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