Line data Source code
1 : /* SPDX-License-Identifier: BSD-3-Clause
2 : * Copyright (C) 2018 Intel Corporation. All rights reserved.
3 : * Copyright (c) 2020, 2021 Mellanox Technologies LTD. All rights reserved.
4 : * Copyright (c) 2021 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
5 : */
6 :
7 : #include "spdk/stdinc.h"
8 :
9 : #if defined(__FreeBSD__)
10 : #include <sys/event.h>
11 : #define SPDK_KEVENT
12 : #else
13 : #include <sys/epoll.h>
14 : #define SPDK_EPOLL
15 : #endif
16 :
17 : #if defined(__linux__)
18 : #include <linux/errqueue.h>
19 : #endif
20 :
21 : #include "spdk/env.h"
22 : #include "spdk/log.h"
23 : #include "spdk/pipe.h"
24 : #include "spdk/sock.h"
25 : #include "spdk/util.h"
26 : #include "spdk/string.h"
27 : #include "spdk/net.h"
28 : #include "spdk/file.h"
29 : #include "spdk_internal/sock.h"
30 : #include "spdk/net.h"
31 :
32 : #include "openssl/crypto.h"
33 : #include "openssl/err.h"
34 : #include "openssl/ssl.h"
35 :
36 : #define MAX_TMPBUF 1024
37 : #define PORTNUMLEN 32
38 :
39 : #if defined(SO_ZEROCOPY) && defined(MSG_ZEROCOPY)
40 : #define SPDK_ZEROCOPY
41 : #endif
42 :
43 : struct spdk_posix_sock {
44 : struct spdk_sock base;
45 : int fd;
46 :
47 : uint32_t sendmsg_idx;
48 :
49 : struct spdk_pipe *recv_pipe;
50 : int recv_buf_sz;
51 : bool pipe_has_data;
52 : bool socket_has_data;
53 : bool zcopy;
54 :
55 : int placement_id;
56 :
57 : SSL_CTX *ctx;
58 : SSL *ssl;
59 :
60 : TAILQ_ENTRY(spdk_posix_sock) link;
61 :
62 : char interface_name[IFNAMSIZ];
63 : };
64 :
65 : TAILQ_HEAD(spdk_has_data_list, spdk_posix_sock);
66 :
67 : struct spdk_posix_sock_group_impl {
68 : struct spdk_sock_group_impl base;
69 : int fd;
70 : struct spdk_interrupt *intr;
71 : struct spdk_has_data_list socks_with_data;
72 : int placement_id;
73 : struct spdk_pipe_group *pipe_group;
74 : };
75 :
76 : static struct spdk_sock_impl_opts g_posix_impl_opts = {
77 : .recv_buf_size = DEFAULT_SO_RCVBUF_SIZE,
78 : .send_buf_size = DEFAULT_SO_SNDBUF_SIZE,
79 : .enable_recv_pipe = true,
80 : .enable_quickack = false,
81 : .enable_placement_id = PLACEMENT_NONE,
82 : .enable_zerocopy_send_server = true,
83 : .enable_zerocopy_send_client = false,
84 : .zerocopy_threshold = 0,
85 : .tls_version = 0,
86 : .enable_ktls = false,
87 : .psk_key = NULL,
88 : .psk_key_size = 0,
89 : .psk_identity = NULL,
90 : .get_key = NULL,
91 : .get_key_ctx = NULL,
92 : .tls_cipher_suites = NULL
93 : };
94 :
95 : static struct spdk_sock_impl_opts g_ssl_impl_opts = {
96 : .recv_buf_size = MIN_SO_RCVBUF_SIZE,
97 : .send_buf_size = MIN_SO_SNDBUF_SIZE,
98 : .enable_recv_pipe = true,
99 : .enable_quickack = false,
100 : .enable_placement_id = PLACEMENT_NONE,
101 : .enable_zerocopy_send_server = true,
102 : .enable_zerocopy_send_client = false,
103 : .zerocopy_threshold = 0,
104 : .tls_version = 0,
105 : .enable_ktls = false,
106 : .psk_key = NULL,
107 : .psk_identity = NULL
108 : };
109 :
110 : static struct spdk_sock_map g_map = {
111 : .entries = STAILQ_HEAD_INITIALIZER(g_map.entries),
112 : .mtx = PTHREAD_MUTEX_INITIALIZER
113 : };
114 :
115 : __attribute((destructor)) static void
116 2 : posix_sock_map_cleanup(void)
117 : {
118 2 : spdk_sock_map_cleanup(&g_map);
119 2 : }
120 :
121 : #define __posix_sock(sock) (struct spdk_posix_sock *)sock
122 : #define __posix_group_impl(group) (struct spdk_posix_sock_group_impl *)group
123 :
124 : static void
125 20 : posix_sock_copy_impl_opts(struct spdk_sock_impl_opts *dest, const struct spdk_sock_impl_opts *src,
126 : size_t len)
127 : {
128 : #define FIELD_OK(field) \
129 : offsetof(struct spdk_sock_impl_opts, field) + sizeof(src->field) <= len
130 :
131 : #define SET_FIELD(field) \
132 : if (FIELD_OK(field)) { \
133 : dest->field = src->field; \
134 : }
135 :
136 20 : SET_FIELD(recv_buf_size);
137 20 : SET_FIELD(send_buf_size);
138 20 : SET_FIELD(enable_recv_pipe);
139 20 : SET_FIELD(enable_zerocopy_send);
140 20 : SET_FIELD(enable_quickack);
141 20 : SET_FIELD(enable_placement_id);
142 20 : SET_FIELD(enable_zerocopy_send_server);
143 20 : SET_FIELD(enable_zerocopy_send_client);
144 20 : SET_FIELD(zerocopy_threshold);
145 20 : SET_FIELD(tls_version);
146 20 : SET_FIELD(enable_ktls);
147 20 : SET_FIELD(psk_key);
148 20 : SET_FIELD(psk_key_size);
149 20 : SET_FIELD(psk_identity);
150 20 : SET_FIELD(get_key);
151 20 : SET_FIELD(get_key_ctx);
152 20 : SET_FIELD(tls_cipher_suites);
153 :
154 : #undef SET_FIELD
155 : #undef FIELD_OK
156 20 : }
157 :
158 : static int
159 11 : _sock_impl_get_opts(struct spdk_sock_impl_opts *opts, struct spdk_sock_impl_opts *impl_opts,
160 : size_t *len)
161 : {
162 11 : if (!opts || !len) {
163 0 : errno = EINVAL;
164 0 : return -1;
165 : }
166 :
167 11 : assert(sizeof(*opts) >= *len);
168 11 : memset(opts, 0, *len);
169 :
170 11 : posix_sock_copy_impl_opts(opts, impl_opts, *len);
171 11 : *len = spdk_min(*len, sizeof(*impl_opts));
172 :
173 11 : return 0;
174 : }
175 :
176 : static int
177 11 : posix_sock_impl_get_opts(struct spdk_sock_impl_opts *opts, size_t *len)
178 : {
179 11 : return _sock_impl_get_opts(opts, &g_posix_impl_opts, len);
180 : }
181 :
182 : static int
183 0 : ssl_sock_impl_get_opts(struct spdk_sock_impl_opts *opts, size_t *len)
184 : {
185 0 : return _sock_impl_get_opts(opts, &g_ssl_impl_opts, len);
186 : }
187 :
188 : static int
189 3 : _sock_impl_set_opts(const struct spdk_sock_impl_opts *opts, struct spdk_sock_impl_opts *impl_opts,
190 : size_t len)
191 : {
192 3 : if (!opts) {
193 0 : errno = EINVAL;
194 0 : return -1;
195 : }
196 :
197 3 : assert(sizeof(*opts) >= len);
198 3 : posix_sock_copy_impl_opts(impl_opts, opts, len);
199 :
200 3 : return 0;
201 : }
202 :
203 : static int
204 3 : posix_sock_impl_set_opts(const struct spdk_sock_impl_opts *opts, size_t len)
205 : {
206 3 : return _sock_impl_set_opts(opts, &g_posix_impl_opts, len);
207 : }
208 :
209 : static int
210 0 : ssl_sock_impl_set_opts(const struct spdk_sock_impl_opts *opts, size_t len)
211 : {
212 0 : return _sock_impl_set_opts(opts, &g_ssl_impl_opts, len);
213 : }
214 :
215 : static void
216 16 : _opts_get_impl_opts(const struct spdk_sock_opts *opts, struct spdk_sock_impl_opts *dest,
217 : const struct spdk_sock_impl_opts *default_impl)
218 : {
219 : /* Copy the default impl_opts first to cover cases when user's impl_opts is smaller */
220 16 : memcpy(dest, default_impl, sizeof(*dest));
221 :
222 16 : if (opts->impl_opts != NULL) {
223 6 : assert(sizeof(*dest) >= opts->impl_opts_size);
224 6 : posix_sock_copy_impl_opts(dest, opts->impl_opts, opts->impl_opts_size);
225 : }
226 16 : }
227 :
228 : static int
229 0 : posix_sock_getaddr(struct spdk_sock *_sock, char *saddr, int slen, uint16_t *sport,
230 : char *caddr, int clen, uint16_t *cport)
231 : {
232 0 : struct spdk_posix_sock *sock = __posix_sock(_sock);
233 :
234 0 : assert(sock != NULL);
235 0 : return spdk_net_getaddr(sock->fd, saddr, slen, sport, caddr, clen, cport);
236 : }
237 :
238 : static const char *
239 6 : posix_sock_get_interface_name(struct spdk_sock *_sock)
240 : {
241 6 : struct spdk_posix_sock *sock = __posix_sock(_sock);
242 6 : char saddr[64];
243 : int rc;
244 :
245 6 : rc = spdk_net_getaddr(sock->fd, saddr, sizeof(saddr), NULL, NULL, 0, NULL);
246 6 : if (rc != 0) {
247 0 : return NULL;
248 : }
249 :
250 6 : rc = spdk_net_get_interface_name(saddr, sock->interface_name,
251 : sizeof(sock->interface_name));
252 6 : if (rc != 0) {
253 0 : return NULL;
254 : }
255 :
256 6 : return sock->interface_name;
257 : }
258 :
259 : static uint32_t
260 0 : posix_sock_get_numa_socket_id(struct spdk_sock *sock)
261 : {
262 : const char *interface_name;
263 0 : uint32_t numa_socket_id;
264 : int rc;
265 :
266 0 : interface_name = posix_sock_get_interface_name(sock);
267 0 : if (interface_name == NULL) {
268 0 : return SPDK_ENV_SOCKET_ID_ANY;
269 : }
270 :
271 0 : rc = spdk_read_sysfs_attribute_uint32(&numa_socket_id,
272 : "/sys/class/net/%s/device/numa_node", interface_name);
273 0 : if (rc == 0) {
274 0 : return numa_socket_id;
275 : } else {
276 0 : return SPDK_ENV_SOCKET_ID_ANY;
277 : }
278 : }
279 :
280 : enum posix_sock_create_type {
281 : SPDK_SOCK_CREATE_LISTEN,
282 : SPDK_SOCK_CREATE_CONNECT,
283 : };
284 :
285 : static int
286 1 : posix_sock_alloc_pipe(struct spdk_posix_sock *sock, int sz)
287 : {
288 1 : uint8_t *new_buf, *old_buf;
289 : struct spdk_pipe *new_pipe;
290 1 : struct iovec siov[2];
291 1 : struct iovec diov[2];
292 : int sbytes;
293 : ssize_t bytes;
294 : int rc;
295 :
296 1 : if (sock->recv_buf_sz == sz) {
297 0 : return 0;
298 : }
299 :
300 : /* If the new size is 0, just free the pipe */
301 1 : if (sz == 0) {
302 0 : old_buf = spdk_pipe_destroy(sock->recv_pipe);
303 0 : free(old_buf);
304 0 : sock->recv_pipe = NULL;
305 0 : return 0;
306 1 : } else if (sz < MIN_SOCK_PIPE_SIZE) {
307 0 : SPDK_ERRLOG("The size of the pipe must be larger than %d\n", MIN_SOCK_PIPE_SIZE);
308 0 : return -1;
309 : }
310 :
311 : /* Round up to next 64 byte multiple */
312 1 : rc = posix_memalign((void **)&new_buf, 64, sz);
313 1 : if (rc != 0) {
314 0 : SPDK_ERRLOG("socket recv buf allocation failed\n");
315 0 : return -ENOMEM;
316 : }
317 1 : memset(new_buf, 0, sz);
318 :
319 1 : new_pipe = spdk_pipe_create(new_buf, sz);
320 1 : if (new_pipe == NULL) {
321 0 : SPDK_ERRLOG("socket pipe allocation failed\n");
322 0 : free(new_buf);
323 0 : return -ENOMEM;
324 : }
325 :
326 1 : if (sock->recv_pipe != NULL) {
327 : /* Pull all of the data out of the old pipe */
328 0 : sbytes = spdk_pipe_reader_get_buffer(sock->recv_pipe, sock->recv_buf_sz, siov);
329 0 : if (sbytes > sz) {
330 : /* Too much data to fit into the new pipe size */
331 0 : old_buf = spdk_pipe_destroy(new_pipe);
332 0 : free(old_buf);
333 0 : return -EINVAL;
334 : }
335 :
336 0 : sbytes = spdk_pipe_writer_get_buffer(new_pipe, sz, diov);
337 0 : assert(sbytes == sz);
338 :
339 0 : bytes = spdk_iovcpy(siov, 2, diov, 2);
340 0 : spdk_pipe_writer_advance(new_pipe, bytes);
341 :
342 0 : old_buf = spdk_pipe_destroy(sock->recv_pipe);
343 0 : free(old_buf);
344 : }
345 :
346 1 : sock->recv_buf_sz = sz;
347 1 : sock->recv_pipe = new_pipe;
348 :
349 1 : if (sock->base.group_impl) {
350 : struct spdk_posix_sock_group_impl *group;
351 :
352 0 : group = __posix_group_impl(sock->base.group_impl);
353 0 : spdk_pipe_group_add(group->pipe_group, sock->recv_pipe);
354 : }
355 :
356 1 : return 0;
357 : }
358 :
359 : static int
360 1 : posix_sock_set_recvbuf(struct spdk_sock *_sock, int sz)
361 : {
362 1 : struct spdk_posix_sock *sock = __posix_sock(_sock);
363 : int min_size;
364 : int rc;
365 :
366 1 : assert(sock != NULL);
367 :
368 1 : if (_sock->impl_opts.enable_recv_pipe) {
369 1 : rc = posix_sock_alloc_pipe(sock, sz);
370 1 : if (rc) {
371 0 : return rc;
372 : }
373 : }
374 :
375 : /* Set kernel buffer size to be at least MIN_SO_RCVBUF_SIZE and
376 : * _sock->impl_opts.recv_buf_size. */
377 1 : min_size = spdk_max(MIN_SO_RCVBUF_SIZE, _sock->impl_opts.recv_buf_size);
378 :
379 1 : if (sz < min_size) {
380 1 : sz = min_size;
381 : }
382 :
383 1 : rc = setsockopt(sock->fd, SOL_SOCKET, SO_RCVBUF, &sz, sizeof(sz));
384 1 : if (rc < 0) {
385 0 : return rc;
386 : }
387 :
388 1 : _sock->impl_opts.recv_buf_size = sz;
389 :
390 1 : return 0;
391 : }
392 :
393 : static int
394 1 : posix_sock_set_sendbuf(struct spdk_sock *_sock, int sz)
395 : {
396 1 : struct spdk_posix_sock *sock = __posix_sock(_sock);
397 : int min_size;
398 : int rc;
399 :
400 1 : assert(sock != NULL);
401 :
402 : /* Set kernel buffer size to be at least MIN_SO_SNDBUF_SIZE and
403 : * _sock->impl_opts.send_buf_size. */
404 1 : min_size = spdk_max(MIN_SO_SNDBUF_SIZE, _sock->impl_opts.send_buf_size);
405 :
406 1 : if (sz < min_size) {
407 1 : sz = min_size;
408 : }
409 :
410 1 : rc = setsockopt(sock->fd, SOL_SOCKET, SO_SNDBUF, &sz, sizeof(sz));
411 1 : if (rc < 0) {
412 0 : return rc;
413 : }
414 :
415 1 : _sock->impl_opts.send_buf_size = sz;
416 :
417 1 : return 0;
418 : }
419 :
420 : static void
421 24 : posix_sock_init(struct spdk_posix_sock *sock, bool enable_zero_copy)
422 : {
423 : #if defined(SPDK_ZEROCOPY) || defined(__linux__)
424 24 : int flag;
425 : int rc;
426 : #endif
427 :
428 : #if defined(SPDK_ZEROCOPY)
429 24 : flag = 1;
430 :
431 24 : if (enable_zero_copy) {
432 : /* Try to turn on zero copy sends */
433 0 : rc = setsockopt(sock->fd, SOL_SOCKET, SO_ZEROCOPY, &flag, sizeof(flag));
434 0 : if (rc == 0) {
435 0 : sock->zcopy = true;
436 : }
437 : }
438 : #endif
439 :
440 : #if defined(__linux__)
441 24 : flag = 1;
442 :
443 24 : if (sock->base.impl_opts.enable_quickack) {
444 0 : rc = setsockopt(sock->fd, IPPROTO_TCP, TCP_QUICKACK, &flag, sizeof(flag));
445 0 : if (rc != 0) {
446 0 : SPDK_ERRLOG("quickack was failed to set\n");
447 : }
448 : }
449 :
450 24 : spdk_sock_get_placement_id(sock->fd, sock->base.impl_opts.enable_placement_id,
451 : &sock->placement_id);
452 :
453 24 : if (sock->base.impl_opts.enable_placement_id == PLACEMENT_MARK) {
454 : /* Save placement_id */
455 0 : spdk_sock_map_insert(&g_map, sock->placement_id, NULL);
456 : }
457 : #endif
458 24 : }
459 :
460 : static struct spdk_posix_sock *
461 24 : posix_sock_alloc(int fd, struct spdk_sock_impl_opts *impl_opts, bool enable_zero_copy)
462 : {
463 : struct spdk_posix_sock *sock;
464 :
465 24 : sock = calloc(1, sizeof(*sock));
466 24 : if (sock == NULL) {
467 0 : SPDK_ERRLOG("sock allocation failed\n");
468 0 : return NULL;
469 : }
470 :
471 24 : sock->fd = fd;
472 24 : memcpy(&sock->base.impl_opts, impl_opts, sizeof(*impl_opts));
473 24 : posix_sock_init(sock, enable_zero_copy);
474 :
475 24 : return sock;
476 : }
477 :
478 : static int
479 16 : posix_fd_create(struct addrinfo *res, struct spdk_sock_opts *opts,
480 : struct spdk_sock_impl_opts *impl_opts)
481 : {
482 : int fd;
483 16 : int val = 1;
484 16 : int rc, sz;
485 : #if defined(__linux__)
486 16 : int to;
487 : #endif
488 :
489 16 : fd = socket(res->ai_family, res->ai_socktype, res->ai_protocol);
490 16 : if (fd < 0) {
491 : /* error */
492 0 : return -1;
493 : }
494 :
495 16 : sz = impl_opts->recv_buf_size;
496 16 : rc = setsockopt(fd, SOL_SOCKET, SO_RCVBUF, &sz, sizeof(sz));
497 : if (rc) {
498 : /* Not fatal */
499 : }
500 :
501 16 : sz = impl_opts->send_buf_size;
502 16 : rc = setsockopt(fd, SOL_SOCKET, SO_SNDBUF, &sz, sizeof(sz));
503 : if (rc) {
504 : /* Not fatal */
505 : }
506 :
507 16 : rc = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &val, sizeof val);
508 16 : if (rc != 0) {
509 0 : close(fd);
510 : /* error */
511 0 : return -1;
512 : }
513 16 : rc = setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, &val, sizeof val);
514 16 : if (rc != 0) {
515 0 : close(fd);
516 : /* error */
517 0 : return -1;
518 : }
519 :
520 : #if defined(SO_PRIORITY)
521 16 : if (opts->priority) {
522 0 : rc = setsockopt(fd, SOL_SOCKET, SO_PRIORITY, &opts->priority, sizeof val);
523 0 : if (rc != 0) {
524 0 : close(fd);
525 : /* error */
526 0 : return -1;
527 : }
528 : }
529 : #endif
530 :
531 16 : if (res->ai_family == AF_INET6) {
532 0 : rc = setsockopt(fd, IPPROTO_IPV6, IPV6_V6ONLY, &val, sizeof val);
533 0 : if (rc != 0) {
534 0 : close(fd);
535 : /* error */
536 0 : return -1;
537 : }
538 : }
539 :
540 16 : if (opts->ack_timeout) {
541 : #if defined(__linux__)
542 0 : to = opts->ack_timeout;
543 0 : rc = setsockopt(fd, IPPROTO_TCP, TCP_USER_TIMEOUT, &to, sizeof(to));
544 0 : if (rc != 0) {
545 0 : close(fd);
546 : /* error */
547 0 : return -1;
548 : }
549 : #else
550 : SPDK_WARNLOG("TCP_USER_TIMEOUT is not supported.\n");
551 : #endif
552 : }
553 :
554 16 : return fd;
555 : }
556 :
557 : static int
558 0 : posix_sock_psk_find_session_server_cb(SSL *ssl, const unsigned char *identity,
559 : size_t identity_len, SSL_SESSION **sess)
560 : {
561 0 : struct spdk_sock_impl_opts *impl_opts = SSL_get_app_data(ssl);
562 0 : uint8_t key[SSL_MAX_MASTER_KEY_LENGTH] = {};
563 : int keylen;
564 : int rc, i;
565 : STACK_OF(SSL_CIPHER) *ciphers;
566 : const SSL_CIPHER *cipher;
567 : const char *cipher_name;
568 0 : const char *user_cipher = NULL;
569 0 : bool found = false;
570 :
571 0 : if (impl_opts->get_key) {
572 0 : rc = impl_opts->get_key(key, sizeof(key), &user_cipher, identity, impl_opts->get_key_ctx);
573 0 : if (rc < 0) {
574 0 : SPDK_ERRLOG("Unable to find PSK for identity: %s\n", identity);
575 0 : return 0;
576 : }
577 0 : keylen = rc;
578 : } else {
579 0 : if (impl_opts->psk_key == NULL) {
580 0 : SPDK_ERRLOG("PSK is not set\n");
581 0 : return 0;
582 : }
583 :
584 0 : SPDK_DEBUGLOG(sock_posix, "Length of Client's PSK ID %lu\n", strlen(impl_opts->psk_identity));
585 0 : if (strcmp(impl_opts->psk_identity, identity) != 0) {
586 0 : SPDK_ERRLOG("Unknown Client's PSK ID\n");
587 0 : return 0;
588 : }
589 0 : keylen = impl_opts->psk_key_size;
590 :
591 0 : memcpy(key, impl_opts->psk_key, keylen);
592 0 : user_cipher = impl_opts->tls_cipher_suites;
593 : }
594 :
595 0 : if (user_cipher == NULL) {
596 0 : SPDK_ERRLOG("Cipher suite not set\n");
597 0 : return 0;
598 : }
599 :
600 0 : *sess = SSL_SESSION_new();
601 0 : if (*sess == NULL) {
602 0 : SPDK_ERRLOG("Unable to allocate new SSL session\n");
603 0 : return 0;
604 : }
605 :
606 0 : ciphers = SSL_get_ciphers(ssl);
607 0 : for (i = 0; i < sk_SSL_CIPHER_num(ciphers); i++) {
608 0 : cipher = sk_SSL_CIPHER_value(ciphers, i);
609 0 : cipher_name = SSL_CIPHER_get_name(cipher);
610 :
611 0 : if (strcmp(user_cipher, cipher_name) == 0) {
612 0 : rc = SSL_SESSION_set_cipher(*sess, cipher);
613 0 : if (rc != 1) {
614 0 : SPDK_ERRLOG("Unable to set cipher: %s\n", cipher_name);
615 0 : goto err;
616 : }
617 0 : found = true;
618 0 : break;
619 : }
620 : }
621 0 : if (found == false) {
622 0 : SPDK_ERRLOG("No suitable cipher found\n");
623 0 : goto err;
624 : }
625 :
626 0 : SPDK_DEBUGLOG(sock_posix, "Cipher selected: %s\n", cipher_name);
627 :
628 0 : rc = SSL_SESSION_set_protocol_version(*sess, TLS1_3_VERSION);
629 0 : if (rc != 1) {
630 0 : SPDK_ERRLOG("Unable to set TLS version: %d\n", TLS1_3_VERSION);
631 0 : goto err;
632 : }
633 :
634 0 : rc = SSL_SESSION_set1_master_key(*sess, key, keylen);
635 0 : if (rc != 1) {
636 0 : SPDK_ERRLOG("Unable to set PSK for session\n");
637 0 : goto err;
638 : }
639 :
640 0 : return 1;
641 :
642 0 : err:
643 0 : SSL_SESSION_free(*sess);
644 0 : *sess = NULL;
645 0 : return 0;
646 : }
647 :
648 : static int
649 0 : posix_sock_psk_use_session_client_cb(SSL *ssl, const EVP_MD *md, const unsigned char **identity,
650 : size_t *identity_len, SSL_SESSION **sess)
651 : {
652 0 : struct spdk_sock_impl_opts *impl_opts = SSL_get_app_data(ssl);
653 : int rc, i;
654 : STACK_OF(SSL_CIPHER) *ciphers;
655 : const SSL_CIPHER *cipher;
656 : const char *cipher_name;
657 : long keylen;
658 0 : bool found = false;
659 :
660 0 : if (impl_opts->psk_key == NULL) {
661 0 : SPDK_ERRLOG("PSK is not set\n");
662 0 : return 0;
663 : }
664 0 : if (impl_opts->psk_key_size > SSL_MAX_MASTER_KEY_LENGTH) {
665 0 : SPDK_ERRLOG("PSK too long\n");
666 0 : return 0;
667 : }
668 0 : keylen = impl_opts->psk_key_size;
669 :
670 0 : if (impl_opts->tls_cipher_suites == NULL) {
671 0 : SPDK_ERRLOG("Cipher suite not set\n");
672 0 : return 0;
673 : }
674 0 : *sess = SSL_SESSION_new();
675 0 : if (*sess == NULL) {
676 0 : SPDK_ERRLOG("Unable to allocate new SSL session\n");
677 0 : return 0;
678 : }
679 :
680 0 : ciphers = SSL_get_ciphers(ssl);
681 0 : for (i = 0; i < sk_SSL_CIPHER_num(ciphers); i++) {
682 0 : cipher = sk_SSL_CIPHER_value(ciphers, i);
683 0 : cipher_name = SSL_CIPHER_get_name(cipher);
684 :
685 0 : if (strcmp(impl_opts->tls_cipher_suites, cipher_name) == 0) {
686 0 : rc = SSL_SESSION_set_cipher(*sess, cipher);
687 0 : if (rc != 1) {
688 0 : SPDK_ERRLOG("Unable to set cipher: %s\n", cipher_name);
689 0 : goto err;
690 : }
691 0 : found = true;
692 0 : break;
693 : }
694 : }
695 0 : if (found == false) {
696 0 : SPDK_ERRLOG("No suitable cipher found\n");
697 0 : goto err;
698 : }
699 :
700 0 : SPDK_DEBUGLOG(sock_posix, "Cipher selected: %s\n", cipher_name);
701 :
702 0 : rc = SSL_SESSION_set_protocol_version(*sess, TLS1_3_VERSION);
703 0 : if (rc != 1) {
704 0 : SPDK_ERRLOG("Unable to set TLS version: %d\n", TLS1_3_VERSION);
705 0 : goto err;
706 : }
707 :
708 0 : rc = SSL_SESSION_set1_master_key(*sess, impl_opts->psk_key, keylen);
709 0 : if (rc != 1) {
710 0 : SPDK_ERRLOG("Unable to set PSK for session\n");
711 0 : goto err;
712 : }
713 :
714 0 : *identity_len = strlen(impl_opts->psk_identity);
715 0 : *identity = impl_opts->psk_identity;
716 :
717 0 : return 1;
718 :
719 0 : err:
720 0 : SSL_SESSION_free(*sess);
721 0 : *sess = NULL;
722 0 : return 0;
723 : }
724 :
725 : static SSL_CTX *
726 0 : posix_sock_create_ssl_context(const SSL_METHOD *method, struct spdk_sock_opts *opts,
727 : struct spdk_sock_impl_opts *impl_opts)
728 : {
729 : SSL_CTX *ctx;
730 0 : int tls_version = 0;
731 0 : bool ktls_enabled = false;
732 : #ifdef SSL_OP_ENABLE_KTLS
733 : long options;
734 : #endif
735 :
736 0 : SSL_library_init();
737 0 : OpenSSL_add_all_algorithms();
738 0 : SSL_load_error_strings();
739 : /* Produce a SSL CTX in SSL V2 and V3 standards compliant way */
740 0 : ctx = SSL_CTX_new(method);
741 0 : if (!ctx) {
742 0 : SPDK_ERRLOG("SSL_CTX_new() failed, msg = %s\n", ERR_error_string(ERR_peek_last_error(), NULL));
743 0 : return NULL;
744 : }
745 0 : SPDK_DEBUGLOG(sock_posix, "SSL context created\n");
746 :
747 0 : switch (impl_opts->tls_version) {
748 0 : case 0:
749 : /* auto-negotiation */
750 0 : break;
751 0 : case SPDK_TLS_VERSION_1_3:
752 0 : tls_version = TLS1_3_VERSION;
753 0 : break;
754 0 : default:
755 0 : SPDK_ERRLOG("Incorrect TLS version provided: %d\n", impl_opts->tls_version);
756 0 : goto err;
757 : }
758 :
759 0 : if (tls_version) {
760 0 : SPDK_DEBUGLOG(sock_posix, "Hardening TLS version to '%d'='0x%X'\n", impl_opts->tls_version,
761 : tls_version);
762 0 : if (!SSL_CTX_set_min_proto_version(ctx, tls_version)) {
763 0 : SPDK_ERRLOG("Unable to set Min TLS version to '%d'='0x%X\n", impl_opts->tls_version, tls_version);
764 0 : goto err;
765 : }
766 0 : if (!SSL_CTX_set_max_proto_version(ctx, tls_version)) {
767 0 : SPDK_ERRLOG("Unable to set Max TLS version to '%d'='0x%X\n", impl_opts->tls_version, tls_version);
768 0 : goto err;
769 : }
770 : }
771 0 : if (impl_opts->enable_ktls) {
772 0 : SPDK_DEBUGLOG(sock_posix, "Enabling kTLS offload\n");
773 : #ifdef SSL_OP_ENABLE_KTLS
774 0 : options = SSL_CTX_set_options(ctx, SSL_OP_ENABLE_KTLS);
775 0 : ktls_enabled = options & SSL_OP_ENABLE_KTLS;
776 : #else
777 : ktls_enabled = false;
778 : #endif
779 0 : if (!ktls_enabled) {
780 0 : SPDK_ERRLOG("Unable to set kTLS offload via SSL_CTX_set_options(). Configure openssl with 'enable-ktls'\n");
781 0 : goto err;
782 : }
783 : }
784 :
785 : /* SSL_CTX_set_ciphersuites() return 1 if the requested
786 : * cipher suite list was configured, and 0 otherwise. */
787 0 : if (impl_opts->tls_cipher_suites != NULL &&
788 0 : SSL_CTX_set_ciphersuites(ctx, impl_opts->tls_cipher_suites) != 1) {
789 0 : SPDK_ERRLOG("Unable to set TLS cipher suites for SSL'\n");
790 0 : goto err;
791 : }
792 :
793 0 : return ctx;
794 :
795 0 : err:
796 0 : SSL_CTX_free(ctx);
797 0 : return NULL;
798 : }
799 :
800 : static SSL *
801 0 : ssl_sock_setup_connect(SSL_CTX *ctx, int fd)
802 : {
803 : SSL *ssl;
804 :
805 0 : ssl = SSL_new(ctx);
806 0 : if (!ssl) {
807 0 : SPDK_ERRLOG("SSL_new() failed, msg = %s\n", ERR_error_string(ERR_peek_last_error(), NULL));
808 0 : return NULL;
809 : }
810 0 : SSL_set_fd(ssl, fd);
811 0 : SSL_set_connect_state(ssl);
812 0 : SSL_set_psk_use_session_callback(ssl, posix_sock_psk_use_session_client_cb);
813 0 : SPDK_DEBUGLOG(sock_posix, "SSL object creation finished: %p\n", ssl);
814 0 : SPDK_DEBUGLOG(sock_posix, "%s = SSL_state_string_long(%p)\n", SSL_state_string_long(ssl), ssl);
815 0 : SPDK_DEBUGLOG(sock_posix, "%s = SSL_state_string_long(%p)\n", SSL_state_string_long(ssl), ssl);
816 0 : SPDK_DEBUGLOG(sock_posix, "Negotiated Cipher suite:%s\n",
817 : SSL_CIPHER_get_name(SSL_get_current_cipher(ssl)));
818 0 : return ssl;
819 : }
820 :
821 : static SSL *
822 0 : ssl_sock_setup_accept(SSL_CTX *ctx, int fd)
823 : {
824 : SSL *ssl;
825 :
826 0 : ssl = SSL_new(ctx);
827 0 : if (!ssl) {
828 0 : SPDK_ERRLOG("SSL_new() failed, msg = %s\n", ERR_error_string(ERR_peek_last_error(), NULL));
829 0 : return NULL;
830 : }
831 0 : SSL_set_fd(ssl, fd);
832 0 : SSL_set_accept_state(ssl);
833 0 : SSL_set_psk_find_session_callback(ssl, posix_sock_psk_find_session_server_cb);
834 0 : SPDK_DEBUGLOG(sock_posix, "SSL object creation finished: %p\n", ssl);
835 0 : SPDK_DEBUGLOG(sock_posix, "%s = SSL_state_string_long(%p)\n", SSL_state_string_long(ssl), ssl);
836 0 : SPDK_DEBUGLOG(sock_posix, "%s = SSL_state_string_long(%p)\n", SSL_state_string_long(ssl), ssl);
837 0 : SPDK_DEBUGLOG(sock_posix, "Negotiated Cipher suite:%s\n",
838 : SSL_CIPHER_get_name(SSL_get_current_cipher(ssl)));
839 0 : return ssl;
840 : }
841 :
842 : static ssize_t
843 0 : SSL_readv(SSL *ssl, const struct iovec *iov, int iovcnt)
844 : {
845 0 : int i, rc = 0;
846 0 : ssize_t total = 0;
847 :
848 0 : for (i = 0; i < iovcnt; i++) {
849 0 : rc = SSL_read(ssl, iov[i].iov_base, iov[i].iov_len);
850 :
851 0 : if (rc > 0) {
852 0 : total += rc;
853 : }
854 0 : if (rc != (int)iov[i].iov_len) {
855 0 : break;
856 : }
857 : }
858 0 : if (total > 0) {
859 0 : errno = 0;
860 0 : return total;
861 : }
862 0 : switch (SSL_get_error(ssl, rc)) {
863 0 : case SSL_ERROR_ZERO_RETURN:
864 0 : errno = ENOTCONN;
865 0 : return 0;
866 0 : case SSL_ERROR_WANT_READ:
867 : case SSL_ERROR_WANT_WRITE:
868 : case SSL_ERROR_WANT_CONNECT:
869 : case SSL_ERROR_WANT_ACCEPT:
870 : case SSL_ERROR_WANT_X509_LOOKUP:
871 : case SSL_ERROR_WANT_ASYNC:
872 : case SSL_ERROR_WANT_ASYNC_JOB:
873 : case SSL_ERROR_WANT_CLIENT_HELLO_CB:
874 0 : errno = EAGAIN;
875 0 : return -1;
876 0 : case SSL_ERROR_SYSCALL:
877 : case SSL_ERROR_SSL:
878 0 : errno = ENOTCONN;
879 0 : return -1;
880 0 : default:
881 0 : errno = ENOTCONN;
882 0 : return -1;
883 : }
884 : }
885 :
886 : static ssize_t
887 0 : SSL_writev(SSL *ssl, struct iovec *iov, int iovcnt)
888 : {
889 0 : int i, rc = 0;
890 0 : ssize_t total = 0;
891 :
892 0 : for (i = 0; i < iovcnt; i++) {
893 0 : rc = SSL_write(ssl, iov[i].iov_base, iov[i].iov_len);
894 :
895 0 : if (rc > 0) {
896 0 : total += rc;
897 : }
898 0 : if (rc != (int)iov[i].iov_len) {
899 0 : break;
900 : }
901 : }
902 0 : if (total > 0) {
903 0 : errno = 0;
904 0 : return total;
905 : }
906 0 : switch (SSL_get_error(ssl, rc)) {
907 0 : case SSL_ERROR_ZERO_RETURN:
908 0 : errno = ENOTCONN;
909 0 : return 0;
910 0 : case SSL_ERROR_WANT_READ:
911 : case SSL_ERROR_WANT_WRITE:
912 : case SSL_ERROR_WANT_CONNECT:
913 : case SSL_ERROR_WANT_ACCEPT:
914 : case SSL_ERROR_WANT_X509_LOOKUP:
915 : case SSL_ERROR_WANT_ASYNC:
916 : case SSL_ERROR_WANT_ASYNC_JOB:
917 : case SSL_ERROR_WANT_CLIENT_HELLO_CB:
918 0 : errno = EAGAIN;
919 0 : return -1;
920 0 : case SSL_ERROR_SYSCALL:
921 : case SSL_ERROR_SSL:
922 0 : errno = ENOTCONN;
923 0 : return -1;
924 0 : default:
925 0 : errno = ENOTCONN;
926 0 : return -1;
927 : }
928 : }
929 :
930 : static struct spdk_sock *
931 16 : posix_sock_create(const char *ip, int port,
932 : enum posix_sock_create_type type,
933 : struct spdk_sock_opts *opts,
934 : bool enable_ssl)
935 : {
936 : struct spdk_posix_sock *sock;
937 16 : struct spdk_sock_impl_opts impl_opts;
938 16 : char buf[MAX_TMPBUF];
939 16 : char portnum[PORTNUMLEN];
940 : char *p;
941 : const char *src_addr;
942 : uint16_t src_port;
943 16 : struct addrinfo hints, *res, *res0, *src_ai;
944 : int fd, flag;
945 : int rc;
946 16 : bool enable_zcopy_user_opts = true;
947 16 : bool enable_zcopy_impl_opts = true;
948 16 : SSL_CTX *ctx = 0;
949 16 : SSL *ssl = 0;
950 :
951 16 : assert(opts != NULL);
952 16 : if (enable_ssl) {
953 0 : _opts_get_impl_opts(opts, &impl_opts, &g_ssl_impl_opts);
954 : } else {
955 16 : _opts_get_impl_opts(opts, &impl_opts, &g_posix_impl_opts);
956 : }
957 :
958 16 : if (ip == NULL) {
959 0 : return NULL;
960 : }
961 16 : if (ip[0] == '[') {
962 0 : snprintf(buf, sizeof(buf), "%s", ip + 1);
963 0 : p = strchr(buf, ']');
964 0 : if (p != NULL) {
965 0 : *p = '\0';
966 : }
967 0 : ip = (const char *) &buf[0];
968 : }
969 :
970 16 : snprintf(portnum, sizeof portnum, "%d", port);
971 16 : memset(&hints, 0, sizeof hints);
972 16 : hints.ai_family = PF_UNSPEC;
973 16 : hints.ai_socktype = SOCK_STREAM;
974 16 : hints.ai_flags = AI_NUMERICSERV;
975 16 : hints.ai_flags |= AI_PASSIVE;
976 16 : hints.ai_flags |= AI_NUMERICHOST;
977 16 : rc = getaddrinfo(ip, portnum, &hints, &res0);
978 16 : if (rc != 0) {
979 0 : SPDK_ERRLOG("getaddrinfo() failed %s (%d)\n", gai_strerror(rc), rc);
980 0 : return NULL;
981 : }
982 :
983 : /* try listen */
984 16 : fd = -1;
985 16 : for (res = res0; res != NULL; res = res->ai_next) {
986 16 : retry:
987 16 : fd = posix_fd_create(res, opts, &impl_opts);
988 16 : if (fd < 0) {
989 0 : continue;
990 : }
991 16 : if (type == SPDK_SOCK_CREATE_LISTEN) {
992 7 : rc = bind(fd, res->ai_addr, res->ai_addrlen);
993 7 : if (rc != 0) {
994 0 : SPDK_ERRLOG("bind() failed at port %d, errno = %d\n", port, errno);
995 0 : switch (errno) {
996 0 : case EINTR:
997 : /* interrupted? */
998 0 : close(fd);
999 0 : goto retry;
1000 0 : case EADDRNOTAVAIL:
1001 0 : SPDK_ERRLOG("IP address %s not available. "
1002 : "Verify IP address in config file "
1003 : "and make sure setup script is "
1004 : "run before starting spdk app.\n", ip);
1005 : /* FALLTHROUGH */
1006 0 : default:
1007 : /* try next family */
1008 0 : close(fd);
1009 0 : fd = -1;
1010 0 : continue;
1011 : }
1012 : }
1013 : /* bind OK */
1014 7 : rc = listen(fd, 512);
1015 7 : if (rc != 0) {
1016 0 : SPDK_ERRLOG("listen() failed, errno = %d\n", errno);
1017 0 : close(fd);
1018 0 : fd = -1;
1019 0 : break;
1020 : }
1021 7 : enable_zcopy_impl_opts = impl_opts.enable_zerocopy_send_server;
1022 9 : } else if (type == SPDK_SOCK_CREATE_CONNECT) {
1023 9 : src_addr = SPDK_GET_FIELD(opts, src_addr, NULL, opts->opts_size);
1024 9 : src_port = SPDK_GET_FIELD(opts, src_port, 0, opts->opts_size);
1025 9 : if (src_addr != NULL || src_port != 0) {
1026 0 : snprintf(portnum, sizeof(portnum), "%"PRIu16, src_port);
1027 0 : memset(&hints, 0, sizeof hints);
1028 0 : hints.ai_family = AF_UNSPEC;
1029 0 : hints.ai_socktype = SOCK_STREAM;
1030 0 : hints.ai_flags = AI_NUMERICSERV | AI_NUMERICHOST | AI_PASSIVE;
1031 0 : rc = getaddrinfo(src_addr, src_port > 0 ? portnum : NULL,
1032 : &hints, &src_ai);
1033 0 : if (rc != 0 || src_ai == NULL) {
1034 0 : SPDK_ERRLOG("getaddrinfo() failed %s (%d)\n",
1035 : rc != 0 ? gai_strerror(rc) : "", rc);
1036 0 : close(fd);
1037 0 : fd = -1;
1038 0 : break;
1039 : }
1040 0 : rc = bind(fd, src_ai->ai_addr, src_ai->ai_addrlen);
1041 0 : if (rc != 0) {
1042 0 : SPDK_ERRLOG("bind() failed errno %d (%s:%s)\n", errno,
1043 : src_addr ? src_addr : "", portnum);
1044 0 : close(fd);
1045 0 : fd = -1;
1046 0 : freeaddrinfo(src_ai);
1047 0 : src_ai = NULL;
1048 0 : break;
1049 : }
1050 0 : freeaddrinfo(src_ai);
1051 0 : src_ai = NULL;
1052 : }
1053 9 : rc = connect(fd, res->ai_addr, res->ai_addrlen);
1054 9 : if (rc != 0) {
1055 0 : SPDK_ERRLOG("connect() failed, errno = %d\n", errno);
1056 : /* try next family */
1057 0 : close(fd);
1058 0 : fd = -1;
1059 0 : continue;
1060 : }
1061 9 : enable_zcopy_impl_opts = impl_opts.enable_zerocopy_send_client;
1062 9 : if (enable_ssl) {
1063 0 : ctx = posix_sock_create_ssl_context(TLS_client_method(), opts, &impl_opts);
1064 0 : if (!ctx) {
1065 0 : SPDK_ERRLOG("posix_sock_create_ssl_context() failed, errno = %d\n", errno);
1066 0 : close(fd);
1067 0 : fd = -1;
1068 0 : break;
1069 : }
1070 0 : ssl = ssl_sock_setup_connect(ctx, fd);
1071 0 : if (!ssl) {
1072 0 : SPDK_ERRLOG("ssl_sock_setup_connect() failed, errno = %d\n", errno);
1073 0 : close(fd);
1074 0 : fd = -1;
1075 0 : SSL_CTX_free(ctx);
1076 0 : break;
1077 : }
1078 : }
1079 : }
1080 :
1081 16 : flag = fcntl(fd, F_GETFL);
1082 16 : if (fcntl(fd, F_SETFL, flag | O_NONBLOCK) < 0) {
1083 0 : SPDK_ERRLOG("fcntl can't set nonblocking mode for socket, fd: %d (%d)\n", fd, errno);
1084 0 : SSL_free(ssl);
1085 0 : SSL_CTX_free(ctx);
1086 0 : close(fd);
1087 0 : fd = -1;
1088 0 : break;
1089 : }
1090 16 : break;
1091 : }
1092 16 : freeaddrinfo(res0);
1093 :
1094 16 : if (fd < 0) {
1095 0 : return NULL;
1096 : }
1097 :
1098 : /* Only enable zero copy for non-loopback and non-ssl sockets. */
1099 16 : enable_zcopy_user_opts = opts->zcopy && !spdk_net_is_loopback(fd) && !enable_ssl;
1100 :
1101 16 : sock = posix_sock_alloc(fd, &impl_opts, enable_zcopy_user_opts && enable_zcopy_impl_opts);
1102 16 : if (sock == NULL) {
1103 0 : SPDK_ERRLOG("sock allocation failed\n");
1104 0 : SSL_free(ssl);
1105 0 : SSL_CTX_free(ctx);
1106 0 : close(fd);
1107 0 : return NULL;
1108 : }
1109 :
1110 16 : if (ctx) {
1111 0 : sock->ctx = ctx;
1112 : }
1113 :
1114 16 : if (ssl) {
1115 0 : sock->ssl = ssl;
1116 0 : SSL_set_app_data(ssl, &sock->base.impl_opts);
1117 : }
1118 :
1119 16 : return &sock->base;
1120 : }
1121 :
1122 : static struct spdk_sock *
1123 7 : posix_sock_listen(const char *ip, int port, struct spdk_sock_opts *opts)
1124 : {
1125 7 : return posix_sock_create(ip, port, SPDK_SOCK_CREATE_LISTEN, opts, false);
1126 : }
1127 :
1128 : static struct spdk_sock *
1129 9 : posix_sock_connect(const char *ip, int port, struct spdk_sock_opts *opts)
1130 : {
1131 9 : return posix_sock_create(ip, port, SPDK_SOCK_CREATE_CONNECT, opts, false);
1132 : }
1133 :
1134 : static struct spdk_sock *
1135 10 : _posix_sock_accept(struct spdk_sock *_sock, bool enable_ssl)
1136 : {
1137 10 : struct spdk_posix_sock *sock = __posix_sock(_sock);
1138 10 : struct spdk_posix_sock_group_impl *group = __posix_group_impl(sock->base.group_impl);
1139 10 : struct sockaddr_storage sa;
1140 10 : socklen_t salen;
1141 : int rc, fd;
1142 : struct spdk_posix_sock *new_sock;
1143 : int flag;
1144 10 : SSL_CTX *ctx = 0;
1145 10 : SSL *ssl = 0;
1146 :
1147 10 : memset(&sa, 0, sizeof(sa));
1148 10 : salen = sizeof(sa);
1149 :
1150 10 : assert(sock != NULL);
1151 :
1152 : /* epoll_wait will trigger again if there is more than one request */
1153 10 : if (group && sock->socket_has_data) {
1154 0 : sock->socket_has_data = false;
1155 0 : TAILQ_REMOVE(&group->socks_with_data, sock, link);
1156 : }
1157 :
1158 10 : rc = accept(sock->fd, (struct sockaddr *)&sa, &salen);
1159 :
1160 10 : if (rc == -1) {
1161 2 : return NULL;
1162 : }
1163 :
1164 8 : fd = rc;
1165 :
1166 8 : flag = fcntl(fd, F_GETFL);
1167 8 : if ((!(flag & O_NONBLOCK)) && (fcntl(fd, F_SETFL, flag | O_NONBLOCK) < 0)) {
1168 0 : SPDK_ERRLOG("fcntl can't set nonblocking mode for socket, fd: %d (%d)\n", fd, errno);
1169 0 : close(fd);
1170 0 : return NULL;
1171 : }
1172 :
1173 : #if defined(SO_PRIORITY)
1174 : /* The priority is not inherited, so call this function again */
1175 8 : if (sock->base.opts.priority) {
1176 0 : rc = setsockopt(fd, SOL_SOCKET, SO_PRIORITY, &sock->base.opts.priority, sizeof(int));
1177 0 : if (rc != 0) {
1178 0 : close(fd);
1179 0 : return NULL;
1180 : }
1181 : }
1182 : #endif
1183 :
1184 : /* Establish SSL connection */
1185 8 : if (enable_ssl) {
1186 0 : ctx = posix_sock_create_ssl_context(TLS_server_method(), &sock->base.opts, &sock->base.impl_opts);
1187 0 : if (!ctx) {
1188 0 : SPDK_ERRLOG("posix_sock_create_ssl_context() failed, errno = %d\n", errno);
1189 0 : close(fd);
1190 0 : return NULL;
1191 : }
1192 0 : ssl = ssl_sock_setup_accept(ctx, fd);
1193 0 : if (!ssl) {
1194 0 : SPDK_ERRLOG("ssl_sock_setup_accept() failed, errno = %d\n", errno);
1195 0 : close(fd);
1196 0 : SSL_CTX_free(ctx);
1197 0 : return NULL;
1198 : }
1199 : }
1200 :
1201 : /* Inherit the zero copy feature from the listen socket */
1202 8 : new_sock = posix_sock_alloc(fd, &sock->base.impl_opts, sock->zcopy);
1203 8 : if (new_sock == NULL) {
1204 0 : close(fd);
1205 0 : SSL_free(ssl);
1206 0 : SSL_CTX_free(ctx);
1207 0 : return NULL;
1208 : }
1209 :
1210 8 : if (ctx) {
1211 0 : new_sock->ctx = ctx;
1212 : }
1213 :
1214 8 : if (ssl) {
1215 0 : new_sock->ssl = ssl;
1216 0 : SSL_set_app_data(ssl, &new_sock->base.impl_opts);
1217 : }
1218 :
1219 8 : return &new_sock->base;
1220 : }
1221 :
1222 : static struct spdk_sock *
1223 10 : posix_sock_accept(struct spdk_sock *_sock)
1224 : {
1225 10 : return _posix_sock_accept(_sock, false);
1226 : }
1227 :
1228 : static int
1229 24 : posix_sock_close(struct spdk_sock *_sock)
1230 : {
1231 24 : struct spdk_posix_sock *sock = __posix_sock(_sock);
1232 : void *pipe_buf;
1233 :
1234 24 : assert(TAILQ_EMPTY(&_sock->pending_reqs));
1235 :
1236 24 : if (sock->ssl != NULL) {
1237 0 : SSL_shutdown(sock->ssl);
1238 : }
1239 :
1240 : /* If the socket fails to close, the best choice is to
1241 : * leak the fd but continue to free the rest of the sock
1242 : * memory. */
1243 24 : close(sock->fd);
1244 :
1245 24 : SSL_free(sock->ssl);
1246 24 : SSL_CTX_free(sock->ctx);
1247 :
1248 24 : pipe_buf = spdk_pipe_destroy(sock->recv_pipe);
1249 24 : free(pipe_buf);
1250 24 : free(sock);
1251 :
1252 24 : return 0;
1253 : }
1254 :
1255 : #ifdef SPDK_ZEROCOPY
1256 : static int
1257 0 : _sock_check_zcopy(struct spdk_sock *sock)
1258 : {
1259 0 : struct spdk_posix_sock *psock = __posix_sock(sock);
1260 0 : struct msghdr msgh = {};
1261 0 : uint8_t buf[sizeof(struct cmsghdr) + sizeof(struct sock_extended_err)];
1262 : ssize_t rc;
1263 : struct sock_extended_err *serr;
1264 : struct cmsghdr *cm;
1265 : uint32_t idx;
1266 : struct spdk_sock_request *req, *treq;
1267 : bool found;
1268 :
1269 0 : msgh.msg_control = buf;
1270 0 : msgh.msg_controllen = sizeof(buf);
1271 :
1272 : while (true) {
1273 0 : rc = recvmsg(psock->fd, &msgh, MSG_ERRQUEUE);
1274 :
1275 0 : if (rc < 0) {
1276 0 : if (errno == EWOULDBLOCK || errno == EAGAIN) {
1277 0 : return 0;
1278 : }
1279 :
1280 0 : if (!TAILQ_EMPTY(&sock->pending_reqs)) {
1281 0 : SPDK_ERRLOG("Attempting to receive from ERRQUEUE yielded error, but pending list still has orphaned entries\n");
1282 : } else {
1283 0 : SPDK_WARNLOG("Recvmsg yielded an error!\n");
1284 : }
1285 0 : return 0;
1286 : }
1287 :
1288 0 : cm = CMSG_FIRSTHDR(&msgh);
1289 0 : if (!(cm &&
1290 0 : ((cm->cmsg_level == SOL_IP && cm->cmsg_type == IP_RECVERR) ||
1291 0 : (cm->cmsg_level == SOL_IPV6 && cm->cmsg_type == IPV6_RECVERR)))) {
1292 0 : SPDK_WARNLOG("Unexpected cmsg level or type!\n");
1293 0 : return 0;
1294 : }
1295 :
1296 0 : serr = (struct sock_extended_err *)CMSG_DATA(cm);
1297 0 : if (serr->ee_errno != 0 || serr->ee_origin != SO_EE_ORIGIN_ZEROCOPY) {
1298 0 : SPDK_WARNLOG("Unexpected extended error origin\n");
1299 0 : return 0;
1300 : }
1301 :
1302 : /* Most of the time, the pending_reqs array is in the exact
1303 : * order we need such that all of the requests to complete are
1304 : * in order, in the front. It is guaranteed that all requests
1305 : * belonging to the same sendmsg call are sequential, so once
1306 : * we encounter one match we can stop looping as soon as a
1307 : * non-match is found.
1308 : */
1309 0 : idx = serr->ee_info;
1310 : while (true) {
1311 0 : found = false;
1312 0 : TAILQ_FOREACH_SAFE(req, &sock->pending_reqs, internal.link, treq) {
1313 0 : if (!req->internal.is_zcopy) {
1314 : /* This wasn't a zcopy request. It was just waiting in line to complete */
1315 0 : rc = spdk_sock_request_put(sock, req, 0);
1316 0 : if (rc < 0) {
1317 0 : return rc;
1318 : }
1319 0 : } else if (req->internal.offset == idx) {
1320 0 : found = true;
1321 0 : rc = spdk_sock_request_put(sock, req, 0);
1322 0 : if (rc < 0) {
1323 0 : return rc;
1324 : }
1325 0 : } else if (found) {
1326 0 : break;
1327 : }
1328 : }
1329 :
1330 0 : if (idx == serr->ee_data) {
1331 0 : break;
1332 : }
1333 :
1334 0 : if (idx == UINT32_MAX) {
1335 0 : idx = 0;
1336 : } else {
1337 0 : idx++;
1338 : }
1339 : }
1340 : }
1341 :
1342 : return 0;
1343 : }
1344 : #endif
1345 :
1346 : static int
1347 29 : _sock_flush(struct spdk_sock *sock)
1348 : {
1349 29 : struct spdk_posix_sock *psock = __posix_sock(sock);
1350 29 : struct msghdr msg = {};
1351 29 : int flags;
1352 29 : struct iovec iovs[IOV_BATCH_SIZE];
1353 : int iovcnt;
1354 : int retval;
1355 : struct spdk_sock_request *req;
1356 : int i;
1357 : ssize_t rc, sent;
1358 : unsigned int offset;
1359 : size_t len;
1360 29 : bool is_zcopy = false;
1361 :
1362 : /* Can't flush from within a callback or we end up with recursive calls */
1363 29 : if (sock->cb_cnt > 0) {
1364 0 : errno = EAGAIN;
1365 0 : return -1;
1366 : }
1367 :
1368 : #ifdef SPDK_ZEROCOPY
1369 29 : if (psock->zcopy) {
1370 0 : flags = MSG_ZEROCOPY | MSG_NOSIGNAL;
1371 : } else
1372 : #endif
1373 : {
1374 29 : flags = MSG_NOSIGNAL;
1375 : }
1376 :
1377 29 : iovcnt = spdk_sock_prep_reqs(sock, iovs, 0, NULL, &flags);
1378 29 : if (iovcnt == 0) {
1379 22 : return 0;
1380 : }
1381 :
1382 : #ifdef SPDK_ZEROCOPY
1383 7 : is_zcopy = flags & MSG_ZEROCOPY;
1384 : #endif
1385 :
1386 : /* Perform the vectored write */
1387 7 : msg.msg_iov = iovs;
1388 7 : msg.msg_iovlen = iovcnt;
1389 :
1390 7 : if (psock->ssl) {
1391 0 : rc = SSL_writev(psock->ssl, iovs, iovcnt);
1392 : } else {
1393 7 : rc = sendmsg(psock->fd, &msg, flags);
1394 : }
1395 7 : if (rc <= 0) {
1396 0 : if (rc == 0 || errno == EAGAIN || errno == EWOULDBLOCK || (errno == ENOBUFS && psock->zcopy)) {
1397 0 : errno = EAGAIN;
1398 : }
1399 0 : return -1;
1400 : }
1401 :
1402 7 : sent = rc;
1403 :
1404 7 : if (is_zcopy) {
1405 : /* Handling overflow case, because we use psock->sendmsg_idx - 1 for the
1406 : * req->internal.offset, so sendmsg_idx should not be zero */
1407 0 : if (spdk_unlikely(psock->sendmsg_idx == UINT32_MAX)) {
1408 0 : psock->sendmsg_idx = 1;
1409 : } else {
1410 0 : psock->sendmsg_idx++;
1411 : }
1412 : }
1413 :
1414 : /* Consume the requests that were actually written */
1415 7 : req = TAILQ_FIRST(&sock->queued_reqs);
1416 8 : while (req) {
1417 8 : offset = req->internal.offset;
1418 :
1419 : /* req->internal.is_zcopy is true when the whole req or part of it is sent with zerocopy */
1420 8 : req->internal.is_zcopy = is_zcopy;
1421 :
1422 20 : for (i = 0; i < req->iovcnt; i++) {
1423 : /* Advance by the offset first */
1424 14 : if (offset >= SPDK_SOCK_REQUEST_IOV(req, i)->iov_len) {
1425 1 : offset -= SPDK_SOCK_REQUEST_IOV(req, i)->iov_len;
1426 1 : continue;
1427 : }
1428 :
1429 : /* Calculate the remaining length of this element */
1430 13 : len = SPDK_SOCK_REQUEST_IOV(req, i)->iov_len - offset;
1431 :
1432 13 : if (len > (size_t)rc) {
1433 : /* This element was partially sent. */
1434 2 : req->internal.offset += rc;
1435 2 : return sent;
1436 : }
1437 :
1438 11 : offset = 0;
1439 11 : req->internal.offset += len;
1440 11 : rc -= len;
1441 : }
1442 :
1443 : /* Handled a full request. */
1444 6 : spdk_sock_request_pend(sock, req);
1445 :
1446 6 : if (!req->internal.is_zcopy && req == TAILQ_FIRST(&sock->pending_reqs)) {
1447 : /* The sendmsg syscall above isn't currently asynchronous,
1448 : * so it's already done. */
1449 6 : retval = spdk_sock_request_put(sock, req, 0);
1450 6 : if (retval) {
1451 1 : break;
1452 : }
1453 : } else {
1454 : /* Re-use the offset field to hold the sendmsg call index. The
1455 : * index is 0 based, so subtract one here because we've already
1456 : * incremented above. */
1457 0 : req->internal.offset = psock->sendmsg_idx - 1;
1458 : }
1459 :
1460 5 : if (rc == 0) {
1461 4 : break;
1462 : }
1463 :
1464 1 : req = TAILQ_FIRST(&sock->queued_reqs);
1465 : }
1466 :
1467 5 : return sent;
1468 : }
1469 :
1470 : static int
1471 1 : posix_sock_flush(struct spdk_sock *sock)
1472 : {
1473 : #ifdef SPDK_ZEROCOPY
1474 1 : struct spdk_posix_sock *psock = __posix_sock(sock);
1475 :
1476 1 : if (psock->zcopy && !TAILQ_EMPTY(&sock->pending_reqs)) {
1477 0 : _sock_check_zcopy(sock);
1478 : }
1479 : #endif
1480 :
1481 1 : return _sock_flush(sock);
1482 : }
1483 :
1484 : static ssize_t
1485 0 : posix_sock_recv_from_pipe(struct spdk_posix_sock *sock, struct iovec *diov, int diovcnt)
1486 : {
1487 0 : struct iovec siov[2];
1488 : int sbytes;
1489 : ssize_t bytes;
1490 : struct spdk_posix_sock_group_impl *group;
1491 :
1492 0 : sbytes = spdk_pipe_reader_get_buffer(sock->recv_pipe, sock->recv_buf_sz, siov);
1493 0 : if (sbytes < 0) {
1494 0 : errno = EINVAL;
1495 0 : return -1;
1496 0 : } else if (sbytes == 0) {
1497 0 : errno = EAGAIN;
1498 0 : return -1;
1499 : }
1500 :
1501 0 : bytes = spdk_iovcpy(siov, 2, diov, diovcnt);
1502 :
1503 0 : if (bytes == 0) {
1504 : /* The only way this happens is if diov is 0 length */
1505 0 : errno = EINVAL;
1506 0 : return -1;
1507 : }
1508 :
1509 0 : spdk_pipe_reader_advance(sock->recv_pipe, bytes);
1510 :
1511 : /* If we drained the pipe, mark it appropriately */
1512 0 : if (spdk_pipe_reader_bytes_available(sock->recv_pipe) == 0) {
1513 0 : assert(sock->pipe_has_data == true);
1514 :
1515 0 : group = __posix_group_impl(sock->base.group_impl);
1516 0 : if (group && !sock->socket_has_data) {
1517 0 : TAILQ_REMOVE(&group->socks_with_data, sock, link);
1518 : }
1519 :
1520 0 : sock->pipe_has_data = false;
1521 : }
1522 :
1523 0 : return bytes;
1524 : }
1525 :
1526 : static inline ssize_t
1527 0 : posix_sock_read(struct spdk_posix_sock *sock)
1528 : {
1529 0 : struct iovec iov[2];
1530 : int bytes_avail, bytes_recvd;
1531 : struct spdk_posix_sock_group_impl *group;
1532 :
1533 0 : bytes_avail = spdk_pipe_writer_get_buffer(sock->recv_pipe, sock->recv_buf_sz, iov);
1534 :
1535 0 : if (bytes_avail <= 0) {
1536 0 : return bytes_avail;
1537 : }
1538 :
1539 0 : if (sock->ssl) {
1540 0 : bytes_recvd = SSL_readv(sock->ssl, iov, 2);
1541 : } else {
1542 0 : bytes_recvd = readv(sock->fd, iov, 2);
1543 : }
1544 :
1545 0 : assert(sock->pipe_has_data == false);
1546 :
1547 0 : if (bytes_recvd <= 0) {
1548 : /* Errors count as draining the socket data */
1549 0 : if (sock->base.group_impl && sock->socket_has_data) {
1550 0 : group = __posix_group_impl(sock->base.group_impl);
1551 0 : TAILQ_REMOVE(&group->socks_with_data, sock, link);
1552 : }
1553 :
1554 0 : sock->socket_has_data = false;
1555 :
1556 0 : return bytes_recvd;
1557 : }
1558 :
1559 0 : spdk_pipe_writer_advance(sock->recv_pipe, bytes_recvd);
1560 :
1561 : #if DEBUG
1562 0 : if (sock->base.group_impl) {
1563 0 : assert(sock->socket_has_data == true);
1564 : }
1565 : #endif
1566 :
1567 0 : sock->pipe_has_data = true;
1568 0 : if (bytes_recvd < bytes_avail) {
1569 : /* We drained the kernel socket entirely. */
1570 0 : sock->socket_has_data = false;
1571 : }
1572 :
1573 0 : return bytes_recvd;
1574 : }
1575 :
1576 : static ssize_t
1577 9 : posix_sock_readv(struct spdk_sock *_sock, struct iovec *iov, int iovcnt)
1578 : {
1579 9 : struct spdk_posix_sock *sock = __posix_sock(_sock);
1580 9 : struct spdk_posix_sock_group_impl *group = __posix_group_impl(sock->base.group_impl);
1581 : int rc, i;
1582 : size_t len;
1583 :
1584 9 : if (sock->recv_pipe == NULL) {
1585 9 : assert(sock->pipe_has_data == false);
1586 9 : if (group && sock->socket_has_data) {
1587 5 : sock->socket_has_data = false;
1588 5 : TAILQ_REMOVE(&group->socks_with_data, sock, link);
1589 : }
1590 9 : if (sock->ssl) {
1591 0 : return SSL_readv(sock->ssl, iov, iovcnt);
1592 : } else {
1593 9 : return readv(sock->fd, iov, iovcnt);
1594 : }
1595 : }
1596 :
1597 : /* If the socket is not in a group, we must assume it always has
1598 : * data waiting for us because it is not epolled */
1599 0 : if (!sock->pipe_has_data && (group == NULL || sock->socket_has_data)) {
1600 : /* If the user is receiving a sufficiently large amount of data,
1601 : * receive directly to their buffers. */
1602 0 : len = 0;
1603 0 : for (i = 0; i < iovcnt; i++) {
1604 0 : len += iov[i].iov_len;
1605 : }
1606 :
1607 0 : if (len >= MIN_SOCK_PIPE_SIZE) {
1608 : /* TODO: Should this detect if kernel socket is drained? */
1609 0 : if (sock->ssl) {
1610 0 : return SSL_readv(sock->ssl, iov, iovcnt);
1611 : } else {
1612 0 : return readv(sock->fd, iov, iovcnt);
1613 : }
1614 : }
1615 :
1616 : /* Otherwise, do a big read into our pipe */
1617 0 : rc = posix_sock_read(sock);
1618 0 : if (rc <= 0) {
1619 0 : return rc;
1620 : }
1621 : }
1622 :
1623 0 : return posix_sock_recv_from_pipe(sock, iov, iovcnt);
1624 : }
1625 :
1626 : static ssize_t
1627 7 : posix_sock_recv(struct spdk_sock *sock, void *buf, size_t len)
1628 : {
1629 7 : struct iovec iov[1];
1630 :
1631 7 : iov[0].iov_base = buf;
1632 7 : iov[0].iov_len = len;
1633 :
1634 7 : return posix_sock_readv(sock, iov, 1);
1635 : }
1636 :
1637 : static ssize_t
1638 7 : posix_sock_writev(struct spdk_sock *_sock, struct iovec *iov, int iovcnt)
1639 : {
1640 7 : struct spdk_posix_sock *sock = __posix_sock(_sock);
1641 : int rc;
1642 :
1643 : /* In order to process a writev, we need to flush any asynchronous writes
1644 : * first. */
1645 7 : rc = _sock_flush(_sock);
1646 7 : if (rc < 0) {
1647 0 : return rc;
1648 : }
1649 :
1650 7 : if (!TAILQ_EMPTY(&_sock->queued_reqs)) {
1651 : /* We weren't able to flush all requests */
1652 0 : errno = EAGAIN;
1653 0 : return -1;
1654 : }
1655 :
1656 7 : if (sock->ssl) {
1657 0 : return SSL_writev(sock->ssl, iov, iovcnt);
1658 : } else {
1659 7 : return writev(sock->fd, iov, iovcnt);
1660 : }
1661 : }
1662 :
1663 : static int
1664 0 : posix_sock_recv_next(struct spdk_sock *_sock, void **buf, void **ctx)
1665 : {
1666 0 : struct spdk_posix_sock *sock = __posix_sock(_sock);
1667 0 : struct iovec iov;
1668 : ssize_t rc;
1669 :
1670 0 : if (sock->recv_pipe != NULL) {
1671 0 : errno = ENOTSUP;
1672 0 : return -1;
1673 : }
1674 :
1675 0 : iov.iov_len = spdk_sock_group_get_buf(_sock->group_impl->group, &iov.iov_base, ctx);
1676 0 : if (iov.iov_len == 0) {
1677 0 : errno = ENOBUFS;
1678 0 : return -1;
1679 : }
1680 :
1681 0 : rc = posix_sock_readv(_sock, &iov, 1);
1682 0 : if (rc <= 0) {
1683 0 : spdk_sock_group_provide_buf(_sock->group_impl->group, iov.iov_base, iov.iov_len, *ctx);
1684 0 : return rc;
1685 : }
1686 :
1687 0 : *buf = iov.iov_base;
1688 :
1689 0 : return rc;
1690 : }
1691 :
1692 : static void
1693 2 : posix_sock_writev_async(struct spdk_sock *sock, struct spdk_sock_request *req)
1694 : {
1695 : int rc;
1696 :
1697 2 : spdk_sock_request_queue(sock, req);
1698 :
1699 : /* If there are a sufficient number queued, just flush them out immediately. */
1700 2 : if (sock->queued_iovcnt >= IOV_BATCH_SIZE) {
1701 0 : rc = _sock_flush(sock);
1702 0 : if (rc < 0 && errno != EAGAIN) {
1703 0 : spdk_sock_abort_requests(sock);
1704 : }
1705 : }
1706 2 : }
1707 :
1708 : static int
1709 1 : posix_sock_set_recvlowat(struct spdk_sock *_sock, int nbytes)
1710 : {
1711 1 : struct spdk_posix_sock *sock = __posix_sock(_sock);
1712 1 : int val;
1713 : int rc;
1714 :
1715 1 : assert(sock != NULL);
1716 :
1717 1 : val = nbytes;
1718 1 : rc = setsockopt(sock->fd, SOL_SOCKET, SO_RCVLOWAT, &val, sizeof val);
1719 1 : if (rc != 0) {
1720 0 : return -1;
1721 : }
1722 1 : return 0;
1723 : }
1724 :
1725 : static bool
1726 1 : posix_sock_is_ipv6(struct spdk_sock *_sock)
1727 : {
1728 1 : struct spdk_posix_sock *sock = __posix_sock(_sock);
1729 1 : struct sockaddr_storage sa;
1730 1 : socklen_t salen;
1731 : int rc;
1732 :
1733 1 : assert(sock != NULL);
1734 :
1735 1 : memset(&sa, 0, sizeof sa);
1736 1 : salen = sizeof sa;
1737 1 : rc = getsockname(sock->fd, (struct sockaddr *) &sa, &salen);
1738 1 : if (rc != 0) {
1739 0 : SPDK_ERRLOG("getsockname() failed (errno=%d)\n", errno);
1740 0 : return false;
1741 : }
1742 :
1743 1 : return (sa.ss_family == AF_INET6);
1744 : }
1745 :
1746 : static bool
1747 1 : posix_sock_is_ipv4(struct spdk_sock *_sock)
1748 : {
1749 1 : struct spdk_posix_sock *sock = __posix_sock(_sock);
1750 1 : struct sockaddr_storage sa;
1751 1 : socklen_t salen;
1752 : int rc;
1753 :
1754 1 : assert(sock != NULL);
1755 :
1756 1 : memset(&sa, 0, sizeof sa);
1757 1 : salen = sizeof sa;
1758 1 : rc = getsockname(sock->fd, (struct sockaddr *) &sa, &salen);
1759 1 : if (rc != 0) {
1760 0 : SPDK_ERRLOG("getsockname() failed (errno=%d)\n", errno);
1761 0 : return false;
1762 : }
1763 :
1764 1 : return (sa.ss_family == AF_INET);
1765 : }
1766 :
1767 : static bool
1768 3 : posix_sock_is_connected(struct spdk_sock *_sock)
1769 : {
1770 3 : struct spdk_posix_sock *sock = __posix_sock(_sock);
1771 3 : uint8_t byte;
1772 : int rc;
1773 :
1774 3 : rc = recv(sock->fd, &byte, 1, MSG_PEEK);
1775 3 : if (rc == 0) {
1776 1 : return false;
1777 : }
1778 :
1779 2 : if (rc < 0) {
1780 2 : if (errno == EAGAIN || errno == EWOULDBLOCK) {
1781 2 : return true;
1782 : }
1783 :
1784 0 : return false;
1785 : }
1786 :
1787 0 : return true;
1788 : }
1789 :
1790 : static struct spdk_sock_group_impl *
1791 1 : posix_sock_group_impl_get_optimal(struct spdk_sock *_sock, struct spdk_sock_group_impl *hint)
1792 : {
1793 1 : struct spdk_posix_sock *sock = __posix_sock(_sock);
1794 1 : struct spdk_sock_group_impl *group_impl;
1795 :
1796 1 : if (sock->placement_id != -1) {
1797 0 : spdk_sock_map_lookup(&g_map, sock->placement_id, &group_impl, hint);
1798 0 : return group_impl;
1799 : }
1800 :
1801 1 : return NULL;
1802 : }
1803 :
1804 : static struct spdk_sock_group_impl *
1805 12 : _sock_group_impl_create(uint32_t enable_placement_id)
1806 : {
1807 : struct spdk_posix_sock_group_impl *group_impl;
1808 : int fd;
1809 :
1810 : #if defined(SPDK_EPOLL)
1811 12 : fd = epoll_create1(0);
1812 : #elif defined(SPDK_KEVENT)
1813 : fd = kqueue();
1814 : #endif
1815 12 : if (fd == -1) {
1816 0 : return NULL;
1817 : }
1818 :
1819 12 : group_impl = calloc(1, sizeof(*group_impl));
1820 12 : if (group_impl == NULL) {
1821 0 : SPDK_ERRLOG("group_impl allocation failed\n");
1822 0 : close(fd);
1823 0 : return NULL;
1824 : }
1825 :
1826 12 : group_impl->pipe_group = spdk_pipe_group_create();
1827 12 : if (group_impl->pipe_group == NULL) {
1828 0 : SPDK_ERRLOG("pipe_group allocation failed\n");
1829 0 : free(group_impl);
1830 0 : close(fd);
1831 0 : return NULL;
1832 : }
1833 :
1834 12 : group_impl->fd = fd;
1835 12 : TAILQ_INIT(&group_impl->socks_with_data);
1836 12 : group_impl->placement_id = -1;
1837 :
1838 12 : if (enable_placement_id == PLACEMENT_CPU) {
1839 0 : spdk_sock_map_insert(&g_map, spdk_env_get_current_core(), &group_impl->base);
1840 0 : group_impl->placement_id = spdk_env_get_current_core();
1841 : }
1842 :
1843 12 : return &group_impl->base;
1844 : }
1845 :
1846 : static struct spdk_sock_group_impl *
1847 6 : posix_sock_group_impl_create(void)
1848 : {
1849 6 : return _sock_group_impl_create(g_posix_impl_opts.enable_placement_id);
1850 : }
1851 :
1852 : static struct spdk_sock_group_impl *
1853 6 : ssl_sock_group_impl_create(void)
1854 : {
1855 6 : return _sock_group_impl_create(g_ssl_impl_opts.enable_placement_id);
1856 : }
1857 :
1858 : static void
1859 0 : posix_sock_mark(struct spdk_posix_sock_group_impl *group, struct spdk_posix_sock *sock,
1860 : int placement_id)
1861 : {
1862 : #if defined(SO_MARK)
1863 : int rc;
1864 :
1865 0 : rc = setsockopt(sock->fd, SOL_SOCKET, SO_MARK,
1866 : &placement_id, sizeof(placement_id));
1867 0 : if (rc != 0) {
1868 : /* Not fatal */
1869 0 : SPDK_ERRLOG("Error setting SO_MARK\n");
1870 0 : return;
1871 : }
1872 :
1873 0 : rc = spdk_sock_map_insert(&g_map, placement_id, &group->base);
1874 0 : if (rc != 0) {
1875 : /* Not fatal */
1876 0 : SPDK_ERRLOG("Failed to insert sock group into map: %d\n", rc);
1877 0 : return;
1878 : }
1879 :
1880 0 : sock->placement_id = placement_id;
1881 : #endif
1882 : }
1883 :
1884 : static void
1885 0 : posix_sock_update_mark(struct spdk_sock_group_impl *_group, struct spdk_sock *_sock)
1886 : {
1887 0 : struct spdk_posix_sock_group_impl *group = __posix_group_impl(_group);
1888 :
1889 0 : if (group->placement_id == -1) {
1890 0 : group->placement_id = spdk_sock_map_find_free(&g_map);
1891 :
1892 : /* If a free placement id is found, update existing sockets in this group */
1893 0 : if (group->placement_id != -1) {
1894 : struct spdk_sock *sock, *tmp;
1895 :
1896 0 : TAILQ_FOREACH_SAFE(sock, &_group->socks, link, tmp) {
1897 0 : posix_sock_mark(group, __posix_sock(sock), group->placement_id);
1898 : }
1899 : }
1900 : }
1901 :
1902 0 : if (group->placement_id != -1) {
1903 : /*
1904 : * group placement id is already determined for this poll group.
1905 : * Mark socket with group's placement id.
1906 : */
1907 0 : posix_sock_mark(group, __posix_sock(_sock), group->placement_id);
1908 : }
1909 0 : }
1910 :
1911 : static int
1912 5 : posix_sock_group_impl_add_sock(struct spdk_sock_group_impl *_group, struct spdk_sock *_sock)
1913 : {
1914 5 : struct spdk_posix_sock_group_impl *group = __posix_group_impl(_group);
1915 5 : struct spdk_posix_sock *sock = __posix_sock(_sock);
1916 : int rc;
1917 :
1918 : #if defined(SPDK_EPOLL)
1919 5 : struct epoll_event event;
1920 :
1921 5 : memset(&event, 0, sizeof(event));
1922 : /* EPOLLERR is always on even if we don't set it, but be explicit for clarity */
1923 5 : event.events = EPOLLIN | EPOLLERR;
1924 5 : if (spdk_interrupt_mode_is_enabled()) {
1925 0 : event.events |= EPOLLOUT;
1926 : }
1927 :
1928 5 : event.data.ptr = sock;
1929 :
1930 5 : rc = epoll_ctl(group->fd, EPOLL_CTL_ADD, sock->fd, &event);
1931 : #elif defined(SPDK_KEVENT)
1932 : struct kevent event;
1933 : struct timespec ts = {0};
1934 :
1935 : EV_SET(&event, sock->fd, EVFILT_READ, EV_ADD, 0, 0, sock);
1936 :
1937 : rc = kevent(group->fd, &event, 1, NULL, 0, &ts);
1938 : #endif
1939 :
1940 5 : if (rc != 0) {
1941 0 : return rc;
1942 : }
1943 :
1944 : /* switched from another polling group due to scheduling */
1945 5 : if (spdk_unlikely(sock->recv_pipe != NULL &&
1946 : (spdk_pipe_reader_bytes_available(sock->recv_pipe) > 0))) {
1947 0 : sock->pipe_has_data = true;
1948 0 : sock->socket_has_data = false;
1949 0 : TAILQ_INSERT_TAIL(&group->socks_with_data, sock, link);
1950 5 : } else if (sock->recv_pipe != NULL) {
1951 0 : rc = spdk_pipe_group_add(group->pipe_group, sock->recv_pipe);
1952 0 : assert(rc == 0);
1953 : }
1954 :
1955 5 : if (_sock->impl_opts.enable_placement_id == PLACEMENT_MARK) {
1956 0 : posix_sock_update_mark(_group, _sock);
1957 5 : } else if (sock->placement_id != -1) {
1958 0 : rc = spdk_sock_map_insert(&g_map, sock->placement_id, &group->base);
1959 0 : if (rc != 0) {
1960 0 : SPDK_ERRLOG("Failed to insert sock group into map: %d\n", rc);
1961 : /* Do not treat this as an error. The system will continue running. */
1962 : }
1963 : }
1964 :
1965 5 : return rc;
1966 : }
1967 :
1968 : static int
1969 5 : posix_sock_group_impl_remove_sock(struct spdk_sock_group_impl *_group, struct spdk_sock *_sock)
1970 : {
1971 5 : struct spdk_posix_sock_group_impl *group = __posix_group_impl(_group);
1972 5 : struct spdk_posix_sock *sock = __posix_sock(_sock);
1973 : int rc;
1974 :
1975 5 : if (sock->pipe_has_data || sock->socket_has_data) {
1976 0 : TAILQ_REMOVE(&group->socks_with_data, sock, link);
1977 0 : sock->pipe_has_data = false;
1978 0 : sock->socket_has_data = false;
1979 5 : } else if (sock->recv_pipe != NULL) {
1980 0 : rc = spdk_pipe_group_remove(group->pipe_group, sock->recv_pipe);
1981 0 : assert(rc == 0);
1982 : }
1983 :
1984 5 : if (sock->placement_id != -1) {
1985 0 : spdk_sock_map_release(&g_map, sock->placement_id);
1986 : }
1987 :
1988 : #if defined(SPDK_EPOLL)
1989 5 : struct epoll_event event;
1990 :
1991 : /* Event parameter is ignored but some old kernel version still require it. */
1992 5 : rc = epoll_ctl(group->fd, EPOLL_CTL_DEL, sock->fd, &event);
1993 : #elif defined(SPDK_KEVENT)
1994 : struct kevent event;
1995 : struct timespec ts = {0};
1996 :
1997 : EV_SET(&event, sock->fd, EVFILT_READ, EV_DELETE, 0, 0, NULL);
1998 :
1999 : rc = kevent(group->fd, &event, 1, NULL, 0, &ts);
2000 : if (rc == 0 && event.flags & EV_ERROR) {
2001 : rc = -1;
2002 : errno = event.data;
2003 : }
2004 : #endif
2005 :
2006 5 : spdk_sock_abort_requests(_sock);
2007 :
2008 5 : return rc;
2009 : }
2010 :
2011 : static int
2012 7 : posix_sock_group_impl_poll(struct spdk_sock_group_impl *_group, int max_events,
2013 : struct spdk_sock **socks)
2014 : {
2015 7 : struct spdk_posix_sock_group_impl *group = __posix_group_impl(_group);
2016 : struct spdk_sock *sock, *tmp;
2017 : int num_events, i, rc;
2018 : struct spdk_posix_sock *psock, *ptmp;
2019 : #if defined(SPDK_EPOLL)
2020 7 : struct epoll_event events[MAX_EVENTS_PER_POLL];
2021 : #elif defined(SPDK_KEVENT)
2022 : struct kevent events[MAX_EVENTS_PER_POLL];
2023 : struct timespec ts = {0};
2024 : #endif
2025 :
2026 : #ifdef SPDK_ZEROCOPY
2027 : /* When all of the following conditions are met
2028 : * - non-blocking socket
2029 : * - zero copy is enabled
2030 : * - interrupts suppressed (i.e. busy polling)
2031 : * - the NIC tx queue is full at the time sendmsg() is called
2032 : * - epoll_wait determines there is an EPOLLIN event for the socket
2033 : * then we can get into a situation where data we've sent is queued
2034 : * up in the kernel network stack, but interrupts have been suppressed
2035 : * because other traffic is flowing so the kernel misses the signal
2036 : * to flush the software tx queue. If there wasn't incoming data
2037 : * pending on the socket, then epoll_wait would have been sufficient
2038 : * to kick off the send operation, but since there is a pending event
2039 : * epoll_wait does not trigger the necessary operation.
2040 : *
2041 : * We deal with this by checking for all of the above conditions and
2042 : * additionally looking for EPOLLIN events that were not consumed from
2043 : * the last poll loop. We take this to mean that the upper layer is
2044 : * unable to consume them because it is blocked waiting for resources
2045 : * to free up, and those resources are most likely freed in response
2046 : * to a pending asynchronous write completing.
2047 : *
2048 : * Additionally, sockets that have the same placement_id actually share
2049 : * an underlying hardware queue. That means polling one of them is
2050 : * equivalent to polling all of them. As a quick mechanism to avoid
2051 : * making extra poll() calls, stash the last placement_id during the loop
2052 : * and only poll if it's not the same. The overwhelmingly common case
2053 : * is that all sockets in this list have the same placement_id because
2054 : * SPDK is intentionally grouping sockets by that value, so even
2055 : * though this won't stop all extra calls to poll(), it's very fast
2056 : * and will catch all of them in practice.
2057 : */
2058 7 : int last_placement_id = -1;
2059 :
2060 7 : TAILQ_FOREACH(psock, &group->socks_with_data, link) {
2061 0 : if (psock->zcopy && psock->placement_id >= 0 &&
2062 0 : psock->placement_id != last_placement_id) {
2063 0 : struct pollfd pfd = {psock->fd, POLLIN | POLLERR, 0};
2064 :
2065 0 : poll(&pfd, 1, 0);
2066 0 : last_placement_id = psock->placement_id;
2067 : }
2068 : }
2069 : #endif
2070 :
2071 : /* This must be a TAILQ_FOREACH_SAFE because while flushing,
2072 : * a completion callback could remove the sock from the
2073 : * group. */
2074 22 : TAILQ_FOREACH_SAFE(sock, &_group->socks, link, tmp) {
2075 15 : rc = _sock_flush(sock);
2076 15 : if (rc < 0 && errno != EAGAIN) {
2077 0 : spdk_sock_abort_requests(sock);
2078 : }
2079 : }
2080 :
2081 7 : assert(max_events > 0);
2082 :
2083 : #if defined(SPDK_EPOLL)
2084 7 : num_events = epoll_wait(group->fd, events, max_events, 0);
2085 : #elif defined(SPDK_KEVENT)
2086 : num_events = kevent(group->fd, NULL, 0, events, max_events, &ts);
2087 : #endif
2088 :
2089 7 : if (num_events == -1) {
2090 0 : return -1;
2091 7 : } else if (num_events == 0 && !TAILQ_EMPTY(&_group->socks)) {
2092 1 : sock = TAILQ_FIRST(&_group->socks);
2093 1 : psock = __posix_sock(sock);
2094 : /* poll() is called here to busy poll the queue associated with
2095 : * first socket in list and potentially reap incoming data.
2096 : */
2097 1 : if (sock->opts.priority) {
2098 0 : struct pollfd pfd = {0, 0, 0};
2099 :
2100 0 : pfd.fd = psock->fd;
2101 0 : pfd.events = POLLIN | POLLERR;
2102 0 : poll(&pfd, 1, 0);
2103 : }
2104 : }
2105 :
2106 12 : for (i = 0; i < num_events; i++) {
2107 : #if defined(SPDK_EPOLL)
2108 5 : sock = events[i].data.ptr;
2109 5 : psock = __posix_sock(sock);
2110 :
2111 : #ifdef SPDK_ZEROCOPY
2112 5 : if (events[i].events & EPOLLERR) {
2113 0 : rc = _sock_check_zcopy(sock);
2114 : /* If the socket was closed or removed from
2115 : * the group in response to a send ack, don't
2116 : * add it to the array here. */
2117 0 : if (rc || sock->cb_fn == NULL) {
2118 0 : continue;
2119 : }
2120 : }
2121 : #endif
2122 5 : if ((events[i].events & EPOLLIN) == 0) {
2123 0 : continue;
2124 : }
2125 :
2126 : #elif defined(SPDK_KEVENT)
2127 : sock = events[i].udata;
2128 : psock = __posix_sock(sock);
2129 : #endif
2130 :
2131 : /* If the socket is not already in the list, add it now */
2132 5 : if (!psock->socket_has_data && !psock->pipe_has_data) {
2133 5 : TAILQ_INSERT_TAIL(&group->socks_with_data, psock, link);
2134 : }
2135 5 : psock->socket_has_data = true;
2136 : }
2137 :
2138 7 : num_events = 0;
2139 :
2140 12 : TAILQ_FOREACH_SAFE(psock, &group->socks_with_data, link, ptmp) {
2141 5 : if (num_events == max_events) {
2142 0 : break;
2143 : }
2144 :
2145 : /* If the socket's cb_fn is NULL, just remove it from the
2146 : * list and do not add it to socks array */
2147 5 : if (spdk_unlikely(psock->base.cb_fn == NULL)) {
2148 0 : psock->socket_has_data = false;
2149 0 : psock->pipe_has_data = false;
2150 0 : TAILQ_REMOVE(&group->socks_with_data, psock, link);
2151 0 : continue;
2152 : }
2153 :
2154 5 : socks[num_events++] = &psock->base;
2155 : }
2156 :
2157 : /* Cycle the has_data list so that each time we poll things aren't
2158 : * in the same order. Say we have 6 sockets in the list, named as follows:
2159 : * A B C D E F
2160 : * And all 6 sockets had epoll events, but max_events is only 3. That means
2161 : * psock currently points at D. We want to rearrange the list to the following:
2162 : * D E F A B C
2163 : *
2164 : * The variables below are named according to this example to make it easier to
2165 : * follow the swaps.
2166 : */
2167 7 : if (psock != NULL) {
2168 : struct spdk_posix_sock *pa, *pc, *pd, *pf;
2169 :
2170 : /* Capture pointers to the elements we need */
2171 0 : pd = psock;
2172 0 : pc = TAILQ_PREV(pd, spdk_has_data_list, link);
2173 0 : pa = TAILQ_FIRST(&group->socks_with_data);
2174 0 : pf = TAILQ_LAST(&group->socks_with_data, spdk_has_data_list);
2175 :
2176 : /* Break the link between C and D */
2177 0 : pc->link.tqe_next = NULL;
2178 :
2179 : /* Connect F to A */
2180 0 : pf->link.tqe_next = pa;
2181 0 : pa->link.tqe_prev = &pf->link.tqe_next;
2182 :
2183 : /* Fix up the list first/last pointers */
2184 0 : group->socks_with_data.tqh_first = pd;
2185 0 : group->socks_with_data.tqh_last = &pc->link.tqe_next;
2186 :
2187 : /* D is in front of the list, make tqe prev pointer point to the head of list */
2188 0 : pd->link.tqe_prev = &group->socks_with_data.tqh_first;
2189 : }
2190 :
2191 7 : return num_events;
2192 : }
2193 :
2194 : static int
2195 0 : posix_sock_group_impl_register_interrupt(struct spdk_sock_group_impl *_group, uint32_t events,
2196 : spdk_interrupt_fn fn, void *arg, const char *name)
2197 : {
2198 0 : struct spdk_posix_sock_group_impl *group = __posix_group_impl(_group);
2199 :
2200 0 : group->intr = spdk_interrupt_register_for_events(group->fd, events, fn, arg, name);
2201 :
2202 0 : return group->intr ? 0 : -1;
2203 : }
2204 :
2205 : static void
2206 0 : posix_sock_group_impl_unregister_interrupt(struct spdk_sock_group_impl *_group)
2207 : {
2208 0 : struct spdk_posix_sock_group_impl *group = __posix_group_impl(_group);
2209 :
2210 0 : spdk_interrupt_unregister(&group->intr);
2211 0 : }
2212 :
2213 : static int
2214 12 : _sock_group_impl_close(struct spdk_sock_group_impl *_group, uint32_t enable_placement_id)
2215 : {
2216 12 : struct spdk_posix_sock_group_impl *group = __posix_group_impl(_group);
2217 : int rc;
2218 :
2219 12 : if (enable_placement_id == PLACEMENT_CPU) {
2220 0 : spdk_sock_map_release(&g_map, spdk_env_get_current_core());
2221 : }
2222 :
2223 12 : spdk_pipe_group_destroy(group->pipe_group);
2224 12 : rc = close(group->fd);
2225 12 : free(group);
2226 12 : return rc;
2227 : }
2228 :
2229 : static int
2230 6 : posix_sock_group_impl_close(struct spdk_sock_group_impl *_group)
2231 : {
2232 6 : return _sock_group_impl_close(_group, g_posix_impl_opts.enable_placement_id);
2233 : }
2234 :
2235 : static int
2236 6 : ssl_sock_group_impl_close(struct spdk_sock_group_impl *_group)
2237 : {
2238 6 : return _sock_group_impl_close(_group, g_ssl_impl_opts.enable_placement_id);
2239 : }
2240 :
2241 : static struct spdk_net_impl g_posix_net_impl = {
2242 : .name = "posix",
2243 : .getaddr = posix_sock_getaddr,
2244 : .get_interface_name = posix_sock_get_interface_name,
2245 : .get_numa_socket_id = posix_sock_get_numa_socket_id,
2246 : .connect = posix_sock_connect,
2247 : .listen = posix_sock_listen,
2248 : .accept = posix_sock_accept,
2249 : .close = posix_sock_close,
2250 : .recv = posix_sock_recv,
2251 : .readv = posix_sock_readv,
2252 : .writev = posix_sock_writev,
2253 : .recv_next = posix_sock_recv_next,
2254 : .writev_async = posix_sock_writev_async,
2255 : .flush = posix_sock_flush,
2256 : .set_recvlowat = posix_sock_set_recvlowat,
2257 : .set_recvbuf = posix_sock_set_recvbuf,
2258 : .set_sendbuf = posix_sock_set_sendbuf,
2259 : .is_ipv6 = posix_sock_is_ipv6,
2260 : .is_ipv4 = posix_sock_is_ipv4,
2261 : .is_connected = posix_sock_is_connected,
2262 : .group_impl_get_optimal = posix_sock_group_impl_get_optimal,
2263 : .group_impl_create = posix_sock_group_impl_create,
2264 : .group_impl_add_sock = posix_sock_group_impl_add_sock,
2265 : .group_impl_remove_sock = posix_sock_group_impl_remove_sock,
2266 : .group_impl_poll = posix_sock_group_impl_poll,
2267 : .group_impl_register_interrupt = posix_sock_group_impl_register_interrupt,
2268 : .group_impl_unregister_interrupt = posix_sock_group_impl_unregister_interrupt,
2269 : .group_impl_close = posix_sock_group_impl_close,
2270 : .get_opts = posix_sock_impl_get_opts,
2271 : .set_opts = posix_sock_impl_set_opts,
2272 : };
2273 :
2274 2 : SPDK_NET_IMPL_REGISTER_DEFAULT(posix, &g_posix_net_impl);
2275 :
2276 : static struct spdk_sock *
2277 0 : ssl_sock_listen(const char *ip, int port, struct spdk_sock_opts *opts)
2278 : {
2279 0 : return posix_sock_create(ip, port, SPDK_SOCK_CREATE_LISTEN, opts, true);
2280 : }
2281 :
2282 : static struct spdk_sock *
2283 0 : ssl_sock_connect(const char *ip, int port, struct spdk_sock_opts *opts)
2284 : {
2285 0 : return posix_sock_create(ip, port, SPDK_SOCK_CREATE_CONNECT, opts, true);
2286 : }
2287 :
2288 : static struct spdk_sock *
2289 0 : ssl_sock_accept(struct spdk_sock *_sock)
2290 : {
2291 0 : return _posix_sock_accept(_sock, true);
2292 : }
2293 :
2294 : static struct spdk_net_impl g_ssl_net_impl = {
2295 : .name = "ssl",
2296 : .getaddr = posix_sock_getaddr,
2297 : .get_interface_name = posix_sock_get_interface_name,
2298 : .get_numa_socket_id = posix_sock_get_numa_socket_id,
2299 : .connect = ssl_sock_connect,
2300 : .listen = ssl_sock_listen,
2301 : .accept = ssl_sock_accept,
2302 : .close = posix_sock_close,
2303 : .recv = posix_sock_recv,
2304 : .readv = posix_sock_readv,
2305 : .writev = posix_sock_writev,
2306 : .recv_next = posix_sock_recv_next,
2307 : .writev_async = posix_sock_writev_async,
2308 : .flush = posix_sock_flush,
2309 : .set_recvlowat = posix_sock_set_recvlowat,
2310 : .set_recvbuf = posix_sock_set_recvbuf,
2311 : .set_sendbuf = posix_sock_set_sendbuf,
2312 : .is_ipv6 = posix_sock_is_ipv6,
2313 : .is_ipv4 = posix_sock_is_ipv4,
2314 : .is_connected = posix_sock_is_connected,
2315 : .group_impl_get_optimal = posix_sock_group_impl_get_optimal,
2316 : .group_impl_create = ssl_sock_group_impl_create,
2317 : .group_impl_add_sock = posix_sock_group_impl_add_sock,
2318 : .group_impl_remove_sock = posix_sock_group_impl_remove_sock,
2319 : .group_impl_poll = posix_sock_group_impl_poll,
2320 : .group_impl_register_interrupt = posix_sock_group_impl_register_interrupt,
2321 : .group_impl_unregister_interrupt = posix_sock_group_impl_unregister_interrupt,
2322 : .group_impl_close = ssl_sock_group_impl_close,
2323 : .get_opts = ssl_sock_impl_get_opts,
2324 : .set_opts = ssl_sock_impl_set_opts,
2325 : };
2326 :
2327 2 : SPDK_NET_IMPL_REGISTER(ssl, &g_ssl_net_impl);
2328 2 : SPDK_LOG_REGISTER_COMPONENT(sock_posix)
|