Line data Source code
1 : /* SPDX-License-Identifier: BSD-3-Clause
2 : * Copyright (C) 2018 Intel Corporation. All rights reserved.
3 : * Copyright (c) 2019, 2020 Mellanox Technologies LTD. All rights reserved.
4 : * Copyright (c) 2022-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
5 : */
6 :
7 : #include "spdk/accel.h"
8 : #include "spdk/stdinc.h"
9 : #include "spdk/crc32.h"
10 : #include "spdk/endian.h"
11 : #include "spdk/assert.h"
12 : #include "spdk/thread.h"
13 : #include "spdk/nvmf_transport.h"
14 : #include "spdk/string.h"
15 : #include "spdk/trace.h"
16 : #include "spdk/util.h"
17 : #include "spdk/log.h"
18 : #include "spdk/keyring.h"
19 :
20 : #include "spdk_internal/assert.h"
21 : #include "spdk_internal/nvme_tcp.h"
22 : #include "spdk_internal/sock.h"
23 :
24 : #include "nvmf_internal.h"
25 : #include "transport.h"
26 :
27 : #include "spdk_internal/trace_defs.h"
28 :
29 : #define NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME 16
30 : #define SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY 16
31 : #define SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY 0
32 : #define SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM 32
33 : #define SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION true
34 :
35 : #define SPDK_NVMF_TCP_MIN_IO_QUEUE_DEPTH 2
36 : #define SPDK_NVMF_TCP_MAX_IO_QUEUE_DEPTH 65535
37 : #define SPDK_NVMF_TCP_MIN_ADMIN_QUEUE_DEPTH 2
38 : #define SPDK_NVMF_TCP_MAX_ADMIN_QUEUE_DEPTH 4096
39 :
40 : #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH 128
41 : #define SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH 128
42 : #define SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR 128
43 : #define SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE 4096
44 : #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE 131072
45 : #define SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE 131072
46 : #define SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS 511
47 : #define SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE UINT32_MAX
48 : #define SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP false
49 : #define SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC 1
50 :
51 : #define TCP_PSK_INVALID_PERMISSIONS 0177
52 :
53 : const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp;
54 : static bool g_tls_log = false;
55 :
56 : /* spdk nvmf related structure */
57 : enum spdk_nvmf_tcp_req_state {
58 :
59 : /* The request is not currently in use */
60 : TCP_REQUEST_STATE_FREE = 0,
61 :
62 : /* Initial state when request first received */
63 : TCP_REQUEST_STATE_NEW = 1,
64 :
65 : /* The request is queued until a data buffer is available. */
66 : TCP_REQUEST_STATE_NEED_BUFFER = 2,
67 :
68 : /* The request has the data buffer available */
69 : TCP_REQUEST_STATE_HAVE_BUFFER = 3,
70 :
71 : /* The request is waiting for zcopy_start to finish */
72 : TCP_REQUEST_STATE_AWAITING_ZCOPY_START = 4,
73 :
74 : /* The request has received a zero-copy buffer */
75 : TCP_REQUEST_STATE_ZCOPY_START_COMPLETED = 5,
76 :
77 : /* The request is currently transferring data from the host to the controller. */
78 : TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER = 6,
79 :
80 : /* The request is waiting for the R2T send acknowledgement. */
81 : TCP_REQUEST_STATE_AWAITING_R2T_ACK = 7,
82 :
83 : /* The request is ready to execute at the block device */
84 : TCP_REQUEST_STATE_READY_TO_EXECUTE = 8,
85 :
86 : /* The request is currently executing at the block device */
87 : TCP_REQUEST_STATE_EXECUTING = 9,
88 :
89 : /* The request is waiting for zcopy buffers to be committed */
90 : TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT = 10,
91 :
92 : /* The request finished executing at the block device */
93 : TCP_REQUEST_STATE_EXECUTED = 11,
94 :
95 : /* The request is ready to send a completion */
96 : TCP_REQUEST_STATE_READY_TO_COMPLETE = 12,
97 :
98 : /* The request is currently transferring final pdus from the controller to the host. */
99 : TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST = 13,
100 :
101 : /* The request is waiting for zcopy buffers to be released (without committing) */
102 : TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE = 14,
103 :
104 : /* The request completed and can be marked free. */
105 : TCP_REQUEST_STATE_COMPLETED = 15,
106 :
107 : /* Terminator */
108 : TCP_REQUEST_NUM_STATES,
109 : };
110 :
111 : static const char *spdk_nvmf_tcp_term_req_fes_str[] = {
112 : "Invalid PDU Header Field",
113 : "PDU Sequence Error",
114 : "Header Digiest Error",
115 : "Data Transfer Out of Range",
116 : "R2T Limit Exceeded",
117 : "Unsupported parameter",
118 : };
119 :
120 1 : SPDK_TRACE_REGISTER_FN(nvmf_tcp_trace, "nvmf_tcp", TRACE_GROUP_NVMF_TCP)
121 : {
122 0 : spdk_trace_register_owner_type(OWNER_TYPE_NVMF_TCP, 't');
123 0 : spdk_trace_register_object(OBJECT_NVMF_TCP_IO, 'r');
124 0 : spdk_trace_register_description("TCP_REQ_NEW",
125 : TRACE_TCP_REQUEST_STATE_NEW,
126 : OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 1,
127 : SPDK_TRACE_ARG_TYPE_INT, "qd");
128 0 : spdk_trace_register_description("TCP_REQ_NEED_BUFFER",
129 : TRACE_TCP_REQUEST_STATE_NEED_BUFFER,
130 : OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0,
131 : SPDK_TRACE_ARG_TYPE_INT, "");
132 0 : spdk_trace_register_description("TCP_REQ_HAVE_BUFFER",
133 : TRACE_TCP_REQUEST_STATE_HAVE_BUFFER,
134 : OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0,
135 : SPDK_TRACE_ARG_TYPE_INT, "");
136 0 : spdk_trace_register_description("TCP_REQ_WAIT_ZCPY_START",
137 : TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_START,
138 : OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0,
139 : SPDK_TRACE_ARG_TYPE_INT, "");
140 0 : spdk_trace_register_description("TCP_REQ_ZCPY_START_CPL",
141 : TRACE_TCP_REQUEST_STATE_ZCOPY_START_COMPLETED,
142 : OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0,
143 : SPDK_TRACE_ARG_TYPE_INT, "");
144 0 : spdk_trace_register_description("TCP_REQ_TX_H_TO_C",
145 : TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER,
146 : OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0,
147 : SPDK_TRACE_ARG_TYPE_INT, "");
148 0 : spdk_trace_register_description("TCP_REQ_RDY_TO_EXECUTE",
149 : TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE,
150 : OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0,
151 : SPDK_TRACE_ARG_TYPE_INT, "");
152 0 : spdk_trace_register_description("TCP_REQ_EXECUTING",
153 : TRACE_TCP_REQUEST_STATE_EXECUTING,
154 : OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0,
155 : SPDK_TRACE_ARG_TYPE_INT, "");
156 0 : spdk_trace_register_description("TCP_REQ_WAIT_ZCPY_CMT",
157 : TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_COMMIT,
158 : OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0,
159 : SPDK_TRACE_ARG_TYPE_INT, "");
160 0 : spdk_trace_register_description("TCP_REQ_EXECUTED",
161 : TRACE_TCP_REQUEST_STATE_EXECUTED,
162 : OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0,
163 : SPDK_TRACE_ARG_TYPE_INT, "");
164 0 : spdk_trace_register_description("TCP_REQ_RDY_TO_COMPLETE",
165 : TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE,
166 : OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0,
167 : SPDK_TRACE_ARG_TYPE_INT, "");
168 0 : spdk_trace_register_description("TCP_REQ_TRANSFER_C2H",
169 : TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST,
170 : OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0,
171 : SPDK_TRACE_ARG_TYPE_INT, "");
172 0 : spdk_trace_register_description("TCP_REQ_AWAIT_ZCPY_RLS",
173 : TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_RELEASE,
174 : OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0,
175 : SPDK_TRACE_ARG_TYPE_INT, "");
176 0 : spdk_trace_register_description("TCP_REQ_COMPLETED",
177 : TRACE_TCP_REQUEST_STATE_COMPLETED,
178 : OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0,
179 : SPDK_TRACE_ARG_TYPE_INT, "qd");
180 0 : spdk_trace_register_description("TCP_READ_DONE",
181 : TRACE_TCP_READ_FROM_SOCKET_DONE,
182 : OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0,
183 : SPDK_TRACE_ARG_TYPE_INT, "");
184 0 : spdk_trace_register_description("TCP_REQ_AWAIT_R2T_ACK",
185 : TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK,
186 : OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0,
187 : SPDK_TRACE_ARG_TYPE_INT, "");
188 :
189 0 : spdk_trace_register_description("TCP_QP_CREATE", TRACE_TCP_QP_CREATE,
190 : OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0,
191 : SPDK_TRACE_ARG_TYPE_INT, "");
192 0 : spdk_trace_register_description("TCP_QP_SOCK_INIT", TRACE_TCP_QP_SOCK_INIT,
193 : OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0,
194 : SPDK_TRACE_ARG_TYPE_INT, "");
195 0 : spdk_trace_register_description("TCP_QP_STATE_CHANGE", TRACE_TCP_QP_STATE_CHANGE,
196 : OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0,
197 : SPDK_TRACE_ARG_TYPE_INT, "state");
198 0 : spdk_trace_register_description("TCP_QP_DISCONNECT", TRACE_TCP_QP_DISCONNECT,
199 : OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0,
200 : SPDK_TRACE_ARG_TYPE_INT, "");
201 0 : spdk_trace_register_description("TCP_QP_DESTROY", TRACE_TCP_QP_DESTROY,
202 : OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0,
203 : SPDK_TRACE_ARG_TYPE_INT, "");
204 0 : spdk_trace_register_description("TCP_QP_ABORT_REQ", TRACE_TCP_QP_ABORT_REQ,
205 : OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0,
206 : SPDK_TRACE_ARG_TYPE_INT, "");
207 0 : spdk_trace_register_description("TCP_QP_RCV_STATE_CHANGE", TRACE_TCP_QP_RCV_STATE_CHANGE,
208 : OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0,
209 : SPDK_TRACE_ARG_TYPE_INT, "state");
210 :
211 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_IO_START, OBJECT_NVMF_TCP_IO, 1);
212 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_IO_DONE, OBJECT_NVMF_TCP_IO, 0);
213 0 : spdk_trace_tpoint_register_relation(TRACE_SOCK_REQ_QUEUE, OBJECT_NVMF_TCP_IO, 0);
214 0 : spdk_trace_tpoint_register_relation(TRACE_SOCK_REQ_PEND, OBJECT_NVMF_TCP_IO, 0);
215 0 : spdk_trace_tpoint_register_relation(TRACE_SOCK_REQ_COMPLETE, OBJECT_NVMF_TCP_IO, 0);
216 0 : }
217 :
218 : struct spdk_nvmf_tcp_req {
219 : struct spdk_nvmf_request req;
220 : struct spdk_nvme_cpl rsp;
221 : struct spdk_nvme_cmd cmd;
222 :
223 : /* A PDU that can be used for sending responses. This is
224 : * not the incoming PDU! */
225 : struct nvme_tcp_pdu *pdu;
226 :
227 : /* In-capsule data buffer */
228 : uint8_t *buf;
229 :
230 : struct spdk_nvmf_tcp_req *fused_pair;
231 :
232 : /*
233 : * The PDU for a request may be used multiple times in serial over
234 : * the request's lifetime. For example, first to send an R2T, then
235 : * to send a completion. To catch mistakes where the PDU is used
236 : * twice at the same time, add a debug flag here for init/fini.
237 : */
238 : bool pdu_in_use;
239 : bool has_in_capsule_data;
240 : bool fused_failed;
241 :
242 : /* transfer_tag */
243 : uint16_t ttag;
244 :
245 : enum spdk_nvmf_tcp_req_state state;
246 :
247 : /*
248 : * h2c_offset is used when we receive the h2c_data PDU.
249 : */
250 : uint32_t h2c_offset;
251 :
252 : STAILQ_ENTRY(spdk_nvmf_tcp_req) link;
253 : TAILQ_ENTRY(spdk_nvmf_tcp_req) state_link;
254 : STAILQ_ENTRY(spdk_nvmf_tcp_req) control_msg_link;
255 : };
256 :
257 : struct spdk_nvmf_tcp_qpair {
258 : struct spdk_nvmf_qpair qpair;
259 : struct spdk_nvmf_tcp_poll_group *group;
260 : struct spdk_sock *sock;
261 :
262 : enum nvme_tcp_pdu_recv_state recv_state;
263 : enum nvme_tcp_qpair_state state;
264 :
265 : /* PDU being actively received */
266 : struct nvme_tcp_pdu *pdu_in_progress;
267 :
268 : struct spdk_nvmf_tcp_req *fused_first;
269 :
270 : /* Queues to track the requests in all states */
271 : TAILQ_HEAD(, spdk_nvmf_tcp_req) tcp_req_working_queue;
272 : TAILQ_HEAD(, spdk_nvmf_tcp_req) tcp_req_free_queue;
273 : SLIST_HEAD(, nvme_tcp_pdu) tcp_pdu_free_queue;
274 : /* Number of working pdus */
275 : uint32_t tcp_pdu_working_count;
276 :
277 : /* Number of requests in each state */
278 : uint32_t state_cntr[TCP_REQUEST_NUM_STATES];
279 :
280 : uint8_t cpda;
281 :
282 : bool host_hdgst_enable;
283 : bool host_ddgst_enable;
284 :
285 : /* This is a spare PDU used for sending special management
286 : * operations. Primarily, this is used for the initial
287 : * connection response and c2h termination request. */
288 : struct nvme_tcp_pdu *mgmt_pdu;
289 :
290 : /* Arrays of in-capsule buffers, requests, and pdus.
291 : * Each array is 'resource_count' number of elements */
292 : void *bufs;
293 : struct spdk_nvmf_tcp_req *reqs;
294 : struct nvme_tcp_pdu *pdus;
295 : uint32_t resource_count;
296 : uint32_t recv_buf_size;
297 :
298 : struct spdk_nvmf_tcp_port *port;
299 :
300 : /* IP address */
301 : char initiator_addr[SPDK_NVMF_TRADDR_MAX_LEN];
302 : char target_addr[SPDK_NVMF_TRADDR_MAX_LEN];
303 :
304 : /* IP port */
305 : uint16_t initiator_port;
306 : uint16_t target_port;
307 :
308 : /* Wait until the host terminates the connection (e.g. after sending C2HTermReq) */
309 : bool wait_terminate;
310 :
311 : /* Timer used to destroy qpair after detecting transport error issue if initiator does
312 : * not close the connection.
313 : */
314 : struct spdk_poller *timeout_poller;
315 :
316 : spdk_nvmf_transport_qpair_fini_cb fini_cb_fn;
317 : void *fini_cb_arg;
318 :
319 : TAILQ_ENTRY(spdk_nvmf_tcp_qpair) link;
320 : };
321 :
322 : struct spdk_nvmf_tcp_control_msg {
323 : STAILQ_ENTRY(spdk_nvmf_tcp_control_msg) link;
324 : };
325 :
326 : struct spdk_nvmf_tcp_control_msg_list {
327 : void *msg_buf;
328 : STAILQ_HEAD(, spdk_nvmf_tcp_control_msg) free_msgs;
329 : STAILQ_HEAD(, spdk_nvmf_tcp_req) waiting_for_msg_reqs;
330 : };
331 :
332 : struct spdk_nvmf_tcp_poll_group {
333 : struct spdk_nvmf_transport_poll_group group;
334 : struct spdk_sock_group *sock_group;
335 :
336 : TAILQ_HEAD(, spdk_nvmf_tcp_qpair) qpairs;
337 : TAILQ_HEAD(, spdk_nvmf_tcp_qpair) await_req;
338 :
339 : struct spdk_io_channel *accel_channel;
340 : struct spdk_nvmf_tcp_control_msg_list *control_msg_list;
341 :
342 : TAILQ_ENTRY(spdk_nvmf_tcp_poll_group) link;
343 : };
344 :
345 : struct spdk_nvmf_tcp_port {
346 : const struct spdk_nvme_transport_id *trid;
347 : struct spdk_sock *listen_sock;
348 : struct spdk_nvmf_transport *transport;
349 : TAILQ_ENTRY(spdk_nvmf_tcp_port) link;
350 : };
351 :
352 : struct tcp_transport_opts {
353 : bool c2h_success;
354 : uint16_t control_msg_num;
355 : uint32_t sock_priority;
356 : };
357 :
358 : struct tcp_psk_entry {
359 : char hostnqn[SPDK_NVMF_NQN_MAX_LEN + 1];
360 : char subnqn[SPDK_NVMF_NQN_MAX_LEN + 1];
361 : char pskid[NVMF_PSK_IDENTITY_LEN];
362 : uint8_t psk[SPDK_TLS_PSK_MAX_LEN];
363 : struct spdk_key *key;
364 :
365 : /* Original path saved to emit SPDK configuration via "save_config". */
366 : char psk_path[PATH_MAX];
367 : uint32_t psk_size;
368 : enum nvme_tcp_cipher_suite tls_cipher_suite;
369 : TAILQ_ENTRY(tcp_psk_entry) link;
370 : };
371 :
372 : struct spdk_nvmf_tcp_transport {
373 : struct spdk_nvmf_transport transport;
374 : struct tcp_transport_opts tcp_opts;
375 : uint32_t ack_timeout;
376 :
377 : struct spdk_nvmf_tcp_poll_group *next_pg;
378 :
379 : struct spdk_poller *accept_poller;
380 : struct spdk_sock_group *listen_sock_group;
381 :
382 : TAILQ_HEAD(, spdk_nvmf_tcp_port) ports;
383 : TAILQ_HEAD(, spdk_nvmf_tcp_poll_group) poll_groups;
384 :
385 : TAILQ_HEAD(, tcp_psk_entry) psks;
386 : };
387 :
388 : static const struct spdk_json_object_decoder tcp_transport_opts_decoder[] = {
389 : {
390 : "c2h_success", offsetof(struct tcp_transport_opts, c2h_success),
391 : spdk_json_decode_bool, true
392 : },
393 : {
394 : "control_msg_num", offsetof(struct tcp_transport_opts, control_msg_num),
395 : spdk_json_decode_uint16, true
396 : },
397 : {
398 : "sock_priority", offsetof(struct tcp_transport_opts, sock_priority),
399 : spdk_json_decode_uint32, true
400 : },
401 : };
402 :
403 : static bool nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport,
404 : struct spdk_nvmf_tcp_req *tcp_req);
405 : static void nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group);
406 :
407 : static void _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair,
408 : struct spdk_nvmf_tcp_req *tcp_req);
409 :
410 : static inline void
411 10 : nvmf_tcp_req_set_state(struct spdk_nvmf_tcp_req *tcp_req,
412 : enum spdk_nvmf_tcp_req_state state)
413 : {
414 : struct spdk_nvmf_qpair *qpair;
415 : struct spdk_nvmf_tcp_qpair *tqpair;
416 :
417 10 : qpair = tcp_req->req.qpair;
418 10 : tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
419 :
420 10 : assert(tqpair->state_cntr[tcp_req->state] > 0);
421 10 : tqpair->state_cntr[tcp_req->state]--;
422 10 : tqpair->state_cntr[state]++;
423 :
424 10 : tcp_req->state = state;
425 10 : }
426 :
427 : static inline struct nvme_tcp_pdu *
428 7 : nvmf_tcp_req_pdu_init(struct spdk_nvmf_tcp_req *tcp_req)
429 : {
430 7 : assert(tcp_req->pdu_in_use == false);
431 :
432 7 : memset(tcp_req->pdu, 0, sizeof(*tcp_req->pdu));
433 7 : tcp_req->pdu->qpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair);
434 :
435 7 : return tcp_req->pdu;
436 : }
437 :
438 : static struct spdk_nvmf_tcp_req *
439 1 : nvmf_tcp_req_get(struct spdk_nvmf_tcp_qpair *tqpair)
440 : {
441 : struct spdk_nvmf_tcp_req *tcp_req;
442 :
443 1 : tcp_req = TAILQ_FIRST(&tqpair->tcp_req_free_queue);
444 1 : if (spdk_unlikely(!tcp_req)) {
445 0 : return NULL;
446 : }
447 :
448 1 : memset(&tcp_req->rsp, 0, sizeof(tcp_req->rsp));
449 1 : tcp_req->h2c_offset = 0;
450 1 : tcp_req->has_in_capsule_data = false;
451 1 : tcp_req->req.dif_enabled = false;
452 1 : tcp_req->req.zcopy_phase = NVMF_ZCOPY_PHASE_NONE;
453 :
454 1 : TAILQ_REMOVE(&tqpair->tcp_req_free_queue, tcp_req, state_link);
455 1 : TAILQ_INSERT_TAIL(&tqpair->tcp_req_working_queue, tcp_req, state_link);
456 1 : tqpair->qpair.queue_depth++;
457 1 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW);
458 1 : return tcp_req;
459 : }
460 :
461 : static inline void
462 0 : nvmf_tcp_req_put(struct spdk_nvmf_tcp_qpair *tqpair, struct spdk_nvmf_tcp_req *tcp_req)
463 : {
464 0 : assert(!tcp_req->pdu_in_use);
465 :
466 0 : TAILQ_REMOVE(&tqpair->tcp_req_working_queue, tcp_req, state_link);
467 0 : TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link);
468 0 : tqpair->qpair.queue_depth--;
469 0 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_FREE);
470 0 : }
471 :
472 : static void
473 1 : nvmf_tcp_req_get_buffers_done(struct spdk_nvmf_request *req)
474 : {
475 : struct spdk_nvmf_tcp_req *tcp_req;
476 : struct spdk_nvmf_transport *transport;
477 : struct spdk_nvmf_tcp_transport *ttransport;
478 :
479 1 : tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req);
480 1 : transport = req->qpair->transport;
481 1 : ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport);
482 :
483 1 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_HAVE_BUFFER);
484 1 : nvmf_tcp_req_process(ttransport, tcp_req);
485 1 : }
486 :
487 : static void
488 0 : nvmf_tcp_request_free(void *cb_arg)
489 : {
490 : struct spdk_nvmf_tcp_transport *ttransport;
491 0 : struct spdk_nvmf_tcp_req *tcp_req = cb_arg;
492 :
493 0 : assert(tcp_req != NULL);
494 :
495 0 : SPDK_DEBUGLOG(nvmf_tcp, "tcp_req=%p will be freed\n", tcp_req);
496 0 : ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport,
497 : struct spdk_nvmf_tcp_transport, transport);
498 0 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED);
499 0 : nvmf_tcp_req_process(ttransport, tcp_req);
500 0 : }
501 :
502 : static int
503 0 : nvmf_tcp_req_free(struct spdk_nvmf_request *req)
504 : {
505 0 : struct spdk_nvmf_tcp_req *tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req);
506 :
507 0 : nvmf_tcp_request_free(tcp_req);
508 :
509 0 : return 0;
510 : }
511 :
512 : static void
513 6 : nvmf_tcp_drain_state_queue(struct spdk_nvmf_tcp_qpair *tqpair,
514 : enum spdk_nvmf_tcp_req_state state)
515 : {
516 : struct spdk_nvmf_tcp_req *tcp_req, *req_tmp;
517 :
518 6 : assert(state != TCP_REQUEST_STATE_FREE);
519 6 : TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) {
520 0 : if (state == tcp_req->state) {
521 0 : nvmf_tcp_request_free(tcp_req);
522 : }
523 : }
524 6 : }
525 :
526 : static inline void
527 0 : nvmf_tcp_request_get_buffers_abort(struct spdk_nvmf_tcp_req *tcp_req)
528 : {
529 : /* Request can wait either for the iobuf or control_msg */
530 0 : struct spdk_nvmf_poll_group *group = tcp_req->req.qpair->group;
531 0 : struct spdk_nvmf_transport *transport = tcp_req->req.qpair->transport;
532 0 : struct spdk_nvmf_transport_poll_group *tgroup = nvmf_get_transport_poll_group(group, transport);
533 0 : struct spdk_nvmf_tcp_poll_group *tcp_group = SPDK_CONTAINEROF(tgroup,
534 : struct spdk_nvmf_tcp_poll_group, group);
535 : struct spdk_nvmf_tcp_req *tmp_req, *abort_req;
536 :
537 0 : assert(tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER);
538 :
539 0 : STAILQ_FOREACH_SAFE(abort_req, &tcp_group->control_msg_list->waiting_for_msg_reqs, control_msg_link,
540 : tmp_req) {
541 0 : if (abort_req == tcp_req) {
542 0 : STAILQ_REMOVE(&tcp_group->control_msg_list->waiting_for_msg_reqs, abort_req, spdk_nvmf_tcp_req,
543 : control_msg_link);
544 0 : return;
545 : }
546 : }
547 :
548 0 : if (!nvmf_request_get_buffers_abort(&tcp_req->req)) {
549 0 : SPDK_ERRLOG("Failed to abort tcp_req=%p\n", tcp_req);
550 0 : assert(0 && "Should never happen");
551 : }
552 : }
553 :
554 : static void
555 1 : nvmf_tcp_cleanup_all_states(struct spdk_nvmf_tcp_qpair *tqpair)
556 : {
557 : struct spdk_nvmf_tcp_req *tcp_req, *req_tmp;
558 :
559 1 : nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST);
560 1 : nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEW);
561 :
562 : /* Wipe the requests waiting for buffer from the waiting list */
563 1 : TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) {
564 0 : if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) {
565 0 : nvmf_tcp_request_get_buffers_abort(tcp_req);
566 : }
567 : }
568 :
569 1 : nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEED_BUFFER);
570 1 : nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_EXECUTING);
571 1 : nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER);
572 1 : nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK);
573 1 : }
574 :
575 : static void
576 0 : nvmf_tcp_dump_qpair_req_contents(struct spdk_nvmf_tcp_qpair *tqpair)
577 : {
578 : int i;
579 : struct spdk_nvmf_tcp_req *tcp_req;
580 :
581 0 : SPDK_ERRLOG("Dumping contents of queue pair (QID %d)\n", tqpair->qpair.qid);
582 0 : for (i = 1; i < TCP_REQUEST_NUM_STATES; i++) {
583 0 : SPDK_ERRLOG("\tNum of requests in state[%d] = %u\n", i, tqpair->state_cntr[i]);
584 0 : TAILQ_FOREACH(tcp_req, &tqpair->tcp_req_working_queue, state_link) {
585 0 : if ((int)tcp_req->state == i) {
586 0 : SPDK_ERRLOG("\t\tRequest Data From Pool: %d\n", tcp_req->req.data_from_pool);
587 0 : SPDK_ERRLOG("\t\tRequest opcode: %d\n", tcp_req->req.cmd->nvmf_cmd.opcode);
588 : }
589 : }
590 : }
591 0 : }
592 :
593 : static void
594 1 : _nvmf_tcp_qpair_destroy(void *_tqpair)
595 : {
596 1 : struct spdk_nvmf_tcp_qpair *tqpair = _tqpair;
597 1 : spdk_nvmf_transport_qpair_fini_cb cb_fn = tqpair->fini_cb_fn;
598 1 : void *cb_arg = tqpair->fini_cb_arg;
599 1 : int err = 0;
600 :
601 1 : spdk_trace_record(TRACE_TCP_QP_DESTROY, tqpair->qpair.trace_id, 0, 0);
602 :
603 1 : SPDK_DEBUGLOG(nvmf_tcp, "enter\n");
604 :
605 1 : err = spdk_sock_close(&tqpair->sock);
606 1 : assert(err == 0);
607 1 : nvmf_tcp_cleanup_all_states(tqpair);
608 :
609 1 : if (tqpair->state_cntr[TCP_REQUEST_STATE_FREE] != tqpair->resource_count) {
610 0 : SPDK_ERRLOG("tqpair(%p) free tcp request num is %u but should be %u\n", tqpair,
611 : tqpair->state_cntr[TCP_REQUEST_STATE_FREE],
612 : tqpair->resource_count);
613 0 : err++;
614 : }
615 :
616 1 : if (err > 0) {
617 0 : nvmf_tcp_dump_qpair_req_contents(tqpair);
618 : }
619 :
620 : /* The timeout poller might still be registered here if we close the qpair before host
621 : * terminates the connection.
622 : */
623 1 : spdk_poller_unregister(&tqpair->timeout_poller);
624 1 : spdk_dma_free(tqpair->pdus);
625 1 : free(tqpair->reqs);
626 1 : spdk_free(tqpair->bufs);
627 1 : spdk_trace_unregister_owner(tqpair->qpair.trace_id);
628 1 : free(tqpair);
629 :
630 1 : if (cb_fn != NULL) {
631 0 : cb_fn(cb_arg);
632 : }
633 :
634 1 : SPDK_DEBUGLOG(nvmf_tcp, "Leave\n");
635 1 : }
636 :
637 : static void
638 1 : nvmf_tcp_qpair_destroy(struct spdk_nvmf_tcp_qpair *tqpair)
639 : {
640 : /* Delay the destruction to make sure it isn't performed from the context of a sock
641 : * callback. Otherwise, spdk_sock_close() might not abort pending requests, causing their
642 : * completions to be executed after the qpair is freed. (Note: this fixed issue #2471.)
643 : */
644 1 : spdk_thread_send_msg(spdk_get_thread(), _nvmf_tcp_qpair_destroy, tqpair);
645 1 : }
646 :
647 : static void
648 0 : nvmf_tcp_dump_opts(struct spdk_nvmf_transport *transport, struct spdk_json_write_ctx *w)
649 : {
650 : struct spdk_nvmf_tcp_transport *ttransport;
651 0 : assert(w != NULL);
652 :
653 0 : ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport);
654 0 : spdk_json_write_named_bool(w, "c2h_success", ttransport->tcp_opts.c2h_success);
655 0 : spdk_json_write_named_uint32(w, "sock_priority", ttransport->tcp_opts.sock_priority);
656 0 : }
657 :
658 : static void
659 1 : nvmf_tcp_free_psk_entry(struct tcp_psk_entry *entry)
660 : {
661 1 : if (entry == NULL) {
662 0 : return;
663 : }
664 :
665 1 : spdk_memset_s(entry->psk, sizeof(entry->psk), 0, sizeof(entry->psk));
666 1 : spdk_keyring_put_key(entry->key);
667 1 : free(entry);
668 : }
669 :
670 : static int
671 5 : nvmf_tcp_destroy(struct spdk_nvmf_transport *transport,
672 : spdk_nvmf_transport_destroy_done_cb cb_fn, void *cb_arg)
673 : {
674 : struct spdk_nvmf_tcp_transport *ttransport;
675 : struct tcp_psk_entry *entry, *tmp;
676 :
677 5 : assert(transport != NULL);
678 5 : ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport);
679 :
680 5 : TAILQ_FOREACH_SAFE(entry, &ttransport->psks, link, tmp) {
681 0 : TAILQ_REMOVE(&ttransport->psks, entry, link);
682 0 : nvmf_tcp_free_psk_entry(entry);
683 : }
684 :
685 5 : spdk_poller_unregister(&ttransport->accept_poller);
686 5 : spdk_sock_group_close(&ttransport->listen_sock_group);
687 5 : free(ttransport);
688 :
689 5 : if (cb_fn) {
690 0 : cb_fn(cb_arg);
691 : }
692 5 : return 0;
693 : }
694 :
695 : static int nvmf_tcp_accept(void *ctx);
696 :
697 : static void nvmf_tcp_accept_cb(void *ctx, struct spdk_sock_group *group, struct spdk_sock *sock);
698 :
699 : static struct spdk_nvmf_transport *
700 6 : nvmf_tcp_create(struct spdk_nvmf_transport_opts *opts)
701 : {
702 : struct spdk_nvmf_tcp_transport *ttransport;
703 : uint32_t sge_count;
704 : uint32_t min_shared_buffers;
705 :
706 6 : ttransport = calloc(1, sizeof(*ttransport));
707 6 : if (!ttransport) {
708 0 : return NULL;
709 : }
710 :
711 6 : TAILQ_INIT(&ttransport->ports);
712 6 : TAILQ_INIT(&ttransport->poll_groups);
713 6 : TAILQ_INIT(&ttransport->psks);
714 :
715 6 : ttransport->transport.ops = &spdk_nvmf_transport_tcp;
716 :
717 6 : ttransport->tcp_opts.c2h_success = SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION;
718 6 : ttransport->tcp_opts.sock_priority = SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY;
719 6 : ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM;
720 6 : if (opts->transport_specific != NULL &&
721 0 : spdk_json_decode_object_relaxed(opts->transport_specific, tcp_transport_opts_decoder,
722 : SPDK_COUNTOF(tcp_transport_opts_decoder),
723 0 : &ttransport->tcp_opts)) {
724 0 : SPDK_ERRLOG("spdk_json_decode_object_relaxed failed\n");
725 0 : free(ttransport);
726 0 : return NULL;
727 : }
728 :
729 6 : SPDK_NOTICELOG("*** TCP Transport Init ***\n");
730 :
731 6 : SPDK_INFOLOG(nvmf_tcp, "*** TCP Transport Init ***\n"
732 : " Transport opts: max_ioq_depth=%d, max_io_size=%d,\n"
733 : " max_io_qpairs_per_ctrlr=%d, io_unit_size=%d,\n"
734 : " in_capsule_data_size=%d, max_aq_depth=%d\n"
735 : " num_shared_buffers=%d, c2h_success=%d,\n"
736 : " dif_insert_or_strip=%d, sock_priority=%d\n"
737 : " abort_timeout_sec=%d, control_msg_num=%hu\n"
738 : " ack_timeout=%d\n",
739 : opts->max_queue_depth,
740 : opts->max_io_size,
741 : opts->max_qpairs_per_ctrlr - 1,
742 : opts->io_unit_size,
743 : opts->in_capsule_data_size,
744 : opts->max_aq_depth,
745 : opts->num_shared_buffers,
746 : ttransport->tcp_opts.c2h_success,
747 : opts->dif_insert_or_strip,
748 : ttransport->tcp_opts.sock_priority,
749 : opts->abort_timeout_sec,
750 : ttransport->tcp_opts.control_msg_num,
751 : opts->ack_timeout);
752 :
753 6 : if (ttransport->tcp_opts.sock_priority > SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY) {
754 0 : SPDK_ERRLOG("Unsupported socket_priority=%d, the current range is: 0 to %d\n"
755 : "you can use man 7 socket to view the range of priority under SO_PRIORITY item\n",
756 : ttransport->tcp_opts.sock_priority, SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY);
757 0 : free(ttransport);
758 0 : return NULL;
759 : }
760 :
761 6 : if (ttransport->tcp_opts.control_msg_num == 0 &&
762 0 : opts->in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) {
763 0 : SPDK_WARNLOG("TCP param control_msg_num can't be 0 if ICD is less than %u bytes. Using default value %u\n",
764 : SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM);
765 0 : ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM;
766 : }
767 :
768 : /* I/O unit size cannot be larger than max I/O size */
769 6 : if (opts->io_unit_size > opts->max_io_size) {
770 1 : SPDK_WARNLOG("TCP param io_unit_size %u can't be larger than max_io_size %u. Using max_io_size as io_unit_size\n",
771 : opts->io_unit_size, opts->max_io_size);
772 1 : opts->io_unit_size = opts->max_io_size;
773 : }
774 :
775 : /* In capsule data size cannot be larger than max I/O size */
776 6 : if (opts->in_capsule_data_size > opts->max_io_size) {
777 0 : SPDK_WARNLOG("TCP param ICD size %u can't be larger than max_io_size %u. Using max_io_size as ICD size\n",
778 : opts->io_unit_size, opts->max_io_size);
779 0 : opts->in_capsule_data_size = opts->max_io_size;
780 : }
781 :
782 : /* max IO queue depth cannot be smaller than 2 or larger than 65535.
783 : * We will not check SPDK_NVMF_TCP_MAX_IO_QUEUE_DEPTH, because max_queue_depth is 16bits and always not larger than 64k. */
784 6 : if (opts->max_queue_depth < SPDK_NVMF_TCP_MIN_IO_QUEUE_DEPTH) {
785 0 : SPDK_WARNLOG("TCP param max_queue_depth %u can't be smaller than %u or larger than %u. Using default value %u\n",
786 : opts->max_queue_depth, SPDK_NVMF_TCP_MIN_IO_QUEUE_DEPTH,
787 : SPDK_NVMF_TCP_MAX_IO_QUEUE_DEPTH, SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH);
788 0 : opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH;
789 : }
790 :
791 : /* max admin queue depth cannot be smaller than 2 or larger than 4096 */
792 6 : if (opts->max_aq_depth < SPDK_NVMF_TCP_MIN_ADMIN_QUEUE_DEPTH ||
793 6 : opts->max_aq_depth > SPDK_NVMF_TCP_MAX_ADMIN_QUEUE_DEPTH) {
794 0 : SPDK_WARNLOG("TCP param max_aq_depth %u can't be smaller than %u or larger than %u. Using default value %u\n",
795 : opts->max_aq_depth, SPDK_NVMF_TCP_MIN_ADMIN_QUEUE_DEPTH,
796 : SPDK_NVMF_TCP_MAX_ADMIN_QUEUE_DEPTH, SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH);
797 0 : opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH;
798 : }
799 :
800 6 : sge_count = opts->max_io_size / opts->io_unit_size;
801 6 : if (sge_count > SPDK_NVMF_MAX_SGL_ENTRIES) {
802 1 : SPDK_ERRLOG("Unsupported IO Unit size specified, %d bytes\n", opts->io_unit_size);
803 1 : free(ttransport);
804 1 : return NULL;
805 : }
806 :
807 : /* If buf_cache_size == UINT32_MAX, we will dynamically pick a cache size later that we know will fit. */
808 5 : if (opts->buf_cache_size < UINT32_MAX) {
809 5 : min_shared_buffers = spdk_env_get_core_count() * opts->buf_cache_size;
810 5 : if (min_shared_buffers > opts->num_shared_buffers) {
811 0 : SPDK_ERRLOG("There are not enough buffers to satisfy "
812 : "per-poll group caches for each thread. (%" PRIu32 ") "
813 : "supplied. (%" PRIu32 ") required\n", opts->num_shared_buffers, min_shared_buffers);
814 0 : SPDK_ERRLOG("Please specify a larger number of shared buffers\n");
815 0 : free(ttransport);
816 0 : return NULL;
817 : }
818 : }
819 :
820 5 : ttransport->accept_poller = SPDK_POLLER_REGISTER(nvmf_tcp_accept, &ttransport->transport,
821 : opts->acceptor_poll_rate);
822 5 : if (!ttransport->accept_poller) {
823 0 : free(ttransport);
824 0 : return NULL;
825 : }
826 :
827 5 : ttransport->listen_sock_group = spdk_sock_group_create(NULL);
828 5 : if (ttransport->listen_sock_group == NULL) {
829 0 : SPDK_ERRLOG("Failed to create socket group for listen sockets\n");
830 0 : spdk_poller_unregister(&ttransport->accept_poller);
831 0 : free(ttransport);
832 0 : return NULL;
833 : }
834 :
835 5 : return &ttransport->transport;
836 : }
837 :
838 : static int
839 0 : nvmf_tcp_trsvcid_to_int(const char *trsvcid)
840 : {
841 : unsigned long long ull;
842 0 : char *end = NULL;
843 :
844 0 : ull = strtoull(trsvcid, &end, 10);
845 0 : if (end == NULL || end == trsvcid || *end != '\0') {
846 0 : return -1;
847 : }
848 :
849 : /* Valid TCP/IP port numbers are in [1, 65535] */
850 0 : if (ull == 0 || ull > 65535) {
851 0 : return -1;
852 : }
853 :
854 0 : return (int)ull;
855 : }
856 :
857 : /**
858 : * Canonicalize a listen address trid.
859 : */
860 : static int
861 0 : nvmf_tcp_canon_listen_trid(struct spdk_nvme_transport_id *canon_trid,
862 : const struct spdk_nvme_transport_id *trid)
863 : {
864 : int trsvcid_int;
865 :
866 0 : trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid);
867 0 : if (trsvcid_int < 0) {
868 0 : return -EINVAL;
869 : }
870 :
871 0 : memset(canon_trid, 0, sizeof(*canon_trid));
872 0 : spdk_nvme_trid_populate_transport(canon_trid, SPDK_NVME_TRANSPORT_TCP);
873 0 : canon_trid->adrfam = trid->adrfam;
874 0 : snprintf(canon_trid->traddr, sizeof(canon_trid->traddr), "%s", trid->traddr);
875 0 : snprintf(canon_trid->trsvcid, sizeof(canon_trid->trsvcid), "%d", trsvcid_int);
876 :
877 0 : return 0;
878 : }
879 :
880 : /**
881 : * Find an existing listening port.
882 : */
883 : static struct spdk_nvmf_tcp_port *
884 0 : nvmf_tcp_find_port(struct spdk_nvmf_tcp_transport *ttransport,
885 : const struct spdk_nvme_transport_id *trid)
886 : {
887 0 : struct spdk_nvme_transport_id canon_trid;
888 : struct spdk_nvmf_tcp_port *port;
889 :
890 0 : if (nvmf_tcp_canon_listen_trid(&canon_trid, trid) != 0) {
891 0 : return NULL;
892 : }
893 :
894 0 : TAILQ_FOREACH(port, &ttransport->ports, link) {
895 0 : if (spdk_nvme_transport_id_compare(&canon_trid, port->trid) == 0) {
896 0 : return port;
897 : }
898 : }
899 :
900 0 : return NULL;
901 : }
902 :
903 : static int
904 0 : tcp_sock_get_key(uint8_t *out, int out_len, const char **cipher, const char *pskid,
905 : void *get_key_ctx)
906 : {
907 : struct tcp_psk_entry *entry;
908 0 : struct spdk_nvmf_tcp_transport *ttransport = get_key_ctx;
909 : size_t psk_len;
910 : int rc;
911 :
912 0 : TAILQ_FOREACH(entry, &ttransport->psks, link) {
913 0 : if (strcmp(pskid, entry->pskid) != 0) {
914 0 : continue;
915 : }
916 :
917 0 : psk_len = entry->psk_size;
918 0 : if ((size_t)out_len < psk_len) {
919 0 : SPDK_ERRLOG("Out buffer of size: %" PRIu32 " cannot fit PSK of len: %lu\n",
920 : out_len, psk_len);
921 0 : return -ENOBUFS;
922 : }
923 :
924 : /* Convert PSK to the TLS PSK format. */
925 0 : rc = nvme_tcp_derive_tls_psk(entry->psk, psk_len, pskid, out, out_len,
926 : entry->tls_cipher_suite);
927 0 : if (rc < 0) {
928 0 : SPDK_ERRLOG("Could not generate TLS PSK\n");
929 : }
930 :
931 0 : switch (entry->tls_cipher_suite) {
932 0 : case NVME_TCP_CIPHER_AES_128_GCM_SHA256:
933 0 : *cipher = "TLS_AES_128_GCM_SHA256";
934 0 : break;
935 0 : case NVME_TCP_CIPHER_AES_256_GCM_SHA384:
936 0 : *cipher = "TLS_AES_256_GCM_SHA384";
937 0 : break;
938 0 : default:
939 0 : *cipher = NULL;
940 0 : return -ENOTSUP;
941 : }
942 :
943 0 : return rc;
944 : }
945 :
946 0 : SPDK_ERRLOG("Could not find PSK for identity: %s\n", pskid);
947 :
948 0 : return -EINVAL;
949 : }
950 :
951 : static int
952 0 : nvmf_tcp_listen(struct spdk_nvmf_transport *transport, const struct spdk_nvme_transport_id *trid,
953 : struct spdk_nvmf_listen_opts *listen_opts)
954 : {
955 : struct spdk_nvmf_tcp_transport *ttransport;
956 : struct spdk_nvmf_tcp_port *port;
957 : int trsvcid_int;
958 : uint8_t adrfam;
959 : const char *sock_impl_name;
960 0 : struct spdk_sock_impl_opts impl_opts;
961 0 : size_t impl_opts_size = sizeof(impl_opts);
962 0 : struct spdk_sock_opts opts;
963 : int rc;
964 :
965 0 : if (!strlen(trid->trsvcid)) {
966 0 : SPDK_ERRLOG("Service id is required\n");
967 0 : return -EINVAL;
968 : }
969 :
970 0 : ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport);
971 :
972 0 : trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid);
973 0 : if (trsvcid_int < 0) {
974 0 : SPDK_ERRLOG("Invalid trsvcid '%s'\n", trid->trsvcid);
975 0 : return -EINVAL;
976 : }
977 :
978 0 : port = calloc(1, sizeof(*port));
979 0 : if (!port) {
980 0 : SPDK_ERRLOG("Port allocation failed\n");
981 0 : return -ENOMEM;
982 : }
983 :
984 0 : port->trid = trid;
985 :
986 0 : sock_impl_name = NULL;
987 :
988 0 : opts.opts_size = sizeof(opts);
989 0 : spdk_sock_get_default_opts(&opts);
990 0 : opts.priority = ttransport->tcp_opts.sock_priority;
991 0 : opts.ack_timeout = transport->opts.ack_timeout;
992 0 : if (listen_opts->secure_channel) {
993 0 : if (listen_opts->sock_impl &&
994 0 : strncmp("ssl", listen_opts->sock_impl, strlen(listen_opts->sock_impl))) {
995 0 : SPDK_ERRLOG("Enabling secure_channel while specifying a sock_impl different from 'ssl' is unsupported");
996 0 : free(port);
997 0 : return -EINVAL;
998 : }
999 0 : listen_opts->sock_impl = "ssl";
1000 : }
1001 :
1002 0 : if (listen_opts->sock_impl) {
1003 0 : sock_impl_name = listen_opts->sock_impl;
1004 0 : spdk_sock_impl_get_opts(sock_impl_name, &impl_opts, &impl_opts_size);
1005 :
1006 0 : if (!strncmp("ssl", sock_impl_name, strlen(sock_impl_name))) {
1007 0 : if (!g_tls_log) {
1008 0 : SPDK_NOTICELOG("TLS support is considered experimental\n");
1009 0 : g_tls_log = true;
1010 : }
1011 0 : impl_opts.tls_version = SPDK_TLS_VERSION_1_3;
1012 0 : impl_opts.get_key = tcp_sock_get_key;
1013 0 : impl_opts.get_key_ctx = ttransport;
1014 0 : impl_opts.tls_cipher_suites = "TLS_AES_256_GCM_SHA384:TLS_AES_128_GCM_SHA256";
1015 : }
1016 :
1017 0 : opts.impl_opts = &impl_opts;
1018 0 : opts.impl_opts_size = sizeof(impl_opts);
1019 : }
1020 :
1021 0 : port->listen_sock = spdk_sock_listen_ext(trid->traddr, trsvcid_int,
1022 : sock_impl_name, &opts);
1023 0 : if (port->listen_sock == NULL) {
1024 0 : SPDK_ERRLOG("spdk_sock_listen(%s, %d) failed: %s (%d)\n",
1025 : trid->traddr, trsvcid_int,
1026 : spdk_strerror(errno), errno);
1027 0 : free(port);
1028 0 : return -errno;
1029 : }
1030 :
1031 0 : if (spdk_sock_is_ipv4(port->listen_sock)) {
1032 0 : adrfam = SPDK_NVMF_ADRFAM_IPV4;
1033 0 : } else if (spdk_sock_is_ipv6(port->listen_sock)) {
1034 0 : adrfam = SPDK_NVMF_ADRFAM_IPV6;
1035 : } else {
1036 0 : SPDK_ERRLOG("Unhandled socket type\n");
1037 0 : adrfam = 0;
1038 : }
1039 :
1040 0 : if (adrfam != trid->adrfam) {
1041 0 : SPDK_ERRLOG("Socket address family mismatch\n");
1042 0 : spdk_sock_close(&port->listen_sock);
1043 0 : free(port);
1044 0 : return -EINVAL;
1045 : }
1046 :
1047 0 : rc = spdk_sock_group_add_sock(ttransport->listen_sock_group, port->listen_sock, nvmf_tcp_accept_cb,
1048 : port);
1049 0 : if (rc < 0) {
1050 0 : SPDK_ERRLOG("Failed to add socket to the listen socket group\n");
1051 0 : spdk_sock_close(&port->listen_sock);
1052 0 : free(port);
1053 0 : return -errno;
1054 : }
1055 :
1056 0 : port->transport = transport;
1057 :
1058 0 : SPDK_NOTICELOG("*** NVMe/TCP Target Listening on %s port %s ***\n",
1059 : trid->traddr, trid->trsvcid);
1060 :
1061 0 : TAILQ_INSERT_TAIL(&ttransport->ports, port, link);
1062 0 : return 0;
1063 : }
1064 :
1065 : static void
1066 0 : nvmf_tcp_stop_listen(struct spdk_nvmf_transport *transport,
1067 : const struct spdk_nvme_transport_id *trid)
1068 : {
1069 : struct spdk_nvmf_tcp_transport *ttransport;
1070 : struct spdk_nvmf_tcp_port *port;
1071 :
1072 0 : ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport);
1073 :
1074 0 : SPDK_DEBUGLOG(nvmf_tcp, "Removing listen address %s port %s\n",
1075 : trid->traddr, trid->trsvcid);
1076 :
1077 0 : port = nvmf_tcp_find_port(ttransport, trid);
1078 0 : if (port) {
1079 0 : spdk_sock_group_remove_sock(ttransport->listen_sock_group, port->listen_sock);
1080 0 : TAILQ_REMOVE(&ttransport->ports, port, link);
1081 0 : spdk_sock_close(&port->listen_sock);
1082 0 : free(port);
1083 : }
1084 0 : }
1085 :
1086 : static void nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair,
1087 : enum nvme_tcp_pdu_recv_state state);
1088 :
1089 : static void
1090 1 : nvmf_tcp_qpair_set_state(struct spdk_nvmf_tcp_qpair *tqpair, enum nvme_tcp_qpair_state state)
1091 : {
1092 1 : tqpair->state = state;
1093 1 : spdk_trace_record(TRACE_TCP_QP_STATE_CHANGE, tqpair->qpair.trace_id, 0, 0,
1094 : (uint64_t)tqpair->state);
1095 1 : }
1096 :
1097 : static void
1098 0 : nvmf_tcp_qpair_disconnect(struct spdk_nvmf_tcp_qpair *tqpair)
1099 : {
1100 0 : SPDK_DEBUGLOG(nvmf_tcp, "Disconnecting qpair %p\n", tqpair);
1101 :
1102 0 : spdk_trace_record(TRACE_TCP_QP_DISCONNECT, tqpair->qpair.trace_id, 0, 0);
1103 :
1104 0 : if (tqpair->state <= NVME_TCP_QPAIR_STATE_RUNNING) {
1105 0 : nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_EXITING);
1106 0 : assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR);
1107 0 : spdk_poller_unregister(&tqpair->timeout_poller);
1108 :
1109 : /* This will end up calling nvmf_tcp_close_qpair */
1110 0 : spdk_nvmf_qpair_disconnect(&tqpair->qpair);
1111 : }
1112 0 : }
1113 :
1114 : static void
1115 16 : _mgmt_pdu_write_done(void *_tqpair, int err)
1116 : {
1117 16 : struct spdk_nvmf_tcp_qpair *tqpair = _tqpair;
1118 16 : struct nvme_tcp_pdu *pdu = tqpair->mgmt_pdu;
1119 :
1120 16 : if (spdk_unlikely(err != 0)) {
1121 16 : nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING);
1122 16 : return;
1123 : }
1124 :
1125 0 : assert(pdu->cb_fn != NULL);
1126 0 : pdu->cb_fn(pdu->cb_arg);
1127 : }
1128 :
1129 : static void
1130 0 : _req_pdu_write_done(void *req, int err)
1131 : {
1132 0 : struct spdk_nvmf_tcp_req *tcp_req = req;
1133 0 : struct nvme_tcp_pdu *pdu = tcp_req->pdu;
1134 0 : struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair;
1135 :
1136 0 : assert(tcp_req->pdu_in_use);
1137 0 : tcp_req->pdu_in_use = false;
1138 :
1139 : /* If the request is in a completed state, we're waiting for write completion to free it */
1140 0 : if (spdk_unlikely(tcp_req->state == TCP_REQUEST_STATE_COMPLETED)) {
1141 0 : nvmf_tcp_request_free(tcp_req);
1142 0 : return;
1143 : }
1144 :
1145 0 : if (spdk_unlikely(err != 0)) {
1146 0 : nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING);
1147 0 : return;
1148 : }
1149 :
1150 0 : assert(pdu->cb_fn != NULL);
1151 0 : pdu->cb_fn(pdu->cb_arg);
1152 : }
1153 :
1154 : static void
1155 16 : _pdu_write_done(struct nvme_tcp_pdu *pdu, int err)
1156 : {
1157 16 : pdu->sock_req.cb_fn(pdu->sock_req.cb_arg, err);
1158 16 : }
1159 :
1160 : static void
1161 23 : _tcp_write_pdu(struct nvme_tcp_pdu *pdu)
1162 : {
1163 : int rc;
1164 23 : uint32_t mapped_length;
1165 23 : struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair;
1166 :
1167 46 : pdu->sock_req.iovcnt = nvme_tcp_build_iovs(pdu->iov, SPDK_COUNTOF(pdu->iov), pdu,
1168 23 : tqpair->host_hdgst_enable, tqpair->host_ddgst_enable, &mapped_length);
1169 23 : spdk_sock_writev_async(tqpair->sock, &pdu->sock_req);
1170 :
1171 23 : if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP ||
1172 22 : pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ) {
1173 : /* Try to force the send immediately. */
1174 16 : rc = spdk_sock_flush(tqpair->sock);
1175 16 : if (rc > 0 && (uint32_t)rc == mapped_length) {
1176 0 : _pdu_write_done(pdu, 0);
1177 : } else {
1178 16 : SPDK_ERRLOG("Could not write %s to socket: rc=%d, errno=%d\n",
1179 : pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP ?
1180 : "IC_RESP" : "TERM_REQ", rc, errno);
1181 16 : _pdu_write_done(pdu, rc >= 0 ? -EAGAIN : -errno);
1182 : }
1183 : }
1184 23 : }
1185 :
1186 : static void
1187 0 : data_crc32_accel_done(void *cb_arg, int status)
1188 : {
1189 0 : struct nvme_tcp_pdu *pdu = cb_arg;
1190 :
1191 0 : if (spdk_unlikely(status)) {
1192 0 : SPDK_ERRLOG("Failed to compute the data digest for pdu =%p\n", pdu);
1193 0 : _pdu_write_done(pdu, status);
1194 0 : return;
1195 : }
1196 :
1197 0 : pdu->data_digest_crc32 ^= SPDK_CRC32C_XOR;
1198 0 : MAKE_DIGEST_WORD(pdu->data_digest, pdu->data_digest_crc32);
1199 :
1200 0 : _tcp_write_pdu(pdu);
1201 : }
1202 :
1203 : static void
1204 23 : pdu_data_crc32_compute(struct nvme_tcp_pdu *pdu)
1205 : {
1206 23 : struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair;
1207 23 : int rc = 0;
1208 :
1209 : /* Data Digest */
1210 23 : if (pdu->data_len > 0 && g_nvme_tcp_ddgst[pdu->hdr.common.pdu_type] && tqpair->host_ddgst_enable) {
1211 : /* Only support this limitated case for the first step */
1212 0 : if (spdk_likely(!pdu->dif_ctx && (pdu->data_len % SPDK_NVME_TCP_DIGEST_ALIGNMENT == 0)
1213 : && tqpair->group)) {
1214 0 : rc = spdk_accel_submit_crc32cv(tqpair->group->accel_channel, &pdu->data_digest_crc32, pdu->data_iov,
1215 : pdu->data_iovcnt, 0, data_crc32_accel_done, pdu);
1216 0 : if (spdk_likely(rc == 0)) {
1217 0 : return;
1218 : }
1219 : } else {
1220 0 : pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu);
1221 : }
1222 0 : data_crc32_accel_done(pdu, rc);
1223 : } else {
1224 23 : _tcp_write_pdu(pdu);
1225 : }
1226 : }
1227 :
1228 : static void
1229 23 : nvmf_tcp_qpair_write_pdu(struct spdk_nvmf_tcp_qpair *tqpair,
1230 : struct nvme_tcp_pdu *pdu,
1231 : nvme_tcp_qpair_xfer_complete_cb cb_fn,
1232 : void *cb_arg)
1233 : {
1234 : int hlen;
1235 : uint32_t crc32c;
1236 :
1237 23 : assert(tqpair->pdu_in_progress != pdu);
1238 :
1239 23 : hlen = pdu->hdr.common.hlen;
1240 23 : pdu->cb_fn = cb_fn;
1241 23 : pdu->cb_arg = cb_arg;
1242 :
1243 23 : pdu->iov[0].iov_base = &pdu->hdr.raw;
1244 23 : pdu->iov[0].iov_len = hlen;
1245 :
1246 : /* Header Digest */
1247 23 : if (g_nvme_tcp_hdgst[pdu->hdr.common.pdu_type] && tqpair->host_hdgst_enable) {
1248 1 : crc32c = nvme_tcp_pdu_calc_header_digest(pdu);
1249 1 : MAKE_DIGEST_WORD((uint8_t *)pdu->hdr.raw + hlen, crc32c);
1250 : }
1251 :
1252 : /* Data Digest */
1253 23 : pdu_data_crc32_compute(pdu);
1254 23 : }
1255 :
1256 : static void
1257 16 : nvmf_tcp_qpair_write_mgmt_pdu(struct spdk_nvmf_tcp_qpair *tqpair,
1258 : nvme_tcp_qpair_xfer_complete_cb cb_fn,
1259 : void *cb_arg)
1260 : {
1261 16 : struct nvme_tcp_pdu *pdu = tqpair->mgmt_pdu;
1262 :
1263 16 : pdu->sock_req.cb_fn = _mgmt_pdu_write_done;
1264 16 : pdu->sock_req.cb_arg = tqpair;
1265 :
1266 16 : nvmf_tcp_qpair_write_pdu(tqpair, pdu, cb_fn, cb_arg);
1267 16 : }
1268 :
1269 : static void
1270 7 : nvmf_tcp_qpair_write_req_pdu(struct spdk_nvmf_tcp_qpair *tqpair,
1271 : struct spdk_nvmf_tcp_req *tcp_req,
1272 : nvme_tcp_qpair_xfer_complete_cb cb_fn,
1273 : void *cb_arg)
1274 : {
1275 7 : struct nvme_tcp_pdu *pdu = tcp_req->pdu;
1276 :
1277 7 : pdu->sock_req.cb_fn = _req_pdu_write_done;
1278 7 : pdu->sock_req.cb_arg = tcp_req;
1279 :
1280 7 : assert(!tcp_req->pdu_in_use);
1281 7 : tcp_req->pdu_in_use = true;
1282 :
1283 7 : nvmf_tcp_qpair_write_pdu(tqpair, pdu, cb_fn, cb_arg);
1284 7 : }
1285 :
1286 : static int
1287 1 : nvmf_tcp_qpair_init_mem_resource(struct spdk_nvmf_tcp_qpair *tqpair)
1288 : {
1289 : uint32_t i;
1290 : struct spdk_nvmf_transport_opts *opts;
1291 : uint32_t in_capsule_data_size;
1292 :
1293 1 : opts = &tqpair->qpair.transport->opts;
1294 :
1295 1 : in_capsule_data_size = opts->in_capsule_data_size;
1296 1 : if (opts->dif_insert_or_strip) {
1297 0 : in_capsule_data_size = SPDK_BDEV_BUF_SIZE_WITH_MD(in_capsule_data_size);
1298 : }
1299 :
1300 1 : tqpair->resource_count = opts->max_queue_depth;
1301 :
1302 1 : tqpair->reqs = calloc(tqpair->resource_count, sizeof(*tqpair->reqs));
1303 1 : if (!tqpair->reqs) {
1304 0 : SPDK_ERRLOG("Unable to allocate reqs on tqpair=%p\n", tqpair);
1305 0 : return -1;
1306 : }
1307 :
1308 1 : if (in_capsule_data_size) {
1309 1 : tqpair->bufs = spdk_zmalloc(tqpair->resource_count * in_capsule_data_size, 0x1000,
1310 : NULL, SPDK_ENV_LCORE_ID_ANY,
1311 : SPDK_MALLOC_DMA);
1312 1 : if (!tqpair->bufs) {
1313 0 : SPDK_ERRLOG("Unable to allocate bufs on tqpair=%p.\n", tqpair);
1314 0 : return -1;
1315 : }
1316 : }
1317 : /* prepare memory space for receiving pdus and tcp_req */
1318 : /* Add additional 1 member, which will be used for mgmt_pdu owned by the tqpair */
1319 1 : tqpair->pdus = spdk_dma_zmalloc((2 * tqpair->resource_count + 1) * sizeof(*tqpair->pdus), 0x1000,
1320 : NULL);
1321 1 : if (!tqpair->pdus) {
1322 0 : SPDK_ERRLOG("Unable to allocate pdu pool on tqpair =%p.\n", tqpair);
1323 0 : return -1;
1324 : }
1325 :
1326 129 : for (i = 0; i < tqpair->resource_count; i++) {
1327 128 : struct spdk_nvmf_tcp_req *tcp_req = &tqpair->reqs[i];
1328 :
1329 128 : tcp_req->ttag = i + 1;
1330 128 : tcp_req->req.qpair = &tqpair->qpair;
1331 :
1332 128 : tcp_req->pdu = &tqpair->pdus[i];
1333 128 : tcp_req->pdu->qpair = tqpair;
1334 :
1335 : /* Set up memory to receive commands */
1336 128 : if (tqpair->bufs) {
1337 128 : tcp_req->buf = (void *)((uintptr_t)tqpair->bufs + (i * in_capsule_data_size));
1338 : }
1339 :
1340 : /* Set the cmdn and rsp */
1341 128 : tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp;
1342 128 : tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd;
1343 :
1344 128 : tcp_req->req.stripped_data = NULL;
1345 :
1346 : /* Initialize request state to FREE */
1347 128 : tcp_req->state = TCP_REQUEST_STATE_FREE;
1348 128 : TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link);
1349 128 : tqpair->state_cntr[TCP_REQUEST_STATE_FREE]++;
1350 : }
1351 :
1352 129 : for (; i < 2 * tqpair->resource_count; i++) {
1353 128 : struct nvme_tcp_pdu *pdu = &tqpair->pdus[i];
1354 :
1355 128 : pdu->qpair = tqpair;
1356 128 : SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, pdu, slist);
1357 : }
1358 :
1359 1 : tqpair->mgmt_pdu = &tqpair->pdus[i];
1360 1 : tqpair->mgmt_pdu->qpair = tqpair;
1361 1 : tqpair->pdu_in_progress = SLIST_FIRST(&tqpair->tcp_pdu_free_queue);
1362 1 : SLIST_REMOVE_HEAD(&tqpair->tcp_pdu_free_queue, slist);
1363 1 : tqpair->tcp_pdu_working_count = 1;
1364 :
1365 1 : tqpair->recv_buf_size = (in_capsule_data_size + sizeof(struct spdk_nvme_tcp_cmd) + 2 *
1366 : SPDK_NVME_TCP_DIGEST_LEN) * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR;
1367 :
1368 1 : return 0;
1369 : }
1370 :
1371 : static int
1372 1 : nvmf_tcp_qpair_init(struct spdk_nvmf_qpair *qpair)
1373 : {
1374 : struct spdk_nvmf_tcp_qpair *tqpair;
1375 :
1376 1 : tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
1377 :
1378 1 : SPDK_DEBUGLOG(nvmf_tcp, "New TCP Connection: %p\n", qpair);
1379 :
1380 1 : spdk_trace_record(TRACE_TCP_QP_CREATE, tqpair->qpair.trace_id, 0, 0);
1381 :
1382 : /* Initialise request state queues of the qpair */
1383 1 : TAILQ_INIT(&tqpair->tcp_req_free_queue);
1384 1 : TAILQ_INIT(&tqpair->tcp_req_working_queue);
1385 1 : SLIST_INIT(&tqpair->tcp_pdu_free_queue);
1386 1 : tqpair->qpair.queue_depth = 0;
1387 :
1388 1 : tqpair->host_hdgst_enable = true;
1389 1 : tqpair->host_ddgst_enable = true;
1390 :
1391 1 : return 0;
1392 : }
1393 :
1394 : static int
1395 0 : nvmf_tcp_qpair_sock_init(struct spdk_nvmf_tcp_qpair *tqpair)
1396 : {
1397 0 : char saddr[32], caddr[32];
1398 0 : uint16_t sport, cport;
1399 0 : char owner[256];
1400 : int rc;
1401 :
1402 0 : rc = spdk_sock_getaddr(tqpair->sock, saddr, sizeof(saddr), &sport,
1403 : caddr, sizeof(caddr), &cport);
1404 0 : if (rc != 0) {
1405 0 : SPDK_ERRLOG("spdk_sock_getaddr() failed\n");
1406 0 : return rc;
1407 : }
1408 0 : snprintf(owner, sizeof(owner), "%s:%d", caddr, cport);
1409 0 : tqpair->qpair.trace_id = spdk_trace_register_owner(OWNER_TYPE_NVMF_TCP, owner);
1410 0 : spdk_trace_record(TRACE_TCP_QP_SOCK_INIT, tqpair->qpair.trace_id, 0, 0);
1411 :
1412 : /* set low water mark */
1413 0 : rc = spdk_sock_set_recvlowat(tqpair->sock, 1);
1414 0 : if (rc != 0) {
1415 0 : SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n");
1416 0 : return rc;
1417 : }
1418 :
1419 0 : return 0;
1420 : }
1421 :
1422 : static void
1423 0 : nvmf_tcp_handle_connect(struct spdk_nvmf_tcp_port *port, struct spdk_sock *sock)
1424 : {
1425 : struct spdk_nvmf_tcp_qpair *tqpair;
1426 : int rc;
1427 :
1428 0 : SPDK_DEBUGLOG(nvmf_tcp, "New connection accepted on %s port %s\n",
1429 : port->trid->traddr, port->trid->trsvcid);
1430 :
1431 0 : tqpair = calloc(1, sizeof(struct spdk_nvmf_tcp_qpair));
1432 0 : if (tqpair == NULL) {
1433 0 : SPDK_ERRLOG("Could not allocate new connection.\n");
1434 0 : spdk_sock_close(&sock);
1435 0 : return;
1436 : }
1437 :
1438 0 : tqpair->sock = sock;
1439 0 : tqpair->state_cntr[TCP_REQUEST_STATE_FREE] = 0;
1440 0 : tqpair->port = port;
1441 0 : tqpair->qpair.transport = port->transport;
1442 :
1443 0 : rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr,
1444 : sizeof(tqpair->target_addr), &tqpair->target_port,
1445 0 : tqpair->initiator_addr, sizeof(tqpair->initiator_addr),
1446 : &tqpair->initiator_port);
1447 0 : if (rc < 0) {
1448 0 : SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair);
1449 0 : nvmf_tcp_qpair_destroy(tqpair);
1450 0 : return;
1451 : }
1452 :
1453 0 : spdk_nvmf_tgt_new_qpair(port->transport->tgt, &tqpair->qpair);
1454 : }
1455 :
1456 : static uint32_t
1457 0 : nvmf_tcp_port_accept(struct spdk_nvmf_tcp_port *port)
1458 : {
1459 : struct spdk_sock *sock;
1460 0 : uint32_t count = 0;
1461 : int i;
1462 :
1463 0 : for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) {
1464 0 : sock = spdk_sock_accept(port->listen_sock);
1465 0 : if (sock == NULL) {
1466 0 : break;
1467 : }
1468 0 : count++;
1469 0 : nvmf_tcp_handle_connect(port, sock);
1470 : }
1471 :
1472 0 : return count;
1473 : }
1474 :
1475 : static int
1476 0 : nvmf_tcp_accept(void *ctx)
1477 : {
1478 0 : struct spdk_nvmf_transport *transport = ctx;
1479 : struct spdk_nvmf_tcp_transport *ttransport;
1480 : int count;
1481 :
1482 0 : ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport);
1483 :
1484 0 : count = spdk_sock_group_poll(ttransport->listen_sock_group);
1485 0 : if (count < 0) {
1486 0 : SPDK_ERRLOG("Fail in TCP listen socket group poll\n");
1487 : }
1488 :
1489 0 : return count != 0 ? SPDK_POLLER_BUSY : SPDK_POLLER_IDLE;
1490 : }
1491 :
1492 : static void
1493 0 : nvmf_tcp_accept_cb(void *ctx, struct spdk_sock_group *group, struct spdk_sock *sock)
1494 : {
1495 0 : struct spdk_nvmf_tcp_port *port = ctx;
1496 :
1497 0 : nvmf_tcp_port_accept(port);
1498 0 : }
1499 :
1500 : static void
1501 0 : nvmf_tcp_discover(struct spdk_nvmf_transport *transport,
1502 : struct spdk_nvme_transport_id *trid,
1503 : struct spdk_nvmf_discovery_log_page_entry *entry)
1504 : {
1505 : struct spdk_nvmf_tcp_port *port;
1506 : struct spdk_nvmf_tcp_transport *ttransport;
1507 :
1508 0 : entry->trtype = SPDK_NVMF_TRTYPE_TCP;
1509 0 : entry->adrfam = trid->adrfam;
1510 :
1511 0 : spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' ');
1512 0 : spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' ');
1513 :
1514 0 : ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport);
1515 0 : port = nvmf_tcp_find_port(ttransport, trid);
1516 :
1517 0 : assert(port != NULL);
1518 :
1519 0 : if (strcmp(spdk_sock_get_impl_name(port->listen_sock), "ssl") == 0) {
1520 0 : entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_REQUIRED;
1521 0 : entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_TLS_1_3;
1522 : } else {
1523 0 : entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_REQUIRED;
1524 0 : entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE;
1525 : }
1526 0 : }
1527 :
1528 : static struct spdk_nvmf_tcp_control_msg_list *
1529 1 : nvmf_tcp_control_msg_list_create(uint16_t num_messages)
1530 : {
1531 : struct spdk_nvmf_tcp_control_msg_list *list;
1532 : struct spdk_nvmf_tcp_control_msg *msg;
1533 : uint16_t i;
1534 :
1535 1 : list = calloc(1, sizeof(*list));
1536 1 : if (!list) {
1537 0 : SPDK_ERRLOG("Failed to allocate memory for list structure\n");
1538 0 : return NULL;
1539 : }
1540 :
1541 1 : list->msg_buf = spdk_zmalloc(num_messages * SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE,
1542 : NVMF_DATA_BUFFER_ALIGNMENT, NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
1543 1 : if (!list->msg_buf) {
1544 0 : SPDK_ERRLOG("Failed to allocate memory for control message buffers\n");
1545 0 : free(list);
1546 0 : return NULL;
1547 : }
1548 :
1549 1 : STAILQ_INIT(&list->free_msgs);
1550 1 : STAILQ_INIT(&list->waiting_for_msg_reqs);
1551 :
1552 33 : for (i = 0; i < num_messages; i++) {
1553 32 : msg = (struct spdk_nvmf_tcp_control_msg *)((char *)list->msg_buf + i *
1554 : SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE);
1555 32 : STAILQ_INSERT_TAIL(&list->free_msgs, msg, link);
1556 : }
1557 :
1558 1 : return list;
1559 : }
1560 :
1561 : static void
1562 1 : nvmf_tcp_control_msg_list_free(struct spdk_nvmf_tcp_control_msg_list *list)
1563 : {
1564 1 : if (!list) {
1565 0 : return;
1566 : }
1567 :
1568 1 : spdk_free(list->msg_buf);
1569 1 : free(list);
1570 : }
1571 :
1572 : static struct spdk_nvmf_transport_poll_group *
1573 1 : nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport,
1574 : struct spdk_nvmf_poll_group *group)
1575 : {
1576 : struct spdk_nvmf_tcp_transport *ttransport;
1577 : struct spdk_nvmf_tcp_poll_group *tgroup;
1578 :
1579 1 : if (spdk_interrupt_mode_is_enabled()) {
1580 0 : SPDK_ERRLOG("TCP transport does not support interrupt mode\n");
1581 0 : return NULL;
1582 : }
1583 :
1584 1 : tgroup = calloc(1, sizeof(*tgroup));
1585 1 : if (!tgroup) {
1586 0 : return NULL;
1587 : }
1588 :
1589 1 : tgroup->sock_group = spdk_sock_group_create(&tgroup->group);
1590 1 : if (!tgroup->sock_group) {
1591 0 : goto cleanup;
1592 : }
1593 :
1594 1 : TAILQ_INIT(&tgroup->qpairs);
1595 1 : TAILQ_INIT(&tgroup->await_req);
1596 :
1597 1 : ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport);
1598 :
1599 1 : if (transport->opts.in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) {
1600 1 : SPDK_DEBUGLOG(nvmf_tcp, "ICD %u is less than min required for admin/fabric commands (%u). "
1601 : "Creating control messages list\n", transport->opts.in_capsule_data_size,
1602 : SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE);
1603 1 : tgroup->control_msg_list = nvmf_tcp_control_msg_list_create(ttransport->tcp_opts.control_msg_num);
1604 1 : if (!tgroup->control_msg_list) {
1605 0 : goto cleanup;
1606 : }
1607 : }
1608 :
1609 1 : tgroup->accel_channel = spdk_accel_get_io_channel();
1610 1 : if (spdk_unlikely(!tgroup->accel_channel)) {
1611 0 : SPDK_ERRLOG("Cannot create accel_channel for tgroup=%p\n", tgroup);
1612 0 : goto cleanup;
1613 : }
1614 :
1615 1 : TAILQ_INSERT_TAIL(&ttransport->poll_groups, tgroup, link);
1616 1 : if (ttransport->next_pg == NULL) {
1617 1 : ttransport->next_pg = tgroup;
1618 : }
1619 :
1620 1 : return &tgroup->group;
1621 :
1622 0 : cleanup:
1623 0 : nvmf_tcp_poll_group_destroy(&tgroup->group);
1624 0 : return NULL;
1625 : }
1626 :
1627 : static struct spdk_nvmf_transport_poll_group *
1628 0 : nvmf_tcp_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair)
1629 : {
1630 : struct spdk_nvmf_tcp_transport *ttransport;
1631 : struct spdk_nvmf_tcp_poll_group **pg;
1632 : struct spdk_nvmf_tcp_qpair *tqpair;
1633 0 : struct spdk_sock_group *group = NULL, *hint = NULL;
1634 : int rc;
1635 :
1636 0 : ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport);
1637 :
1638 0 : if (TAILQ_EMPTY(&ttransport->poll_groups)) {
1639 0 : return NULL;
1640 : }
1641 :
1642 0 : pg = &ttransport->next_pg;
1643 0 : assert(*pg != NULL);
1644 0 : hint = (*pg)->sock_group;
1645 :
1646 0 : tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
1647 0 : rc = spdk_sock_get_optimal_sock_group(tqpair->sock, &group, hint);
1648 0 : if (rc != 0) {
1649 0 : return NULL;
1650 0 : } else if (group != NULL) {
1651 : /* Optimal poll group was found */
1652 0 : return spdk_sock_group_get_ctx(group);
1653 : }
1654 :
1655 : /* The hint was used for optimal poll group, advance next_pg. */
1656 0 : *pg = TAILQ_NEXT(*pg, link);
1657 0 : if (*pg == NULL) {
1658 0 : *pg = TAILQ_FIRST(&ttransport->poll_groups);
1659 : }
1660 :
1661 0 : return spdk_sock_group_get_ctx(hint);
1662 : }
1663 :
1664 : static void
1665 1 : nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group)
1666 : {
1667 : struct spdk_nvmf_tcp_poll_group *tgroup, *next_tgroup;
1668 : struct spdk_nvmf_tcp_transport *ttransport;
1669 :
1670 1 : tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group);
1671 1 : spdk_sock_group_close(&tgroup->sock_group);
1672 1 : if (tgroup->control_msg_list) {
1673 1 : nvmf_tcp_control_msg_list_free(tgroup->control_msg_list);
1674 : }
1675 :
1676 1 : if (tgroup->accel_channel) {
1677 1 : spdk_put_io_channel(tgroup->accel_channel);
1678 : }
1679 :
1680 1 : if (tgroup->group.transport == NULL) {
1681 : /* Transport can be NULL when nvmf_tcp_poll_group_create()
1682 : * calls this function directly in a failure path. */
1683 0 : free(tgroup);
1684 0 : return;
1685 : }
1686 :
1687 1 : ttransport = SPDK_CONTAINEROF(tgroup->group.transport, struct spdk_nvmf_tcp_transport, transport);
1688 :
1689 1 : next_tgroup = TAILQ_NEXT(tgroup, link);
1690 1 : TAILQ_REMOVE(&ttransport->poll_groups, tgroup, link);
1691 1 : if (next_tgroup == NULL) {
1692 1 : next_tgroup = TAILQ_FIRST(&ttransport->poll_groups);
1693 : }
1694 1 : if (ttransport->next_pg == tgroup) {
1695 1 : ttransport->next_pg = next_tgroup;
1696 : }
1697 :
1698 1 : free(tgroup);
1699 : }
1700 :
1701 : static void
1702 37 : nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair,
1703 : enum nvme_tcp_pdu_recv_state state)
1704 : {
1705 37 : if (tqpair->recv_state == state) {
1706 19 : SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n",
1707 : tqpair, state);
1708 19 : return;
1709 : }
1710 :
1711 18 : if (spdk_unlikely(state == NVME_TCP_PDU_RECV_STATE_QUIESCING)) {
1712 13 : if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH && tqpair->pdu_in_progress) {
1713 10 : SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, tqpair->pdu_in_progress, slist);
1714 10 : tqpair->tcp_pdu_working_count--;
1715 : }
1716 : }
1717 :
1718 18 : if (spdk_unlikely(state == NVME_TCP_PDU_RECV_STATE_ERROR)) {
1719 0 : assert(tqpair->tcp_pdu_working_count == 0);
1720 : }
1721 :
1722 18 : if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) {
1723 : /* When leaving the await req state, move the qpair to the main list */
1724 0 : TAILQ_REMOVE(&tqpair->group->await_req, tqpair, link);
1725 0 : TAILQ_INSERT_TAIL(&tqpair->group->qpairs, tqpair, link);
1726 18 : } else if (state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) {
1727 0 : TAILQ_REMOVE(&tqpair->group->qpairs, tqpair, link);
1728 0 : TAILQ_INSERT_TAIL(&tqpair->group->await_req, tqpair, link);
1729 : }
1730 :
1731 18 : SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv state=%d\n", tqpair, state);
1732 18 : tqpair->recv_state = state;
1733 :
1734 18 : spdk_trace_record(TRACE_TCP_QP_RCV_STATE_CHANGE, tqpair->qpair.trace_id, 0, 0,
1735 : (uint64_t)tqpair->recv_state);
1736 : }
1737 :
1738 : static int
1739 0 : nvmf_tcp_qpair_handle_timeout(void *ctx)
1740 : {
1741 0 : struct spdk_nvmf_tcp_qpair *tqpair = ctx;
1742 :
1743 0 : assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR);
1744 :
1745 0 : SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair,
1746 : SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT);
1747 :
1748 0 : nvmf_tcp_qpair_disconnect(tqpair);
1749 0 : return SPDK_POLLER_BUSY;
1750 : }
1751 :
1752 : static void
1753 0 : nvmf_tcp_send_c2h_term_req_complete(void *cb_arg)
1754 : {
1755 0 : struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg;
1756 :
1757 0 : if (!tqpair->timeout_poller) {
1758 0 : tqpair->timeout_poller = SPDK_POLLER_REGISTER(nvmf_tcp_qpair_handle_timeout, tqpair,
1759 : SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000);
1760 : }
1761 0 : }
1762 :
1763 : static void
1764 15 : nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu,
1765 : enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset)
1766 : {
1767 : struct nvme_tcp_pdu *rsp_pdu;
1768 : struct spdk_nvme_tcp_term_req_hdr *c2h_term_req;
1769 15 : uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req);
1770 : uint32_t copy_len;
1771 :
1772 15 : rsp_pdu = tqpair->mgmt_pdu;
1773 :
1774 15 : c2h_term_req = &rsp_pdu->hdr.term_req;
1775 15 : c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ;
1776 15 : c2h_term_req->common.hlen = c2h_term_req_hdr_len;
1777 15 : c2h_term_req->fes = fes;
1778 :
1779 15 : if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) ||
1780 : (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) {
1781 12 : DSET32(&c2h_term_req->fei, error_offset);
1782 : }
1783 :
1784 15 : copy_len = spdk_min(pdu->hdr.common.hlen, SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE);
1785 :
1786 : /* Copy the error info into the buffer */
1787 15 : memcpy((uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, pdu->hdr.raw, copy_len);
1788 15 : nvme_tcp_pdu_set_data(rsp_pdu, (uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, copy_len);
1789 :
1790 : /* Contain the header of the wrong received pdu */
1791 15 : c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len;
1792 15 : tqpair->wait_terminate = true;
1793 15 : nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING);
1794 15 : nvmf_tcp_qpair_write_mgmt_pdu(tqpair, nvmf_tcp_send_c2h_term_req_complete, tqpair);
1795 15 : }
1796 :
1797 : static void
1798 1 : nvmf_tcp_capsule_cmd_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport,
1799 : struct spdk_nvmf_tcp_qpair *tqpair,
1800 : struct nvme_tcp_pdu *pdu)
1801 : {
1802 : struct spdk_nvmf_tcp_req *tcp_req;
1803 :
1804 1 : assert(pdu->psh_valid_bytes == pdu->psh_len);
1805 1 : assert(pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD);
1806 :
1807 1 : tcp_req = nvmf_tcp_req_get(tqpair);
1808 1 : if (!tcp_req) {
1809 : /* Directly return and make the allocation retry again. This can happen if we're
1810 : * using asynchronous writes to send the response to the host or when releasing
1811 : * zero-copy buffers after a response has been sent. In both cases, the host might
1812 : * receive the response before we've finished processing the request and is free to
1813 : * send another one.
1814 : */
1815 0 : if (tqpair->state_cntr[TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST] > 0 ||
1816 0 : tqpair->state_cntr[TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE] > 0) {
1817 0 : return;
1818 : }
1819 :
1820 : /* The host sent more commands than the maximum queue depth. */
1821 0 : SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair);
1822 0 : nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING);
1823 0 : return;
1824 : }
1825 :
1826 1 : pdu->req = tcp_req;
1827 1 : assert(tcp_req->state == TCP_REQUEST_STATE_NEW);
1828 1 : nvmf_tcp_req_process(ttransport, tcp_req);
1829 : }
1830 :
1831 : static void
1832 0 : nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport,
1833 : struct spdk_nvmf_tcp_qpair *tqpair,
1834 : struct nvme_tcp_pdu *pdu)
1835 : {
1836 : struct spdk_nvmf_tcp_req *tcp_req;
1837 : struct spdk_nvme_tcp_cmd *capsule_cmd;
1838 0 : uint32_t error_offset = 0;
1839 : enum spdk_nvme_tcp_term_req_fes fes;
1840 : struct spdk_nvme_cpl *rsp;
1841 :
1842 0 : capsule_cmd = &pdu->hdr.capsule_cmd;
1843 0 : tcp_req = pdu->req;
1844 0 : assert(tcp_req != NULL);
1845 :
1846 : /* Zero-copy requests don't support ICD */
1847 0 : assert(!spdk_nvmf_request_using_zcopy(&tcp_req->req));
1848 :
1849 0 : if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) {
1850 0 : SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n",
1851 : SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo);
1852 0 : fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
1853 0 : error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo);
1854 0 : goto err;
1855 : }
1856 :
1857 0 : rsp = &tcp_req->req.rsp->nvme_cpl;
1858 0 : if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) {
1859 0 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE);
1860 : } else {
1861 0 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE);
1862 : }
1863 :
1864 0 : nvmf_tcp_req_process(ttransport, tcp_req);
1865 :
1866 0 : return;
1867 0 : err:
1868 0 : nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
1869 : }
1870 :
1871 : static void
1872 1 : nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport,
1873 : struct spdk_nvmf_tcp_qpair *tqpair,
1874 : struct nvme_tcp_pdu *pdu)
1875 : {
1876 : struct spdk_nvmf_tcp_req *tcp_req;
1877 1 : uint32_t error_offset = 0;
1878 1 : enum spdk_nvme_tcp_term_req_fes fes = 0;
1879 : struct spdk_nvme_tcp_h2c_data_hdr *h2c_data;
1880 :
1881 1 : h2c_data = &pdu->hdr.h2c_data;
1882 :
1883 1 : SPDK_DEBUGLOG(nvmf_tcp, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n",
1884 : tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag);
1885 :
1886 1 : if (h2c_data->ttag > tqpair->resource_count) {
1887 0 : SPDK_DEBUGLOG(nvmf_tcp, "ttag %u is larger than allowed %u.\n", h2c_data->ttag,
1888 : tqpair->resource_count);
1889 0 : fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR;
1890 0 : error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag);
1891 0 : goto err;
1892 : }
1893 :
1894 1 : tcp_req = &tqpair->reqs[h2c_data->ttag - 1];
1895 :
1896 1 : if (spdk_unlikely(tcp_req->state != TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER &&
1897 : tcp_req->state != TCP_REQUEST_STATE_AWAITING_R2T_ACK)) {
1898 0 : SPDK_DEBUGLOG(nvmf_tcp, "tcp_req(%p), tqpair=%p, has error state in %d\n", tcp_req, tqpair,
1899 : tcp_req->state);
1900 0 : fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
1901 0 : error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag);
1902 0 : goto err;
1903 : }
1904 :
1905 1 : if (spdk_unlikely(tcp_req->req.cmd->nvme_cmd.cid != h2c_data->cccid)) {
1906 0 : SPDK_DEBUGLOG(nvmf_tcp, "tcp_req(%p), tqpair=%p, expected %u but %u for cccid.\n", tcp_req, tqpair,
1907 : tcp_req->req.cmd->nvme_cmd.cid, h2c_data->cccid);
1908 0 : fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR;
1909 0 : error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid);
1910 0 : goto err;
1911 : }
1912 :
1913 1 : if (tcp_req->h2c_offset != h2c_data->datao) {
1914 0 : SPDK_DEBUGLOG(nvmf_tcp,
1915 : "tcp_req(%p), tqpair=%p, expected data offset %u, but data offset is %u\n",
1916 : tcp_req, tqpair, tcp_req->h2c_offset, h2c_data->datao);
1917 0 : fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE;
1918 0 : goto err;
1919 : }
1920 :
1921 1 : if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) {
1922 0 : SPDK_DEBUGLOG(nvmf_tcp,
1923 : "tcp_req(%p), tqpair=%p, (datao=%u + datal=%u) exceeds requested length=%u\n",
1924 : tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length);
1925 0 : fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE;
1926 0 : goto err;
1927 : }
1928 :
1929 1 : pdu->req = tcp_req;
1930 :
1931 1 : if (spdk_unlikely(tcp_req->req.dif_enabled)) {
1932 0 : pdu->dif_ctx = &tcp_req->req.dif.dif_ctx;
1933 : }
1934 :
1935 1 : nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt,
1936 : h2c_data->datao, h2c_data->datal);
1937 1 : nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD);
1938 1 : return;
1939 :
1940 0 : err:
1941 0 : nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
1942 : }
1943 :
1944 : static void
1945 3 : nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req,
1946 : struct spdk_nvmf_tcp_qpair *tqpair)
1947 : {
1948 : struct nvme_tcp_pdu *rsp_pdu;
1949 : struct spdk_nvme_tcp_rsp *capsule_resp;
1950 :
1951 3 : SPDK_DEBUGLOG(nvmf_tcp, "enter, tqpair=%p\n", tqpair);
1952 :
1953 3 : rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req);
1954 3 : assert(rsp_pdu != NULL);
1955 :
1956 3 : capsule_resp = &rsp_pdu->hdr.capsule_resp;
1957 3 : capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP;
1958 3 : capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp);
1959 3 : capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl;
1960 3 : if (tqpair->host_hdgst_enable) {
1961 1 : capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF;
1962 1 : capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN;
1963 : }
1964 :
1965 3 : nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_request_free, tcp_req);
1966 3 : }
1967 :
1968 : static void
1969 0 : nvmf_tcp_pdu_c2h_data_complete(void *cb_arg)
1970 : {
1971 0 : struct spdk_nvmf_tcp_req *tcp_req = cb_arg;
1972 0 : struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair,
1973 : struct spdk_nvmf_tcp_qpair, qpair);
1974 :
1975 0 : assert(tqpair != NULL);
1976 :
1977 0 : if (spdk_unlikely(tcp_req->pdu->rw_offset < tcp_req->req.length)) {
1978 0 : SPDK_DEBUGLOG(nvmf_tcp, "sending another C2H part, offset %u length %u\n", tcp_req->pdu->rw_offset,
1979 : tcp_req->req.length);
1980 0 : _nvmf_tcp_send_c2h_data(tqpair, tcp_req);
1981 0 : return;
1982 : }
1983 :
1984 0 : if (tcp_req->pdu->hdr.c2h_data.common.flags & SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS) {
1985 0 : nvmf_tcp_request_free(tcp_req);
1986 : } else {
1987 0 : nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair);
1988 : }
1989 : }
1990 :
1991 : static void
1992 0 : nvmf_tcp_r2t_complete(void *cb_arg)
1993 : {
1994 0 : struct spdk_nvmf_tcp_req *tcp_req = cb_arg;
1995 : struct spdk_nvmf_tcp_transport *ttransport;
1996 :
1997 0 : ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport,
1998 : struct spdk_nvmf_tcp_transport, transport);
1999 :
2000 0 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER);
2001 :
2002 0 : if (tcp_req->h2c_offset == tcp_req->req.length) {
2003 0 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE);
2004 0 : nvmf_tcp_req_process(ttransport, tcp_req);
2005 : }
2006 0 : }
2007 :
2008 : static void
2009 0 : nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair,
2010 : struct spdk_nvmf_tcp_req *tcp_req)
2011 : {
2012 : struct nvme_tcp_pdu *rsp_pdu;
2013 : struct spdk_nvme_tcp_r2t_hdr *r2t;
2014 :
2015 0 : rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req);
2016 0 : assert(rsp_pdu != NULL);
2017 :
2018 0 : r2t = &rsp_pdu->hdr.r2t;
2019 0 : r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T;
2020 0 : r2t->common.plen = r2t->common.hlen = sizeof(*r2t);
2021 :
2022 0 : if (tqpair->host_hdgst_enable) {
2023 0 : r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF;
2024 0 : r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN;
2025 : }
2026 :
2027 0 : r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid;
2028 0 : r2t->ttag = tcp_req->ttag;
2029 0 : r2t->r2to = tcp_req->h2c_offset;
2030 0 : r2t->r2tl = tcp_req->req.length;
2031 :
2032 0 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_R2T_ACK);
2033 :
2034 0 : SPDK_DEBUGLOG(nvmf_tcp,
2035 : "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n",
2036 : tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl);
2037 0 : nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_r2t_complete, tcp_req);
2038 0 : }
2039 :
2040 : static void
2041 0 : nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport,
2042 : struct spdk_nvmf_tcp_qpair *tqpair,
2043 : struct nvme_tcp_pdu *pdu)
2044 : {
2045 : struct spdk_nvmf_tcp_req *tcp_req;
2046 : struct spdk_nvme_cpl *rsp;
2047 :
2048 0 : tcp_req = pdu->req;
2049 0 : assert(tcp_req != NULL);
2050 :
2051 0 : SPDK_DEBUGLOG(nvmf_tcp, "enter\n");
2052 :
2053 0 : tcp_req->h2c_offset += pdu->data_len;
2054 :
2055 : /* Wait for all of the data to arrive AND for the initial R2T PDU send to be
2056 : * acknowledged before moving on. */
2057 0 : if (tcp_req->h2c_offset == tcp_req->req.length &&
2058 0 : tcp_req->state == TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER) {
2059 : /* After receiving all the h2c data, we need to check whether there is
2060 : * transient transport error */
2061 0 : rsp = &tcp_req->req.rsp->nvme_cpl;
2062 0 : if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) {
2063 0 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE);
2064 : } else {
2065 0 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE);
2066 : }
2067 0 : nvmf_tcp_req_process(ttransport, tcp_req);
2068 : }
2069 0 : }
2070 :
2071 : static void
2072 0 : nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req)
2073 : {
2074 0 : SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req,
2075 : spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]);
2076 0 : if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) ||
2077 0 : (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) {
2078 0 : SPDK_DEBUGLOG(nvmf_tcp, "The offset from the start of the PDU header is %u\n",
2079 : DGET32(h2c_term_req->fei));
2080 : }
2081 0 : }
2082 :
2083 : static void
2084 0 : nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair,
2085 : struct nvme_tcp_pdu *pdu)
2086 : {
2087 0 : struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req;
2088 0 : uint32_t error_offset = 0;
2089 : enum spdk_nvme_tcp_term_req_fes fes;
2090 :
2091 0 : if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) {
2092 0 : SPDK_ERRLOG("Fatal Error Status(FES) is unknown for h2c_term_req pdu=%p\n", pdu);
2093 0 : fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
2094 0 : error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes);
2095 0 : goto end;
2096 : }
2097 :
2098 : /* set the data buffer */
2099 0 : nvme_tcp_pdu_set_data(pdu, (uint8_t *)pdu->hdr.raw + h2c_term_req->common.hlen,
2100 0 : h2c_term_req->common.plen - h2c_term_req->common.hlen);
2101 0 : nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD);
2102 0 : return;
2103 0 : end:
2104 0 : nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
2105 : }
2106 :
2107 : static void
2108 0 : nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair,
2109 : struct nvme_tcp_pdu *pdu)
2110 : {
2111 0 : struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req;
2112 :
2113 0 : nvmf_tcp_h2c_term_req_dump(h2c_term_req);
2114 0 : nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING);
2115 0 : }
2116 :
2117 : static void
2118 0 : _nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu)
2119 : {
2120 0 : struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport,
2121 : struct spdk_nvmf_tcp_transport, transport);
2122 :
2123 0 : switch (pdu->hdr.common.pdu_type) {
2124 0 : case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD:
2125 0 : nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu);
2126 0 : break;
2127 0 : case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA:
2128 0 : nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu);
2129 0 : break;
2130 :
2131 0 : case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ:
2132 0 : nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu);
2133 0 : break;
2134 :
2135 0 : default:
2136 : /* The code should not go to here */
2137 0 : SPDK_ERRLOG("ERROR pdu type %d\n", pdu->hdr.common.pdu_type);
2138 0 : break;
2139 : }
2140 0 : SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, pdu, slist);
2141 0 : tqpair->tcp_pdu_working_count--;
2142 0 : }
2143 :
2144 : static inline void
2145 1 : nvmf_tcp_req_set_cpl(struct spdk_nvmf_tcp_req *treq, int sct, int sc)
2146 : {
2147 1 : treq->req.rsp->nvme_cpl.status.sct = sct;
2148 1 : treq->req.rsp->nvme_cpl.status.sc = sc;
2149 1 : treq->req.rsp->nvme_cpl.cid = treq->req.cmd->nvme_cmd.cid;
2150 1 : }
2151 :
2152 : static void
2153 0 : data_crc32_calc_done(void *cb_arg, int status)
2154 : {
2155 0 : struct nvme_tcp_pdu *pdu = cb_arg;
2156 0 : struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair;
2157 :
2158 : /* async crc32 calculation is failed and use direct calculation to check */
2159 0 : if (spdk_unlikely(status)) {
2160 0 : SPDK_ERRLOG("Data digest on tqpair=(%p) with pdu=%p failed to be calculated asynchronously\n",
2161 : tqpair, pdu);
2162 0 : pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu);
2163 : }
2164 0 : pdu->data_digest_crc32 ^= SPDK_CRC32C_XOR;
2165 0 : if (!MATCH_DIGEST_WORD(pdu->data_digest, pdu->data_digest_crc32)) {
2166 0 : SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu);
2167 0 : assert(pdu->req != NULL);
2168 0 : nvmf_tcp_req_set_cpl(pdu->req, SPDK_NVME_SCT_GENERIC,
2169 : SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR);
2170 : }
2171 0 : _nvmf_tcp_pdu_payload_handle(tqpair, pdu);
2172 0 : }
2173 :
2174 : static void
2175 0 : nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu)
2176 : {
2177 0 : int rc = 0;
2178 0 : assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD);
2179 0 : tqpair->pdu_in_progress = NULL;
2180 0 : nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY);
2181 0 : SPDK_DEBUGLOG(nvmf_tcp, "enter\n");
2182 : /* check data digest if need */
2183 0 : if (pdu->ddgst_enable) {
2184 0 : if (tqpair->qpair.qid != 0 && !pdu->dif_ctx && tqpair->group &&
2185 0 : (pdu->data_len % SPDK_NVME_TCP_DIGEST_ALIGNMENT == 0)) {
2186 0 : rc = spdk_accel_submit_crc32cv(tqpair->group->accel_channel, &pdu->data_digest_crc32, pdu->data_iov,
2187 : pdu->data_iovcnt, 0, data_crc32_calc_done, pdu);
2188 0 : if (spdk_likely(rc == 0)) {
2189 0 : return;
2190 : }
2191 : } else {
2192 0 : pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu);
2193 : }
2194 0 : data_crc32_calc_done(pdu, rc);
2195 : } else {
2196 0 : _nvmf_tcp_pdu_payload_handle(tqpair, pdu);
2197 : }
2198 : }
2199 :
2200 : static void
2201 0 : nvmf_tcp_send_icresp_complete(void *cb_arg)
2202 : {
2203 0 : struct spdk_nvmf_tcp_qpair *tqpair = cb_arg;
2204 :
2205 0 : nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_RUNNING);
2206 0 : }
2207 :
2208 : static void
2209 3 : nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport,
2210 : struct spdk_nvmf_tcp_qpair *tqpair,
2211 : struct nvme_tcp_pdu *pdu)
2212 : {
2213 3 : struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr.ic_req;
2214 : struct nvme_tcp_pdu *rsp_pdu;
2215 : struct spdk_nvme_tcp_ic_resp *ic_resp;
2216 3 : uint32_t error_offset = 0;
2217 : enum spdk_nvme_tcp_term_req_fes fes;
2218 :
2219 : /* Only PFV 0 is defined currently */
2220 3 : if (ic_req->pfv != 0) {
2221 2 : SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv);
2222 2 : fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
2223 2 : error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv);
2224 2 : goto end;
2225 : }
2226 :
2227 : /* This value is 0’s based value in units of dwords should not be larger than SPDK_NVME_TCP_HPDA_MAX */
2228 1 : if (ic_req->hpda > SPDK_NVME_TCP_HPDA_MAX) {
2229 0 : SPDK_ERRLOG("ICReq HPDA out of range 0 to 31, got %u\n", ic_req->hpda);
2230 0 : fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
2231 0 : error_offset = offsetof(struct spdk_nvme_tcp_ic_req, hpda);
2232 0 : goto end;
2233 : }
2234 :
2235 : /* MAXR2T is 0's based */
2236 1 : SPDK_DEBUGLOG(nvmf_tcp, "maxr2t =%u\n", (ic_req->maxr2t + 1u));
2237 :
2238 1 : tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false;
2239 1 : if (!tqpair->host_hdgst_enable) {
2240 1 : tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR;
2241 : }
2242 :
2243 1 : tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false;
2244 1 : if (!tqpair->host_ddgst_enable) {
2245 1 : tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR;
2246 : }
2247 :
2248 1 : tqpair->recv_buf_size = spdk_max(tqpair->recv_buf_size, MIN_SOCK_PIPE_SIZE);
2249 : /* Now that we know whether digests are enabled, properly size the receive buffer */
2250 1 : if (spdk_sock_set_recvbuf(tqpair->sock, tqpair->recv_buf_size) < 0) {
2251 0 : SPDK_WARNLOG("Unable to allocate enough memory for receive buffer on tqpair=%p with size=%d\n",
2252 : tqpair,
2253 : tqpair->recv_buf_size);
2254 : /* Not fatal. */
2255 : }
2256 :
2257 1 : tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX);
2258 1 : SPDK_DEBUGLOG(nvmf_tcp, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda);
2259 :
2260 1 : rsp_pdu = tqpair->mgmt_pdu;
2261 :
2262 1 : ic_resp = &rsp_pdu->hdr.ic_resp;
2263 1 : ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP;
2264 1 : ic_resp->common.hlen = ic_resp->common.plen = sizeof(*ic_resp);
2265 1 : ic_resp->pfv = 0;
2266 1 : ic_resp->cpda = tqpair->cpda;
2267 1 : ic_resp->maxh2cdata = ttransport->transport.opts.max_io_size;
2268 1 : ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0;
2269 1 : ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0;
2270 :
2271 1 : SPDK_DEBUGLOG(nvmf_tcp, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable);
2272 1 : SPDK_DEBUGLOG(nvmf_tcp, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable);
2273 :
2274 1 : nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_INITIALIZING);
2275 1 : nvmf_tcp_qpair_write_mgmt_pdu(tqpair, nvmf_tcp_send_icresp_complete, tqpair);
2276 1 : nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY);
2277 1 : return;
2278 2 : end:
2279 2 : nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
2280 : }
2281 :
2282 : static void
2283 0 : nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair,
2284 : struct spdk_nvmf_tcp_transport *ttransport)
2285 : {
2286 : struct nvme_tcp_pdu *pdu;
2287 : int rc;
2288 0 : uint32_t crc32c, error_offset = 0;
2289 : enum spdk_nvme_tcp_term_req_fes fes;
2290 :
2291 0 : assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH);
2292 0 : pdu = tqpair->pdu_in_progress;
2293 :
2294 0 : SPDK_DEBUGLOG(nvmf_tcp, "pdu type of tqpair(%p) is %d\n", tqpair,
2295 : pdu->hdr.common.pdu_type);
2296 : /* check header digest if needed */
2297 0 : if (pdu->has_hdgst) {
2298 0 : SPDK_DEBUGLOG(nvmf_tcp, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair);
2299 0 : crc32c = nvme_tcp_pdu_calc_header_digest(pdu);
2300 0 : rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, crc32c);
2301 0 : if (rc == 0) {
2302 0 : SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu);
2303 0 : fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR;
2304 0 : nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
2305 0 : return;
2306 :
2307 : }
2308 : }
2309 :
2310 0 : switch (pdu->hdr.common.pdu_type) {
2311 0 : case SPDK_NVME_TCP_PDU_TYPE_IC_REQ:
2312 0 : nvmf_tcp_icreq_handle(ttransport, tqpair, pdu);
2313 0 : break;
2314 0 : case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD:
2315 0 : nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_REQ);
2316 0 : break;
2317 0 : case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA:
2318 0 : nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu);
2319 0 : break;
2320 :
2321 0 : case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ:
2322 0 : nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu);
2323 0 : break;
2324 :
2325 0 : default:
2326 0 : SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress->hdr.common.pdu_type);
2327 0 : fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
2328 0 : error_offset = 1;
2329 0 : nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
2330 0 : break;
2331 : }
2332 : }
2333 :
2334 : static void
2335 11 : nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair)
2336 : {
2337 : struct nvme_tcp_pdu *pdu;
2338 11 : uint32_t error_offset = 0;
2339 : enum spdk_nvme_tcp_term_req_fes fes;
2340 : uint8_t expected_hlen, pdo;
2341 11 : bool plen_error = false, pdo_error = false;
2342 :
2343 11 : assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH);
2344 11 : pdu = tqpair->pdu_in_progress;
2345 11 : assert(pdu);
2346 11 : if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) {
2347 4 : if (tqpair->state != NVME_TCP_QPAIR_STATE_INVALID) {
2348 1 : SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu);
2349 1 : fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR;
2350 1 : goto err;
2351 : }
2352 3 : expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req);
2353 3 : if (pdu->hdr.common.plen != expected_hlen) {
2354 1 : plen_error = true;
2355 : }
2356 : } else {
2357 7 : if (tqpair->state != NVME_TCP_QPAIR_STATE_RUNNING) {
2358 1 : SPDK_ERRLOG("The TCP/IP connection is not negotiated\n");
2359 1 : fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR;
2360 1 : goto err;
2361 : }
2362 :
2363 6 : switch (pdu->hdr.common.pdu_type) {
2364 2 : case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD:
2365 2 : expected_hlen = sizeof(struct spdk_nvme_tcp_cmd);
2366 2 : pdo = pdu->hdr.common.pdo;
2367 2 : if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) {
2368 1 : pdo_error = true;
2369 1 : break;
2370 : }
2371 :
2372 1 : if (pdu->hdr.common.plen < expected_hlen) {
2373 1 : plen_error = true;
2374 : }
2375 1 : break;
2376 2 : case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA:
2377 2 : expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr);
2378 2 : pdo = pdu->hdr.common.pdo;
2379 2 : if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) {
2380 1 : pdo_error = true;
2381 1 : break;
2382 : }
2383 1 : if (pdu->hdr.common.plen < expected_hlen) {
2384 1 : plen_error = true;
2385 : }
2386 1 : break;
2387 :
2388 1 : case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ:
2389 1 : expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr);
2390 1 : if ((pdu->hdr.common.plen <= expected_hlen) ||
2391 0 : (pdu->hdr.common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) {
2392 1 : plen_error = true;
2393 : }
2394 1 : break;
2395 :
2396 1 : default:
2397 1 : SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr.common.pdu_type);
2398 1 : fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
2399 1 : error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type);
2400 1 : goto err;
2401 : }
2402 : }
2403 :
2404 8 : if (pdu->hdr.common.hlen != expected_hlen) {
2405 1 : SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n",
2406 : pdu->hdr.common.pdu_type,
2407 : expected_hlen, pdu->hdr.common.hlen, tqpair);
2408 1 : fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
2409 1 : error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen);
2410 1 : goto err;
2411 7 : } else if (pdo_error) {
2412 2 : fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
2413 2 : error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo);
2414 5 : } else if (plen_error) {
2415 4 : fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
2416 4 : error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen);
2417 4 : goto err;
2418 : } else {
2419 1 : nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH);
2420 1 : nvme_tcp_pdu_calc_psh_len(tqpair->pdu_in_progress, tqpair->host_hdgst_enable);
2421 1 : return;
2422 : }
2423 10 : err:
2424 10 : nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset);
2425 : }
2426 :
2427 : static int
2428 0 : nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair)
2429 : {
2430 0 : int rc = 0;
2431 : struct nvme_tcp_pdu *pdu;
2432 : enum nvme_tcp_pdu_recv_state prev_state;
2433 : uint32_t data_len;
2434 0 : struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport,
2435 : struct spdk_nvmf_tcp_transport, transport);
2436 :
2437 : /* The loop here is to allow for several back-to-back state changes. */
2438 : do {
2439 0 : prev_state = tqpair->recv_state;
2440 0 : SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state);
2441 :
2442 0 : pdu = tqpair->pdu_in_progress;
2443 0 : assert(pdu != NULL ||
2444 : tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY ||
2445 : tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_QUIESCING ||
2446 : tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR);
2447 :
2448 0 : switch (tqpair->recv_state) {
2449 : /* Wait for the common header */
2450 0 : case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY:
2451 0 : if (!pdu) {
2452 0 : pdu = SLIST_FIRST(&tqpair->tcp_pdu_free_queue);
2453 0 : if (spdk_unlikely(!pdu)) {
2454 0 : return NVME_TCP_PDU_IN_PROGRESS;
2455 : }
2456 0 : SLIST_REMOVE_HEAD(&tqpair->tcp_pdu_free_queue, slist);
2457 0 : tqpair->pdu_in_progress = pdu;
2458 0 : tqpair->tcp_pdu_working_count++;
2459 : }
2460 0 : memset(pdu, 0, offsetof(struct nvme_tcp_pdu, qpair));
2461 0 : nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH);
2462 : /* FALLTHROUGH */
2463 0 : case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH:
2464 0 : if (spdk_unlikely(tqpair->state == NVME_TCP_QPAIR_STATE_INITIALIZING)) {
2465 0 : return rc;
2466 : }
2467 :
2468 0 : rc = nvme_tcp_read_data(tqpair->sock,
2469 0 : sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes,
2470 0 : (void *)&pdu->hdr.common + pdu->ch_valid_bytes);
2471 0 : if (rc < 0) {
2472 0 : SPDK_DEBUGLOG(nvmf_tcp, "will disconnect tqpair=%p\n", tqpair);
2473 0 : nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING);
2474 0 : break;
2475 0 : } else if (rc > 0) {
2476 0 : pdu->ch_valid_bytes += rc;
2477 0 : spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, tqpair->qpair.trace_id, rc, 0);
2478 : }
2479 :
2480 0 : if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) {
2481 0 : return NVME_TCP_PDU_IN_PROGRESS;
2482 : }
2483 :
2484 : /* The command header of this PDU has now been read from the socket. */
2485 0 : nvmf_tcp_pdu_ch_handle(tqpair);
2486 0 : break;
2487 : /* Wait for the pdu specific header */
2488 0 : case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH:
2489 0 : rc = nvme_tcp_read_data(tqpair->sock,
2490 0 : pdu->psh_len - pdu->psh_valid_bytes,
2491 0 : (void *)&pdu->hdr.raw + sizeof(struct spdk_nvme_tcp_common_pdu_hdr) + pdu->psh_valid_bytes);
2492 0 : if (rc < 0) {
2493 0 : nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING);
2494 0 : break;
2495 0 : } else if (rc > 0) {
2496 0 : spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, tqpair->qpair.trace_id, rc, 0);
2497 0 : pdu->psh_valid_bytes += rc;
2498 : }
2499 :
2500 0 : if (pdu->psh_valid_bytes < pdu->psh_len) {
2501 0 : return NVME_TCP_PDU_IN_PROGRESS;
2502 : }
2503 :
2504 : /* All header(ch, psh, head digits) of this PDU has now been read from the socket. */
2505 0 : nvmf_tcp_pdu_psh_handle(tqpair, ttransport);
2506 0 : break;
2507 : /* Wait for the req slot */
2508 0 : case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ:
2509 0 : nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu);
2510 0 : break;
2511 : /* Wait for the request processing loop to acquire a buffer for the PDU */
2512 0 : case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_BUF:
2513 0 : break;
2514 0 : case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD:
2515 : /* check whether the data is valid, if not we just return */
2516 0 : if (!pdu->data_len) {
2517 0 : return NVME_TCP_PDU_IN_PROGRESS;
2518 : }
2519 :
2520 0 : data_len = pdu->data_len;
2521 : /* data digest */
2522 0 : if (spdk_unlikely((pdu->hdr.common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) &&
2523 : tqpair->host_ddgst_enable)) {
2524 0 : data_len += SPDK_NVME_TCP_DIGEST_LEN;
2525 0 : pdu->ddgst_enable = true;
2526 : }
2527 :
2528 0 : rc = nvme_tcp_read_payload_data(tqpair->sock, pdu);
2529 0 : if (rc < 0) {
2530 0 : nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING);
2531 0 : break;
2532 : }
2533 0 : pdu->rw_offset += rc;
2534 :
2535 0 : if (pdu->rw_offset < data_len) {
2536 0 : return NVME_TCP_PDU_IN_PROGRESS;
2537 : }
2538 :
2539 : /* Generate and insert DIF to whole data block received if DIF is enabled */
2540 0 : if (spdk_unlikely(pdu->dif_ctx != NULL) &&
2541 0 : spdk_dif_generate_stream(pdu->data_iov, pdu->data_iovcnt, 0, data_len,
2542 : pdu->dif_ctx) != 0) {
2543 0 : SPDK_ERRLOG("DIF generate failed\n");
2544 0 : nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING);
2545 0 : break;
2546 : }
2547 :
2548 : /* All of this PDU has now been read from the socket. */
2549 0 : nvmf_tcp_pdu_payload_handle(tqpair, pdu);
2550 0 : break;
2551 0 : case NVME_TCP_PDU_RECV_STATE_QUIESCING:
2552 0 : if (tqpair->tcp_pdu_working_count != 0) {
2553 0 : return NVME_TCP_PDU_IN_PROGRESS;
2554 : }
2555 0 : nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR);
2556 0 : break;
2557 0 : case NVME_TCP_PDU_RECV_STATE_ERROR:
2558 0 : if (spdk_sock_is_connected(tqpair->sock) && tqpair->wait_terminate) {
2559 0 : return NVME_TCP_PDU_IN_PROGRESS;
2560 : }
2561 0 : return NVME_TCP_PDU_FATAL;
2562 0 : default:
2563 0 : SPDK_ERRLOG("The state(%d) is invalid\n", tqpair->recv_state);
2564 0 : abort();
2565 : break;
2566 : }
2567 0 : } while (tqpair->recv_state != prev_state);
2568 :
2569 0 : return rc;
2570 : }
2571 :
2572 : static inline void *
2573 0 : nvmf_tcp_control_msg_get(struct spdk_nvmf_tcp_control_msg_list *list,
2574 : struct spdk_nvmf_tcp_req *tcp_req)
2575 : {
2576 : struct spdk_nvmf_tcp_control_msg *msg;
2577 :
2578 0 : assert(list);
2579 :
2580 0 : msg = STAILQ_FIRST(&list->free_msgs);
2581 0 : if (!msg) {
2582 0 : SPDK_DEBUGLOG(nvmf_tcp, "Out of control messages\n");
2583 0 : STAILQ_INSERT_TAIL(&list->waiting_for_msg_reqs, tcp_req, control_msg_link);
2584 0 : return NULL;
2585 : }
2586 0 : STAILQ_REMOVE_HEAD(&list->free_msgs, link);
2587 0 : return msg;
2588 : }
2589 :
2590 : static inline void
2591 0 : nvmf_tcp_control_msg_put(struct spdk_nvmf_tcp_control_msg_list *list, void *_msg)
2592 : {
2593 0 : struct spdk_nvmf_tcp_control_msg *msg = _msg;
2594 : struct spdk_nvmf_tcp_req *tcp_req;
2595 : struct spdk_nvmf_tcp_transport *ttransport;
2596 :
2597 0 : assert(list);
2598 0 : STAILQ_INSERT_HEAD(&list->free_msgs, msg, link);
2599 0 : if (!STAILQ_EMPTY(&list->waiting_for_msg_reqs)) {
2600 0 : tcp_req = STAILQ_FIRST(&list->waiting_for_msg_reqs);
2601 0 : STAILQ_REMOVE_HEAD(&list->waiting_for_msg_reqs, control_msg_link);
2602 0 : ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport,
2603 : struct spdk_nvmf_tcp_transport, transport);
2604 0 : nvmf_tcp_req_process(ttransport, tcp_req);
2605 : }
2606 0 : }
2607 :
2608 : static void
2609 3 : nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_req *tcp_req,
2610 : struct spdk_nvmf_transport *transport,
2611 : struct spdk_nvmf_transport_poll_group *group)
2612 : {
2613 3 : struct spdk_nvmf_request *req = &tcp_req->req;
2614 : struct spdk_nvme_cmd *cmd;
2615 : struct spdk_nvme_sgl_descriptor *sgl;
2616 : struct spdk_nvmf_tcp_poll_group *tgroup;
2617 : enum spdk_nvme_tcp_term_req_fes fes;
2618 : struct nvme_tcp_pdu *pdu;
2619 : struct spdk_nvmf_tcp_qpair *tqpair;
2620 3 : uint32_t length, error_offset = 0;
2621 :
2622 3 : cmd = &req->cmd->nvme_cmd;
2623 3 : sgl = &cmd->dptr.sgl1;
2624 :
2625 3 : if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK &&
2626 3 : sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) {
2627 : /* get request length from sgl */
2628 3 : length = sgl->unkeyed.length;
2629 3 : if (spdk_unlikely(length > transport->opts.max_io_size)) {
2630 1 : SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n",
2631 : length, transport->opts.max_io_size);
2632 1 : fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_LIMIT_EXCEEDED;
2633 1 : goto fatal_err;
2634 : }
2635 :
2636 : /* fill request length and populate iovs */
2637 2 : req->length = length;
2638 :
2639 2 : SPDK_DEBUGLOG(nvmf_tcp, "Data requested length= 0x%x\n", length);
2640 :
2641 2 : if (spdk_unlikely(req->dif_enabled)) {
2642 0 : req->dif.orig_length = length;
2643 0 : length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx);
2644 0 : req->dif.elba_length = length;
2645 : }
2646 :
2647 2 : if (nvmf_ctrlr_use_zcopy(req)) {
2648 0 : SPDK_DEBUGLOG(nvmf_tcp, "Using zero-copy to execute request %p\n", tcp_req);
2649 0 : req->data_from_pool = false;
2650 0 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_HAVE_BUFFER);
2651 0 : return;
2652 : }
2653 :
2654 2 : if (spdk_nvmf_request_get_buffers(req, group, transport, length)) {
2655 : /* No available buffers. Queue this request up. */
2656 1 : SPDK_DEBUGLOG(nvmf_tcp, "No available large data buffers. Queueing request %p\n",
2657 : tcp_req);
2658 1 : return;
2659 : }
2660 :
2661 1 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_HAVE_BUFFER);
2662 1 : SPDK_DEBUGLOG(nvmf_tcp, "Request %p took %d buffer/s from central pool, and data=%p\n",
2663 : tcp_req, req->iovcnt, req->iov[0].iov_base);
2664 :
2665 1 : return;
2666 0 : } else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK &&
2667 0 : sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) {
2668 0 : uint64_t offset = sgl->address;
2669 0 : uint32_t max_len = transport->opts.in_capsule_data_size;
2670 :
2671 0 : assert(tcp_req->has_in_capsule_data);
2672 : /* Capsule Cmd with In-capsule Data should get data length from pdu header */
2673 0 : tqpair = tcp_req->pdu->qpair;
2674 : /* receiving pdu is not same with the pdu in tcp_req */
2675 0 : pdu = tqpair->pdu_in_progress;
2676 0 : length = pdu->hdr.common.plen - pdu->psh_len - sizeof(struct spdk_nvme_tcp_common_pdu_hdr);
2677 0 : if (tqpair->host_ddgst_enable) {
2678 0 : length -= SPDK_NVME_TCP_DIGEST_LEN;
2679 : }
2680 : /* This error is not defined in NVMe/TCP spec, take this error as fatal error */
2681 0 : if (spdk_unlikely(length != sgl->unkeyed.length)) {
2682 0 : SPDK_ERRLOG("In-Capsule Data length 0x%x is not equal to SGL data length 0x%x\n",
2683 : length, sgl->unkeyed.length);
2684 0 : fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD;
2685 0 : error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen);
2686 0 : goto fatal_err;
2687 : }
2688 :
2689 0 : SPDK_DEBUGLOG(nvmf_tcp, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n",
2690 : offset, length);
2691 :
2692 : /* The NVMe/TCP transport does not use ICDOFF to control the in-capsule data offset. ICDOFF should be '0' */
2693 0 : if (spdk_unlikely(offset != 0)) {
2694 : /* Not defined fatal error in NVMe/TCP spec, handle this error as a fatal error */
2695 0 : SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " should be ZERO in NVMe/TCP\n", offset);
2696 0 : fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER;
2697 0 : error_offset = offsetof(struct spdk_nvme_tcp_cmd, ccsqe.dptr.sgl1.address);
2698 0 : goto fatal_err;
2699 : }
2700 :
2701 0 : if (spdk_unlikely(length > max_len)) {
2702 : /* According to the SPEC we should support ICD up to 8192 bytes for admin and fabric commands */
2703 0 : if (length <= SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE &&
2704 0 : (cmd->opc == SPDK_NVME_OPC_FABRIC || req->qpair->qid == 0)) {
2705 :
2706 : /* Get a buffer from dedicated list */
2707 0 : SPDK_DEBUGLOG(nvmf_tcp, "Getting a buffer from control msg list\n");
2708 0 : tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group);
2709 0 : assert(tgroup->control_msg_list);
2710 0 : req->iov[0].iov_base = nvmf_tcp_control_msg_get(tgroup->control_msg_list, tcp_req);
2711 0 : if (!req->iov[0].iov_base) {
2712 : /* No available buffers. Queue this request up. */
2713 0 : SPDK_DEBUGLOG(nvmf_tcp, "No available ICD buffers. Queueing request %p\n", tcp_req);
2714 0 : nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_BUF);
2715 0 : return;
2716 : }
2717 : } else {
2718 0 : SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n",
2719 : length, max_len);
2720 0 : fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_LIMIT_EXCEEDED;
2721 0 : goto fatal_err;
2722 : }
2723 : } else {
2724 0 : req->iov[0].iov_base = tcp_req->buf;
2725 : }
2726 :
2727 0 : req->length = length;
2728 0 : req->data_from_pool = false;
2729 :
2730 0 : if (spdk_unlikely(req->dif_enabled)) {
2731 0 : length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx);
2732 0 : req->dif.elba_length = length;
2733 : }
2734 :
2735 0 : req->iov[0].iov_len = length;
2736 0 : req->iovcnt = 1;
2737 0 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_HAVE_BUFFER);
2738 :
2739 0 : return;
2740 : }
2741 : /* If we want to handle the problem here, then we can't skip the following data segment.
2742 : * Because this function runs before reading data part, now handle all errors as fatal errors. */
2743 0 : SPDK_ERRLOG("Invalid NVMf I/O Command SGL: Type 0x%x, Subtype 0x%x\n",
2744 : sgl->generic.type, sgl->generic.subtype);
2745 0 : fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER;
2746 0 : error_offset = offsetof(struct spdk_nvme_tcp_cmd, ccsqe.dptr.sgl1.generic);
2747 1 : fatal_err:
2748 1 : nvmf_tcp_send_c2h_term_req(tcp_req->pdu->qpair, tcp_req->pdu, fes, error_offset);
2749 : }
2750 :
2751 : static inline enum spdk_nvme_media_error_status_code
2752 0 : nvmf_tcp_dif_error_to_compl_status(uint8_t err_type) {
2753 : enum spdk_nvme_media_error_status_code result;
2754 :
2755 0 : switch (err_type)
2756 : {
2757 0 : case SPDK_DIF_REFTAG_ERROR:
2758 0 : result = SPDK_NVME_SC_REFERENCE_TAG_CHECK_ERROR;
2759 0 : break;
2760 0 : case SPDK_DIF_APPTAG_ERROR:
2761 0 : result = SPDK_NVME_SC_APPLICATION_TAG_CHECK_ERROR;
2762 0 : break;
2763 0 : case SPDK_DIF_GUARD_ERROR:
2764 0 : result = SPDK_NVME_SC_GUARD_CHECK_ERROR;
2765 0 : break;
2766 0 : default:
2767 0 : SPDK_UNREACHABLE();
2768 : break;
2769 : }
2770 :
2771 0 : return result;
2772 : }
2773 :
2774 : static void
2775 4 : _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair,
2776 : struct spdk_nvmf_tcp_req *tcp_req)
2777 : {
2778 4 : struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(
2779 : tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport);
2780 : struct nvme_tcp_pdu *rsp_pdu;
2781 : struct spdk_nvme_tcp_c2h_data_hdr *c2h_data;
2782 : uint32_t plen, pdo, alignment;
2783 : int rc;
2784 :
2785 4 : SPDK_DEBUGLOG(nvmf_tcp, "enter\n");
2786 :
2787 4 : rsp_pdu = tcp_req->pdu;
2788 4 : assert(rsp_pdu != NULL);
2789 :
2790 4 : c2h_data = &rsp_pdu->hdr.c2h_data;
2791 4 : c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA;
2792 4 : plen = c2h_data->common.hlen = sizeof(*c2h_data);
2793 :
2794 4 : if (tqpair->host_hdgst_enable) {
2795 0 : plen += SPDK_NVME_TCP_DIGEST_LEN;
2796 0 : c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF;
2797 : }
2798 :
2799 : /* set the psh */
2800 4 : c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid;
2801 4 : c2h_data->datal = tcp_req->req.length - tcp_req->pdu->rw_offset;
2802 4 : c2h_data->datao = tcp_req->pdu->rw_offset;
2803 :
2804 : /* set the padding */
2805 4 : rsp_pdu->padding_len = 0;
2806 4 : pdo = plen;
2807 4 : if (tqpair->cpda) {
2808 0 : alignment = (tqpair->cpda + 1) << 2;
2809 0 : if (plen % alignment != 0) {
2810 0 : pdo = (plen + alignment) / alignment * alignment;
2811 0 : rsp_pdu->padding_len = pdo - plen;
2812 0 : plen = pdo;
2813 : }
2814 : }
2815 :
2816 4 : c2h_data->common.pdo = pdo;
2817 4 : plen += c2h_data->datal;
2818 4 : if (tqpair->host_ddgst_enable) {
2819 0 : c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF;
2820 0 : plen += SPDK_NVME_TCP_DIGEST_LEN;
2821 : }
2822 :
2823 4 : c2h_data->common.plen = plen;
2824 :
2825 4 : if (spdk_unlikely(tcp_req->req.dif_enabled)) {
2826 0 : rsp_pdu->dif_ctx = &tcp_req->req.dif.dif_ctx;
2827 : }
2828 :
2829 4 : nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt,
2830 : c2h_data->datao, c2h_data->datal);
2831 :
2832 :
2833 4 : c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU;
2834 : /* Need to send the capsule response if response is not all 0 */
2835 4 : if (ttransport->tcp_opts.c2h_success &&
2836 2 : tcp_req->rsp.cdw0 == 0 && tcp_req->rsp.cdw1 == 0) {
2837 1 : c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS;
2838 : }
2839 :
2840 4 : if (spdk_unlikely(tcp_req->req.dif_enabled)) {
2841 0 : struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl;
2842 0 : struct spdk_dif_error err_blk = {};
2843 0 : uint32_t mapped_length = 0;
2844 0 : uint32_t available_iovs = SPDK_COUNTOF(rsp_pdu->iov);
2845 0 : uint32_t ddgst_len = 0;
2846 :
2847 0 : if (tqpair->host_ddgst_enable) {
2848 : /* Data digest consumes additional iov entry */
2849 0 : available_iovs--;
2850 : /* plen needs to be updated since nvme_tcp_build_iovs compares expected and actual plen */
2851 0 : ddgst_len = SPDK_NVME_TCP_DIGEST_LEN;
2852 0 : c2h_data->common.plen -= ddgst_len;
2853 : }
2854 : /* Temp call to estimate if data can be described by limited number of iovs.
2855 : * iov vector will be rebuilt in nvmf_tcp_qpair_write_pdu */
2856 0 : nvme_tcp_build_iovs(rsp_pdu->iov, available_iovs, rsp_pdu, tqpair->host_hdgst_enable,
2857 : false, &mapped_length);
2858 :
2859 0 : if (mapped_length != c2h_data->common.plen) {
2860 0 : c2h_data->datal = mapped_length - (c2h_data->common.plen - c2h_data->datal);
2861 0 : SPDK_DEBUGLOG(nvmf_tcp,
2862 : "Part C2H, data_len %u (of %u), PDU len %u, updated PDU len %u, offset %u\n",
2863 : c2h_data->datal, tcp_req->req.length, c2h_data->common.plen, mapped_length, rsp_pdu->rw_offset);
2864 0 : c2h_data->common.plen = mapped_length;
2865 :
2866 : /* Rebuild pdu->data_iov since data length is changed */
2867 0 : nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, c2h_data->datao,
2868 : c2h_data->datal);
2869 :
2870 0 : c2h_data->common.flags &= ~(SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU |
2871 : SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS);
2872 : }
2873 :
2874 0 : c2h_data->common.plen += ddgst_len;
2875 :
2876 0 : assert(rsp_pdu->rw_offset <= tcp_req->req.length);
2877 :
2878 0 : rc = spdk_dif_verify_stream(rsp_pdu->data_iov, rsp_pdu->data_iovcnt,
2879 : 0, rsp_pdu->data_len, rsp_pdu->dif_ctx, &err_blk);
2880 0 : if (rc != 0) {
2881 0 : SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n",
2882 : err_blk.err_type, err_blk.err_offset);
2883 0 : rsp->status.sct = SPDK_NVME_SCT_MEDIA_ERROR;
2884 0 : rsp->status.sc = nvmf_tcp_dif_error_to_compl_status(err_blk.err_type);
2885 0 : nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair);
2886 0 : return;
2887 : }
2888 : }
2889 :
2890 4 : rsp_pdu->rw_offset += c2h_data->datal;
2891 4 : nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_pdu_c2h_data_complete, tcp_req);
2892 : }
2893 :
2894 : static void
2895 4 : nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair,
2896 : struct spdk_nvmf_tcp_req *tcp_req)
2897 : {
2898 4 : nvmf_tcp_req_pdu_init(tcp_req);
2899 4 : _nvmf_tcp_send_c2h_data(tqpair, tcp_req);
2900 4 : }
2901 :
2902 : static int
2903 1 : request_transfer_out(struct spdk_nvmf_request *req)
2904 : {
2905 : struct spdk_nvmf_tcp_req *tcp_req;
2906 : struct spdk_nvmf_qpair *qpair;
2907 : struct spdk_nvmf_tcp_qpair *tqpair;
2908 : struct spdk_nvme_cpl *rsp;
2909 :
2910 1 : SPDK_DEBUGLOG(nvmf_tcp, "enter\n");
2911 :
2912 1 : qpair = req->qpair;
2913 1 : rsp = &req->rsp->nvme_cpl;
2914 1 : tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req);
2915 :
2916 : /* Advance our sq_head pointer */
2917 1 : if (qpair->sq_head == qpair->sq_head_max) {
2918 1 : qpair->sq_head = 0;
2919 : } else {
2920 0 : qpair->sq_head++;
2921 : }
2922 1 : rsp->sqhd = qpair->sq_head;
2923 :
2924 1 : tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair);
2925 1 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST);
2926 1 : if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) {
2927 0 : nvmf_tcp_send_c2h_data(tqpair, tcp_req);
2928 : } else {
2929 1 : nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair);
2930 : }
2931 :
2932 1 : return 0;
2933 : }
2934 :
2935 : static void
2936 4 : nvmf_tcp_check_fused_ordering(struct spdk_nvmf_tcp_transport *ttransport,
2937 : struct spdk_nvmf_tcp_qpair *tqpair,
2938 : struct spdk_nvmf_tcp_req *tcp_req)
2939 : {
2940 : enum spdk_nvme_cmd_fuse last, next;
2941 :
2942 4 : last = tqpair->fused_first ? tqpair->fused_first->cmd.fuse : SPDK_NVME_CMD_FUSE_NONE;
2943 4 : next = tcp_req->cmd.fuse;
2944 :
2945 4 : assert(last != SPDK_NVME_CMD_FUSE_SECOND);
2946 :
2947 4 : if (spdk_likely(last == SPDK_NVME_CMD_FUSE_NONE && next == SPDK_NVME_CMD_FUSE_NONE)) {
2948 4 : return;
2949 : }
2950 :
2951 0 : if (last == SPDK_NVME_CMD_FUSE_FIRST) {
2952 0 : if (next == SPDK_NVME_CMD_FUSE_SECOND) {
2953 : /* This is a valid pair of fused commands. Point them at each other
2954 : * so they can be submitted consecutively once ready to be executed.
2955 : */
2956 0 : tqpair->fused_first->fused_pair = tcp_req;
2957 0 : tcp_req->fused_pair = tqpair->fused_first;
2958 0 : tqpair->fused_first = NULL;
2959 0 : return;
2960 : } else {
2961 : /* Mark the last req as failed since it wasn't followed by a SECOND. */
2962 0 : tqpair->fused_first->fused_failed = true;
2963 :
2964 : /*
2965 : * If the last req is in READY_TO_EXECUTE state, then call
2966 : * nvmf_tcp_req_process(), otherwise nothing else will kick it.
2967 : */
2968 0 : if (tqpair->fused_first->state == TCP_REQUEST_STATE_READY_TO_EXECUTE) {
2969 0 : nvmf_tcp_req_process(ttransport, tqpair->fused_first);
2970 : }
2971 :
2972 0 : tqpair->fused_first = NULL;
2973 : }
2974 : }
2975 :
2976 0 : if (next == SPDK_NVME_CMD_FUSE_FIRST) {
2977 : /* Set tqpair->fused_first here so that we know to check that the next request
2978 : * is a SECOND (and to fail this one if it isn't).
2979 : */
2980 0 : tqpair->fused_first = tcp_req;
2981 0 : } else if (next == SPDK_NVME_CMD_FUSE_SECOND) {
2982 : /* Mark this req failed since it is a SECOND and the last one was not a FIRST. */
2983 0 : tcp_req->fused_failed = true;
2984 : }
2985 : }
2986 :
2987 : static bool
2988 5 : nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport,
2989 : struct spdk_nvmf_tcp_req *tcp_req)
2990 : {
2991 : struct spdk_nvmf_tcp_qpair *tqpair;
2992 : uint32_t plen;
2993 : struct nvme_tcp_pdu *pdu;
2994 : enum spdk_nvmf_tcp_req_state prev_state;
2995 5 : bool progress = false;
2996 5 : struct spdk_nvmf_transport *transport = &ttransport->transport;
2997 : struct spdk_nvmf_transport_poll_group *group;
2998 : struct spdk_nvmf_tcp_poll_group *tgroup;
2999 :
3000 5 : tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair);
3001 5 : group = &tqpair->group->group;
3002 5 : assert(tcp_req->state != TCP_REQUEST_STATE_FREE);
3003 :
3004 : /* If the qpair is not active, we need to abort the outstanding requests. */
3005 5 : if (!spdk_nvmf_qpair_is_active(&tqpair->qpair)) {
3006 0 : if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) {
3007 0 : nvmf_tcp_request_get_buffers_abort(tcp_req);
3008 : }
3009 0 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED);
3010 : }
3011 :
3012 : /* The loop here is to allow for several back-to-back state changes. */
3013 : do {
3014 13 : prev_state = tcp_req->state;
3015 :
3016 13 : SPDK_DEBUGLOG(nvmf_tcp, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state,
3017 : tqpair);
3018 :
3019 13 : switch (tcp_req->state) {
3020 0 : case TCP_REQUEST_STATE_FREE:
3021 : /* Some external code must kick a request into TCP_REQUEST_STATE_NEW
3022 : * to escape this state. */
3023 0 : break;
3024 4 : case TCP_REQUEST_STATE_NEW:
3025 4 : spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req,
3026 : tqpair->qpair.queue_depth);
3027 :
3028 : /* copy the cmd from the receive pdu */
3029 4 : tcp_req->cmd = tqpair->pdu_in_progress->hdr.capsule_cmd.ccsqe;
3030 :
3031 4 : if (spdk_unlikely(spdk_nvmf_request_get_dif_ctx(&tcp_req->req, &tcp_req->req.dif.dif_ctx))) {
3032 0 : tcp_req->req.dif_enabled = true;
3033 0 : tqpair->pdu_in_progress->dif_ctx = &tcp_req->req.dif.dif_ctx;
3034 : }
3035 :
3036 4 : nvmf_tcp_check_fused_ordering(ttransport, tqpair, tcp_req);
3037 :
3038 : /* The next state transition depends on the data transfer needs of this request. */
3039 4 : tcp_req->req.xfer = spdk_nvmf_req_get_xfer(&tcp_req->req);
3040 :
3041 4 : if (spdk_unlikely(tcp_req->req.xfer == SPDK_NVME_DATA_BIDIRECTIONAL)) {
3042 1 : nvmf_tcp_req_set_cpl(tcp_req, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_INVALID_OPCODE);
3043 1 : nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY);
3044 1 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE);
3045 1 : SPDK_DEBUGLOG(nvmf_tcp, "Request %p: invalid xfer type (BIDIRECTIONAL)\n", tcp_req);
3046 1 : break;
3047 : }
3048 :
3049 : /* If no data to transfer, ready to execute. */
3050 3 : if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) {
3051 : /* Reset the tqpair receiving pdu state */
3052 0 : nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY);
3053 0 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE);
3054 0 : break;
3055 : }
3056 :
3057 3 : pdu = tqpair->pdu_in_progress;
3058 3 : plen = pdu->hdr.common.hlen;
3059 3 : if (tqpair->host_hdgst_enable) {
3060 0 : plen += SPDK_NVME_TCP_DIGEST_LEN;
3061 : }
3062 3 : if (pdu->hdr.common.plen != plen) {
3063 3 : tcp_req->has_in_capsule_data = true;
3064 : } else {
3065 : /* Data is transmitted by C2H PDUs */
3066 0 : nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY);
3067 : }
3068 :
3069 3 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER);
3070 3 : break;
3071 3 : case TCP_REQUEST_STATE_NEED_BUFFER:
3072 3 : spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, tqpair->qpair.trace_id, 0,
3073 : (uintptr_t)tcp_req);
3074 :
3075 3 : assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE);
3076 :
3077 : /* Try to get a data buffer */
3078 3 : nvmf_tcp_req_parse_sgl(tcp_req, transport, group);
3079 3 : break;
3080 2 : case TCP_REQUEST_STATE_HAVE_BUFFER:
3081 2 : spdk_trace_record(TRACE_TCP_REQUEST_STATE_HAVE_BUFFER, tqpair->qpair.trace_id, 0,
3082 : (uintptr_t)tcp_req);
3083 : /* Get a zcopy buffer if the request can be serviced through zcopy */
3084 2 : if (spdk_nvmf_request_using_zcopy(&tcp_req->req)) {
3085 0 : if (spdk_unlikely(tcp_req->req.dif_enabled)) {
3086 0 : assert(tcp_req->req.dif.elba_length >= tcp_req->req.length);
3087 0 : tcp_req->req.length = tcp_req->req.dif.elba_length;
3088 : }
3089 :
3090 0 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_START);
3091 0 : spdk_nvmf_request_zcopy_start(&tcp_req->req);
3092 0 : break;
3093 : }
3094 :
3095 2 : assert(tcp_req->req.iovcnt > 0);
3096 :
3097 : /* If data is transferring from host to controller, we need to do a transfer from the host. */
3098 2 : if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) {
3099 2 : if (tcp_req->req.data_from_pool) {
3100 0 : SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair);
3101 0 : nvmf_tcp_send_r2t_pdu(tqpair, tcp_req);
3102 : } else {
3103 : struct nvme_tcp_pdu *pdu;
3104 :
3105 2 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER);
3106 :
3107 2 : pdu = tqpair->pdu_in_progress;
3108 2 : SPDK_DEBUGLOG(nvmf_tcp, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req,
3109 : tqpair);
3110 : /* No need to send r2t, contained in the capsuled data */
3111 2 : nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt,
3112 : 0, tcp_req->req.length);
3113 2 : nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD);
3114 : }
3115 2 : break;
3116 : }
3117 :
3118 0 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE);
3119 0 : break;
3120 0 : case TCP_REQUEST_STATE_AWAITING_ZCOPY_START:
3121 0 : spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_START, tqpair->qpair.trace_id, 0,
3122 : (uintptr_t)tcp_req);
3123 : /* Some external code must kick a request into TCP_REQUEST_STATE_ZCOPY_START_COMPLETED
3124 : * to escape this state. */
3125 0 : break;
3126 0 : case TCP_REQUEST_STATE_ZCOPY_START_COMPLETED:
3127 0 : spdk_trace_record(TRACE_TCP_REQUEST_STATE_ZCOPY_START_COMPLETED, tqpair->qpair.trace_id, 0,
3128 : (uintptr_t)tcp_req);
3129 0 : if (spdk_unlikely(spdk_nvme_cpl_is_error(&tcp_req->req.rsp->nvme_cpl))) {
3130 0 : SPDK_DEBUGLOG(nvmf_tcp, "Zero-copy start failed for tcp_req(%p) on tqpair=%p\n",
3131 : tcp_req, tqpair);
3132 0 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE);
3133 0 : break;
3134 : }
3135 0 : if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) {
3136 0 : SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair);
3137 0 : nvmf_tcp_send_r2t_pdu(tqpair, tcp_req);
3138 : } else {
3139 0 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED);
3140 : }
3141 0 : break;
3142 0 : case TCP_REQUEST_STATE_AWAITING_R2T_ACK:
3143 0 : spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, tqpair->qpair.trace_id, 0,
3144 : (uintptr_t)tcp_req);
3145 : /* The R2T completion or the h2c data incoming will kick it out of this state. */
3146 0 : break;
3147 2 : case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER:
3148 :
3149 2 : spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, tqpair->qpair.trace_id,
3150 : 0, (uintptr_t)tcp_req);
3151 : /* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE
3152 : * to escape this state. */
3153 2 : break;
3154 0 : case TCP_REQUEST_STATE_READY_TO_EXECUTE:
3155 0 : spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, tqpair->qpair.trace_id, 0,
3156 : (uintptr_t)tcp_req);
3157 :
3158 0 : if (spdk_unlikely(tcp_req->req.dif_enabled)) {
3159 0 : assert(tcp_req->req.dif.elba_length >= tcp_req->req.length);
3160 0 : tcp_req->req.length = tcp_req->req.dif.elba_length;
3161 : }
3162 :
3163 0 : if (tcp_req->cmd.fuse != SPDK_NVME_CMD_FUSE_NONE) {
3164 0 : if (tcp_req->fused_failed) {
3165 : /* This request failed FUSED semantics. Fail it immediately, without
3166 : * even sending it to the target layer.
3167 : */
3168 0 : nvmf_tcp_req_set_cpl(tcp_req, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_ABORTED_MISSING_FUSED);
3169 0 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE);
3170 0 : break;
3171 : }
3172 :
3173 0 : if (tcp_req->fused_pair == NULL ||
3174 0 : tcp_req->fused_pair->state != TCP_REQUEST_STATE_READY_TO_EXECUTE) {
3175 : /* This request is ready to execute, but either we don't know yet if it's
3176 : * valid - i.e. this is a FIRST but we haven't received the next request yet),
3177 : * or the other request of this fused pair isn't ready to execute. So
3178 : * break here and this request will get processed later either when the
3179 : * other request is ready or we find that this request isn't valid.
3180 : */
3181 : break;
3182 : }
3183 : }
3184 :
3185 0 : if (!spdk_nvmf_request_using_zcopy(&tcp_req->req)) {
3186 0 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING);
3187 : /* If we get to this point, and this request is a fused command, we know that
3188 : * it is part of a valid sequence (FIRST followed by a SECOND) and that both
3189 : * requests are READY_TO_EXECUTE. So call spdk_nvmf_request_exec() both on this
3190 : * request, and the other request of the fused pair, in the correct order.
3191 : * Also clear the ->fused_pair pointers on both requests, since after this point
3192 : * we no longer need to maintain the relationship between these two requests.
3193 : */
3194 0 : if (tcp_req->cmd.fuse == SPDK_NVME_CMD_FUSE_SECOND) {
3195 0 : assert(tcp_req->fused_pair != NULL);
3196 0 : assert(tcp_req->fused_pair->fused_pair == tcp_req);
3197 0 : nvmf_tcp_req_set_state(tcp_req->fused_pair, TCP_REQUEST_STATE_EXECUTING);
3198 0 : spdk_nvmf_request_exec(&tcp_req->fused_pair->req);
3199 0 : tcp_req->fused_pair->fused_pair = NULL;
3200 0 : tcp_req->fused_pair = NULL;
3201 : }
3202 0 : spdk_nvmf_request_exec(&tcp_req->req);
3203 0 : if (tcp_req->cmd.fuse == SPDK_NVME_CMD_FUSE_FIRST) {
3204 0 : assert(tcp_req->fused_pair != NULL);
3205 0 : assert(tcp_req->fused_pair->fused_pair == tcp_req);
3206 0 : nvmf_tcp_req_set_state(tcp_req->fused_pair, TCP_REQUEST_STATE_EXECUTING);
3207 0 : spdk_nvmf_request_exec(&tcp_req->fused_pair->req);
3208 0 : tcp_req->fused_pair->fused_pair = NULL;
3209 0 : tcp_req->fused_pair = NULL;
3210 : }
3211 : } else {
3212 : /* For zero-copy, only requests with data coming from host to the
3213 : * controller can end up here. */
3214 0 : assert(tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER);
3215 0 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT);
3216 0 : spdk_nvmf_request_zcopy_end(&tcp_req->req, true);
3217 : }
3218 :
3219 0 : break;
3220 0 : case TCP_REQUEST_STATE_EXECUTING:
3221 0 : spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req);
3222 : /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED
3223 : * to escape this state. */
3224 0 : break;
3225 0 : case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT:
3226 0 : spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_COMMIT, tqpair->qpair.trace_id, 0,
3227 : (uintptr_t)tcp_req);
3228 : /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED
3229 : * to escape this state. */
3230 0 : break;
3231 0 : case TCP_REQUEST_STATE_EXECUTED:
3232 0 : spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req);
3233 :
3234 0 : if (spdk_unlikely(tcp_req->req.dif_enabled)) {
3235 0 : tcp_req->req.length = tcp_req->req.dif.orig_length;
3236 : }
3237 :
3238 0 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE);
3239 0 : break;
3240 1 : case TCP_REQUEST_STATE_READY_TO_COMPLETE:
3241 1 : spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, tqpair->qpair.trace_id, 0,
3242 : (uintptr_t)tcp_req);
3243 1 : if (request_transfer_out(&tcp_req->req) != 0) {
3244 0 : assert(0); /* No good way to handle this currently */
3245 : }
3246 1 : break;
3247 1 : case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST:
3248 1 : spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, tqpair->qpair.trace_id,
3249 : 0, (uintptr_t)tcp_req);
3250 : /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED
3251 : * to escape this state. */
3252 1 : break;
3253 0 : case TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE:
3254 0 : spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_RELEASE, tqpair->qpair.trace_id, 0,
3255 : (uintptr_t)tcp_req);
3256 : /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED
3257 : * to escape this state. */
3258 0 : break;
3259 0 : case TCP_REQUEST_STATE_COMPLETED:
3260 0 : spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req,
3261 : tqpair->qpair.queue_depth);
3262 : /* If there's an outstanding PDU sent to the host, the request is completed
3263 : * due to the qpair being disconnected. We must delay the completion until
3264 : * that write is done to avoid freeing the request twice. */
3265 0 : if (spdk_unlikely(tcp_req->pdu_in_use)) {
3266 0 : SPDK_DEBUGLOG(nvmf_tcp, "Delaying completion due to outstanding "
3267 : "write on req=%p\n", tcp_req);
3268 : /* This can only happen for zcopy requests */
3269 0 : assert(spdk_nvmf_request_using_zcopy(&tcp_req->req));
3270 0 : assert(!spdk_nvmf_qpair_is_active(&tqpair->qpair));
3271 0 : break;
3272 : }
3273 :
3274 0 : if (tcp_req->req.data_from_pool) {
3275 0 : spdk_nvmf_request_free_buffers(&tcp_req->req, group, transport);
3276 0 : } else if (spdk_unlikely(tcp_req->has_in_capsule_data &&
3277 : (tcp_req->cmd.opc == SPDK_NVME_OPC_FABRIC ||
3278 : tqpair->qpair.qid == 0) && tcp_req->req.length > transport->opts.in_capsule_data_size)) {
3279 0 : tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group);
3280 0 : assert(tgroup->control_msg_list);
3281 0 : SPDK_DEBUGLOG(nvmf_tcp, "Put buf to control msg list\n");
3282 0 : nvmf_tcp_control_msg_put(tgroup->control_msg_list,
3283 : tcp_req->req.iov[0].iov_base);
3284 0 : } else if (tcp_req->req.zcopy_bdev_io != NULL) {
3285 : /* If the request has an unreleased zcopy bdev_io, it's either a
3286 : * read, a failed write, or the qpair is being disconnected */
3287 0 : assert(spdk_nvmf_request_using_zcopy(&tcp_req->req));
3288 0 : assert(tcp_req->req.xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST ||
3289 : spdk_nvme_cpl_is_error(&tcp_req->req.rsp->nvme_cpl) ||
3290 : !spdk_nvmf_qpair_is_active(&tqpair->qpair));
3291 0 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE);
3292 0 : spdk_nvmf_request_zcopy_end(&tcp_req->req, false);
3293 0 : break;
3294 : }
3295 0 : tcp_req->req.length = 0;
3296 0 : tcp_req->req.iovcnt = 0;
3297 0 : tcp_req->fused_failed = false;
3298 0 : if (tcp_req->fused_pair) {
3299 : /* This req was part of a valid fused pair, but failed before it got to
3300 : * READ_TO_EXECUTE state. This means we need to fail the other request
3301 : * in the pair, because it is no longer part of a valid pair. If the pair
3302 : * already reached READY_TO_EXECUTE state, we need to kick it.
3303 : */
3304 0 : tcp_req->fused_pair->fused_failed = true;
3305 0 : if (tcp_req->fused_pair->state == TCP_REQUEST_STATE_READY_TO_EXECUTE) {
3306 0 : nvmf_tcp_req_process(ttransport, tcp_req->fused_pair);
3307 : }
3308 0 : tcp_req->fused_pair = NULL;
3309 : }
3310 :
3311 0 : nvmf_tcp_req_put(tqpair, tcp_req);
3312 0 : break;
3313 0 : case TCP_REQUEST_NUM_STATES:
3314 : default:
3315 0 : assert(0);
3316 : break;
3317 : }
3318 :
3319 13 : if (tcp_req->state != prev_state) {
3320 8 : progress = true;
3321 : }
3322 13 : } while (tcp_req->state != prev_state);
3323 :
3324 5 : return progress;
3325 : }
3326 :
3327 : static void
3328 0 : nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock)
3329 : {
3330 0 : struct spdk_nvmf_tcp_qpair *tqpair = arg;
3331 : int rc;
3332 :
3333 0 : assert(tqpair != NULL);
3334 0 : rc = nvmf_tcp_sock_process(tqpair);
3335 :
3336 : /* If there was a new socket error, disconnect */
3337 0 : if (rc < 0) {
3338 0 : nvmf_tcp_qpair_disconnect(tqpair);
3339 : }
3340 0 : }
3341 :
3342 : static int
3343 0 : nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group,
3344 : struct spdk_nvmf_qpair *qpair)
3345 : {
3346 : struct spdk_nvmf_tcp_poll_group *tgroup;
3347 : struct spdk_nvmf_tcp_qpair *tqpair;
3348 : int rc;
3349 :
3350 0 : tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group);
3351 0 : tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
3352 :
3353 0 : rc = nvmf_tcp_qpair_sock_init(tqpair);
3354 0 : if (rc != 0) {
3355 0 : SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair);
3356 0 : return -1;
3357 : }
3358 :
3359 0 : rc = nvmf_tcp_qpair_init(&tqpair->qpair);
3360 0 : if (rc < 0) {
3361 0 : SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair);
3362 0 : return -1;
3363 : }
3364 :
3365 0 : rc = nvmf_tcp_qpair_init_mem_resource(tqpair);
3366 0 : if (rc < 0) {
3367 0 : SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair);
3368 0 : return -1;
3369 : }
3370 :
3371 0 : rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock,
3372 : nvmf_tcp_sock_cb, tqpair);
3373 0 : if (rc != 0) {
3374 0 : SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n",
3375 : spdk_strerror(errno), errno);
3376 0 : return -1;
3377 : }
3378 :
3379 0 : tqpair->group = tgroup;
3380 0 : nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_INVALID);
3381 0 : TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link);
3382 :
3383 0 : return 0;
3384 : }
3385 :
3386 : static int
3387 0 : nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group,
3388 : struct spdk_nvmf_qpair *qpair)
3389 : {
3390 : struct spdk_nvmf_tcp_poll_group *tgroup;
3391 : struct spdk_nvmf_tcp_qpair *tqpair;
3392 : int rc;
3393 :
3394 0 : tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group);
3395 0 : tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
3396 :
3397 0 : assert(tqpair->group == tgroup);
3398 :
3399 0 : SPDK_DEBUGLOG(nvmf_tcp, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup);
3400 0 : if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) {
3401 : /* Change the state to move the qpair from the await_req list to the main list
3402 : * and prevent adding it again later by nvmf_tcp_qpair_set_recv_state() */
3403 0 : nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING);
3404 : }
3405 0 : TAILQ_REMOVE(&tgroup->qpairs, tqpair, link);
3406 :
3407 : /* Try to force out any pending writes */
3408 0 : spdk_sock_flush(tqpair->sock);
3409 :
3410 0 : rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock);
3411 0 : if (rc != 0) {
3412 0 : SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n",
3413 : spdk_strerror(errno), errno);
3414 : }
3415 :
3416 0 : return rc;
3417 : }
3418 :
3419 : static int
3420 0 : nvmf_tcp_req_complete(struct spdk_nvmf_request *req)
3421 : {
3422 : struct spdk_nvmf_tcp_transport *ttransport;
3423 : struct spdk_nvmf_tcp_req *tcp_req;
3424 :
3425 0 : ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport);
3426 0 : tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req);
3427 :
3428 0 : switch (tcp_req->state) {
3429 0 : case TCP_REQUEST_STATE_EXECUTING:
3430 : case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT:
3431 0 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED);
3432 0 : break;
3433 0 : case TCP_REQUEST_STATE_AWAITING_ZCOPY_START:
3434 0 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_ZCOPY_START_COMPLETED);
3435 0 : break;
3436 0 : case TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE:
3437 0 : nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED);
3438 0 : break;
3439 0 : default:
3440 0 : SPDK_ERRLOG("Unexpected request state %d (cntlid:%d, qid:%d)\n",
3441 : tcp_req->state, req->qpair->ctrlr->cntlid, req->qpair->qid);
3442 0 : assert(0 && "Unexpected request state");
3443 : break;
3444 : }
3445 :
3446 0 : nvmf_tcp_req_process(ttransport, tcp_req);
3447 :
3448 0 : return 0;
3449 : }
3450 :
3451 : static void
3452 0 : nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair,
3453 : spdk_nvmf_transport_qpair_fini_cb cb_fn, void *cb_arg)
3454 : {
3455 : struct spdk_nvmf_tcp_qpair *tqpair;
3456 :
3457 0 : SPDK_DEBUGLOG(nvmf_tcp, "Qpair: %p\n", qpair);
3458 :
3459 0 : tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
3460 :
3461 0 : assert(tqpair->fini_cb_fn == NULL);
3462 0 : tqpair->fini_cb_fn = cb_fn;
3463 0 : tqpair->fini_cb_arg = cb_arg;
3464 :
3465 0 : nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_EXITED);
3466 0 : nvmf_tcp_qpair_destroy(tqpair);
3467 0 : }
3468 :
3469 : static int
3470 0 : nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group)
3471 : {
3472 : struct spdk_nvmf_tcp_poll_group *tgroup;
3473 0 : int num_events, rc = 0, rc2;
3474 : struct spdk_nvmf_tcp_qpair *tqpair, *tqpair_tmp;
3475 :
3476 0 : tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group);
3477 :
3478 0 : if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs) && TAILQ_EMPTY(&tgroup->await_req))) {
3479 0 : return 0;
3480 : }
3481 :
3482 0 : num_events = spdk_sock_group_poll(tgroup->sock_group);
3483 0 : if (spdk_unlikely(num_events < 0)) {
3484 0 : SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group);
3485 : }
3486 :
3487 0 : TAILQ_FOREACH_SAFE(tqpair, &tgroup->await_req, link, tqpair_tmp) {
3488 0 : rc2 = nvmf_tcp_sock_process(tqpair);
3489 :
3490 : /* If there was a new socket error, disconnect */
3491 0 : if (spdk_unlikely(rc2 < 0)) {
3492 0 : nvmf_tcp_qpair_disconnect(tqpair);
3493 0 : if (rc == 0) {
3494 0 : rc = rc2;
3495 : }
3496 : }
3497 : }
3498 :
3499 0 : return rc == 0 ? num_events : rc;
3500 : }
3501 :
3502 : static int
3503 0 : nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair,
3504 : struct spdk_nvme_transport_id *trid, bool peer)
3505 : {
3506 : struct spdk_nvmf_tcp_qpair *tqpair;
3507 : uint16_t port;
3508 :
3509 0 : tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
3510 0 : spdk_nvme_trid_populate_transport(trid, SPDK_NVME_TRANSPORT_TCP);
3511 :
3512 0 : if (peer) {
3513 0 : snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr);
3514 0 : port = tqpair->initiator_port;
3515 : } else {
3516 0 : snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr);
3517 0 : port = tqpair->target_port;
3518 : }
3519 :
3520 0 : if (spdk_sock_is_ipv4(tqpair->sock)) {
3521 0 : trid->adrfam = SPDK_NVMF_ADRFAM_IPV4;
3522 0 : } else if (spdk_sock_is_ipv6(tqpair->sock)) {
3523 0 : trid->adrfam = SPDK_NVMF_ADRFAM_IPV6;
3524 : } else {
3525 0 : return -1;
3526 : }
3527 :
3528 0 : snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port);
3529 0 : return 0;
3530 : }
3531 :
3532 : static int
3533 0 : nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair,
3534 : struct spdk_nvme_transport_id *trid)
3535 : {
3536 0 : return nvmf_tcp_qpair_get_trid(qpair, trid, 0);
3537 : }
3538 :
3539 : static int
3540 0 : nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair,
3541 : struct spdk_nvme_transport_id *trid)
3542 : {
3543 0 : return nvmf_tcp_qpair_get_trid(qpair, trid, 1);
3544 : }
3545 :
3546 : static int
3547 0 : nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair,
3548 : struct spdk_nvme_transport_id *trid)
3549 : {
3550 0 : return nvmf_tcp_qpair_get_trid(qpair, trid, 0);
3551 : }
3552 :
3553 : static void
3554 0 : nvmf_tcp_req_set_abort_status(struct spdk_nvmf_request *req,
3555 : struct spdk_nvmf_tcp_req *tcp_req_to_abort)
3556 : {
3557 0 : nvmf_tcp_req_set_cpl(tcp_req_to_abort, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_ABORTED_BY_REQUEST);
3558 0 : nvmf_tcp_req_set_state(tcp_req_to_abort, TCP_REQUEST_STATE_READY_TO_COMPLETE);
3559 :
3560 0 : req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command was successfully aborted. */
3561 0 : }
3562 :
3563 : static int
3564 0 : _nvmf_tcp_qpair_abort_request(void *ctx)
3565 : {
3566 0 : struct spdk_nvmf_request *req = ctx;
3567 0 : struct spdk_nvmf_tcp_req *tcp_req_to_abort = SPDK_CONTAINEROF(req->req_to_abort,
3568 : struct spdk_nvmf_tcp_req, req);
3569 0 : struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(req->req_to_abort->qpair,
3570 : struct spdk_nvmf_tcp_qpair, qpair);
3571 0 : struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport,
3572 : struct spdk_nvmf_tcp_transport, transport);
3573 : int rc;
3574 :
3575 0 : spdk_poller_unregister(&req->poller);
3576 :
3577 0 : switch (tcp_req_to_abort->state) {
3578 0 : case TCP_REQUEST_STATE_EXECUTING:
3579 : case TCP_REQUEST_STATE_AWAITING_ZCOPY_START:
3580 : case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT:
3581 0 : rc = nvmf_ctrlr_abort_request(req);
3582 0 : if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS) {
3583 0 : return SPDK_POLLER_BUSY;
3584 : }
3585 0 : break;
3586 :
3587 0 : case TCP_REQUEST_STATE_NEED_BUFFER:
3588 0 : nvmf_tcp_request_get_buffers_abort(tcp_req_to_abort);
3589 0 : nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort);
3590 0 : nvmf_tcp_req_process(ttransport, tcp_req_to_abort);
3591 0 : break;
3592 :
3593 0 : case TCP_REQUEST_STATE_AWAITING_R2T_ACK:
3594 : case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER:
3595 0 : if (spdk_get_ticks() < req->timeout_tsc) {
3596 0 : req->poller = SPDK_POLLER_REGISTER(_nvmf_tcp_qpair_abort_request, req, 0);
3597 0 : return SPDK_POLLER_BUSY;
3598 : }
3599 0 : break;
3600 :
3601 0 : default:
3602 : /* Requests in other states are either un-abortable (e.g.
3603 : * TRANSFERRING_CONTROLLER_TO_HOST) or should never end up here, as they're
3604 : * immediately transitioned to other states in nvmf_tcp_req_process() (e.g.
3605 : * READY_TO_EXECUTE). But it is fine to end up here, as we'll simply complete the
3606 : * abort request with the bit0 of dword0 set (command not aborted).
3607 : */
3608 0 : break;
3609 : }
3610 :
3611 0 : spdk_nvmf_request_complete(req);
3612 0 : return SPDK_POLLER_BUSY;
3613 : }
3614 :
3615 : static void
3616 0 : nvmf_tcp_qpair_abort_request(struct spdk_nvmf_qpair *qpair,
3617 : struct spdk_nvmf_request *req)
3618 : {
3619 : struct spdk_nvmf_tcp_qpair *tqpair;
3620 : struct spdk_nvmf_tcp_transport *ttransport;
3621 : struct spdk_nvmf_transport *transport;
3622 : uint16_t cid;
3623 : uint32_t i;
3624 0 : struct spdk_nvmf_tcp_req *tcp_req_to_abort = NULL;
3625 :
3626 0 : tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair);
3627 0 : ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport);
3628 0 : transport = &ttransport->transport;
3629 :
3630 0 : cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid;
3631 :
3632 0 : for (i = 0; i < tqpair->resource_count; i++) {
3633 0 : if (tqpair->reqs[i].state != TCP_REQUEST_STATE_FREE &&
3634 0 : tqpair->reqs[i].req.cmd->nvme_cmd.cid == cid) {
3635 0 : tcp_req_to_abort = &tqpair->reqs[i];
3636 0 : break;
3637 : }
3638 : }
3639 :
3640 0 : spdk_trace_record(TRACE_TCP_QP_ABORT_REQ, tqpair->qpair.trace_id, 0, (uintptr_t)req);
3641 :
3642 0 : if (tcp_req_to_abort == NULL) {
3643 0 : spdk_nvmf_request_complete(req);
3644 0 : return;
3645 : }
3646 :
3647 0 : req->req_to_abort = &tcp_req_to_abort->req;
3648 0 : req->timeout_tsc = spdk_get_ticks() +
3649 0 : transport->opts.abort_timeout_sec * spdk_get_ticks_hz();
3650 0 : req->poller = NULL;
3651 :
3652 0 : _nvmf_tcp_qpair_abort_request(req);
3653 : }
3654 :
3655 : struct tcp_subsystem_add_host_opts {
3656 : char *psk;
3657 : };
3658 :
3659 : static const struct spdk_json_object_decoder tcp_subsystem_add_host_opts_decoder[] = {
3660 : {"psk", offsetof(struct tcp_subsystem_add_host_opts, psk), spdk_json_decode_string, true},
3661 : };
3662 :
3663 : static int
3664 1 : tcp_load_psk(const char *fname, char *buf, size_t bufsz)
3665 : {
3666 : FILE *psk_file;
3667 1 : struct stat statbuf;
3668 : int rc;
3669 :
3670 1 : if (stat(fname, &statbuf) != 0) {
3671 0 : SPDK_ERRLOG("Could not read permissions for PSK file\n");
3672 0 : return -EACCES;
3673 : }
3674 :
3675 1 : if ((statbuf.st_mode & TCP_PSK_INVALID_PERMISSIONS) != 0) {
3676 0 : SPDK_ERRLOG("Incorrect permissions for PSK file\n");
3677 0 : return -EPERM;
3678 : }
3679 1 : if ((size_t)statbuf.st_size > bufsz) {
3680 0 : SPDK_ERRLOG("Invalid PSK: too long\n");
3681 0 : return -EINVAL;
3682 : }
3683 1 : psk_file = fopen(fname, "r");
3684 1 : if (psk_file == NULL) {
3685 0 : SPDK_ERRLOG("Could not open PSK file\n");
3686 0 : return -EINVAL;
3687 : }
3688 :
3689 1 : rc = fread(buf, 1, statbuf.st_size, psk_file);
3690 1 : if (rc != statbuf.st_size) {
3691 0 : SPDK_ERRLOG("Failed to read PSK\n");
3692 0 : fclose(psk_file);
3693 0 : return -EINVAL;
3694 : }
3695 :
3696 1 : fclose(psk_file);
3697 1 : return 0;
3698 : }
3699 :
3700 1 : SPDK_LOG_DEPRECATION_REGISTER(nvmf_tcp_psk_path, "PSK path", "v24.09", 0);
3701 :
3702 : static int
3703 1 : nvmf_tcp_subsystem_add_host(struct spdk_nvmf_transport *transport,
3704 : const struct spdk_nvmf_subsystem *subsystem,
3705 : const char *hostnqn,
3706 : const struct spdk_json_val *transport_specific)
3707 : {
3708 1 : struct tcp_subsystem_add_host_opts opts;
3709 : struct spdk_nvmf_tcp_transport *ttransport;
3710 1 : struct tcp_psk_entry *tmp, *entry = NULL;
3711 1 : uint8_t psk_configured[SPDK_TLS_PSK_MAX_LEN] = {};
3712 1 : char psk_interchange[SPDK_TLS_PSK_MAX_LEN + 1] = {};
3713 : uint8_t tls_cipher_suite;
3714 1 : int rc = 0;
3715 1 : uint8_t psk_retained_hash;
3716 1 : uint64_t psk_configured_size;
3717 :
3718 1 : if (transport_specific == NULL) {
3719 0 : return 0;
3720 : }
3721 :
3722 1 : assert(transport != NULL);
3723 1 : assert(subsystem != NULL);
3724 :
3725 1 : memset(&opts, 0, sizeof(opts));
3726 :
3727 : /* Decode PSK (either name of a key or file path) */
3728 1 : if (spdk_json_decode_object_relaxed(transport_specific, tcp_subsystem_add_host_opts_decoder,
3729 : SPDK_COUNTOF(tcp_subsystem_add_host_opts_decoder), &opts)) {
3730 0 : SPDK_ERRLOG("spdk_json_decode_object failed\n");
3731 0 : return -EINVAL;
3732 : }
3733 :
3734 1 : if (opts.psk == NULL) {
3735 0 : return 0;
3736 : }
3737 :
3738 1 : entry = calloc(1, sizeof(struct tcp_psk_entry));
3739 1 : if (entry == NULL) {
3740 0 : SPDK_ERRLOG("Unable to allocate memory for PSK entry!\n");
3741 0 : rc = -ENOMEM;
3742 0 : goto end;
3743 : }
3744 :
3745 1 : entry->key = spdk_keyring_get_key(opts.psk);
3746 1 : if (entry->key != NULL) {
3747 0 : rc = spdk_key_get_key(entry->key, psk_interchange, SPDK_TLS_PSK_MAX_LEN);
3748 0 : if (rc < 0) {
3749 0 : SPDK_ERRLOG("Failed to retrieve PSK '%s'\n", opts.psk);
3750 0 : rc = -EINVAL;
3751 0 : goto end;
3752 : }
3753 : } else {
3754 1 : if (strlen(opts.psk) >= sizeof(entry->psk)) {
3755 0 : SPDK_ERRLOG("PSK path too long\n");
3756 0 : rc = -EINVAL;
3757 0 : goto end;
3758 : }
3759 :
3760 1 : rc = tcp_load_psk(opts.psk, psk_interchange, SPDK_TLS_PSK_MAX_LEN);
3761 1 : if (rc) {
3762 0 : SPDK_ERRLOG("Could not retrieve PSK from file\n");
3763 0 : goto end;
3764 : }
3765 :
3766 1 : SPDK_LOG_DEPRECATED(nvmf_tcp_psk_path);
3767 : }
3768 :
3769 : /* Parse PSK interchange to get length of base64 encoded data.
3770 : * This is then used to decide which cipher suite should be used
3771 : * to generate PSK identity and TLS PSK later on. */
3772 1 : rc = nvme_tcp_parse_interchange_psk(psk_interchange, psk_configured, sizeof(psk_configured),
3773 : &psk_configured_size, &psk_retained_hash);
3774 1 : if (rc < 0) {
3775 0 : SPDK_ERRLOG("Failed to parse PSK interchange!\n");
3776 0 : goto end;
3777 : }
3778 :
3779 : /* The Base64 string encodes the configured PSK (32 or 48 bytes binary).
3780 : * This check also ensures that psk_configured_size is smaller than
3781 : * psk_retained buffer size. */
3782 1 : if (psk_configured_size == SHA256_DIGEST_LENGTH) {
3783 1 : tls_cipher_suite = NVME_TCP_CIPHER_AES_128_GCM_SHA256;
3784 0 : } else if (psk_configured_size == SHA384_DIGEST_LENGTH) {
3785 0 : tls_cipher_suite = NVME_TCP_CIPHER_AES_256_GCM_SHA384;
3786 : } else {
3787 0 : SPDK_ERRLOG("Unrecognized cipher suite!\n");
3788 0 : rc = -EINVAL;
3789 0 : goto end;
3790 : }
3791 :
3792 1 : ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport);
3793 : /* Generate PSK identity. */
3794 1 : rc = nvme_tcp_generate_psk_identity(entry->pskid, sizeof(entry->pskid), hostnqn,
3795 1 : subsystem->subnqn, tls_cipher_suite);
3796 1 : if (rc) {
3797 0 : rc = -EINVAL;
3798 0 : goto end;
3799 : }
3800 : /* Check if PSK identity entry already exists. */
3801 1 : TAILQ_FOREACH(tmp, &ttransport->psks, link) {
3802 0 : if (strncmp(tmp->pskid, entry->pskid, NVMF_PSK_IDENTITY_LEN) == 0) {
3803 0 : SPDK_ERRLOG("Given PSK identity: %s entry already exists!\n", entry->pskid);
3804 0 : rc = -EEXIST;
3805 0 : goto end;
3806 : }
3807 : }
3808 :
3809 1 : if (snprintf(entry->hostnqn, sizeof(entry->hostnqn), "%s", hostnqn) < 0) {
3810 0 : SPDK_ERRLOG("Could not write hostnqn string!\n");
3811 0 : rc = -EINVAL;
3812 0 : goto end;
3813 : }
3814 1 : if (snprintf(entry->subnqn, sizeof(entry->subnqn), "%s", subsystem->subnqn) < 0) {
3815 0 : SPDK_ERRLOG("Could not write subnqn string!\n");
3816 0 : rc = -EINVAL;
3817 0 : goto end;
3818 : }
3819 :
3820 1 : entry->tls_cipher_suite = tls_cipher_suite;
3821 :
3822 : /* No hash indicates that Configured PSK must be used as Retained PSK. */
3823 1 : if (psk_retained_hash == NVME_TCP_HASH_ALGORITHM_NONE) {
3824 : /* Psk configured is either 32 or 48 bytes long. */
3825 0 : memcpy(entry->psk, psk_configured, psk_configured_size);
3826 0 : entry->psk_size = psk_configured_size;
3827 : } else {
3828 : /* Derive retained PSK. */
3829 1 : rc = nvme_tcp_derive_retained_psk(psk_configured, psk_configured_size, hostnqn, entry->psk,
3830 : SPDK_TLS_PSK_MAX_LEN, psk_retained_hash);
3831 1 : if (rc < 0) {
3832 0 : SPDK_ERRLOG("Unable to derive retained PSK!\n");
3833 0 : goto end;
3834 : }
3835 1 : entry->psk_size = rc;
3836 : }
3837 :
3838 1 : if (entry->key == NULL) {
3839 1 : rc = snprintf(entry->psk_path, sizeof(entry->psk_path), "%s", opts.psk);
3840 1 : if (rc < 0 || (size_t)rc >= sizeof(entry->psk_path)) {
3841 0 : SPDK_ERRLOG("Could not save PSK path!\n");
3842 0 : rc = -ENAMETOOLONG;
3843 0 : goto end;
3844 : }
3845 : }
3846 :
3847 1 : TAILQ_INSERT_TAIL(&ttransport->psks, entry, link);
3848 1 : rc = 0;
3849 :
3850 1 : end:
3851 1 : spdk_memset_s(psk_configured, sizeof(psk_configured), 0, sizeof(psk_configured));
3852 1 : spdk_memset_s(psk_interchange, sizeof(psk_interchange), 0, sizeof(psk_interchange));
3853 :
3854 1 : free(opts.psk);
3855 1 : if (rc != 0) {
3856 0 : nvmf_tcp_free_psk_entry(entry);
3857 : }
3858 :
3859 1 : return rc;
3860 : }
3861 :
3862 : static void
3863 1 : nvmf_tcp_subsystem_remove_host(struct spdk_nvmf_transport *transport,
3864 : const struct spdk_nvmf_subsystem *subsystem,
3865 : const char *hostnqn)
3866 : {
3867 : struct spdk_nvmf_tcp_transport *ttransport;
3868 : struct tcp_psk_entry *entry, *tmp;
3869 :
3870 1 : assert(transport != NULL);
3871 1 : assert(subsystem != NULL);
3872 :
3873 1 : ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport);
3874 1 : TAILQ_FOREACH_SAFE(entry, &ttransport->psks, link, tmp) {
3875 1 : if ((strncmp(entry->hostnqn, hostnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0 &&
3876 1 : (strncmp(entry->subnqn, subsystem->subnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0) {
3877 1 : TAILQ_REMOVE(&ttransport->psks, entry, link);
3878 1 : nvmf_tcp_free_psk_entry(entry);
3879 1 : break;
3880 : }
3881 : }
3882 1 : }
3883 :
3884 : static void
3885 0 : nvmf_tcp_subsystem_dump_host(struct spdk_nvmf_transport *transport,
3886 : const struct spdk_nvmf_subsystem *subsystem, const char *hostnqn,
3887 : struct spdk_json_write_ctx *w)
3888 : {
3889 : struct spdk_nvmf_tcp_transport *ttransport;
3890 : struct tcp_psk_entry *entry;
3891 :
3892 0 : assert(transport != NULL);
3893 0 : assert(subsystem != NULL);
3894 :
3895 0 : ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport);
3896 0 : TAILQ_FOREACH(entry, &ttransport->psks, link) {
3897 0 : if ((strncmp(entry->hostnqn, hostnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0 &&
3898 0 : (strncmp(entry->subnqn, subsystem->subnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0) {
3899 0 : spdk_json_write_named_string(w, "psk", entry->key ?
3900 0 : spdk_key_get_name(entry->key) : entry->psk_path);
3901 0 : break;
3902 : }
3903 : }
3904 0 : }
3905 :
3906 : static void
3907 1 : nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts)
3908 : {
3909 1 : opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH;
3910 1 : opts->max_qpairs_per_ctrlr = SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR;
3911 1 : opts->in_capsule_data_size = SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE;
3912 1 : opts->max_io_size = SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE;
3913 1 : opts->io_unit_size = SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE;
3914 1 : opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH;
3915 1 : opts->num_shared_buffers = SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS;
3916 1 : opts->buf_cache_size = SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE;
3917 1 : opts->dif_insert_or_strip = SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP;
3918 1 : opts->abort_timeout_sec = SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC;
3919 1 : opts->transport_specific = NULL;
3920 1 : }
3921 :
3922 : const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = {
3923 : .name = "TCP",
3924 : .type = SPDK_NVME_TRANSPORT_TCP,
3925 : .opts_init = nvmf_tcp_opts_init,
3926 : .create = nvmf_tcp_create,
3927 : .dump_opts = nvmf_tcp_dump_opts,
3928 : .destroy = nvmf_tcp_destroy,
3929 :
3930 : .listen = nvmf_tcp_listen,
3931 : .stop_listen = nvmf_tcp_stop_listen,
3932 :
3933 : .listener_discover = nvmf_tcp_discover,
3934 :
3935 : .poll_group_create = nvmf_tcp_poll_group_create,
3936 : .get_optimal_poll_group = nvmf_tcp_get_optimal_poll_group,
3937 : .poll_group_destroy = nvmf_tcp_poll_group_destroy,
3938 : .poll_group_add = nvmf_tcp_poll_group_add,
3939 : .poll_group_remove = nvmf_tcp_poll_group_remove,
3940 : .poll_group_poll = nvmf_tcp_poll_group_poll,
3941 :
3942 : .req_free = nvmf_tcp_req_free,
3943 : .req_complete = nvmf_tcp_req_complete,
3944 : .req_get_buffers_done = nvmf_tcp_req_get_buffers_done,
3945 :
3946 : .qpair_fini = nvmf_tcp_close_qpair,
3947 : .qpair_get_local_trid = nvmf_tcp_qpair_get_local_trid,
3948 : .qpair_get_peer_trid = nvmf_tcp_qpair_get_peer_trid,
3949 : .qpair_get_listen_trid = nvmf_tcp_qpair_get_listen_trid,
3950 : .qpair_abort_request = nvmf_tcp_qpair_abort_request,
3951 : .subsystem_add_host = nvmf_tcp_subsystem_add_host,
3952 : .subsystem_remove_host = nvmf_tcp_subsystem_remove_host,
3953 : .subsystem_dump_host = nvmf_tcp_subsystem_dump_host,
3954 : };
3955 :
3956 1 : SPDK_NVMF_TRANSPORT_REGISTER(tcp, &spdk_nvmf_transport_tcp);
3957 1 : SPDK_LOG_REGISTER_COMPONENT(nvmf_tcp)
|