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