1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2016-2018 Intel Corporation
3 */
4
5 #include <dirent.h>
6 #include <errno.h>
7 #include <fcntl.h>
8 #include <fnmatch.h>
9 #include <inttypes.h>
10 #include <libgen.h>
11 #include <limits.h>
12 #include <pthread.h>
13 #include <stdio.h>
14 #include <stdlib.h>
15 #include <string.h>
16 #include <sys/file.h>
17 #include <sys/time.h>
18 #include <sys/socket.h>
19 #include <sys/un.h>
20 #include <unistd.h>
21
22 #include <rte_alarm.h>
23 #include <rte_common.h>
24 #include <rte_cycles.h>
25 #include <rte_eal.h>
26 #include <rte_errno.h>
27 #include <rte_lcore.h>
28 #include <rte_log.h>
29 #include <rte_thread.h>
30
31 #include "eal_memcfg.h"
32 #include "eal_private.h"
33 #include "eal_filesystem.h"
34 #include "eal_internal_cfg.h"
35
36 static RTE_ATOMIC(int) mp_fd = -1;
37 static rte_thread_t mp_handle_tid;
38 static char mp_filter[PATH_MAX]; /* Filter for secondary process sockets */
39 static char mp_dir_path[PATH_MAX]; /* The directory path for all mp sockets */
40 static pthread_mutex_t mp_mutex_action = PTHREAD_MUTEX_INITIALIZER;
41 static char peer_name[PATH_MAX];
42
43 struct action_entry {
44 TAILQ_ENTRY(action_entry) next;
45 char action_name[RTE_MP_MAX_NAME_LEN];
46 rte_mp_t action;
47 };
48
49 /** Double linked list of actions. */
50 TAILQ_HEAD(action_entry_list, action_entry);
51
52 static struct action_entry_list action_entry_list =
53 TAILQ_HEAD_INITIALIZER(action_entry_list);
54
55 enum mp_type {
56 MP_MSG, /* Share message with peers, will not block */
57 MP_REQ, /* Request for information, Will block for a reply */
58 MP_REP, /* Response to previously-received request */
59 MP_IGN, /* Response telling requester to ignore this response */
60 };
61
62 struct mp_msg_internal {
63 int type;
64 struct rte_mp_msg msg;
65 };
66
67 struct async_request_param {
68 rte_mp_async_reply_t clb;
69 struct rte_mp_reply user_reply;
70 struct timespec end;
71 int n_responses_processed;
72 };
73
74 struct pending_request {
75 TAILQ_ENTRY(pending_request) next;
76 enum {
77 REQUEST_TYPE_SYNC,
78 REQUEST_TYPE_ASYNC
79 } type;
80 char dst[PATH_MAX];
81 struct rte_mp_msg *request;
82 struct rte_mp_msg *reply;
83 int reply_received;
84 union {
85 struct {
86 struct async_request_param *param;
87 } async;
88 struct {
89 pthread_cond_t cond;
90 } sync;
91 };
92 };
93
94 TAILQ_HEAD(pending_request_list, pending_request);
95
96 static struct {
97 struct pending_request_list requests;
98 pthread_mutex_t lock;
99 } pending_requests = {
100 .requests = TAILQ_HEAD_INITIALIZER(pending_requests.requests),
101 .lock = PTHREAD_MUTEX_INITIALIZER,
102 /**< used in async requests only */
103 };
104
105 /* forward declarations */
106 static int
107 mp_send(struct rte_mp_msg *msg, const char *peer, int type);
108
109 /* for use with alarm callback */
110 static void
111 async_reply_handle(void *arg);
112
113 /* for use with process_msg */
114 static struct pending_request *
115 async_reply_handle_thread_unsafe(void *arg);
116
117 static void
118 trigger_async_action(struct pending_request *req);
119
120 static struct pending_request *
find_pending_request(const char * dst,const char * act_name)121 find_pending_request(const char *dst, const char *act_name)
122 {
123 struct pending_request *r;
124
125 TAILQ_FOREACH(r, &pending_requests.requests, next) {
126 if (!strcmp(r->dst, dst) &&
127 !strcmp(r->request->name, act_name))
128 break;
129 }
130
131 return r;
132 }
133
134 static void
create_socket_path(const char * name,char * buf,int len)135 create_socket_path(const char *name, char *buf, int len)
136 {
137 const char *prefix = eal_mp_socket_path();
138
139 if (strlen(name) > 0)
140 snprintf(buf, len, "%s_%s", prefix, name);
141 else
142 strlcpy(buf, prefix, len);
143 }
144
145 int
rte_eal_primary_proc_alive(const char * config_file_path)146 rte_eal_primary_proc_alive(const char *config_file_path)
147 {
148 int config_fd;
149
150 if (config_file_path)
151 config_fd = open(config_file_path, O_RDONLY);
152 else {
153 const char *path;
154
155 path = eal_runtime_config_path();
156 config_fd = open(path, O_RDONLY);
157 }
158 if (config_fd < 0)
159 return 0;
160
161 int ret = lockf(config_fd, F_TEST, 0);
162 close(config_fd);
163
164 return !!ret;
165 }
166
167 static struct action_entry *
find_action_entry_by_name(const char * name)168 find_action_entry_by_name(const char *name)
169 {
170 struct action_entry *entry;
171
172 TAILQ_FOREACH(entry, &action_entry_list, next) {
173 if (strncmp(entry->action_name, name, RTE_MP_MAX_NAME_LEN) == 0)
174 break;
175 }
176
177 return entry;
178 }
179
180 static int
validate_action_name(const char * name)181 validate_action_name(const char *name)
182 {
183 if (name == NULL) {
184 EAL_LOG(ERR, "Action name cannot be NULL");
185 rte_errno = EINVAL;
186 return -1;
187 }
188 if (strnlen(name, RTE_MP_MAX_NAME_LEN) == 0) {
189 EAL_LOG(ERR, "Length of action name is zero");
190 rte_errno = EINVAL;
191 return -1;
192 }
193 if (strnlen(name, RTE_MP_MAX_NAME_LEN) == RTE_MP_MAX_NAME_LEN) {
194 rte_errno = E2BIG;
195 return -1;
196 }
197 return 0;
198 }
199
200 int
rte_mp_action_register(const char * name,rte_mp_t action)201 rte_mp_action_register(const char *name, rte_mp_t action)
202 {
203 struct action_entry *entry;
204 const struct internal_config *internal_conf =
205 eal_get_internal_configuration();
206
207 if (validate_action_name(name) != 0)
208 return -1;
209
210 if (internal_conf->no_shconf) {
211 EAL_LOG(DEBUG, "No shared files mode enabled, IPC is disabled");
212 rte_errno = ENOTSUP;
213 return -1;
214 }
215
216 entry = malloc(sizeof(struct action_entry));
217 if (entry == NULL) {
218 rte_errno = ENOMEM;
219 return -1;
220 }
221 strlcpy(entry->action_name, name, sizeof(entry->action_name));
222 entry->action = action;
223
224 pthread_mutex_lock(&mp_mutex_action);
225 if (find_action_entry_by_name(name) != NULL) {
226 pthread_mutex_unlock(&mp_mutex_action);
227 rte_errno = EEXIST;
228 free(entry);
229 return -1;
230 }
231 TAILQ_INSERT_TAIL(&action_entry_list, entry, next);
232 pthread_mutex_unlock(&mp_mutex_action);
233 return 0;
234 }
235
236 void
rte_mp_action_unregister(const char * name)237 rte_mp_action_unregister(const char *name)
238 {
239 struct action_entry *entry;
240 const struct internal_config *internal_conf =
241 eal_get_internal_configuration();
242
243 if (validate_action_name(name) != 0)
244 return;
245
246 if (internal_conf->no_shconf) {
247 EAL_LOG(DEBUG, "No shared files mode enabled, IPC is disabled");
248 return;
249 }
250
251 pthread_mutex_lock(&mp_mutex_action);
252 entry = find_action_entry_by_name(name);
253 if (entry == NULL) {
254 pthread_mutex_unlock(&mp_mutex_action);
255 return;
256 }
257 TAILQ_REMOVE(&action_entry_list, entry, next);
258 pthread_mutex_unlock(&mp_mutex_action);
259 free(entry);
260 }
261
262 static int
read_msg(int fd,struct mp_msg_internal * m,struct sockaddr_un * s)263 read_msg(int fd, struct mp_msg_internal *m, struct sockaddr_un *s)
264 {
265 int msglen;
266 struct iovec iov;
267 struct msghdr msgh;
268 char control[CMSG_SPACE(sizeof(m->msg.fds))];
269 struct cmsghdr *cmsg;
270 int buflen = sizeof(*m) - sizeof(m->msg.fds);
271
272 memset(&msgh, 0, sizeof(msgh));
273 iov.iov_base = m;
274 iov.iov_len = buflen;
275
276 msgh.msg_name = s;
277 msgh.msg_namelen = sizeof(*s);
278 msgh.msg_iov = &iov;
279 msgh.msg_iovlen = 1;
280 msgh.msg_control = control;
281 msgh.msg_controllen = sizeof(control);
282
283 retry:
284 msglen = recvmsg(fd, &msgh, 0);
285
286 /* zero length message means socket was closed */
287 if (msglen == 0)
288 return 0;
289
290 if (msglen < 0) {
291 if (errno == EINTR)
292 goto retry;
293
294 EAL_LOG(ERR, "recvmsg failed, %s", strerror(errno));
295 return -1;
296 }
297
298 if (msglen != buflen || (msgh.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
299 EAL_LOG(ERR, "truncated msg");
300 return -1;
301 }
302
303 /* read auxiliary FDs if any */
304 for (cmsg = CMSG_FIRSTHDR(&msgh); cmsg != NULL;
305 cmsg = CMSG_NXTHDR(&msgh, cmsg)) {
306 if ((cmsg->cmsg_level == SOL_SOCKET) &&
307 (cmsg->cmsg_type == SCM_RIGHTS)) {
308 memcpy(m->msg.fds, CMSG_DATA(cmsg), sizeof(m->msg.fds));
309 break;
310 }
311 }
312 /* sanity-check the response */
313 if (m->msg.num_fds < 0 || m->msg.num_fds > RTE_MP_MAX_FD_NUM) {
314 EAL_LOG(ERR, "invalid number of fd's received");
315 return -1;
316 }
317 if (m->msg.len_param < 0 || m->msg.len_param > RTE_MP_MAX_PARAM_LEN) {
318 EAL_LOG(ERR, "invalid received data length");
319 return -1;
320 }
321 return msglen;
322 }
323
324 static void
cleanup_msg_fds(const struct rte_mp_msg * msg)325 cleanup_msg_fds(const struct rte_mp_msg *msg)
326 {
327 int i;
328
329 for (i = 0; i < msg->num_fds; i++)
330 close(msg->fds[i]);
331 }
332
333 static void
process_msg(struct mp_msg_internal * m,struct sockaddr_un * s)334 process_msg(struct mp_msg_internal *m, struct sockaddr_un *s)
335 {
336 struct pending_request *pending_req;
337 struct action_entry *entry;
338 struct rte_mp_msg *msg = &m->msg;
339 rte_mp_t action = NULL;
340 const struct internal_config *internal_conf =
341 eal_get_internal_configuration();
342
343 EAL_LOG(DEBUG, "msg: %s", msg->name);
344
345 if (m->type == MP_REP || m->type == MP_IGN) {
346 struct pending_request *req = NULL;
347
348 pthread_mutex_lock(&pending_requests.lock);
349 pending_req = find_pending_request(s->sun_path, msg->name);
350 if (pending_req) {
351 memcpy(pending_req->reply, msg, sizeof(*msg));
352 /* -1 indicates that we've been asked to ignore */
353 pending_req->reply_received =
354 m->type == MP_REP ? 1 : -1;
355
356 if (pending_req->type == REQUEST_TYPE_SYNC)
357 pthread_cond_signal(&pending_req->sync.cond);
358 else if (pending_req->type == REQUEST_TYPE_ASYNC)
359 req = async_reply_handle_thread_unsafe(
360 pending_req);
361 } else {
362 EAL_LOG(ERR, "Drop mp reply: %s", msg->name);
363 cleanup_msg_fds(msg);
364 }
365 pthread_mutex_unlock(&pending_requests.lock);
366
367 if (req != NULL)
368 trigger_async_action(req);
369 return;
370 }
371
372 pthread_mutex_lock(&mp_mutex_action);
373 entry = find_action_entry_by_name(msg->name);
374 if (entry != NULL)
375 action = entry->action;
376 pthread_mutex_unlock(&mp_mutex_action);
377
378 if (!action) {
379 if (m->type == MP_REQ && !internal_conf->init_complete) {
380 /* if this is a request, and init is not yet complete,
381 * and callback wasn't registered, we should tell the
382 * requester to ignore our existence because we're not
383 * yet ready to process this request.
384 */
385 struct rte_mp_msg dummy;
386
387 memset(&dummy, 0, sizeof(dummy));
388 strlcpy(dummy.name, msg->name, sizeof(dummy.name));
389 mp_send(&dummy, s->sun_path, MP_IGN);
390 } else {
391 EAL_LOG(ERR, "Cannot find action: %s",
392 msg->name);
393 }
394 cleanup_msg_fds(msg);
395 } else if (action(msg, s->sun_path) < 0) {
396 EAL_LOG(ERR, "Fail to handle message: %s", msg->name);
397 }
398 }
399
400 static uint32_t
mp_handle(void * arg __rte_unused)401 mp_handle(void *arg __rte_unused)
402 {
403 struct mp_msg_internal msg;
404 struct sockaddr_un sa;
405 int fd;
406
407 while ((fd = rte_atomic_load_explicit(&mp_fd, rte_memory_order_relaxed)) >= 0) {
408 int ret;
409
410 ret = read_msg(fd, &msg, &sa);
411 if (ret <= 0)
412 break;
413
414 process_msg(&msg, &sa);
415 }
416
417 return 0;
418 }
419
420 static int
timespec_cmp(const struct timespec * a,const struct timespec * b)421 timespec_cmp(const struct timespec *a, const struct timespec *b)
422 {
423 if (a->tv_sec < b->tv_sec)
424 return -1;
425 if (a->tv_sec > b->tv_sec)
426 return 1;
427 if (a->tv_nsec < b->tv_nsec)
428 return -1;
429 if (a->tv_nsec > b->tv_nsec)
430 return 1;
431 return 0;
432 }
433
434 enum async_action {
435 ACTION_FREE, /**< free the action entry, but don't trigger callback */
436 ACTION_TRIGGER /**< trigger callback, then free action entry */
437 };
438
439 static enum async_action
process_async_request(struct pending_request * sr,const struct timespec * now)440 process_async_request(struct pending_request *sr, const struct timespec *now)
441 {
442 struct async_request_param *param;
443 struct rte_mp_reply *reply;
444 bool timeout, last_msg;
445
446 param = sr->async.param;
447 reply = ¶m->user_reply;
448
449 /* did we timeout? */
450 timeout = timespec_cmp(¶m->end, now) <= 0;
451
452 /* if we received a response, adjust relevant data and copy message. */
453 if (sr->reply_received == 1 && sr->reply) {
454 struct rte_mp_msg *msg, *user_msgs, *tmp;
455
456 msg = sr->reply;
457 user_msgs = reply->msgs;
458
459 tmp = realloc(user_msgs, sizeof(*msg) *
460 (reply->nb_received + 1));
461 if (!tmp) {
462 EAL_LOG(ERR, "Fail to alloc reply for request %s:%s",
463 sr->dst, sr->request->name);
464 /* this entry is going to be removed and its message
465 * dropped, but we don't want to leak memory, so
466 * continue.
467 */
468 } else {
469 user_msgs = tmp;
470 reply->msgs = user_msgs;
471 memcpy(&user_msgs[reply->nb_received],
472 msg, sizeof(*msg));
473 reply->nb_received++;
474 }
475
476 /* mark this request as processed */
477 param->n_responses_processed++;
478 } else if (sr->reply_received == -1) {
479 /* we were asked to ignore this process */
480 reply->nb_sent--;
481 } else if (timeout) {
482 /* count it as processed response, but don't increment
483 * nb_received.
484 */
485 param->n_responses_processed++;
486 }
487
488 free(sr->reply);
489
490 last_msg = param->n_responses_processed == reply->nb_sent;
491
492 return last_msg ? ACTION_TRIGGER : ACTION_FREE;
493 }
494
495 static void
trigger_async_action(struct pending_request * sr)496 trigger_async_action(struct pending_request *sr)
497 {
498 struct async_request_param *param;
499 struct rte_mp_reply *reply;
500
501 param = sr->async.param;
502 reply = ¶m->user_reply;
503
504 param->clb(sr->request, reply);
505
506 /* clean up */
507 free(sr->async.param->user_reply.msgs);
508 free(sr->async.param);
509 free(sr->request);
510 free(sr);
511 }
512
513 static struct pending_request *
async_reply_handle_thread_unsafe(void * arg)514 async_reply_handle_thread_unsafe(void *arg)
515 {
516 struct pending_request *req = (struct pending_request *)arg;
517 enum async_action action;
518 struct timespec ts_now;
519
520 if (clock_gettime(CLOCK_MONOTONIC, &ts_now) < 0) {
521 EAL_LOG(ERR, "Cannot get current time");
522 goto no_trigger;
523 }
524
525 action = process_async_request(req, &ts_now);
526
527 TAILQ_REMOVE(&pending_requests.requests, req, next);
528
529 if (rte_eal_alarm_cancel(async_reply_handle, req) < 0) {
530 /* if we failed to cancel the alarm because it's already in
531 * progress, don't proceed because otherwise we will end up
532 * handling the same message twice.
533 */
534 if (rte_errno == EINPROGRESS) {
535 EAL_LOG(DEBUG, "Request handling is already in progress");
536 goto no_trigger;
537 }
538 EAL_LOG(ERR, "Failed to cancel alarm");
539 }
540
541 if (action == ACTION_TRIGGER)
542 return req;
543 no_trigger:
544 free(req);
545 return NULL;
546 }
547
548 static void
async_reply_handle(void * arg)549 async_reply_handle(void *arg)
550 {
551 struct pending_request *req;
552
553 pthread_mutex_lock(&pending_requests.lock);
554 req = async_reply_handle_thread_unsafe(arg);
555 pthread_mutex_unlock(&pending_requests.lock);
556
557 if (req != NULL)
558 trigger_async_action(req);
559 }
560
561 static int
open_socket_fd(void)562 open_socket_fd(void)
563 {
564 struct sockaddr_un un;
565
566 peer_name[0] = '\0';
567 if (rte_eal_process_type() == RTE_PROC_SECONDARY)
568 snprintf(peer_name, sizeof(peer_name),
569 "%d_%"PRIx64, getpid(), rte_rdtsc());
570
571 mp_fd = socket(AF_UNIX, SOCK_DGRAM, 0);
572 if (mp_fd < 0) {
573 EAL_LOG(ERR, "failed to create unix socket");
574 return -1;
575 }
576
577 memset(&un, 0, sizeof(un));
578 un.sun_family = AF_UNIX;
579
580 create_socket_path(peer_name, un.sun_path, sizeof(un.sun_path));
581
582 unlink(un.sun_path); /* May still exist since last run */
583
584 if (bind(mp_fd, (struct sockaddr *)&un, sizeof(un)) < 0) {
585 EAL_LOG(ERR, "failed to bind %s: %s",
586 un.sun_path, strerror(errno));
587 close(mp_fd);
588 return -1;
589 }
590
591 EAL_LOG(INFO, "Multi-process socket %s", un.sun_path);
592 return mp_fd;
593 }
594
595 static void
close_socket_fd(int fd)596 close_socket_fd(int fd)
597 {
598 char path[PATH_MAX];
599
600 close(fd);
601 create_socket_path(peer_name, path, sizeof(path));
602 unlink(path);
603 }
604
605 int
rte_mp_channel_init(void)606 rte_mp_channel_init(void)
607 {
608 char path[PATH_MAX];
609 int dir_fd;
610 const struct internal_config *internal_conf =
611 eal_get_internal_configuration();
612
613 /* in no shared files mode, we do not have secondary processes support,
614 * so no need to initialize IPC.
615 */
616 if (internal_conf->no_shconf) {
617 EAL_LOG(DEBUG, "No shared files mode enabled, IPC will be disabled");
618 rte_errno = ENOTSUP;
619 return -1;
620 }
621
622 /* create filter path */
623 create_socket_path("*", path, sizeof(path));
624 strlcpy(mp_filter, basename(path), sizeof(mp_filter));
625
626 /* path may have been modified, so recreate it */
627 create_socket_path("*", path, sizeof(path));
628 strlcpy(mp_dir_path, dirname(path), sizeof(mp_dir_path));
629
630 /* lock the directory */
631 dir_fd = open(mp_dir_path, O_RDONLY);
632 if (dir_fd < 0) {
633 EAL_LOG(ERR, "failed to open %s: %s",
634 mp_dir_path, strerror(errno));
635 return -1;
636 }
637
638 if (flock(dir_fd, LOCK_EX)) {
639 EAL_LOG(ERR, "failed to lock %s: %s",
640 mp_dir_path, strerror(errno));
641 close(dir_fd);
642 return -1;
643 }
644
645 if (open_socket_fd() < 0) {
646 close(dir_fd);
647 return -1;
648 }
649
650 if (rte_thread_create_internal_control(&mp_handle_tid, "mp-msg",
651 mp_handle, NULL) < 0) {
652 EAL_LOG(ERR, "failed to create mp thread: %s",
653 strerror(errno));
654 close(dir_fd);
655 close(rte_atomic_exchange_explicit(&mp_fd, -1, rte_memory_order_relaxed));
656 return -1;
657 }
658
659 /* unlock the directory */
660 flock(dir_fd, LOCK_UN);
661 close(dir_fd);
662
663 return 0;
664 }
665
666 void
rte_mp_channel_cleanup(void)667 rte_mp_channel_cleanup(void)
668 {
669 int fd;
670
671 fd = rte_atomic_exchange_explicit(&mp_fd, -1, rte_memory_order_relaxed);
672 if (fd < 0)
673 return;
674
675 pthread_cancel((pthread_t)mp_handle_tid.opaque_id);
676 rte_thread_join(mp_handle_tid, NULL);
677 close_socket_fd(fd);
678 }
679
680 /**
681 * Return -1, as fail to send message and it's caused by the local side.
682 * Return 0, as fail to send message and it's caused by the remote side.
683 * Return 1, as succeed to send message.
684 */
685 static int
send_msg(const char * dst_path,struct rte_mp_msg * msg,int type)686 send_msg(const char *dst_path, struct rte_mp_msg *msg, int type)
687 {
688 int snd;
689 struct iovec iov;
690 struct msghdr msgh;
691 struct cmsghdr *cmsg;
692 struct sockaddr_un dst;
693 struct mp_msg_internal m;
694 int fd_size = msg->num_fds * sizeof(int);
695 char control[CMSG_SPACE(fd_size)];
696
697 m.type = type;
698 memcpy(&m.msg, msg, sizeof(*msg));
699
700 memset(&dst, 0, sizeof(dst));
701 dst.sun_family = AF_UNIX;
702 strlcpy(dst.sun_path, dst_path, sizeof(dst.sun_path));
703
704 memset(&msgh, 0, sizeof(msgh));
705 memset(control, 0, sizeof(control));
706
707 iov.iov_base = &m;
708 iov.iov_len = sizeof(m) - sizeof(msg->fds);
709
710 msgh.msg_name = &dst;
711 msgh.msg_namelen = sizeof(dst);
712 msgh.msg_iov = &iov;
713 msgh.msg_iovlen = 1;
714 msgh.msg_control = control;
715 msgh.msg_controllen = sizeof(control);
716
717 cmsg = CMSG_FIRSTHDR(&msgh);
718 cmsg->cmsg_len = CMSG_LEN(fd_size);
719 cmsg->cmsg_level = SOL_SOCKET;
720 cmsg->cmsg_type = SCM_RIGHTS;
721 memcpy(CMSG_DATA(cmsg), msg->fds, fd_size);
722
723 do {
724 snd = sendmsg(mp_fd, &msgh, 0);
725 } while (snd < 0 && errno == EINTR);
726
727 if (snd < 0) {
728 rte_errno = errno;
729 /* Check if it caused by peer process exits */
730 if (errno == ECONNREFUSED &&
731 rte_eal_process_type() == RTE_PROC_PRIMARY) {
732 unlink(dst_path);
733 return 0;
734 }
735 EAL_LOG(ERR, "failed to send to (%s) due to %s",
736 dst_path, strerror(errno));
737 return -1;
738 }
739
740 return 1;
741 }
742
743 static int
mp_send(struct rte_mp_msg * msg,const char * peer,int type)744 mp_send(struct rte_mp_msg *msg, const char *peer, int type)
745 {
746 int dir_fd, ret = 0;
747 DIR *mp_dir;
748 struct dirent *ent;
749
750 if (!peer && (rte_eal_process_type() == RTE_PROC_SECONDARY))
751 peer = eal_mp_socket_path();
752
753 if (peer) {
754 if (send_msg(peer, msg, type) < 0)
755 return -1;
756 else
757 return 0;
758 }
759
760 /* broadcast to all secondary processes */
761 mp_dir = opendir(mp_dir_path);
762 if (!mp_dir) {
763 EAL_LOG(ERR, "Unable to open directory %s",
764 mp_dir_path);
765 rte_errno = errno;
766 return -1;
767 }
768
769 dir_fd = dirfd(mp_dir);
770 /* lock the directory to prevent processes spinning up while we send */
771 if (flock(dir_fd, LOCK_SH)) {
772 EAL_LOG(ERR, "Unable to lock directory %s",
773 mp_dir_path);
774 rte_errno = errno;
775 closedir(mp_dir);
776 return -1;
777 }
778
779 while ((ent = readdir(mp_dir))) {
780 char path[PATH_MAX];
781
782 if (fnmatch(mp_filter, ent->d_name, 0) != 0)
783 continue;
784
785 snprintf(path, sizeof(path), "%s/%s", mp_dir_path,
786 ent->d_name);
787 if (send_msg(path, msg, type) < 0)
788 ret = -1;
789 }
790 /* unlock the dir */
791 flock(dir_fd, LOCK_UN);
792
793 /* dir_fd automatically closed on closedir */
794 closedir(mp_dir);
795 return ret;
796 }
797
798 static int
check_input(const struct rte_mp_msg * msg)799 check_input(const struct rte_mp_msg *msg)
800 {
801 if (msg == NULL) {
802 EAL_LOG(ERR, "Msg cannot be NULL");
803 rte_errno = EINVAL;
804 return -1;
805 }
806
807 if (validate_action_name(msg->name) != 0)
808 return -1;
809
810 if (msg->len_param < 0) {
811 EAL_LOG(ERR, "Message data length is negative");
812 rte_errno = EINVAL;
813 return -1;
814 }
815
816 if (msg->num_fds < 0) {
817 EAL_LOG(ERR, "Number of fd's is negative");
818 rte_errno = EINVAL;
819 return -1;
820 }
821
822 if (msg->len_param > RTE_MP_MAX_PARAM_LEN) {
823 EAL_LOG(ERR, "Message data is too long");
824 rte_errno = E2BIG;
825 return -1;
826 }
827
828 if (msg->num_fds > RTE_MP_MAX_FD_NUM) {
829 EAL_LOG(ERR, "Cannot send more than %d FDs",
830 RTE_MP_MAX_FD_NUM);
831 rte_errno = E2BIG;
832 return -1;
833 }
834
835 return 0;
836 }
837
838 int
rte_mp_sendmsg(struct rte_mp_msg * msg)839 rte_mp_sendmsg(struct rte_mp_msg *msg)
840 {
841 const struct internal_config *internal_conf =
842 eal_get_internal_configuration();
843
844 if (check_input(msg) != 0)
845 return -1;
846
847 if (internal_conf->no_shconf) {
848 EAL_LOG(DEBUG, "No shared files mode enabled, IPC is disabled");
849 rte_errno = ENOTSUP;
850 return -1;
851 }
852
853 EAL_LOG(DEBUG, "sendmsg: %s", msg->name);
854 return mp_send(msg, NULL, MP_MSG);
855 }
856
857 static int
mp_request_async(const char * dst,struct rte_mp_msg * req,struct async_request_param * param,const struct timespec * ts)858 mp_request_async(const char *dst, struct rte_mp_msg *req,
859 struct async_request_param *param, const struct timespec *ts)
860 {
861 struct rte_mp_msg *reply_msg;
862 struct pending_request *pending_req, *exist;
863 int ret = -1;
864
865 pending_req = calloc(1, sizeof(*pending_req));
866 reply_msg = calloc(1, sizeof(*reply_msg));
867 if (pending_req == NULL || reply_msg == NULL) {
868 EAL_LOG(ERR, "Could not allocate space for sync request");
869 rte_errno = ENOMEM;
870 ret = -1;
871 goto fail;
872 }
873
874 pending_req->type = REQUEST_TYPE_ASYNC;
875 strlcpy(pending_req->dst, dst, sizeof(pending_req->dst));
876 pending_req->request = req;
877 pending_req->reply = reply_msg;
878 pending_req->async.param = param;
879
880 /* queue already locked by caller */
881
882 exist = find_pending_request(dst, req->name);
883 if (exist) {
884 EAL_LOG(ERR, "A pending request %s:%s", dst, req->name);
885 rte_errno = EEXIST;
886 ret = -1;
887 goto fail;
888 }
889
890 ret = send_msg(dst, req, MP_REQ);
891 if (ret < 0) {
892 EAL_LOG(ERR, "Fail to send request %s:%s",
893 dst, req->name);
894 ret = -1;
895 goto fail;
896 } else if (ret == 0) {
897 ret = 0;
898 goto fail;
899 }
900 param->user_reply.nb_sent++;
901
902 /* if alarm set fails, we simply ignore the reply */
903 if (rte_eal_alarm_set(ts->tv_sec * 1000000 + ts->tv_nsec / 1000,
904 async_reply_handle, pending_req) < 0) {
905 EAL_LOG(ERR, "Fail to set alarm for request %s:%s",
906 dst, req->name);
907 ret = -1;
908 goto fail;
909 }
910 TAILQ_INSERT_TAIL(&pending_requests.requests, pending_req, next);
911
912 return 0;
913 fail:
914 free(pending_req);
915 free(reply_msg);
916 return ret;
917 }
918
919 static int
mp_request_sync(const char * dst,struct rte_mp_msg * req,struct rte_mp_reply * reply,const struct timespec * ts)920 mp_request_sync(const char *dst, struct rte_mp_msg *req,
921 struct rte_mp_reply *reply, const struct timespec *ts)
922 {
923 int ret;
924 pthread_condattr_t attr;
925 struct rte_mp_msg msg, *tmp;
926 struct pending_request pending_req, *exist;
927
928 pending_req.type = REQUEST_TYPE_SYNC;
929 pending_req.reply_received = 0;
930 strlcpy(pending_req.dst, dst, sizeof(pending_req.dst));
931 pending_req.request = req;
932 pending_req.reply = &msg;
933 pthread_condattr_init(&attr);
934 pthread_condattr_setclock(&attr, CLOCK_MONOTONIC);
935 pthread_cond_init(&pending_req.sync.cond, &attr);
936
937 exist = find_pending_request(dst, req->name);
938 if (exist) {
939 EAL_LOG(ERR, "A pending request %s:%s", dst, req->name);
940 rte_errno = EEXIST;
941 return -1;
942 }
943
944 ret = send_msg(dst, req, MP_REQ);
945 if (ret < 0) {
946 EAL_LOG(ERR, "Fail to send request %s:%s",
947 dst, req->name);
948 return -1;
949 } else if (ret == 0)
950 return 0;
951
952 TAILQ_INSERT_TAIL(&pending_requests.requests, &pending_req, next);
953
954 reply->nb_sent++;
955
956 do {
957 ret = pthread_cond_timedwait(&pending_req.sync.cond,
958 &pending_requests.lock, ts);
959 } while (ret != 0 && ret != ETIMEDOUT);
960
961 TAILQ_REMOVE(&pending_requests.requests, &pending_req, next);
962
963 if (pending_req.reply_received == 0) {
964 EAL_LOG(ERR, "Fail to recv reply for request %s:%s",
965 dst, req->name);
966 rte_errno = ETIMEDOUT;
967 return -1;
968 }
969 if (pending_req.reply_received == -1) {
970 EAL_LOG(DEBUG, "Asked to ignore response");
971 /* not receiving this message is not an error, so decrement
972 * number of sent messages
973 */
974 reply->nb_sent--;
975 return 0;
976 }
977
978 tmp = realloc(reply->msgs, sizeof(msg) * (reply->nb_received + 1));
979 if (!tmp) {
980 EAL_LOG(ERR, "Fail to alloc reply for request %s:%s",
981 dst, req->name);
982 rte_errno = ENOMEM;
983 return -1;
984 }
985 memcpy(&tmp[reply->nb_received], &msg, sizeof(msg));
986 reply->msgs = tmp;
987 reply->nb_received++;
988 return 0;
989 }
990
991 int
rte_mp_request_sync(struct rte_mp_msg * req,struct rte_mp_reply * reply,const struct timespec * ts)992 rte_mp_request_sync(struct rte_mp_msg *req, struct rte_mp_reply *reply,
993 const struct timespec *ts)
994 {
995 int dir_fd, ret = -1;
996 DIR *mp_dir;
997 struct dirent *ent;
998 struct timespec now, end;
999 const struct internal_config *internal_conf =
1000 eal_get_internal_configuration();
1001
1002 EAL_LOG(DEBUG, "request: %s", req->name);
1003
1004 reply->nb_sent = 0;
1005 reply->nb_received = 0;
1006 reply->msgs = NULL;
1007
1008 if (check_input(req) != 0)
1009 goto end;
1010
1011 if (internal_conf->no_shconf) {
1012 EAL_LOG(DEBUG, "No shared files mode enabled, IPC is disabled");
1013 rte_errno = ENOTSUP;
1014 return -1;
1015 }
1016
1017 if (clock_gettime(CLOCK_MONOTONIC, &now) < 0) {
1018 EAL_LOG(ERR, "Failed to get current time");
1019 rte_errno = errno;
1020 goto end;
1021 }
1022
1023 end.tv_nsec = (now.tv_nsec + ts->tv_nsec) % 1000000000;
1024 end.tv_sec = now.tv_sec + ts->tv_sec +
1025 (now.tv_nsec + ts->tv_nsec) / 1000000000;
1026
1027 /* for secondary process, send request to the primary process only */
1028 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1029 pthread_mutex_lock(&pending_requests.lock);
1030 ret = mp_request_sync(eal_mp_socket_path(), req, reply, &end);
1031 pthread_mutex_unlock(&pending_requests.lock);
1032 goto end;
1033 }
1034
1035 /* for primary process, broadcast request, and collect reply 1 by 1 */
1036 mp_dir = opendir(mp_dir_path);
1037 if (!mp_dir) {
1038 EAL_LOG(ERR, "Unable to open directory %s", mp_dir_path);
1039 rte_errno = errno;
1040 goto end;
1041 }
1042
1043 dir_fd = dirfd(mp_dir);
1044 /* lock the directory to prevent processes spinning up while we send */
1045 if (flock(dir_fd, LOCK_SH)) {
1046 EAL_LOG(ERR, "Unable to lock directory %s",
1047 mp_dir_path);
1048 rte_errno = errno;
1049 goto close_end;
1050 }
1051
1052 pthread_mutex_lock(&pending_requests.lock);
1053 while ((ent = readdir(mp_dir))) {
1054 char path[PATH_MAX];
1055
1056 if (fnmatch(mp_filter, ent->d_name, 0) != 0)
1057 continue;
1058
1059 snprintf(path, sizeof(path), "%s/%s", mp_dir_path,
1060 ent->d_name);
1061
1062 /* unlocks the mutex while waiting for response,
1063 * locks on receive
1064 */
1065 if (mp_request_sync(path, req, reply, &end))
1066 goto unlock_end;
1067 }
1068 ret = 0;
1069
1070 unlock_end:
1071 pthread_mutex_unlock(&pending_requests.lock);
1072 /* unlock the directory */
1073 flock(dir_fd, LOCK_UN);
1074
1075 close_end:
1076 /* dir_fd automatically closed on closedir */
1077 closedir(mp_dir);
1078
1079 end:
1080 if (ret) {
1081 free(reply->msgs);
1082 reply->nb_received = 0;
1083 reply->msgs = NULL;
1084 }
1085 return ret;
1086 }
1087
1088 int
rte_mp_request_async(struct rte_mp_msg * req,const struct timespec * ts,rte_mp_async_reply_t clb)1089 rte_mp_request_async(struct rte_mp_msg *req, const struct timespec *ts,
1090 rte_mp_async_reply_t clb)
1091 {
1092 struct rte_mp_msg *copy;
1093 struct pending_request *dummy;
1094 struct async_request_param *param;
1095 struct rte_mp_reply *reply;
1096 int dir_fd, ret = 0;
1097 DIR *mp_dir;
1098 struct dirent *ent;
1099 struct timespec now;
1100 struct timespec *end;
1101 bool dummy_used = false;
1102 const struct internal_config *internal_conf =
1103 eal_get_internal_configuration();
1104
1105 EAL_LOG(DEBUG, "request: %s", req->name);
1106
1107 if (check_input(req) != 0)
1108 return -1;
1109
1110 if (internal_conf->no_shconf) {
1111 EAL_LOG(DEBUG, "No shared files mode enabled, IPC is disabled");
1112 rte_errno = ENOTSUP;
1113 return -1;
1114 }
1115
1116 if (clock_gettime(CLOCK_MONOTONIC, &now) < 0) {
1117 EAL_LOG(ERR, "Failed to get current time");
1118 rte_errno = errno;
1119 return -1;
1120 }
1121 copy = calloc(1, sizeof(*copy));
1122 dummy = calloc(1, sizeof(*dummy));
1123 param = calloc(1, sizeof(*param));
1124 if (copy == NULL || dummy == NULL || param == NULL) {
1125 EAL_LOG(ERR, "Failed to allocate memory for async reply");
1126 rte_errno = ENOMEM;
1127 goto fail;
1128 }
1129
1130 /* copy message */
1131 memcpy(copy, req, sizeof(*copy));
1132
1133 param->n_responses_processed = 0;
1134 param->clb = clb;
1135 end = ¶m->end;
1136 reply = ¶m->user_reply;
1137
1138 end->tv_nsec = (now.tv_nsec + ts->tv_nsec) % 1000000000;
1139 end->tv_sec = now.tv_sec + ts->tv_sec +
1140 (now.tv_nsec + ts->tv_nsec) / 1000000000;
1141 reply->nb_sent = 0;
1142 reply->nb_received = 0;
1143 reply->msgs = NULL;
1144
1145 /* we have to lock the request queue here, as we will be adding a bunch
1146 * of requests to the queue at once, and some of the replies may arrive
1147 * before we add all of the requests to the queue.
1148 */
1149 pthread_mutex_lock(&pending_requests.lock);
1150
1151 /* we have to ensure that callback gets triggered even if we don't send
1152 * anything, therefore earlier we have allocated a dummy request. fill
1153 * it, and put it on the queue if we don't send any requests.
1154 */
1155 dummy->type = REQUEST_TYPE_ASYNC;
1156 dummy->request = copy;
1157 dummy->reply = NULL;
1158 dummy->async.param = param;
1159 dummy->reply_received = 1; /* short-circuit the timeout */
1160
1161 /* for secondary process, send request to the primary process only */
1162 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1163 ret = mp_request_async(eal_mp_socket_path(), copy, param, ts);
1164
1165 /* if we didn't send anything, put dummy request on the queue */
1166 if (ret == 0 && reply->nb_sent == 0) {
1167 TAILQ_INSERT_TAIL(&pending_requests.requests, dummy,
1168 next);
1169 dummy_used = true;
1170 }
1171
1172 pthread_mutex_unlock(&pending_requests.lock);
1173
1174 /* if we couldn't send anything, clean up */
1175 if (ret != 0)
1176 goto fail;
1177 return 0;
1178 }
1179
1180 /* for primary process, broadcast request */
1181 mp_dir = opendir(mp_dir_path);
1182 if (!mp_dir) {
1183 EAL_LOG(ERR, "Unable to open directory %s", mp_dir_path);
1184 rte_errno = errno;
1185 goto unlock_fail;
1186 }
1187 dir_fd = dirfd(mp_dir);
1188
1189 /* lock the directory to prevent processes spinning up while we send */
1190 if (flock(dir_fd, LOCK_SH)) {
1191 EAL_LOG(ERR, "Unable to lock directory %s",
1192 mp_dir_path);
1193 rte_errno = errno;
1194 goto closedir_fail;
1195 }
1196
1197 while ((ent = readdir(mp_dir))) {
1198 char path[PATH_MAX];
1199
1200 if (fnmatch(mp_filter, ent->d_name, 0) != 0)
1201 continue;
1202
1203 snprintf(path, sizeof(path), "%s/%s", mp_dir_path,
1204 ent->d_name);
1205
1206 if (mp_request_async(path, copy, param, ts))
1207 ret = -1;
1208 }
1209 /* if we didn't send anything, put dummy request on the queue */
1210 if (ret == 0 && reply->nb_sent == 0) {
1211 TAILQ_INSERT_HEAD(&pending_requests.requests, dummy, next);
1212 dummy_used = true;
1213 }
1214
1215 /* finally, unlock the queue */
1216 pthread_mutex_unlock(&pending_requests.lock);
1217
1218 /* unlock the directory */
1219 flock(dir_fd, LOCK_UN);
1220
1221 /* dir_fd automatically closed on closedir */
1222 closedir(mp_dir);
1223
1224 /* if dummy was unused, free it */
1225 if (!dummy_used)
1226 free(dummy);
1227
1228 return ret;
1229 closedir_fail:
1230 closedir(mp_dir);
1231 unlock_fail:
1232 pthread_mutex_unlock(&pending_requests.lock);
1233 fail:
1234 free(dummy);
1235 free(param);
1236 free(copy);
1237 return -1;
1238 }
1239
1240 int
rte_mp_reply(struct rte_mp_msg * msg,const char * peer)1241 rte_mp_reply(struct rte_mp_msg *msg, const char *peer)
1242 {
1243 EAL_LOG(DEBUG, "reply: %s", msg->name);
1244 const struct internal_config *internal_conf =
1245 eal_get_internal_configuration();
1246
1247 if (check_input(msg) != 0)
1248 return -1;
1249
1250 if (peer == NULL) {
1251 EAL_LOG(ERR, "peer is not specified");
1252 rte_errno = EINVAL;
1253 return -1;
1254 }
1255
1256 if (internal_conf->no_shconf) {
1257 EAL_LOG(DEBUG, "No shared files mode enabled, IPC is disabled");
1258 return 0;
1259 }
1260
1261 return mp_send(msg, peer, MP_REP);
1262 }
1263
1264 /* Internally, the status of the mp feature is represented as a three-state:
1265 * - "unknown" as long as no secondary process attached to a primary process
1266 * and there was no call to rte_mp_disable yet,
1267 * - "enabled" as soon as a secondary process attaches to a primary process,
1268 * - "disabled" when a primary process successfully called rte_mp_disable,
1269 */
1270 enum mp_status {
1271 MP_STATUS_UNKNOWN,
1272 MP_STATUS_DISABLED,
1273 MP_STATUS_ENABLED,
1274 };
1275
1276 static bool
set_mp_status(enum mp_status status)1277 set_mp_status(enum mp_status status)
1278 {
1279 struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
1280 uint8_t expected;
1281 uint8_t desired;
1282
1283 expected = MP_STATUS_UNKNOWN;
1284 desired = status;
1285 if (rte_atomic_compare_exchange_strong_explicit(&mcfg->mp_status, &expected, desired,
1286 rte_memory_order_relaxed, rte_memory_order_relaxed))
1287 return true;
1288
1289 return rte_atomic_load_explicit(&mcfg->mp_status, rte_memory_order_relaxed) == desired;
1290 }
1291
1292 bool
rte_mp_disable(void)1293 rte_mp_disable(void)
1294 {
1295 return set_mp_status(MP_STATUS_DISABLED);
1296 }
1297
1298 bool
__rte_mp_enable(void)1299 __rte_mp_enable(void)
1300 {
1301 return set_mp_status(MP_STATUS_ENABLED);
1302 }
1303