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 : #define SPDK_EPOLL
14 : #endif
15 :
16 : #if defined(__linux__)
17 : #include <linux/errqueue.h>
18 : #endif
19 :
20 : #include "spdk/env.h"
21 : #include "spdk/log.h"
22 : #include "spdk/pipe.h"
23 : #include "spdk/sock.h"
24 : #include "spdk/util.h"
25 : #include "spdk/string.h"
26 : #include "spdk/net.h"
27 : #include "spdk/file.h"
28 : #include "spdk_internal/sock.h"
29 : #include "spdk/net.h"
30 :
31 : #include "openssl/crypto.h"
32 : #include "openssl/err.h"
33 : #include "openssl/ssl.h"
34 :
35 : #define MAX_TMPBUF 1024
36 : #define PORTNUMLEN 32
37 :
38 : #if defined(SO_ZEROCOPY) && defined(MSG_ZEROCOPY)
39 : #define SPDK_ZEROCOPY
40 : #endif
41 :
42 : struct spdk_posix_sock {
43 : struct spdk_sock base;
44 : int fd;
45 :
46 : uint32_t sendmsg_idx;
47 :
48 : struct spdk_pipe *recv_pipe;
49 : int recv_buf_sz;
50 : bool pipe_has_data;
51 : bool socket_has_data;
52 : bool zcopy;
53 :
54 : int placement_id;
55 :
56 : SSL_CTX *ctx;
57 : SSL *ssl;
58 :
59 : TAILQ_ENTRY(spdk_posix_sock) link;
60 :
61 : char interface_name[IFNAMSIZ];
62 : };
63 :
64 : TAILQ_HEAD(spdk_has_data_list, spdk_posix_sock);
65 :
66 : struct spdk_posix_sock_group_impl {
67 : struct spdk_sock_group_impl base;
68 : int fd;
69 : struct spdk_interrupt *intr;
70 : struct spdk_has_data_list socks_with_data;
71 : int placement_id;
72 : struct spdk_pipe_group *pipe_group;
73 : };
74 :
75 : static struct spdk_sock_impl_opts g_posix_impl_opts = {
76 : .recv_buf_size = DEFAULT_SO_RCVBUF_SIZE,
77 : .send_buf_size = DEFAULT_SO_SNDBUF_SIZE,
78 : .enable_recv_pipe = true,
79 : .enable_quickack = false,
80 : .enable_placement_id = PLACEMENT_NONE,
81 : .enable_zerocopy_send_server = true,
82 : .enable_zerocopy_send_client = false,
83 : .zerocopy_threshold = 0,
84 : .tls_version = 0,
85 : .enable_ktls = false,
86 : .psk_key = NULL,
87 : .psk_key_size = 0,
88 : .psk_identity = NULL,
89 : .get_key = NULL,
90 : .get_key_ctx = NULL,
91 : .tls_cipher_suites = NULL
92 : };
93 :
94 : static struct spdk_sock_impl_opts g_ssl_impl_opts = {
95 : .recv_buf_size = MIN_SO_RCVBUF_SIZE,
96 : .send_buf_size = MIN_SO_SNDBUF_SIZE,
97 : .enable_recv_pipe = true,
98 : .enable_quickack = false,
99 : .enable_placement_id = PLACEMENT_NONE,
100 : .enable_zerocopy_send_server = true,
101 : .enable_zerocopy_send_client = false,
102 : .zerocopy_threshold = 0,
103 : .tls_version = 0,
104 : .enable_ktls = false,
105 : .psk_key = NULL,
106 : .psk_identity = NULL
107 : };
108 :
109 : static struct spdk_sock_map g_map = {
110 : .entries = STAILQ_HEAD_INITIALIZER(g_map.entries),
111 : .mtx = PTHREAD_MUTEX_INITIALIZER
112 : };
113 :
114 : __attribute((destructor)) static void
115 2 : posix_sock_map_cleanup(void)
116 : {
117 2 : spdk_sock_map_cleanup(&g_map);
118 2 : }
119 :
120 : #define __posix_sock(sock) (struct spdk_posix_sock *)sock
121 : #define __posix_group_impl(group) (struct spdk_posix_sock_group_impl *)group
122 :
123 : static void
124 20 : posix_sock_copy_impl_opts(struct spdk_sock_impl_opts *dest, const struct spdk_sock_impl_opts *src,
125 : size_t len)
126 : {
127 : #define FIELD_OK(field) \
128 : offsetof(struct spdk_sock_impl_opts, field) + sizeof(src->field) <= len
129 :
130 : #define SET_FIELD(field) \
131 : if (FIELD_OK(field)) { \
132 : dest->field = src->field; \
133 : }
134 :
135 20 : SET_FIELD(recv_buf_size);
136 20 : SET_FIELD(send_buf_size);
137 20 : SET_FIELD(enable_recv_pipe);
138 20 : SET_FIELD(enable_zerocopy_send);
139 20 : SET_FIELD(enable_quickack);
140 20 : SET_FIELD(enable_placement_id);
141 20 : SET_FIELD(enable_zerocopy_send_server);
142 20 : SET_FIELD(enable_zerocopy_send_client);
143 20 : SET_FIELD(zerocopy_threshold);
144 20 : SET_FIELD(tls_version);
145 20 : SET_FIELD(enable_ktls);
146 20 : SET_FIELD(psk_key);
147 20 : SET_FIELD(psk_key_size);
148 20 : SET_FIELD(psk_identity);
149 20 : SET_FIELD(get_key);
150 20 : SET_FIELD(get_key_ctx);
151 20 : SET_FIELD(tls_cipher_suites);
152 :
153 : #undef SET_FIELD
154 : #undef FIELD_OK
155 20 : }
156 :
157 : static int
158 11 : _sock_impl_get_opts(struct spdk_sock_impl_opts *opts, struct spdk_sock_impl_opts *impl_opts,
159 : size_t *len)
160 : {
161 11 : if (!opts || !len) {
162 0 : errno = EINVAL;
163 0 : return -1;
164 : }
165 :
166 11 : assert(sizeof(*opts) >= *len);
167 11 : memset(opts, 0, *len);
168 :
169 11 : posix_sock_copy_impl_opts(opts, impl_opts, *len);
170 11 : *len = spdk_min(*len, sizeof(*impl_opts));
171 :
172 11 : return 0;
173 : }
174 :
175 : static int
176 11 : posix_sock_impl_get_opts(struct spdk_sock_impl_opts *opts, size_t *len)
177 : {
178 11 : return _sock_impl_get_opts(opts, &g_posix_impl_opts, len);
179 : }
180 :
181 : static int
182 0 : ssl_sock_impl_get_opts(struct spdk_sock_impl_opts *opts, size_t *len)
183 : {
184 0 : return _sock_impl_get_opts(opts, &g_ssl_impl_opts, len);
185 : }
186 :
187 : static int
188 3 : _sock_impl_set_opts(const struct spdk_sock_impl_opts *opts, struct spdk_sock_impl_opts *impl_opts,
189 : size_t len)
190 : {
191 3 : if (!opts) {
192 0 : errno = EINVAL;
193 0 : return -1;
194 : }
195 :
196 3 : assert(sizeof(*opts) >= len);
197 3 : posix_sock_copy_impl_opts(impl_opts, opts, len);
198 :
199 3 : return 0;
200 : }
201 :
202 : static int
203 3 : posix_sock_impl_set_opts(const struct spdk_sock_impl_opts *opts, size_t len)
204 : {
205 3 : return _sock_impl_set_opts(opts, &g_posix_impl_opts, len);
206 : }
207 :
208 : static int
209 0 : ssl_sock_impl_set_opts(const struct spdk_sock_impl_opts *opts, size_t len)
210 : {
211 0 : return _sock_impl_set_opts(opts, &g_ssl_impl_opts, len);
212 : }
213 :
214 : static void
215 16 : _opts_get_impl_opts(const struct spdk_sock_opts *opts, struct spdk_sock_impl_opts *dest,
216 : const struct spdk_sock_impl_opts *default_impl)
217 : {
218 : /* Copy the default impl_opts first to cover cases when user's impl_opts is smaller */
219 16 : memcpy(dest, default_impl, sizeof(*dest));
220 :
221 16 : if (opts->impl_opts != NULL) {
222 6 : assert(sizeof(*dest) >= opts->impl_opts_size);
223 6 : posix_sock_copy_impl_opts(dest, opts->impl_opts, opts->impl_opts_size);
224 : }
225 16 : }
226 :
227 : static int
228 0 : posix_sock_getaddr(struct spdk_sock *_sock, char *saddr, int slen, uint16_t *sport,
229 : char *caddr, int clen, uint16_t *cport)
230 : {
231 0 : struct spdk_posix_sock *sock = __posix_sock(_sock);
232 :
233 0 : assert(sock != NULL);
234 0 : return spdk_net_getaddr(sock->fd, saddr, slen, sport, caddr, clen, cport);
235 : }
236 :
237 : static const char *
238 6 : posix_sock_get_interface_name(struct spdk_sock *_sock)
239 : {
240 6 : struct spdk_posix_sock *sock = __posix_sock(_sock);
241 6 : char saddr[64];
242 : int rc;
243 :
244 6 : rc = spdk_net_getaddr(sock->fd, saddr, sizeof(saddr), NULL, NULL, 0, NULL);
245 6 : if (rc != 0) {
246 0 : return NULL;
247 : }
248 :
249 6 : rc = spdk_net_get_interface_name(saddr, sock->interface_name,
250 : sizeof(sock->interface_name));
251 6 : if (rc != 0) {
252 0 : return NULL;
253 : }
254 :
255 6 : return sock->interface_name;
256 : }
257 :
258 : static int32_t
259 0 : posix_sock_get_numa_id(struct spdk_sock *sock)
260 : {
261 : const char *interface_name;
262 0 : uint32_t numa_id;
263 : int rc;
264 :
265 0 : interface_name = posix_sock_get_interface_name(sock);
266 0 : if (interface_name == NULL) {
267 0 : return SPDK_ENV_NUMA_ID_ANY;
268 : }
269 :
270 0 : rc = spdk_read_sysfs_attribute_uint32(&numa_id,
271 : "/sys/class/net/%s/device/numa_node", interface_name);
272 0 : if (rc == 0 && numa_id <= INT32_MAX) {
273 0 : return (int32_t)numa_id;
274 : } else {
275 0 : return SPDK_ENV_NUMA_ID_ANY;
276 : }
277 : }
278 :
279 : enum posix_sock_create_type {
280 : SPDK_SOCK_CREATE_LISTEN,
281 : SPDK_SOCK_CREATE_CONNECT,
282 : };
283 :
284 : static int
285 1 : posix_sock_alloc_pipe(struct spdk_posix_sock *sock, int sz)
286 : {
287 1 : uint8_t *new_buf, *old_buf;
288 : struct spdk_pipe *new_pipe;
289 1 : struct iovec siov[2];
290 1 : struct iovec diov[2];
291 : int sbytes;
292 : ssize_t bytes;
293 : int rc;
294 :
295 1 : if (sock->recv_buf_sz == sz) {
296 0 : return 0;
297 : }
298 :
299 : /* If the new size is 0, just free the pipe */
300 1 : if (sz == 0) {
301 0 : old_buf = spdk_pipe_destroy(sock->recv_pipe);
302 0 : free(old_buf);
303 0 : sock->recv_pipe = NULL;
304 0 : return 0;
305 1 : } else if (sz < MIN_SOCK_PIPE_SIZE) {
306 0 : SPDK_ERRLOG("The size of the pipe must be larger than %d\n", MIN_SOCK_PIPE_SIZE);
307 0 : return -1;
308 : }
309 :
310 : /* Round up to next 64 byte multiple */
311 1 : rc = posix_memalign((void **)&new_buf, 64, sz);
312 1 : if (rc != 0) {
313 0 : SPDK_ERRLOG("socket recv buf allocation failed\n");
314 0 : return -ENOMEM;
315 : }
316 1 : memset(new_buf, 0, sz);
317 :
318 1 : new_pipe = spdk_pipe_create(new_buf, sz);
319 1 : if (new_pipe == NULL) {
320 0 : SPDK_ERRLOG("socket pipe allocation failed\n");
321 0 : free(new_buf);
322 0 : return -ENOMEM;
323 : }
324 :
325 1 : if (sock->recv_pipe != NULL) {
326 : /* Pull all of the data out of the old pipe */
327 0 : sbytes = spdk_pipe_reader_get_buffer(sock->recv_pipe, sock->recv_buf_sz, siov);
328 0 : if (sbytes > sz) {
329 : /* Too much data to fit into the new pipe size */
330 0 : old_buf = spdk_pipe_destroy(new_pipe);
331 0 : free(old_buf);
332 0 : return -EINVAL;
333 : }
334 :
335 0 : sbytes = spdk_pipe_writer_get_buffer(new_pipe, sz, diov);
336 0 : assert(sbytes == sz);
337 :
338 0 : bytes = spdk_iovcpy(siov, 2, diov, 2);
339 0 : spdk_pipe_writer_advance(new_pipe, bytes);
340 :
341 0 : old_buf = spdk_pipe_destroy(sock->recv_pipe);
342 0 : free(old_buf);
343 : }
344 :
345 1 : sock->recv_buf_sz = sz;
346 1 : sock->recv_pipe = new_pipe;
347 :
348 1 : if (sock->base.group_impl) {
349 : struct spdk_posix_sock_group_impl *group;
350 :
351 0 : group = __posix_group_impl(sock->base.group_impl);
352 0 : spdk_pipe_group_add(group->pipe_group, sock->recv_pipe);
353 : }
354 :
355 1 : return 0;
356 : }
357 :
358 : static int
359 1 : posix_sock_set_recvbuf(struct spdk_sock *_sock, int sz)
360 : {
361 1 : struct spdk_posix_sock *sock = __posix_sock(_sock);
362 : int min_size;
363 : int rc;
364 :
365 1 : assert(sock != NULL);
366 :
367 1 : if (_sock->impl_opts.enable_recv_pipe) {
368 1 : rc = posix_sock_alloc_pipe(sock, sz);
369 1 : if (rc) {
370 0 : return rc;
371 : }
372 : }
373 :
374 : /* Set kernel buffer size to be at least MIN_SO_RCVBUF_SIZE and
375 : * _sock->impl_opts.recv_buf_size. */
376 1 : min_size = spdk_max(MIN_SO_RCVBUF_SIZE, _sock->impl_opts.recv_buf_size);
377 :
378 1 : if (sz < min_size) {
379 1 : sz = min_size;
380 : }
381 :
382 1 : rc = setsockopt(sock->fd, SOL_SOCKET, SO_RCVBUF, &sz, sizeof(sz));
383 1 : if (rc < 0) {
384 0 : return rc;
385 : }
386 :
387 1 : _sock->impl_opts.recv_buf_size = sz;
388 :
389 1 : return 0;
390 : }
391 :
392 : static int
393 1 : posix_sock_set_sendbuf(struct spdk_sock *_sock, int sz)
394 : {
395 1 : struct spdk_posix_sock *sock = __posix_sock(_sock);
396 : int min_size;
397 : int rc;
398 :
399 1 : assert(sock != NULL);
400 :
401 : /* Set kernel buffer size to be at least MIN_SO_SNDBUF_SIZE and
402 : * _sock->impl_opts.send_buf_size. */
403 1 : min_size = spdk_max(MIN_SO_SNDBUF_SIZE, _sock->impl_opts.send_buf_size);
404 :
405 1 : if (sz < min_size) {
406 1 : sz = min_size;
407 : }
408 :
409 1 : rc = setsockopt(sock->fd, SOL_SOCKET, SO_SNDBUF, &sz, sizeof(sz));
410 1 : if (rc < 0) {
411 0 : return rc;
412 : }
413 :
414 1 : _sock->impl_opts.send_buf_size = sz;
415 :
416 1 : return 0;
417 : }
418 :
419 : static void
420 24 : posix_sock_init(struct spdk_posix_sock *sock, bool enable_zero_copy)
421 : {
422 : #if defined(SPDK_ZEROCOPY) || defined(__linux__)
423 24 : int flag;
424 : int rc;
425 : #endif
426 :
427 : #if defined(SPDK_ZEROCOPY)
428 24 : flag = 1;
429 :
430 24 : if (enable_zero_copy) {
431 : /* Try to turn on zero copy sends */
432 0 : rc = setsockopt(sock->fd, SOL_SOCKET, SO_ZEROCOPY, &flag, sizeof(flag));
433 0 : if (rc == 0) {
434 0 : sock->zcopy = true;
435 : }
436 : }
437 : #endif
438 :
439 : #if defined(__linux__)
440 24 : flag = 1;
441 :
442 24 : if (sock->base.impl_opts.enable_quickack) {
443 0 : rc = setsockopt(sock->fd, IPPROTO_TCP, TCP_QUICKACK, &flag, sizeof(flag));
444 0 : if (rc != 0) {
445 0 : SPDK_ERRLOG("quickack was failed to set\n");
446 : }
447 : }
448 :
449 24 : spdk_sock_get_placement_id(sock->fd, sock->base.impl_opts.enable_placement_id,
450 : &sock->placement_id);
451 :
452 24 : if (sock->base.impl_opts.enable_placement_id == PLACEMENT_MARK) {
453 : /* Save placement_id */
454 0 : spdk_sock_map_insert(&g_map, sock->placement_id, NULL);
455 : }
456 : #endif
457 24 : }
458 :
459 : static struct spdk_posix_sock *
460 24 : posix_sock_alloc(int fd, struct spdk_sock_impl_opts *impl_opts, bool enable_zero_copy)
461 : {
462 : struct spdk_posix_sock *sock;
463 :
464 24 : sock = calloc(1, sizeof(*sock));
465 24 : if (sock == NULL) {
466 0 : SPDK_ERRLOG("sock allocation failed\n");
467 0 : return NULL;
468 : }
469 :
470 24 : sock->fd = fd;
471 24 : memcpy(&sock->base.impl_opts, impl_opts, sizeof(*impl_opts));
472 24 : posix_sock_init(sock, enable_zero_copy);
473 :
474 24 : return sock;
475 : }
476 :
477 : static int
478 16 : posix_fd_create(struct addrinfo *res, struct spdk_sock_opts *opts,
479 : struct spdk_sock_impl_opts *impl_opts)
480 : {
481 : int fd;
482 16 : int val = 1;
483 16 : int rc, sz;
484 : #if defined(__linux__)
485 16 : int to;
486 : #endif
487 :
488 16 : fd = socket(res->ai_family, res->ai_socktype, res->ai_protocol);
489 16 : if (fd < 0) {
490 : /* error */
491 0 : return -1;
492 : }
493 :
494 16 : sz = impl_opts->recv_buf_size;
495 16 : rc = setsockopt(fd, SOL_SOCKET, SO_RCVBUF, &sz, sizeof(sz));
496 : if (rc) {
497 : /* Not fatal */
498 : }
499 :
500 16 : sz = impl_opts->send_buf_size;
501 16 : rc = setsockopt(fd, SOL_SOCKET, SO_SNDBUF, &sz, sizeof(sz));
502 : if (rc) {
503 : /* Not fatal */
504 : }
505 :
506 16 : rc = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &val, sizeof val);
507 16 : if (rc != 0) {
508 0 : close(fd);
509 : /* error */
510 0 : return -1;
511 : }
512 16 : rc = setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, &val, sizeof val);
513 16 : if (rc != 0) {
514 0 : close(fd);
515 : /* error */
516 0 : return -1;
517 : }
518 :
519 : #if defined(SO_PRIORITY)
520 16 : if (opts->priority) {
521 0 : rc = setsockopt(fd, SOL_SOCKET, SO_PRIORITY, &opts->priority, sizeof val);
522 0 : if (rc != 0) {
523 0 : close(fd);
524 : /* error */
525 0 : return -1;
526 : }
527 : }
528 : #endif
529 :
530 16 : if (res->ai_family == AF_INET6) {
531 0 : rc = setsockopt(fd, IPPROTO_IPV6, IPV6_V6ONLY, &val, sizeof val);
532 0 : if (rc != 0) {
533 0 : close(fd);
534 : /* error */
535 0 : return -1;
536 : }
537 : }
538 :
539 16 : if (opts->ack_timeout) {
540 : #if defined(__linux__)
541 0 : to = opts->ack_timeout;
542 0 : rc = setsockopt(fd, IPPROTO_TCP, TCP_USER_TIMEOUT, &to, sizeof(to));
543 0 : if (rc != 0) {
544 0 : close(fd);
545 : /* error */
546 0 : return -1;
547 : }
548 : #else
549 : SPDK_WARNLOG("TCP_USER_TIMEOUT is not supported.\n");
550 : #endif
551 : }
552 :
553 16 : return fd;
554 : }
555 :
556 : static int
557 0 : posix_sock_psk_find_session_server_cb(SSL *ssl, const unsigned char *identity,
558 : size_t identity_len, SSL_SESSION **sess)
559 : {
560 0 : struct spdk_sock_impl_opts *impl_opts = SSL_get_app_data(ssl);
561 0 : uint8_t key[SSL_MAX_MASTER_KEY_LENGTH] = {};
562 : int keylen;
563 : int rc, i;
564 : STACK_OF(SSL_CIPHER) *ciphers;
565 : const SSL_CIPHER *cipher;
566 : const char *cipher_name;
567 0 : const char *user_cipher = NULL;
568 0 : bool found = false;
569 :
570 0 : if (impl_opts->get_key) {
571 0 : rc = impl_opts->get_key(key, sizeof(key), &user_cipher, identity, impl_opts->get_key_ctx);
572 0 : if (rc < 0) {
573 0 : SPDK_ERRLOG("Unable to find PSK for identity: %s\n", identity);
574 0 : return 0;
575 : }
576 0 : keylen = rc;
577 : } else {
578 0 : if (impl_opts->psk_key == NULL) {
579 0 : SPDK_ERRLOG("PSK is not set\n");
580 0 : return 0;
581 : }
582 :
583 0 : SPDK_DEBUGLOG(sock_posix, "Length of Client's PSK ID %lu\n", strlen(impl_opts->psk_identity));
584 0 : if (strcmp(impl_opts->psk_identity, identity) != 0) {
585 0 : SPDK_ERRLOG("Unknown Client's PSK ID\n");
586 0 : return 0;
587 : }
588 0 : keylen = impl_opts->psk_key_size;
589 :
590 0 : memcpy(key, impl_opts->psk_key, keylen);
591 0 : user_cipher = impl_opts->tls_cipher_suites;
592 : }
593 :
594 0 : if (user_cipher == NULL) {
595 0 : SPDK_ERRLOG("Cipher suite not set\n");
596 0 : return 0;
597 : }
598 :
599 0 : *sess = SSL_SESSION_new();
600 0 : if (*sess == NULL) {
601 0 : SPDK_ERRLOG("Unable to allocate new SSL session\n");
602 0 : return 0;
603 : }
604 :
605 0 : ciphers = SSL_get_ciphers(ssl);
606 0 : for (i = 0; i < sk_SSL_CIPHER_num(ciphers); i++) {
607 0 : cipher = sk_SSL_CIPHER_value(ciphers, i);
608 0 : cipher_name = SSL_CIPHER_get_name(cipher);
609 :
610 0 : if (strcmp(user_cipher, cipher_name) == 0) {
611 0 : rc = SSL_SESSION_set_cipher(*sess, cipher);
612 0 : if (rc != 1) {
613 0 : SPDK_ERRLOG("Unable to set cipher: %s\n", cipher_name);
614 0 : goto err;
615 : }
616 0 : found = true;
617 0 : break;
618 : }
619 : }
620 0 : if (found == false) {
621 0 : SPDK_ERRLOG("No suitable cipher found\n");
622 0 : goto err;
623 : }
624 :
625 0 : SPDK_DEBUGLOG(sock_posix, "Cipher selected: %s\n", cipher_name);
626 :
627 0 : rc = SSL_SESSION_set_protocol_version(*sess, TLS1_3_VERSION);
628 0 : if (rc != 1) {
629 0 : SPDK_ERRLOG("Unable to set TLS version: %d\n", TLS1_3_VERSION);
630 0 : goto err;
631 : }
632 :
633 0 : rc = SSL_SESSION_set1_master_key(*sess, key, keylen);
634 0 : if (rc != 1) {
635 0 : SPDK_ERRLOG("Unable to set PSK for session\n");
636 0 : goto err;
637 : }
638 :
639 0 : return 1;
640 :
641 0 : err:
642 0 : SSL_SESSION_free(*sess);
643 0 : *sess = NULL;
644 0 : return 0;
645 : }
646 :
647 : static int
648 0 : posix_sock_psk_use_session_client_cb(SSL *ssl, const EVP_MD *md, const unsigned char **identity,
649 : size_t *identity_len, SSL_SESSION **sess)
650 : {
651 0 : struct spdk_sock_impl_opts *impl_opts = SSL_get_app_data(ssl);
652 : int rc, i;
653 : STACK_OF(SSL_CIPHER) *ciphers;
654 : const SSL_CIPHER *cipher;
655 : const char *cipher_name;
656 : long keylen;
657 0 : bool found = false;
658 :
659 0 : if (impl_opts->psk_key == NULL) {
660 0 : SPDK_ERRLOG("PSK is not set\n");
661 0 : return 0;
662 : }
663 0 : if (impl_opts->psk_key_size > SSL_MAX_MASTER_KEY_LENGTH) {
664 0 : SPDK_ERRLOG("PSK too long\n");
665 0 : return 0;
666 : }
667 0 : keylen = impl_opts->psk_key_size;
668 :
669 0 : if (impl_opts->tls_cipher_suites == NULL) {
670 0 : SPDK_ERRLOG("Cipher suite not set\n");
671 0 : return 0;
672 : }
673 0 : *sess = SSL_SESSION_new();
674 0 : if (*sess == NULL) {
675 0 : SPDK_ERRLOG("Unable to allocate new SSL session\n");
676 0 : return 0;
677 : }
678 :
679 0 : ciphers = SSL_get_ciphers(ssl);
680 0 : for (i = 0; i < sk_SSL_CIPHER_num(ciphers); i++) {
681 0 : cipher = sk_SSL_CIPHER_value(ciphers, i);
682 0 : cipher_name = SSL_CIPHER_get_name(cipher);
683 :
684 0 : if (strcmp(impl_opts->tls_cipher_suites, cipher_name) == 0) {
685 0 : rc = SSL_SESSION_set_cipher(*sess, cipher);
686 0 : if (rc != 1) {
687 0 : SPDK_ERRLOG("Unable to set cipher: %s\n", cipher_name);
688 0 : goto err;
689 : }
690 0 : found = true;
691 0 : break;
692 : }
693 : }
694 0 : if (found == false) {
695 0 : SPDK_ERRLOG("No suitable cipher found\n");
696 0 : goto err;
697 : }
698 :
699 0 : SPDK_DEBUGLOG(sock_posix, "Cipher selected: %s\n", cipher_name);
700 :
701 0 : rc = SSL_SESSION_set_protocol_version(*sess, TLS1_3_VERSION);
702 0 : if (rc != 1) {
703 0 : SPDK_ERRLOG("Unable to set TLS version: %d\n", TLS1_3_VERSION);
704 0 : goto err;
705 : }
706 :
707 0 : rc = SSL_SESSION_set1_master_key(*sess, impl_opts->psk_key, keylen);
708 0 : if (rc != 1) {
709 0 : SPDK_ERRLOG("Unable to set PSK for session\n");
710 0 : goto err;
711 : }
712 :
713 0 : *identity_len = strlen(impl_opts->psk_identity);
714 0 : *identity = impl_opts->psk_identity;
715 :
716 0 : return 1;
717 :
718 0 : err:
719 0 : SSL_SESSION_free(*sess);
720 0 : *sess = NULL;
721 0 : return 0;
722 : }
723 :
724 : static SSL_CTX *
725 0 : posix_sock_create_ssl_context(const SSL_METHOD *method, struct spdk_sock_opts *opts,
726 : struct spdk_sock_impl_opts *impl_opts)
727 : {
728 : SSL_CTX *ctx;
729 0 : int tls_version = 0;
730 0 : bool ktls_enabled = false;
731 : #ifdef SSL_OP_ENABLE_KTLS
732 : long options;
733 : #endif
734 :
735 0 : SSL_library_init();
736 0 : OpenSSL_add_all_algorithms();
737 0 : SSL_load_error_strings();
738 : /* Produce a SSL CTX in SSL V2 and V3 standards compliant way */
739 0 : ctx = SSL_CTX_new(method);
740 0 : if (!ctx) {
741 0 : SPDK_ERRLOG("SSL_CTX_new() failed, msg = %s\n", ERR_error_string(ERR_peek_last_error(), NULL));
742 0 : return NULL;
743 : }
744 0 : SPDK_DEBUGLOG(sock_posix, "SSL context created\n");
745 :
746 0 : switch (impl_opts->tls_version) {
747 0 : case 0:
748 : /* auto-negotiation */
749 0 : break;
750 0 : case SPDK_TLS_VERSION_1_3:
751 0 : tls_version = TLS1_3_VERSION;
752 0 : break;
753 0 : default:
754 0 : SPDK_ERRLOG("Incorrect TLS version provided: %d\n", impl_opts->tls_version);
755 0 : goto err;
756 : }
757 :
758 0 : if (tls_version) {
759 0 : SPDK_DEBUGLOG(sock_posix, "Hardening TLS version to '%d'='0x%X'\n", impl_opts->tls_version,
760 : tls_version);
761 0 : if (!SSL_CTX_set_min_proto_version(ctx, tls_version)) {
762 0 : SPDK_ERRLOG("Unable to set Min TLS version to '%d'='0x%X\n", impl_opts->tls_version, tls_version);
763 0 : goto err;
764 : }
765 0 : if (!SSL_CTX_set_max_proto_version(ctx, tls_version)) {
766 0 : SPDK_ERRLOG("Unable to set Max TLS version to '%d'='0x%X\n", impl_opts->tls_version, tls_version);
767 0 : goto err;
768 : }
769 : }
770 0 : if (impl_opts->enable_ktls) {
771 0 : SPDK_DEBUGLOG(sock_posix, "Enabling kTLS offload\n");
772 : #ifdef SSL_OP_ENABLE_KTLS
773 0 : options = SSL_CTX_set_options(ctx, SSL_OP_ENABLE_KTLS);
774 0 : ktls_enabled = options & SSL_OP_ENABLE_KTLS;
775 : #else
776 : ktls_enabled = false;
777 : #endif
778 0 : if (!ktls_enabled) {
779 0 : SPDK_ERRLOG("Unable to set kTLS offload via SSL_CTX_set_options(). Configure openssl with 'enable-ktls'\n");
780 0 : goto err;
781 : }
782 : }
783 :
784 : /* SSL_CTX_set_ciphersuites() return 1 if the requested
785 : * cipher suite list was configured, and 0 otherwise. */
786 0 : if (impl_opts->tls_cipher_suites != NULL &&
787 0 : SSL_CTX_set_ciphersuites(ctx, impl_opts->tls_cipher_suites) != 1) {
788 0 : SPDK_ERRLOG("Unable to set TLS cipher suites for SSL'\n");
789 0 : goto err;
790 : }
791 :
792 0 : return ctx;
793 :
794 0 : err:
795 0 : SSL_CTX_free(ctx);
796 0 : return NULL;
797 : }
798 :
799 : static SSL *
800 0 : ssl_sock_setup_connect(SSL_CTX *ctx, int fd)
801 : {
802 : SSL *ssl;
803 :
804 0 : ssl = SSL_new(ctx);
805 0 : if (!ssl) {
806 0 : SPDK_ERRLOG("SSL_new() failed, msg = %s\n", ERR_error_string(ERR_peek_last_error(), NULL));
807 0 : return NULL;
808 : }
809 0 : SSL_set_fd(ssl, fd);
810 0 : SSL_set_connect_state(ssl);
811 0 : SSL_set_psk_use_session_callback(ssl, posix_sock_psk_use_session_client_cb);
812 0 : SPDK_DEBUGLOG(sock_posix, "SSL object creation finished: %p\n", ssl);
813 0 : SPDK_DEBUGLOG(sock_posix, "%s = SSL_state_string_long(%p)\n", SSL_state_string_long(ssl), 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, "Negotiated Cipher suite:%s\n",
816 : SSL_CIPHER_get_name(SSL_get_current_cipher(ssl)));
817 0 : return ssl;
818 : }
819 :
820 : static SSL *
821 0 : ssl_sock_setup_accept(SSL_CTX *ctx, int fd)
822 : {
823 : SSL *ssl;
824 :
825 0 : ssl = SSL_new(ctx);
826 0 : if (!ssl) {
827 0 : SPDK_ERRLOG("SSL_new() failed, msg = %s\n", ERR_error_string(ERR_peek_last_error(), NULL));
828 0 : return NULL;
829 : }
830 0 : SSL_set_fd(ssl, fd);
831 0 : SSL_set_accept_state(ssl);
832 0 : SSL_set_psk_find_session_callback(ssl, posix_sock_psk_find_session_server_cb);
833 0 : SPDK_DEBUGLOG(sock_posix, "SSL object creation finished: %p\n", ssl);
834 0 : SPDK_DEBUGLOG(sock_posix, "%s = SSL_state_string_long(%p)\n", SSL_state_string_long(ssl), 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, "Negotiated Cipher suite:%s\n",
837 : SSL_CIPHER_get_name(SSL_get_current_cipher(ssl)));
838 0 : return ssl;
839 : }
840 :
841 : static ssize_t
842 0 : SSL_readv(SSL *ssl, const struct iovec *iov, int iovcnt)
843 : {
844 0 : int i, rc = 0;
845 0 : ssize_t total = 0;
846 :
847 0 : for (i = 0; i < iovcnt; i++) {
848 0 : rc = SSL_read(ssl, iov[i].iov_base, iov[i].iov_len);
849 :
850 0 : if (rc > 0) {
851 0 : total += rc;
852 : }
853 0 : if (rc != (int)iov[i].iov_len) {
854 0 : break;
855 : }
856 : }
857 0 : if (total > 0) {
858 0 : errno = 0;
859 0 : return total;
860 : }
861 0 : switch (SSL_get_error(ssl, rc)) {
862 0 : case SSL_ERROR_ZERO_RETURN:
863 0 : errno = ENOTCONN;
864 0 : return 0;
865 0 : case SSL_ERROR_WANT_READ:
866 : case SSL_ERROR_WANT_WRITE:
867 : case SSL_ERROR_WANT_CONNECT:
868 : case SSL_ERROR_WANT_ACCEPT:
869 : case SSL_ERROR_WANT_X509_LOOKUP:
870 : case SSL_ERROR_WANT_ASYNC:
871 : case SSL_ERROR_WANT_ASYNC_JOB:
872 : case SSL_ERROR_WANT_CLIENT_HELLO_CB:
873 0 : errno = EAGAIN;
874 0 : return -1;
875 0 : case SSL_ERROR_SYSCALL:
876 : case SSL_ERROR_SSL:
877 0 : errno = ENOTCONN;
878 0 : return -1;
879 0 : default:
880 0 : errno = ENOTCONN;
881 0 : return -1;
882 : }
883 : }
884 :
885 : static ssize_t
886 0 : SSL_writev(SSL *ssl, struct iovec *iov, int iovcnt)
887 : {
888 0 : int i, rc = 0;
889 0 : ssize_t total = 0;
890 :
891 0 : for (i = 0; i < iovcnt; i++) {
892 0 : rc = SSL_write(ssl, iov[i].iov_base, iov[i].iov_len);
893 :
894 0 : if (rc > 0) {
895 0 : total += rc;
896 : }
897 0 : if (rc != (int)iov[i].iov_len) {
898 0 : break;
899 : }
900 : }
901 0 : if (total > 0) {
902 0 : errno = 0;
903 0 : return total;
904 : }
905 0 : switch (SSL_get_error(ssl, rc)) {
906 0 : case SSL_ERROR_ZERO_RETURN:
907 0 : errno = ENOTCONN;
908 0 : return 0;
909 0 : case SSL_ERROR_WANT_READ:
910 : case SSL_ERROR_WANT_WRITE:
911 : case SSL_ERROR_WANT_CONNECT:
912 : case SSL_ERROR_WANT_ACCEPT:
913 : case SSL_ERROR_WANT_X509_LOOKUP:
914 : case SSL_ERROR_WANT_ASYNC:
915 : case SSL_ERROR_WANT_ASYNC_JOB:
916 : case SSL_ERROR_WANT_CLIENT_HELLO_CB:
917 0 : errno = EAGAIN;
918 0 : return -1;
919 0 : case SSL_ERROR_SYSCALL:
920 : case SSL_ERROR_SSL:
921 0 : errno = ENOTCONN;
922 0 : return -1;
923 0 : default:
924 0 : errno = ENOTCONN;
925 0 : return -1;
926 : }
927 : }
928 :
929 : static struct spdk_sock *
930 16 : posix_sock_create(const char *ip, int port,
931 : enum posix_sock_create_type type,
932 : struct spdk_sock_opts *opts,
933 : bool enable_ssl)
934 : {
935 : struct spdk_posix_sock *sock;
936 16 : struct spdk_sock_impl_opts impl_opts;
937 16 : char buf[MAX_TMPBUF];
938 16 : char portnum[PORTNUMLEN];
939 : char *p;
940 : const char *src_addr;
941 : uint16_t src_port;
942 16 : struct addrinfo hints, *res, *res0, *src_ai;
943 : int fd, flag;
944 : int rc;
945 16 : bool enable_zcopy_user_opts = true;
946 16 : bool enable_zcopy_impl_opts = true;
947 16 : SSL_CTX *ctx = 0;
948 16 : SSL *ssl = 0;
949 :
950 16 : assert(opts != NULL);
951 16 : if (enable_ssl) {
952 0 : _opts_get_impl_opts(opts, &impl_opts, &g_ssl_impl_opts);
953 : } else {
954 16 : _opts_get_impl_opts(opts, &impl_opts, &g_posix_impl_opts);
955 : }
956 :
957 16 : if (ip == NULL) {
958 0 : return NULL;
959 : }
960 16 : if (ip[0] == '[') {
961 0 : snprintf(buf, sizeof(buf), "%s", ip + 1);
962 0 : p = strchr(buf, ']');
963 0 : if (p != NULL) {
964 0 : *p = '\0';
965 : }
966 0 : ip = (const char *) &buf[0];
967 : }
968 :
969 16 : snprintf(portnum, sizeof portnum, "%d", port);
970 16 : memset(&hints, 0, sizeof hints);
971 16 : hints.ai_family = PF_UNSPEC;
972 16 : hints.ai_socktype = SOCK_STREAM;
973 16 : hints.ai_flags = AI_NUMERICSERV;
974 16 : hints.ai_flags |= AI_PASSIVE;
975 16 : hints.ai_flags |= AI_NUMERICHOST;
976 16 : rc = getaddrinfo(ip, portnum, &hints, &res0);
977 16 : if (rc != 0) {
978 0 : SPDK_ERRLOG("getaddrinfo() failed %s (%d)\n", gai_strerror(rc), rc);
979 0 : return NULL;
980 : }
981 :
982 : /* try listen */
983 16 : fd = -1;
984 16 : for (res = res0; res != NULL; res = res->ai_next) {
985 16 : retry:
986 16 : fd = posix_fd_create(res, opts, &impl_opts);
987 16 : if (fd < 0) {
988 0 : continue;
989 : }
990 16 : if (type == SPDK_SOCK_CREATE_LISTEN) {
991 7 : rc = bind(fd, res->ai_addr, res->ai_addrlen);
992 7 : if (rc != 0) {
993 0 : SPDK_ERRLOG("bind() failed at port %d, errno = %d\n", port, errno);
994 0 : switch (errno) {
995 0 : case EINTR:
996 : /* interrupted? */
997 0 : close(fd);
998 0 : goto retry;
999 0 : case EADDRNOTAVAIL:
1000 0 : SPDK_ERRLOG("IP address %s not available. "
1001 : "Verify IP address in config file "
1002 : "and make sure setup script is "
1003 : "run before starting spdk app.\n", ip);
1004 : /* FALLTHROUGH */
1005 0 : default:
1006 : /* try next family */
1007 0 : close(fd);
1008 0 : fd = -1;
1009 0 : continue;
1010 : }
1011 : }
1012 : /* bind OK */
1013 7 : rc = listen(fd, 512);
1014 7 : if (rc != 0) {
1015 0 : SPDK_ERRLOG("listen() failed, errno = %d\n", errno);
1016 0 : close(fd);
1017 0 : fd = -1;
1018 0 : break;
1019 : }
1020 7 : enable_zcopy_impl_opts = impl_opts.enable_zerocopy_send_server;
1021 9 : } else if (type == SPDK_SOCK_CREATE_CONNECT) {
1022 9 : src_addr = SPDK_GET_FIELD(opts, src_addr, NULL, opts->opts_size);
1023 9 : src_port = SPDK_GET_FIELD(opts, src_port, 0, opts->opts_size);
1024 9 : if (src_addr != NULL || src_port != 0) {
1025 0 : snprintf(portnum, sizeof(portnum), "%"PRIu16, src_port);
1026 0 : memset(&hints, 0, sizeof hints);
1027 0 : hints.ai_family = AF_UNSPEC;
1028 0 : hints.ai_socktype = SOCK_STREAM;
1029 0 : hints.ai_flags = AI_NUMERICSERV | AI_NUMERICHOST | AI_PASSIVE;
1030 0 : rc = getaddrinfo(src_addr, src_port > 0 ? portnum : NULL,
1031 : &hints, &src_ai);
1032 0 : if (rc != 0 || src_ai == NULL) {
1033 0 : SPDK_ERRLOG("getaddrinfo() failed %s (%d)\n",
1034 : rc != 0 ? gai_strerror(rc) : "", rc);
1035 0 : close(fd);
1036 0 : fd = -1;
1037 0 : break;
1038 : }
1039 0 : rc = bind(fd, src_ai->ai_addr, src_ai->ai_addrlen);
1040 0 : if (rc != 0) {
1041 0 : SPDK_ERRLOG("bind() failed errno %d (%s:%s)\n", errno,
1042 : src_addr ? src_addr : "", portnum);
1043 0 : close(fd);
1044 0 : fd = -1;
1045 0 : freeaddrinfo(src_ai);
1046 0 : src_ai = NULL;
1047 0 : break;
1048 : }
1049 0 : freeaddrinfo(src_ai);
1050 0 : src_ai = NULL;
1051 : }
1052 9 : rc = connect(fd, res->ai_addr, res->ai_addrlen);
1053 9 : if (rc != 0) {
1054 0 : SPDK_ERRLOG("connect() failed, errno = %d\n", errno);
1055 : /* try next family */
1056 0 : close(fd);
1057 0 : fd = -1;
1058 0 : continue;
1059 : }
1060 9 : enable_zcopy_impl_opts = impl_opts.enable_zerocopy_send_client;
1061 9 : if (enable_ssl) {
1062 0 : ctx = posix_sock_create_ssl_context(TLS_client_method(), opts, &impl_opts);
1063 0 : if (!ctx) {
1064 0 : SPDK_ERRLOG("posix_sock_create_ssl_context() failed, errno = %d\n", errno);
1065 0 : close(fd);
1066 0 : fd = -1;
1067 0 : break;
1068 : }
1069 0 : ssl = ssl_sock_setup_connect(ctx, fd);
1070 0 : if (!ssl) {
1071 0 : SPDK_ERRLOG("ssl_sock_setup_connect() failed, errno = %d\n", errno);
1072 0 : close(fd);
1073 0 : fd = -1;
1074 0 : SSL_CTX_free(ctx);
1075 0 : break;
1076 : }
1077 : }
1078 : }
1079 :
1080 16 : flag = fcntl(fd, F_GETFL);
1081 16 : if (fcntl(fd, F_SETFL, flag | O_NONBLOCK) < 0) {
1082 0 : SPDK_ERRLOG("fcntl can't set nonblocking mode for socket, fd: %d (%d)\n", fd, errno);
1083 0 : SSL_free(ssl);
1084 0 : SSL_CTX_free(ctx);
1085 0 : close(fd);
1086 0 : fd = -1;
1087 0 : break;
1088 : }
1089 16 : break;
1090 : }
1091 16 : freeaddrinfo(res0);
1092 :
1093 16 : if (fd < 0) {
1094 0 : return NULL;
1095 : }
1096 :
1097 : /* Only enable zero copy for non-loopback and non-ssl sockets. */
1098 16 : enable_zcopy_user_opts = opts->zcopy && !spdk_net_is_loopback(fd) && !enable_ssl;
1099 :
1100 16 : sock = posix_sock_alloc(fd, &impl_opts, enable_zcopy_user_opts && enable_zcopy_impl_opts);
1101 16 : if (sock == NULL) {
1102 0 : SPDK_ERRLOG("sock allocation failed\n");
1103 0 : SSL_free(ssl);
1104 0 : SSL_CTX_free(ctx);
1105 0 : close(fd);
1106 0 : return NULL;
1107 : }
1108 :
1109 16 : if (ctx) {
1110 0 : sock->ctx = ctx;
1111 : }
1112 :
1113 16 : if (ssl) {
1114 0 : sock->ssl = ssl;
1115 0 : SSL_set_app_data(ssl, &sock->base.impl_opts);
1116 : }
1117 :
1118 16 : return &sock->base;
1119 : }
1120 :
1121 : static struct spdk_sock *
1122 7 : posix_sock_listen(const char *ip, int port, struct spdk_sock_opts *opts)
1123 : {
1124 7 : return posix_sock_create(ip, port, SPDK_SOCK_CREATE_LISTEN, opts, false);
1125 : }
1126 :
1127 : static struct spdk_sock *
1128 9 : posix_sock_connect(const char *ip, int port, struct spdk_sock_opts *opts)
1129 : {
1130 9 : return posix_sock_create(ip, port, SPDK_SOCK_CREATE_CONNECT, opts, false);
1131 : }
1132 :
1133 : static struct spdk_sock *
1134 10 : _posix_sock_accept(struct spdk_sock *_sock, bool enable_ssl)
1135 : {
1136 10 : struct spdk_posix_sock *sock = __posix_sock(_sock);
1137 10 : struct spdk_posix_sock_group_impl *group = __posix_group_impl(sock->base.group_impl);
1138 10 : struct sockaddr_storage sa;
1139 10 : socklen_t salen;
1140 : int rc, fd;
1141 : struct spdk_posix_sock *new_sock;
1142 : int flag;
1143 10 : SSL_CTX *ctx = 0;
1144 10 : SSL *ssl = 0;
1145 :
1146 10 : memset(&sa, 0, sizeof(sa));
1147 10 : salen = sizeof(sa);
1148 :
1149 10 : assert(sock != NULL);
1150 :
1151 : /* epoll_wait will trigger again if there is more than one request */
1152 10 : if (group && sock->socket_has_data) {
1153 0 : sock->socket_has_data = false;
1154 0 : TAILQ_REMOVE(&group->socks_with_data, sock, link);
1155 : }
1156 :
1157 10 : rc = accept(sock->fd, (struct sockaddr *)&sa, &salen);
1158 :
1159 10 : if (rc == -1) {
1160 2 : return NULL;
1161 : }
1162 :
1163 8 : fd = rc;
1164 :
1165 8 : flag = fcntl(fd, F_GETFL);
1166 8 : if ((!(flag & O_NONBLOCK)) && (fcntl(fd, F_SETFL, flag | O_NONBLOCK) < 0)) {
1167 0 : SPDK_ERRLOG("fcntl can't set nonblocking mode for socket, fd: %d (%d)\n", fd, errno);
1168 0 : close(fd);
1169 0 : return NULL;
1170 : }
1171 :
1172 : #if defined(SO_PRIORITY)
1173 : /* The priority is not inherited, so call this function again */
1174 8 : if (sock->base.opts.priority) {
1175 0 : rc = setsockopt(fd, SOL_SOCKET, SO_PRIORITY, &sock->base.opts.priority, sizeof(int));
1176 0 : if (rc != 0) {
1177 0 : close(fd);
1178 0 : return NULL;
1179 : }
1180 : }
1181 : #endif
1182 :
1183 : /* Establish SSL connection */
1184 8 : if (enable_ssl) {
1185 0 : ctx = posix_sock_create_ssl_context(TLS_server_method(), &sock->base.opts, &sock->base.impl_opts);
1186 0 : if (!ctx) {
1187 0 : SPDK_ERRLOG("posix_sock_create_ssl_context() failed, errno = %d\n", errno);
1188 0 : close(fd);
1189 0 : return NULL;
1190 : }
1191 0 : ssl = ssl_sock_setup_accept(ctx, fd);
1192 0 : if (!ssl) {
1193 0 : SPDK_ERRLOG("ssl_sock_setup_accept() failed, errno = %d\n", errno);
1194 0 : close(fd);
1195 0 : SSL_CTX_free(ctx);
1196 0 : return NULL;
1197 : }
1198 : }
1199 :
1200 : /* Inherit the zero copy feature from the listen socket */
1201 8 : new_sock = posix_sock_alloc(fd, &sock->base.impl_opts, sock->zcopy);
1202 8 : if (new_sock == NULL) {
1203 0 : close(fd);
1204 0 : SSL_free(ssl);
1205 0 : SSL_CTX_free(ctx);
1206 0 : return NULL;
1207 : }
1208 :
1209 8 : if (ctx) {
1210 0 : new_sock->ctx = ctx;
1211 : }
1212 :
1213 8 : if (ssl) {
1214 0 : new_sock->ssl = ssl;
1215 0 : SSL_set_app_data(ssl, &new_sock->base.impl_opts);
1216 : }
1217 :
1218 8 : return &new_sock->base;
1219 : }
1220 :
1221 : static struct spdk_sock *
1222 10 : posix_sock_accept(struct spdk_sock *_sock)
1223 : {
1224 10 : return _posix_sock_accept(_sock, false);
1225 : }
1226 :
1227 : static int
1228 24 : posix_sock_close(struct spdk_sock *_sock)
1229 : {
1230 24 : struct spdk_posix_sock *sock = __posix_sock(_sock);
1231 : void *pipe_buf;
1232 :
1233 24 : assert(TAILQ_EMPTY(&_sock->pending_reqs));
1234 :
1235 24 : if (sock->ssl != NULL) {
1236 0 : SSL_shutdown(sock->ssl);
1237 : }
1238 :
1239 : /* If the socket fails to close, the best choice is to
1240 : * leak the fd but continue to free the rest of the sock
1241 : * memory. */
1242 24 : close(sock->fd);
1243 :
1244 24 : SSL_free(sock->ssl);
1245 24 : SSL_CTX_free(sock->ctx);
1246 :
1247 24 : pipe_buf = spdk_pipe_destroy(sock->recv_pipe);
1248 24 : free(pipe_buf);
1249 24 : free(sock);
1250 :
1251 24 : return 0;
1252 : }
1253 :
1254 : #ifdef SPDK_ZEROCOPY
1255 : static int
1256 0 : _sock_check_zcopy(struct spdk_sock *sock)
1257 : {
1258 0 : struct spdk_posix_sock *psock = __posix_sock(sock);
1259 0 : struct msghdr msgh = {};
1260 0 : uint8_t buf[sizeof(struct cmsghdr) + sizeof(struct sock_extended_err)];
1261 : ssize_t rc;
1262 : struct sock_extended_err *serr;
1263 : struct cmsghdr *cm;
1264 : uint32_t idx;
1265 : struct spdk_sock_request *req, *treq;
1266 : bool found;
1267 :
1268 0 : msgh.msg_control = buf;
1269 0 : msgh.msg_controllen = sizeof(buf);
1270 :
1271 : while (true) {
1272 0 : rc = recvmsg(psock->fd, &msgh, MSG_ERRQUEUE);
1273 :
1274 0 : if (rc < 0) {
1275 0 : if (errno == EWOULDBLOCK || errno == EAGAIN) {
1276 0 : return 0;
1277 : }
1278 :
1279 0 : if (!TAILQ_EMPTY(&sock->pending_reqs)) {
1280 0 : SPDK_ERRLOG("Attempting to receive from ERRQUEUE yielded error, but pending list still has orphaned entries\n");
1281 : } else {
1282 0 : SPDK_WARNLOG("Recvmsg yielded an error!\n");
1283 : }
1284 0 : return 0;
1285 : }
1286 :
1287 0 : cm = CMSG_FIRSTHDR(&msgh);
1288 0 : if (!(cm &&
1289 0 : ((cm->cmsg_level == SOL_IP && cm->cmsg_type == IP_RECVERR) ||
1290 0 : (cm->cmsg_level == SOL_IPV6 && cm->cmsg_type == IPV6_RECVERR)))) {
1291 0 : SPDK_WARNLOG("Unexpected cmsg level or type!\n");
1292 0 : return 0;
1293 : }
1294 :
1295 0 : serr = (struct sock_extended_err *)CMSG_DATA(cm);
1296 0 : if (serr->ee_errno != 0 || serr->ee_origin != SO_EE_ORIGIN_ZEROCOPY) {
1297 0 : SPDK_WARNLOG("Unexpected extended error origin\n");
1298 0 : return 0;
1299 : }
1300 :
1301 : /* Most of the time, the pending_reqs array is in the exact
1302 : * order we need such that all of the requests to complete are
1303 : * in order, in the front. It is guaranteed that all requests
1304 : * belonging to the same sendmsg call are sequential, so once
1305 : * we encounter one match we can stop looping as soon as a
1306 : * non-match is found.
1307 : */
1308 0 : idx = serr->ee_info;
1309 : while (true) {
1310 0 : found = false;
1311 0 : TAILQ_FOREACH_SAFE(req, &sock->pending_reqs, internal.link, treq) {
1312 0 : if (!req->internal.is_zcopy) {
1313 : /* This wasn't a zcopy request. It was just waiting in line to complete */
1314 0 : rc = spdk_sock_request_put(sock, req, 0);
1315 0 : if (rc < 0) {
1316 0 : return rc;
1317 : }
1318 0 : } else if (req->internal.offset == idx) {
1319 0 : found = true;
1320 0 : rc = spdk_sock_request_put(sock, req, 0);
1321 0 : if (rc < 0) {
1322 0 : return rc;
1323 : }
1324 0 : } else if (found) {
1325 0 : break;
1326 : }
1327 : }
1328 :
1329 0 : if (idx == serr->ee_data) {
1330 0 : break;
1331 : }
1332 :
1333 0 : if (idx == UINT32_MAX) {
1334 0 : idx = 0;
1335 : } else {
1336 0 : idx++;
1337 : }
1338 : }
1339 : }
1340 :
1341 : return 0;
1342 : }
1343 : #endif
1344 :
1345 : static int
1346 29 : _sock_flush(struct spdk_sock *sock)
1347 : {
1348 29 : struct spdk_posix_sock *psock = __posix_sock(sock);
1349 29 : struct msghdr msg = {};
1350 29 : int flags;
1351 29 : struct iovec iovs[IOV_BATCH_SIZE];
1352 : int iovcnt;
1353 : int retval;
1354 : struct spdk_sock_request *req;
1355 : int i;
1356 : ssize_t rc, sent;
1357 : unsigned int offset;
1358 : size_t len;
1359 29 : bool is_zcopy = false;
1360 :
1361 : /* Can't flush from within a callback or we end up with recursive calls */
1362 29 : if (sock->cb_cnt > 0) {
1363 0 : errno = EAGAIN;
1364 0 : return -1;
1365 : }
1366 :
1367 : #ifdef SPDK_ZEROCOPY
1368 29 : if (psock->zcopy) {
1369 0 : flags = MSG_ZEROCOPY | MSG_NOSIGNAL;
1370 : } else
1371 : #endif
1372 : {
1373 29 : flags = MSG_NOSIGNAL;
1374 : }
1375 :
1376 29 : iovcnt = spdk_sock_prep_reqs(sock, iovs, 0, NULL, &flags);
1377 29 : if (iovcnt == 0) {
1378 22 : return 0;
1379 : }
1380 :
1381 : #ifdef SPDK_ZEROCOPY
1382 7 : is_zcopy = flags & MSG_ZEROCOPY;
1383 : #endif
1384 :
1385 : /* Perform the vectored write */
1386 7 : msg.msg_iov = iovs;
1387 7 : msg.msg_iovlen = iovcnt;
1388 :
1389 7 : if (psock->ssl) {
1390 0 : rc = SSL_writev(psock->ssl, iovs, iovcnt);
1391 : } else {
1392 7 : rc = sendmsg(psock->fd, &msg, flags);
1393 : }
1394 7 : if (rc <= 0) {
1395 0 : if (rc == 0 || errno == EAGAIN || errno == EWOULDBLOCK || (errno == ENOBUFS && psock->zcopy)) {
1396 0 : errno = EAGAIN;
1397 : }
1398 0 : return -1;
1399 : }
1400 :
1401 7 : sent = rc;
1402 :
1403 7 : if (is_zcopy) {
1404 : /* Handling overflow case, because we use psock->sendmsg_idx - 1 for the
1405 : * req->internal.offset, so sendmsg_idx should not be zero */
1406 0 : if (spdk_unlikely(psock->sendmsg_idx == UINT32_MAX)) {
1407 0 : psock->sendmsg_idx = 1;
1408 : } else {
1409 0 : psock->sendmsg_idx++;
1410 : }
1411 : }
1412 :
1413 : /* Consume the requests that were actually written */
1414 7 : req = TAILQ_FIRST(&sock->queued_reqs);
1415 8 : while (req) {
1416 8 : offset = req->internal.offset;
1417 :
1418 : /* req->internal.is_zcopy is true when the whole req or part of it is sent with zerocopy */
1419 8 : req->internal.is_zcopy = is_zcopy;
1420 :
1421 20 : for (i = 0; i < req->iovcnt; i++) {
1422 : /* Advance by the offset first */
1423 14 : if (offset >= SPDK_SOCK_REQUEST_IOV(req, i)->iov_len) {
1424 1 : offset -= SPDK_SOCK_REQUEST_IOV(req, i)->iov_len;
1425 1 : continue;
1426 : }
1427 :
1428 : /* Calculate the remaining length of this element */
1429 13 : len = SPDK_SOCK_REQUEST_IOV(req, i)->iov_len - offset;
1430 :
1431 13 : if (len > (size_t)rc) {
1432 : /* This element was partially sent. */
1433 2 : req->internal.offset += rc;
1434 2 : return sent;
1435 : }
1436 :
1437 11 : offset = 0;
1438 11 : req->internal.offset += len;
1439 11 : rc -= len;
1440 : }
1441 :
1442 : /* Handled a full request. */
1443 6 : spdk_sock_request_pend(sock, req);
1444 :
1445 6 : if (!req->internal.is_zcopy && req == TAILQ_FIRST(&sock->pending_reqs)) {
1446 : /* The sendmsg syscall above isn't currently asynchronous,
1447 : * so it's already done. */
1448 6 : retval = spdk_sock_request_put(sock, req, 0);
1449 6 : if (retval) {
1450 1 : break;
1451 : }
1452 : } else {
1453 : /* Re-use the offset field to hold the sendmsg call index. The
1454 : * index is 0 based, so subtract one here because we've already
1455 : * incremented above. */
1456 0 : req->internal.offset = psock->sendmsg_idx - 1;
1457 : }
1458 :
1459 5 : if (rc == 0) {
1460 4 : break;
1461 : }
1462 :
1463 1 : req = TAILQ_FIRST(&sock->queued_reqs);
1464 : }
1465 :
1466 5 : return sent;
1467 : }
1468 :
1469 : static int
1470 1 : posix_sock_flush(struct spdk_sock *sock)
1471 : {
1472 : #ifdef SPDK_ZEROCOPY
1473 1 : struct spdk_posix_sock *psock = __posix_sock(sock);
1474 :
1475 1 : if (psock->zcopy && !TAILQ_EMPTY(&sock->pending_reqs)) {
1476 0 : _sock_check_zcopy(sock);
1477 : }
1478 : #endif
1479 :
1480 1 : return _sock_flush(sock);
1481 : }
1482 :
1483 : static ssize_t
1484 0 : posix_sock_recv_from_pipe(struct spdk_posix_sock *sock, struct iovec *diov, int diovcnt)
1485 : {
1486 0 : struct iovec siov[2];
1487 : int sbytes;
1488 : ssize_t bytes;
1489 : struct spdk_posix_sock_group_impl *group;
1490 :
1491 0 : sbytes = spdk_pipe_reader_get_buffer(sock->recv_pipe, sock->recv_buf_sz, siov);
1492 0 : if (sbytes < 0) {
1493 0 : errno = EINVAL;
1494 0 : return -1;
1495 0 : } else if (sbytes == 0) {
1496 0 : errno = EAGAIN;
1497 0 : return -1;
1498 : }
1499 :
1500 0 : bytes = spdk_iovcpy(siov, 2, diov, diovcnt);
1501 :
1502 0 : if (bytes == 0) {
1503 : /* The only way this happens is if diov is 0 length */
1504 0 : errno = EINVAL;
1505 0 : return -1;
1506 : }
1507 :
1508 0 : spdk_pipe_reader_advance(sock->recv_pipe, bytes);
1509 :
1510 : /* If we drained the pipe, mark it appropriately */
1511 0 : if (spdk_pipe_reader_bytes_available(sock->recv_pipe) == 0) {
1512 0 : assert(sock->pipe_has_data == true);
1513 :
1514 0 : group = __posix_group_impl(sock->base.group_impl);
1515 0 : if (group && !sock->socket_has_data) {
1516 0 : TAILQ_REMOVE(&group->socks_with_data, sock, link);
1517 : }
1518 :
1519 0 : sock->pipe_has_data = false;
1520 : }
1521 :
1522 0 : return bytes;
1523 : }
1524 :
1525 : static inline ssize_t
1526 0 : posix_sock_read(struct spdk_posix_sock *sock)
1527 : {
1528 0 : struct iovec iov[2];
1529 : int bytes_avail, bytes_recvd;
1530 : struct spdk_posix_sock_group_impl *group;
1531 :
1532 0 : bytes_avail = spdk_pipe_writer_get_buffer(sock->recv_pipe, sock->recv_buf_sz, iov);
1533 :
1534 0 : if (bytes_avail <= 0) {
1535 0 : return bytes_avail;
1536 : }
1537 :
1538 0 : if (sock->ssl) {
1539 0 : bytes_recvd = SSL_readv(sock->ssl, iov, 2);
1540 : } else {
1541 0 : bytes_recvd = readv(sock->fd, iov, 2);
1542 : }
1543 :
1544 0 : assert(sock->pipe_has_data == false);
1545 :
1546 0 : if (bytes_recvd <= 0) {
1547 : /* Errors count as draining the socket data */
1548 0 : if (sock->base.group_impl && sock->socket_has_data) {
1549 0 : group = __posix_group_impl(sock->base.group_impl);
1550 0 : TAILQ_REMOVE(&group->socks_with_data, sock, link);
1551 : }
1552 :
1553 0 : sock->socket_has_data = false;
1554 :
1555 0 : return bytes_recvd;
1556 : }
1557 :
1558 0 : spdk_pipe_writer_advance(sock->recv_pipe, bytes_recvd);
1559 :
1560 : #if DEBUG
1561 0 : if (sock->base.group_impl) {
1562 0 : assert(sock->socket_has_data == true);
1563 : }
1564 : #endif
1565 :
1566 0 : sock->pipe_has_data = true;
1567 0 : if (bytes_recvd < bytes_avail) {
1568 : /* We drained the kernel socket entirely. */
1569 0 : sock->socket_has_data = false;
1570 : }
1571 :
1572 0 : return bytes_recvd;
1573 : }
1574 :
1575 : static ssize_t
1576 9 : posix_sock_readv(struct spdk_sock *_sock, struct iovec *iov, int iovcnt)
1577 : {
1578 9 : struct spdk_posix_sock *sock = __posix_sock(_sock);
1579 9 : struct spdk_posix_sock_group_impl *group = __posix_group_impl(sock->base.group_impl);
1580 : int rc, i;
1581 : size_t len;
1582 :
1583 9 : if (sock->recv_pipe == NULL) {
1584 9 : assert(sock->pipe_has_data == false);
1585 9 : if (group && sock->socket_has_data) {
1586 5 : sock->socket_has_data = false;
1587 5 : TAILQ_REMOVE(&group->socks_with_data, sock, link);
1588 : }
1589 9 : if (sock->ssl) {
1590 0 : return SSL_readv(sock->ssl, iov, iovcnt);
1591 : } else {
1592 9 : return readv(sock->fd, iov, iovcnt);
1593 : }
1594 : }
1595 :
1596 : /* If the socket is not in a group, we must assume it always has
1597 : * data waiting for us because it is not epolled */
1598 0 : if (!sock->pipe_has_data && (group == NULL || sock->socket_has_data)) {
1599 : /* If the user is receiving a sufficiently large amount of data,
1600 : * receive directly to their buffers. */
1601 0 : len = 0;
1602 0 : for (i = 0; i < iovcnt; i++) {
1603 0 : len += iov[i].iov_len;
1604 : }
1605 :
1606 0 : if (len >= MIN_SOCK_PIPE_SIZE) {
1607 : /* TODO: Should this detect if kernel socket is drained? */
1608 0 : if (sock->ssl) {
1609 0 : return SSL_readv(sock->ssl, iov, iovcnt);
1610 : } else {
1611 0 : return readv(sock->fd, iov, iovcnt);
1612 : }
1613 : }
1614 :
1615 : /* Otherwise, do a big read into our pipe */
1616 0 : rc = posix_sock_read(sock);
1617 0 : if (rc <= 0) {
1618 0 : return rc;
1619 : }
1620 : }
1621 :
1622 0 : return posix_sock_recv_from_pipe(sock, iov, iovcnt);
1623 : }
1624 :
1625 : static ssize_t
1626 7 : posix_sock_recv(struct spdk_sock *sock, void *buf, size_t len)
1627 : {
1628 7 : struct iovec iov[1];
1629 :
1630 7 : iov[0].iov_base = buf;
1631 7 : iov[0].iov_len = len;
1632 :
1633 7 : return posix_sock_readv(sock, iov, 1);
1634 : }
1635 :
1636 : static ssize_t
1637 7 : posix_sock_writev(struct spdk_sock *_sock, struct iovec *iov, int iovcnt)
1638 : {
1639 7 : struct spdk_posix_sock *sock = __posix_sock(_sock);
1640 : int rc;
1641 :
1642 : /* In order to process a writev, we need to flush any asynchronous writes
1643 : * first. */
1644 7 : rc = _sock_flush(_sock);
1645 7 : if (rc < 0) {
1646 0 : return rc;
1647 : }
1648 :
1649 7 : if (!TAILQ_EMPTY(&_sock->queued_reqs)) {
1650 : /* We weren't able to flush all requests */
1651 0 : errno = EAGAIN;
1652 0 : return -1;
1653 : }
1654 :
1655 7 : if (sock->ssl) {
1656 0 : return SSL_writev(sock->ssl, iov, iovcnt);
1657 : } else {
1658 7 : return writev(sock->fd, iov, iovcnt);
1659 : }
1660 : }
1661 :
1662 : static int
1663 0 : posix_sock_recv_next(struct spdk_sock *_sock, void **buf, void **ctx)
1664 : {
1665 0 : struct spdk_posix_sock *sock = __posix_sock(_sock);
1666 0 : struct iovec iov;
1667 : ssize_t rc;
1668 :
1669 0 : if (sock->recv_pipe != NULL) {
1670 0 : errno = ENOTSUP;
1671 0 : return -1;
1672 : }
1673 :
1674 0 : iov.iov_len = spdk_sock_group_get_buf(_sock->group_impl->group, &iov.iov_base, ctx);
1675 0 : if (iov.iov_len == 0) {
1676 0 : errno = ENOBUFS;
1677 0 : return -1;
1678 : }
1679 :
1680 0 : rc = posix_sock_readv(_sock, &iov, 1);
1681 0 : if (rc <= 0) {
1682 0 : spdk_sock_group_provide_buf(_sock->group_impl->group, iov.iov_base, iov.iov_len, *ctx);
1683 0 : return rc;
1684 : }
1685 :
1686 0 : *buf = iov.iov_base;
1687 :
1688 0 : return rc;
1689 : }
1690 :
1691 : static void
1692 2 : posix_sock_writev_async(struct spdk_sock *sock, struct spdk_sock_request *req)
1693 : {
1694 : int rc;
1695 :
1696 2 : spdk_sock_request_queue(sock, req);
1697 :
1698 : /* If there are a sufficient number queued, just flush them out immediately. */
1699 2 : if (sock->queued_iovcnt >= IOV_BATCH_SIZE) {
1700 0 : rc = _sock_flush(sock);
1701 0 : if (rc < 0 && errno != EAGAIN) {
1702 0 : spdk_sock_abort_requests(sock);
1703 : }
1704 : }
1705 2 : }
1706 :
1707 : static int
1708 1 : posix_sock_set_recvlowat(struct spdk_sock *_sock, int nbytes)
1709 : {
1710 1 : struct spdk_posix_sock *sock = __posix_sock(_sock);
1711 1 : int val;
1712 : int rc;
1713 :
1714 1 : assert(sock != NULL);
1715 :
1716 1 : val = nbytes;
1717 1 : rc = setsockopt(sock->fd, SOL_SOCKET, SO_RCVLOWAT, &val, sizeof val);
1718 1 : if (rc != 0) {
1719 0 : return -1;
1720 : }
1721 1 : return 0;
1722 : }
1723 :
1724 : static bool
1725 1 : posix_sock_is_ipv6(struct spdk_sock *_sock)
1726 : {
1727 1 : struct spdk_posix_sock *sock = __posix_sock(_sock);
1728 1 : struct sockaddr_storage sa;
1729 1 : socklen_t salen;
1730 : int rc;
1731 :
1732 1 : assert(sock != NULL);
1733 :
1734 1 : memset(&sa, 0, sizeof sa);
1735 1 : salen = sizeof sa;
1736 1 : rc = getsockname(sock->fd, (struct sockaddr *) &sa, &salen);
1737 1 : if (rc != 0) {
1738 0 : SPDK_ERRLOG("getsockname() failed (errno=%d)\n", errno);
1739 0 : return false;
1740 : }
1741 :
1742 1 : return (sa.ss_family == AF_INET6);
1743 : }
1744 :
1745 : static bool
1746 1 : posix_sock_is_ipv4(struct spdk_sock *_sock)
1747 : {
1748 1 : struct spdk_posix_sock *sock = __posix_sock(_sock);
1749 1 : struct sockaddr_storage sa;
1750 1 : socklen_t salen;
1751 : int rc;
1752 :
1753 1 : assert(sock != NULL);
1754 :
1755 1 : memset(&sa, 0, sizeof sa);
1756 1 : salen = sizeof sa;
1757 1 : rc = getsockname(sock->fd, (struct sockaddr *) &sa, &salen);
1758 1 : if (rc != 0) {
1759 0 : SPDK_ERRLOG("getsockname() failed (errno=%d)\n", errno);
1760 0 : return false;
1761 : }
1762 :
1763 1 : return (sa.ss_family == AF_INET);
1764 : }
1765 :
1766 : static bool
1767 3 : posix_sock_is_connected(struct spdk_sock *_sock)
1768 : {
1769 3 : struct spdk_posix_sock *sock = __posix_sock(_sock);
1770 3 : uint8_t byte;
1771 : int rc;
1772 :
1773 3 : rc = recv(sock->fd, &byte, 1, MSG_PEEK);
1774 3 : if (rc == 0) {
1775 1 : return false;
1776 : }
1777 :
1778 2 : if (rc < 0) {
1779 2 : if (errno == EAGAIN || errno == EWOULDBLOCK) {
1780 2 : return true;
1781 : }
1782 :
1783 0 : return false;
1784 : }
1785 :
1786 0 : return true;
1787 : }
1788 :
1789 : static struct spdk_sock_group_impl *
1790 1 : posix_sock_group_impl_get_optimal(struct spdk_sock *_sock, struct spdk_sock_group_impl *hint)
1791 : {
1792 1 : struct spdk_posix_sock *sock = __posix_sock(_sock);
1793 1 : struct spdk_sock_group_impl *group_impl;
1794 :
1795 1 : if (sock->placement_id != -1) {
1796 0 : spdk_sock_map_lookup(&g_map, sock->placement_id, &group_impl, hint);
1797 0 : return group_impl;
1798 : }
1799 :
1800 1 : return NULL;
1801 : }
1802 :
1803 : static struct spdk_sock_group_impl *
1804 12 : _sock_group_impl_create(uint32_t enable_placement_id)
1805 : {
1806 : struct spdk_posix_sock_group_impl *group_impl;
1807 : int fd;
1808 :
1809 : #if defined(SPDK_EPOLL)
1810 12 : fd = epoll_create1(0);
1811 : #elif defined(SPDK_KEVENT)
1812 : fd = kqueue();
1813 : #endif
1814 12 : if (fd == -1) {
1815 0 : return NULL;
1816 : }
1817 :
1818 12 : group_impl = calloc(1, sizeof(*group_impl));
1819 12 : if (group_impl == NULL) {
1820 0 : SPDK_ERRLOG("group_impl allocation failed\n");
1821 0 : close(fd);
1822 0 : return NULL;
1823 : }
1824 :
1825 12 : group_impl->pipe_group = spdk_pipe_group_create();
1826 12 : if (group_impl->pipe_group == NULL) {
1827 0 : SPDK_ERRLOG("pipe_group allocation failed\n");
1828 0 : free(group_impl);
1829 0 : close(fd);
1830 0 : return NULL;
1831 : }
1832 :
1833 12 : group_impl->fd = fd;
1834 12 : TAILQ_INIT(&group_impl->socks_with_data);
1835 12 : group_impl->placement_id = -1;
1836 :
1837 12 : if (enable_placement_id == PLACEMENT_CPU) {
1838 0 : spdk_sock_map_insert(&g_map, spdk_env_get_current_core(), &group_impl->base);
1839 0 : group_impl->placement_id = spdk_env_get_current_core();
1840 : }
1841 :
1842 12 : return &group_impl->base;
1843 : }
1844 :
1845 : static struct spdk_sock_group_impl *
1846 6 : posix_sock_group_impl_create(void)
1847 : {
1848 6 : return _sock_group_impl_create(g_posix_impl_opts.enable_placement_id);
1849 : }
1850 :
1851 : static struct spdk_sock_group_impl *
1852 6 : ssl_sock_group_impl_create(void)
1853 : {
1854 6 : return _sock_group_impl_create(g_ssl_impl_opts.enable_placement_id);
1855 : }
1856 :
1857 : static void
1858 0 : posix_sock_mark(struct spdk_posix_sock_group_impl *group, struct spdk_posix_sock *sock,
1859 : int placement_id)
1860 : {
1861 : #if defined(SO_MARK)
1862 : int rc;
1863 :
1864 0 : rc = setsockopt(sock->fd, SOL_SOCKET, SO_MARK,
1865 : &placement_id, sizeof(placement_id));
1866 0 : if (rc != 0) {
1867 : /* Not fatal */
1868 0 : SPDK_ERRLOG("Error setting SO_MARK\n");
1869 0 : return;
1870 : }
1871 :
1872 0 : rc = spdk_sock_map_insert(&g_map, placement_id, &group->base);
1873 0 : if (rc != 0) {
1874 : /* Not fatal */
1875 0 : SPDK_ERRLOG("Failed to insert sock group into map: %d\n", rc);
1876 0 : return;
1877 : }
1878 :
1879 0 : sock->placement_id = placement_id;
1880 : #endif
1881 : }
1882 :
1883 : static void
1884 0 : posix_sock_update_mark(struct spdk_sock_group_impl *_group, struct spdk_sock *_sock)
1885 : {
1886 0 : struct spdk_posix_sock_group_impl *group = __posix_group_impl(_group);
1887 :
1888 0 : if (group->placement_id == -1) {
1889 0 : group->placement_id = spdk_sock_map_find_free(&g_map);
1890 :
1891 : /* If a free placement id is found, update existing sockets in this group */
1892 0 : if (group->placement_id != -1) {
1893 : struct spdk_sock *sock, *tmp;
1894 :
1895 0 : TAILQ_FOREACH_SAFE(sock, &_group->socks, link, tmp) {
1896 0 : posix_sock_mark(group, __posix_sock(sock), group->placement_id);
1897 : }
1898 : }
1899 : }
1900 :
1901 0 : if (group->placement_id != -1) {
1902 : /*
1903 : * group placement id is already determined for this poll group.
1904 : * Mark socket with group's placement id.
1905 : */
1906 0 : posix_sock_mark(group, __posix_sock(_sock), group->placement_id);
1907 : }
1908 0 : }
1909 :
1910 : static int
1911 5 : posix_sock_group_impl_add_sock(struct spdk_sock_group_impl *_group, struct spdk_sock *_sock)
1912 : {
1913 5 : struct spdk_posix_sock_group_impl *group = __posix_group_impl(_group);
1914 5 : struct spdk_posix_sock *sock = __posix_sock(_sock);
1915 : int rc;
1916 :
1917 : #if defined(SPDK_EPOLL)
1918 5 : struct epoll_event event;
1919 :
1920 5 : memset(&event, 0, sizeof(event));
1921 : /* EPOLLERR is always on even if we don't set it, but be explicit for clarity */
1922 5 : event.events = EPOLLIN | EPOLLERR;
1923 5 : event.data.ptr = sock;
1924 :
1925 5 : rc = epoll_ctl(group->fd, EPOLL_CTL_ADD, sock->fd, &event);
1926 : #elif defined(SPDK_KEVENT)
1927 : struct kevent event;
1928 : struct timespec ts = {0};
1929 :
1930 : EV_SET(&event, sock->fd, EVFILT_READ, EV_ADD, 0, 0, sock);
1931 :
1932 : rc = kevent(group->fd, &event, 1, NULL, 0, &ts);
1933 : #endif
1934 :
1935 5 : if (rc != 0) {
1936 0 : return rc;
1937 : }
1938 :
1939 : /* switched from another polling group due to scheduling */
1940 5 : if (spdk_unlikely(sock->recv_pipe != NULL &&
1941 : (spdk_pipe_reader_bytes_available(sock->recv_pipe) > 0))) {
1942 0 : sock->pipe_has_data = true;
1943 0 : sock->socket_has_data = false;
1944 0 : TAILQ_INSERT_TAIL(&group->socks_with_data, sock, link);
1945 5 : } else if (sock->recv_pipe != NULL) {
1946 0 : rc = spdk_pipe_group_add(group->pipe_group, sock->recv_pipe);
1947 0 : assert(rc == 0);
1948 : }
1949 :
1950 5 : if (_sock->impl_opts.enable_placement_id == PLACEMENT_MARK) {
1951 0 : posix_sock_update_mark(_group, _sock);
1952 5 : } else if (sock->placement_id != -1) {
1953 0 : rc = spdk_sock_map_insert(&g_map, sock->placement_id, &group->base);
1954 0 : if (rc != 0) {
1955 0 : SPDK_ERRLOG("Failed to insert sock group into map: %d\n", rc);
1956 : /* Do not treat this as an error. The system will continue running. */
1957 : }
1958 : }
1959 :
1960 5 : return rc;
1961 : }
1962 :
1963 : static int
1964 5 : posix_sock_group_impl_remove_sock(struct spdk_sock_group_impl *_group, struct spdk_sock *_sock)
1965 : {
1966 5 : struct spdk_posix_sock_group_impl *group = __posix_group_impl(_group);
1967 5 : struct spdk_posix_sock *sock = __posix_sock(_sock);
1968 : int rc;
1969 :
1970 5 : if (sock->pipe_has_data || sock->socket_has_data) {
1971 0 : TAILQ_REMOVE(&group->socks_with_data, sock, link);
1972 0 : sock->pipe_has_data = false;
1973 0 : sock->socket_has_data = false;
1974 5 : } else if (sock->recv_pipe != NULL) {
1975 0 : rc = spdk_pipe_group_remove(group->pipe_group, sock->recv_pipe);
1976 0 : assert(rc == 0);
1977 : }
1978 :
1979 5 : if (sock->placement_id != -1) {
1980 0 : spdk_sock_map_release(&g_map, sock->placement_id);
1981 : }
1982 :
1983 : #if defined(SPDK_EPOLL)
1984 5 : struct epoll_event event;
1985 :
1986 : /* Event parameter is ignored but some old kernel version still require it. */
1987 5 : rc = epoll_ctl(group->fd, EPOLL_CTL_DEL, sock->fd, &event);
1988 : #elif defined(SPDK_KEVENT)
1989 : struct kevent event;
1990 : struct timespec ts = {0};
1991 :
1992 : EV_SET(&event, sock->fd, EVFILT_READ, EV_DELETE, 0, 0, NULL);
1993 :
1994 : rc = kevent(group->fd, &event, 1, NULL, 0, &ts);
1995 : if (rc == 0 && event.flags & EV_ERROR) {
1996 : rc = -1;
1997 : errno = event.data;
1998 : }
1999 : #endif
2000 :
2001 5 : spdk_sock_abort_requests(_sock);
2002 :
2003 5 : return rc;
2004 : }
2005 :
2006 : static int
2007 7 : posix_sock_group_impl_poll(struct spdk_sock_group_impl *_group, int max_events,
2008 : struct spdk_sock **socks)
2009 : {
2010 7 : struct spdk_posix_sock_group_impl *group = __posix_group_impl(_group);
2011 : struct spdk_sock *sock, *tmp;
2012 : int num_events, i, rc;
2013 : struct spdk_posix_sock *psock, *ptmp;
2014 : #if defined(SPDK_EPOLL)
2015 7 : struct epoll_event events[MAX_EVENTS_PER_POLL];
2016 : #elif defined(SPDK_KEVENT)
2017 : struct kevent events[MAX_EVENTS_PER_POLL];
2018 : struct timespec ts = {0};
2019 : #endif
2020 :
2021 : #ifdef SPDK_ZEROCOPY
2022 : /* When all of the following conditions are met
2023 : * - non-blocking socket
2024 : * - zero copy is enabled
2025 : * - interrupts suppressed (i.e. busy polling)
2026 : * - the NIC tx queue is full at the time sendmsg() is called
2027 : * - epoll_wait determines there is an EPOLLIN event for the socket
2028 : * then we can get into a situation where data we've sent is queued
2029 : * up in the kernel network stack, but interrupts have been suppressed
2030 : * because other traffic is flowing so the kernel misses the signal
2031 : * to flush the software tx queue. If there wasn't incoming data
2032 : * pending on the socket, then epoll_wait would have been sufficient
2033 : * to kick off the send operation, but since there is a pending event
2034 : * epoll_wait does not trigger the necessary operation.
2035 : *
2036 : * We deal with this by checking for all of the above conditions and
2037 : * additionally looking for EPOLLIN events that were not consumed from
2038 : * the last poll loop. We take this to mean that the upper layer is
2039 : * unable to consume them because it is blocked waiting for resources
2040 : * to free up, and those resources are most likely freed in response
2041 : * to a pending asynchronous write completing.
2042 : *
2043 : * Additionally, sockets that have the same placement_id actually share
2044 : * an underlying hardware queue. That means polling one of them is
2045 : * equivalent to polling all of them. As a quick mechanism to avoid
2046 : * making extra poll() calls, stash the last placement_id during the loop
2047 : * and only poll if it's not the same. The overwhelmingly common case
2048 : * is that all sockets in this list have the same placement_id because
2049 : * SPDK is intentionally grouping sockets by that value, so even
2050 : * though this won't stop all extra calls to poll(), it's very fast
2051 : * and will catch all of them in practice.
2052 : */
2053 7 : int last_placement_id = -1;
2054 :
2055 7 : TAILQ_FOREACH(psock, &group->socks_with_data, link) {
2056 0 : if (psock->zcopy && psock->placement_id >= 0 &&
2057 0 : psock->placement_id != last_placement_id) {
2058 0 : struct pollfd pfd = {psock->fd, POLLIN | POLLERR, 0};
2059 :
2060 0 : poll(&pfd, 1, 0);
2061 0 : last_placement_id = psock->placement_id;
2062 : }
2063 : }
2064 : #endif
2065 :
2066 : /* This must be a TAILQ_FOREACH_SAFE because while flushing,
2067 : * a completion callback could remove the sock from the
2068 : * group. */
2069 22 : TAILQ_FOREACH_SAFE(sock, &_group->socks, link, tmp) {
2070 15 : rc = _sock_flush(sock);
2071 15 : if (rc < 0 && errno != EAGAIN) {
2072 0 : spdk_sock_abort_requests(sock);
2073 : }
2074 : }
2075 :
2076 7 : assert(max_events > 0);
2077 :
2078 : #if defined(SPDK_EPOLL)
2079 7 : num_events = epoll_wait(group->fd, events, max_events, 0);
2080 : #elif defined(SPDK_KEVENT)
2081 : num_events = kevent(group->fd, NULL, 0, events, max_events, &ts);
2082 : #endif
2083 :
2084 7 : if (num_events == -1) {
2085 0 : return -1;
2086 7 : } else if (num_events == 0 && !TAILQ_EMPTY(&_group->socks)) {
2087 1 : sock = TAILQ_FIRST(&_group->socks);
2088 1 : psock = __posix_sock(sock);
2089 : /* poll() is called here to busy poll the queue associated with
2090 : * first socket in list and potentially reap incoming data.
2091 : */
2092 1 : if (sock->opts.priority) {
2093 0 : struct pollfd pfd = {0, 0, 0};
2094 :
2095 0 : pfd.fd = psock->fd;
2096 0 : pfd.events = POLLIN | POLLERR;
2097 0 : poll(&pfd, 1, 0);
2098 : }
2099 : }
2100 :
2101 12 : for (i = 0; i < num_events; i++) {
2102 : #if defined(SPDK_EPOLL)
2103 5 : sock = events[i].data.ptr;
2104 5 : psock = __posix_sock(sock);
2105 :
2106 : #ifdef SPDK_ZEROCOPY
2107 5 : if (events[i].events & EPOLLERR) {
2108 0 : rc = _sock_check_zcopy(sock);
2109 : /* If the socket was closed or removed from
2110 : * the group in response to a send ack, don't
2111 : * add it to the array here. */
2112 0 : if (rc || sock->cb_fn == NULL) {
2113 0 : continue;
2114 : }
2115 : }
2116 : #endif
2117 5 : if ((events[i].events & EPOLLIN) == 0) {
2118 0 : continue;
2119 : }
2120 :
2121 : #elif defined(SPDK_KEVENT)
2122 : sock = events[i].udata;
2123 : psock = __posix_sock(sock);
2124 : #endif
2125 :
2126 : /* If the socket is not already in the list, add it now */
2127 5 : if (!psock->socket_has_data && !psock->pipe_has_data) {
2128 5 : TAILQ_INSERT_TAIL(&group->socks_with_data, psock, link);
2129 : }
2130 5 : psock->socket_has_data = true;
2131 : }
2132 :
2133 7 : num_events = 0;
2134 :
2135 12 : TAILQ_FOREACH_SAFE(psock, &group->socks_with_data, link, ptmp) {
2136 5 : if (num_events == max_events) {
2137 0 : break;
2138 : }
2139 :
2140 : /* If the socket's cb_fn is NULL, just remove it from the
2141 : * list and do not add it to socks array */
2142 5 : if (spdk_unlikely(psock->base.cb_fn == NULL)) {
2143 0 : psock->socket_has_data = false;
2144 0 : psock->pipe_has_data = false;
2145 0 : TAILQ_REMOVE(&group->socks_with_data, psock, link);
2146 0 : continue;
2147 : }
2148 :
2149 5 : socks[num_events++] = &psock->base;
2150 : }
2151 :
2152 : /* Cycle the has_data list so that each time we poll things aren't
2153 : * in the same order. Say we have 6 sockets in the list, named as follows:
2154 : * A B C D E F
2155 : * And all 6 sockets had epoll events, but max_events is only 3. That means
2156 : * psock currently points at D. We want to rearrange the list to the following:
2157 : * D E F A B C
2158 : *
2159 : * The variables below are named according to this example to make it easier to
2160 : * follow the swaps.
2161 : */
2162 7 : if (psock != NULL) {
2163 : struct spdk_posix_sock *pa, *pc, *pd, *pf;
2164 :
2165 : /* Capture pointers to the elements we need */
2166 0 : pd = psock;
2167 0 : pc = TAILQ_PREV(pd, spdk_has_data_list, link);
2168 0 : pa = TAILQ_FIRST(&group->socks_with_data);
2169 0 : pf = TAILQ_LAST(&group->socks_with_data, spdk_has_data_list);
2170 :
2171 : /* Break the link between C and D */
2172 0 : pc->link.tqe_next = NULL;
2173 :
2174 : /* Connect F to A */
2175 0 : pf->link.tqe_next = pa;
2176 0 : pa->link.tqe_prev = &pf->link.tqe_next;
2177 :
2178 : /* Fix up the list first/last pointers */
2179 0 : group->socks_with_data.tqh_first = pd;
2180 0 : group->socks_with_data.tqh_last = &pc->link.tqe_next;
2181 :
2182 : /* D is in front of the list, make tqe prev pointer point to the head of list */
2183 0 : pd->link.tqe_prev = &group->socks_with_data.tqh_first;
2184 : }
2185 :
2186 7 : return num_events;
2187 : }
2188 :
2189 : static int
2190 0 : posix_sock_group_impl_register_interrupt(struct spdk_sock_group_impl *_group, uint32_t events,
2191 : spdk_interrupt_fn fn, void *arg, const char *name)
2192 : {
2193 0 : struct spdk_posix_sock_group_impl *group = __posix_group_impl(_group);
2194 :
2195 0 : group->intr = spdk_interrupt_register_for_events(group->fd, events, fn, arg, name);
2196 :
2197 0 : return group->intr ? 0 : -1;
2198 : }
2199 :
2200 : static void
2201 0 : posix_sock_group_impl_unregister_interrupt(struct spdk_sock_group_impl *_group)
2202 : {
2203 0 : struct spdk_posix_sock_group_impl *group = __posix_group_impl(_group);
2204 :
2205 0 : spdk_interrupt_unregister(&group->intr);
2206 0 : }
2207 :
2208 : static int
2209 12 : _sock_group_impl_close(struct spdk_sock_group_impl *_group, uint32_t enable_placement_id)
2210 : {
2211 12 : struct spdk_posix_sock_group_impl *group = __posix_group_impl(_group);
2212 : int rc;
2213 :
2214 12 : if (enable_placement_id == PLACEMENT_CPU) {
2215 0 : spdk_sock_map_release(&g_map, spdk_env_get_current_core());
2216 : }
2217 :
2218 12 : spdk_pipe_group_destroy(group->pipe_group);
2219 12 : rc = close(group->fd);
2220 12 : free(group);
2221 12 : return rc;
2222 : }
2223 :
2224 : static int
2225 6 : posix_sock_group_impl_close(struct spdk_sock_group_impl *_group)
2226 : {
2227 6 : return _sock_group_impl_close(_group, g_posix_impl_opts.enable_placement_id);
2228 : }
2229 :
2230 : static int
2231 6 : ssl_sock_group_impl_close(struct spdk_sock_group_impl *_group)
2232 : {
2233 6 : return _sock_group_impl_close(_group, g_ssl_impl_opts.enable_placement_id);
2234 : }
2235 :
2236 : static struct spdk_net_impl g_posix_net_impl = {
2237 : .name = "posix",
2238 : .getaddr = posix_sock_getaddr,
2239 : .get_interface_name = posix_sock_get_interface_name,
2240 : .get_numa_id = posix_sock_get_numa_id,
2241 : .connect = posix_sock_connect,
2242 : .listen = posix_sock_listen,
2243 : .accept = posix_sock_accept,
2244 : .close = posix_sock_close,
2245 : .recv = posix_sock_recv,
2246 : .readv = posix_sock_readv,
2247 : .writev = posix_sock_writev,
2248 : .recv_next = posix_sock_recv_next,
2249 : .writev_async = posix_sock_writev_async,
2250 : .flush = posix_sock_flush,
2251 : .set_recvlowat = posix_sock_set_recvlowat,
2252 : .set_recvbuf = posix_sock_set_recvbuf,
2253 : .set_sendbuf = posix_sock_set_sendbuf,
2254 : .is_ipv6 = posix_sock_is_ipv6,
2255 : .is_ipv4 = posix_sock_is_ipv4,
2256 : .is_connected = posix_sock_is_connected,
2257 : .group_impl_get_optimal = posix_sock_group_impl_get_optimal,
2258 : .group_impl_create = posix_sock_group_impl_create,
2259 : .group_impl_add_sock = posix_sock_group_impl_add_sock,
2260 : .group_impl_remove_sock = posix_sock_group_impl_remove_sock,
2261 : .group_impl_poll = posix_sock_group_impl_poll,
2262 : .group_impl_register_interrupt = posix_sock_group_impl_register_interrupt,
2263 : .group_impl_unregister_interrupt = posix_sock_group_impl_unregister_interrupt,
2264 : .group_impl_close = posix_sock_group_impl_close,
2265 : .get_opts = posix_sock_impl_get_opts,
2266 : .set_opts = posix_sock_impl_set_opts,
2267 : };
2268 :
2269 2 : SPDK_NET_IMPL_REGISTER_DEFAULT(posix, &g_posix_net_impl);
2270 :
2271 : static struct spdk_sock *
2272 0 : ssl_sock_listen(const char *ip, int port, struct spdk_sock_opts *opts)
2273 : {
2274 0 : return posix_sock_create(ip, port, SPDK_SOCK_CREATE_LISTEN, opts, true);
2275 : }
2276 :
2277 : static struct spdk_sock *
2278 0 : ssl_sock_connect(const char *ip, int port, struct spdk_sock_opts *opts)
2279 : {
2280 0 : return posix_sock_create(ip, port, SPDK_SOCK_CREATE_CONNECT, opts, true);
2281 : }
2282 :
2283 : static struct spdk_sock *
2284 0 : ssl_sock_accept(struct spdk_sock *_sock)
2285 : {
2286 0 : return _posix_sock_accept(_sock, true);
2287 : }
2288 :
2289 : static struct spdk_net_impl g_ssl_net_impl = {
2290 : .name = "ssl",
2291 : .getaddr = posix_sock_getaddr,
2292 : .get_interface_name = posix_sock_get_interface_name,
2293 : .get_numa_id = posix_sock_get_numa_id,
2294 : .connect = ssl_sock_connect,
2295 : .listen = ssl_sock_listen,
2296 : .accept = ssl_sock_accept,
2297 : .close = posix_sock_close,
2298 : .recv = posix_sock_recv,
2299 : .readv = posix_sock_readv,
2300 : .writev = posix_sock_writev,
2301 : .recv_next = posix_sock_recv_next,
2302 : .writev_async = posix_sock_writev_async,
2303 : .flush = posix_sock_flush,
2304 : .set_recvlowat = posix_sock_set_recvlowat,
2305 : .set_recvbuf = posix_sock_set_recvbuf,
2306 : .set_sendbuf = posix_sock_set_sendbuf,
2307 : .is_ipv6 = posix_sock_is_ipv6,
2308 : .is_ipv4 = posix_sock_is_ipv4,
2309 : .is_connected = posix_sock_is_connected,
2310 : .group_impl_get_optimal = posix_sock_group_impl_get_optimal,
2311 : .group_impl_create = ssl_sock_group_impl_create,
2312 : .group_impl_add_sock = posix_sock_group_impl_add_sock,
2313 : .group_impl_remove_sock = posix_sock_group_impl_remove_sock,
2314 : .group_impl_poll = posix_sock_group_impl_poll,
2315 : .group_impl_register_interrupt = posix_sock_group_impl_register_interrupt,
2316 : .group_impl_unregister_interrupt = posix_sock_group_impl_unregister_interrupt,
2317 : .group_impl_close = ssl_sock_group_impl_close,
2318 : .get_opts = ssl_sock_impl_get_opts,
2319 : .set_opts = ssl_sock_impl_set_opts,
2320 : };
2321 :
2322 2 : SPDK_NET_IMPL_REGISTER(ssl, &g_ssl_net_impl);
2323 2 : SPDK_LOG_REGISTER_COMPONENT(sock_posix)
|
