xref: /spdk/lib/nvmf/tcp.c (revision 1fa071d332db21bf893d581a8e93b425ba788a24)
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/sock.h"
41 #include "spdk/string.h"
42 #include "spdk/trace.h"
43 #include "spdk/util.h"
44 
45 #include "spdk_internal/assert.h"
46 #include "spdk_internal/log.h"
47 #include "spdk_internal/nvme_tcp.h"
48 
49 #define NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME 16
50 #define SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY 6
51 #define SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR 4
52 
53 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp;
54 
55 /* spdk nvmf related structure */
56 enum spdk_nvmf_tcp_req_state {
57 
58 	/* The request is not currently in use */
59 	TCP_REQUEST_STATE_FREE = 0,
60 
61 	/* Initial state when request first received */
62 	TCP_REQUEST_STATE_NEW,
63 
64 	/* The request is queued until a data buffer is available. */
65 	TCP_REQUEST_STATE_NEED_BUFFER,
66 
67 	/* The request is currently transferring data from the host to the controller. */
68 	TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER,
69 
70 	/* The request is waiting for the R2T send acknowledgement. */
71 	TCP_REQUEST_STATE_AWAITING_R2T_ACK,
72 
73 	/* The request is ready to execute at the block device */
74 	TCP_REQUEST_STATE_READY_TO_EXECUTE,
75 
76 	/* The request is currently executing at the block device */
77 	TCP_REQUEST_STATE_EXECUTING,
78 
79 	/* The request finished executing at the block device */
80 	TCP_REQUEST_STATE_EXECUTED,
81 
82 	/* The request is ready to send a completion */
83 	TCP_REQUEST_STATE_READY_TO_COMPLETE,
84 
85 	/* The request is currently transferring final pdus from the controller to the host. */
86 	TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST,
87 
88 	/* The request completed and can be marked free. */
89 	TCP_REQUEST_STATE_COMPLETED,
90 
91 	/* Terminator */
92 	TCP_REQUEST_NUM_STATES,
93 };
94 
95 static const char *spdk_nvmf_tcp_term_req_fes_str[] = {
96 	"Invalid PDU Header Field",
97 	"PDU Sequence Error",
98 	"Header Digiest Error",
99 	"Data Transfer Out of Range",
100 	"R2T Limit Exceeded",
101 	"Unsupported parameter",
102 };
103 
104 #define OBJECT_NVMF_TCP_IO				0x80
105 
106 #define TRACE_GROUP_NVMF_TCP				0x5
107 #define TRACE_TCP_REQUEST_STATE_NEW					SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x0)
108 #define TRACE_TCP_REQUEST_STATE_NEED_BUFFER				SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x1)
109 #define TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER		SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x2)
110 #define TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE			SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x3)
111 #define TRACE_TCP_REQUEST_STATE_EXECUTING				SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x4)
112 #define TRACE_TCP_REQUEST_STATE_EXECUTED				SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x5)
113 #define TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE			SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x6)
114 #define TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST		SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x7)
115 #define TRACE_TCP_REQUEST_STATE_COMPLETED				SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x8)
116 #define TRACE_TCP_FLUSH_WRITEBUF_START					SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x9)
117 #define TRACE_TCP_FLUSH_WRITEBUF_DONE					SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xA)
118 #define TRACE_TCP_READ_FROM_SOCKET_DONE					SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xB)
119 #define TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK				SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xC)
120 
121 SPDK_TRACE_REGISTER_FN(nvmf_tcp_trace, "nvmf_tcp", TRACE_GROUP_NVMF_TCP)
122 {
123 	spdk_trace_register_object(OBJECT_NVMF_TCP_IO, 'r');
124 	spdk_trace_register_description("TCP_REQ_NEW",
125 					TRACE_TCP_REQUEST_STATE_NEW,
126 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 1, 1, "");
127 	spdk_trace_register_description("TCP_REQ_NEED_BUFFER",
128 					TRACE_TCP_REQUEST_STATE_NEED_BUFFER,
129 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, "");
130 	spdk_trace_register_description("TCP_REQ_TX_H_TO_C",
131 					TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER,
132 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, "");
133 	spdk_trace_register_description("TCP_REQ_RDY_TO_EXECUTE",
134 					TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE,
135 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, "");
136 	spdk_trace_register_description("TCP_REQ_EXECUTING",
137 					TRACE_TCP_REQUEST_STATE_EXECUTING,
138 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, "");
139 	spdk_trace_register_description("TCP_REQ_EXECUTED",
140 					TRACE_TCP_REQUEST_STATE_EXECUTED,
141 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, "");
142 	spdk_trace_register_description("TCP_REQ_RDY_TO_COMPLETE",
143 					TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE,
144 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, "");
145 	spdk_trace_register_description("TCP_REQ_TRANSFER_C2H",
146 					TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST,
147 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, "");
148 	spdk_trace_register_description("TCP_REQ_COMPLETED",
149 					TRACE_TCP_REQUEST_STATE_COMPLETED,
150 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, "");
151 	spdk_trace_register_description("TCP_WRITE_START",
152 					TRACE_TCP_FLUSH_WRITEBUF_START,
153 					OWNER_NONE, OBJECT_NONE, 0, 0, "");
154 	spdk_trace_register_description("TCP_WRITE_DONE",
155 					TRACE_TCP_FLUSH_WRITEBUF_DONE,
156 					OWNER_NONE, OBJECT_NONE, 0, 0, "");
157 	spdk_trace_register_description("TCP_READ_DONE",
158 					TRACE_TCP_READ_FROM_SOCKET_DONE,
159 					OWNER_NONE, OBJECT_NONE, 0, 0, "");
160 	spdk_trace_register_description("TCP_REQ_AWAIT_R2T_ACK",
161 					TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK,
162 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, "");
163 }
164 
165 struct spdk_nvmf_tcp_req  {
166 	struct spdk_nvmf_request		req;
167 	struct spdk_nvme_cpl			rsp;
168 	struct spdk_nvme_cmd			cmd;
169 
170 	/* A PDU that can be used for sending responses. This is
171 	 * not the incoming PDU! */
172 	struct nvme_tcp_pdu			*pdu;
173 
174 	/*
175 	 * The PDU for a request may be used multiple times in serial over
176 	 * the request's lifetime. For example, first to send an R2T, then
177 	 * to send a completion. To catch mistakes where the PDU is used
178 	 * twice at the same time, add a debug flag here for init/fini.
179 	 */
180 	bool					pdu_in_use;
181 
182 	/* In-capsule data buffer */
183 	uint8_t					*buf;
184 
185 	bool					has_incapsule_data;
186 
187 	/* transfer_tag */
188 	uint16_t				ttag;
189 
190 	enum spdk_nvmf_tcp_req_state		state;
191 
192 	/*
193 	 * h2c_offset is used when we receive the h2c_data PDU.
194 	 */
195 	uint32_t				h2c_offset;
196 
197 	STAILQ_ENTRY(spdk_nvmf_tcp_req)		link;
198 	TAILQ_ENTRY(spdk_nvmf_tcp_req)		state_link;
199 };
200 
201 struct spdk_nvmf_tcp_qpair {
202 	struct spdk_nvmf_qpair			qpair;
203 	struct spdk_nvmf_tcp_poll_group		*group;
204 	struct spdk_nvmf_tcp_port		*port;
205 	struct spdk_sock			*sock;
206 
207 	enum nvme_tcp_pdu_recv_state		recv_state;
208 	enum nvme_tcp_qpair_state		state;
209 
210 	/* PDU being actively received */
211 	struct nvme_tcp_pdu			pdu_in_progress;
212 	uint32_t				recv_buf_size;
213 
214 	/* This is a spare PDU used for sending special management
215 	 * operations. Primarily, this is used for the initial
216 	 * connection response and c2h termination request. */
217 	struct nvme_tcp_pdu			mgmt_pdu;
218 
219 	TAILQ_HEAD(, nvme_tcp_pdu)		send_queue;
220 
221 	/* Arrays of in-capsule buffers, requests, and pdus.
222 	 * Each array is 'resource_count' number of elements */
223 	void					*bufs;
224 	struct spdk_nvmf_tcp_req		*reqs;
225 	struct nvme_tcp_pdu			*pdus;
226 	uint32_t				resource_count;
227 
228 	/* Queues to track the requests in all states */
229 	TAILQ_HEAD(, spdk_nvmf_tcp_req)		state_queue[TCP_REQUEST_NUM_STATES];
230 	/* Number of requests in each state */
231 	uint32_t				state_cntr[TCP_REQUEST_NUM_STATES];
232 
233 	uint8_t					cpda;
234 
235 	bool					host_hdgst_enable;
236 	bool					host_ddgst_enable;
237 
238 	/* IP address */
239 	char					initiator_addr[SPDK_NVMF_TRADDR_MAX_LEN];
240 	char					target_addr[SPDK_NVMF_TRADDR_MAX_LEN];
241 
242 	/* IP port */
243 	uint16_t				initiator_port;
244 	uint16_t				target_port;
245 
246 	/* Timer used to destroy qpair after detecting transport error issue if initiator does
247 	 *  not close the connection.
248 	 */
249 	struct spdk_poller			*timeout_poller;
250 
251 	TAILQ_ENTRY(spdk_nvmf_tcp_qpair)	link;
252 };
253 
254 struct spdk_nvmf_tcp_poll_group {
255 	struct spdk_nvmf_transport_poll_group	group;
256 	struct spdk_sock_group			*sock_group;
257 
258 	TAILQ_HEAD(, spdk_nvmf_tcp_qpair)	qpairs;
259 	TAILQ_HEAD(, spdk_nvmf_tcp_qpair)	await_req;
260 };
261 
262 struct spdk_nvmf_tcp_port {
263 	const struct spdk_nvme_transport_id	*trid;
264 	struct spdk_sock			*listen_sock;
265 	TAILQ_ENTRY(spdk_nvmf_tcp_port)		link;
266 };
267 
268 struct spdk_nvmf_tcp_transport {
269 	struct spdk_nvmf_transport		transport;
270 
271 	pthread_mutex_t				lock;
272 
273 	TAILQ_HEAD(, spdk_nvmf_tcp_port)	ports;
274 };
275 
276 static bool spdk_nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport,
277 				      struct spdk_nvmf_tcp_req *tcp_req);
278 
279 static void
280 spdk_nvmf_tcp_req_set_state(struct spdk_nvmf_tcp_req *tcp_req,
281 			    enum spdk_nvmf_tcp_req_state state)
282 {
283 	struct spdk_nvmf_qpair *qpair;
284 	struct spdk_nvmf_tcp_qpair *tqpair;
285 
286 	qpair = tcp_req->req.qpair;
287 	tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
288 
289 	TAILQ_REMOVE(&tqpair->state_queue[tcp_req->state], tcp_req, state_link);
290 	assert(tqpair->state_cntr[tcp_req->state] > 0);
291 	tqpair->state_cntr[tcp_req->state]--;
292 
293 	TAILQ_INSERT_TAIL(&tqpair->state_queue[state], tcp_req, state_link);
294 	tqpair->state_cntr[state]++;
295 
296 	tcp_req->state = state;
297 }
298 
299 static inline struct nvme_tcp_pdu *
300 nvmf_tcp_req_pdu_init(struct spdk_nvmf_tcp_req *tcp_req)
301 {
302 	assert(tcp_req->pdu_in_use == false);
303 	tcp_req->pdu_in_use = true;
304 
305 	memset(tcp_req->pdu, 0, sizeof(*tcp_req->pdu));
306 	tcp_req->pdu->qpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair);
307 
308 	return tcp_req->pdu;
309 }
310 
311 static inline void
312 nvmf_tcp_req_pdu_fini(struct spdk_nvmf_tcp_req *tcp_req)
313 {
314 	tcp_req->pdu_in_use = false;
315 }
316 
317 static struct spdk_nvmf_tcp_req *
318 spdk_nvmf_tcp_req_get(struct spdk_nvmf_tcp_qpair *tqpair)
319 {
320 	struct spdk_nvmf_tcp_req *tcp_req;
321 
322 	tcp_req = TAILQ_FIRST(&tqpair->state_queue[TCP_REQUEST_STATE_FREE]);
323 	if (!tcp_req) {
324 		return NULL;
325 	}
326 
327 	memset(&tcp_req->rsp, 0, sizeof(tcp_req->rsp));
328 	tcp_req->h2c_offset = 0;
329 	tcp_req->has_incapsule_data = false;
330 	tcp_req->req.dif.dif_insert_or_strip = false;
331 
332 	spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW);
333 	return tcp_req;
334 }
335 
336 static void
337 nvmf_tcp_request_free(struct spdk_nvmf_tcp_req *tcp_req)
338 {
339 	struct spdk_nvmf_tcp_transport *ttransport;
340 
341 	assert(tcp_req != NULL);
342 
343 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tcp_req=%p will be freed\n", tcp_req);
344 	ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport,
345 				      struct spdk_nvmf_tcp_transport, transport);
346 	spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED);
347 	spdk_nvmf_tcp_req_process(ttransport, tcp_req);
348 }
349 
350 static int
351 spdk_nvmf_tcp_req_free(struct spdk_nvmf_request *req)
352 {
353 	struct spdk_nvmf_tcp_req *tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req);
354 
355 	nvmf_tcp_request_free(tcp_req);
356 
357 	return 0;
358 }
359 
360 static void
361 spdk_nvmf_tcp_drain_state_queue(struct spdk_nvmf_tcp_qpair *tqpair,
362 				enum spdk_nvmf_tcp_req_state state)
363 {
364 	struct spdk_nvmf_tcp_req *tcp_req, *req_tmp;
365 
366 	TAILQ_FOREACH_SAFE(tcp_req, &tqpair->state_queue[state], state_link, req_tmp) {
367 		nvmf_tcp_request_free(tcp_req);
368 	}
369 }
370 
371 static void
372 spdk_nvmf_tcp_cleanup_all_states(struct spdk_nvmf_tcp_qpair *tqpair)
373 {
374 	struct spdk_nvmf_tcp_req *tcp_req, *req_tmp;
375 
376 	assert(TAILQ_EMPTY(&tqpair->send_queue));
377 
378 	spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST);
379 	spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEW);
380 
381 	/* Wipe the requests waiting for buffer from the global list */
382 	TAILQ_FOREACH_SAFE(tcp_req, &tqpair->state_queue[TCP_REQUEST_STATE_NEED_BUFFER], state_link,
383 			   req_tmp) {
384 		STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, &tcp_req->req,
385 			      spdk_nvmf_request, buf_link);
386 	}
387 
388 	spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEED_BUFFER);
389 	spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_EXECUTING);
390 	spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER);
391 	spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK);
392 }
393 
394 static void
395 nvmf_tcp_dump_qpair_req_contents(struct spdk_nvmf_tcp_qpair *tqpair)
396 {
397 	int i;
398 	struct spdk_nvmf_tcp_req *tcp_req;
399 
400 	SPDK_ERRLOG("Dumping contents of queue pair (QID %d)\n", tqpair->qpair.qid);
401 	for (i = 1; i < TCP_REQUEST_NUM_STATES; i++) {
402 		SPDK_ERRLOG("\tNum of requests in state[%d] = %u\n", i, tqpair->state_cntr[i]);
403 		TAILQ_FOREACH(tcp_req, &tqpair->state_queue[i], state_link) {
404 			SPDK_ERRLOG("\t\tRequest Data From Pool: %d\n", tcp_req->req.data_from_pool);
405 			SPDK_ERRLOG("\t\tRequest opcode: %d\n", tcp_req->req.cmd->nvmf_cmd.opcode);
406 		}
407 	}
408 }
409 
410 static void
411 spdk_nvmf_tcp_qpair_destroy(struct spdk_nvmf_tcp_qpair *tqpair)
412 {
413 	int err = 0;
414 
415 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n");
416 
417 	err = spdk_sock_close(&tqpair->sock);
418 	assert(err == 0);
419 	spdk_nvmf_tcp_cleanup_all_states(tqpair);
420 
421 	if (tqpair->state_cntr[TCP_REQUEST_STATE_FREE] != tqpair->resource_count) {
422 		SPDK_ERRLOG("tqpair(%p) free tcp request num is %u but should be %u\n", tqpair,
423 			    tqpair->state_cntr[TCP_REQUEST_STATE_FREE],
424 			    tqpair->resource_count);
425 		err++;
426 	}
427 
428 	if (err > 0) {
429 		nvmf_tcp_dump_qpair_req_contents(tqpair);
430 	}
431 
432 	spdk_dma_free(tqpair->pdus);
433 	free(tqpair->reqs);
434 	spdk_free(tqpair->bufs);
435 	free(tqpair);
436 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Leave\n");
437 }
438 
439 static int
440 spdk_nvmf_tcp_destroy(struct spdk_nvmf_transport *transport)
441 {
442 	struct spdk_nvmf_tcp_transport	*ttransport;
443 
444 	assert(transport != NULL);
445 	ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport);
446 
447 	pthread_mutex_destroy(&ttransport->lock);
448 	free(ttransport);
449 	return 0;
450 }
451 
452 static struct spdk_nvmf_transport *
453 spdk_nvmf_tcp_create(struct spdk_nvmf_transport_opts *opts)
454 {
455 	struct spdk_nvmf_tcp_transport *ttransport;
456 	uint32_t sge_count;
457 	uint32_t min_shared_buffers;
458 
459 	ttransport = calloc(1, sizeof(*ttransport));
460 	if (!ttransport) {
461 		return NULL;
462 	}
463 
464 	TAILQ_INIT(&ttransport->ports);
465 
466 	ttransport->transport.ops = &spdk_nvmf_transport_tcp;
467 
468 	SPDK_NOTICELOG("*** TCP Transport Init ***\n");
469 
470 	SPDK_INFOLOG(SPDK_LOG_NVMF_TCP, "*** TCP Transport Init ***\n"
471 		     "  Transport opts:  max_ioq_depth=%d, max_io_size=%d,\n"
472 		     "  max_qpairs_per_ctrlr=%d, io_unit_size=%d,\n"
473 		     "  in_capsule_data_size=%d, max_aq_depth=%d\n"
474 		     "  num_shared_buffers=%d, c2h_success=%d,\n"
475 		     "  dif_insert_or_strip=%d, sock_priority=%d\n",
476 		     opts->max_queue_depth,
477 		     opts->max_io_size,
478 		     opts->max_qpairs_per_ctrlr,
479 		     opts->io_unit_size,
480 		     opts->in_capsule_data_size,
481 		     opts->max_aq_depth,
482 		     opts->num_shared_buffers,
483 		     opts->c2h_success,
484 		     opts->dif_insert_or_strip,
485 		     opts->sock_priority);
486 
487 	if (opts->sock_priority > SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY) {
488 		SPDK_ERRLOG("Unsupported socket_priority=%d, the current range is: 0 to %d\n"
489 			    "you can use man 7 socket to view the range of priority under SO_PRIORITY item\n",
490 			    opts->sock_priority, SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY);
491 		free(ttransport);
492 		return NULL;
493 	}
494 
495 	/* I/O unit size cannot be larger than max I/O size */
496 	if (opts->io_unit_size > opts->max_io_size) {
497 		opts->io_unit_size = opts->max_io_size;
498 	}
499 
500 	sge_count = opts->max_io_size / opts->io_unit_size;
501 	if (sge_count > SPDK_NVMF_MAX_SGL_ENTRIES) {
502 		SPDK_ERRLOG("Unsupported IO Unit size specified, %d bytes\n", opts->io_unit_size);
503 		free(ttransport);
504 		return NULL;
505 	}
506 
507 	min_shared_buffers = spdk_thread_get_count() * opts->buf_cache_size;
508 	if (min_shared_buffers > opts->num_shared_buffers) {
509 		SPDK_ERRLOG("There are not enough buffers to satisfy"
510 			    "per-poll group caches for each thread. (%" PRIu32 ")"
511 			    "supplied. (%" PRIu32 ") required\n", opts->num_shared_buffers, min_shared_buffers);
512 		SPDK_ERRLOG("Please specify a larger number of shared buffers\n");
513 		spdk_nvmf_tcp_destroy(&ttransport->transport);
514 		return NULL;
515 	}
516 
517 	pthread_mutex_init(&ttransport->lock, NULL);
518 
519 	return &ttransport->transport;
520 }
521 
522 static int
523 _spdk_nvmf_tcp_trsvcid_to_int(const char *trsvcid)
524 {
525 	unsigned long long ull;
526 	char *end = NULL;
527 
528 	ull = strtoull(trsvcid, &end, 10);
529 	if (end == NULL || end == trsvcid || *end != '\0') {
530 		return -1;
531 	}
532 
533 	/* Valid TCP/IP port numbers are in [0, 65535] */
534 	if (ull > 65535) {
535 		return -1;
536 	}
537 
538 	return (int)ull;
539 }
540 
541 /**
542  * Canonicalize a listen address trid.
543  */
544 static int
545 _spdk_nvmf_tcp_canon_listen_trid(struct spdk_nvme_transport_id *canon_trid,
546 				 const struct spdk_nvme_transport_id *trid)
547 {
548 	int trsvcid_int;
549 
550 	trsvcid_int = _spdk_nvmf_tcp_trsvcid_to_int(trid->trsvcid);
551 	if (trsvcid_int < 0) {
552 		return -EINVAL;
553 	}
554 
555 	memset(canon_trid, 0, sizeof(*canon_trid));
556 	spdk_nvme_trid_populate_transport(canon_trid, SPDK_NVME_TRANSPORT_TCP);
557 	canon_trid->adrfam = trid->adrfam;
558 	snprintf(canon_trid->traddr, sizeof(canon_trid->traddr), "%s", trid->traddr);
559 	snprintf(canon_trid->trsvcid, sizeof(canon_trid->trsvcid), "%d", trsvcid_int);
560 
561 	return 0;
562 }
563 
564 /**
565  * Find an existing listening port.
566  *
567  * Caller must hold ttransport->lock.
568  */
569 static struct spdk_nvmf_tcp_port *
570 _spdk_nvmf_tcp_find_port(struct spdk_nvmf_tcp_transport *ttransport,
571 			 const struct spdk_nvme_transport_id *trid)
572 {
573 	struct spdk_nvme_transport_id canon_trid;
574 	struct spdk_nvmf_tcp_port *port;
575 
576 	if (_spdk_nvmf_tcp_canon_listen_trid(&canon_trid, trid) != 0) {
577 		return NULL;
578 	}
579 
580 	TAILQ_FOREACH(port, &ttransport->ports, link) {
581 		if (spdk_nvme_transport_id_compare(&canon_trid, port->trid) == 0) {
582 			return port;
583 		}
584 	}
585 
586 	return NULL;
587 }
588 
589 static int
590 spdk_nvmf_tcp_listen(struct spdk_nvmf_transport *transport,
591 		     const struct spdk_nvme_transport_id *trid)
592 {
593 	struct spdk_nvmf_tcp_transport *ttransport;
594 	struct spdk_nvmf_tcp_port *port;
595 	int trsvcid_int;
596 	uint8_t adrfam;
597 
598 	ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport);
599 
600 	trsvcid_int = _spdk_nvmf_tcp_trsvcid_to_int(trid->trsvcid);
601 	if (trsvcid_int < 0) {
602 		SPDK_ERRLOG("Invalid trsvcid '%s'\n", trid->trsvcid);
603 		return -EINVAL;
604 	}
605 
606 	pthread_mutex_lock(&ttransport->lock);
607 	port = calloc(1, sizeof(*port));
608 	if (!port) {
609 		SPDK_ERRLOG("Port allocation failed\n");
610 		pthread_mutex_unlock(&ttransport->lock);
611 		return -ENOMEM;
612 	}
613 
614 	port->trid = trid;
615 	port->listen_sock = spdk_sock_listen(trid->traddr, trsvcid_int, NULL);
616 	if (port->listen_sock == NULL) {
617 		SPDK_ERRLOG("spdk_sock_listen(%s, %d) failed: %s (%d)\n",
618 			    trid->traddr, trsvcid_int,
619 			    spdk_strerror(errno), errno);
620 		free(port);
621 		pthread_mutex_unlock(&ttransport->lock);
622 		return -errno;
623 	}
624 
625 	if (spdk_sock_is_ipv4(port->listen_sock)) {
626 		adrfam = SPDK_NVMF_ADRFAM_IPV4;
627 	} else if (spdk_sock_is_ipv6(port->listen_sock)) {
628 		adrfam = SPDK_NVMF_ADRFAM_IPV6;
629 	} else {
630 		SPDK_ERRLOG("Unhandled socket type\n");
631 		adrfam = 0;
632 	}
633 
634 	if (adrfam != trid->adrfam) {
635 		SPDK_ERRLOG("Socket address family mismatch\n");
636 		spdk_sock_close(&port->listen_sock);
637 		free(port);
638 		pthread_mutex_unlock(&ttransport->lock);
639 		return -EINVAL;
640 	}
641 
642 	SPDK_NOTICELOG("*** NVMe/TCP Target Listening on %s port %s ***\n",
643 		       trid->traddr, trid->trsvcid);
644 
645 	TAILQ_INSERT_TAIL(&ttransport->ports, port, link);
646 	pthread_mutex_unlock(&ttransport->lock);
647 	return 0;
648 }
649 
650 static void
651 spdk_nvmf_tcp_stop_listen(struct spdk_nvmf_transport *transport,
652 			  const struct spdk_nvme_transport_id *trid)
653 {
654 	struct spdk_nvmf_tcp_transport *ttransport;
655 	struct spdk_nvmf_tcp_port *port;
656 
657 	ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport);
658 
659 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Removing listen address %s port %s\n",
660 		      trid->traddr, trid->trsvcid);
661 
662 	pthread_mutex_lock(&ttransport->lock);
663 	port = _spdk_nvmf_tcp_find_port(ttransport, trid);
664 	if (port) {
665 		TAILQ_REMOVE(&ttransport->ports, port, link);
666 		spdk_sock_close(&port->listen_sock);
667 		free(port);
668 	}
669 
670 	pthread_mutex_unlock(&ttransport->lock);
671 }
672 
673 static void spdk_nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair,
674 		enum nvme_tcp_pdu_recv_state state);
675 
676 static void
677 spdk_nvmf_tcp_qpair_disconnect(struct spdk_nvmf_tcp_qpair *tqpair)
678 {
679 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Disconnecting qpair %p\n", tqpair);
680 
681 	if (tqpair->state <= NVME_TCP_QPAIR_STATE_RUNNING) {
682 		tqpair->state = NVME_TCP_QPAIR_STATE_EXITING;
683 		spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR);
684 		spdk_poller_unregister(&tqpair->timeout_poller);
685 
686 		/* This will end up calling spdk_nvmf_tcp_close_qpair */
687 		spdk_nvmf_qpair_disconnect(&tqpair->qpair, NULL, NULL);
688 	}
689 }
690 
691 static void
692 _pdu_write_done(void *_pdu, int err)
693 {
694 	struct nvme_tcp_pdu			*pdu = _pdu;
695 	struct spdk_nvmf_tcp_qpair		*tqpair = pdu->qpair;
696 
697 	TAILQ_REMOVE(&tqpair->send_queue, pdu, tailq);
698 
699 	if (err != 0) {
700 		spdk_nvmf_tcp_qpair_disconnect(tqpair);
701 		return;
702 	}
703 
704 	assert(pdu->cb_fn != NULL);
705 	pdu->cb_fn(pdu->cb_arg);
706 }
707 
708 static void
709 spdk_nvmf_tcp_qpair_write_pdu(struct spdk_nvmf_tcp_qpair *tqpair,
710 			      struct nvme_tcp_pdu *pdu,
711 			      nvme_tcp_qpair_xfer_complete_cb cb_fn,
712 			      void *cb_arg)
713 {
714 	int hlen;
715 	uint32_t crc32c;
716 	uint32_t mapped_length = 0;
717 	ssize_t rc;
718 
719 	assert(&tqpair->pdu_in_progress != pdu);
720 
721 	hlen = pdu->hdr.common.hlen;
722 
723 	/* Header Digest */
724 	if (g_nvme_tcp_hdgst[pdu->hdr.common.pdu_type] && tqpair->host_hdgst_enable) {
725 		crc32c = nvme_tcp_pdu_calc_header_digest(pdu);
726 		MAKE_DIGEST_WORD((uint8_t *)pdu->hdr.raw + hlen, crc32c);
727 	}
728 
729 	/* Data Digest */
730 	if (pdu->data_len > 0 && g_nvme_tcp_ddgst[pdu->hdr.common.pdu_type] && tqpair->host_ddgst_enable) {
731 		crc32c = nvme_tcp_pdu_calc_data_digest(pdu);
732 		MAKE_DIGEST_WORD(pdu->data_digest, crc32c);
733 	}
734 
735 	pdu->cb_fn = cb_fn;
736 	pdu->cb_arg = cb_arg;
737 
738 	pdu->sock_req.iovcnt = nvme_tcp_build_iovs(pdu->iov, SPDK_COUNTOF(pdu->iov), pdu,
739 			       tqpair->host_hdgst_enable, tqpair->host_ddgst_enable,
740 			       &mapped_length);
741 	pdu->sock_req.cb_fn = _pdu_write_done;
742 	pdu->sock_req.cb_arg = pdu;
743 	TAILQ_INSERT_TAIL(&tqpair->send_queue, pdu, tailq);
744 	if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP ||
745 	    pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ) {
746 		rc = spdk_sock_writev(tqpair->sock, pdu->iov, pdu->sock_req.iovcnt);
747 		if (rc == mapped_length) {
748 			_pdu_write_done(pdu, 0);
749 		} else {
750 			SPDK_ERRLOG("IC_RESP or TERM_REQ could not write to socket.\n");
751 			_pdu_write_done(pdu, -1);
752 		}
753 	} else {
754 		spdk_sock_writev_async(tqpair->sock, &pdu->sock_req);
755 	}
756 }
757 
758 static int
759 spdk_nvmf_tcp_qpair_init_mem_resource(struct spdk_nvmf_tcp_qpair *tqpair)
760 {
761 	uint32_t i;
762 	struct spdk_nvmf_transport_opts *opts;
763 	uint32_t in_capsule_data_size;
764 
765 	opts = &tqpair->qpair.transport->opts;
766 
767 	in_capsule_data_size = opts->in_capsule_data_size;
768 	if (opts->dif_insert_or_strip) {
769 		in_capsule_data_size = SPDK_BDEV_BUF_SIZE_WITH_MD(in_capsule_data_size);
770 	}
771 
772 	tqpair->resource_count = opts->max_queue_depth;
773 
774 	tqpair->mgmt_pdu.qpair = tqpair;
775 
776 	tqpair->reqs = calloc(tqpair->resource_count, sizeof(*tqpair->reqs));
777 	if (!tqpair->reqs) {
778 		SPDK_ERRLOG("Unable to allocate reqs on tqpair=%p\n", tqpair);
779 		return -1;
780 	}
781 
782 	if (in_capsule_data_size) {
783 		tqpair->bufs = spdk_zmalloc(tqpair->resource_count * in_capsule_data_size, 0x1000,
784 					    NULL, SPDK_ENV_LCORE_ID_ANY,
785 					    SPDK_MALLOC_DMA);
786 		if (!tqpair->bufs) {
787 			SPDK_ERRLOG("Unable to allocate bufs on tqpair=%p.\n", tqpair);
788 			return -1;
789 		}
790 	}
791 
792 	tqpair->pdus = spdk_dma_malloc(tqpair->resource_count * sizeof(*tqpair->pdus), 0x1000, NULL);
793 	if (!tqpair->pdus) {
794 		SPDK_ERRLOG("Unable to allocate pdu pool on tqpair =%p.\n", tqpair);
795 		return -1;
796 	}
797 
798 	for (i = 0; i < tqpair->resource_count; i++) {
799 		struct spdk_nvmf_tcp_req *tcp_req = &tqpair->reqs[i];
800 
801 		tcp_req->ttag = i + 1;
802 		tcp_req->req.qpair = &tqpair->qpair;
803 
804 		tcp_req->pdu = &tqpair->pdus[i];
805 		tcp_req->pdu->qpair = tqpair;
806 
807 		/* Set up memory to receive commands */
808 		if (tqpair->bufs) {
809 			tcp_req->buf = (void *)((uintptr_t)tqpair->bufs + (i * in_capsule_data_size));
810 		}
811 
812 		/* Set the cmdn and rsp */
813 		tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp;
814 		tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd;
815 
816 		/* Initialize request state to FREE */
817 		tcp_req->state = TCP_REQUEST_STATE_FREE;
818 		TAILQ_INSERT_TAIL(&tqpair->state_queue[tcp_req->state], tcp_req, state_link);
819 		tqpair->state_cntr[TCP_REQUEST_STATE_FREE]++;
820 	}
821 
822 	tqpair->recv_buf_size = (in_capsule_data_size + sizeof(struct spdk_nvme_tcp_cmd) + 2 *
823 				 SPDK_NVME_TCP_DIGEST_LEN) * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR;
824 
825 	return 0;
826 }
827 
828 static int
829 spdk_nvmf_tcp_qpair_init(struct spdk_nvmf_qpair *qpair)
830 {
831 	struct spdk_nvmf_tcp_qpair *tqpair;
832 	int i;
833 
834 	tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
835 
836 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "New TCP Connection: %p\n", qpair);
837 
838 	TAILQ_INIT(&tqpair->send_queue);
839 
840 	/* Initialise request state queues of the qpair */
841 	for (i = TCP_REQUEST_STATE_FREE; i < TCP_REQUEST_NUM_STATES; i++) {
842 		TAILQ_INIT(&tqpair->state_queue[i]);
843 	}
844 
845 	tqpair->host_hdgst_enable = true;
846 	tqpair->host_ddgst_enable = true;
847 
848 	return 0;
849 }
850 
851 static int
852 spdk_nvmf_tcp_qpair_sock_init(struct spdk_nvmf_tcp_qpair *tqpair)
853 {
854 	int rc;
855 
856 	/* set low water mark */
857 	rc = spdk_sock_set_recvlowat(tqpair->sock, sizeof(struct spdk_nvme_tcp_c2h_data_hdr));
858 	if (rc != 0) {
859 		SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n");
860 		return rc;
861 	}
862 
863 	return 0;
864 }
865 
866 static void
867 _spdk_nvmf_tcp_handle_connect(struct spdk_nvmf_transport *transport,
868 			      struct spdk_nvmf_tcp_port *port,
869 			      struct spdk_sock *sock,
870 			      new_qpair_fn cb_fn, void *cb_arg)
871 {
872 	struct spdk_nvmf_tcp_qpair *tqpair;
873 	int rc;
874 
875 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "New connection accepted on %s port %s\n",
876 		      port->trid->traddr, port->trid->trsvcid);
877 
878 	tqpair = calloc(1, sizeof(struct spdk_nvmf_tcp_qpair));
879 	if (tqpair == NULL) {
880 		SPDK_ERRLOG("Could not allocate new connection.\n");
881 		spdk_sock_close(&sock);
882 		return;
883 	}
884 
885 	tqpair->sock = sock;
886 	tqpair->state_cntr[TCP_REQUEST_STATE_FREE] = 0;
887 	tqpair->port = port;
888 	tqpair->qpair.transport = transport;
889 
890 	rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr,
891 			       sizeof(tqpair->target_addr), &tqpair->target_port,
892 			       tqpair->initiator_addr, sizeof(tqpair->initiator_addr),
893 			       &tqpair->initiator_port);
894 	if (rc < 0) {
895 		SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair);
896 		spdk_nvmf_tcp_qpair_destroy(tqpair);
897 		return;
898 	}
899 
900 	cb_fn(&tqpair->qpair, cb_arg);
901 }
902 
903 static void
904 spdk_nvmf_tcp_port_accept(struct spdk_nvmf_transport *transport, struct spdk_nvmf_tcp_port *port,
905 			  new_qpair_fn cb_fn, void *cb_arg)
906 {
907 	struct spdk_sock *sock;
908 	int i;
909 
910 	for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) {
911 		sock = spdk_sock_accept(port->listen_sock);
912 		if (sock == NULL) {
913 			break;
914 		}
915 		_spdk_nvmf_tcp_handle_connect(transport, port, sock, cb_fn, cb_arg);
916 	}
917 }
918 
919 static void
920 spdk_nvmf_tcp_accept(struct spdk_nvmf_transport *transport, new_qpair_fn cb_fn, void *cb_arg)
921 {
922 	struct spdk_nvmf_tcp_transport *ttransport;
923 	struct spdk_nvmf_tcp_port *port;
924 
925 	ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport);
926 
927 	TAILQ_FOREACH(port, &ttransport->ports, link) {
928 		spdk_nvmf_tcp_port_accept(transport, port, cb_fn, cb_arg);
929 	}
930 }
931 
932 static void
933 spdk_nvmf_tcp_discover(struct spdk_nvmf_transport *transport,
934 		       struct spdk_nvme_transport_id *trid,
935 		       struct spdk_nvmf_discovery_log_page_entry *entry)
936 {
937 	entry->trtype = SPDK_NVMF_TRTYPE_TCP;
938 	entry->adrfam = trid->adrfam;
939 	entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_REQUIRED;
940 
941 	spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' ');
942 	spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' ');
943 
944 	entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE;
945 }
946 
947 static struct spdk_nvmf_transport_poll_group *
948 spdk_nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport)
949 {
950 	struct spdk_nvmf_tcp_poll_group *tgroup;
951 
952 	tgroup = calloc(1, sizeof(*tgroup));
953 	if (!tgroup) {
954 		return NULL;
955 	}
956 
957 	tgroup->sock_group = spdk_sock_group_create(&tgroup->group);
958 	if (!tgroup->sock_group) {
959 		goto cleanup;
960 	}
961 
962 	TAILQ_INIT(&tgroup->qpairs);
963 	TAILQ_INIT(&tgroup->await_req);
964 
965 	return &tgroup->group;
966 
967 cleanup:
968 	free(tgroup);
969 	return NULL;
970 }
971 
972 static struct spdk_nvmf_transport_poll_group *
973 spdk_nvmf_tcp_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair)
974 {
975 	struct spdk_nvmf_tcp_qpair *tqpair;
976 	struct spdk_sock_group *group = NULL;
977 	int rc;
978 
979 	tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
980 	rc = spdk_sock_get_optimal_sock_group(tqpair->sock, &group);
981 	if (!rc && group != NULL) {
982 		return spdk_sock_group_get_ctx(group);
983 	}
984 
985 	return NULL;
986 }
987 
988 static void
989 spdk_nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group)
990 {
991 	struct spdk_nvmf_tcp_poll_group *tgroup;
992 
993 	tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group);
994 	spdk_sock_group_close(&tgroup->sock_group);
995 
996 	free(tgroup);
997 }
998 
999 static void
1000 spdk_nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair,
1001 				   enum nvme_tcp_pdu_recv_state state)
1002 {
1003 	if (tqpair->recv_state == state) {
1004 		SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n",
1005 			    tqpair, state);
1006 		return;
1007 	}
1008 
1009 	if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) {
1010 		/* When leaving the await req state, move the qpair to the main list */
1011 		TAILQ_REMOVE(&tqpair->group->await_req, tqpair, link);
1012 		TAILQ_INSERT_TAIL(&tqpair->group->qpairs, tqpair, link);
1013 	}
1014 
1015 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tqpair(%p) recv state=%d\n", tqpair, state);
1016 	tqpair->recv_state = state;
1017 
1018 	switch (state) {
1019 	case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH:
1020 	case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH:
1021 	case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD:
1022 		break;
1023 	case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ:
1024 		TAILQ_REMOVE(&tqpair->group->qpairs, tqpair, link);
1025 		TAILQ_INSERT_TAIL(&tqpair->group->await_req, tqpair, link);
1026 		break;
1027 	case NVME_TCP_PDU_RECV_STATE_ERROR:
1028 	case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY:
1029 		memset(&tqpair->pdu_in_progress, 0, sizeof(tqpair->pdu_in_progress));
1030 		break;
1031 	default:
1032 		SPDK_ERRLOG("The state(%d) is invalid\n", state);
1033 		abort();
1034 		break;
1035 	}
1036 }
1037 
1038 static int
1039 spdk_nvmf_tcp_qpair_handle_timeout(void *ctx)
1040 {
1041 	struct spdk_nvmf_tcp_qpair *tqpair = ctx;
1042 
1043 	assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR);
1044 
1045 	SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair,
1046 		    SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT);
1047 
1048 	spdk_nvmf_tcp_qpair_disconnect(tqpair);
1049 	return 0;
1050 }
1051 
1052 static void
1053 spdk_nvmf_tcp_send_c2h_term_req_complete(void *cb_arg)
1054 {
1055 	struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg;
1056 
1057 	if (!tqpair->timeout_poller) {
1058 		tqpair->timeout_poller = SPDK_POLLER_REGISTER(spdk_nvmf_tcp_qpair_handle_timeout, tqpair,
1059 					 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000);
1060 	}
1061 }
1062 
1063 static void
1064 spdk_nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu,
1065 				enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset)
1066 {
1067 	struct nvme_tcp_pdu *rsp_pdu;
1068 	struct spdk_nvme_tcp_term_req_hdr *c2h_term_req;
1069 	uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req);
1070 	uint32_t copy_len;
1071 
1072 	rsp_pdu = &tqpair->mgmt_pdu;
1073 
1074 	c2h_term_req = &rsp_pdu->hdr.term_req;
1075 	c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ;
1076 	c2h_term_req->common.hlen = c2h_term_req_hdr_len;
1077 
1078 	if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) ||
1079 	    (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) {
1080 		DSET32(&c2h_term_req->fei, error_offset);
1081 	}
1082 
1083 	copy_len = spdk_min(pdu->hdr.common.hlen, SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE);
1084 
1085 	/* Copy the error info into the buffer */
1086 	memcpy((uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, pdu->hdr.raw, copy_len);
1087 	nvme_tcp_pdu_set_data(rsp_pdu, (uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, copy_len);
1088 
1089 	/* Contain the header of the wrong received pdu */
1090 	c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len;
1091 	spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR);
1092 	spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_send_c2h_term_req_complete, tqpair);
1093 }
1094 
1095 static void
1096 spdk_nvmf_tcp_capsule_cmd_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport,
1097 				     struct spdk_nvmf_tcp_qpair *tqpair,
1098 				     struct nvme_tcp_pdu *pdu)
1099 {
1100 	struct spdk_nvmf_tcp_req *tcp_req;
1101 
1102 	assert(pdu->psh_valid_bytes == pdu->psh_len);
1103 	assert(pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD);
1104 
1105 	tcp_req = spdk_nvmf_tcp_req_get(tqpair);
1106 	if (!tcp_req) {
1107 		/* Directly return and make the allocation retry again */
1108 		if (tqpair->state_cntr[TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST] > 0) {
1109 			return;
1110 		}
1111 
1112 		/* The host sent more commands than the maximum queue depth. */
1113 		SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair);
1114 		spdk_nvmf_tcp_qpair_disconnect(tqpair);
1115 		return;
1116 	}
1117 
1118 	pdu->req = tcp_req;
1119 	assert(tcp_req->state == TCP_REQUEST_STATE_NEW);
1120 	spdk_nvmf_tcp_req_process(ttransport, tcp_req);
1121 }
1122 
1123 static void
1124 spdk_nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport,
1125 		struct spdk_nvmf_tcp_qpair *tqpair,
1126 		struct nvme_tcp_pdu *pdu)
1127 {
1128 	struct spdk_nvmf_tcp_req *tcp_req;
1129 	struct spdk_nvme_tcp_cmd *capsule_cmd;
1130 	uint32_t error_offset = 0;
1131 	enum spdk_nvme_tcp_term_req_fes fes;
1132 
1133 	capsule_cmd = &pdu->hdr.capsule_cmd;
1134 	tcp_req = pdu->req;
1135 	assert(tcp_req != NULL);
1136 	if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) {
1137 		SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n",
1138 			    SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo);
1139 		fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
1140 		error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo);
1141 		goto err;
1142 	}
1143 
1144 	spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY);
1145 	spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE);
1146 	spdk_nvmf_tcp_req_process(ttransport, tcp_req);
1147 
1148 	return;
1149 err:
1150 	spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
1151 }
1152 
1153 static int
1154 nvmf_tcp_find_req_in_state(struct spdk_nvmf_tcp_qpair *tqpair,
1155 			   enum spdk_nvmf_tcp_req_state state,
1156 			   uint16_t cid, uint16_t tag,
1157 			   struct spdk_nvmf_tcp_req **req)
1158 {
1159 	struct spdk_nvmf_tcp_req *tcp_req = NULL;
1160 
1161 	TAILQ_FOREACH(tcp_req, &tqpair->state_queue[state], state_link) {
1162 		if (tcp_req->req.cmd->nvme_cmd.cid != cid) {
1163 			continue;
1164 		}
1165 
1166 		if (tcp_req->ttag == tag) {
1167 			*req = tcp_req;
1168 			return 0;
1169 		}
1170 
1171 		*req = NULL;
1172 		return -1;
1173 	}
1174 
1175 	/* Didn't find it, but not an error */
1176 	*req = NULL;
1177 	return 0;
1178 }
1179 
1180 static void
1181 spdk_nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport,
1182 				  struct spdk_nvmf_tcp_qpair *tqpair,
1183 				  struct nvme_tcp_pdu *pdu)
1184 {
1185 	struct spdk_nvmf_tcp_req *tcp_req;
1186 	uint32_t error_offset = 0;
1187 	enum spdk_nvme_tcp_term_req_fes fes = 0;
1188 	struct spdk_nvme_tcp_h2c_data_hdr *h2c_data;
1189 	int rc;
1190 
1191 	h2c_data = &pdu->hdr.h2c_data;
1192 
1193 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n",
1194 		      tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag);
1195 
1196 	rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER,
1197 					h2c_data->cccid, h2c_data->ttag, &tcp_req);
1198 	if (rc == 0 && tcp_req == NULL) {
1199 		rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK, h2c_data->cccid,
1200 						h2c_data->ttag, &tcp_req);
1201 	}
1202 
1203 	if (!tcp_req) {
1204 		SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tcp_req is not found for tqpair=%p\n", tqpair);
1205 		fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER;
1206 		if (rc == 0) {
1207 			error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid);
1208 		} else {
1209 			error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag);
1210 		}
1211 		goto err;
1212 	}
1213 
1214 	if (tcp_req->h2c_offset != h2c_data->datao) {
1215 		SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP,
1216 			      "tcp_req(%p), tqpair=%p, expected data offset %u, but data offset is %u\n",
1217 			      tcp_req, tqpair, tcp_req->h2c_offset, h2c_data->datao);
1218 		fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE;
1219 		goto err;
1220 	}
1221 
1222 	if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) {
1223 		SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP,
1224 			      "tcp_req(%p), tqpair=%p,  (datao=%u + datal=%u) execeeds requested length=%u\n",
1225 			      tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length);
1226 		fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE;
1227 		goto err;
1228 	}
1229 
1230 	pdu->req = tcp_req;
1231 
1232 	if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) {
1233 		pdu->dif_ctx = &tcp_req->req.dif.dif_ctx;
1234 	}
1235 
1236 	nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt,
1237 				  h2c_data->datao, h2c_data->datal);
1238 	spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD);
1239 	return;
1240 
1241 err:
1242 	spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
1243 }
1244 
1245 static void
1246 spdk_nvmf_tcp_pdu_cmd_complete(void *cb_arg)
1247 {
1248 	struct spdk_nvmf_tcp_req *tcp_req = cb_arg;
1249 	nvmf_tcp_request_free(tcp_req);
1250 }
1251 
1252 static void
1253 spdk_nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req,
1254 				    struct spdk_nvmf_tcp_qpair *tqpair)
1255 {
1256 	struct nvme_tcp_pdu *rsp_pdu;
1257 	struct spdk_nvme_tcp_rsp *capsule_resp;
1258 
1259 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter, tqpair=%p\n", tqpair);
1260 
1261 	rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req);
1262 	assert(rsp_pdu != NULL);
1263 
1264 	capsule_resp = &rsp_pdu->hdr.capsule_resp;
1265 	capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP;
1266 	capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp);
1267 	capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl;
1268 	if (tqpair->host_hdgst_enable) {
1269 		capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF;
1270 		capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN;
1271 	}
1272 
1273 	spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_pdu_cmd_complete, tcp_req);
1274 }
1275 
1276 static void
1277 spdk_nvmf_tcp_pdu_c2h_data_complete(void *cb_arg)
1278 {
1279 	struct spdk_nvmf_tcp_req *tcp_req = cb_arg;
1280 	struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair,
1281 					     struct spdk_nvmf_tcp_qpair, qpair);
1282 
1283 	assert(tqpair != NULL);
1284 	if (tqpair->qpair.transport->opts.c2h_success) {
1285 		nvmf_tcp_request_free(tcp_req);
1286 	} else {
1287 		nvmf_tcp_req_pdu_fini(tcp_req);
1288 		spdk_nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair);
1289 	}
1290 }
1291 
1292 static void
1293 spdk_nvmf_tcp_r2t_complete(void *cb_arg)
1294 {
1295 	struct spdk_nvmf_tcp_req *tcp_req = cb_arg;
1296 	struct spdk_nvmf_tcp_transport *ttransport;
1297 
1298 	nvmf_tcp_req_pdu_fini(tcp_req);
1299 
1300 	ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport,
1301 				      struct spdk_nvmf_tcp_transport, transport);
1302 
1303 	spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER);
1304 
1305 	if (tcp_req->h2c_offset == tcp_req->req.length) {
1306 		spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE);
1307 		spdk_nvmf_tcp_req_process(ttransport, tcp_req);
1308 	}
1309 }
1310 
1311 static void
1312 spdk_nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair,
1313 			   struct spdk_nvmf_tcp_req *tcp_req)
1314 {
1315 	struct nvme_tcp_pdu *rsp_pdu;
1316 	struct spdk_nvme_tcp_r2t_hdr *r2t;
1317 
1318 	rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req);
1319 	assert(rsp_pdu != NULL);
1320 
1321 	r2t = &rsp_pdu->hdr.r2t;
1322 	r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T;
1323 	r2t->common.plen = r2t->common.hlen = sizeof(*r2t);
1324 
1325 	if (tqpair->host_hdgst_enable) {
1326 		r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF;
1327 		r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN;
1328 	}
1329 
1330 	r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid;
1331 	r2t->ttag = tcp_req->ttag;
1332 	r2t->r2to = tcp_req->h2c_offset;
1333 	r2t->r2tl = tcp_req->req.length;
1334 
1335 	spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_R2T_ACK);
1336 
1337 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP,
1338 		      "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n",
1339 		      tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl);
1340 	spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_r2t_complete, tcp_req);
1341 }
1342 
1343 static void
1344 spdk_nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport,
1345 				      struct spdk_nvmf_tcp_qpair *tqpair,
1346 				      struct nvme_tcp_pdu *pdu)
1347 {
1348 	struct spdk_nvmf_tcp_req *tcp_req;
1349 
1350 	tcp_req = pdu->req;
1351 	assert(tcp_req != NULL);
1352 
1353 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n");
1354 
1355 	tcp_req->h2c_offset += pdu->data_len;
1356 
1357 	spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY);
1358 
1359 	/* Wait for all of the data to arrive AND for the initial R2T PDU send to be
1360 	 * acknowledged before moving on. */
1361 	if (tcp_req->h2c_offset == tcp_req->req.length &&
1362 	    tcp_req->state == TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER) {
1363 		spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE);
1364 		spdk_nvmf_tcp_req_process(ttransport, tcp_req);
1365 	}
1366 }
1367 
1368 static void
1369 spdk_nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req)
1370 {
1371 	SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req,
1372 		    spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]);
1373 	if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) ||
1374 	    (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) {
1375 		SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "The offset from the start of the PDU header is %u\n",
1376 			      DGET32(h2c_term_req->fei));
1377 	}
1378 }
1379 
1380 static void
1381 spdk_nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair,
1382 				      struct nvme_tcp_pdu *pdu)
1383 {
1384 	struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req;
1385 	uint32_t error_offset = 0;
1386 	enum spdk_nvme_tcp_term_req_fes fes;
1387 
1388 
1389 	if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) {
1390 		SPDK_ERRLOG("Fatal Error Stauts(FES) is unknown for h2c_term_req pdu=%p\n", pdu);
1391 		fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
1392 		error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes);
1393 		goto end;
1394 	}
1395 
1396 	/* set the data buffer */
1397 	nvme_tcp_pdu_set_data(pdu, (uint8_t *)pdu->hdr.raw + h2c_term_req->common.hlen,
1398 			      h2c_term_req->common.plen - h2c_term_req->common.hlen);
1399 	spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD);
1400 	return;
1401 end:
1402 	spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
1403 }
1404 
1405 static void
1406 spdk_nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair,
1407 		struct nvme_tcp_pdu *pdu)
1408 {
1409 	struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req;
1410 
1411 	spdk_nvmf_tcp_h2c_term_req_dump(h2c_term_req);
1412 	spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR);
1413 }
1414 
1415 static void
1416 spdk_nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair,
1417 				 struct spdk_nvmf_tcp_transport *ttransport)
1418 {
1419 	int rc = 0;
1420 	struct nvme_tcp_pdu *pdu;
1421 	uint32_t crc32c, error_offset = 0;
1422 	enum spdk_nvme_tcp_term_req_fes fes;
1423 
1424 	assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD);
1425 	pdu = &tqpair->pdu_in_progress;
1426 
1427 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n");
1428 	/* check data digest if need */
1429 	if (pdu->ddgst_enable) {
1430 		crc32c = nvme_tcp_pdu_calc_data_digest(pdu);
1431 		rc = MATCH_DIGEST_WORD(pdu->data_digest, crc32c);
1432 		if (rc == 0) {
1433 			SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu);
1434 			fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR;
1435 			spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
1436 			return;
1437 
1438 		}
1439 	}
1440 
1441 	switch (pdu->hdr.common.pdu_type) {
1442 	case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD:
1443 		spdk_nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu);
1444 		break;
1445 	case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA:
1446 		spdk_nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu);
1447 		break;
1448 
1449 	case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ:
1450 		spdk_nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu);
1451 		break;
1452 
1453 	default:
1454 		/* The code should not go to here */
1455 		SPDK_ERRLOG("The code should not go to here\n");
1456 		break;
1457 	}
1458 }
1459 
1460 static void
1461 spdk_nvmf_tcp_send_icresp_complete(void *cb_arg)
1462 {
1463 	struct spdk_nvmf_tcp_qpair *tqpair = cb_arg;
1464 
1465 	tqpair->state = NVME_TCP_QPAIR_STATE_RUNNING;
1466 }
1467 
1468 static void
1469 spdk_nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport,
1470 			   struct spdk_nvmf_tcp_qpair *tqpair,
1471 			   struct nvme_tcp_pdu *pdu)
1472 {
1473 	struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr.ic_req;
1474 	struct nvme_tcp_pdu *rsp_pdu;
1475 	struct spdk_nvme_tcp_ic_resp *ic_resp;
1476 	uint32_t error_offset = 0;
1477 	enum spdk_nvme_tcp_term_req_fes fes;
1478 
1479 	/* Only PFV 0 is defined currently */
1480 	if (ic_req->pfv != 0) {
1481 		SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv);
1482 		fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
1483 		error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv);
1484 		goto end;
1485 	}
1486 
1487 	/* MAXR2T is 0's based */
1488 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "maxr2t =%u\n", (ic_req->maxr2t + 1u));
1489 
1490 	tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false;
1491 	if (!tqpair->host_hdgst_enable) {
1492 		tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR;
1493 	}
1494 
1495 	tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false;
1496 	if (!tqpair->host_ddgst_enable) {
1497 		tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR;
1498 	}
1499 
1500 	/* Now that we know whether digests are enabled, properly size the receive buffer */
1501 	if (spdk_sock_set_recvbuf(tqpair->sock, tqpair->recv_buf_size) < 0) {
1502 		SPDK_WARNLOG("Unable to allocate enough memory for receive buffer on tqpair=%p with size=%d\n",
1503 			     tqpair,
1504 			     tqpair->recv_buf_size);
1505 		/* Not fatal. */
1506 	}
1507 
1508 	tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX);
1509 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda);
1510 
1511 	rsp_pdu = &tqpair->mgmt_pdu;
1512 
1513 	ic_resp = &rsp_pdu->hdr.ic_resp;
1514 	ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP;
1515 	ic_resp->common.hlen = ic_resp->common.plen =  sizeof(*ic_resp);
1516 	ic_resp->pfv = 0;
1517 	ic_resp->cpda = tqpair->cpda;
1518 	ic_resp->maxh2cdata = ttransport->transport.opts.max_io_size;
1519 	ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0;
1520 	ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0;
1521 
1522 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable);
1523 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable);
1524 
1525 	tqpair->state = NVME_TCP_QPAIR_STATE_INITIALIZING;
1526 	spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_send_icresp_complete, tqpair);
1527 	spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY);
1528 	return;
1529 end:
1530 	spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
1531 }
1532 
1533 static void
1534 spdk_nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair,
1535 			     struct spdk_nvmf_tcp_transport *ttransport)
1536 {
1537 	struct nvme_tcp_pdu *pdu;
1538 	int rc;
1539 	uint32_t crc32c, error_offset = 0;
1540 	enum spdk_nvme_tcp_term_req_fes fes;
1541 
1542 	assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH);
1543 	pdu = &tqpair->pdu_in_progress;
1544 
1545 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "pdu type of tqpair(%p) is %d\n", tqpair,
1546 		      pdu->hdr.common.pdu_type);
1547 	/* check header digest if needed */
1548 	if (pdu->has_hdgst) {
1549 		SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair);
1550 		crc32c = nvme_tcp_pdu_calc_header_digest(pdu);
1551 		rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, crc32c);
1552 		if (rc == 0) {
1553 			SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu);
1554 			fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR;
1555 			spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
1556 			return;
1557 
1558 		}
1559 	}
1560 
1561 	switch (pdu->hdr.common.pdu_type) {
1562 	case SPDK_NVME_TCP_PDU_TYPE_IC_REQ:
1563 		spdk_nvmf_tcp_icreq_handle(ttransport, tqpair, pdu);
1564 		break;
1565 	case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD:
1566 		spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_REQ);
1567 		break;
1568 	case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA:
1569 		spdk_nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu);
1570 		break;
1571 
1572 	case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ:
1573 		spdk_nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu);
1574 		break;
1575 
1576 	default:
1577 		SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress.hdr.common.pdu_type);
1578 		fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
1579 		error_offset = 1;
1580 		spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
1581 		break;
1582 	}
1583 }
1584 
1585 static void
1586 spdk_nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair)
1587 {
1588 	struct nvme_tcp_pdu *pdu;
1589 	uint32_t error_offset = 0;
1590 	enum spdk_nvme_tcp_term_req_fes fes;
1591 	uint8_t expected_hlen, pdo;
1592 	bool plen_error = false, pdo_error = false;
1593 
1594 	assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH);
1595 	pdu = &tqpair->pdu_in_progress;
1596 
1597 	if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) {
1598 		if (tqpair->state != NVME_TCP_QPAIR_STATE_INVALID) {
1599 			SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu);
1600 			fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR;
1601 			goto err;
1602 		}
1603 		expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req);
1604 		if (pdu->hdr.common.plen != expected_hlen) {
1605 			plen_error = true;
1606 		}
1607 	} else {
1608 		if (tqpair->state != NVME_TCP_QPAIR_STATE_RUNNING) {
1609 			SPDK_ERRLOG("The TCP/IP connection is not negotitated\n");
1610 			fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR;
1611 			goto err;
1612 		}
1613 
1614 		switch (pdu->hdr.common.pdu_type) {
1615 		case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD:
1616 			expected_hlen = sizeof(struct spdk_nvme_tcp_cmd);
1617 			pdo = pdu->hdr.common.pdo;
1618 			if ((tqpair->cpda != 0) && (pdo != ((tqpair->cpda + 1) << 2))) {
1619 				pdo_error = true;
1620 				break;
1621 			}
1622 
1623 			if (pdu->hdr.common.plen < expected_hlen) {
1624 				plen_error = true;
1625 			}
1626 			break;
1627 		case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA:
1628 			expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr);
1629 			pdo = pdu->hdr.common.pdo;
1630 			if ((tqpair->cpda != 0) && (pdo != ((tqpair->cpda + 1) << 2))) {
1631 				pdo_error = true;
1632 				break;
1633 			}
1634 			if (pdu->hdr.common.plen < expected_hlen) {
1635 				plen_error = true;
1636 			}
1637 			break;
1638 
1639 		case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ:
1640 			expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr);
1641 			if ((pdu->hdr.common.plen <= expected_hlen) ||
1642 			    (pdu->hdr.common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) {
1643 				plen_error = true;
1644 			}
1645 			break;
1646 
1647 		default:
1648 			SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr.common.pdu_type);
1649 			fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
1650 			error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type);
1651 			goto err;
1652 		}
1653 	}
1654 
1655 	if (pdu->hdr.common.hlen != expected_hlen) {
1656 		SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n",
1657 			    pdu->hdr.common.pdu_type,
1658 			    expected_hlen, pdu->hdr.common.hlen, tqpair);
1659 		fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
1660 		error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen);
1661 		goto err;
1662 	} else if (pdo_error) {
1663 		fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
1664 		error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo);
1665 	} else if (plen_error) {
1666 		fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
1667 		error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen);
1668 		goto err;
1669 	} else {
1670 		spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH);
1671 		nvme_tcp_pdu_calc_psh_len(&tqpair->pdu_in_progress, tqpair->host_hdgst_enable);
1672 		return;
1673 	}
1674 err:
1675 	spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
1676 }
1677 
1678 static int
1679 nvmf_tcp_pdu_payload_insert_dif(struct nvme_tcp_pdu *pdu, uint32_t read_offset,
1680 				int read_len)
1681 {
1682 	int rc;
1683 
1684 	rc = spdk_dif_generate_stream(pdu->data_iov, pdu->data_iovcnt,
1685 				      read_offset, read_len, pdu->dif_ctx);
1686 	if (rc != 0) {
1687 		SPDK_ERRLOG("DIF generate failed\n");
1688 	}
1689 
1690 	return rc;
1691 }
1692 
1693 static int
1694 spdk_nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair)
1695 {
1696 	int rc = 0;
1697 	struct nvme_tcp_pdu *pdu;
1698 	enum nvme_tcp_pdu_recv_state prev_state;
1699 	uint32_t data_len;
1700 	struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport,
1701 			struct spdk_nvmf_tcp_transport, transport);
1702 
1703 	/* The loop here is to allow for several back-to-back state changes. */
1704 	do {
1705 		prev_state = tqpair->recv_state;
1706 		SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state);
1707 
1708 		pdu = &tqpair->pdu_in_progress;
1709 		switch (tqpair->recv_state) {
1710 		/* Wait for the common header  */
1711 		case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY:
1712 		case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH:
1713 			if (spdk_unlikely(tqpair->state == NVME_TCP_QPAIR_STATE_INITIALIZING)) {
1714 				return rc;
1715 			}
1716 
1717 			rc = nvme_tcp_read_data(tqpair->sock,
1718 						sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes,
1719 						(void *)&pdu->hdr.common + pdu->ch_valid_bytes);
1720 			if (rc < 0) {
1721 				SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "will disconnect tqpair=%p\n", tqpair);
1722 				return NVME_TCP_PDU_FATAL;
1723 			} else if (rc > 0) {
1724 				pdu->ch_valid_bytes += rc;
1725 				spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, 0, rc, 0, 0);
1726 				if (spdk_likely(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY)) {
1727 					spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH);
1728 				}
1729 			}
1730 
1731 			if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) {
1732 				return NVME_TCP_PDU_IN_PROGRESS;
1733 			}
1734 
1735 			/* The command header of this PDU has now been read from the socket. */
1736 			spdk_nvmf_tcp_pdu_ch_handle(tqpair);
1737 			break;
1738 		/* Wait for the pdu specific header  */
1739 		case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH:
1740 			rc = nvme_tcp_read_data(tqpair->sock,
1741 						pdu->psh_len - pdu->psh_valid_bytes,
1742 						(void *)&pdu->hdr.raw + sizeof(struct spdk_nvme_tcp_common_pdu_hdr) + pdu->psh_valid_bytes);
1743 			if (rc < 0) {
1744 				return NVME_TCP_PDU_FATAL;
1745 			} else if (rc > 0) {
1746 				spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE,
1747 						  0, rc, 0, 0);
1748 				pdu->psh_valid_bytes += rc;
1749 			}
1750 
1751 			if (pdu->psh_valid_bytes < pdu->psh_len) {
1752 				return NVME_TCP_PDU_IN_PROGRESS;
1753 			}
1754 
1755 			/* All header(ch, psh, head digist) of this PDU has now been read from the socket. */
1756 			spdk_nvmf_tcp_pdu_psh_handle(tqpair, ttransport);
1757 			break;
1758 		/* Wait for the req slot */
1759 		case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ:
1760 			spdk_nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu);
1761 			break;
1762 		case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD:
1763 			/* check whether the data is valid, if not we just return */
1764 			if (!pdu->data_len) {
1765 				return NVME_TCP_PDU_IN_PROGRESS;
1766 			}
1767 
1768 			data_len = pdu->data_len;
1769 			/* data digest */
1770 			if (spdk_unlikely((pdu->hdr.common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) &&
1771 					  tqpair->host_ddgst_enable)) {
1772 				data_len += SPDK_NVME_TCP_DIGEST_LEN;
1773 				pdu->ddgst_enable = true;
1774 			}
1775 
1776 			rc = nvme_tcp_read_payload_data(tqpair->sock, pdu);
1777 			if (rc < 0) {
1778 				return NVME_TCP_PDU_IN_PROGRESS;
1779 			}
1780 			pdu->readv_offset += rc;
1781 
1782 			if (spdk_unlikely(pdu->dif_ctx != NULL)) {
1783 				rc = nvmf_tcp_pdu_payload_insert_dif(pdu, pdu->readv_offset - rc, rc);
1784 				if (rc != 0) {
1785 					return NVME_TCP_PDU_FATAL;
1786 				}
1787 			}
1788 
1789 			if (pdu->readv_offset < data_len) {
1790 				return NVME_TCP_PDU_IN_PROGRESS;
1791 			}
1792 
1793 			/* All of this PDU has now been read from the socket. */
1794 			spdk_nvmf_tcp_pdu_payload_handle(tqpair, ttransport);
1795 			break;
1796 		case NVME_TCP_PDU_RECV_STATE_ERROR:
1797 			if (!spdk_sock_is_connected(tqpair->sock)) {
1798 				return NVME_TCP_PDU_FATAL;
1799 			}
1800 			break;
1801 		default:
1802 			assert(0);
1803 			SPDK_ERRLOG("code should not come to here");
1804 			break;
1805 		}
1806 	} while (tqpair->recv_state != prev_state);
1807 
1808 	return rc;
1809 }
1810 
1811 static int
1812 spdk_nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_req *tcp_req,
1813 			    struct spdk_nvmf_transport *transport,
1814 			    struct spdk_nvmf_transport_poll_group *group)
1815 {
1816 	struct spdk_nvmf_request		*req = &tcp_req->req;
1817 	struct spdk_nvme_cmd			*cmd;
1818 	struct spdk_nvme_cpl			*rsp;
1819 	struct spdk_nvme_sgl_descriptor		*sgl;
1820 	uint32_t				length;
1821 
1822 	cmd = &req->cmd->nvme_cmd;
1823 	rsp = &req->rsp->nvme_cpl;
1824 	sgl = &cmd->dptr.sgl1;
1825 
1826 	length = sgl->unkeyed.length;
1827 
1828 	if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK &&
1829 	    sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) {
1830 		if (length > transport->opts.max_io_size) {
1831 			SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n",
1832 				    length, transport->opts.max_io_size);
1833 			rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID;
1834 			return -1;
1835 		}
1836 
1837 		/* fill request length and populate iovs */
1838 		req->length = length;
1839 
1840 		SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Data requested length= 0x%x\n", length);
1841 
1842 		if (spdk_unlikely(req->dif.dif_insert_or_strip)) {
1843 			req->dif.orig_length = length;
1844 			length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx);
1845 			req->dif.elba_length = length;
1846 		}
1847 
1848 		if (spdk_nvmf_request_get_buffers(req, group, transport, length)) {
1849 			/* No available buffers. Queue this request up. */
1850 			SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "No available large data buffers. Queueing request %p\n",
1851 				      tcp_req);
1852 			return 0;
1853 		}
1854 
1855 		/* backward compatible */
1856 		req->data = req->iov[0].iov_base;
1857 
1858 		SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Request %p took %d buffer/s from central pool, and data=%p\n",
1859 			      tcp_req, req->iovcnt, req->data);
1860 
1861 		return 0;
1862 	} else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK &&
1863 		   sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) {
1864 		uint64_t offset = sgl->address;
1865 		uint32_t max_len = transport->opts.in_capsule_data_size;
1866 
1867 		SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n",
1868 			      offset, length);
1869 
1870 		if (offset > max_len) {
1871 			SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " exceeds capsule length 0x%x\n",
1872 				    offset, max_len);
1873 			rsp->status.sc = SPDK_NVME_SC_INVALID_SGL_OFFSET;
1874 			return -1;
1875 		}
1876 		max_len -= (uint32_t)offset;
1877 
1878 		if (length > max_len) {
1879 			SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n",
1880 				    length, max_len);
1881 			rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID;
1882 			return -1;
1883 		}
1884 
1885 		req->data = tcp_req->buf + offset;
1886 		req->data_from_pool = false;
1887 		req->length = length;
1888 
1889 		if (spdk_unlikely(req->dif.dif_insert_or_strip)) {
1890 			length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx);
1891 			req->dif.elba_length = length;
1892 		}
1893 
1894 		req->iov[0].iov_base = req->data;
1895 		req->iov[0].iov_len = length;
1896 		req->iovcnt = 1;
1897 
1898 		return 0;
1899 	}
1900 
1901 	SPDK_ERRLOG("Invalid NVMf I/O Command SGL:  Type 0x%x, Subtype 0x%x\n",
1902 		    sgl->generic.type, sgl->generic.subtype);
1903 	rsp->status.sc = SPDK_NVME_SC_SGL_DESCRIPTOR_TYPE_INVALID;
1904 	return -1;
1905 }
1906 
1907 static inline enum spdk_nvme_media_error_status_code
1908 nvmf_tcp_dif_error_to_compl_status(uint8_t err_type) {
1909 	enum spdk_nvme_media_error_status_code result;
1910 
1911 	switch (err_type)
1912 	{
1913 	case SPDK_DIF_REFTAG_ERROR:
1914 		result = SPDK_NVME_SC_REFERENCE_TAG_CHECK_ERROR;
1915 		break;
1916 	case SPDK_DIF_APPTAG_ERROR:
1917 		result = SPDK_NVME_SC_APPLICATION_TAG_CHECK_ERROR;
1918 		break;
1919 	case SPDK_DIF_GUARD_ERROR:
1920 		result = SPDK_NVME_SC_GUARD_CHECK_ERROR;
1921 		break;
1922 	default:
1923 		SPDK_UNREACHABLE();
1924 		break;
1925 	}
1926 
1927 	return result;
1928 }
1929 
1930 static void
1931 spdk_nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair,
1932 			    struct spdk_nvmf_tcp_req *tcp_req)
1933 {
1934 	struct nvme_tcp_pdu *rsp_pdu;
1935 	struct spdk_nvme_tcp_c2h_data_hdr *c2h_data;
1936 	uint32_t plen, pdo, alignment;
1937 	int rc;
1938 
1939 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n");
1940 
1941 	rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req);
1942 	assert(rsp_pdu != NULL);
1943 
1944 	c2h_data = &rsp_pdu->hdr.c2h_data;
1945 	c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA;
1946 	plen = c2h_data->common.hlen = sizeof(*c2h_data);
1947 
1948 	if (tqpair->host_hdgst_enable) {
1949 		plen += SPDK_NVME_TCP_DIGEST_LEN;
1950 		c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF;
1951 	}
1952 
1953 	/* set the psh */
1954 	c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid;
1955 	c2h_data->datal = tcp_req->req.length;
1956 	c2h_data->datao = 0;
1957 
1958 	/* set the padding */
1959 	rsp_pdu->padding_len = 0;
1960 	pdo = plen;
1961 	if (tqpair->cpda) {
1962 		alignment = (tqpair->cpda + 1) << 2;
1963 		if (alignment > plen) {
1964 			rsp_pdu->padding_len = alignment - plen;
1965 			pdo = plen = alignment;
1966 		}
1967 	}
1968 
1969 	c2h_data->common.pdo = pdo;
1970 	plen += c2h_data->datal;
1971 	if (tqpair->host_ddgst_enable) {
1972 		c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF;
1973 		plen += SPDK_NVME_TCP_DIGEST_LEN;
1974 	}
1975 
1976 	c2h_data->common.plen = plen;
1977 
1978 	if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) {
1979 		rsp_pdu->dif_ctx = &tcp_req->req.dif.dif_ctx;
1980 	}
1981 
1982 	nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt,
1983 				  c2h_data->datao, c2h_data->datal);
1984 
1985 	if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) {
1986 		struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl;
1987 		struct spdk_dif_error err_blk = {};
1988 
1989 		rc = spdk_dif_verify_stream(rsp_pdu->data_iov, rsp_pdu->data_iovcnt,
1990 					    0, rsp_pdu->data_len, rsp_pdu->dif_ctx, &err_blk);
1991 		if (rc != 0) {
1992 			SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n",
1993 				    err_blk.err_type, err_blk.err_offset);
1994 			rsp->status.sct = SPDK_NVME_SCT_MEDIA_ERROR;
1995 			rsp->status.sc = nvmf_tcp_dif_error_to_compl_status(err_blk.err_type);
1996 			nvmf_tcp_req_pdu_fini(tcp_req);
1997 			spdk_nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair);
1998 			return;
1999 		}
2000 	}
2001 
2002 	c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU;
2003 	if (tqpair->qpair.transport->opts.c2h_success) {
2004 		c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS;
2005 	}
2006 
2007 	spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_pdu_c2h_data_complete, tcp_req);
2008 }
2009 
2010 static int
2011 request_transfer_out(struct spdk_nvmf_request *req)
2012 {
2013 	struct spdk_nvmf_tcp_req	*tcp_req;
2014 	struct spdk_nvmf_qpair		*qpair;
2015 	struct spdk_nvmf_tcp_qpair	*tqpair;
2016 	struct spdk_nvme_cpl		*rsp;
2017 
2018 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n");
2019 
2020 	qpair = req->qpair;
2021 	rsp = &req->rsp->nvme_cpl;
2022 	tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req);
2023 
2024 	/* Advance our sq_head pointer */
2025 	if (qpair->sq_head == qpair->sq_head_max) {
2026 		qpair->sq_head = 0;
2027 	} else {
2028 		qpair->sq_head++;
2029 	}
2030 	rsp->sqhd = qpair->sq_head;
2031 
2032 	tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair);
2033 	spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST);
2034 	if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) {
2035 		spdk_nvmf_tcp_send_c2h_data(tqpair, tcp_req);
2036 	} else {
2037 		spdk_nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair);
2038 	}
2039 
2040 	return 0;
2041 }
2042 
2043 static void
2044 spdk_nvmf_tcp_set_incapsule_data(struct spdk_nvmf_tcp_qpair *tqpair,
2045 				 struct spdk_nvmf_tcp_req *tcp_req)
2046 {
2047 	struct nvme_tcp_pdu *pdu;
2048 	uint32_t plen = 0;
2049 
2050 	pdu = &tqpair->pdu_in_progress;
2051 	plen = pdu->hdr.common.hlen;
2052 
2053 	if (tqpair->host_hdgst_enable) {
2054 		plen += SPDK_NVME_TCP_DIGEST_LEN;
2055 	}
2056 
2057 	if (pdu->hdr.common.plen != plen) {
2058 		tcp_req->has_incapsule_data = true;
2059 	}
2060 }
2061 
2062 static bool
2063 spdk_nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport,
2064 			  struct spdk_nvmf_tcp_req *tcp_req)
2065 {
2066 	struct spdk_nvmf_tcp_qpair		*tqpair;
2067 	int					rc;
2068 	enum spdk_nvmf_tcp_req_state		prev_state;
2069 	bool					progress = false;
2070 	struct spdk_nvmf_transport		*transport = &ttransport->transport;
2071 	struct spdk_nvmf_transport_poll_group	*group;
2072 
2073 	tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair);
2074 	group = &tqpair->group->group;
2075 	assert(tcp_req->state != TCP_REQUEST_STATE_FREE);
2076 
2077 	/* The loop here is to allow for several back-to-back state changes. */
2078 	do {
2079 		prev_state = tcp_req->state;
2080 
2081 		SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state,
2082 			      tqpair);
2083 
2084 		switch (tcp_req->state) {
2085 		case TCP_REQUEST_STATE_FREE:
2086 			/* Some external code must kick a request into TCP_REQUEST_STATE_NEW
2087 			 * to escape this state. */
2088 			break;
2089 		case TCP_REQUEST_STATE_NEW:
2090 			spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, 0, 0, (uintptr_t)tcp_req, 0);
2091 
2092 			/* copy the cmd from the receive pdu */
2093 			tcp_req->cmd = tqpair->pdu_in_progress.hdr.capsule_cmd.ccsqe;
2094 
2095 			if (spdk_unlikely(spdk_nvmf_request_get_dif_ctx(&tcp_req->req, &tcp_req->req.dif.dif_ctx))) {
2096 				tcp_req->req.dif.dif_insert_or_strip = true;
2097 				tqpair->pdu_in_progress.dif_ctx = &tcp_req->req.dif.dif_ctx;
2098 			}
2099 
2100 			/* The next state transition depends on the data transfer needs of this request. */
2101 			tcp_req->req.xfer = spdk_nvmf_req_get_xfer(&tcp_req->req);
2102 
2103 			/* If no data to transfer, ready to execute. */
2104 			if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) {
2105 				/* Reset the tqpair receving pdu state */
2106 				spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY);
2107 				spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE);
2108 				break;
2109 			}
2110 
2111 			spdk_nvmf_tcp_set_incapsule_data(tqpair, tcp_req);
2112 
2113 			if (!tcp_req->has_incapsule_data) {
2114 				spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY);
2115 			}
2116 
2117 			spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER);
2118 			STAILQ_INSERT_TAIL(&group->pending_buf_queue, &tcp_req->req, buf_link);
2119 			break;
2120 		case TCP_REQUEST_STATE_NEED_BUFFER:
2121 			spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 0, 0, (uintptr_t)tcp_req, 0);
2122 
2123 			assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE);
2124 
2125 			if (!tcp_req->has_incapsule_data && (&tcp_req->req != STAILQ_FIRST(&group->pending_buf_queue))) {
2126 				SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP,
2127 					      "Not the first element to wait for the buf for tcp_req(%p) on tqpair=%p\n",
2128 					      tcp_req, tqpair);
2129 				/* This request needs to wait in line to obtain a buffer */
2130 				break;
2131 			}
2132 
2133 			/* Try to get a data buffer */
2134 			rc = spdk_nvmf_tcp_req_parse_sgl(tcp_req, transport, group);
2135 			if (rc < 0) {
2136 				STAILQ_REMOVE_HEAD(&group->pending_buf_queue, buf_link);
2137 				/* Reset the tqpair receving pdu state */
2138 				spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR);
2139 				spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE);
2140 				break;
2141 			}
2142 
2143 			if (!tcp_req->req.data) {
2144 				SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "No buffer allocated for tcp_req(%p) on tqpair(%p\n)",
2145 					      tcp_req, tqpair);
2146 				/* No buffers available. */
2147 				break;
2148 			}
2149 
2150 			STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link);
2151 
2152 			/* If data is transferring from host to controller, we need to do a transfer from the host. */
2153 			if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) {
2154 				if (tcp_req->req.data_from_pool) {
2155 					SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair);
2156 					spdk_nvmf_tcp_send_r2t_pdu(tqpair, tcp_req);
2157 				} else {
2158 					struct nvme_tcp_pdu *pdu;
2159 
2160 					spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER);
2161 
2162 					pdu = &tqpair->pdu_in_progress;
2163 					SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req,
2164 						      tqpair);
2165 					/* No need to send r2t, contained in the capsuled data */
2166 					nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt,
2167 								  0, tcp_req->req.length);
2168 					spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD);
2169 				}
2170 				break;
2171 			}
2172 
2173 			spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE);
2174 			break;
2175 		case TCP_REQUEST_STATE_AWAITING_R2T_ACK:
2176 			spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 0, 0, (uintptr_t)tcp_req, 0);
2177 			/* The R2T completion or the h2c data incoming will kick it out of this state. */
2178 			break;
2179 		case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER:
2180 
2181 			spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 0, 0,
2182 					  (uintptr_t)tcp_req, 0);
2183 			/* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE
2184 			 * to escape this state. */
2185 			break;
2186 		case TCP_REQUEST_STATE_READY_TO_EXECUTE:
2187 			spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 0, 0, (uintptr_t)tcp_req, 0);
2188 
2189 			if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) {
2190 				assert(tcp_req->req.dif.elba_length >= tcp_req->req.length);
2191 				tcp_req->req.length = tcp_req->req.dif.elba_length;
2192 			}
2193 
2194 			spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING);
2195 			spdk_nvmf_request_exec(&tcp_req->req);
2196 			break;
2197 		case TCP_REQUEST_STATE_EXECUTING:
2198 			spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, 0, 0, (uintptr_t)tcp_req, 0);
2199 			/* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED
2200 			 * to escape this state. */
2201 			break;
2202 		case TCP_REQUEST_STATE_EXECUTED:
2203 			spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, 0, 0, (uintptr_t)tcp_req, 0);
2204 
2205 			if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) {
2206 				tcp_req->req.length = tcp_req->req.dif.orig_length;
2207 			}
2208 
2209 			spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE);
2210 			break;
2211 		case TCP_REQUEST_STATE_READY_TO_COMPLETE:
2212 			spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 0, 0, (uintptr_t)tcp_req, 0);
2213 			rc = request_transfer_out(&tcp_req->req);
2214 			assert(rc == 0); /* No good way to handle this currently */
2215 			break;
2216 		case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST:
2217 			spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 0, 0,
2218 					  (uintptr_t)tcp_req,
2219 					  0);
2220 			/* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED
2221 			 * to escape this state. */
2222 			break;
2223 		case TCP_REQUEST_STATE_COMPLETED:
2224 			spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, 0, 0, (uintptr_t)tcp_req, 0);
2225 			if (tcp_req->req.data_from_pool) {
2226 				spdk_nvmf_request_free_buffers(&tcp_req->req, group, transport);
2227 			}
2228 			tcp_req->req.length = 0;
2229 			tcp_req->req.iovcnt = 0;
2230 			tcp_req->req.data = NULL;
2231 
2232 			nvmf_tcp_req_pdu_fini(tcp_req);
2233 
2234 			spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_FREE);
2235 			break;
2236 		case TCP_REQUEST_NUM_STATES:
2237 		default:
2238 			assert(0);
2239 			break;
2240 		}
2241 
2242 		if (tcp_req->state != prev_state) {
2243 			progress = true;
2244 		}
2245 	} while (tcp_req->state != prev_state);
2246 
2247 	return progress;
2248 }
2249 
2250 static void
2251 spdk_nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock)
2252 {
2253 	struct spdk_nvmf_tcp_qpair *tqpair = arg;
2254 	int rc;
2255 
2256 	assert(tqpair != NULL);
2257 	rc = spdk_nvmf_tcp_sock_process(tqpair);
2258 
2259 	/* If there was a new socket error, disconnect */
2260 	if (rc < 0) {
2261 		spdk_nvmf_tcp_qpair_disconnect(tqpair);
2262 	}
2263 }
2264 
2265 static int
2266 spdk_nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group,
2267 			     struct spdk_nvmf_qpair *qpair)
2268 {
2269 	struct spdk_nvmf_tcp_poll_group	*tgroup;
2270 	struct spdk_nvmf_tcp_qpair	*tqpair;
2271 	int				rc;
2272 
2273 	tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group);
2274 	tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
2275 
2276 	rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock,
2277 				      spdk_nvmf_tcp_sock_cb, tqpair);
2278 	if (rc != 0) {
2279 		SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n",
2280 			    spdk_strerror(errno), errno);
2281 		return -1;
2282 	}
2283 
2284 	rc =  spdk_nvmf_tcp_qpair_sock_init(tqpair);
2285 	if (rc != 0) {
2286 		SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair);
2287 		return -1;
2288 	}
2289 
2290 	rc = spdk_nvmf_tcp_qpair_init(&tqpair->qpair);
2291 	if (rc < 0) {
2292 		SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair);
2293 		return -1;
2294 	}
2295 
2296 	rc = spdk_nvmf_tcp_qpair_init_mem_resource(tqpair);
2297 	if (rc < 0) {
2298 		SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair);
2299 		return -1;
2300 	}
2301 
2302 	tqpair->group = tgroup;
2303 	tqpair->state = NVME_TCP_QPAIR_STATE_INVALID;
2304 	TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link);
2305 
2306 	return 0;
2307 }
2308 
2309 static int
2310 spdk_nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group,
2311 				struct spdk_nvmf_qpair *qpair)
2312 {
2313 	struct spdk_nvmf_tcp_poll_group	*tgroup;
2314 	struct spdk_nvmf_tcp_qpair		*tqpair;
2315 	int				rc;
2316 
2317 	tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group);
2318 	tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
2319 
2320 	assert(tqpair->group == tgroup);
2321 
2322 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup);
2323 	if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) {
2324 		TAILQ_REMOVE(&tgroup->await_req, tqpair, link);
2325 	} else {
2326 		TAILQ_REMOVE(&tgroup->qpairs, tqpair, link);
2327 	}
2328 
2329 	rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock);
2330 	if (rc != 0) {
2331 		SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n",
2332 			    spdk_strerror(errno), errno);
2333 	}
2334 
2335 	return rc;
2336 }
2337 
2338 static int
2339 spdk_nvmf_tcp_req_complete(struct spdk_nvmf_request *req)
2340 {
2341 	struct spdk_nvmf_tcp_transport *ttransport;
2342 	struct spdk_nvmf_tcp_req *tcp_req;
2343 
2344 	ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport);
2345 	tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req);
2346 
2347 	spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED);
2348 	spdk_nvmf_tcp_req_process(ttransport, tcp_req);
2349 
2350 	return 0;
2351 }
2352 
2353 static void
2354 spdk_nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair)
2355 {
2356 	struct spdk_nvmf_tcp_qpair *tqpair;
2357 
2358 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Qpair: %p\n", qpair);
2359 
2360 	tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
2361 	tqpair->state = NVME_TCP_QPAIR_STATE_EXITED;
2362 	spdk_nvmf_tcp_qpair_destroy(tqpair);
2363 }
2364 
2365 static int
2366 spdk_nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group)
2367 {
2368 	struct spdk_nvmf_tcp_poll_group *tgroup;
2369 	int rc;
2370 	struct spdk_nvmf_request *req, *req_tmp;
2371 	struct spdk_nvmf_tcp_req *tcp_req;
2372 	struct spdk_nvmf_tcp_qpair *tqpair, *tqpair_tmp;
2373 	struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(group->transport,
2374 			struct spdk_nvmf_tcp_transport, transport);
2375 
2376 	tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group);
2377 
2378 	if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs) && TAILQ_EMPTY(&tgroup->await_req))) {
2379 		return 0;
2380 	}
2381 
2382 	STAILQ_FOREACH_SAFE(req, &group->pending_buf_queue, buf_link, req_tmp) {
2383 		tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req);
2384 		if (spdk_nvmf_tcp_req_process(ttransport, tcp_req) == false) {
2385 			break;
2386 		}
2387 	}
2388 
2389 	rc = spdk_sock_group_poll(tgroup->sock_group);
2390 	if (rc < 0) {
2391 		SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group);
2392 	}
2393 
2394 	TAILQ_FOREACH_SAFE(tqpair, &tgroup->await_req, link, tqpair_tmp) {
2395 		spdk_nvmf_tcp_sock_process(tqpair);
2396 	}
2397 
2398 	return rc;
2399 }
2400 
2401 static int
2402 spdk_nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair,
2403 			     struct spdk_nvme_transport_id *trid, bool peer)
2404 {
2405 	struct spdk_nvmf_tcp_qpair     *tqpair;
2406 	uint16_t			port;
2407 
2408 	tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
2409 	spdk_nvme_trid_populate_transport(trid, SPDK_NVME_TRANSPORT_TCP);
2410 
2411 	if (peer) {
2412 		snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr);
2413 		port = tqpair->initiator_port;
2414 	} else {
2415 		snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr);
2416 		port = tqpair->target_port;
2417 	}
2418 
2419 	if (spdk_sock_is_ipv4(tqpair->sock)) {
2420 		trid->adrfam = SPDK_NVMF_ADRFAM_IPV4;
2421 	} else if (spdk_sock_is_ipv6(tqpair->sock)) {
2422 		trid->adrfam = SPDK_NVMF_ADRFAM_IPV6;
2423 	} else {
2424 		return -1;
2425 	}
2426 
2427 	snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port);
2428 	return 0;
2429 }
2430 
2431 static int
2432 spdk_nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair,
2433 				   struct spdk_nvme_transport_id *trid)
2434 {
2435 	return spdk_nvmf_tcp_qpair_get_trid(qpair, trid, 0);
2436 }
2437 
2438 static int
2439 spdk_nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair,
2440 				  struct spdk_nvme_transport_id *trid)
2441 {
2442 	return spdk_nvmf_tcp_qpair_get_trid(qpair, trid, 1);
2443 }
2444 
2445 static int
2446 spdk_nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair,
2447 				    struct spdk_nvme_transport_id *trid)
2448 {
2449 	return spdk_nvmf_tcp_qpair_get_trid(qpair, trid, 0);
2450 }
2451 
2452 #define SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH 128
2453 #define SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH 128
2454 #define SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR 128
2455 #define SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE 4096
2456 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE 131072
2457 #define SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE 131072
2458 #define SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS 511
2459 #define SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE 32
2460 #define SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION true
2461 #define SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP false
2462 #define SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY 0
2463 
2464 static void
2465 spdk_nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts)
2466 {
2467 	opts->max_queue_depth =		SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH;
2468 	opts->max_qpairs_per_ctrlr =	SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR;
2469 	opts->in_capsule_data_size =	SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE;
2470 	opts->max_io_size =		SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE;
2471 	opts->io_unit_size =		SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE;
2472 	opts->max_aq_depth =		SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH;
2473 	opts->num_shared_buffers =	SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS;
2474 	opts->buf_cache_size =		SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE;
2475 	opts->c2h_success =		SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION;
2476 	opts->dif_insert_or_strip =	SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP;
2477 	opts->sock_priority =		SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY;
2478 }
2479 
2480 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = {
2481 	.name = "TCP",
2482 	.type = SPDK_NVME_TRANSPORT_TCP,
2483 	.opts_init = spdk_nvmf_tcp_opts_init,
2484 	.create = spdk_nvmf_tcp_create,
2485 	.destroy = spdk_nvmf_tcp_destroy,
2486 
2487 	.listen = spdk_nvmf_tcp_listen,
2488 	.stop_listen = spdk_nvmf_tcp_stop_listen,
2489 	.accept = spdk_nvmf_tcp_accept,
2490 
2491 	.listener_discover = spdk_nvmf_tcp_discover,
2492 
2493 	.poll_group_create = spdk_nvmf_tcp_poll_group_create,
2494 	.get_optimal_poll_group = spdk_nvmf_tcp_get_optimal_poll_group,
2495 	.poll_group_destroy = spdk_nvmf_tcp_poll_group_destroy,
2496 	.poll_group_add = spdk_nvmf_tcp_poll_group_add,
2497 	.poll_group_remove = spdk_nvmf_tcp_poll_group_remove,
2498 	.poll_group_poll = spdk_nvmf_tcp_poll_group_poll,
2499 
2500 	.req_free = spdk_nvmf_tcp_req_free,
2501 	.req_complete = spdk_nvmf_tcp_req_complete,
2502 
2503 	.qpair_fini = spdk_nvmf_tcp_close_qpair,
2504 	.qpair_get_local_trid = spdk_nvmf_tcp_qpair_get_local_trid,
2505 	.qpair_get_peer_trid = spdk_nvmf_tcp_qpair_get_peer_trid,
2506 	.qpair_get_listen_trid = spdk_nvmf_tcp_qpair_get_listen_trid,
2507 };
2508 
2509 SPDK_NVMF_TRANSPORT_REGISTER(tcp, &spdk_nvmf_transport_tcp);
2510 SPDK_LOG_REGISTER_COMPONENT("nvmf_tcp", SPDK_LOG_NVMF_TCP)
2511