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