2265 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2266 group = &tqpair->group->group; 2267 assert(tcp_req->state != TCP_REQUEST_STATE_FREE); 2268 2269 /* If the qpair is not active, we need to abort the outstanding requests. */ 2270 if (tqpair->qpair.state != SPDK_NVMF_QPAIR_ACTIVE) { 2271 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 2272 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2273 } 2274 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 2275 } 2276 2277 /* The loop here is to allow for several back-to-back state changes. */ 2278 do { 2279 prev_state = tcp_req->state; 2280 2281 SPDK_DEBUGLOG(nvmf_tcp, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state, 2282 tqpair); 2283 2284 switch (tcp_req->state) { 2285 case TCP_REQUEST_STATE_FREE: 2286 /* Some external code must kick a request into TCP_REQUEST_STATE_NEW 2287 * to escape this state. */ 2288 break; 2289 case TCP_REQUEST_STATE_NEW: 2290 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, 0, 0, (uintptr_t)tcp_req, 0); 2291 2292 /* copy the cmd from the receive pdu */ 2293 tcp_req->cmd = tqpair->pdu_in_progress.hdr.capsule_cmd.ccsqe; 2294 2295 if (spdk_unlikely(spdk_nvmf_request_get_dif_ctx(&tcp_req->req, &tcp_req->req.dif.dif_ctx))) { 2296 tcp_req->req.dif.dif_insert_or_strip = true; 2297 tqpair->pdu_in_progress.dif_ctx = &tcp_req->req.dif.dif_ctx; 2298 } 2299 2300 /* The next state transition depends on the data transfer needs of this request. */ 2301 tcp_req->req.xfer = spdk_nvmf_req_get_xfer(&tcp_req->req); 2302 2303 if (spdk_unlikely(tcp_req->req.xfer == SPDK_NVME_DATA_BIDIRECTIONAL)) { 2304 tcp_req->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2305 tcp_req->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SC_INVALID_OPCODE; 2306 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2307 SPDK_DEBUGLOG(nvmf_tcp, "Request %p: invalid xfer type (BIDIRECTIONAL)\n", tcp_req); 2308 break; 2309 } 2310 2311 /* If no data to transfer, ready to execute. */ 2312 if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) { 2313 /* Reset the tqpair receving pdu state */ 2314 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2315 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2316 break; 2317 } 2318 2319 nvmf_tcp_set_incapsule_data(tqpair, tcp_req); 2320 2321 if (!tcp_req->has_incapsule_data) { 2322 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2323 } 2324 2325 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER); 2326 STAILQ_INSERT_TAIL(&group->pending_buf_queue, &tcp_req->req, buf_link); 2327 break; 2328 case TCP_REQUEST_STATE_NEED_BUFFER: 2329 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 0, 0, (uintptr_t)tcp_req, 0); 2330 2331 assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE); 2332 2333 if (!tcp_req->has_incapsule_data && (&tcp_req->req != STAILQ_FIRST(&group->pending_buf_queue))) { 2334 SPDK_DEBUGLOG(nvmf_tcp, 2335 "Not the first element to wait for the buf for tcp_req(%p) on tqpair=%p\n", 2336 tcp_req, tqpair); 2337 /* This request needs to wait in line to obtain a buffer */ 2338 break; 2339 } 2340 2341 /* Try to get a data buffer */ 2342 rc = nvmf_tcp_req_parse_sgl(tcp_req, transport, group); 2343 if (rc < 0) { 2344 STAILQ_REMOVE_HEAD(&group->pending_buf_queue, buf_link); 2345 /* Reset the tqpair receving pdu state */ 2346 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 2347 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2348 break; 2349 } 2350 2351 if (!tcp_req->req.data) { 2352 SPDK_DEBUGLOG(nvmf_tcp, "No buffer allocated for tcp_req(%p) on tqpair(%p\n)", 2353 tcp_req, tqpair); 2354 /* No buffers available. */ 2355 break; 2356 } 2357 2358 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2359 2360 /* If data is transferring from host to controller, we need to do a transfer from the host. */ 2361 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 2362 if (tcp_req->req.data_from_pool) { 2363 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 2364 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 2365 } else { 2366 struct nvme_tcp_pdu *pdu; 2367 2368 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 2369 2370 pdu = &tqpair->pdu_in_progress; 2371 SPDK_DEBUGLOG(nvmf_tcp, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, 2372 tqpair); 2373 /* No need to send r2t, contained in the capsuled data */ 2374 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2375 0, tcp_req->req.length); 2376 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2377 } 2378 break; 2379 } 2380 2381 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2382 break; 2383 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 2384 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 0, 0, (uintptr_t)tcp_req, 0); 2385 /* The R2T completion or the h2c data incoming will kick it out of this state. */ 2386 break; 2387 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 2388 2389 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 0, 0, 2390 (uintptr_t)tcp_req, 0); 2391 /* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE 2392 * to escape this state. */ 2393 break; 2394 case TCP_REQUEST_STATE_READY_TO_EXECUTE: 2395 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 0, 0, (uintptr_t)tcp_req, 0); 2396 2397 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2398 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 2399 tcp_req->req.length = tcp_req->req.dif.elba_length; 2400 } 2401 2402 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING); 2403 spdk_nvmf_request_exec(&tcp_req->req); 2404 break; 2405 case TCP_REQUEST_STATE_EXECUTING: 2406 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, 0, 0, (uintptr_t)tcp_req, 0); 2407 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 2408 * to escape this state. */ 2409 break; 2410 case TCP_REQUEST_STATE_EXECUTED: 2411 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, 0, 0, (uintptr_t)tcp_req, 0); 2412 2413 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2414 tcp_req->req.length = tcp_req->req.dif.orig_length; 2415 } 2416 2417 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2418 break; 2419 case TCP_REQUEST_STATE_READY_TO_COMPLETE: 2420 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 0, 0, (uintptr_t)tcp_req, 0); 2421 rc = request_transfer_out(&tcp_req->req); 2422 assert(rc == 0); /* No good way to handle this currently */ 2423 break; 2424 case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST: 2425 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 0, 0, 2426 (uintptr_t)tcp_req, 2427 0); 2428 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 2429 * to escape this state. */ 2430 break; 2431 case TCP_REQUEST_STATE_COMPLETED: 2432 spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, 0, 0, (uintptr_t)tcp_req, 0); 2433 if (tcp_req->req.data_from_pool) { 2434 spdk_nvmf_request_free_buffers(&tcp_req->req, group, transport); 2435 } else if (spdk_unlikely(tcp_req->has_incapsule_data && (tcp_req->cmd.opc == SPDK_NVME_OPC_FABRIC || 2436 tqpair->qpair.qid == 0) && tcp_req->req.length > transport->opts.in_capsule_data_size)) { 2437 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2438 assert(tgroup->control_msg_list); 2439 SPDK_DEBUGLOG(nvmf_tcp, "Put buf to control msg list\n"); 2440 nvmf_tcp_control_msg_put(tgroup->control_msg_list, tcp_req->req.data); 2441 } 2442 tcp_req->req.length = 0; 2443 tcp_req->req.iovcnt = 0; 2444 tcp_req->req.data = NULL; 2445 2446 nvmf_tcp_req_pdu_fini(tcp_req); 2447 2448 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_FREE); 2449 break; 2450 case TCP_REQUEST_NUM_STATES: 2451 default: 2452 assert(0); 2453 break; 2454 } 2455 2456 if (tcp_req->state != prev_state) { 2457 progress = true; 2458 } 2459 } while (tcp_req->state != prev_state); 2460 2461 return progress; 2462 } 2463 2464 static void 2465 nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock) 2466 { 2467 struct spdk_nvmf_tcp_qpair *tqpair = arg; 2468 int rc; 2469 2470 assert(tqpair != NULL); 2471 rc = nvmf_tcp_sock_process(tqpair); 2472 2473 /* If there was a new socket error, disconnect */ 2474 if (rc < 0) { 2475 nvmf_tcp_qpair_disconnect(tqpair); 2476 } 2477 } 2478 2479 static int 2480 nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group, 2481 struct spdk_nvmf_qpair *qpair) 2482 { 2483 struct spdk_nvmf_tcp_poll_group *tgroup; 2484 struct spdk_nvmf_tcp_qpair *tqpair; 2485 int rc; 2486 2487 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2488 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2489 2490 rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock, 2491 nvmf_tcp_sock_cb, tqpair); 2492 if (rc != 0) { 2493 SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n", 2494 spdk_strerror(errno), errno); 2495 return -1; 2496 } 2497 2498 rc = nvmf_tcp_qpair_sock_init(tqpair); 2499 if (rc != 0) { 2500 SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair); 2501 return -1; 2502 } 2503 2504 rc = nvmf_tcp_qpair_init(&tqpair->qpair); 2505 if (rc < 0) { 2506 SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair); 2507 return -1; 2508 } 2509 2510 rc = nvmf_tcp_qpair_init_mem_resource(tqpair); 2511 if (rc < 0) { 2512 SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair); 2513 return -1; 2514 } 2515 2516 tqpair->group = tgroup; 2517 tqpair->state = NVME_TCP_QPAIR_STATE_INVALID; 2518 TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link); 2519 2520 return 0; 2521 } 2522 2523 static int 2524 nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group, 2525 struct spdk_nvmf_qpair *qpair) 2526 { 2527 struct spdk_nvmf_tcp_poll_group *tgroup; 2528 struct spdk_nvmf_tcp_qpair *tqpair; 2529 int rc; 2530 2531 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2532 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2533 2534 assert(tqpair->group == tgroup); 2535 2536 SPDK_DEBUGLOG(nvmf_tcp, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup); 2537 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 2538 TAILQ_REMOVE(&tgroup->await_req, tqpair, link); 2539 } else { 2540 TAILQ_REMOVE(&tgroup->qpairs, tqpair, link); 2541 } 2542 2543 rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock); 2544 if (rc != 0) { 2545 SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n", 2546 spdk_strerror(errno), errno); 2547 } 2548 2549 return rc; 2550 } 2551 2552 static int 2553 nvmf_tcp_req_complete(struct spdk_nvmf_request *req) 2554 { 2555 struct spdk_nvmf_tcp_transport *ttransport; 2556 struct spdk_nvmf_tcp_req *tcp_req; 2557 2558 ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport); 2559 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2560 2561 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 2562 nvmf_tcp_req_process(ttransport, tcp_req); 2563 2564 return 0; 2565 } 2566 2567 static void 2568 nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair) 2569 { 2570 struct spdk_nvmf_tcp_qpair *tqpair; 2571 2572 SPDK_DEBUGLOG(nvmf_tcp, "Qpair: %p\n", qpair); 2573 2574 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2575 tqpair->state = NVME_TCP_QPAIR_STATE_EXITED; 2576 nvmf_tcp_qpair_destroy(tqpair); 2577 } 2578 2579 static int 2580 nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group) 2581 { 2582 struct spdk_nvmf_tcp_poll_group *tgroup; 2583 int rc; 2584 struct spdk_nvmf_request *req, *req_tmp; 2585 struct spdk_nvmf_tcp_req *tcp_req; 2586 struct spdk_nvmf_tcp_qpair *tqpair, *tqpair_tmp; 2587 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(group->transport, 2588 struct spdk_nvmf_tcp_transport, transport); 2589 2590 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2591 2592 if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs) && TAILQ_EMPTY(&tgroup->await_req))) { 2593 return 0; 2594 } 2595 2596 STAILQ_FOREACH_SAFE(req, &group->pending_buf_queue, buf_link, req_tmp) { 2597 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2598 if (nvmf_tcp_req_process(ttransport, tcp_req) == false) { 2599 break; 2600 } 2601 } 2602 2603 rc = spdk_sock_group_poll(tgroup->sock_group); 2604 if (rc < 0) { 2605 SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group); 2606 } 2607 2608 TAILQ_FOREACH_SAFE(tqpair, &tgroup->await_req, link, tqpair_tmp) { 2609 nvmf_tcp_sock_process(tqpair); 2610 } 2611 2612 return rc; 2613 } 2614 2615 static int 2616 nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair, 2617 struct spdk_nvme_transport_id *trid, bool peer) 2618 { 2619 struct spdk_nvmf_tcp_qpair *tqpair; 2620 uint16_t port; 2621 2622 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2623 spdk_nvme_trid_populate_transport(trid, SPDK_NVME_TRANSPORT_TCP); 2624 2625 if (peer) { 2626 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr); 2627 port = tqpair->initiator_port; 2628 } else { 2629 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr); 2630 port = tqpair->target_port; 2631 } 2632 2633 if (spdk_sock_is_ipv4(tqpair->sock)) { 2634 trid->adrfam = SPDK_NVMF_ADRFAM_IPV4; 2635 } else if (spdk_sock_is_ipv6(tqpair->sock)) { 2636 trid->adrfam = SPDK_NVMF_ADRFAM_IPV6; 2637 } else { 2638 return -1; 2639 } 2640 2641 snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port); 2642 return 0; 2643 } 2644 2645 static int 2646 nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair, 2647 struct spdk_nvme_transport_id *trid) 2648 { 2649 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2650 } 2651 2652 static int 2653 nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair, 2654 struct spdk_nvme_transport_id *trid) 2655 { 2656 return nvmf_tcp_qpair_get_trid(qpair, trid, 1); 2657 } 2658 2659 static int 2660 nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair, 2661 struct spdk_nvme_transport_id *trid) 2662 { 2663 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2664 } 2665 2666 static void 2667 nvmf_tcp_req_set_abort_status(struct spdk_nvmf_request *req, 2668 struct spdk_nvmf_tcp_req *tcp_req_to_abort) 2669 { 2670 tcp_req_to_abort->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2671 tcp_req_to_abort->req.rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST; 2672 2673 nvmf_tcp_req_set_state(tcp_req_to_abort, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2674 2675 req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command was successfully aborted. */ 2676 } 2677 2678 static int 2679 _nvmf_tcp_qpair_abort_request(void *ctx) 2680 { 2681 struct spdk_nvmf_request *req = ctx; 2682 struct spdk_nvmf_tcp_req *tcp_req_to_abort = SPDK_CONTAINEROF(req->req_to_abort, 2683 struct spdk_nvmf_tcp_req, req); 2684 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(req->req_to_abort->qpair, 2685 struct spdk_nvmf_tcp_qpair, qpair); 2686 int rc; 2687 2688 spdk_poller_unregister(&req->poller); 2689 2690 switch (tcp_req_to_abort->state) { 2691 case TCP_REQUEST_STATE_EXECUTING: 2692 rc = nvmf_ctrlr_abort_request(req); 2693 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS) { 2694 return SPDK_POLLER_BUSY; 2695 } 2696 break; 2697 2698 case TCP_REQUEST_STATE_NEED_BUFFER: 2699 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, 2700 &tcp_req_to_abort->req, spdk_nvmf_request, buf_link); 2701 2702 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 2703 break; 2704 2705 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 2706 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 2707 break; 2708 2709 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 2710 if (spdk_get_ticks() < req->timeout_tsc) { 2711 req->poller = SPDK_POLLER_REGISTER(_nvmf_tcp_qpair_abort_request, req, 0); 2712 return SPDK_POLLER_BUSY; 2713 } 2714 break; 2715 2716 default: 2717 break; 2718 } 2719 2720 spdk_nvmf_request_complete(req); 2721 return SPDK_POLLER_BUSY; 2722 } 2723 2724 static void 2725 nvmf_tcp_qpair_abort_request(struct spdk_nvmf_qpair *qpair, 2726 struct spdk_nvmf_request *req) 2727 { 2728 struct spdk_nvmf_tcp_qpair *tqpair; 2729 struct spdk_nvmf_tcp_transport *ttransport; 2730 struct spdk_nvmf_transport *transport; 2731 uint16_t cid; 2732 uint32_t i; 2733 struct spdk_nvmf_tcp_req *tcp_req_to_abort = NULL; 2734 2735 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2736 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 2737 transport = &ttransport->transport; 2738 2739 cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid; 2740 2741 for (i = 0; i < tqpair->resource_count; i++) { 2742 tcp_req_to_abort = &tqpair->reqs[i]; 2743 2744 if (tcp_req_to_abort->state != TCP_REQUEST_STATE_FREE && 2745 tcp_req_to_abort->req.cmd->nvme_cmd.cid == cid) { 2746 break; 2747 } 2748 } 2749 2750 if (tcp_req_to_abort == NULL) { 2751 spdk_nvmf_request_complete(req); 2752 return; 2753 } 2754 2755 req->req_to_abort = &tcp_req_to_abort->req; 2756 req->timeout_tsc = spdk_get_ticks() + 2757 transport->opts.abort_timeout_sec * spdk_get_ticks_hz(); 2758 req->poller = NULL; 2759 2760 _nvmf_tcp_qpair_abort_request(req); 2761 } 2762 2763 #define SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH 128 2764 #define SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH 128 2765 #define SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR 128 2766 #define SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE 4096 2767 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE 131072 2768 #define SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE 131072 2769 #define SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS 511 2770 #define SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE 32 2771 #define SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP false 2772 #define SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC 1 2773 2774 static void 2775 nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts) 2776 { 2777 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH; 2778 opts->max_qpairs_per_ctrlr = SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR; 2779 opts->in_capsule_data_size = SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE; 2780 opts->max_io_size = SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE; 2781 opts->io_unit_size = SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE; 2782 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH; 2783 opts->num_shared_buffers = SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS; 2784 opts->buf_cache_size = SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE; 2785 opts->dif_insert_or_strip = SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP; 2786 opts->abort_timeout_sec = SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC; 2787 opts->transport_specific = NULL; 2788 } 2789 2790 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = { 2791 .name = "TCP", 2792 .type = SPDK_NVME_TRANSPORT_TCP, 2793 .opts_init = nvmf_tcp_opts_init, 2794 .create = nvmf_tcp_create, 2795 .dump_opts = nvmf_tcp_dump_opts, 2796 .destroy = nvmf_tcp_destroy, 2797 2798 .listen = nvmf_tcp_listen, 2799 .stop_listen = nvmf_tcp_stop_listen, 2800 .accept = nvmf_tcp_accept, 2801 2802 .listener_discover = nvmf_tcp_discover, 2803 2804 .poll_group_create = nvmf_tcp_poll_group_create, 2805 .get_optimal_poll_group = nvmf_tcp_get_optimal_poll_group, 2806 .poll_group_destroy = nvmf_tcp_poll_group_destroy, 2807 .poll_group_add = nvmf_tcp_poll_group_add, 2808 .poll_group_remove = nvmf_tcp_poll_group_remove, 2809 .poll_group_poll = nvmf_tcp_poll_group_poll, 2810 2811 .req_free = nvmf_tcp_req_free, 2812 .req_complete = nvmf_tcp_req_complete, 2813 2814 .qpair_fini = nvmf_tcp_close_qpair, 2815 .qpair_get_local_trid = nvmf_tcp_qpair_get_local_trid, 2816 .qpair_get_peer_trid = nvmf_tcp_qpair_get_peer_trid, 2817 .qpair_get_listen_trid = nvmf_tcp_qpair_get_listen_trid, 2818 .qpair_abort_request = nvmf_tcp_qpair_abort_request, 2819 }; 2820 2821 SPDK_NVMF_TRANSPORT_REGISTER(tcp, &spdk_nvmf_transport_tcp); 2822 SPDK_LOG_REGISTER_COMPONENT(nvmf_tcp) 2823