xref: /spdk/lib/nvmf/tcp.c (revision 56e12b00711b40d1e2ff45d4147193f45fdd9af0)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright (c) Intel Corporation.
5  *   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.h"
40 #include "spdk/nvmf_spec.h"
41 #include "spdk/sock.h"
42 #include "spdk/string.h"
43 #include "spdk/trace.h"
44 #include "spdk/util.h"
45 
46 #include "nvmf_internal.h"
47 #include "transport.h"
48 
49 #include "spdk_internal/log.h"
50 #include "spdk_internal/nvme_tcp.h"
51 
52 #define NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME 16
53 
54 #define NVMF_TCP_PDU_MAX_H2C_DATA_SIZE	131072
55 #define NVMF_TCP_PDU_MAX_C2H_DATA_SIZE	131072
56 #define NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM  64  /* Maximal c2h_data pdu number for ecah tqpair */
57 
58 /* This is used to support the Linux kernel NVMe-oF initiator */
59 #define LINUX_KERNEL_SUPPORT_NOT_SENDING_RESP_FOR_C2H 0
60 
61 /* spdk nvmf related structure */
62 enum spdk_nvmf_tcp_req_state {
63 
64 	/* The request is not currently in use */
65 	TCP_REQUEST_STATE_FREE = 0,
66 
67 	/* Initial state when request first received */
68 	TCP_REQUEST_STATE_NEW,
69 
70 	/* The request is queued until a data buffer is available. */
71 	TCP_REQUEST_STATE_NEED_BUFFER,
72 
73 	/* The request is pending on r2t slots */
74 	TCP_REQUEST_STATE_DATA_PENDING_FOR_R2T,
75 
76 	/* The request is currently transferring data from the host to the controller. */
77 	TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER,
78 
79 	/* The request is ready to execute at the block device */
80 	TCP_REQUEST_STATE_READY_TO_EXECUTE,
81 
82 	/* The request is currently executing at the block device */
83 	TCP_REQUEST_STATE_EXECUTING,
84 
85 	/* The request finished executing at the block device */
86 	TCP_REQUEST_STATE_EXECUTED,
87 
88 	/* The request is ready to send a completion */
89 	TCP_REQUEST_STATE_READY_TO_COMPLETE,
90 
91 	/* The request is currently transferring final pdus from the controller to the host. */
92 	TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST,
93 
94 	/* The request completed and can be marked free. */
95 	TCP_REQUEST_STATE_COMPLETED,
96 
97 	/* Terminator */
98 	TCP_REQUEST_NUM_STATES,
99 };
100 
101 static const char *spdk_nvmf_tcp_term_req_fes_str[] = {
102 	"Invalid PDU Header Field",
103 	"PDU Sequence Error",
104 	"Header Digiest Error",
105 	"Data Transfer Out of Range",
106 	"R2T Limit Exceeded",
107 	"Unsupported parameter",
108 };
109 
110 #define OBJECT_NVMF_TCP_IO				0x80
111 
112 #define TRACE_GROUP_NVMF_TCP				0x5
113 #define TRACE_TCP_REQUEST_STATE_NEW					SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x0)
114 #define TRACE_TCP_REQUEST_STATE_NEED_BUFFER				SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x1)
115 #define TRACE_TCP_REQUEST_STATE_DATA_PENDING_FOR_R2T			SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x2)
116 #define TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER		SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x3)
117 #define TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE			SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x4)
118 #define TRACE_TCP_REQUEST_STATE_EXECUTING				SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x5)
119 #define TRACE_TCP_REQUEST_STATE_EXECUTED				SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x6)
120 #define TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE			SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x7)
121 #define TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST		SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x8)
122 #define TRACE_TCP_REQUEST_STATE_COMPLETED				SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x9)
123 #define TRACE_TCP_FLUSH_WRITEBUF_START					SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xA)
124 #define TRACE_TCP_FLUSH_WRITEBUF_DONE					SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xB)
125 #define TRACE_TCP_FLUSH_WRITEBUF_PDU_DONE				SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xC)
126 
127 SPDK_TRACE_REGISTER_FN(nvmf_tcp_trace, "nvmf_tcp", TRACE_GROUP_NVMF_TCP)
128 {
129 	spdk_trace_register_object(OBJECT_NVMF_TCP_IO, 'r');
130 	spdk_trace_register_description("TCP_REQ_NEW", "",
131 					TRACE_TCP_REQUEST_STATE_NEW,
132 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 1, 1, "");
133 	spdk_trace_register_description("TCP_REQ_NEED_BUFFER", "",
134 					TRACE_TCP_REQUEST_STATE_NEED_BUFFER,
135 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, "");
136 	spdk_trace_register_description("TCP_REQ_TX_PENDING_R2T", "",
137 					TRACE_TCP_REQUEST_STATE_DATA_PENDING_FOR_R2T,
138 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, "");
139 	spdk_trace_register_description("TCP_REQ_TX_H_TO_C", "",
140 					TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER,
141 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, "");
142 	spdk_trace_register_description("TCP_REQ_RDY_TO_EXECUTE", "",
143 					TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE,
144 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, "");
145 	spdk_trace_register_description("TCP_REQ_EXECUTING", "",
146 					TRACE_TCP_REQUEST_STATE_EXECUTING,
147 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, "");
148 	spdk_trace_register_description("TCP_REQ_EXECUTED", "",
149 					TRACE_TCP_REQUEST_STATE_EXECUTED,
150 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, "");
151 	spdk_trace_register_description("TCP_REQ_RDY_TO_COMPLETE", "",
152 					TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE,
153 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, "");
154 	spdk_trace_register_description("TCP_REQ_COMPLETING_INCAPSULE", "",
155 					TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST,
156 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, "");
157 	spdk_trace_register_description("TCP_REQ_COMPLETED", "",
158 					TRACE_TCP_REQUEST_STATE_COMPLETED,
159 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, "");
160 	spdk_trace_register_description("TCP_FLUSH_WRITEBUF_START", "",
161 					TRACE_TCP_FLUSH_WRITEBUF_START,
162 					OWNER_NONE, OBJECT_NONE, 0, 0, "");
163 	spdk_trace_register_description("TCP_FLUSH_WRITEBUF_DONE", "",
164 					TRACE_TCP_FLUSH_WRITEBUF_DONE,
165 					OWNER_NONE, OBJECT_NONE, 0, 0, "");
166 	spdk_trace_register_description("TCP_FLUSH_WRITEBUF_PDU_DONE", "",
167 					TRACE_TCP_FLUSH_WRITEBUF_PDU_DONE,
168 					OWNER_NONE, OBJECT_NONE, 0, 0, "");
169 }
170 
171 struct spdk_nvmf_tcp_req  {
172 	struct spdk_nvmf_request		req;
173 	struct spdk_nvme_cpl			rsp;
174 	struct spdk_nvme_cmd			cmd;
175 
176 	/* In-capsule data buffer */
177 	uint8_t					*buf;
178 
179 	bool					data_from_pool;
180 	void					*buffers[SPDK_NVMF_MAX_SGL_ENTRIES];
181 
182 	/* transfer_tag */
183 	uint16_t				ttag;
184 
185 	/*
186 	 * next_expected_r2t_offset is used when we receive the h2c_data PDU.
187 	 */
188 	uint32_t				next_expected_r2t_offset;
189 	uint32_t				r2tl_remain;
190 
191 	/*
192 	 * c2h_data_offset is used when we send the c2h_data PDU.
193 	 */
194 	uint32_t				c2h_data_offset;
195 	uint32_t				c2h_data_pdu_num;
196 
197 	enum spdk_nvmf_tcp_req_state		state;
198 	bool					has_incapsule_data;
199 
200 	TAILQ_ENTRY(spdk_nvmf_tcp_req)		link;
201 	TAILQ_ENTRY(spdk_nvmf_tcp_req)		state_link;
202 };
203 
204 struct spdk_nvmf_tcp_qpair {
205 	struct spdk_nvmf_qpair			qpair;
206 	struct spdk_nvmf_tcp_poll_group		*group;
207 	struct spdk_nvmf_tcp_port		*port;
208 	struct spdk_sock			*sock;
209 	struct spdk_poller			*flush_poller;
210 
211 	enum nvme_tcp_pdu_recv_state		recv_state;
212 	enum nvme_tcp_qpair_state		state;
213 
214 	struct nvme_tcp_pdu			pdu_in_progress;
215 
216 	TAILQ_HEAD(, nvme_tcp_pdu)		send_queue;
217 	TAILQ_HEAD(, nvme_tcp_pdu)		free_queue;
218 
219 	struct nvme_tcp_pdu			*pdu;
220 	struct nvme_tcp_pdu			*pdu_pool;
221 	uint16_t				free_pdu_num;
222 
223 	/* Queues to track the requests in all states */
224 	TAILQ_HEAD(, spdk_nvmf_tcp_req)		state_queue[TCP_REQUEST_NUM_STATES];
225 	/* Number of requests in each state */
226 	int32_t					state_cntr[TCP_REQUEST_NUM_STATES];
227 
228 	uint32_t				maxr2t;
229 	uint32_t				pending_r2t;
230 	TAILQ_HEAD(, spdk_nvmf_tcp_req)		queued_c2h_data_tcp_req;
231 
232 	uint8_t					cpda;
233 
234 	/* Array of size "max_queue_depth * InCapsuleDataSize" containing
235 	 * buffers to be used for in capsule data.
236 	 */
237 	void					*buf;
238 	void					*bufs;
239 	struct spdk_nvmf_tcp_req			*req;
240 	struct spdk_nvmf_tcp_req			*reqs;
241 
242 	bool					host_hdgst_enable;
243 	bool					host_ddgst_enable;
244 
245 
246 	/* The maximum number of I/O outstanding on this connection at one time */
247 	uint16_t				max_queue_depth;
248 
249 
250 	/** Specifies the maximum number of PDU-Data bytes per H2C Data Transfer PDU */
251 	uint32_t				maxh2cdata;
252 
253 	uint32_t				c2h_data_pdu_cnt;
254 
255 	/* IP address */
256 	char					initiator_addr[SPDK_NVMF_TRADDR_MAX_LEN];
257 	char					target_addr[SPDK_NVMF_TRADDR_MAX_LEN];
258 
259 	/* IP port */
260 	uint16_t				initiator_port;
261 	uint16_t				target_port;
262 
263 	/* Timer used to destroy qpair after detecting transport error issue if initiator does
264 	 *  not close the connection.
265 	 */
266 	struct spdk_poller			*timeout_poller;
267 
268 	TAILQ_ENTRY(spdk_nvmf_tcp_qpair)	link;
269 };
270 
271 struct spdk_nvmf_tcp_poll_group {
272 	struct spdk_nvmf_transport_poll_group	group;
273 	struct spdk_sock_group			*sock_group;
274 
275 	/* Requests that are waiting to obtain a data buffer */
276 	TAILQ_HEAD(, spdk_nvmf_tcp_req)		pending_data_buf_queue;
277 
278 	TAILQ_HEAD(, spdk_nvmf_tcp_qpair)	qpairs;
279 };
280 
281 struct spdk_nvmf_tcp_port {
282 	struct spdk_nvme_transport_id		trid;
283 	struct spdk_sock			*listen_sock;
284 	uint32_t				ref;
285 	TAILQ_ENTRY(spdk_nvmf_tcp_port)		link;
286 };
287 
288 struct spdk_nvmf_tcp_transport {
289 	struct spdk_nvmf_transport		transport;
290 
291 	pthread_mutex_t				lock;
292 
293 	TAILQ_HEAD(, spdk_nvmf_tcp_port)	ports;
294 };
295 
296 static void spdk_nvmf_tcp_qpair_process_pending(struct spdk_nvmf_tcp_transport *ttransport,
297 		struct spdk_nvmf_tcp_qpair *tqpair);
298 static bool spdk_nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport,
299 				      struct spdk_nvmf_tcp_req *tcp_req);
300 static void spdk_nvmf_tcp_handle_pending_c2h_data_queue(struct spdk_nvmf_tcp_qpair *tqpair);
301 
302 static void
303 spdk_nvmf_tcp_req_set_state(struct spdk_nvmf_tcp_req *tcp_req,
304 			    enum spdk_nvmf_tcp_req_state state)
305 {
306 	struct spdk_nvmf_qpair *qpair;
307 	struct spdk_nvmf_tcp_qpair *tqpair;
308 
309 	qpair = tcp_req->req.qpair;
310 	tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
311 
312 	TAILQ_REMOVE(&tqpair->state_queue[tcp_req->state], tcp_req, state_link);
313 	tqpair->state_cntr[tcp_req->state]--;
314 	assert(tqpair->state_cntr[tcp_req->state] >= 0);
315 
316 	TAILQ_INSERT_TAIL(&tqpair->state_queue[state], tcp_req, state_link);
317 	tqpair->state_cntr[state]++;
318 
319 	tcp_req->state = state;
320 }
321 
322 static struct nvme_tcp_pdu *
323 spdk_nvmf_tcp_pdu_get(struct spdk_nvmf_tcp_qpair *tqpair)
324 {
325 	struct nvme_tcp_pdu *pdu;
326 
327 	pdu = TAILQ_FIRST(&tqpair->free_queue);
328 	if (!pdu) {
329 		SPDK_ERRLOG("Unable to get PDU for tqpair=%p\n", tqpair);
330 		abort();
331 		return NULL;
332 	}
333 
334 	tqpair->free_pdu_num--;
335 	TAILQ_REMOVE(&tqpair->free_queue, pdu, tailq);
336 	memset(pdu, 0, sizeof(*pdu));
337 	pdu->ref = 1;
338 
339 	return pdu;
340 }
341 
342 static void
343 spdk_nvmf_tcp_pdu_put(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu)
344 {
345 	if (!pdu) {
346 		return;
347 	}
348 
349 	assert(pdu->ref > 0);
350 
351 	pdu->ref--;
352 	if (pdu->ref == 0) {
353 		tqpair->free_pdu_num++;
354 		TAILQ_INSERT_HEAD(&tqpair->free_queue, pdu, tailq);
355 	}
356 }
357 
358 static struct spdk_nvmf_tcp_req *
359 spdk_nvmf_tcp_req_get(struct spdk_nvmf_tcp_qpair *tqpair)
360 {
361 	struct spdk_nvmf_tcp_req *tcp_req;
362 
363 	tcp_req = TAILQ_FIRST(&tqpair->state_queue[TCP_REQUEST_STATE_FREE]);
364 	if (!tcp_req) {
365 		SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair);
366 		return NULL;
367 	}
368 
369 	memset(&tcp_req->cmd, 0, sizeof(tcp_req->cmd));
370 	memset(&tcp_req->rsp, 0, sizeof(tcp_req->rsp));
371 	tcp_req->next_expected_r2t_offset = 0;
372 	tcp_req->r2tl_remain = 0;
373 	tcp_req->c2h_data_offset = 0;
374 	tcp_req->has_incapsule_data = false;
375 
376 	spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW);
377 	return tcp_req;
378 }
379 
380 static void
381 nvmf_tcp_request_free(struct spdk_nvmf_tcp_req *tcp_req)
382 {
383 	struct spdk_nvmf_tcp_transport *ttransport;
384 
385 	if (!tcp_req) {
386 		return;
387 	}
388 
389 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tcp_req=%p will be freed\n", tcp_req);
390 	ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport,
391 				      struct spdk_nvmf_tcp_transport, transport);
392 	spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED);
393 	spdk_nvmf_tcp_req_process(ttransport, tcp_req);
394 }
395 
396 static int
397 spdk_nvmf_tcp_req_free(struct spdk_nvmf_request *req)
398 {
399 	struct spdk_nvmf_tcp_req *tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req);
400 
401 	nvmf_tcp_request_free(tcp_req);
402 
403 	return 0;
404 }
405 
406 static void
407 spdk_nvmf_tcp_drain_state_queue(struct spdk_nvmf_tcp_qpair *tqpair,
408 				enum spdk_nvmf_tcp_req_state state)
409 {
410 	struct spdk_nvmf_tcp_req *tcp_req, *req_tmp;
411 
412 	TAILQ_FOREACH_SAFE(tcp_req, &tqpair->state_queue[state], state_link, req_tmp) {
413 		nvmf_tcp_request_free(tcp_req);
414 	}
415 }
416 
417 static void
418 spdk_nvmf_tcp_cleanup_all_states(struct spdk_nvmf_tcp_qpair *tqpair)
419 {
420 	struct spdk_nvmf_tcp_req *tcp_req, *req_tmp;
421 	struct nvme_tcp_pdu *pdu, *tmp_pdu;
422 
423 	/* Free the pdus in the send_queue */
424 	TAILQ_FOREACH_SAFE(pdu, &tqpair->send_queue, tailq, tmp_pdu) {
425 		TAILQ_REMOVE(&tqpair->send_queue, pdu, tailq);
426 		/* Also check the pdu type, we need to calculte the c2h_data_pdu_cnt later */
427 		if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_DATA) {
428 			assert(tqpair->c2h_data_pdu_cnt > 0);
429 			tqpair->c2h_data_pdu_cnt--;
430 		}
431 		spdk_nvmf_tcp_pdu_put(tqpair, pdu);
432 	}
433 
434 	TAILQ_FOREACH_SAFE(tcp_req, &tqpair->queued_c2h_data_tcp_req, link, req_tmp) {
435 		TAILQ_REMOVE(&tqpair->queued_c2h_data_tcp_req, tcp_req, link);
436 	}
437 	spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST);
438 
439 	spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEW);
440 
441 	spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_DATA_PENDING_FOR_R2T);
442 
443 	/* Wipe the requests waiting for buffer from the global list */
444 	TAILQ_FOREACH_SAFE(tcp_req, &tqpair->state_queue[TCP_REQUEST_STATE_NEED_BUFFER], state_link,
445 			   req_tmp) {
446 		TAILQ_REMOVE(&tqpair->group->pending_data_buf_queue, tcp_req, link);
447 	}
448 
449 	spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEED_BUFFER);
450 	spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_EXECUTING);
451 	spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER);
452 }
453 
454 static void
455 nvmf_tcp_dump_qpair_req_contents(struct spdk_nvmf_tcp_qpair *tqpair)
456 {
457 	int i;
458 	struct spdk_nvmf_tcp_req *tcp_req;
459 
460 	SPDK_ERRLOG("Dumping contents of queue pair (QID %d)\n", tqpair->qpair.qid);
461 	for (i = 1; i < TCP_REQUEST_NUM_STATES; i++) {
462 		SPDK_ERRLOG("\tNum of requests in state[%d] = %d\n", i, tqpair->state_cntr[i]);
463 		TAILQ_FOREACH(tcp_req, &tqpair->state_queue[i], state_link) {
464 			SPDK_ERRLOG("\t\tRequest Data From Pool: %d\n", tcp_req->data_from_pool);
465 			SPDK_ERRLOG("\t\tRequest opcode: %d\n", tcp_req->req.cmd->nvmf_cmd.opcode);
466 		}
467 	}
468 }
469 
470 static void
471 spdk_nvmf_tcp_qpair_destroy(struct spdk_nvmf_tcp_qpair *tqpair)
472 {
473 	int err = 0;
474 
475 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n");
476 
477 	spdk_poller_unregister(&tqpair->flush_poller);
478 	spdk_sock_close(&tqpair->sock);
479 	spdk_nvmf_tcp_cleanup_all_states(tqpair);
480 
481 	if (tqpair->free_pdu_num != (tqpair->max_queue_depth + NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM)) {
482 		SPDK_ERRLOG("tqpair(%p) free pdu pool num is %u but should be %u\n", tqpair,
483 			    tqpair->free_pdu_num,
484 			    (tqpair->max_queue_depth + NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM));
485 		err++;
486 	}
487 
488 	if (tqpair->state_cntr[TCP_REQUEST_STATE_FREE] != tqpair->max_queue_depth) {
489 		SPDK_ERRLOG("tqpair(%p) free tcp request num is %u but should be %u\n", tqpair,
490 			    tqpair->state_cntr[TCP_REQUEST_STATE_FREE],
491 			    tqpair->max_queue_depth);
492 		err++;
493 	}
494 
495 	if (tqpair->c2h_data_pdu_cnt != 0) {
496 		SPDK_ERRLOG("tqpair(%p) free c2h_data_pdu cnt is %u but should be 0\n", tqpair,
497 			    tqpair->c2h_data_pdu_cnt);
498 		err++;
499 	}
500 
501 	if (err > 0) {
502 		nvmf_tcp_dump_qpair_req_contents(tqpair);
503 	}
504 	free(tqpair->pdu);
505 	free(tqpair->pdu_pool);
506 	free(tqpair->req);
507 	free(tqpair->reqs);
508 	spdk_dma_free(tqpair->buf);
509 	spdk_dma_free(tqpair->bufs);
510 	free(tqpair);
511 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Leave\n");
512 }
513 
514 static int
515 spdk_nvmf_tcp_destroy(struct spdk_nvmf_transport *transport)
516 {
517 	struct spdk_nvmf_tcp_transport	*ttransport;
518 
519 	assert(transport != NULL);
520 	ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport);
521 
522 	pthread_mutex_destroy(&ttransport->lock);
523 	free(ttransport);
524 	return 0;
525 }
526 
527 static struct spdk_nvmf_transport *
528 spdk_nvmf_tcp_create(struct spdk_nvmf_transport_opts *opts)
529 {
530 	struct spdk_nvmf_tcp_transport *ttransport;
531 	uint32_t sge_count;
532 	uint32_t min_shared_buffers;
533 
534 	ttransport = calloc(1, sizeof(*ttransport));
535 	if (!ttransport) {
536 		return NULL;
537 	}
538 
539 	TAILQ_INIT(&ttransport->ports);
540 
541 	ttransport->transport.ops = &spdk_nvmf_transport_tcp;
542 
543 	SPDK_NOTICELOG("*** TCP Transport Init ***\n");
544 
545 	SPDK_INFOLOG(SPDK_LOG_NVMF_TCP, "*** TCP Transport Init ***\n"
546 		     "  Transport opts:  max_ioq_depth=%d, max_io_size=%d,\n"
547 		     "  max_qpairs_per_ctrlr=%d, io_unit_size=%d,\n"
548 		     "  in_capsule_data_size=%d, max_aq_depth=%d\n"
549 		     "  num_shared_buffers=%d\n",
550 		     opts->max_queue_depth,
551 		     opts->max_io_size,
552 		     opts->max_qpairs_per_ctrlr,
553 		     opts->io_unit_size,
554 		     opts->in_capsule_data_size,
555 		     opts->max_aq_depth,
556 		     opts->num_shared_buffers);
557 
558 	/* I/O unit size cannot be larger than max I/O size */
559 	if (opts->io_unit_size > opts->max_io_size) {
560 		opts->io_unit_size = opts->max_io_size;
561 	}
562 
563 	sge_count = opts->max_io_size / opts->io_unit_size;
564 	if (sge_count > SPDK_NVMF_MAX_SGL_ENTRIES) {
565 		SPDK_ERRLOG("Unsupported IO Unit size specified, %d bytes\n", opts->io_unit_size);
566 		free(ttransport);
567 		return NULL;
568 	}
569 
570 	min_shared_buffers = spdk_thread_get_count() * opts->buf_cache_size;
571 	if (min_shared_buffers > opts->num_shared_buffers) {
572 		SPDK_ERRLOG("There are not enough buffers to satisfy"
573 			    "per-poll group caches for each thread. (%" PRIu32 ")"
574 			    "supplied. (%" PRIu32 ") required\n", opts->num_shared_buffers, min_shared_buffers);
575 		SPDK_ERRLOG("Please specify a larger number of shared buffers\n");
576 		spdk_nvmf_tcp_destroy(&ttransport->transport);
577 		return NULL;
578 	}
579 
580 	pthread_mutex_init(&ttransport->lock, NULL);
581 
582 	return &ttransport->transport;
583 }
584 
585 static int
586 _spdk_nvmf_tcp_trsvcid_to_int(const char *trsvcid)
587 {
588 	unsigned long long ull;
589 	char *end = NULL;
590 
591 	ull = strtoull(trsvcid, &end, 10);
592 	if (end == NULL || end == trsvcid || *end != '\0') {
593 		return -1;
594 	}
595 
596 	/* Valid TCP/IP port numbers are in [0, 65535] */
597 	if (ull > 65535) {
598 		return -1;
599 	}
600 
601 	return (int)ull;
602 }
603 
604 /**
605  * Canonicalize a listen address trid.
606  */
607 static int
608 _spdk_nvmf_tcp_canon_listen_trid(struct spdk_nvme_transport_id *canon_trid,
609 				 const struct spdk_nvme_transport_id *trid)
610 {
611 	int trsvcid_int;
612 
613 	trsvcid_int = _spdk_nvmf_tcp_trsvcid_to_int(trid->trsvcid);
614 	if (trsvcid_int < 0) {
615 		return -EINVAL;
616 	}
617 
618 	memset(canon_trid, 0, sizeof(*canon_trid));
619 	canon_trid->trtype = SPDK_NVME_TRANSPORT_TCP;
620 	canon_trid->adrfam = trid->adrfam;
621 	snprintf(canon_trid->traddr, sizeof(canon_trid->traddr), "%s", trid->traddr);
622 	snprintf(canon_trid->trsvcid, sizeof(canon_trid->trsvcid), "%d", trsvcid_int);
623 
624 	return 0;
625 }
626 
627 /**
628  * Find an existing listening port.
629  *
630  * Caller must hold ttransport->lock.
631  */
632 static struct spdk_nvmf_tcp_port *
633 _spdk_nvmf_tcp_find_port(struct spdk_nvmf_tcp_transport *ttransport,
634 			 const struct spdk_nvme_transport_id *trid)
635 {
636 	struct spdk_nvme_transport_id canon_trid;
637 	struct spdk_nvmf_tcp_port *port;
638 
639 	if (_spdk_nvmf_tcp_canon_listen_trid(&canon_trid, trid) != 0) {
640 		return NULL;
641 	}
642 
643 	TAILQ_FOREACH(port, &ttransport->ports, link) {
644 		if (spdk_nvme_transport_id_compare(&canon_trid, &port->trid) == 0) {
645 			return port;
646 		}
647 	}
648 
649 	return NULL;
650 }
651 
652 static int
653 spdk_nvmf_tcp_listen(struct spdk_nvmf_transport *transport,
654 		     const struct spdk_nvme_transport_id *trid)
655 {
656 	struct spdk_nvmf_tcp_transport *ttransport;
657 	struct spdk_nvmf_tcp_port *port;
658 	int trsvcid_int;
659 	uint8_t adrfam;
660 
661 	ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport);
662 
663 	trsvcid_int = _spdk_nvmf_tcp_trsvcid_to_int(trid->trsvcid);
664 	if (trsvcid_int < 0) {
665 		SPDK_ERRLOG("Invalid trsvcid '%s'\n", trid->trsvcid);
666 		return -EINVAL;
667 	}
668 
669 	pthread_mutex_lock(&ttransport->lock);
670 
671 	port = _spdk_nvmf_tcp_find_port(ttransport, trid);
672 	if (port) {
673 		SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Already listening on %s port %s\n",
674 			      trid->traddr, trid->trsvcid);
675 		port->ref++;
676 		pthread_mutex_unlock(&ttransport->lock);
677 		return 0;
678 	}
679 
680 	port = calloc(1, sizeof(*port));
681 	if (!port) {
682 		SPDK_ERRLOG("Port allocation failed\n");
683 		free(port);
684 		pthread_mutex_unlock(&ttransport->lock);
685 		return -ENOMEM;
686 	}
687 
688 	port->ref = 1;
689 
690 	if (_spdk_nvmf_tcp_canon_listen_trid(&port->trid, trid) != 0) {
691 		SPDK_ERRLOG("Invalid traddr %s / trsvcid %s\n",
692 			    trid->traddr, trid->trsvcid);
693 		free(port);
694 		pthread_mutex_unlock(&ttransport->lock);
695 		return -ENOMEM;
696 	}
697 
698 	port->listen_sock = spdk_sock_listen(trid->traddr, trsvcid_int);
699 	if (port->listen_sock == NULL) {
700 		SPDK_ERRLOG("spdk_sock_listen(%s, %d) failed: %s (%d)\n",
701 			    trid->traddr, trsvcid_int,
702 			    spdk_strerror(errno), errno);
703 		free(port);
704 		pthread_mutex_unlock(&ttransport->lock);
705 		return -errno;
706 	}
707 
708 	if (spdk_sock_is_ipv4(port->listen_sock)) {
709 		adrfam = SPDK_NVMF_ADRFAM_IPV4;
710 	} else if (spdk_sock_is_ipv6(port->listen_sock)) {
711 		adrfam = SPDK_NVMF_ADRFAM_IPV6;
712 	} else {
713 		SPDK_ERRLOG("Unhandled socket type\n");
714 		adrfam = 0;
715 	}
716 
717 	if (adrfam != trid->adrfam) {
718 		SPDK_ERRLOG("Socket address family mismatch\n");
719 		spdk_sock_close(&port->listen_sock);
720 		free(port);
721 		pthread_mutex_unlock(&ttransport->lock);
722 		return -EINVAL;
723 	}
724 
725 	SPDK_NOTICELOG("*** NVMe/TCP Target Listening on %s port %d ***\n",
726 		       trid->traddr, trsvcid_int);
727 
728 	TAILQ_INSERT_TAIL(&ttransport->ports, port, link);
729 	pthread_mutex_unlock(&ttransport->lock);
730 
731 	return 0;
732 }
733 
734 static int
735 spdk_nvmf_tcp_stop_listen(struct spdk_nvmf_transport *transport,
736 			  const struct spdk_nvme_transport_id *trid)
737 {
738 	struct spdk_nvmf_tcp_transport *ttransport;
739 	struct spdk_nvmf_tcp_port *port;
740 	int rc;
741 
742 	ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport);
743 
744 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Removing listen address %s port %s\n",
745 		      trid->traddr, trid->trsvcid);
746 
747 	pthread_mutex_lock(&ttransport->lock);
748 	port = _spdk_nvmf_tcp_find_port(ttransport, trid);
749 	if (port) {
750 		assert(port->ref > 0);
751 		port->ref--;
752 		if (port->ref == 0) {
753 			TAILQ_REMOVE(&ttransport->ports, port, link);
754 			spdk_sock_close(&port->listen_sock);
755 			free(port);
756 		}
757 		rc = 0;
758 	} else {
759 		SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Port not found\n");
760 		rc = -ENOENT;
761 	}
762 	pthread_mutex_unlock(&ttransport->lock);
763 
764 	return rc;
765 }
766 
767 static int
768 spdk_nvmf_tcp_qpair_flush_pdus_internal(struct spdk_nvmf_tcp_qpair *tqpair)
769 {
770 	const int array_size = 32;
771 	struct iovec	iovec_array[array_size];
772 	struct iovec	*iov = iovec_array;
773 	int iovec_cnt = 0;
774 	int bytes = 0;
775 	int total_length = 0;
776 	uint32_t writev_offset;
777 	struct nvme_tcp_pdu *pdu;
778 	int pdu_length;
779 	TAILQ_HEAD(, nvme_tcp_pdu) completed_pdus_list;
780 	struct spdk_nvmf_tcp_transport *ttransport;
781 
782 	pdu = TAILQ_FIRST(&tqpair->send_queue);
783 
784 	if (pdu == NULL) {
785 		return 0;
786 	}
787 
788 	/*
789 	 * Build up a list of iovecs for the first few PDUs in the
790 	 *  tqpair 's send_queue.
791 	 */
792 	while (pdu != NULL && ((array_size - iovec_cnt) >= 3)) {
793 		iovec_cnt += nvme_tcp_build_iovecs(&iovec_array[iovec_cnt],
794 						   pdu,
795 						   tqpair->host_hdgst_enable,
796 						   tqpair->host_ddgst_enable);
797 		total_length += pdu->hdr.common.plen;
798 		pdu = TAILQ_NEXT(pdu, tailq);
799 	}
800 
801 	/*
802 	 * Check if the first PDU was partially written out the last time
803 	 *  this function was called, and if so adjust the iovec array
804 	 *  accordingly.
805 	 */
806 	writev_offset = TAILQ_FIRST(&tqpair->send_queue)->writev_offset;
807 	total_length -= writev_offset;
808 	while ((writev_offset > 0) && (iovec_cnt > 0)) {
809 		if (writev_offset >= iov->iov_len) {
810 			writev_offset -= iov->iov_len;
811 			iov++;
812 			iovec_cnt--;
813 		} else {
814 			iov->iov_len -= writev_offset;
815 			iov->iov_base = (char *)iov->iov_base + writev_offset;
816 			writev_offset = 0;
817 		}
818 	}
819 
820 	spdk_trace_record(TRACE_TCP_FLUSH_WRITEBUF_START, 0, total_length, 0, iovec_cnt);
821 
822 	bytes = spdk_sock_writev(tqpair->sock, iov, iovec_cnt);
823 	if (bytes == -1) {
824 		if (errno == EWOULDBLOCK || errno == EAGAIN) {
825 			return 1;
826 		} else {
827 			SPDK_ERRLOG("spdk_sock_writev() failed, errno %d: %s\n",
828 				    errno, spdk_strerror(errno));
829 			return -1;
830 		}
831 	}
832 
833 	spdk_trace_record(TRACE_TCP_FLUSH_WRITEBUF_DONE, 0, bytes, 0, 0);
834 
835 	pdu = TAILQ_FIRST(&tqpair->send_queue);
836 
837 	/*
838 	 * Free any PDUs that were fully written.  If a PDU was only
839 	 *  partially written, update its writev_offset so that next
840 	 *  time only the unwritten portion will be sent to writev().
841 	 */
842 	TAILQ_INIT(&completed_pdus_list);
843 	while (bytes > 0) {
844 		pdu_length = pdu->hdr.common.plen - pdu->writev_offset;
845 		if (bytes >= pdu_length) {
846 			bytes -= pdu_length;
847 			TAILQ_REMOVE(&tqpair->send_queue, pdu, tailq);
848 			TAILQ_INSERT_TAIL(&completed_pdus_list, pdu, tailq);
849 			pdu = TAILQ_FIRST(&tqpair->send_queue);
850 
851 		} else {
852 			pdu->writev_offset += bytes;
853 			bytes = 0;
854 		}
855 	}
856 
857 	while (!TAILQ_EMPTY(&completed_pdus_list)) {
858 		pdu = TAILQ_FIRST(&completed_pdus_list);
859 		TAILQ_REMOVE(&completed_pdus_list, pdu, tailq);
860 		assert(pdu->cb_fn != NULL);
861 		pdu->cb_fn(pdu->cb_arg);
862 		spdk_nvmf_tcp_pdu_put(tqpair, pdu);
863 	}
864 
865 	ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport);
866 	spdk_nvmf_tcp_qpair_process_pending(ttransport, tqpair);
867 
868 	return TAILQ_EMPTY(&tqpair->send_queue) ? 0 : 1;
869 }
870 
871 static int
872 spdk_nvmf_tcp_qpair_flush_pdus(void *_tqpair)
873 {
874 	struct spdk_nvmf_tcp_qpair *tqpair = _tqpair;
875 	int rc;
876 
877 	if (tqpair->state == NVME_TCP_QPAIR_STATE_RUNNING) {
878 		rc = spdk_nvmf_tcp_qpair_flush_pdus_internal(tqpair);
879 		if (rc == 0 && tqpair->flush_poller != NULL) {
880 			spdk_poller_unregister(&tqpair->flush_poller);
881 		} else if (rc == 1 && tqpair->flush_poller == NULL) {
882 			tqpair->flush_poller = spdk_poller_register(spdk_nvmf_tcp_qpair_flush_pdus,
883 					       tqpair, 50);
884 		}
885 	} else {
886 		/*
887 		 * If the tqpair state is not RUNNING, then
888 		 * keep trying to flush PDUs until our list is
889 		 * empty - to make sure all data is sent before
890 		 * closing the connection.
891 		 */
892 		do {
893 			rc = spdk_nvmf_tcp_qpair_flush_pdus_internal(tqpair);
894 		} while (rc == 1);
895 	}
896 
897 	if (rc < 0 && tqpair->state < NVME_TCP_QPAIR_STATE_EXITING) {
898 		/*
899 		 * If the poller has already started destruction of the tqpair,
900 		 *  i.e. the socket read failed, then the connection state may already
901 		 *  be EXITED.  We don't want to set it back to EXITING in that case.
902 		 */
903 		tqpair->state = NVME_TCP_QPAIR_STATE_EXITING;
904 	}
905 
906 	return -1;
907 }
908 
909 static void
910 spdk_nvmf_tcp_qpair_write_pdu(struct spdk_nvmf_tcp_qpair *tqpair,
911 			      struct nvme_tcp_pdu *pdu,
912 			      nvme_tcp_qpair_xfer_complete_cb cb_fn,
913 			      void *cb_arg)
914 {
915 	int enable_digest;
916 	int hlen;
917 	uint32_t crc32c;
918 
919 	hlen = pdu->hdr.common.hlen;
920 	enable_digest = 1;
921 	if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP ||
922 	    pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ) {
923 		/* this PDU should be sent without digest */
924 		enable_digest = 0;
925 	}
926 
927 	/* Header Digest */
928 	if (enable_digest && tqpair->host_hdgst_enable) {
929 		crc32c = nvme_tcp_pdu_calc_header_digest(pdu);
930 		MAKE_DIGEST_WORD((uint8_t *)pdu->hdr.raw + hlen, crc32c);
931 	}
932 
933 	/* Data Digest */
934 	if (pdu->data_len > 0 && enable_digest && tqpair->host_ddgst_enable) {
935 		crc32c = nvme_tcp_pdu_calc_data_digest(pdu);
936 		MAKE_DIGEST_WORD(pdu->data_digest, crc32c);
937 	}
938 
939 	pdu->cb_fn = cb_fn;
940 	pdu->cb_arg = cb_arg;
941 	TAILQ_INSERT_TAIL(&tqpair->send_queue, pdu, tailq);
942 	spdk_nvmf_tcp_qpair_flush_pdus(tqpair);
943 }
944 
945 static int
946 spdk_nvmf_tcp_qpair_init_mem_resource(struct spdk_nvmf_tcp_qpair *tqpair, uint16_t size)
947 {
948 	int i;
949 	struct spdk_nvmf_tcp_req *tcp_req;
950 	struct spdk_nvmf_transport *transport = tqpair->qpair.transport;
951 	struct spdk_nvmf_tcp_transport *ttransport;
952 	ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport);
953 
954 	if (!tqpair->qpair.sq_head_max) {
955 		tqpair->req = calloc(1, sizeof(*tqpair->req));
956 		if (!tqpair->req) {
957 			SPDK_ERRLOG("Unable to allocate req on tqpair=%p.\n", tqpair);
958 			return -1;
959 		}
960 
961 		if (transport->opts.in_capsule_data_size) {
962 			tqpair->buf = spdk_dma_zmalloc(ttransport->transport.opts.in_capsule_data_size, 0x1000, NULL);
963 			if (!tqpair->buf) {
964 				SPDK_ERRLOG("Unable to allocate buf on tqpair=%p.\n", tqpair);
965 				return -1;
966 			}
967 		}
968 
969 		tcp_req = tqpair->req;
970 		tcp_req->ttag = 0;
971 		tcp_req->req.qpair = &tqpair->qpair;
972 
973 		/* Set up memory to receive commands */
974 		if (tqpair->buf) {
975 			tcp_req->buf = tqpair->buf;
976 		}
977 
978 		/* Set the cmdn and rsp */
979 		tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp;
980 		tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd;
981 
982 		/* Initialize request state to FREE */
983 		tcp_req->state = TCP_REQUEST_STATE_FREE;
984 		TAILQ_INSERT_TAIL(&tqpair->state_queue[tcp_req->state], tcp_req, state_link);
985 
986 		tqpair->pdu = calloc(NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM + 1, sizeof(*tqpair->pdu));
987 		if (!tqpair->pdu) {
988 			SPDK_ERRLOG("Unable to allocate pdu on tqpair=%p.\n", tqpair);
989 			return -1;
990 		}
991 
992 		for (i = 0; i < 1 + NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM; i++) {
993 			TAILQ_INSERT_TAIL(&tqpair->free_queue, &tqpair->pdu[i], tailq);
994 		}
995 
996 	} else {
997 		tqpair->reqs = calloc(size, sizeof(*tqpair->reqs));
998 		if (!tqpair->reqs) {
999 			SPDK_ERRLOG("Unable to allocate reqs on tqpair=%p\n", tqpair);
1000 			return -1;
1001 		}
1002 
1003 		if (transport->opts.in_capsule_data_size) {
1004 			tqpair->bufs = spdk_dma_zmalloc(size * transport->opts.in_capsule_data_size,
1005 							0x1000, NULL);
1006 			if (!tqpair->bufs) {
1007 				SPDK_ERRLOG("Unable to allocate bufs on tqpair=%p.\n", tqpair);
1008 				return -1;
1009 			}
1010 		}
1011 
1012 		for (i = 0; i < size; i++) {
1013 			struct spdk_nvmf_tcp_req *tcp_req = &tqpair->reqs[i];
1014 
1015 			tcp_req->ttag = i + 1;
1016 			tcp_req->req.qpair = &tqpair->qpair;
1017 
1018 			/* Set up memory to receive commands */
1019 			if (tqpair->bufs) {
1020 				tcp_req->buf = (void *)((uintptr_t)tqpair->bufs + (i * transport->opts.in_capsule_data_size));
1021 			}
1022 
1023 			/* Set the cmdn and rsp */
1024 			tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp;
1025 			tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd;
1026 
1027 			/* Initialize request state to FREE */
1028 			tcp_req->state = TCP_REQUEST_STATE_FREE;
1029 			TAILQ_INSERT_TAIL(&tqpair->state_queue[tcp_req->state], tcp_req, state_link);
1030 		}
1031 
1032 		tqpair->pdu_pool = calloc(size, sizeof(*tqpair->pdu_pool));
1033 		if (!tqpair->pdu_pool) {
1034 			SPDK_ERRLOG("Unable to allocate pdu pool on tqpair =%p.\n", tqpair);
1035 			return -1;
1036 		}
1037 
1038 		for (i = 0; i < size; i++) {
1039 			TAILQ_INSERT_TAIL(&tqpair->free_queue, &tqpair->pdu_pool[i], tailq);
1040 		}
1041 	}
1042 
1043 	return 0;
1044 }
1045 
1046 static int
1047 spdk_nvmf_tcp_qpair_init(struct spdk_nvmf_qpair *qpair)
1048 {
1049 	struct spdk_nvmf_tcp_qpair *tqpair;
1050 	int i;
1051 
1052 	tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
1053 
1054 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "New TCP Connection: %p\n", qpair);
1055 
1056 	TAILQ_INIT(&tqpair->send_queue);
1057 	TAILQ_INIT(&tqpair->free_queue);
1058 	TAILQ_INIT(&tqpair->queued_c2h_data_tcp_req);
1059 
1060 	/* Initialise request state queues of the qpair */
1061 	for (i = TCP_REQUEST_STATE_FREE; i < TCP_REQUEST_NUM_STATES; i++) {
1062 		TAILQ_INIT(&tqpair->state_queue[i]);
1063 	}
1064 
1065 	tqpair->host_hdgst_enable = true;
1066 	tqpair->host_ddgst_enable = true;
1067 
1068 	return 0;
1069 }
1070 
1071 static int
1072 spdk_nvmf_tcp_qpair_sock_init(struct spdk_nvmf_tcp_qpair *tqpair)
1073 {
1074 
1075 	int rc;
1076 	int buf_size;
1077 
1078 	/* set recv buffer size */
1079 	buf_size = 2 * 1024 * 1024;
1080 	rc = spdk_sock_set_recvbuf(tqpair->sock, buf_size);
1081 	if (rc != 0) {
1082 		SPDK_ERRLOG("spdk_sock_set_recvbuf failed\n");
1083 		return rc;
1084 	}
1085 
1086 	/* set send buffer size */
1087 	rc = spdk_sock_set_sendbuf(tqpair->sock, buf_size);
1088 	if (rc != 0) {
1089 		SPDK_ERRLOG("spdk_sock_set_sendbuf failed\n");
1090 		return rc;
1091 	}
1092 
1093 	/* set low water mark */
1094 	rc = spdk_sock_set_recvlowat(tqpair->sock, sizeof(struct spdk_nvme_tcp_c2h_data_hdr));
1095 	if (rc != 0) {
1096 		SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n");
1097 		return rc;
1098 	}
1099 
1100 	return 0;
1101 }
1102 
1103 static void
1104 _spdk_nvmf_tcp_handle_connect(struct spdk_nvmf_transport *transport,
1105 			      struct spdk_nvmf_tcp_port *port,
1106 			      struct spdk_sock *sock, new_qpair_fn cb_fn)
1107 {
1108 	struct spdk_nvmf_tcp_qpair *tqpair;
1109 	int rc;
1110 
1111 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "New connection accepted on %s port %s\n",
1112 		      port->trid.traddr, port->trid.trsvcid);
1113 
1114 	tqpair = calloc(1, sizeof(struct spdk_nvmf_tcp_qpair));
1115 	if (tqpair == NULL) {
1116 		SPDK_ERRLOG("Could not allocate new connection.\n");
1117 		spdk_sock_close(&sock);
1118 		return;
1119 	}
1120 
1121 	tqpair->sock = sock;
1122 	tqpair->max_queue_depth = 1;
1123 	tqpair->free_pdu_num = tqpair->max_queue_depth + NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM;
1124 	tqpair->state_cntr[TCP_REQUEST_STATE_FREE] = tqpair->max_queue_depth;
1125 	tqpair->port = port;
1126 	tqpair->qpair.transport = transport;
1127 
1128 	rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr,
1129 			       sizeof(tqpair->target_addr), &tqpair->target_port,
1130 			       tqpair->initiator_addr, sizeof(tqpair->initiator_addr),
1131 			       &tqpair->initiator_port);
1132 	if (rc < 0) {
1133 		SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair);
1134 		spdk_nvmf_tcp_qpair_destroy(tqpair);
1135 		return;
1136 	}
1137 
1138 	cb_fn(&tqpair->qpair);
1139 }
1140 
1141 static void
1142 spdk_nvmf_tcp_port_accept(struct spdk_nvmf_transport *transport, struct spdk_nvmf_tcp_port *port,
1143 			  new_qpair_fn cb_fn)
1144 {
1145 	struct spdk_sock *sock;
1146 	int i;
1147 
1148 	for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) {
1149 		sock = spdk_sock_accept(port->listen_sock);
1150 		if (sock) {
1151 			_spdk_nvmf_tcp_handle_connect(transport, port, sock, cb_fn);
1152 		}
1153 	}
1154 }
1155 
1156 static void
1157 spdk_nvmf_tcp_accept(struct spdk_nvmf_transport *transport, new_qpair_fn cb_fn)
1158 {
1159 	struct spdk_nvmf_tcp_transport *ttransport;
1160 	struct spdk_nvmf_tcp_port *port;
1161 
1162 	ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport);
1163 
1164 	TAILQ_FOREACH(port, &ttransport->ports, link) {
1165 		spdk_nvmf_tcp_port_accept(transport, port, cb_fn);
1166 	}
1167 }
1168 
1169 static void
1170 spdk_nvmf_tcp_discover(struct spdk_nvmf_transport *transport,
1171 		       struct spdk_nvme_transport_id *trid,
1172 		       struct spdk_nvmf_discovery_log_page_entry *entry)
1173 {
1174 	entry->trtype = SPDK_NVMF_TRTYPE_TCP;
1175 	entry->adrfam = trid->adrfam;
1176 	entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_SPECIFIED;
1177 
1178 	spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' ');
1179 	spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' ');
1180 
1181 	entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE;
1182 }
1183 
1184 static struct spdk_nvmf_transport_poll_group *
1185 spdk_nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport)
1186 {
1187 	struct spdk_nvmf_tcp_poll_group *tgroup;
1188 
1189 	tgroup = calloc(1, sizeof(*tgroup));
1190 	if (!tgroup) {
1191 		return NULL;
1192 	}
1193 
1194 	tgroup->sock_group = spdk_sock_group_create();
1195 	if (!tgroup->sock_group) {
1196 		goto cleanup;
1197 	}
1198 
1199 	TAILQ_INIT(&tgroup->qpairs);
1200 	TAILQ_INIT(&tgroup->pending_data_buf_queue);
1201 
1202 	return &tgroup->group;
1203 
1204 cleanup:
1205 	free(tgroup);
1206 	return NULL;
1207 }
1208 
1209 static void
1210 spdk_nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group)
1211 {
1212 	struct spdk_nvmf_tcp_poll_group *tgroup;
1213 
1214 	tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group);
1215 	spdk_sock_group_close(&tgroup->sock_group);
1216 
1217 	if (!TAILQ_EMPTY(&tgroup->pending_data_buf_queue)) {
1218 		SPDK_ERRLOG("Pending I/O list wasn't empty on poll group destruction\n");
1219 	}
1220 
1221 	free(tgroup);
1222 }
1223 
1224 static void
1225 spdk_nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair,
1226 				   enum nvme_tcp_pdu_recv_state state)
1227 {
1228 	if (tqpair->recv_state == state) {
1229 		SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n",
1230 			    tqpair, state);
1231 		return;
1232 	}
1233 
1234 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tqpair(%p) recv state=%d\n", tqpair, state);
1235 	tqpair->recv_state = state;
1236 
1237 	switch (state) {
1238 	case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH:
1239 	case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH:
1240 	case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD:
1241 		break;
1242 	case NVME_TCP_PDU_RECV_STATE_ERROR:
1243 	case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY:
1244 		memset(&tqpair->pdu_in_progress, 0, sizeof(tqpair->pdu_in_progress));
1245 		break;
1246 	default:
1247 		SPDK_ERRLOG("The state(%d) is invalid\n", state);
1248 		abort();
1249 		break;
1250 	}
1251 }
1252 
1253 static int
1254 spdk_nvmf_tcp_qpair_handle_timeout(void *ctx)
1255 {
1256 	struct spdk_nvmf_tcp_qpair *tqpair = ctx;
1257 
1258 	assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR);
1259 
1260 	SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair,
1261 		    SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT);
1262 	tqpair->state = NVME_TCP_QPAIR_STATE_EXITED;
1263 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "will disconect the tqpair=%p\n", tqpair);
1264 	spdk_poller_unregister(&tqpair->timeout_poller);
1265 	spdk_nvmf_qpair_disconnect(&tqpair->qpair, NULL, NULL);
1266 
1267 	return 0;
1268 }
1269 
1270 static void
1271 spdk_nvmf_tcp_send_c2h_term_req_complete(void *cb_arg)
1272 {
1273 	struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg;
1274 
1275 	if (!tqpair->timeout_poller) {
1276 		tqpair->timeout_poller = spdk_poller_register(spdk_nvmf_tcp_qpair_handle_timeout, tqpair,
1277 					 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000);
1278 	}
1279 }
1280 
1281 static void
1282 spdk_nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu,
1283 				enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset)
1284 {
1285 	struct nvme_tcp_pdu *rsp_pdu;
1286 	struct spdk_nvme_tcp_term_req_hdr *c2h_term_req;
1287 	uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req);
1288 	uint32_t copy_len;
1289 
1290 	rsp_pdu = spdk_nvmf_tcp_pdu_get(tqpair);
1291 	if (!rsp_pdu) {
1292 		tqpair->state = NVME_TCP_QPAIR_STATE_EXITING;
1293 		spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR);
1294 		return;
1295 	}
1296 
1297 	c2h_term_req = &rsp_pdu->hdr.term_req;
1298 	c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ;
1299 	c2h_term_req->common.hlen = c2h_term_req_hdr_len;
1300 
1301 	if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) ||
1302 	    (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) {
1303 		DSET32(&c2h_term_req->fei, error_offset);
1304 	}
1305 
1306 	rsp_pdu->data = (uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len;
1307 	copy_len = pdu->hdr.common.hlen;
1308 	if (copy_len > SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE) {
1309 		copy_len = SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE;
1310 	}
1311 
1312 	/* Copy the error info into the buffer */
1313 	memcpy((uint8_t *)rsp_pdu->data, pdu->hdr.raw, copy_len);
1314 	rsp_pdu->data_len = copy_len;
1315 
1316 	/* Contain the header of the wrong received pdu */
1317 	c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len;
1318 	spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR);
1319 	spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_send_c2h_term_req_complete, tqpair);
1320 }
1321 
1322 static void
1323 spdk_nvmf_tcp_capsule_cmd_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 
1329 	tcp_req = spdk_nvmf_tcp_req_get(tqpair);
1330 	if (!tcp_req) {
1331 		SPDK_ERRLOG("Cannot allocate tcp_req\n");
1332 		tqpair->state = NVME_TCP_QPAIR_STATE_EXITING;
1333 		spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR);
1334 		return;
1335 	}
1336 
1337 	pdu->ctx = tcp_req;
1338 	spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW);
1339 	spdk_nvmf_tcp_req_process(ttransport, tcp_req);
1340 	return;
1341 }
1342 
1343 static void
1344 spdk_nvmf_tcp_capsule_cmd_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 	struct spdk_nvme_tcp_cmd *capsule_cmd;
1350 	uint32_t error_offset = 0;
1351 	enum spdk_nvme_tcp_term_req_fes fes;
1352 
1353 	capsule_cmd = &pdu->hdr.capsule_cmd;
1354 	tcp_req = pdu->ctx;
1355 	assert(tcp_req != NULL);
1356 	if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) {
1357 		SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n",
1358 			    SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo);
1359 		fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
1360 		error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo);
1361 		goto err;
1362 	}
1363 
1364 	spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY);
1365 	spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE);
1366 	spdk_nvmf_tcp_req_process(ttransport, tcp_req);
1367 
1368 	return;
1369 err:
1370 	spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
1371 }
1372 
1373 static void
1374 spdk_nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport,
1375 				  struct spdk_nvmf_tcp_qpair *tqpair,
1376 				  struct nvme_tcp_pdu *pdu)
1377 {
1378 	struct spdk_nvmf_tcp_req *tcp_req;
1379 	uint32_t error_offset = 0;
1380 	enum spdk_nvme_tcp_term_req_fes fes = 0;
1381 	struct spdk_nvme_tcp_h2c_data_hdr *h2c_data;
1382 	uint32_t iov_index;
1383 	bool ttag_offset_error = false;
1384 
1385 	h2c_data = &pdu->hdr.h2c_data;
1386 
1387 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n",
1388 		      tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag);
1389 
1390 	/* According to the information in the pdu to find the req */
1391 	TAILQ_FOREACH(tcp_req, &tqpair->state_queue[TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER],
1392 		      state_link) {
1393 		if ((tcp_req->req.cmd->nvme_cmd.cid == h2c_data->cccid) && (tcp_req->ttag == h2c_data->ttag)) {
1394 			break;
1395 		}
1396 
1397 		if (!ttag_offset_error && (tcp_req->req.cmd->nvme_cmd.cid == h2c_data->cccid)) {
1398 			ttag_offset_error = true;
1399 		}
1400 	}
1401 
1402 	if (!tcp_req) {
1403 		SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tcp_req is not found for tqpair=%p\n", tqpair);
1404 		fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER;
1405 		if (!ttag_offset_error) {
1406 			error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid);
1407 		} else {
1408 			error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag);
1409 		}
1410 		goto err;
1411 	}
1412 
1413 	if (tcp_req->next_expected_r2t_offset != h2c_data->datao) {
1414 		SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP,
1415 			      "tcp_req(%p), tqpair=%p,  expected_r2t_offset=%u, but data offset =%u\n",
1416 			      tcp_req, tqpair, tcp_req->next_expected_r2t_offset, h2c_data->datao);
1417 		fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE;
1418 		goto err;
1419 	}
1420 
1421 	if (h2c_data->datal > tqpair->maxh2cdata) {
1422 		SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tcp_req(%p), tqpair=%p,  datao=%u execeeds maxh2cdata size=%u\n",
1423 			      tcp_req, tqpair, h2c_data->datao, tqpair->maxh2cdata);
1424 		fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE;
1425 		goto err;
1426 	}
1427 
1428 	if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) {
1429 		SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP,
1430 			      "tcp_req(%p), tqpair=%p,  (datao=%u + datal=%u) execeeds requested length=%u\n",
1431 			      tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length);
1432 		fes = SPDK_NVME_TCP_TERM_REQ_FES_R2T_LIMIT_EXCEEDED;
1433 		goto err;
1434 	}
1435 
1436 	pdu->ctx = tcp_req;
1437 	pdu->data_len = h2c_data->datal;
1438 	iov_index = pdu->hdr.h2c_data.datao / ttransport->transport.opts.io_unit_size;
1439 	pdu->data = tcp_req->req.iov[iov_index].iov_base + (pdu->hdr.h2c_data.datao %
1440 			ttransport->transport.opts.io_unit_size);
1441 	spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD);
1442 	return;
1443 
1444 err:
1445 	spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
1446 }
1447 
1448 static void
1449 spdk_nvmf_tcp_pdu_cmd_complete(void *cb_arg)
1450 {
1451 	struct spdk_nvmf_tcp_req *tcp_req = cb_arg;
1452 	nvmf_tcp_request_free(tcp_req);
1453 }
1454 
1455 static void
1456 spdk_nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req,
1457 				    struct spdk_nvmf_tcp_qpair *tqpair)
1458 {
1459 	struct nvme_tcp_pdu *rsp_pdu;
1460 	struct spdk_nvme_tcp_rsp *capsule_resp;
1461 
1462 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter, tqpair=%p\n", tqpair);
1463 	rsp_pdu = spdk_nvmf_tcp_pdu_get(tqpair);
1464 	if (!rsp_pdu) {
1465 		spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR);
1466 		tqpair->state = NVME_TCP_QPAIR_STATE_EXITING;
1467 		return;
1468 	}
1469 
1470 	capsule_resp = &rsp_pdu->hdr.capsule_resp;
1471 	capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP;
1472 	capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp);
1473 	capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl;
1474 	if (tqpair->host_hdgst_enable) {
1475 		capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF;
1476 		capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN;
1477 	}
1478 
1479 	spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_pdu_cmd_complete, tcp_req);
1480 }
1481 
1482 static void
1483 spdk_nvmf_tcp_pdu_c2h_data_complete(void *cb_arg)
1484 {
1485 	struct spdk_nvmf_tcp_req *tcp_req = cb_arg;
1486 	struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair,
1487 					     struct spdk_nvmf_tcp_qpair, qpair);
1488 
1489 	assert(tqpair != NULL);
1490 	assert(tcp_req->c2h_data_pdu_num > 0);
1491 	tcp_req->c2h_data_pdu_num--;
1492 	if (!tcp_req->c2h_data_pdu_num) {
1493 #if LINUX_KERNEL_SUPPORT_NOT_SENDING_RESP_FOR_C2H
1494 		nvmf_tcp_request_free(tcp_req);
1495 #else
1496 		spdk_nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair);
1497 #endif
1498 	}
1499 
1500 	tqpair->c2h_data_pdu_cnt--;
1501 	spdk_nvmf_tcp_handle_pending_c2h_data_queue(tqpair);
1502 }
1503 
1504 static void
1505 spdk_nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair,
1506 			   struct spdk_nvmf_tcp_req *tcp_req)
1507 {
1508 	struct nvme_tcp_pdu *rsp_pdu;
1509 	struct spdk_nvme_tcp_r2t_hdr *r2t;
1510 
1511 	rsp_pdu = spdk_nvmf_tcp_pdu_get(tqpair);
1512 	if (!rsp_pdu) {
1513 		tqpair->state = NVME_TCP_QPAIR_STATE_EXITING;
1514 		spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR);
1515 		return;
1516 	}
1517 
1518 	r2t = &rsp_pdu->hdr.r2t;
1519 	r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T;
1520 	r2t->common.plen = r2t->common.hlen = sizeof(*r2t);
1521 
1522 	if (tqpair->host_hdgst_enable) {
1523 		r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF;
1524 		r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN;
1525 	}
1526 
1527 	r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid;
1528 	r2t->ttag = tcp_req->ttag;
1529 	r2t->r2to = tcp_req->next_expected_r2t_offset;
1530 	r2t->r2tl = spdk_min(tcp_req->req.length - tcp_req->next_expected_r2t_offset, tqpair->maxh2cdata);
1531 	tcp_req->r2tl_remain = r2t->r2tl;
1532 
1533 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP,
1534 		      "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n",
1535 		      tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl);
1536 	spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_pdu_cmd_complete, NULL);
1537 }
1538 
1539 static void
1540 spdk_nvmf_tcp_handle_queued_r2t_req(struct spdk_nvmf_tcp_qpair *tqpair)
1541 {
1542 	struct spdk_nvmf_tcp_req *tcp_req, *req_tmp;
1543 
1544 	TAILQ_FOREACH_SAFE(tcp_req, &tqpair->state_queue[TCP_REQUEST_STATE_DATA_PENDING_FOR_R2T],
1545 			   state_link, req_tmp) {
1546 		if (tqpair->pending_r2t < tqpair->maxr2t) {
1547 			tqpair->pending_r2t++;
1548 			spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER);
1549 			spdk_nvmf_tcp_send_r2t_pdu(tqpair, tcp_req);
1550 		} else {
1551 			break;
1552 		}
1553 	}
1554 }
1555 
1556 static void
1557 spdk_nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport,
1558 				      struct spdk_nvmf_tcp_qpair *tqpair,
1559 				      struct nvme_tcp_pdu *pdu)
1560 {
1561 	struct spdk_nvmf_tcp_req *tcp_req;
1562 
1563 	tcp_req = pdu->ctx;
1564 	assert(tcp_req != NULL);
1565 
1566 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n");
1567 
1568 	tcp_req->next_expected_r2t_offset += pdu->data_len;
1569 	tcp_req->r2tl_remain -= pdu->data_len;
1570 	spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY);
1571 
1572 	if (!tcp_req->r2tl_remain) {
1573 		if (tcp_req->next_expected_r2t_offset == tcp_req->req.length) {
1574 			assert(tqpair->pending_r2t > 0);
1575 			tqpair->pending_r2t--;
1576 			assert(tqpair->pending_r2t < tqpair->maxr2t);
1577 			spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE);
1578 			spdk_nvmf_tcp_req_process(ttransport, tcp_req);
1579 
1580 			spdk_nvmf_tcp_handle_queued_r2t_req(tqpair);
1581 		} else {
1582 			SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Send r2t pdu for tcp_req=%p on tqpair=%p\n", tcp_req, tqpair);
1583 			spdk_nvmf_tcp_send_r2t_pdu(tqpair, tcp_req);
1584 		}
1585 	}
1586 }
1587 
1588 static void
1589 spdk_nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req)
1590 {
1591 	SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req,
1592 		    spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]);
1593 	if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) ||
1594 	    (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) {
1595 		SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "The offset from the start of the PDU header is %u\n",
1596 			      DGET32(h2c_term_req->fei));
1597 	}
1598 }
1599 
1600 static void
1601 spdk_nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair,
1602 				      struct nvme_tcp_pdu *pdu)
1603 {
1604 	struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req;
1605 	uint32_t error_offset = 0;
1606 	enum spdk_nvme_tcp_term_req_fes fes;
1607 
1608 
1609 	if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) {
1610 		SPDK_ERRLOG("Fatal Error Stauts(FES) is unknown for h2c_term_req pdu=%p\n", pdu);
1611 		fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
1612 		error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes);
1613 		goto end;
1614 	}
1615 
1616 	/* set the data buffer */
1617 	pdu->data = (uint8_t *)pdu->hdr.raw + h2c_term_req->common.hlen;
1618 	pdu->data_len = h2c_term_req->common.plen - h2c_term_req->common.hlen;
1619 	spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD);
1620 	return;
1621 end:
1622 	spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
1623 	return;
1624 }
1625 
1626 static void
1627 spdk_nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair,
1628 		struct nvme_tcp_pdu *pdu)
1629 {
1630 	struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req;
1631 
1632 	spdk_nvmf_tcp_h2c_term_req_dump(h2c_term_req);
1633 	spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR);
1634 	return;
1635 }
1636 
1637 static void
1638 spdk_nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair)
1639 {
1640 	int rc = 0;
1641 	struct nvme_tcp_pdu *pdu;
1642 	uint32_t crc32c, error_offset = 0;
1643 	enum spdk_nvme_tcp_term_req_fes fes;
1644 	struct spdk_nvmf_tcp_transport *ttransport;
1645 
1646 	assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD);
1647 	pdu = &tqpair->pdu_in_progress;
1648 
1649 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n");
1650 	/* check data digest if need */
1651 	if (pdu->ddigest_valid_bytes) {
1652 		crc32c = nvme_tcp_pdu_calc_data_digest(pdu);
1653 		rc = MATCH_DIGEST_WORD(pdu->data_digest, crc32c);
1654 		if (rc == 0) {
1655 			SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu);
1656 			fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR;
1657 			spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
1658 			return;
1659 
1660 		}
1661 	}
1662 
1663 	ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport);
1664 	switch (pdu->hdr.common.pdu_type) {
1665 	case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD:
1666 		spdk_nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu);
1667 		break;
1668 	case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA:
1669 		spdk_nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu);
1670 		break;
1671 
1672 	case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ:
1673 		spdk_nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu);
1674 		break;
1675 
1676 	default:
1677 		/* The code should not go to here */
1678 		SPDK_ERRLOG("The code should not go to here\n");
1679 		break;
1680 	}
1681 }
1682 
1683 static void
1684 spdk_nvmf_tcp_send_icresp_complete(void *cb_arg)
1685 {
1686 	struct spdk_nvmf_tcp_qpair *tqpair = cb_arg;
1687 
1688 	tqpair->state = NVME_TCP_QPAIR_STATE_RUNNING;
1689 }
1690 
1691 static void
1692 spdk_nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport,
1693 			   struct spdk_nvmf_tcp_qpair *tqpair,
1694 			   struct nvme_tcp_pdu *pdu)
1695 {
1696 	struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr.ic_req;
1697 	struct nvme_tcp_pdu *rsp_pdu;
1698 	struct spdk_nvme_tcp_ic_resp *ic_resp;
1699 	uint32_t error_offset = 0;
1700 	enum spdk_nvme_tcp_term_req_fes fes;
1701 
1702 	/* Only PFV 0 is defined currently */
1703 	if (ic_req->pfv != 0) {
1704 		SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv);
1705 		fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
1706 		error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv);
1707 		goto end;
1708 	}
1709 
1710 	/* MAXR2T is 0's based */
1711 	tqpair->maxr2t = ic_req->maxr2t + 1ull;
1712 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "maxr2t =%u\n", tqpair->maxr2t);
1713 
1714 	tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false;
1715 	tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false;
1716 
1717 	tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX);
1718 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda);
1719 
1720 	rsp_pdu = spdk_nvmf_tcp_pdu_get(tqpair);
1721 	if (!rsp_pdu) {
1722 		tqpair->state = NVME_TCP_QPAIR_STATE_EXITING;
1723 		spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR);
1724 		return;
1725 	}
1726 
1727 	ic_resp = &rsp_pdu->hdr.ic_resp;
1728 	ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP;
1729 	ic_resp->common.hlen = ic_resp->common.plen =  sizeof(*ic_resp);
1730 	ic_resp->pfv = 0;
1731 	ic_resp->cpda = tqpair->cpda;
1732 	tqpair->maxh2cdata = spdk_min(NVMF_TCP_PDU_MAX_H2C_DATA_SIZE,
1733 				      ttransport->transport.opts.io_unit_size);
1734 	ic_resp->maxh2cdata = tqpair->maxh2cdata;
1735 	ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0;
1736 	ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0;
1737 
1738 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable);
1739 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable);
1740 
1741 	spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_send_icresp_complete, tqpair);
1742 	spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY);
1743 	return;
1744 end:
1745 	spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
1746 	return;
1747 }
1748 
1749 static void
1750 spdk_nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair)
1751 {
1752 	struct nvme_tcp_pdu *pdu;
1753 	int rc;
1754 	uint32_t crc32c, error_offset = 0;
1755 	enum spdk_nvme_tcp_term_req_fes fes;
1756 	struct spdk_nvmf_tcp_transport *ttransport;
1757 
1758 	assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH);
1759 	pdu = &tqpair->pdu_in_progress;
1760 
1761 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "pdu type of tqpair(%p) is %d\n", tqpair,
1762 		      pdu->hdr.common.pdu_type);
1763 	/* check header digest if needed */
1764 	if (pdu->has_hdgst) {
1765 		SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair);
1766 		crc32c = nvme_tcp_pdu_calc_header_digest(pdu);
1767 		rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, crc32c);
1768 		if (rc == 0) {
1769 			SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu);
1770 			fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR;
1771 			spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
1772 			return;
1773 
1774 		}
1775 	}
1776 
1777 	ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport);
1778 	switch (pdu->hdr.common.pdu_type) {
1779 	case SPDK_NVME_TCP_PDU_TYPE_IC_REQ:
1780 		spdk_nvmf_tcp_icreq_handle(ttransport, tqpair, pdu);
1781 		break;
1782 	case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD:
1783 		spdk_nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu);
1784 		break;
1785 	case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA:
1786 		spdk_nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu);
1787 		break;
1788 
1789 	case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ:
1790 		spdk_nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu);
1791 		break;
1792 
1793 	default:
1794 		SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress.hdr.common.pdu_type);
1795 		fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
1796 		error_offset = 1;
1797 		spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
1798 		break;
1799 	}
1800 }
1801 
1802 static void
1803 spdk_nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair)
1804 {
1805 	struct nvme_tcp_pdu *pdu;
1806 	uint32_t error_offset = 0;
1807 	enum spdk_nvme_tcp_term_req_fes fes;
1808 	uint8_t expected_hlen, pdo;
1809 	bool plen_error = false, pdo_error = false;
1810 
1811 	assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH);
1812 	pdu = &tqpair->pdu_in_progress;
1813 
1814 	if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) {
1815 		if (tqpair->state != NVME_TCP_QPAIR_STATE_INVALID) {
1816 			SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu);
1817 			fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR;
1818 			goto err;
1819 		}
1820 		expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req);
1821 		if (pdu->hdr.common.plen != expected_hlen) {
1822 			plen_error = true;
1823 		}
1824 	} else {
1825 		if (tqpair->state != NVME_TCP_QPAIR_STATE_RUNNING) {
1826 			SPDK_ERRLOG("The TCP/IP connection is not negotitated\n");
1827 			fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR;
1828 			goto err;
1829 		}
1830 
1831 		switch (pdu->hdr.common.pdu_type) {
1832 		case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD:
1833 			expected_hlen = sizeof(struct spdk_nvme_tcp_cmd);
1834 			pdo = pdu->hdr.common.pdo;
1835 			if ((tqpair->cpda != 0) && (pdo != ((tqpair->cpda + 1) << 2))) {
1836 				pdo_error = true;
1837 				break;
1838 			}
1839 
1840 			if (pdu->hdr.common.plen < expected_hlen) {
1841 				plen_error = true;
1842 			}
1843 			break;
1844 		case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA:
1845 			expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr);
1846 			pdo = pdu->hdr.common.pdo;
1847 			if ((tqpair->cpda != 0) && (pdo != ((tqpair->cpda + 1) << 2))) {
1848 				pdo_error = true;
1849 				break;
1850 			}
1851 			if (pdu->hdr.common.plen < expected_hlen) {
1852 				plen_error = true;
1853 			}
1854 			break;
1855 
1856 		case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ:
1857 			expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr);
1858 			if ((pdu->hdr.common.plen <= expected_hlen) ||
1859 			    (pdu->hdr.common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) {
1860 				plen_error = true;
1861 			}
1862 			break;
1863 
1864 		default:
1865 			SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr.common.pdu_type);
1866 			fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
1867 			error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type);
1868 			goto err;
1869 		}
1870 	}
1871 
1872 	if (pdu->hdr.common.hlen != expected_hlen) {
1873 		SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n",
1874 			    pdu->hdr.common.pdu_type,
1875 			    expected_hlen, pdu->hdr.common.hlen, tqpair);
1876 		fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
1877 		error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen);
1878 		goto err;
1879 	} else if (pdo_error) {
1880 		fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
1881 		error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo);
1882 	} else if (plen_error) {
1883 		fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
1884 		error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen);
1885 		goto err;
1886 	} else {
1887 		spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH);
1888 		return;
1889 	}
1890 err:
1891 	spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
1892 }
1893 
1894 static int
1895 spdk_nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair)
1896 {
1897 	int rc = 0;
1898 	struct nvme_tcp_pdu *pdu;
1899 	enum nvme_tcp_pdu_recv_state prev_state;
1900 	uint32_t data_len;
1901 	uint8_t psh_len, pdo, hlen;
1902 	int8_t  padding_len;
1903 
1904 	/* The loop here is to allow for several back-to-back state changes. */
1905 	do {
1906 		prev_state = tqpair->recv_state;
1907 
1908 		SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state);
1909 
1910 		switch (tqpair->recv_state) {
1911 		/* Wait for the common header  */
1912 		case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY:
1913 		case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH:
1914 			pdu = &tqpair->pdu_in_progress;
1915 
1916 			rc = nvme_tcp_read_data(tqpair->sock,
1917 						sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes,
1918 						(void *)&pdu->hdr.common + pdu->ch_valid_bytes);
1919 			if (rc < 0) {
1920 				SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "will disconnect tqpair=%p\n", tqpair);
1921 				return NVME_TCP_PDU_FATAL;
1922 			} else if (rc > 0) {
1923 				pdu->ch_valid_bytes += rc;
1924 				spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH);
1925 			}
1926 
1927 			if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) {
1928 				return NVME_TCP_PDU_IN_PROGRESS;
1929 			}
1930 
1931 			/* The command header of this PDU has now been read from the socket. */
1932 			spdk_nvmf_tcp_pdu_ch_handle(tqpair);
1933 			break;
1934 		/* Wait for the pdu specific header  */
1935 		case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH:
1936 			pdu = &tqpair->pdu_in_progress;
1937 			psh_len = hlen = pdu->hdr.common.hlen;
1938 			/* Only capsule_cmd and h2c_data has header digest */
1939 			if (((pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD) ||
1940 			     (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_H2C_DATA)) &&
1941 			    tqpair->host_hdgst_enable) {
1942 				pdu->has_hdgst = true;
1943 				psh_len += SPDK_NVME_TCP_DIGEST_LEN;
1944 				if (pdu->hdr.common.plen > psh_len) {
1945 					pdo = pdu->hdr.common.pdo;
1946 					padding_len = pdo - psh_len;
1947 					SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "padding length is =%d for pdu=%p on tqpair=%p\n", padding_len,
1948 						      pdu, tqpair);
1949 					if (padding_len > 0) {
1950 						psh_len = pdo;
1951 					}
1952 				}
1953 			}
1954 
1955 			psh_len -= sizeof(struct spdk_nvme_tcp_common_pdu_hdr);
1956 			/* The following will read psh + hdgest (if possbile) + padding (if posssible) */
1957 			if (pdu->psh_valid_bytes < psh_len) {
1958 				rc = nvme_tcp_read_data(tqpair->sock,
1959 							psh_len - pdu->psh_valid_bytes,
1960 							(void *)&pdu->hdr.raw + sizeof(struct spdk_nvme_tcp_common_pdu_hdr) + pdu->psh_valid_bytes);
1961 				if (rc < 0) {
1962 					return NVME_TCP_PDU_FATAL;
1963 				}
1964 
1965 				pdu->psh_valid_bytes += rc;
1966 				if (pdu->psh_valid_bytes < psh_len) {
1967 					return NVME_TCP_PDU_IN_PROGRESS;
1968 				}
1969 			}
1970 
1971 			/* All header(ch, psh, head digist) of this PDU has now been read from the socket. */
1972 			spdk_nvmf_tcp_pdu_psh_handle(tqpair);
1973 			break;
1974 		case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD:
1975 			pdu = &tqpair->pdu_in_progress;
1976 
1977 			/* check whether the data is valid, if not we just return */
1978 			if (!pdu->data) {
1979 				return NVME_TCP_PDU_IN_PROGRESS;
1980 			}
1981 
1982 			data_len = pdu->data_len;
1983 			/* data len */
1984 			if (pdu->data_valid_bytes < data_len) {
1985 				rc = nvme_tcp_read_data(tqpair->sock, data_len - pdu->data_valid_bytes,
1986 							(void *)pdu->data + pdu->data_valid_bytes);
1987 				if (rc < 0) {
1988 					return NVME_TCP_PDU_FATAL;
1989 				}
1990 
1991 				pdu->data_valid_bytes += rc;
1992 				if (pdu->data_valid_bytes < data_len) {
1993 					return NVME_TCP_PDU_IN_PROGRESS;
1994 				}
1995 			}
1996 
1997 			/* data digest */
1998 			if ((pdu->hdr.common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) &&
1999 			    tqpair->host_ddgst_enable && (pdu->ddigest_valid_bytes < SPDK_NVME_TCP_DIGEST_LEN)) {
2000 				rc = nvme_tcp_read_data(tqpair->sock,
2001 							SPDK_NVME_TCP_DIGEST_LEN - pdu->ddigest_valid_bytes,
2002 							pdu->data_digest + pdu->ddigest_valid_bytes);
2003 				if (rc < 0) {
2004 					return NVME_TCP_PDU_FATAL;
2005 				}
2006 
2007 				pdu->ddigest_valid_bytes += rc;
2008 				if (pdu->ddigest_valid_bytes < SPDK_NVME_TCP_DIGEST_LEN) {
2009 					return NVME_TCP_PDU_IN_PROGRESS;
2010 				}
2011 			}
2012 
2013 			/* All of this PDU has now been read from the socket. */
2014 			spdk_nvmf_tcp_pdu_payload_handle(tqpair);
2015 			break;
2016 		case NVME_TCP_PDU_RECV_STATE_ERROR:
2017 			pdu = &tqpair->pdu_in_progress;
2018 			/* Check whether the connection is closed. Each time, we only read 1 byte every time */
2019 			rc = nvme_tcp_read_data(tqpair->sock, 1, (void *)&pdu->hdr.common);
2020 			if (rc < 0) {
2021 				return NVME_TCP_PDU_FATAL;
2022 			}
2023 			break;
2024 		default:
2025 			assert(0);
2026 			SPDK_ERRLOG("code should not come to here");
2027 			break;
2028 		}
2029 	} while (tqpair->recv_state != prev_state);
2030 
2031 	return rc;
2032 }
2033 
2034 static enum spdk_nvme_data_transfer
2035 spdk_nvmf_tcp_req_get_xfer(struct spdk_nvmf_tcp_req *tcp_req) {
2036 	enum spdk_nvme_data_transfer xfer;
2037 	struct spdk_nvme_cmd *cmd = &tcp_req->req.cmd->nvme_cmd;
2038 	struct spdk_nvme_sgl_descriptor *sgl = &cmd->dptr.sgl1;
2039 
2040 	/* Figure out data transfer direction */
2041 	if (cmd->opc == SPDK_NVME_OPC_FABRIC)
2042 	{
2043 		xfer = spdk_nvme_opc_get_data_transfer(tcp_req->req.cmd->nvmf_cmd.fctype);
2044 	} else
2045 	{
2046 		xfer = spdk_nvme_opc_get_data_transfer(cmd->opc);
2047 
2048 		/* Some admin commands are special cases */
2049 		if ((tcp_req->req.qpair->qid == 0) &&
2050 		    ((cmd->opc == SPDK_NVME_OPC_GET_FEATURES) ||
2051 		     (cmd->opc == SPDK_NVME_OPC_SET_FEATURES))) {
2052 			switch (cmd->cdw10 & 0xff) {
2053 			case SPDK_NVME_FEAT_LBA_RANGE_TYPE:
2054 			case SPDK_NVME_FEAT_AUTONOMOUS_POWER_STATE_TRANSITION:
2055 			case SPDK_NVME_FEAT_HOST_IDENTIFIER:
2056 				break;
2057 			default:
2058 				xfer = SPDK_NVME_DATA_NONE;
2059 			}
2060 		}
2061 	}
2062 
2063 	if (xfer == SPDK_NVME_DATA_NONE)
2064 	{
2065 		return xfer;
2066 	}
2067 
2068 	/* Even for commands that may transfer data, they could have specified 0 length.
2069 	 * We want those to show up with xfer SPDK_NVME_DATA_NONE.
2070 	 */
2071 	switch (sgl->generic.type)
2072 	{
2073 	case SPDK_NVME_SGL_TYPE_DATA_BLOCK:
2074 	case SPDK_NVME_SGL_TYPE_BIT_BUCKET:
2075 	case SPDK_NVME_SGL_TYPE_SEGMENT:
2076 	case SPDK_NVME_SGL_TYPE_LAST_SEGMENT:
2077 	case SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK:
2078 		if (sgl->unkeyed.length == 0) {
2079 			xfer = SPDK_NVME_DATA_NONE;
2080 		}
2081 		break;
2082 	case SPDK_NVME_SGL_TYPE_KEYED_DATA_BLOCK:
2083 		if (sgl->keyed.length == 0) {
2084 			xfer = SPDK_NVME_DATA_NONE;
2085 		}
2086 		break;
2087 	}
2088 
2089 	return xfer;
2090 }
2091 
2092 static void
2093 spdk_nvmf_tcp_request_free_buffers(struct spdk_nvmf_tcp_req *tcp_req,
2094 				   struct spdk_nvmf_transport_poll_group *group, struct spdk_nvmf_transport *transport)
2095 {
2096 	for (uint32_t i = 0; i < tcp_req->req.iovcnt; i++) {
2097 		assert(tcp_req->buffers[i] != NULL);
2098 		if (group->buf_cache_count < group->buf_cache_size) {
2099 			STAILQ_INSERT_HEAD(&group->buf_cache,
2100 					   (struct spdk_nvmf_transport_pg_cache_buf *)tcp_req->buffers[i], link);
2101 			group->buf_cache_count++;
2102 		} else {
2103 			spdk_mempool_put(transport->data_buf_pool, tcp_req->buffers[i]);
2104 		}
2105 		tcp_req->req.iov[i].iov_base = NULL;
2106 		tcp_req->buffers[i] = NULL;
2107 		tcp_req->req.iov[i].iov_len = 0;
2108 	}
2109 	tcp_req->data_from_pool = false;
2110 }
2111 
2112 static int
2113 spdk_nvmf_tcp_req_fill_iovs(struct spdk_nvmf_tcp_transport *ttransport,
2114 			    struct spdk_nvmf_tcp_req *tcp_req)
2115 {
2116 	void					*buf = NULL;
2117 	uint32_t				length = tcp_req->req.length;
2118 	uint32_t				i = 0;
2119 	struct spdk_nvmf_tcp_qpair		*tqpair;
2120 	struct spdk_nvmf_transport_poll_group	*group;
2121 
2122 	tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair);
2123 	group = &tqpair->group->group;
2124 
2125 	tcp_req->req.iovcnt = 0;
2126 	while (length) {
2127 		if (!(STAILQ_EMPTY(&group->buf_cache))) {
2128 			group->buf_cache_count--;
2129 			buf = STAILQ_FIRST(&group->buf_cache);
2130 			STAILQ_REMOVE_HEAD(&group->buf_cache, link);
2131 		} else {
2132 			buf = spdk_mempool_get(ttransport->transport.data_buf_pool);
2133 			if (!buf) {
2134 				goto nomem;
2135 			}
2136 		}
2137 
2138 		tcp_req->req.iov[i].iov_base = (void *)((uintptr_t)(buf + NVMF_DATA_BUFFER_MASK) &
2139 							~NVMF_DATA_BUFFER_MASK);
2140 		tcp_req->req.iov[i].iov_len  = spdk_min(length, ttransport->transport.opts.io_unit_size);
2141 		tcp_req->req.iovcnt++;
2142 		tcp_req->buffers[i] = buf;
2143 		length -= tcp_req->req.iov[i].iov_len;
2144 		i++;
2145 	}
2146 
2147 	assert(tcp_req->req.iovcnt < SPDK_NVMF_MAX_SGL_ENTRIES);
2148 	tcp_req->data_from_pool = true;
2149 	return 0;
2150 
2151 nomem:
2152 	spdk_nvmf_tcp_request_free_buffers(tcp_req, group, &ttransport->transport);
2153 	tcp_req->req.iovcnt = 0;
2154 	return -ENOMEM;
2155 }
2156 
2157 static int
2158 spdk_nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_transport *ttransport,
2159 			    struct spdk_nvmf_tcp_req *tcp_req)
2160 {
2161 	struct spdk_nvme_cmd			*cmd;
2162 	struct spdk_nvme_cpl			*rsp;
2163 	struct spdk_nvme_sgl_descriptor		*sgl;
2164 
2165 	cmd = &tcp_req->req.cmd->nvme_cmd;
2166 	rsp = &tcp_req->req.rsp->nvme_cpl;
2167 	sgl = &cmd->dptr.sgl1;
2168 
2169 	if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK &&
2170 	    sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) {
2171 		if (sgl->unkeyed.length > ttransport->transport.opts.max_io_size) {
2172 			SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n",
2173 				    sgl->unkeyed.length, ttransport->transport.opts.max_io_size);
2174 			rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID;
2175 			return -1;
2176 		}
2177 
2178 		/* fill request length and populate iovs */
2179 		tcp_req->req.length = sgl->unkeyed.length;
2180 
2181 		SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Data requested length= 0x%x\n",
2182 			      sgl->unkeyed.length);
2183 
2184 		if (spdk_nvmf_tcp_req_fill_iovs(ttransport, tcp_req) < 0) {
2185 			/* No available buffers. Queue this request up. */
2186 			SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "No available large data buffers. Queueing request %p\n", tcp_req);
2187 			return 0;
2188 		}
2189 
2190 		/* backward compatible */
2191 		tcp_req->req.data = tcp_req->req.iov[0].iov_base;
2192 
2193 
2194 		SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Request %p took %d buffer/s from central pool, and data=%p\n",
2195 			      tcp_req,
2196 			      tcp_req->req.iovcnt, tcp_req->req.data);
2197 
2198 		return 0;
2199 	} else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK &&
2200 		   sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) {
2201 		uint64_t offset = sgl->address;
2202 		uint32_t max_len = ttransport->transport.opts.in_capsule_data_size;
2203 
2204 		SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n",
2205 			      offset, sgl->unkeyed.length);
2206 
2207 		if (offset > max_len) {
2208 			SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " exceeds capsule length 0x%x\n",
2209 				    offset, max_len);
2210 			rsp->status.sc = SPDK_NVME_SC_INVALID_SGL_OFFSET;
2211 			return -1;
2212 		}
2213 		max_len -= (uint32_t)offset;
2214 
2215 		if (sgl->unkeyed.length > max_len) {
2216 			SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n",
2217 				    sgl->unkeyed.length, max_len);
2218 			rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID;
2219 			return -1;
2220 		}
2221 
2222 		tcp_req->req.data = tcp_req->buf + offset;
2223 		tcp_req->data_from_pool = false;
2224 		tcp_req->req.length = sgl->unkeyed.length;
2225 
2226 		tcp_req->req.iov[0].iov_base = tcp_req->req.data;
2227 		tcp_req->req.iov[0].iov_len = tcp_req->req.length;
2228 		tcp_req->req.iovcnt = 1;
2229 
2230 		return 0;
2231 	}
2232 
2233 	SPDK_ERRLOG("Invalid NVMf I/O Command SGL:  Type 0x%x, Subtype 0x%x\n",
2234 		    sgl->generic.type, sgl->generic.subtype);
2235 	rsp->status.sc = SPDK_NVME_SC_SGL_DESCRIPTOR_TYPE_INVALID;
2236 	return -1;
2237 }
2238 
2239 static void
2240 spdk_nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair,
2241 			    struct spdk_nvmf_tcp_req *tcp_req)
2242 {
2243 	struct nvme_tcp_pdu *rsp_pdu;
2244 	struct spdk_nvme_tcp_c2h_data_hdr *c2h_data;
2245 	uint32_t plen, pdo, alignment, offset, iov_index;
2246 
2247 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n");
2248 
2249 	/* always use the first iov_len, which is correct */
2250 	iov_index = tcp_req->c2h_data_offset / tcp_req->req.iov[0].iov_len;
2251 	offset = tcp_req->c2h_data_offset % tcp_req->req.iov[0].iov_len;
2252 
2253 	rsp_pdu = spdk_nvmf_tcp_pdu_get(tqpair);
2254 	assert(rsp_pdu != NULL);
2255 
2256 	c2h_data = &rsp_pdu->hdr.c2h_data;
2257 	c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA;
2258 	plen = c2h_data->common.hlen = sizeof(*c2h_data);
2259 
2260 	if (tqpair->host_hdgst_enable) {
2261 		plen += SPDK_NVME_TCP_DIGEST_LEN;
2262 		c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF;
2263 	}
2264 
2265 	/* set the psh */
2266 	c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid;
2267 	c2h_data->datal = spdk_min(NVMF_TCP_PDU_MAX_C2H_DATA_SIZE,
2268 				   (tcp_req->req.iov[iov_index].iov_len - offset));
2269 	c2h_data->datao = tcp_req->c2h_data_offset;
2270 
2271 	/* set the padding */
2272 	rsp_pdu->padding_len = 0;
2273 	pdo = plen;
2274 	if (tqpair->cpda) {
2275 		alignment = (tqpair->cpda + 1) << 2;
2276 		if (alignment > plen) {
2277 			rsp_pdu->padding_len = alignment - plen;
2278 			pdo = plen = alignment;
2279 		}
2280 	}
2281 
2282 	c2h_data->common.pdo = pdo;
2283 	plen += c2h_data->datal;
2284 	if (tqpair->host_ddgst_enable) {
2285 		c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF;
2286 		plen += SPDK_NVME_TCP_DIGEST_LEN;
2287 	}
2288 
2289 	c2h_data->common.plen = plen;
2290 
2291 	rsp_pdu->data = tcp_req->req.iov[iov_index].iov_base + offset;
2292 	rsp_pdu->data_len = c2h_data->datal;
2293 
2294 	tcp_req->c2h_data_offset += c2h_data->datal;
2295 	if (iov_index == (tcp_req->req.iovcnt - 1) && (tcp_req->c2h_data_offset == tcp_req->req.length)) {
2296 		SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Last pdu for tcp_req=%p on tqpair=%p\n", tcp_req, tqpair);
2297 		c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU;
2298 		/* The linux kernel does not support this yet */
2299 #if LINUX_KERNEL_SUPPORT_NOT_SENDING_RESP_FOR_C2H
2300 		c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS;
2301 #endif
2302 		TAILQ_REMOVE(&tqpair->queued_c2h_data_tcp_req, tcp_req, link);
2303 	}
2304 
2305 	tqpair->c2h_data_pdu_cnt += 1;
2306 	spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_pdu_c2h_data_complete, tcp_req);
2307 }
2308 
2309 static int
2310 spdk_nvmf_tcp_calc_c2h_data_pdu_num(struct spdk_nvmf_tcp_req *tcp_req)
2311 {
2312 	uint32_t i, iov_cnt, pdu_num = 0;
2313 
2314 	iov_cnt = tcp_req->req.iovcnt;
2315 	for (i = 0; i < iov_cnt; i++) {
2316 		pdu_num += (tcp_req->req.iov[i].iov_len + NVMF_TCP_PDU_MAX_C2H_DATA_SIZE - 1) /
2317 			   NVMF_TCP_PDU_MAX_C2H_DATA_SIZE;
2318 	}
2319 
2320 	return pdu_num;
2321 }
2322 
2323 static void
2324 spdk_nvmf_tcp_handle_pending_c2h_data_queue(struct spdk_nvmf_tcp_qpair *tqpair)
2325 {
2326 	struct spdk_nvmf_tcp_req *tcp_req;
2327 
2328 	while (!TAILQ_EMPTY(&tqpair->queued_c2h_data_tcp_req) &&
2329 	       (tqpair->c2h_data_pdu_cnt < NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM)) {
2330 		tcp_req = TAILQ_FIRST(&tqpair->queued_c2h_data_tcp_req);
2331 		spdk_nvmf_tcp_send_c2h_data(tqpair, tcp_req);
2332 	}
2333 }
2334 
2335 static void
2336 spdk_nvmf_tcp_queue_c2h_data(struct spdk_nvmf_tcp_req *tcp_req,
2337 			     struct spdk_nvmf_tcp_qpair *tqpair)
2338 {
2339 	tcp_req->c2h_data_pdu_num = spdk_nvmf_tcp_calc_c2h_data_pdu_num(tcp_req);
2340 
2341 	assert(tcp_req->c2h_data_pdu_num < NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM);
2342 
2343 	TAILQ_INSERT_TAIL(&tqpair->queued_c2h_data_tcp_req, tcp_req, link);
2344 	spdk_nvmf_tcp_handle_pending_c2h_data_queue(tqpair);
2345 }
2346 
2347 static int
2348 request_transfer_out(struct spdk_nvmf_request *req)
2349 {
2350 	struct spdk_nvmf_tcp_req	*tcp_req;
2351 	struct spdk_nvmf_qpair		*qpair;
2352 	struct spdk_nvmf_tcp_qpair	*tqpair;
2353 	struct spdk_nvme_cpl		*rsp;
2354 
2355 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n");
2356 
2357 	qpair = req->qpair;
2358 	rsp = &req->rsp->nvme_cpl;
2359 	tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req);
2360 
2361 	/* Advance our sq_head pointer */
2362 	if (qpair->sq_head == qpair->sq_head_max) {
2363 		qpair->sq_head = 0;
2364 	} else {
2365 		qpair->sq_head++;
2366 	}
2367 	rsp->sqhd = qpair->sq_head;
2368 
2369 	tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair);
2370 	spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST);
2371 	if (rsp->status.sc == SPDK_NVME_SC_SUCCESS &&
2372 	    req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) {
2373 		spdk_nvmf_tcp_queue_c2h_data(tcp_req, tqpair);
2374 	} else {
2375 		spdk_nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair);
2376 	}
2377 
2378 	return 0;
2379 }
2380 
2381 static void
2382 spdk_nvmf_tcp_pdu_set_buf_from_req(struct spdk_nvmf_tcp_qpair *tqpair,
2383 				   struct spdk_nvmf_tcp_req *tcp_req)
2384 {
2385 	struct nvme_tcp_pdu *pdu;
2386 
2387 	if (tcp_req->data_from_pool) {
2388 		SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Will send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair);
2389 		tcp_req->next_expected_r2t_offset = 0;
2390 		spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_DATA_PENDING_FOR_R2T);
2391 		spdk_nvmf_tcp_handle_queued_r2t_req(tqpair);
2392 	} else {
2393 		pdu = &tqpair->pdu_in_progress;
2394 		SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req,
2395 			      tqpair);
2396 		/* No need to send r2t, contained in the capsuled data */
2397 		pdu->data = tcp_req->req.data;
2398 		pdu->data_len = tcp_req->req.length;
2399 		spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD);
2400 		spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER);
2401 	}
2402 }
2403 
2404 static void
2405 spdk_nvmf_tcp_set_incapsule_data(struct spdk_nvmf_tcp_qpair *tqpair,
2406 				 struct spdk_nvmf_tcp_req *tcp_req)
2407 {
2408 	struct nvme_tcp_pdu *pdu;
2409 	uint32_t plen = 0;
2410 
2411 	pdu = &tqpair->pdu_in_progress;
2412 	plen = pdu->hdr.common.hlen;
2413 
2414 	if (tqpair->host_hdgst_enable) {
2415 		plen += SPDK_NVME_TCP_DIGEST_LEN;
2416 	}
2417 
2418 	if (pdu->hdr.common.plen != plen) {
2419 		tcp_req->has_incapsule_data = true;
2420 	}
2421 }
2422 
2423 static bool
2424 spdk_nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport,
2425 			  struct spdk_nvmf_tcp_req *tcp_req)
2426 {
2427 	struct spdk_nvmf_tcp_qpair		*tqpair;
2428 	struct spdk_nvme_cpl			*rsp = &tcp_req->req.rsp->nvme_cpl;
2429 	int					rc;
2430 	enum spdk_nvmf_tcp_req_state		prev_state;
2431 	bool					progress = false;
2432 	struct spdk_nvmf_transport_poll_group	*group;
2433 
2434 	tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair);
2435 	group = &tqpair->group->group;
2436 	assert(tcp_req->state != TCP_REQUEST_STATE_FREE);
2437 
2438 	/* The loop here is to allow for several back-to-back state changes. */
2439 	do {
2440 		prev_state = tcp_req->state;
2441 
2442 		SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state,
2443 			      tqpair);
2444 
2445 		switch (tcp_req->state) {
2446 		case TCP_REQUEST_STATE_FREE:
2447 			/* Some external code must kick a request into TCP_REQUEST_STATE_NEW
2448 			 * to escape this state. */
2449 			break;
2450 		case TCP_REQUEST_STATE_NEW:
2451 			spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, 0, 0, (uintptr_t)tcp_req, 0);
2452 
2453 			/* copy the cmd from the receive pdu */
2454 			tcp_req->cmd = tqpair->pdu_in_progress.hdr.capsule_cmd.ccsqe;
2455 
2456 			/* The next state transition depends on the data transfer needs of this request. */
2457 			tcp_req->req.xfer = spdk_nvmf_tcp_req_get_xfer(tcp_req);
2458 
2459 			/* If no data to transfer, ready to execute. */
2460 			if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) {
2461 				/* Reset the tqpair receving pdu state */
2462 				spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY);
2463 				spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE);
2464 				break;
2465 			}
2466 
2467 			spdk_nvmf_tcp_set_incapsule_data(tqpair, tcp_req);
2468 
2469 			if (!tcp_req->has_incapsule_data) {
2470 				spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY);
2471 			}
2472 
2473 			spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER);
2474 			TAILQ_INSERT_TAIL(&tqpair->group->pending_data_buf_queue, tcp_req, link);
2475 			break;
2476 		case TCP_REQUEST_STATE_NEED_BUFFER:
2477 			spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 0, 0, (uintptr_t)tcp_req, 0);
2478 
2479 			assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE);
2480 
2481 			if (!tcp_req->has_incapsule_data &&
2482 			    (tcp_req != TAILQ_FIRST(&tqpair->group->pending_data_buf_queue))) {
2483 				SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP,
2484 					      "Not the first element to wait for the buf for tcp_req(%p) on tqpair=%p\n",
2485 					      tcp_req, tqpair);
2486 				/* This request needs to wait in line to obtain a buffer */
2487 				break;
2488 			}
2489 
2490 			/* Try to get a data buffer */
2491 			rc = spdk_nvmf_tcp_req_parse_sgl(ttransport, tcp_req);
2492 			if (rc < 0) {
2493 				TAILQ_REMOVE(&tqpair->group->pending_data_buf_queue, tcp_req, link);
2494 				rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
2495 				/* Reset the tqpair receving pdu state */
2496 				spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR);
2497 				spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE);
2498 				break;
2499 			}
2500 
2501 			if (!tcp_req->req.data) {
2502 				SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "No buffer allocated for tcp_req(%p) on tqpair(%p\n)",
2503 					      tcp_req, tqpair);
2504 				/* No buffers available. */
2505 				break;
2506 			}
2507 
2508 			TAILQ_REMOVE(&tqpair->group->pending_data_buf_queue, tcp_req, link);
2509 
2510 			/* If data is transferring from host to controller, we need to do a transfer from the host. */
2511 			if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) {
2512 				spdk_nvmf_tcp_pdu_set_buf_from_req(tqpair, tcp_req);
2513 				break;
2514 			}
2515 
2516 			spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE);
2517 			break;
2518 		case TCP_REQUEST_STATE_DATA_PENDING_FOR_R2T:
2519 			spdk_trace_record(TCP_REQUEST_STATE_DATA_PENDING_FOR_R2T, 0, 0,
2520 					  (uintptr_t)tcp_req, 0);
2521 			/* Some external code must kick a request into TCP_REQUEST_STATE_DATA_PENDING_R2T
2522 			 * to escape this state. */
2523 			break;
2524 
2525 		case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER:
2526 			spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 0, 0,
2527 					  (uintptr_t)tcp_req, 0);
2528 			/* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE
2529 			 * to escape this state. */
2530 			break;
2531 		case TCP_REQUEST_STATE_READY_TO_EXECUTE:
2532 			spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 0, 0, (uintptr_t)tcp_req, 0);
2533 			spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING);
2534 			spdk_nvmf_request_exec(&tcp_req->req);
2535 			break;
2536 		case TCP_REQUEST_STATE_EXECUTING:
2537 			spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, 0, 0, (uintptr_t)tcp_req, 0);
2538 			/* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED
2539 			 * to escape this state. */
2540 			break;
2541 		case TCP_REQUEST_STATE_EXECUTED:
2542 			spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, 0, 0, (uintptr_t)tcp_req, 0);
2543 			spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE);
2544 			break;
2545 		case TCP_REQUEST_STATE_READY_TO_COMPLETE:
2546 			spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 0, 0, (uintptr_t)tcp_req, 0);
2547 			rc = request_transfer_out(&tcp_req->req);
2548 			assert(rc == 0); /* No good way to handle this currently */
2549 			break;
2550 		case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST:
2551 			spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 0, 0,
2552 					  (uintptr_t)tcp_req,
2553 					  0);
2554 			/* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED
2555 			 * to escape this state. */
2556 			break;
2557 		case TCP_REQUEST_STATE_COMPLETED:
2558 			spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, 0, 0, (uintptr_t)tcp_req, 0);
2559 			if (tcp_req->data_from_pool) {
2560 				spdk_nvmf_tcp_request_free_buffers(tcp_req, group, &ttransport->transport);
2561 			}
2562 			tcp_req->req.length = 0;
2563 			tcp_req->req.iovcnt = 0;
2564 			tcp_req->req.data = NULL;
2565 			spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_FREE);
2566 			break;
2567 		case TCP_REQUEST_NUM_STATES:
2568 		default:
2569 			assert(0);
2570 			break;
2571 		}
2572 
2573 		if (tcp_req->state != prev_state) {
2574 			progress = true;
2575 		}
2576 	} while (tcp_req->state != prev_state);
2577 
2578 	return progress;
2579 }
2580 
2581 static void
2582 spdk_nvmf_tcp_qpair_process_pending(struct spdk_nvmf_tcp_transport *ttransport,
2583 				    struct spdk_nvmf_tcp_qpair *tqpair)
2584 {
2585 	struct spdk_nvmf_tcp_req *tcp_req, *req_tmp;
2586 
2587 	/* Tqpair is not in a good state, so return it */
2588 	if (spdk_unlikely(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR)) {
2589 		return;
2590 	}
2591 
2592 	spdk_nvmf_tcp_handle_queued_r2t_req(tqpair);
2593 
2594 	TAILQ_FOREACH_SAFE(tcp_req, &tqpair->group->pending_data_buf_queue, link, req_tmp) {
2595 		if (spdk_nvmf_tcp_req_process(ttransport, tcp_req) == false) {
2596 			break;
2597 		}
2598 	}
2599 }
2600 
2601 static void
2602 spdk_nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock)
2603 {
2604 	struct spdk_nvmf_tcp_qpair *tqpair = arg;
2605 	struct spdk_nvmf_tcp_transport *ttransport;
2606 	int rc;
2607 
2608 	assert(tqpair != NULL);
2609 
2610 	ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport);
2611 	spdk_nvmf_tcp_qpair_process_pending(ttransport, tqpair);
2612 	rc = spdk_nvmf_tcp_sock_process(tqpair);
2613 
2614 	/* check the following two factors:
2615 	 * rc: The socket is closed
2616 	 * State of tqpair: The tqpair is in EXITING state due to internal error
2617 	 */
2618 	if ((rc < 0) || (tqpair->state == NVME_TCP_QPAIR_STATE_EXITING)) {
2619 		tqpair->state = NVME_TCP_QPAIR_STATE_EXITED;
2620 		spdk_nvmf_tcp_qpair_flush_pdus(tqpair);
2621 		SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "will disconect the tqpair=%p\n", tqpair);
2622 		spdk_poller_unregister(&tqpair->timeout_poller);
2623 		spdk_nvmf_qpair_disconnect(&tqpair->qpair, NULL, NULL);
2624 	}
2625 }
2626 
2627 static int
2628 spdk_nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group,
2629 			     struct spdk_nvmf_qpair *qpair)
2630 {
2631 	struct spdk_nvmf_tcp_poll_group	*tgroup;
2632 	struct spdk_nvmf_tcp_qpair	*tqpair;
2633 	int				rc;
2634 
2635 	tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group);
2636 	tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
2637 
2638 	rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock,
2639 				      spdk_nvmf_tcp_sock_cb, tqpair);
2640 	if (rc != 0) {
2641 		SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n",
2642 			    spdk_strerror(errno), errno);
2643 		spdk_nvmf_tcp_qpair_destroy(tqpair);
2644 		return -1;
2645 	}
2646 
2647 	rc =  spdk_nvmf_tcp_qpair_sock_init(tqpair);
2648 	if (rc != 0) {
2649 		SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair);
2650 		spdk_nvmf_tcp_qpair_destroy(tqpair);
2651 		return -1;
2652 	}
2653 
2654 	rc = spdk_nvmf_tcp_qpair_init(&tqpair->qpair);
2655 	if (rc < 0) {
2656 		SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair);
2657 		spdk_nvmf_tcp_qpair_destroy(tqpair);
2658 		return -1;
2659 	}
2660 
2661 	rc = spdk_nvmf_tcp_qpair_init_mem_resource(tqpair, 1);
2662 	if (rc < 0) {
2663 		SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair);
2664 		spdk_nvmf_tcp_qpair_destroy(tqpair);
2665 		return -1;
2666 	}
2667 
2668 	tqpair->group = tgroup;
2669 	tqpair->state = NVME_TCP_QPAIR_STATE_INVALID;
2670 	TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link);
2671 
2672 	return 0;
2673 }
2674 
2675 static int
2676 spdk_nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group,
2677 				struct spdk_nvmf_qpair *qpair)
2678 {
2679 	struct spdk_nvmf_tcp_poll_group	*tgroup;
2680 	struct spdk_nvmf_tcp_qpair		*tqpair;
2681 	int				rc;
2682 
2683 	tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group);
2684 	tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
2685 
2686 	assert(tqpair->group == tgroup);
2687 	tqpair->group = NULL;
2688 
2689 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup);
2690 	TAILQ_REMOVE(&tgroup->qpairs, tqpair, link);
2691 	rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock);
2692 	if (rc != 0) {
2693 		SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n",
2694 			    spdk_strerror(errno), errno);
2695 	}
2696 
2697 	return rc;
2698 }
2699 
2700 static int
2701 spdk_nvmf_tcp_req_complete(struct spdk_nvmf_request *req)
2702 {
2703 	struct spdk_nvmf_tcp_transport *ttransport;
2704 	struct spdk_nvmf_tcp_req *tcp_req;
2705 
2706 	ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport);
2707 	tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req);
2708 
2709 	spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED);
2710 	spdk_nvmf_tcp_req_process(ttransport, tcp_req);
2711 
2712 	return 0;
2713 }
2714 
2715 static void
2716 spdk_nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair)
2717 {
2718 	SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n");
2719 
2720 	spdk_nvmf_tcp_qpair_destroy(SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair));
2721 }
2722 
2723 static int
2724 spdk_nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group)
2725 {
2726 	struct spdk_nvmf_tcp_poll_group *tgroup;
2727 	int rc;
2728 
2729 	tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group);
2730 
2731 	if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs))) {
2732 		return 0;
2733 	}
2734 
2735 	rc = spdk_sock_group_poll(tgroup->sock_group);
2736 	if (rc < 0) {
2737 		SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group);
2738 		return rc;
2739 	}
2740 
2741 	return 0;
2742 }
2743 
2744 static bool
2745 spdk_nvmf_tcp_qpair_is_idle(struct spdk_nvmf_qpair *qpair)
2746 {
2747 	struct spdk_nvmf_tcp_qpair *tqpair;
2748 
2749 	tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
2750 	if (tqpair->state_cntr[TCP_REQUEST_STATE_FREE] == tqpair->max_queue_depth) {
2751 		return true;
2752 	}
2753 
2754 	return false;
2755 }
2756 
2757 static int
2758 spdk_nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair,
2759 			     struct spdk_nvme_transport_id *trid, bool peer)
2760 {
2761 	struct spdk_nvmf_tcp_qpair     *tqpair;
2762 	uint16_t			port;
2763 
2764 	tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
2765 	trid->trtype = SPDK_NVME_TRANSPORT_TCP;
2766 
2767 	if (peer) {
2768 		snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr);
2769 		port = tqpair->initiator_port;
2770 	} else {
2771 		snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr);
2772 		port = tqpair->target_port;
2773 	}
2774 
2775 	if (spdk_sock_is_ipv4(tqpair->sock)) {
2776 		trid->adrfam = SPDK_NVMF_ADRFAM_IPV4;
2777 	} else if (spdk_sock_is_ipv4(tqpair->sock)) {
2778 		trid->adrfam = SPDK_NVMF_ADRFAM_IPV6;
2779 	} else {
2780 		return -1;
2781 	}
2782 
2783 	snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port);
2784 	return 0;
2785 }
2786 
2787 static int
2788 spdk_nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair,
2789 				   struct spdk_nvme_transport_id *trid)
2790 {
2791 	return spdk_nvmf_tcp_qpair_get_trid(qpair, trid, 0);
2792 }
2793 
2794 static int
2795 spdk_nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair,
2796 				  struct spdk_nvme_transport_id *trid)
2797 {
2798 	return spdk_nvmf_tcp_qpair_get_trid(qpair, trid, 1);
2799 }
2800 
2801 static int
2802 spdk_nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair,
2803 				    struct spdk_nvme_transport_id *trid)
2804 {
2805 	return spdk_nvmf_tcp_qpair_get_trid(qpair, trid, 0);
2806 }
2807 
2808 static int
2809 spdk_nvmf_tcp_qpair_set_sq_size(struct spdk_nvmf_qpair *qpair)
2810 {
2811 	struct spdk_nvmf_tcp_qpair     *tqpair;
2812 	int rc;
2813 	tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
2814 
2815 	rc = spdk_nvmf_tcp_qpair_init_mem_resource(tqpair, tqpair->qpair.sq_head_max);
2816 	if (!rc) {
2817 		tqpair->max_queue_depth += tqpair->qpair.sq_head_max;
2818 		tqpair->free_pdu_num += tqpair->qpair.sq_head_max;
2819 		tqpair->state_cntr[TCP_REQUEST_STATE_FREE] += tqpair->qpair.sq_head_max;
2820 		SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "The queue depth=%u for tqpair=%p\n",
2821 			      tqpair->max_queue_depth, tqpair);
2822 	}
2823 
2824 	return rc;
2825 
2826 }
2827 
2828 #define SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH 128
2829 #define SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH 128
2830 #define SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR 64
2831 #define SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE 4096
2832 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE 131072
2833 #define SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE 131072
2834 #define SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS 512
2835 #define SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE 32
2836 
2837 static void
2838 spdk_nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts)
2839 {
2840 	opts->max_queue_depth =		SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH;
2841 	opts->max_qpairs_per_ctrlr =	SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR;
2842 	opts->in_capsule_data_size =	SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE;
2843 	opts->max_io_size =		SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE;
2844 	opts->io_unit_size =		SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE;
2845 	opts->max_aq_depth =		SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH;
2846 	opts->num_shared_buffers =	SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS;
2847 	opts->buf_cache_size =		SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE;
2848 }
2849 
2850 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = {
2851 	.type = SPDK_NVME_TRANSPORT_TCP,
2852 	.opts_init = spdk_nvmf_tcp_opts_init,
2853 	.create = spdk_nvmf_tcp_create,
2854 	.destroy = spdk_nvmf_tcp_destroy,
2855 
2856 	.listen = spdk_nvmf_tcp_listen,
2857 	.stop_listen = spdk_nvmf_tcp_stop_listen,
2858 	.accept = spdk_nvmf_tcp_accept,
2859 
2860 	.listener_discover = spdk_nvmf_tcp_discover,
2861 
2862 	.poll_group_create = spdk_nvmf_tcp_poll_group_create,
2863 	.poll_group_destroy = spdk_nvmf_tcp_poll_group_destroy,
2864 	.poll_group_add = spdk_nvmf_tcp_poll_group_add,
2865 	.poll_group_remove = spdk_nvmf_tcp_poll_group_remove,
2866 	.poll_group_poll = spdk_nvmf_tcp_poll_group_poll,
2867 
2868 	.req_free = spdk_nvmf_tcp_req_free,
2869 	.req_complete = spdk_nvmf_tcp_req_complete,
2870 
2871 	.qpair_fini = spdk_nvmf_tcp_close_qpair,
2872 	.qpair_is_idle = spdk_nvmf_tcp_qpair_is_idle,
2873 	.qpair_get_local_trid = spdk_nvmf_tcp_qpair_get_local_trid,
2874 	.qpair_get_peer_trid = spdk_nvmf_tcp_qpair_get_peer_trid,
2875 	.qpair_get_listen_trid = spdk_nvmf_tcp_qpair_get_listen_trid,
2876 	.qpair_set_sqsize = spdk_nvmf_tcp_qpair_set_sq_size,
2877 };
2878 
2879 SPDK_LOG_REGISTER_COMPONENT("nvmf_tcp", SPDK_LOG_NVMF_TCP)
2880