xref: /spdk/lib/nvmf/tcp.c (revision ceea3088870a3919d6bdfe61d7adba11b9733fb7)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright (c) Intel Corporation. All rights reserved.
5  *   Copyright (c) 2019, 2020 Mellanox Technologies LTD. All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Intel Corporation nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include "spdk/stdinc.h"
35 #include "spdk/crc32.h"
36 #include "spdk/endian.h"
37 #include "spdk/assert.h"
38 #include "spdk/thread.h"
39 #include "spdk/nvmf_transport.h"
40 #include "spdk/sock.h"
41 #include "spdk/string.h"
42 #include "spdk/trace.h"
43 #include "spdk/util.h"
44 
45 #include "spdk_internal/assert.h"
46 #include "spdk/log.h"
47 #include "spdk_internal/nvme_tcp.h"
48 
49 #include "nvmf_internal.h"
50 
51 #define NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME 16
52 #define SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY 16
53 #define SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY 0
54 #define SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION true
55 
56 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp;
57 
58 /* spdk nvmf related structure */
59 enum spdk_nvmf_tcp_req_state {
60 
61 	/* The request is not currently in use */
62 	TCP_REQUEST_STATE_FREE = 0,
63 
64 	/* Initial state when request first received */
65 	TCP_REQUEST_STATE_NEW,
66 
67 	/* The request is queued until a data buffer is available. */
68 	TCP_REQUEST_STATE_NEED_BUFFER,
69 
70 	/* The request is currently transferring data from the host to the controller. */
71 	TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER,
72 
73 	/* The request is waiting for the R2T send acknowledgement. */
74 	TCP_REQUEST_STATE_AWAITING_R2T_ACK,
75 
76 	/* The request is ready to execute at the block device */
77 	TCP_REQUEST_STATE_READY_TO_EXECUTE,
78 
79 	/* The request is currently executing at the block device */
80 	TCP_REQUEST_STATE_EXECUTING,
81 
82 	/* The request finished executing at the block device */
83 	TCP_REQUEST_STATE_EXECUTED,
84 
85 	/* The request is ready to send a completion */
86 	TCP_REQUEST_STATE_READY_TO_COMPLETE,
87 
88 	/* The request is currently transferring final pdus from the controller to the host. */
89 	TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST,
90 
91 	/* The request completed and can be marked free. */
92 	TCP_REQUEST_STATE_COMPLETED,
93 
94 	/* Terminator */
95 	TCP_REQUEST_NUM_STATES,
96 };
97 
98 static const char *spdk_nvmf_tcp_term_req_fes_str[] = {
99 	"Invalid PDU Header Field",
100 	"PDU Sequence Error",
101 	"Header Digiest Error",
102 	"Data Transfer Out of Range",
103 	"R2T Limit Exceeded",
104 	"Unsupported parameter",
105 };
106 
107 #define OBJECT_NVMF_TCP_IO				0x80
108 
109 #define TRACE_GROUP_NVMF_TCP				0x5
110 #define TRACE_TCP_REQUEST_STATE_NEW					SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x0)
111 #define TRACE_TCP_REQUEST_STATE_NEED_BUFFER				SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x1)
112 #define TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER		SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x2)
113 #define TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE			SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x3)
114 #define TRACE_TCP_REQUEST_STATE_EXECUTING				SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x4)
115 #define TRACE_TCP_REQUEST_STATE_EXECUTED				SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x5)
116 #define TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE			SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x6)
117 #define TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST		SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x7)
118 #define TRACE_TCP_REQUEST_STATE_COMPLETED				SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x8)
119 #define TRACE_TCP_FLUSH_WRITEBUF_START					SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x9)
120 #define TRACE_TCP_FLUSH_WRITEBUF_DONE					SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xA)
121 #define TRACE_TCP_READ_FROM_SOCKET_DONE					SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xB)
122 #define TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK				SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xC)
123 
124 SPDK_TRACE_REGISTER_FN(nvmf_tcp_trace, "nvmf_tcp", TRACE_GROUP_NVMF_TCP)
125 {
126 	spdk_trace_register_object(OBJECT_NVMF_TCP_IO, 'r');
127 	spdk_trace_register_description("TCP_REQ_NEW",
128 					TRACE_TCP_REQUEST_STATE_NEW,
129 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 1, 1, "");
130 	spdk_trace_register_description("TCP_REQ_NEED_BUFFER",
131 					TRACE_TCP_REQUEST_STATE_NEED_BUFFER,
132 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, "");
133 	spdk_trace_register_description("TCP_REQ_TX_H_TO_C",
134 					TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER,
135 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, "");
136 	spdk_trace_register_description("TCP_REQ_RDY_TO_EXECUTE",
137 					TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE,
138 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, "");
139 	spdk_trace_register_description("TCP_REQ_EXECUTING",
140 					TRACE_TCP_REQUEST_STATE_EXECUTING,
141 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, "");
142 	spdk_trace_register_description("TCP_REQ_EXECUTED",
143 					TRACE_TCP_REQUEST_STATE_EXECUTED,
144 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, "");
145 	spdk_trace_register_description("TCP_REQ_RDY_TO_COMPLETE",
146 					TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE,
147 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, "");
148 	spdk_trace_register_description("TCP_REQ_TRANSFER_C2H",
149 					TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST,
150 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, "");
151 	spdk_trace_register_description("TCP_REQ_COMPLETED",
152 					TRACE_TCP_REQUEST_STATE_COMPLETED,
153 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, "");
154 	spdk_trace_register_description("TCP_WRITE_START",
155 					TRACE_TCP_FLUSH_WRITEBUF_START,
156 					OWNER_NONE, OBJECT_NONE, 0, 0, "");
157 	spdk_trace_register_description("TCP_WRITE_DONE",
158 					TRACE_TCP_FLUSH_WRITEBUF_DONE,
159 					OWNER_NONE, OBJECT_NONE, 0, 0, "");
160 	spdk_trace_register_description("TCP_READ_DONE",
161 					TRACE_TCP_READ_FROM_SOCKET_DONE,
162 					OWNER_NONE, OBJECT_NONE, 0, 0, "");
163 	spdk_trace_register_description("TCP_REQ_AWAIT_R2T_ACK",
164 					TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK,
165 					OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, "");
166 }
167 
168 struct spdk_nvmf_tcp_req  {
169 	struct spdk_nvmf_request		req;
170 	struct spdk_nvme_cpl			rsp;
171 	struct spdk_nvme_cmd			cmd;
172 
173 	/* A PDU that can be used for sending responses. This is
174 	 * not the incoming PDU! */
175 	struct nvme_tcp_pdu			*pdu;
176 
177 	/*
178 	 * The PDU for a request may be used multiple times in serial over
179 	 * the request's lifetime. For example, first to send an R2T, then
180 	 * to send a completion. To catch mistakes where the PDU is used
181 	 * twice at the same time, add a debug flag here for init/fini.
182 	 */
183 	bool					pdu_in_use;
184 
185 	/* In-capsule data buffer */
186 	uint8_t					*buf;
187 
188 	bool					has_incapsule_data;
189 
190 	/* transfer_tag */
191 	uint16_t				ttag;
192 
193 	enum spdk_nvmf_tcp_req_state		state;
194 
195 	/*
196 	 * h2c_offset is used when we receive the h2c_data PDU.
197 	 */
198 	uint32_t				h2c_offset;
199 
200 	STAILQ_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 
210 	enum nvme_tcp_pdu_recv_state		recv_state;
211 	enum nvme_tcp_qpair_state		state;
212 
213 	/* PDU being actively received */
214 	struct nvme_tcp_pdu			pdu_in_progress;
215 	uint32_t				recv_buf_size;
216 
217 	/* This is a spare PDU used for sending special management
218 	 * operations. Primarily, this is used for the initial
219 	 * connection response and c2h termination request. */
220 	struct nvme_tcp_pdu			*mgmt_pdu;
221 
222 	TAILQ_HEAD(, nvme_tcp_pdu)		send_queue;
223 
224 	/* Arrays of in-capsule buffers, requests, and pdus.
225 	 * Each array is 'resource_count' number of elements */
226 	void					*bufs;
227 	struct spdk_nvmf_tcp_req		*reqs;
228 	struct nvme_tcp_pdu			*pdus;
229 	uint32_t				resource_count;
230 
231 	/* Queues to track the requests in all states */
232 	TAILQ_HEAD(, spdk_nvmf_tcp_req)		state_queue[TCP_REQUEST_NUM_STATES];
233 	/* Number of requests in each state */
234 	uint32_t				state_cntr[TCP_REQUEST_NUM_STATES];
235 
236 	uint8_t					cpda;
237 
238 	bool					host_hdgst_enable;
239 	bool					host_ddgst_enable;
240 
241 	/* IP address */
242 	char					initiator_addr[SPDK_NVMF_TRADDR_MAX_LEN];
243 	char					target_addr[SPDK_NVMF_TRADDR_MAX_LEN];
244 
245 	/* IP port */
246 	uint16_t				initiator_port;
247 	uint16_t				target_port;
248 
249 	/* Timer used to destroy qpair after detecting transport error issue if initiator does
250 	 *  not close the connection.
251 	 */
252 	struct spdk_poller			*timeout_poller;
253 
254 	TAILQ_ENTRY(spdk_nvmf_tcp_qpair)	link;
255 };
256 
257 struct spdk_nvmf_tcp_poll_group {
258 	struct spdk_nvmf_transport_poll_group	group;
259 	struct spdk_sock_group			*sock_group;
260 
261 	TAILQ_HEAD(, spdk_nvmf_tcp_qpair)	qpairs;
262 	TAILQ_HEAD(, spdk_nvmf_tcp_qpair)	await_req;
263 };
264 
265 struct spdk_nvmf_tcp_port {
266 	const struct spdk_nvme_transport_id	*trid;
267 	struct spdk_sock			*listen_sock;
268 	TAILQ_ENTRY(spdk_nvmf_tcp_port)		link;
269 };
270 
271 struct tcp_transport_opts {
272 	bool		c2h_success;
273 	uint32_t	sock_priority;
274 };
275 
276 struct spdk_nvmf_tcp_transport {
277 	struct spdk_nvmf_transport		transport;
278 	struct tcp_transport_opts               tcp_opts;
279 
280 	pthread_mutex_t				lock;
281 
282 	TAILQ_HEAD(, spdk_nvmf_tcp_port)	ports;
283 };
284 
285 static const struct spdk_json_object_decoder tcp_transport_opts_decoder[] = {
286 	{
287 		"c2h_success", offsetof(struct tcp_transport_opts, c2h_success),
288 		spdk_json_decode_bool, true
289 	},
290 	{
291 		"sock_priority", offsetof(struct tcp_transport_opts, sock_priority),
292 		spdk_json_decode_uint32, true
293 	},
294 };
295 
296 static bool nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport,
297 				 struct spdk_nvmf_tcp_req *tcp_req);
298 
299 static void
300 nvmf_tcp_req_set_state(struct spdk_nvmf_tcp_req *tcp_req,
301 		       enum spdk_nvmf_tcp_req_state state)
302 {
303 	struct spdk_nvmf_qpair *qpair;
304 	struct spdk_nvmf_tcp_qpair *tqpair;
305 
306 	qpair = tcp_req->req.qpair;
307 	tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
308 
309 	TAILQ_REMOVE(&tqpair->state_queue[tcp_req->state], tcp_req, state_link);
310 	assert(tqpair->state_cntr[tcp_req->state] > 0);
311 	tqpair->state_cntr[tcp_req->state]--;
312 
313 	TAILQ_INSERT_TAIL(&tqpair->state_queue[state], tcp_req, state_link);
314 	tqpair->state_cntr[state]++;
315 
316 	tcp_req->state = state;
317 }
318 
319 static inline struct nvme_tcp_pdu *
320 nvmf_tcp_req_pdu_init(struct spdk_nvmf_tcp_req *tcp_req)
321 {
322 	assert(tcp_req->pdu_in_use == false);
323 	tcp_req->pdu_in_use = true;
324 
325 	memset(tcp_req->pdu, 0, sizeof(*tcp_req->pdu));
326 	tcp_req->pdu->qpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair);
327 
328 	return tcp_req->pdu;
329 }
330 
331 static inline void
332 nvmf_tcp_req_pdu_fini(struct spdk_nvmf_tcp_req *tcp_req)
333 {
334 	tcp_req->pdu_in_use = false;
335 }
336 
337 static struct spdk_nvmf_tcp_req *
338 nvmf_tcp_req_get(struct spdk_nvmf_tcp_qpair *tqpair)
339 {
340 	struct spdk_nvmf_tcp_req *tcp_req;
341 
342 	tcp_req = TAILQ_FIRST(&tqpair->state_queue[TCP_REQUEST_STATE_FREE]);
343 	if (!tcp_req) {
344 		return NULL;
345 	}
346 
347 	memset(&tcp_req->rsp, 0, sizeof(tcp_req->rsp));
348 	tcp_req->h2c_offset = 0;
349 	tcp_req->has_incapsule_data = false;
350 	tcp_req->req.dif.dif_insert_or_strip = false;
351 
352 	nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW);
353 	return tcp_req;
354 }
355 
356 static void
357 nvmf_tcp_request_free(struct spdk_nvmf_tcp_req *tcp_req)
358 {
359 	struct spdk_nvmf_tcp_transport *ttransport;
360 
361 	assert(tcp_req != NULL);
362 
363 	SPDK_DEBUGLOG(nvmf_tcp, "tcp_req=%p will be freed\n", tcp_req);
364 	ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport,
365 				      struct spdk_nvmf_tcp_transport, transport);
366 	nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED);
367 	nvmf_tcp_req_process(ttransport, tcp_req);
368 }
369 
370 static int
371 nvmf_tcp_req_free(struct spdk_nvmf_request *req)
372 {
373 	struct spdk_nvmf_tcp_req *tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req);
374 
375 	nvmf_tcp_request_free(tcp_req);
376 
377 	return 0;
378 }
379 
380 static void
381 nvmf_tcp_drain_state_queue(struct spdk_nvmf_tcp_qpair *tqpair,
382 			   enum spdk_nvmf_tcp_req_state state)
383 {
384 	struct spdk_nvmf_tcp_req *tcp_req, *req_tmp;
385 
386 	TAILQ_FOREACH_SAFE(tcp_req, &tqpair->state_queue[state], state_link, req_tmp) {
387 		nvmf_tcp_request_free(tcp_req);
388 	}
389 }
390 
391 static void
392 nvmf_tcp_cleanup_all_states(struct spdk_nvmf_tcp_qpair *tqpair)
393 {
394 	struct spdk_nvmf_tcp_req *tcp_req, *req_tmp;
395 
396 	assert(TAILQ_EMPTY(&tqpair->send_queue));
397 
398 	nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST);
399 	nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEW);
400 
401 	/* Wipe the requests waiting for buffer from the global list */
402 	TAILQ_FOREACH_SAFE(tcp_req, &tqpair->state_queue[TCP_REQUEST_STATE_NEED_BUFFER], state_link,
403 			   req_tmp) {
404 		STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, &tcp_req->req,
405 			      spdk_nvmf_request, buf_link);
406 	}
407 
408 	nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEED_BUFFER);
409 	nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_EXECUTING);
410 	nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER);
411 	nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK);
412 }
413 
414 static void
415 nvmf_tcp_dump_qpair_req_contents(struct spdk_nvmf_tcp_qpair *tqpair)
416 {
417 	int i;
418 	struct spdk_nvmf_tcp_req *tcp_req;
419 
420 	SPDK_ERRLOG("Dumping contents of queue pair (QID %d)\n", tqpair->qpair.qid);
421 	for (i = 1; i < TCP_REQUEST_NUM_STATES; i++) {
422 		SPDK_ERRLOG("\tNum of requests in state[%d] = %u\n", i, tqpair->state_cntr[i]);
423 		TAILQ_FOREACH(tcp_req, &tqpair->state_queue[i], state_link) {
424 			SPDK_ERRLOG("\t\tRequest Data From Pool: %d\n", tcp_req->req.data_from_pool);
425 			SPDK_ERRLOG("\t\tRequest opcode: %d\n", tcp_req->req.cmd->nvmf_cmd.opcode);
426 		}
427 	}
428 }
429 
430 static void
431 nvmf_tcp_qpair_destroy(struct spdk_nvmf_tcp_qpair *tqpair)
432 {
433 	int err = 0;
434 
435 	SPDK_DEBUGLOG(nvmf_tcp, "enter\n");
436 
437 	err = spdk_sock_close(&tqpair->sock);
438 	assert(err == 0);
439 	nvmf_tcp_cleanup_all_states(tqpair);
440 
441 	if (tqpair->state_cntr[TCP_REQUEST_STATE_FREE] != tqpair->resource_count) {
442 		SPDK_ERRLOG("tqpair(%p) free tcp request num is %u but should be %u\n", tqpair,
443 			    tqpair->state_cntr[TCP_REQUEST_STATE_FREE],
444 			    tqpair->resource_count);
445 		err++;
446 	}
447 
448 	if (err > 0) {
449 		nvmf_tcp_dump_qpair_req_contents(tqpair);
450 	}
451 
452 	spdk_dma_free(tqpair->pdus);
453 	free(tqpair->reqs);
454 	spdk_free(tqpair->bufs);
455 	free(tqpair);
456 	SPDK_DEBUGLOG(nvmf_tcp, "Leave\n");
457 }
458 
459 static void
460 nvmf_tcp_dump_opts(struct spdk_nvmf_transport *transport, struct spdk_json_write_ctx *w)
461 {
462 	struct spdk_nvmf_tcp_transport	*ttransport;
463 	assert(w != NULL);
464 
465 	ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport);
466 	spdk_json_write_named_bool(w, "c2h_success", ttransport->tcp_opts.c2h_success);
467 	spdk_json_write_named_uint32(w, "sock_priority", ttransport->tcp_opts.sock_priority);
468 }
469 
470 static int
471 nvmf_tcp_destroy(struct spdk_nvmf_transport *transport)
472 {
473 	struct spdk_nvmf_tcp_transport	*ttransport;
474 
475 	assert(transport != NULL);
476 	ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport);
477 
478 	pthread_mutex_destroy(&ttransport->lock);
479 	free(ttransport);
480 	return 0;
481 }
482 
483 static struct spdk_nvmf_transport *
484 nvmf_tcp_create(struct spdk_nvmf_transport_opts *opts)
485 {
486 	struct spdk_nvmf_tcp_transport *ttransport;
487 	uint32_t sge_count;
488 	uint32_t min_shared_buffers;
489 
490 	ttransport = calloc(1, sizeof(*ttransport));
491 	if (!ttransport) {
492 		return NULL;
493 	}
494 
495 	TAILQ_INIT(&ttransport->ports);
496 
497 	ttransport->transport.ops = &spdk_nvmf_transport_tcp;
498 
499 	ttransport->tcp_opts.c2h_success = SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION;
500 	ttransport->tcp_opts.sock_priority = SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY;
501 	if (opts->transport_specific != NULL &&
502 	    spdk_json_decode_object_relaxed(opts->transport_specific, tcp_transport_opts_decoder,
503 					    SPDK_COUNTOF(tcp_transport_opts_decoder),
504 					    &ttransport->tcp_opts)) {
505 		SPDK_ERRLOG("spdk_json_decode_object_relaxed failed\n");
506 		free(ttransport);
507 		return NULL;
508 	}
509 
510 	SPDK_NOTICELOG("*** TCP Transport Init ***\n");
511 
512 	SPDK_INFOLOG(nvmf_tcp, "*** TCP Transport Init ***\n"
513 		     "  Transport opts:  max_ioq_depth=%d, max_io_size=%d,\n"
514 		     "  max_io_qpairs_per_ctrlr=%d, io_unit_size=%d,\n"
515 		     "  in_capsule_data_size=%d, max_aq_depth=%d\n"
516 		     "  num_shared_buffers=%d, c2h_success=%d,\n"
517 		     "  dif_insert_or_strip=%d, sock_priority=%d\n"
518 		     "  abort_timeout_sec=%d\n",
519 		     opts->max_queue_depth,
520 		     opts->max_io_size,
521 		     opts->max_qpairs_per_ctrlr - 1,
522 		     opts->io_unit_size,
523 		     opts->in_capsule_data_size,
524 		     opts->max_aq_depth,
525 		     opts->num_shared_buffers,
526 		     ttransport->tcp_opts.c2h_success,
527 		     opts->dif_insert_or_strip,
528 		     ttransport->tcp_opts.sock_priority,
529 		     opts->abort_timeout_sec);
530 
531 	if (ttransport->tcp_opts.sock_priority > SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY) {
532 		SPDK_ERRLOG("Unsupported socket_priority=%d, the current range is: 0 to %d\n"
533 			    "you can use man 7 socket to view the range of priority under SO_PRIORITY item\n",
534 			    ttransport->tcp_opts.sock_priority, SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY);
535 		free(ttransport);
536 		return NULL;
537 	}
538 
539 	/* I/O unit size cannot be larger than max I/O size */
540 	if (opts->io_unit_size > opts->max_io_size) {
541 		opts->io_unit_size = opts->max_io_size;
542 	}
543 
544 	sge_count = opts->max_io_size / opts->io_unit_size;
545 	if (sge_count > SPDK_NVMF_MAX_SGL_ENTRIES) {
546 		SPDK_ERRLOG("Unsupported IO Unit size specified, %d bytes\n", opts->io_unit_size);
547 		free(ttransport);
548 		return NULL;
549 	}
550 
551 	min_shared_buffers = spdk_env_get_core_count() * opts->buf_cache_size;
552 	if (min_shared_buffers > opts->num_shared_buffers) {
553 		SPDK_ERRLOG("There are not enough buffers to satisfy"
554 			    "per-poll group caches for each thread. (%" PRIu32 ")"
555 			    "supplied. (%" PRIu32 ") required\n", opts->num_shared_buffers, min_shared_buffers);
556 		SPDK_ERRLOG("Please specify a larger number of shared buffers\n");
557 		nvmf_tcp_destroy(&ttransport->transport);
558 		return NULL;
559 	}
560 
561 	pthread_mutex_init(&ttransport->lock, NULL);
562 
563 	return &ttransport->transport;
564 }
565 
566 static int
567 nvmf_tcp_trsvcid_to_int(const char *trsvcid)
568 {
569 	unsigned long long ull;
570 	char *end = NULL;
571 
572 	ull = strtoull(trsvcid, &end, 10);
573 	if (end == NULL || end == trsvcid || *end != '\0') {
574 		return -1;
575 	}
576 
577 	/* Valid TCP/IP port numbers are in [0, 65535] */
578 	if (ull > 65535) {
579 		return -1;
580 	}
581 
582 	return (int)ull;
583 }
584 
585 /**
586  * Canonicalize a listen address trid.
587  */
588 static int
589 nvmf_tcp_canon_listen_trid(struct spdk_nvme_transport_id *canon_trid,
590 			   const struct spdk_nvme_transport_id *trid)
591 {
592 	int trsvcid_int;
593 
594 	trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid);
595 	if (trsvcid_int < 0) {
596 		return -EINVAL;
597 	}
598 
599 	memset(canon_trid, 0, sizeof(*canon_trid));
600 	spdk_nvme_trid_populate_transport(canon_trid, SPDK_NVME_TRANSPORT_TCP);
601 	canon_trid->adrfam = trid->adrfam;
602 	snprintf(canon_trid->traddr, sizeof(canon_trid->traddr), "%s", trid->traddr);
603 	snprintf(canon_trid->trsvcid, sizeof(canon_trid->trsvcid), "%d", trsvcid_int);
604 
605 	return 0;
606 }
607 
608 /**
609  * Find an existing listening port.
610  *
611  * Caller must hold ttransport->lock.
612  */
613 static struct spdk_nvmf_tcp_port *
614 nvmf_tcp_find_port(struct spdk_nvmf_tcp_transport *ttransport,
615 		   const struct spdk_nvme_transport_id *trid)
616 {
617 	struct spdk_nvme_transport_id canon_trid;
618 	struct spdk_nvmf_tcp_port *port;
619 
620 	if (nvmf_tcp_canon_listen_trid(&canon_trid, trid) != 0) {
621 		return NULL;
622 	}
623 
624 	TAILQ_FOREACH(port, &ttransport->ports, link) {
625 		if (spdk_nvme_transport_id_compare(&canon_trid, port->trid) == 0) {
626 			return port;
627 		}
628 	}
629 
630 	return NULL;
631 }
632 
633 static int
634 nvmf_tcp_listen(struct spdk_nvmf_transport *transport,
635 		const struct spdk_nvme_transport_id *trid)
636 {
637 	struct spdk_nvmf_tcp_transport *ttransport;
638 	struct spdk_nvmf_tcp_port *port;
639 	int trsvcid_int;
640 	uint8_t adrfam;
641 	struct spdk_sock_opts opts;
642 
643 	ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport);
644 
645 	trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid);
646 	if (trsvcid_int < 0) {
647 		SPDK_ERRLOG("Invalid trsvcid '%s'\n", trid->trsvcid);
648 		return -EINVAL;
649 	}
650 
651 	pthread_mutex_lock(&ttransport->lock);
652 	port = calloc(1, sizeof(*port));
653 	if (!port) {
654 		SPDK_ERRLOG("Port allocation failed\n");
655 		pthread_mutex_unlock(&ttransport->lock);
656 		return -ENOMEM;
657 	}
658 
659 	port->trid = trid;
660 	opts.opts_size = sizeof(opts);
661 	spdk_sock_get_default_opts(&opts);
662 	opts.priority = ttransport->tcp_opts.sock_priority;
663 	port->listen_sock = spdk_sock_listen_ext(trid->traddr, trsvcid_int,
664 			    NULL, &opts);
665 	if (port->listen_sock == NULL) {
666 		SPDK_ERRLOG("spdk_sock_listen(%s, %d) failed: %s (%d)\n",
667 			    trid->traddr, trsvcid_int,
668 			    spdk_strerror(errno), errno);
669 		free(port);
670 		pthread_mutex_unlock(&ttransport->lock);
671 		return -errno;
672 	}
673 
674 	if (spdk_sock_is_ipv4(port->listen_sock)) {
675 		adrfam = SPDK_NVMF_ADRFAM_IPV4;
676 	} else if (spdk_sock_is_ipv6(port->listen_sock)) {
677 		adrfam = SPDK_NVMF_ADRFAM_IPV6;
678 	} else {
679 		SPDK_ERRLOG("Unhandled socket type\n");
680 		adrfam = 0;
681 	}
682 
683 	if (adrfam != trid->adrfam) {
684 		SPDK_ERRLOG("Socket address family mismatch\n");
685 		spdk_sock_close(&port->listen_sock);
686 		free(port);
687 		pthread_mutex_unlock(&ttransport->lock);
688 		return -EINVAL;
689 	}
690 
691 	SPDK_NOTICELOG("*** NVMe/TCP Target Listening on %s port %s ***\n",
692 		       trid->traddr, trid->trsvcid);
693 
694 	TAILQ_INSERT_TAIL(&ttransport->ports, port, link);
695 	pthread_mutex_unlock(&ttransport->lock);
696 	return 0;
697 }
698 
699 static void
700 nvmf_tcp_stop_listen(struct spdk_nvmf_transport *transport,
701 		     const struct spdk_nvme_transport_id *trid)
702 {
703 	struct spdk_nvmf_tcp_transport *ttransport;
704 	struct spdk_nvmf_tcp_port *port;
705 
706 	ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport);
707 
708 	SPDK_DEBUGLOG(nvmf_tcp, "Removing listen address %s port %s\n",
709 		      trid->traddr, trid->trsvcid);
710 
711 	pthread_mutex_lock(&ttransport->lock);
712 	port = nvmf_tcp_find_port(ttransport, trid);
713 	if (port) {
714 		TAILQ_REMOVE(&ttransport->ports, port, link);
715 		spdk_sock_close(&port->listen_sock);
716 		free(port);
717 	}
718 
719 	pthread_mutex_unlock(&ttransport->lock);
720 }
721 
722 static void nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair,
723 		enum nvme_tcp_pdu_recv_state state);
724 
725 static void
726 nvmf_tcp_qpair_disconnect(struct spdk_nvmf_tcp_qpair *tqpair)
727 {
728 	SPDK_DEBUGLOG(nvmf_tcp, "Disconnecting qpair %p\n", tqpair);
729 
730 	if (tqpair->state <= NVME_TCP_QPAIR_STATE_RUNNING) {
731 		tqpair->state = NVME_TCP_QPAIR_STATE_EXITING;
732 		nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR);
733 		spdk_poller_unregister(&tqpair->timeout_poller);
734 
735 		/* This will end up calling nvmf_tcp_close_qpair */
736 		spdk_nvmf_qpair_disconnect(&tqpair->qpair, NULL, NULL);
737 	}
738 }
739 
740 static void
741 _pdu_write_done(void *_pdu, int err)
742 {
743 	struct nvme_tcp_pdu			*pdu = _pdu;
744 	struct spdk_nvmf_tcp_qpair		*tqpair = pdu->qpair;
745 
746 	TAILQ_REMOVE(&tqpair->send_queue, pdu, tailq);
747 
748 	if (err != 0) {
749 		nvmf_tcp_qpair_disconnect(tqpair);
750 		return;
751 	}
752 
753 	assert(pdu->cb_fn != NULL);
754 	pdu->cb_fn(pdu->cb_arg);
755 }
756 
757 static void
758 nvmf_tcp_qpair_write_pdu(struct spdk_nvmf_tcp_qpair *tqpair,
759 			 struct nvme_tcp_pdu *pdu,
760 			 nvme_tcp_qpair_xfer_complete_cb cb_fn,
761 			 void *cb_arg)
762 {
763 	int hlen;
764 	uint32_t crc32c;
765 	uint32_t mapped_length = 0;
766 	ssize_t rc;
767 
768 	assert(&tqpair->pdu_in_progress != pdu);
769 
770 	hlen = pdu->hdr.common.hlen;
771 
772 	/* Header Digest */
773 	if (g_nvme_tcp_hdgst[pdu->hdr.common.pdu_type] && tqpair->host_hdgst_enable) {
774 		crc32c = nvme_tcp_pdu_calc_header_digest(pdu);
775 		MAKE_DIGEST_WORD((uint8_t *)pdu->hdr.raw + hlen, crc32c);
776 	}
777 
778 	/* Data Digest */
779 	if (pdu->data_len > 0 && g_nvme_tcp_ddgst[pdu->hdr.common.pdu_type] && tqpair->host_ddgst_enable) {
780 		crc32c = nvme_tcp_pdu_calc_data_digest(pdu);
781 		MAKE_DIGEST_WORD(pdu->data_digest, crc32c);
782 	}
783 
784 	pdu->cb_fn = cb_fn;
785 	pdu->cb_arg = cb_arg;
786 
787 	pdu->sock_req.iovcnt = nvme_tcp_build_iovs(pdu->iov, SPDK_COUNTOF(pdu->iov), pdu,
788 			       tqpair->host_hdgst_enable, tqpair->host_ddgst_enable,
789 			       &mapped_length);
790 	pdu->sock_req.cb_fn = _pdu_write_done;
791 	pdu->sock_req.cb_arg = pdu;
792 	TAILQ_INSERT_TAIL(&tqpair->send_queue, pdu, tailq);
793 	if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP ||
794 	    pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ) {
795 		rc = spdk_sock_writev(tqpair->sock, pdu->iov, pdu->sock_req.iovcnt);
796 		if (rc == mapped_length) {
797 			_pdu_write_done(pdu, 0);
798 		} else {
799 			SPDK_ERRLOG("IC_RESP or TERM_REQ could not write to socket.\n");
800 			_pdu_write_done(pdu, -1);
801 		}
802 	} else {
803 		spdk_sock_writev_async(tqpair->sock, &pdu->sock_req);
804 	}
805 }
806 
807 static int
808 nvmf_tcp_qpair_init_mem_resource(struct spdk_nvmf_tcp_qpair *tqpair)
809 {
810 	uint32_t i;
811 	struct spdk_nvmf_transport_opts *opts;
812 	uint32_t in_capsule_data_size;
813 
814 	opts = &tqpair->qpair.transport->opts;
815 
816 	in_capsule_data_size = opts->in_capsule_data_size;
817 	if (opts->dif_insert_or_strip) {
818 		in_capsule_data_size = SPDK_BDEV_BUF_SIZE_WITH_MD(in_capsule_data_size);
819 	}
820 
821 	tqpair->resource_count = opts->max_queue_depth;
822 
823 	tqpair->reqs = calloc(tqpair->resource_count, sizeof(*tqpair->reqs));
824 	if (!tqpair->reqs) {
825 		SPDK_ERRLOG("Unable to allocate reqs on tqpair=%p\n", tqpair);
826 		return -1;
827 	}
828 
829 	if (in_capsule_data_size) {
830 		tqpair->bufs = spdk_zmalloc(tqpair->resource_count * in_capsule_data_size, 0x1000,
831 					    NULL, SPDK_ENV_LCORE_ID_ANY,
832 					    SPDK_MALLOC_DMA);
833 		if (!tqpair->bufs) {
834 			SPDK_ERRLOG("Unable to allocate bufs on tqpair=%p.\n", tqpair);
835 			return -1;
836 		}
837 	}
838 
839 	/* Add addtional one member, which will be used for mgmt_pdu owned by the tqpair */
840 	tqpair->pdus = spdk_dma_malloc((tqpair->resource_count + 1) * sizeof(*tqpair->pdus), 0x1000, NULL);
841 	if (!tqpair->pdus) {
842 		SPDK_ERRLOG("Unable to allocate pdu pool on tqpair =%p.\n", tqpair);
843 		return -1;
844 	}
845 
846 	for (i = 0; i < tqpair->resource_count; i++) {
847 		struct spdk_nvmf_tcp_req *tcp_req = &tqpair->reqs[i];
848 
849 		tcp_req->ttag = i + 1;
850 		tcp_req->req.qpair = &tqpair->qpair;
851 
852 		tcp_req->pdu = &tqpair->pdus[i];
853 		tcp_req->pdu->qpair = tqpair;
854 
855 		/* Set up memory to receive commands */
856 		if (tqpair->bufs) {
857 			tcp_req->buf = (void *)((uintptr_t)tqpair->bufs + (i * in_capsule_data_size));
858 		}
859 
860 		/* Set the cmdn and rsp */
861 		tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp;
862 		tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd;
863 
864 		/* Initialize request state to FREE */
865 		tcp_req->state = TCP_REQUEST_STATE_FREE;
866 		TAILQ_INSERT_TAIL(&tqpair->state_queue[tcp_req->state], tcp_req, state_link);
867 		tqpair->state_cntr[TCP_REQUEST_STATE_FREE]++;
868 	}
869 
870 	tqpair->mgmt_pdu = &tqpair->pdus[i];
871 	tqpair->mgmt_pdu->qpair = tqpair;
872 
873 	tqpair->recv_buf_size = (in_capsule_data_size + sizeof(struct spdk_nvme_tcp_cmd) + 2 *
874 				 SPDK_NVME_TCP_DIGEST_LEN) * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR;
875 
876 	return 0;
877 }
878 
879 static int
880 nvmf_tcp_qpair_init(struct spdk_nvmf_qpair *qpair)
881 {
882 	struct spdk_nvmf_tcp_qpair *tqpair;
883 	int i;
884 
885 	tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
886 
887 	SPDK_DEBUGLOG(nvmf_tcp, "New TCP Connection: %p\n", qpair);
888 
889 	TAILQ_INIT(&tqpair->send_queue);
890 
891 	/* Initialise request state queues of the qpair */
892 	for (i = TCP_REQUEST_STATE_FREE; i < TCP_REQUEST_NUM_STATES; i++) {
893 		TAILQ_INIT(&tqpair->state_queue[i]);
894 	}
895 
896 	tqpair->host_hdgst_enable = true;
897 	tqpair->host_ddgst_enable = true;
898 
899 	return 0;
900 }
901 
902 static int
903 nvmf_tcp_qpair_sock_init(struct spdk_nvmf_tcp_qpair *tqpair)
904 {
905 	int rc;
906 
907 	/* set low water mark */
908 	rc = spdk_sock_set_recvlowat(tqpair->sock, sizeof(struct spdk_nvme_tcp_common_pdu_hdr));
909 	if (rc != 0) {
910 		SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n");
911 		return rc;
912 	}
913 
914 	return 0;
915 }
916 
917 static void
918 nvmf_tcp_handle_connect(struct spdk_nvmf_transport *transport,
919 			struct spdk_nvmf_tcp_port *port,
920 			struct spdk_sock *sock)
921 {
922 	struct spdk_nvmf_tcp_qpair *tqpair;
923 	int rc;
924 
925 	SPDK_DEBUGLOG(nvmf_tcp, "New connection accepted on %s port %s\n",
926 		      port->trid->traddr, port->trid->trsvcid);
927 
928 	tqpair = calloc(1, sizeof(struct spdk_nvmf_tcp_qpair));
929 	if (tqpair == NULL) {
930 		SPDK_ERRLOG("Could not allocate new connection.\n");
931 		spdk_sock_close(&sock);
932 		return;
933 	}
934 
935 	tqpair->sock = sock;
936 	tqpair->state_cntr[TCP_REQUEST_STATE_FREE] = 0;
937 	tqpair->port = port;
938 	tqpair->qpair.transport = transport;
939 	tqpair->qpair.trid = port->trid;
940 
941 	rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr,
942 			       sizeof(tqpair->target_addr), &tqpair->target_port,
943 			       tqpair->initiator_addr, sizeof(tqpair->initiator_addr),
944 			       &tqpair->initiator_port);
945 	if (rc < 0) {
946 		SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair);
947 		nvmf_tcp_qpair_destroy(tqpair);
948 		return;
949 	}
950 
951 	spdk_nvmf_tgt_new_qpair(transport->tgt, &tqpair->qpair);
952 }
953 
954 static uint32_t
955 nvmf_tcp_port_accept(struct spdk_nvmf_transport *transport, struct spdk_nvmf_tcp_port *port)
956 {
957 	struct spdk_sock *sock;
958 	uint32_t count = 0;
959 	int i;
960 
961 	for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) {
962 		sock = spdk_sock_accept(port->listen_sock);
963 		if (sock == NULL) {
964 			break;
965 		}
966 		count++;
967 		nvmf_tcp_handle_connect(transport, port, sock);
968 	}
969 
970 	return count;
971 }
972 
973 static uint32_t
974 nvmf_tcp_accept(struct spdk_nvmf_transport *transport)
975 {
976 	struct spdk_nvmf_tcp_transport *ttransport;
977 	struct spdk_nvmf_tcp_port *port;
978 	uint32_t count = 0;
979 
980 	ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport);
981 
982 	TAILQ_FOREACH(port, &ttransport->ports, link) {
983 		count += nvmf_tcp_port_accept(transport, port);
984 	}
985 
986 	return count;
987 }
988 
989 static void
990 nvmf_tcp_discover(struct spdk_nvmf_transport *transport,
991 		  struct spdk_nvme_transport_id *trid,
992 		  struct spdk_nvmf_discovery_log_page_entry *entry)
993 {
994 	entry->trtype = SPDK_NVMF_TRTYPE_TCP;
995 	entry->adrfam = trid->adrfam;
996 	entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_REQUIRED;
997 
998 	spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' ');
999 	spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' ');
1000 
1001 	entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE;
1002 }
1003 
1004 static struct spdk_nvmf_transport_poll_group *
1005 nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport)
1006 {
1007 	struct spdk_nvmf_tcp_poll_group *tgroup;
1008 
1009 	tgroup = calloc(1, sizeof(*tgroup));
1010 	if (!tgroup) {
1011 		return NULL;
1012 	}
1013 
1014 	tgroup->sock_group = spdk_sock_group_create(&tgroup->group);
1015 	if (!tgroup->sock_group) {
1016 		goto cleanup;
1017 	}
1018 
1019 	TAILQ_INIT(&tgroup->qpairs);
1020 	TAILQ_INIT(&tgroup->await_req);
1021 
1022 	return &tgroup->group;
1023 
1024 cleanup:
1025 	free(tgroup);
1026 	return NULL;
1027 }
1028 
1029 static struct spdk_nvmf_transport_poll_group *
1030 nvmf_tcp_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair)
1031 {
1032 	struct spdk_nvmf_tcp_qpair *tqpair;
1033 	struct spdk_sock_group *group = NULL;
1034 	int rc;
1035 
1036 	tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
1037 	rc = spdk_sock_get_optimal_sock_group(tqpair->sock, &group);
1038 	if (!rc && group != NULL) {
1039 		return spdk_sock_group_get_ctx(group);
1040 	}
1041 
1042 	return NULL;
1043 }
1044 
1045 static void
1046 nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group)
1047 {
1048 	struct spdk_nvmf_tcp_poll_group *tgroup;
1049 
1050 	tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group);
1051 	spdk_sock_group_close(&tgroup->sock_group);
1052 
1053 	free(tgroup);
1054 }
1055 
1056 static void
1057 nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair,
1058 			      enum nvme_tcp_pdu_recv_state state)
1059 {
1060 	if (tqpair->recv_state == state) {
1061 		SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n",
1062 			    tqpair, state);
1063 		return;
1064 	}
1065 
1066 	if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) {
1067 		/* When leaving the await req state, move the qpair to the main list */
1068 		TAILQ_REMOVE(&tqpair->group->await_req, tqpair, link);
1069 		TAILQ_INSERT_TAIL(&tqpair->group->qpairs, tqpair, link);
1070 	}
1071 
1072 	SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv state=%d\n", tqpair, state);
1073 	tqpair->recv_state = state;
1074 
1075 	switch (state) {
1076 	case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH:
1077 	case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH:
1078 	case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD:
1079 		break;
1080 	case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ:
1081 		TAILQ_REMOVE(&tqpair->group->qpairs, tqpair, link);
1082 		TAILQ_INSERT_TAIL(&tqpair->group->await_req, tqpair, link);
1083 		break;
1084 	case NVME_TCP_PDU_RECV_STATE_ERROR:
1085 	case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY:
1086 		memset(&tqpair->pdu_in_progress, 0, sizeof(tqpair->pdu_in_progress));
1087 		break;
1088 	default:
1089 		SPDK_ERRLOG("The state(%d) is invalid\n", state);
1090 		abort();
1091 		break;
1092 	}
1093 }
1094 
1095 static int
1096 nvmf_tcp_qpair_handle_timeout(void *ctx)
1097 {
1098 	struct spdk_nvmf_tcp_qpair *tqpair = ctx;
1099 
1100 	assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR);
1101 
1102 	SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair,
1103 		    SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT);
1104 
1105 	nvmf_tcp_qpair_disconnect(tqpair);
1106 	return SPDK_POLLER_BUSY;
1107 }
1108 
1109 static void
1110 nvmf_tcp_send_c2h_term_req_complete(void *cb_arg)
1111 {
1112 	struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg;
1113 
1114 	if (!tqpair->timeout_poller) {
1115 		tqpair->timeout_poller = SPDK_POLLER_REGISTER(nvmf_tcp_qpair_handle_timeout, tqpair,
1116 					 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000);
1117 	}
1118 }
1119 
1120 static void
1121 nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu,
1122 			   enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset)
1123 {
1124 	struct nvme_tcp_pdu *rsp_pdu;
1125 	struct spdk_nvme_tcp_term_req_hdr *c2h_term_req;
1126 	uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req);
1127 	uint32_t copy_len;
1128 
1129 	rsp_pdu = tqpair->mgmt_pdu;
1130 
1131 	c2h_term_req = &rsp_pdu->hdr.term_req;
1132 	c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ;
1133 	c2h_term_req->common.hlen = c2h_term_req_hdr_len;
1134 
1135 	if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) ||
1136 	    (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) {
1137 		DSET32(&c2h_term_req->fei, error_offset);
1138 	}
1139 
1140 	copy_len = spdk_min(pdu->hdr.common.hlen, SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE);
1141 
1142 	/* Copy the error info into the buffer */
1143 	memcpy((uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, pdu->hdr.raw, copy_len);
1144 	nvme_tcp_pdu_set_data(rsp_pdu, (uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, copy_len);
1145 
1146 	/* Contain the header of the wrong received pdu */
1147 	c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len;
1148 	nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR);
1149 	nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_send_c2h_term_req_complete, tqpair);
1150 }
1151 
1152 static void
1153 nvmf_tcp_capsule_cmd_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport,
1154 				struct spdk_nvmf_tcp_qpair *tqpair,
1155 				struct nvme_tcp_pdu *pdu)
1156 {
1157 	struct spdk_nvmf_tcp_req *tcp_req;
1158 
1159 	assert(pdu->psh_valid_bytes == pdu->psh_len);
1160 	assert(pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD);
1161 
1162 	tcp_req = nvmf_tcp_req_get(tqpair);
1163 	if (!tcp_req) {
1164 		/* Directly return and make the allocation retry again */
1165 		if (tqpair->state_cntr[TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST] > 0) {
1166 			return;
1167 		}
1168 
1169 		/* The host sent more commands than the maximum queue depth. */
1170 		SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair);
1171 		nvmf_tcp_qpair_disconnect(tqpair);
1172 		return;
1173 	}
1174 
1175 	pdu->req = tcp_req;
1176 	assert(tcp_req->state == TCP_REQUEST_STATE_NEW);
1177 	nvmf_tcp_req_process(ttransport, tcp_req);
1178 }
1179 
1180 static void
1181 nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport,
1182 				    struct spdk_nvmf_tcp_qpair *tqpair,
1183 				    struct nvme_tcp_pdu *pdu)
1184 {
1185 	struct spdk_nvmf_tcp_req *tcp_req;
1186 	struct spdk_nvme_tcp_cmd *capsule_cmd;
1187 	uint32_t error_offset = 0;
1188 	enum spdk_nvme_tcp_term_req_fes fes;
1189 
1190 	capsule_cmd = &pdu->hdr.capsule_cmd;
1191 	tcp_req = pdu->req;
1192 	assert(tcp_req != NULL);
1193 	if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) {
1194 		SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n",
1195 			    SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo);
1196 		fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
1197 		error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo);
1198 		goto err;
1199 	}
1200 
1201 	nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY);
1202 	nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE);
1203 	nvmf_tcp_req_process(ttransport, tcp_req);
1204 
1205 	return;
1206 err:
1207 	nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
1208 }
1209 
1210 static int
1211 nvmf_tcp_find_req_in_state(struct spdk_nvmf_tcp_qpair *tqpair,
1212 			   enum spdk_nvmf_tcp_req_state state,
1213 			   uint16_t cid, uint16_t tag,
1214 			   struct spdk_nvmf_tcp_req **req)
1215 {
1216 	struct spdk_nvmf_tcp_req *tcp_req = NULL;
1217 
1218 	TAILQ_FOREACH(tcp_req, &tqpair->state_queue[state], state_link) {
1219 		if (tcp_req->req.cmd->nvme_cmd.cid != cid) {
1220 			continue;
1221 		}
1222 
1223 		if (tcp_req->ttag == tag) {
1224 			*req = tcp_req;
1225 			return 0;
1226 		}
1227 
1228 		*req = NULL;
1229 		return -1;
1230 	}
1231 
1232 	/* Didn't find it, but not an error */
1233 	*req = NULL;
1234 	return 0;
1235 }
1236 
1237 static void
1238 nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport,
1239 			     struct spdk_nvmf_tcp_qpair *tqpair,
1240 			     struct nvme_tcp_pdu *pdu)
1241 {
1242 	struct spdk_nvmf_tcp_req *tcp_req;
1243 	uint32_t error_offset = 0;
1244 	enum spdk_nvme_tcp_term_req_fes fes = 0;
1245 	struct spdk_nvme_tcp_h2c_data_hdr *h2c_data;
1246 	int rc;
1247 
1248 	h2c_data = &pdu->hdr.h2c_data;
1249 
1250 	SPDK_DEBUGLOG(nvmf_tcp, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n",
1251 		      tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag);
1252 
1253 	rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER,
1254 					h2c_data->cccid, h2c_data->ttag, &tcp_req);
1255 	if (rc == 0 && tcp_req == NULL) {
1256 		rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK, h2c_data->cccid,
1257 						h2c_data->ttag, &tcp_req);
1258 	}
1259 
1260 	if (!tcp_req) {
1261 		SPDK_DEBUGLOG(nvmf_tcp, "tcp_req is not found for tqpair=%p\n", tqpair);
1262 		fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER;
1263 		if (rc == 0) {
1264 			error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid);
1265 		} else {
1266 			error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag);
1267 		}
1268 		goto err;
1269 	}
1270 
1271 	if (tcp_req->h2c_offset != h2c_data->datao) {
1272 		SPDK_DEBUGLOG(nvmf_tcp,
1273 			      "tcp_req(%p), tqpair=%p, expected data offset %u, but data offset is %u\n",
1274 			      tcp_req, tqpair, tcp_req->h2c_offset, h2c_data->datao);
1275 		fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE;
1276 		goto err;
1277 	}
1278 
1279 	if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) {
1280 		SPDK_DEBUGLOG(nvmf_tcp,
1281 			      "tcp_req(%p), tqpair=%p,  (datao=%u + datal=%u) execeeds requested length=%u\n",
1282 			      tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length);
1283 		fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE;
1284 		goto err;
1285 	}
1286 
1287 	pdu->req = tcp_req;
1288 
1289 	if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) {
1290 		pdu->dif_ctx = &tcp_req->req.dif.dif_ctx;
1291 	}
1292 
1293 	nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt,
1294 				  h2c_data->datao, h2c_data->datal);
1295 	nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD);
1296 	return;
1297 
1298 err:
1299 	nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
1300 }
1301 
1302 static void
1303 nvmf_tcp_pdu_cmd_complete(void *cb_arg)
1304 {
1305 	struct spdk_nvmf_tcp_req *tcp_req = cb_arg;
1306 	nvmf_tcp_request_free(tcp_req);
1307 }
1308 
1309 static void
1310 nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req,
1311 			       struct spdk_nvmf_tcp_qpair *tqpair)
1312 {
1313 	struct nvme_tcp_pdu *rsp_pdu;
1314 	struct spdk_nvme_tcp_rsp *capsule_resp;
1315 
1316 	SPDK_DEBUGLOG(nvmf_tcp, "enter, tqpair=%p\n", tqpair);
1317 
1318 	rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req);
1319 	assert(rsp_pdu != NULL);
1320 
1321 	capsule_resp = &rsp_pdu->hdr.capsule_resp;
1322 	capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP;
1323 	capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp);
1324 	capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl;
1325 	if (tqpair->host_hdgst_enable) {
1326 		capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF;
1327 		capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN;
1328 	}
1329 
1330 	nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_pdu_cmd_complete, tcp_req);
1331 }
1332 
1333 static void
1334 nvmf_tcp_pdu_c2h_data_complete(void *cb_arg)
1335 {
1336 	struct spdk_nvmf_tcp_req *tcp_req = cb_arg;
1337 	struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair,
1338 					     struct spdk_nvmf_tcp_qpair, qpair);
1339 	struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(
1340 				tcp_req->req.qpair->transport, struct spdk_nvmf_tcp_transport, transport);
1341 
1342 	assert(tqpair != NULL);
1343 	if (ttransport->tcp_opts.c2h_success) {
1344 		nvmf_tcp_request_free(tcp_req);
1345 	} else {
1346 		nvmf_tcp_req_pdu_fini(tcp_req);
1347 		nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair);
1348 	}
1349 }
1350 
1351 static void
1352 nvmf_tcp_r2t_complete(void *cb_arg)
1353 {
1354 	struct spdk_nvmf_tcp_req *tcp_req = cb_arg;
1355 	struct spdk_nvmf_tcp_transport *ttransport;
1356 
1357 	nvmf_tcp_req_pdu_fini(tcp_req);
1358 
1359 	ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport,
1360 				      struct spdk_nvmf_tcp_transport, transport);
1361 
1362 	nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER);
1363 
1364 	if (tcp_req->h2c_offset == tcp_req->req.length) {
1365 		nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE);
1366 		nvmf_tcp_req_process(ttransport, tcp_req);
1367 	}
1368 }
1369 
1370 static void
1371 nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair,
1372 		      struct spdk_nvmf_tcp_req *tcp_req)
1373 {
1374 	struct nvme_tcp_pdu *rsp_pdu;
1375 	struct spdk_nvme_tcp_r2t_hdr *r2t;
1376 
1377 	rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req);
1378 	assert(rsp_pdu != NULL);
1379 
1380 	r2t = &rsp_pdu->hdr.r2t;
1381 	r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T;
1382 	r2t->common.plen = r2t->common.hlen = sizeof(*r2t);
1383 
1384 	if (tqpair->host_hdgst_enable) {
1385 		r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF;
1386 		r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN;
1387 	}
1388 
1389 	r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid;
1390 	r2t->ttag = tcp_req->ttag;
1391 	r2t->r2to = tcp_req->h2c_offset;
1392 	r2t->r2tl = tcp_req->req.length;
1393 
1394 	nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_R2T_ACK);
1395 
1396 	SPDK_DEBUGLOG(nvmf_tcp,
1397 		      "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n",
1398 		      tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl);
1399 	nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_r2t_complete, tcp_req);
1400 }
1401 
1402 static void
1403 nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport,
1404 				 struct spdk_nvmf_tcp_qpair *tqpair,
1405 				 struct nvme_tcp_pdu *pdu)
1406 {
1407 	struct spdk_nvmf_tcp_req *tcp_req;
1408 
1409 	tcp_req = pdu->req;
1410 	assert(tcp_req != NULL);
1411 
1412 	SPDK_DEBUGLOG(nvmf_tcp, "enter\n");
1413 
1414 	tcp_req->h2c_offset += pdu->data_len;
1415 
1416 	nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY);
1417 
1418 	/* Wait for all of the data to arrive AND for the initial R2T PDU send to be
1419 	 * acknowledged before moving on. */
1420 	if (tcp_req->h2c_offset == tcp_req->req.length &&
1421 	    tcp_req->state == TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER) {
1422 		nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE);
1423 		nvmf_tcp_req_process(ttransport, tcp_req);
1424 	}
1425 }
1426 
1427 static void
1428 nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req)
1429 {
1430 	SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req,
1431 		    spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]);
1432 	if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) ||
1433 	    (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) {
1434 		SPDK_DEBUGLOG(nvmf_tcp, "The offset from the start of the PDU header is %u\n",
1435 			      DGET32(h2c_term_req->fei));
1436 	}
1437 }
1438 
1439 static void
1440 nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair,
1441 				 struct nvme_tcp_pdu *pdu)
1442 {
1443 	struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req;
1444 	uint32_t error_offset = 0;
1445 	enum spdk_nvme_tcp_term_req_fes fes;
1446 
1447 
1448 	if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) {
1449 		SPDK_ERRLOG("Fatal Error Stauts(FES) is unknown for h2c_term_req pdu=%p\n", pdu);
1450 		fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
1451 		error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes);
1452 		goto end;
1453 	}
1454 
1455 	/* set the data buffer */
1456 	nvme_tcp_pdu_set_data(pdu, (uint8_t *)pdu->hdr.raw + h2c_term_req->common.hlen,
1457 			      h2c_term_req->common.plen - h2c_term_req->common.hlen);
1458 	nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD);
1459 	return;
1460 end:
1461 	nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
1462 }
1463 
1464 static void
1465 nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair,
1466 				     struct nvme_tcp_pdu *pdu)
1467 {
1468 	struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req;
1469 
1470 	nvmf_tcp_h2c_term_req_dump(h2c_term_req);
1471 	nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR);
1472 }
1473 
1474 static void
1475 nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair,
1476 			    struct spdk_nvmf_tcp_transport *ttransport)
1477 {
1478 	int rc = 0;
1479 	struct nvme_tcp_pdu *pdu;
1480 	uint32_t crc32c, error_offset = 0;
1481 	enum spdk_nvme_tcp_term_req_fes fes;
1482 
1483 	assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD);
1484 	pdu = &tqpair->pdu_in_progress;
1485 
1486 	SPDK_DEBUGLOG(nvmf_tcp, "enter\n");
1487 	/* check data digest if need */
1488 	if (pdu->ddgst_enable) {
1489 		crc32c = nvme_tcp_pdu_calc_data_digest(pdu);
1490 		rc = MATCH_DIGEST_WORD(pdu->data_digest, crc32c);
1491 		if (rc == 0) {
1492 			SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu);
1493 			fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR;
1494 			nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
1495 			return;
1496 
1497 		}
1498 	}
1499 
1500 	switch (pdu->hdr.common.pdu_type) {
1501 	case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD:
1502 		nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu);
1503 		break;
1504 	case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA:
1505 		nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu);
1506 		break;
1507 
1508 	case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ:
1509 		nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu);
1510 		break;
1511 
1512 	default:
1513 		/* The code should not go to here */
1514 		SPDK_ERRLOG("The code should not go to here\n");
1515 		break;
1516 	}
1517 }
1518 
1519 static void
1520 nvmf_tcp_send_icresp_complete(void *cb_arg)
1521 {
1522 	struct spdk_nvmf_tcp_qpair *tqpair = cb_arg;
1523 
1524 	tqpair->state = NVME_TCP_QPAIR_STATE_RUNNING;
1525 }
1526 
1527 static void
1528 nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport,
1529 		      struct spdk_nvmf_tcp_qpair *tqpair,
1530 		      struct nvme_tcp_pdu *pdu)
1531 {
1532 	struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr.ic_req;
1533 	struct nvme_tcp_pdu *rsp_pdu;
1534 	struct spdk_nvme_tcp_ic_resp *ic_resp;
1535 	uint32_t error_offset = 0;
1536 	enum spdk_nvme_tcp_term_req_fes fes;
1537 
1538 	/* Only PFV 0 is defined currently */
1539 	if (ic_req->pfv != 0) {
1540 		SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv);
1541 		fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
1542 		error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv);
1543 		goto end;
1544 	}
1545 
1546 	/* MAXR2T is 0's based */
1547 	SPDK_DEBUGLOG(nvmf_tcp, "maxr2t =%u\n", (ic_req->maxr2t + 1u));
1548 
1549 	tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false;
1550 	if (!tqpair->host_hdgst_enable) {
1551 		tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR;
1552 	}
1553 
1554 	tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false;
1555 	if (!tqpair->host_ddgst_enable) {
1556 		tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR;
1557 	}
1558 
1559 	/* Now that we know whether digests are enabled, properly size the receive buffer */
1560 	if (spdk_sock_set_recvbuf(tqpair->sock, tqpair->recv_buf_size) < 0) {
1561 		SPDK_WARNLOG("Unable to allocate enough memory for receive buffer on tqpair=%p with size=%d\n",
1562 			     tqpair,
1563 			     tqpair->recv_buf_size);
1564 		/* Not fatal. */
1565 	}
1566 
1567 	tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX);
1568 	SPDK_DEBUGLOG(nvmf_tcp, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda);
1569 
1570 	rsp_pdu = tqpair->mgmt_pdu;
1571 
1572 	ic_resp = &rsp_pdu->hdr.ic_resp;
1573 	ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP;
1574 	ic_resp->common.hlen = ic_resp->common.plen =  sizeof(*ic_resp);
1575 	ic_resp->pfv = 0;
1576 	ic_resp->cpda = tqpair->cpda;
1577 	ic_resp->maxh2cdata = ttransport->transport.opts.max_io_size;
1578 	ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0;
1579 	ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0;
1580 
1581 	SPDK_DEBUGLOG(nvmf_tcp, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable);
1582 	SPDK_DEBUGLOG(nvmf_tcp, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable);
1583 
1584 	tqpair->state = NVME_TCP_QPAIR_STATE_INITIALIZING;
1585 	nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_send_icresp_complete, tqpair);
1586 	nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY);
1587 	return;
1588 end:
1589 	nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
1590 }
1591 
1592 static void
1593 nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair,
1594 			struct spdk_nvmf_tcp_transport *ttransport)
1595 {
1596 	struct nvme_tcp_pdu *pdu;
1597 	int rc;
1598 	uint32_t crc32c, error_offset = 0;
1599 	enum spdk_nvme_tcp_term_req_fes fes;
1600 
1601 	assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH);
1602 	pdu = &tqpair->pdu_in_progress;
1603 
1604 	SPDK_DEBUGLOG(nvmf_tcp, "pdu type of tqpair(%p) is %d\n", tqpair,
1605 		      pdu->hdr.common.pdu_type);
1606 	/* check header digest if needed */
1607 	if (pdu->has_hdgst) {
1608 		SPDK_DEBUGLOG(nvmf_tcp, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair);
1609 		crc32c = nvme_tcp_pdu_calc_header_digest(pdu);
1610 		rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, crc32c);
1611 		if (rc == 0) {
1612 			SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu);
1613 			fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR;
1614 			nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
1615 			return;
1616 
1617 		}
1618 	}
1619 
1620 	switch (pdu->hdr.common.pdu_type) {
1621 	case SPDK_NVME_TCP_PDU_TYPE_IC_REQ:
1622 		nvmf_tcp_icreq_handle(ttransport, tqpair, pdu);
1623 		break;
1624 	case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD:
1625 		nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_REQ);
1626 		break;
1627 	case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA:
1628 		nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu);
1629 		break;
1630 
1631 	case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ:
1632 		nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu);
1633 		break;
1634 
1635 	default:
1636 		SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress.hdr.common.pdu_type);
1637 		fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
1638 		error_offset = 1;
1639 		nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
1640 		break;
1641 	}
1642 }
1643 
1644 static void
1645 nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair)
1646 {
1647 	struct nvme_tcp_pdu *pdu;
1648 	uint32_t error_offset = 0;
1649 	enum spdk_nvme_tcp_term_req_fes fes;
1650 	uint8_t expected_hlen, pdo;
1651 	bool plen_error = false, pdo_error = false;
1652 
1653 	assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH);
1654 	pdu = &tqpair->pdu_in_progress;
1655 
1656 	if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) {
1657 		if (tqpair->state != NVME_TCP_QPAIR_STATE_INVALID) {
1658 			SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu);
1659 			fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR;
1660 			goto err;
1661 		}
1662 		expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req);
1663 		if (pdu->hdr.common.plen != expected_hlen) {
1664 			plen_error = true;
1665 		}
1666 	} else {
1667 		if (tqpair->state != NVME_TCP_QPAIR_STATE_RUNNING) {
1668 			SPDK_ERRLOG("The TCP/IP connection is not negotitated\n");
1669 			fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR;
1670 			goto err;
1671 		}
1672 
1673 		switch (pdu->hdr.common.pdu_type) {
1674 		case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD:
1675 			expected_hlen = sizeof(struct spdk_nvme_tcp_cmd);
1676 			pdo = pdu->hdr.common.pdo;
1677 			if ((tqpair->cpda != 0) && (pdo != ((tqpair->cpda + 1) << 2))) {
1678 				pdo_error = true;
1679 				break;
1680 			}
1681 
1682 			if (pdu->hdr.common.plen < expected_hlen) {
1683 				plen_error = true;
1684 			}
1685 			break;
1686 		case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA:
1687 			expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr);
1688 			pdo = pdu->hdr.common.pdo;
1689 			if ((tqpair->cpda != 0) && (pdo != ((tqpair->cpda + 1) << 2))) {
1690 				pdo_error = true;
1691 				break;
1692 			}
1693 			if (pdu->hdr.common.plen < expected_hlen) {
1694 				plen_error = true;
1695 			}
1696 			break;
1697 
1698 		case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ:
1699 			expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr);
1700 			if ((pdu->hdr.common.plen <= expected_hlen) ||
1701 			    (pdu->hdr.common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) {
1702 				plen_error = true;
1703 			}
1704 			break;
1705 
1706 		default:
1707 			SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr.common.pdu_type);
1708 			fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
1709 			error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type);
1710 			goto err;
1711 		}
1712 	}
1713 
1714 	if (pdu->hdr.common.hlen != expected_hlen) {
1715 		SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n",
1716 			    pdu->hdr.common.pdu_type,
1717 			    expected_hlen, pdu->hdr.common.hlen, tqpair);
1718 		fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
1719 		error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen);
1720 		goto err;
1721 	} else if (pdo_error) {
1722 		fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
1723 		error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo);
1724 	} else if (plen_error) {
1725 		fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
1726 		error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen);
1727 		goto err;
1728 	} else {
1729 		nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH);
1730 		nvme_tcp_pdu_calc_psh_len(&tqpair->pdu_in_progress, tqpair->host_hdgst_enable);
1731 		return;
1732 	}
1733 err:
1734 	nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
1735 }
1736 
1737 static int
1738 nvmf_tcp_pdu_payload_insert_dif(struct nvme_tcp_pdu *pdu, uint32_t read_offset,
1739 				int read_len)
1740 {
1741 	int rc;
1742 
1743 	rc = spdk_dif_generate_stream(pdu->data_iov, pdu->data_iovcnt,
1744 				      read_offset, read_len, pdu->dif_ctx);
1745 	if (rc != 0) {
1746 		SPDK_ERRLOG("DIF generate failed\n");
1747 	}
1748 
1749 	return rc;
1750 }
1751 
1752 static int
1753 nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair)
1754 {
1755 	int rc = 0;
1756 	struct nvme_tcp_pdu *pdu;
1757 	enum nvme_tcp_pdu_recv_state prev_state;
1758 	uint32_t data_len;
1759 	struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport,
1760 			struct spdk_nvmf_tcp_transport, transport);
1761 
1762 	/* The loop here is to allow for several back-to-back state changes. */
1763 	do {
1764 		prev_state = tqpair->recv_state;
1765 		SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state);
1766 
1767 		pdu = &tqpair->pdu_in_progress;
1768 		switch (tqpair->recv_state) {
1769 		/* Wait for the common header  */
1770 		case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY:
1771 		case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH:
1772 			if (spdk_unlikely(tqpair->state == NVME_TCP_QPAIR_STATE_INITIALIZING)) {
1773 				return rc;
1774 			}
1775 
1776 			rc = nvme_tcp_read_data(tqpair->sock,
1777 						sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes,
1778 						(void *)&pdu->hdr.common + pdu->ch_valid_bytes);
1779 			if (rc < 0) {
1780 				SPDK_DEBUGLOG(nvmf_tcp, "will disconnect tqpair=%p\n", tqpair);
1781 				return NVME_TCP_PDU_FATAL;
1782 			} else if (rc > 0) {
1783 				pdu->ch_valid_bytes += rc;
1784 				spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, 0, rc, 0, 0);
1785 				if (spdk_likely(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY)) {
1786 					nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH);
1787 				}
1788 			}
1789 
1790 			if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) {
1791 				return NVME_TCP_PDU_IN_PROGRESS;
1792 			}
1793 
1794 			/* The command header of this PDU has now been read from the socket. */
1795 			nvmf_tcp_pdu_ch_handle(tqpair);
1796 			break;
1797 		/* Wait for the pdu specific header  */
1798 		case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH:
1799 			rc = nvme_tcp_read_data(tqpair->sock,
1800 						pdu->psh_len - pdu->psh_valid_bytes,
1801 						(void *)&pdu->hdr.raw + sizeof(struct spdk_nvme_tcp_common_pdu_hdr) + pdu->psh_valid_bytes);
1802 			if (rc < 0) {
1803 				return NVME_TCP_PDU_FATAL;
1804 			} else if (rc > 0) {
1805 				spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE,
1806 						  0, rc, 0, 0);
1807 				pdu->psh_valid_bytes += rc;
1808 			}
1809 
1810 			if (pdu->psh_valid_bytes < pdu->psh_len) {
1811 				return NVME_TCP_PDU_IN_PROGRESS;
1812 			}
1813 
1814 			/* All header(ch, psh, head digist) of this PDU has now been read from the socket. */
1815 			nvmf_tcp_pdu_psh_handle(tqpair, ttransport);
1816 			break;
1817 		/* Wait for the req slot */
1818 		case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ:
1819 			nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu);
1820 			break;
1821 		case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD:
1822 			/* check whether the data is valid, if not we just return */
1823 			if (!pdu->data_len) {
1824 				return NVME_TCP_PDU_IN_PROGRESS;
1825 			}
1826 
1827 			data_len = pdu->data_len;
1828 			/* data digest */
1829 			if (spdk_unlikely((pdu->hdr.common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) &&
1830 					  tqpair->host_ddgst_enable)) {
1831 				data_len += SPDK_NVME_TCP_DIGEST_LEN;
1832 				pdu->ddgst_enable = true;
1833 			}
1834 
1835 			rc = nvme_tcp_read_payload_data(tqpair->sock, pdu);
1836 			if (rc < 0) {
1837 				return NVME_TCP_PDU_FATAL;
1838 			}
1839 			pdu->readv_offset += rc;
1840 
1841 			if (spdk_unlikely(pdu->dif_ctx != NULL)) {
1842 				rc = nvmf_tcp_pdu_payload_insert_dif(pdu, pdu->readv_offset - rc, rc);
1843 				if (rc != 0) {
1844 					return NVME_TCP_PDU_FATAL;
1845 				}
1846 			}
1847 
1848 			if (pdu->readv_offset < data_len) {
1849 				return NVME_TCP_PDU_IN_PROGRESS;
1850 			}
1851 
1852 			/* All of this PDU has now been read from the socket. */
1853 			nvmf_tcp_pdu_payload_handle(tqpair, ttransport);
1854 			break;
1855 		case NVME_TCP_PDU_RECV_STATE_ERROR:
1856 			if (!spdk_sock_is_connected(tqpair->sock)) {
1857 				return NVME_TCP_PDU_FATAL;
1858 			}
1859 			break;
1860 		default:
1861 			assert(0);
1862 			SPDK_ERRLOG("code should not come to here");
1863 			break;
1864 		}
1865 	} while (tqpair->recv_state != prev_state);
1866 
1867 	return rc;
1868 }
1869 
1870 static int
1871 nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_req *tcp_req,
1872 		       struct spdk_nvmf_transport *transport,
1873 		       struct spdk_nvmf_transport_poll_group *group)
1874 {
1875 	struct spdk_nvmf_request		*req = &tcp_req->req;
1876 	struct spdk_nvme_cmd			*cmd;
1877 	struct spdk_nvme_cpl			*rsp;
1878 	struct spdk_nvme_sgl_descriptor		*sgl;
1879 	uint32_t				length;
1880 
1881 	cmd = &req->cmd->nvme_cmd;
1882 	rsp = &req->rsp->nvme_cpl;
1883 	sgl = &cmd->dptr.sgl1;
1884 
1885 	length = sgl->unkeyed.length;
1886 
1887 	if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK &&
1888 	    sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) {
1889 		if (length > transport->opts.max_io_size) {
1890 			SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n",
1891 				    length, transport->opts.max_io_size);
1892 			rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID;
1893 			return -1;
1894 		}
1895 
1896 		/* fill request length and populate iovs */
1897 		req->length = length;
1898 
1899 		SPDK_DEBUGLOG(nvmf_tcp, "Data requested length= 0x%x\n", length);
1900 
1901 		if (spdk_unlikely(req->dif.dif_insert_or_strip)) {
1902 			req->dif.orig_length = length;
1903 			length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx);
1904 			req->dif.elba_length = length;
1905 		}
1906 
1907 		if (spdk_nvmf_request_get_buffers(req, group, transport, length)) {
1908 			/* No available buffers. Queue this request up. */
1909 			SPDK_DEBUGLOG(nvmf_tcp, "No available large data buffers. Queueing request %p\n",
1910 				      tcp_req);
1911 			return 0;
1912 		}
1913 
1914 		/* backward compatible */
1915 		req->data = req->iov[0].iov_base;
1916 
1917 		SPDK_DEBUGLOG(nvmf_tcp, "Request %p took %d buffer/s from central pool, and data=%p\n",
1918 			      tcp_req, req->iovcnt, req->data);
1919 
1920 		return 0;
1921 	} else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK &&
1922 		   sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) {
1923 		uint64_t offset = sgl->address;
1924 		uint32_t max_len = transport->opts.in_capsule_data_size;
1925 
1926 		SPDK_DEBUGLOG(nvmf_tcp, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n",
1927 			      offset, length);
1928 
1929 		if (offset > max_len) {
1930 			SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " exceeds capsule length 0x%x\n",
1931 				    offset, max_len);
1932 			rsp->status.sc = SPDK_NVME_SC_INVALID_SGL_OFFSET;
1933 			return -1;
1934 		}
1935 		max_len -= (uint32_t)offset;
1936 
1937 		if (length > max_len) {
1938 			SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n",
1939 				    length, max_len);
1940 			rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID;
1941 			return -1;
1942 		}
1943 
1944 		req->data = tcp_req->buf + offset;
1945 		req->data_from_pool = false;
1946 		req->length = length;
1947 
1948 		if (spdk_unlikely(req->dif.dif_insert_or_strip)) {
1949 			length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx);
1950 			req->dif.elba_length = length;
1951 		}
1952 
1953 		req->iov[0].iov_base = req->data;
1954 		req->iov[0].iov_len = length;
1955 		req->iovcnt = 1;
1956 
1957 		return 0;
1958 	}
1959 
1960 	SPDK_ERRLOG("Invalid NVMf I/O Command SGL:  Type 0x%x, Subtype 0x%x\n",
1961 		    sgl->generic.type, sgl->generic.subtype);
1962 	rsp->status.sc = SPDK_NVME_SC_SGL_DESCRIPTOR_TYPE_INVALID;
1963 	return -1;
1964 }
1965 
1966 static inline enum spdk_nvme_media_error_status_code
1967 nvmf_tcp_dif_error_to_compl_status(uint8_t err_type) {
1968 	enum spdk_nvme_media_error_status_code result;
1969 
1970 	switch (err_type)
1971 	{
1972 	case SPDK_DIF_REFTAG_ERROR:
1973 		result = SPDK_NVME_SC_REFERENCE_TAG_CHECK_ERROR;
1974 		break;
1975 	case SPDK_DIF_APPTAG_ERROR:
1976 		result = SPDK_NVME_SC_APPLICATION_TAG_CHECK_ERROR;
1977 		break;
1978 	case SPDK_DIF_GUARD_ERROR:
1979 		result = SPDK_NVME_SC_GUARD_CHECK_ERROR;
1980 		break;
1981 	default:
1982 		SPDK_UNREACHABLE();
1983 		break;
1984 	}
1985 
1986 	return result;
1987 }
1988 
1989 static void
1990 nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair,
1991 		       struct spdk_nvmf_tcp_req *tcp_req)
1992 {
1993 	struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(
1994 				tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport);
1995 	struct nvme_tcp_pdu *rsp_pdu;
1996 	struct spdk_nvme_tcp_c2h_data_hdr *c2h_data;
1997 	uint32_t plen, pdo, alignment;
1998 	int rc;
1999 
2000 	SPDK_DEBUGLOG(nvmf_tcp, "enter\n");
2001 
2002 	rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req);
2003 	assert(rsp_pdu != NULL);
2004 
2005 	c2h_data = &rsp_pdu->hdr.c2h_data;
2006 	c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA;
2007 	plen = c2h_data->common.hlen = sizeof(*c2h_data);
2008 
2009 	if (tqpair->host_hdgst_enable) {
2010 		plen += SPDK_NVME_TCP_DIGEST_LEN;
2011 		c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF;
2012 	}
2013 
2014 	/* set the psh */
2015 	c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid;
2016 	c2h_data->datal = tcp_req->req.length;
2017 	c2h_data->datao = 0;
2018 
2019 	/* set the padding */
2020 	rsp_pdu->padding_len = 0;
2021 	pdo = plen;
2022 	if (tqpair->cpda) {
2023 		alignment = (tqpair->cpda + 1) << 2;
2024 		if (alignment > plen) {
2025 			rsp_pdu->padding_len = alignment - plen;
2026 			pdo = plen = alignment;
2027 		}
2028 	}
2029 
2030 	c2h_data->common.pdo = pdo;
2031 	plen += c2h_data->datal;
2032 	if (tqpair->host_ddgst_enable) {
2033 		c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF;
2034 		plen += SPDK_NVME_TCP_DIGEST_LEN;
2035 	}
2036 
2037 	c2h_data->common.plen = plen;
2038 
2039 	if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) {
2040 		rsp_pdu->dif_ctx = &tcp_req->req.dif.dif_ctx;
2041 	}
2042 
2043 	nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt,
2044 				  c2h_data->datao, c2h_data->datal);
2045 
2046 	if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) {
2047 		struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl;
2048 		struct spdk_dif_error err_blk = {};
2049 
2050 		rc = spdk_dif_verify_stream(rsp_pdu->data_iov, rsp_pdu->data_iovcnt,
2051 					    0, rsp_pdu->data_len, rsp_pdu->dif_ctx, &err_blk);
2052 		if (rc != 0) {
2053 			SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n",
2054 				    err_blk.err_type, err_blk.err_offset);
2055 			rsp->status.sct = SPDK_NVME_SCT_MEDIA_ERROR;
2056 			rsp->status.sc = nvmf_tcp_dif_error_to_compl_status(err_blk.err_type);
2057 			nvmf_tcp_req_pdu_fini(tcp_req);
2058 			nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair);
2059 			return;
2060 		}
2061 	}
2062 
2063 	c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU;
2064 	if (ttransport->tcp_opts.c2h_success) {
2065 		c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS;
2066 	}
2067 
2068 	nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_pdu_c2h_data_complete, tcp_req);
2069 }
2070 
2071 static int
2072 request_transfer_out(struct spdk_nvmf_request *req)
2073 {
2074 	struct spdk_nvmf_tcp_req	*tcp_req;
2075 	struct spdk_nvmf_qpair		*qpair;
2076 	struct spdk_nvmf_tcp_qpair	*tqpair;
2077 	struct spdk_nvme_cpl		*rsp;
2078 
2079 	SPDK_DEBUGLOG(nvmf_tcp, "enter\n");
2080 
2081 	qpair = req->qpair;
2082 	rsp = &req->rsp->nvme_cpl;
2083 	tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req);
2084 
2085 	/* Advance our sq_head pointer */
2086 	if (qpair->sq_head == qpair->sq_head_max) {
2087 		qpair->sq_head = 0;
2088 	} else {
2089 		qpair->sq_head++;
2090 	}
2091 	rsp->sqhd = qpair->sq_head;
2092 
2093 	tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair);
2094 	nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST);
2095 	if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) {
2096 		nvmf_tcp_send_c2h_data(tqpair, tcp_req);
2097 	} else {
2098 		nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair);
2099 	}
2100 
2101 	return 0;
2102 }
2103 
2104 static void
2105 nvmf_tcp_set_incapsule_data(struct spdk_nvmf_tcp_qpair *tqpair,
2106 			    struct spdk_nvmf_tcp_req *tcp_req)
2107 {
2108 	struct nvme_tcp_pdu *pdu;
2109 	uint32_t plen = 0;
2110 
2111 	pdu = &tqpair->pdu_in_progress;
2112 	plen = pdu->hdr.common.hlen;
2113 
2114 	if (tqpair->host_hdgst_enable) {
2115 		plen += SPDK_NVME_TCP_DIGEST_LEN;
2116 	}
2117 
2118 	if (pdu->hdr.common.plen != plen) {
2119 		tcp_req->has_incapsule_data = true;
2120 	}
2121 }
2122 
2123 static bool
2124 nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport,
2125 		     struct spdk_nvmf_tcp_req *tcp_req)
2126 {
2127 	struct spdk_nvmf_tcp_qpair		*tqpair;
2128 	int					rc;
2129 	enum spdk_nvmf_tcp_req_state		prev_state;
2130 	bool					progress = false;
2131 	struct spdk_nvmf_transport		*transport = &ttransport->transport;
2132 	struct spdk_nvmf_transport_poll_group	*group;
2133 
2134 	tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair);
2135 	group = &tqpair->group->group;
2136 	assert(tcp_req->state != TCP_REQUEST_STATE_FREE);
2137 
2138 	/* If the qpair is not active, we need to abort the outstanding requests. */
2139 	if (tqpair->qpair.state != SPDK_NVMF_QPAIR_ACTIVE) {
2140 		if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) {
2141 			STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link);
2142 		}
2143 		nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED);
2144 	}
2145 
2146 	/* The loop here is to allow for several back-to-back state changes. */
2147 	do {
2148 		prev_state = tcp_req->state;
2149 
2150 		SPDK_DEBUGLOG(nvmf_tcp, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state,
2151 			      tqpair);
2152 
2153 		switch (tcp_req->state) {
2154 		case TCP_REQUEST_STATE_FREE:
2155 			/* Some external code must kick a request into TCP_REQUEST_STATE_NEW
2156 			 * to escape this state. */
2157 			break;
2158 		case TCP_REQUEST_STATE_NEW:
2159 			spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, 0, 0, (uintptr_t)tcp_req, 0);
2160 
2161 			/* copy the cmd from the receive pdu */
2162 			tcp_req->cmd = tqpair->pdu_in_progress.hdr.capsule_cmd.ccsqe;
2163 
2164 			if (spdk_unlikely(spdk_nvmf_request_get_dif_ctx(&tcp_req->req, &tcp_req->req.dif.dif_ctx))) {
2165 				tcp_req->req.dif.dif_insert_or_strip = true;
2166 				tqpair->pdu_in_progress.dif_ctx = &tcp_req->req.dif.dif_ctx;
2167 			}
2168 
2169 			/* The next state transition depends on the data transfer needs of this request. */
2170 			tcp_req->req.xfer = spdk_nvmf_req_get_xfer(&tcp_req->req);
2171 
2172 			if (spdk_unlikely(tcp_req->req.xfer == SPDK_NVME_DATA_BIDIRECTIONAL)) {
2173 				tcp_req->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
2174 				tcp_req->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SC_INVALID_OPCODE;
2175 				nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE);
2176 				SPDK_DEBUGLOG(nvmf_tcp, "Request %p: invalid xfer type (BIDIRECTIONAL)\n", tcp_req);
2177 				break;
2178 			}
2179 
2180 			/* If no data to transfer, ready to execute. */
2181 			if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) {
2182 				/* Reset the tqpair receving pdu state */
2183 				nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY);
2184 				nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE);
2185 				break;
2186 			}
2187 
2188 			nvmf_tcp_set_incapsule_data(tqpair, tcp_req);
2189 
2190 			if (!tcp_req->has_incapsule_data) {
2191 				nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY);
2192 			}
2193 
2194 			nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER);
2195 			STAILQ_INSERT_TAIL(&group->pending_buf_queue, &tcp_req->req, buf_link);
2196 			break;
2197 		case TCP_REQUEST_STATE_NEED_BUFFER:
2198 			spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 0, 0, (uintptr_t)tcp_req, 0);
2199 
2200 			assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE);
2201 
2202 			if (!tcp_req->has_incapsule_data && (&tcp_req->req != STAILQ_FIRST(&group->pending_buf_queue))) {
2203 				SPDK_DEBUGLOG(nvmf_tcp,
2204 					      "Not the first element to wait for the buf for tcp_req(%p) on tqpair=%p\n",
2205 					      tcp_req, tqpair);
2206 				/* This request needs to wait in line to obtain a buffer */
2207 				break;
2208 			}
2209 
2210 			/* Try to get a data buffer */
2211 			rc = nvmf_tcp_req_parse_sgl(tcp_req, transport, group);
2212 			if (rc < 0) {
2213 				STAILQ_REMOVE_HEAD(&group->pending_buf_queue, buf_link);
2214 				/* Reset the tqpair receving pdu state */
2215 				nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR);
2216 				nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE);
2217 				break;
2218 			}
2219 
2220 			if (!tcp_req->req.data) {
2221 				SPDK_DEBUGLOG(nvmf_tcp, "No buffer allocated for tcp_req(%p) on tqpair(%p\n)",
2222 					      tcp_req, tqpair);
2223 				/* No buffers available. */
2224 				break;
2225 			}
2226 
2227 			STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link);
2228 
2229 			/* If data is transferring from host to controller, we need to do a transfer from the host. */
2230 			if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) {
2231 				if (tcp_req->req.data_from_pool) {
2232 					SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair);
2233 					nvmf_tcp_send_r2t_pdu(tqpair, tcp_req);
2234 				} else {
2235 					struct nvme_tcp_pdu *pdu;
2236 
2237 					nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER);
2238 
2239 					pdu = &tqpair->pdu_in_progress;
2240 					SPDK_DEBUGLOG(nvmf_tcp, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req,
2241 						      tqpair);
2242 					/* No need to send r2t, contained in the capsuled data */
2243 					nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt,
2244 								  0, tcp_req->req.length);
2245 					nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD);
2246 				}
2247 				break;
2248 			}
2249 
2250 			nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE);
2251 			break;
2252 		case TCP_REQUEST_STATE_AWAITING_R2T_ACK:
2253 			spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 0, 0, (uintptr_t)tcp_req, 0);
2254 			/* The R2T completion or the h2c data incoming will kick it out of this state. */
2255 			break;
2256 		case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER:
2257 
2258 			spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 0, 0,
2259 					  (uintptr_t)tcp_req, 0);
2260 			/* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE
2261 			 * to escape this state. */
2262 			break;
2263 		case TCP_REQUEST_STATE_READY_TO_EXECUTE:
2264 			spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 0, 0, (uintptr_t)tcp_req, 0);
2265 
2266 			if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) {
2267 				assert(tcp_req->req.dif.elba_length >= tcp_req->req.length);
2268 				tcp_req->req.length = tcp_req->req.dif.elba_length;
2269 			}
2270 
2271 			nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING);
2272 			spdk_nvmf_request_exec(&tcp_req->req);
2273 			break;
2274 		case TCP_REQUEST_STATE_EXECUTING:
2275 			spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, 0, 0, (uintptr_t)tcp_req, 0);
2276 			/* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED
2277 			 * to escape this state. */
2278 			break;
2279 		case TCP_REQUEST_STATE_EXECUTED:
2280 			spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, 0, 0, (uintptr_t)tcp_req, 0);
2281 
2282 			if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) {
2283 				tcp_req->req.length = tcp_req->req.dif.orig_length;
2284 			}
2285 
2286 			nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE);
2287 			break;
2288 		case TCP_REQUEST_STATE_READY_TO_COMPLETE:
2289 			spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 0, 0, (uintptr_t)tcp_req, 0);
2290 			rc = request_transfer_out(&tcp_req->req);
2291 			assert(rc == 0); /* No good way to handle this currently */
2292 			break;
2293 		case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST:
2294 			spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 0, 0,
2295 					  (uintptr_t)tcp_req,
2296 					  0);
2297 			/* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED
2298 			 * to escape this state. */
2299 			break;
2300 		case TCP_REQUEST_STATE_COMPLETED:
2301 			spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, 0, 0, (uintptr_t)tcp_req, 0);
2302 			if (tcp_req->req.data_from_pool) {
2303 				spdk_nvmf_request_free_buffers(&tcp_req->req, group, transport);
2304 			}
2305 			tcp_req->req.length = 0;
2306 			tcp_req->req.iovcnt = 0;
2307 			tcp_req->req.data = NULL;
2308 
2309 			nvmf_tcp_req_pdu_fini(tcp_req);
2310 
2311 			nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_FREE);
2312 			break;
2313 		case TCP_REQUEST_NUM_STATES:
2314 		default:
2315 			assert(0);
2316 			break;
2317 		}
2318 
2319 		if (tcp_req->state != prev_state) {
2320 			progress = true;
2321 		}
2322 	} while (tcp_req->state != prev_state);
2323 
2324 	return progress;
2325 }
2326 
2327 static void
2328 nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock)
2329 {
2330 	struct spdk_nvmf_tcp_qpair *tqpair = arg;
2331 	int rc;
2332 
2333 	assert(tqpair != NULL);
2334 	rc = nvmf_tcp_sock_process(tqpair);
2335 
2336 	/* If there was a new socket error, disconnect */
2337 	if (rc < 0) {
2338 		nvmf_tcp_qpair_disconnect(tqpair);
2339 	}
2340 }
2341 
2342 static int
2343 nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group,
2344 			struct spdk_nvmf_qpair *qpair)
2345 {
2346 	struct spdk_nvmf_tcp_poll_group	*tgroup;
2347 	struct spdk_nvmf_tcp_qpair	*tqpair;
2348 	int				rc;
2349 
2350 	tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group);
2351 	tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
2352 
2353 	rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock,
2354 				      nvmf_tcp_sock_cb, tqpair);
2355 	if (rc != 0) {
2356 		SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n",
2357 			    spdk_strerror(errno), errno);
2358 		return -1;
2359 	}
2360 
2361 	rc =  nvmf_tcp_qpair_sock_init(tqpair);
2362 	if (rc != 0) {
2363 		SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair);
2364 		return -1;
2365 	}
2366 
2367 	rc = nvmf_tcp_qpair_init(&tqpair->qpair);
2368 	if (rc < 0) {
2369 		SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair);
2370 		return -1;
2371 	}
2372 
2373 	rc = nvmf_tcp_qpair_init_mem_resource(tqpair);
2374 	if (rc < 0) {
2375 		SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair);
2376 		return -1;
2377 	}
2378 
2379 	tqpair->group = tgroup;
2380 	tqpair->state = NVME_TCP_QPAIR_STATE_INVALID;
2381 	TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link);
2382 
2383 	return 0;
2384 }
2385 
2386 static int
2387 nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group,
2388 			   struct spdk_nvmf_qpair *qpair)
2389 {
2390 	struct spdk_nvmf_tcp_poll_group	*tgroup;
2391 	struct spdk_nvmf_tcp_qpair		*tqpair;
2392 	int				rc;
2393 
2394 	tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group);
2395 	tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
2396 
2397 	assert(tqpair->group == tgroup);
2398 
2399 	SPDK_DEBUGLOG(nvmf_tcp, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup);
2400 	if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) {
2401 		TAILQ_REMOVE(&tgroup->await_req, tqpair, link);
2402 	} else {
2403 		TAILQ_REMOVE(&tgroup->qpairs, tqpair, link);
2404 	}
2405 
2406 	rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock);
2407 	if (rc != 0) {
2408 		SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n",
2409 			    spdk_strerror(errno), errno);
2410 	}
2411 
2412 	return rc;
2413 }
2414 
2415 static int
2416 nvmf_tcp_req_complete(struct spdk_nvmf_request *req)
2417 {
2418 	struct spdk_nvmf_tcp_transport *ttransport;
2419 	struct spdk_nvmf_tcp_req *tcp_req;
2420 
2421 	ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport);
2422 	tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req);
2423 
2424 	nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED);
2425 	nvmf_tcp_req_process(ttransport, tcp_req);
2426 
2427 	return 0;
2428 }
2429 
2430 static void
2431 nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair)
2432 {
2433 	struct spdk_nvmf_tcp_qpair *tqpair;
2434 
2435 	SPDK_DEBUGLOG(nvmf_tcp, "Qpair: %p\n", qpair);
2436 
2437 	tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
2438 	tqpair->state = NVME_TCP_QPAIR_STATE_EXITED;
2439 	nvmf_tcp_qpair_destroy(tqpair);
2440 }
2441 
2442 static int
2443 nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group)
2444 {
2445 	struct spdk_nvmf_tcp_poll_group *tgroup;
2446 	int rc;
2447 	struct spdk_nvmf_request *req, *req_tmp;
2448 	struct spdk_nvmf_tcp_req *tcp_req;
2449 	struct spdk_nvmf_tcp_qpair *tqpair, *tqpair_tmp;
2450 	struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(group->transport,
2451 			struct spdk_nvmf_tcp_transport, transport);
2452 
2453 	tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group);
2454 
2455 	if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs) && TAILQ_EMPTY(&tgroup->await_req))) {
2456 		return 0;
2457 	}
2458 
2459 	STAILQ_FOREACH_SAFE(req, &group->pending_buf_queue, buf_link, req_tmp) {
2460 		tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req);
2461 		if (nvmf_tcp_req_process(ttransport, tcp_req) == false) {
2462 			break;
2463 		}
2464 	}
2465 
2466 	rc = spdk_sock_group_poll(tgroup->sock_group);
2467 	if (rc < 0) {
2468 		SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group);
2469 	}
2470 
2471 	TAILQ_FOREACH_SAFE(tqpair, &tgroup->await_req, link, tqpair_tmp) {
2472 		nvmf_tcp_sock_process(tqpair);
2473 	}
2474 
2475 	return rc;
2476 }
2477 
2478 static int
2479 nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair,
2480 			struct spdk_nvme_transport_id *trid, bool peer)
2481 {
2482 	struct spdk_nvmf_tcp_qpair     *tqpair;
2483 	uint16_t			port;
2484 
2485 	tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
2486 	spdk_nvme_trid_populate_transport(trid, SPDK_NVME_TRANSPORT_TCP);
2487 
2488 	if (peer) {
2489 		snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr);
2490 		port = tqpair->initiator_port;
2491 	} else {
2492 		snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr);
2493 		port = tqpair->target_port;
2494 	}
2495 
2496 	if (spdk_sock_is_ipv4(tqpair->sock)) {
2497 		trid->adrfam = SPDK_NVMF_ADRFAM_IPV4;
2498 	} else if (spdk_sock_is_ipv6(tqpair->sock)) {
2499 		trid->adrfam = SPDK_NVMF_ADRFAM_IPV6;
2500 	} else {
2501 		return -1;
2502 	}
2503 
2504 	snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port);
2505 	return 0;
2506 }
2507 
2508 static int
2509 nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair,
2510 			      struct spdk_nvme_transport_id *trid)
2511 {
2512 	return nvmf_tcp_qpair_get_trid(qpair, trid, 0);
2513 }
2514 
2515 static int
2516 nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair,
2517 			     struct spdk_nvme_transport_id *trid)
2518 {
2519 	return nvmf_tcp_qpair_get_trid(qpair, trid, 1);
2520 }
2521 
2522 static int
2523 nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair,
2524 			       struct spdk_nvme_transport_id *trid)
2525 {
2526 	return nvmf_tcp_qpair_get_trid(qpair, trid, 0);
2527 }
2528 
2529 static void
2530 nvmf_tcp_req_set_abort_status(struct spdk_nvmf_request *req,
2531 			      struct spdk_nvmf_tcp_req *tcp_req_to_abort)
2532 {
2533 	tcp_req_to_abort->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
2534 	tcp_req_to_abort->req.rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
2535 
2536 	nvmf_tcp_req_set_state(tcp_req_to_abort, TCP_REQUEST_STATE_READY_TO_COMPLETE);
2537 
2538 	req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command was successfully aborted. */
2539 }
2540 
2541 static int
2542 _nvmf_tcp_qpair_abort_request(void *ctx)
2543 {
2544 	struct spdk_nvmf_request *req = ctx;
2545 	struct spdk_nvmf_tcp_req *tcp_req_to_abort = SPDK_CONTAINEROF(req->req_to_abort,
2546 			struct spdk_nvmf_tcp_req, req);
2547 	struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(req->req_to_abort->qpair,
2548 					     struct spdk_nvmf_tcp_qpair, qpair);
2549 	int rc;
2550 
2551 	spdk_poller_unregister(&req->poller);
2552 
2553 	switch (tcp_req_to_abort->state) {
2554 	case TCP_REQUEST_STATE_EXECUTING:
2555 		rc = nvmf_ctrlr_abort_request(req);
2556 		if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS) {
2557 			return SPDK_POLLER_BUSY;
2558 		}
2559 		break;
2560 
2561 	case TCP_REQUEST_STATE_NEED_BUFFER:
2562 		STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue,
2563 			      &tcp_req_to_abort->req, spdk_nvmf_request, buf_link);
2564 
2565 		nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort);
2566 		break;
2567 
2568 	case TCP_REQUEST_STATE_AWAITING_R2T_ACK:
2569 		nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort);
2570 		break;
2571 
2572 	case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER:
2573 		if (spdk_get_ticks() < req->timeout_tsc) {
2574 			req->poller = SPDK_POLLER_REGISTER(_nvmf_tcp_qpair_abort_request, req, 0);
2575 			return SPDK_POLLER_BUSY;
2576 		}
2577 		break;
2578 
2579 	default:
2580 		break;
2581 	}
2582 
2583 	spdk_nvmf_request_complete(req);
2584 	return SPDK_POLLER_BUSY;
2585 }
2586 
2587 static void
2588 nvmf_tcp_qpair_abort_request(struct spdk_nvmf_qpair *qpair,
2589 			     struct spdk_nvmf_request *req)
2590 {
2591 	struct spdk_nvmf_tcp_qpair *tqpair;
2592 	struct spdk_nvmf_tcp_transport *ttransport;
2593 	struct spdk_nvmf_transport *transport;
2594 	uint16_t cid;
2595 	uint32_t i;
2596 	struct spdk_nvmf_tcp_req *tcp_req_to_abort = NULL;
2597 
2598 	tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
2599 	ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport);
2600 	transport = &ttransport->transport;
2601 
2602 	cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid;
2603 
2604 	for (i = 0; i < tqpair->resource_count; i++) {
2605 		tcp_req_to_abort = &tqpair->reqs[i];
2606 
2607 		if (tcp_req_to_abort->state != TCP_REQUEST_STATE_FREE &&
2608 		    tcp_req_to_abort->req.cmd->nvme_cmd.cid == cid) {
2609 			break;
2610 		}
2611 	}
2612 
2613 	if (tcp_req_to_abort == NULL) {
2614 		spdk_nvmf_request_complete(req);
2615 		return;
2616 	}
2617 
2618 	req->req_to_abort = &tcp_req_to_abort->req;
2619 	req->timeout_tsc = spdk_get_ticks() +
2620 			   transport->opts.abort_timeout_sec * spdk_get_ticks_hz();
2621 	req->poller = NULL;
2622 
2623 	_nvmf_tcp_qpair_abort_request(req);
2624 }
2625 
2626 #define SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH 128
2627 #define SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH 128
2628 #define SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR 128
2629 #define SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE 4096
2630 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE 131072
2631 #define SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE 131072
2632 #define SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS 511
2633 #define SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE 32
2634 #define SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP false
2635 #define SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC 1
2636 
2637 static void
2638 nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts)
2639 {
2640 	opts->max_queue_depth =		SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH;
2641 	opts->max_qpairs_per_ctrlr =	SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR;
2642 	opts->in_capsule_data_size =	SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE;
2643 	opts->max_io_size =		SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE;
2644 	opts->io_unit_size =		SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE;
2645 	opts->max_aq_depth =		SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH;
2646 	opts->num_shared_buffers =	SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS;
2647 	opts->buf_cache_size =		SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE;
2648 	opts->dif_insert_or_strip =	SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP;
2649 	opts->abort_timeout_sec =	SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC;
2650 	opts->transport_specific =      NULL;
2651 }
2652 
2653 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = {
2654 	.name = "TCP",
2655 	.type = SPDK_NVME_TRANSPORT_TCP,
2656 	.opts_init = nvmf_tcp_opts_init,
2657 	.create = nvmf_tcp_create,
2658 	.dump_opts = nvmf_tcp_dump_opts,
2659 	.destroy = nvmf_tcp_destroy,
2660 
2661 	.listen = nvmf_tcp_listen,
2662 	.stop_listen = nvmf_tcp_stop_listen,
2663 	.accept = nvmf_tcp_accept,
2664 
2665 	.listener_discover = nvmf_tcp_discover,
2666 
2667 	.poll_group_create = nvmf_tcp_poll_group_create,
2668 	.get_optimal_poll_group = nvmf_tcp_get_optimal_poll_group,
2669 	.poll_group_destroy = nvmf_tcp_poll_group_destroy,
2670 	.poll_group_add = nvmf_tcp_poll_group_add,
2671 	.poll_group_remove = nvmf_tcp_poll_group_remove,
2672 	.poll_group_poll = nvmf_tcp_poll_group_poll,
2673 
2674 	.req_free = nvmf_tcp_req_free,
2675 	.req_complete = nvmf_tcp_req_complete,
2676 
2677 	.qpair_fini = nvmf_tcp_close_qpair,
2678 	.qpair_get_local_trid = nvmf_tcp_qpair_get_local_trid,
2679 	.qpair_get_peer_trid = nvmf_tcp_qpair_get_peer_trid,
2680 	.qpair_get_listen_trid = nvmf_tcp_qpair_get_listen_trid,
2681 	.qpair_abort_request = nvmf_tcp_qpair_abort_request,
2682 };
2683 
2684 SPDK_NVMF_TRANSPORT_REGISTER(tcp, &spdk_nvmf_transport_tcp);
2685 SPDK_LOG_REGISTER_COMPONENT(nvmf_tcp)
2686