4 * Copyright (c) 2003-2008 Fabrice Bellard
5 * Copyright (c) 2012-2014 Cisco Systems
7 * Permission is hereby granted, free of charge, to any person obtaining a copy
8 * of this software and associated documentation files (the "Software"), to deal
9 * in the Software without restriction, including without limitation the rights
10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11 * copies of the Software, and to permit persons to whom the Software is
12 * furnished to do so, subject to the following conditions:
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
26 #include "qemu/osdep.h"
31 #include "qapi/error.h"
32 #include "qemu/error-report.h"
33 #include "qemu/option.h"
34 #include "qemu/sockets.h"
36 #include "qemu/main-loop.h"
37 #include "qemu/memalign.h"
39 /* The buffer size needs to be investigated for optimum numbers and
40 * optimum means of paging in on different systems. This size is
41 * chosen to be sufficient to accommodate one packet with some headers
44 #define BUFFER_ALIGN sysconf(_SC_PAGESIZE)
45 #define BUFFER_SIZE 2048
47 #define MAX_L2TPV3_MSGCNT 64
48 #define MAX_L2TPV3_IOVCNT (MAX_L2TPV3_MSGCNT * IOVSIZE)
50 /* Header set to 0x30000 signifies a data packet */
52 #define L2TPV3_DATA_PACKET 0x30000
54 /* IANA-assigned IP protocol ID for L2TPv3 */
57 #define IPPROTO_L2TP 0x73
60 typedef struct NetL2TPV3State {
65 * these are used for xmit - that happens packet a time
66 * and for first sign of life packet (easier to parse that once)
73 * these are used for receive - try to "eat" up to 32 packets at a time
76 struct mmsghdr *msgvec;
82 struct sockaddr_storage *dgram_dst;
97 * DOS avoidance in error handling
100 bool header_mismatch;
103 * Ring buffer handling
111 * Precomputed offsets
115 uint32_t cookie_offset;
116 uint32_t counter_offset;
117 uint32_t session_offset;
135 static void net_l2tpv3_send(void *opaque);
136 static void l2tpv3_writable(void *opaque);
138 static void l2tpv3_update_fd_handler(NetL2TPV3State *s)
140 qemu_set_fd_handler(s->fd,
141 s->read_poll ? net_l2tpv3_send : NULL,
142 s->write_poll ? l2tpv3_writable : NULL,
146 static void l2tpv3_read_poll(NetL2TPV3State *s, bool enable)
148 if (s->read_poll != enable) {
149 s->read_poll = enable;
150 l2tpv3_update_fd_handler(s);
154 static void l2tpv3_write_poll(NetL2TPV3State *s, bool enable)
156 if (s->write_poll != enable) {
157 s->write_poll = enable;
158 l2tpv3_update_fd_handler(s);
162 static void l2tpv3_writable(void *opaque)
164 NetL2TPV3State *s = opaque;
165 l2tpv3_write_poll(s, false);
166 qemu_flush_queued_packets(&s->nc);
169 static void l2tpv3_send_completed(NetClientState *nc, ssize_t len)
171 NetL2TPV3State *s = DO_UPCAST(NetL2TPV3State, nc, nc);
172 l2tpv3_read_poll(s, true);
175 static void l2tpv3_poll(NetClientState *nc, bool enable)
177 NetL2TPV3State *s = DO_UPCAST(NetL2TPV3State, nc, nc);
178 l2tpv3_write_poll(s, enable);
179 l2tpv3_read_poll(s, enable);
182 static void l2tpv3_form_header(NetL2TPV3State *s)
187 stl_be_p((uint32_t *) s->header_buf, L2TPV3_DATA_PACKET);
190 (uint32_t *) (s->header_buf + s->session_offset),
194 if (s->cookie_is_64) {
196 (uint64_t *)(s->header_buf + s->cookie_offset),
201 (uint32_t *) (s->header_buf + s->cookie_offset),
206 if (s->has_counter) {
207 counter = (uint32_t *)(s->header_buf + s->counter_offset);
208 if (s->pin_counter) {
211 stl_be_p(counter, ++s->counter);
216 static ssize_t net_l2tpv3_receive_dgram_iov(NetClientState *nc,
217 const struct iovec *iov,
220 NetL2TPV3State *s = DO_UPCAST(NetL2TPV3State, nc, nc);
222 struct msghdr message;
225 if (iovcnt > MAX_L2TPV3_IOVCNT - 1) {
227 "iovec too long %d > %d, change l2tpv3.h",
228 iovcnt, MAX_L2TPV3_IOVCNT
232 l2tpv3_form_header(s);
233 memcpy(s->vec + 1, iov, iovcnt * sizeof(struct iovec));
234 s->vec->iov_base = s->header_buf;
235 s->vec->iov_len = s->offset;
236 message.msg_name = s->dgram_dst;
237 message.msg_namelen = s->dst_size;
238 message.msg_iov = s->vec;
239 message.msg_iovlen = iovcnt + 1;
240 message.msg_control = NULL;
241 message.msg_controllen = 0;
242 message.msg_flags = 0;
244 ret = sendmsg(s->fd, &message, 0);
245 } while ((ret == -1) && (errno == EINTR));
248 } else if (ret == 0) {
249 /* belt and braces - should not occur on DGRAM
250 * we should get an error and never a 0 send
252 ret = iov_size(iov, iovcnt);
254 /* signal upper layer that socket buffer is full */
256 if (ret == -EAGAIN || ret == -ENOBUFS) {
257 l2tpv3_write_poll(s, true);
264 static ssize_t net_l2tpv3_receive_dgram(NetClientState *nc,
268 NetL2TPV3State *s = DO_UPCAST(NetL2TPV3State, nc, nc);
271 struct msghdr message;
274 l2tpv3_form_header(s);
276 vec->iov_base = s->header_buf;
277 vec->iov_len = s->offset;
279 vec->iov_base = (void *) buf;
281 message.msg_name = s->dgram_dst;
282 message.msg_namelen = s->dst_size;
283 message.msg_iov = s->vec;
284 message.msg_iovlen = 2;
285 message.msg_control = NULL;
286 message.msg_controllen = 0;
287 message.msg_flags = 0;
289 ret = sendmsg(s->fd, &message, 0);
290 } while ((ret == -1) && (errno == EINTR));
293 } else if (ret == 0) {
294 /* belt and braces - should not occur on DGRAM
295 * we should get an error and never a 0 send
300 if (ret == -EAGAIN || ret == -ENOBUFS) {
301 /* signal upper layer that socket buffer is full */
302 l2tpv3_write_poll(s, true);
309 static int l2tpv3_verify_header(NetL2TPV3State *s, uint8_t *buf)
315 if ((!s->udp) && (!s->ipv6)) {
316 buf += sizeof(struct iphdr) /* fix for ipv4 raw */;
319 /* we do not do a strict check for "data" packets as per
320 * the RFC spec because the pure IP spec does not have
325 if (s->cookie_is_64) {
326 cookie = ldq_be_p(buf + s->cookie_offset);
328 cookie = ldl_be_p(buf + s->cookie_offset) & 0xffffffffULL;
330 if (cookie != s->rx_cookie) {
331 if (!s->header_mismatch) {
332 error_report("unknown cookie id");
337 session = (uint32_t *) (buf + s->session_offset);
338 if (ldl_be_p(session) != s->rx_session) {
339 if (!s->header_mismatch) {
340 error_report("session mismatch");
347 static void net_l2tpv3_process_queue(NetL2TPV3State *s)
353 struct mmsghdr *msgvec;
355 /* go into ring mode only if there is a "pending" tail */
356 if (s->queue_depth > 0) {
358 msgvec = s->msgvec + s->queue_tail;
359 if (msgvec->msg_len > 0) {
360 data_size = msgvec->msg_len - s->header_size;
361 vec = msgvec->msg_hdr.msg_iov;
362 if ((data_size > 0) &&
363 (l2tpv3_verify_header(s, vec->iov_base) == 0)) {
365 /* Use the legacy delivery for now, we will
366 * switch to using our own ring as a queueing mechanism
369 size = qemu_send_packet_async(
373 l2tpv3_send_completed
376 l2tpv3_read_poll(s, false);
381 if (!s->header_mismatch) {
382 /* report error only once */
383 error_report("l2tpv3 header verification failed");
384 s->header_mismatch = true;
390 s->queue_tail = (s->queue_tail + 1) % MAX_L2TPV3_MSGCNT;
393 (s->queue_depth > 0) &&
394 qemu_can_send_packet(&s->nc) &&
395 ((size > 0) || bad_read)
400 static void net_l2tpv3_send(void *opaque)
402 NetL2TPV3State *s = opaque;
403 int target_count, count;
404 struct mmsghdr *msgvec;
406 /* go into ring mode only if there is a "pending" tail */
408 if (s->queue_depth) {
410 /* The ring buffer we use has variable intake
411 * count of how much we can read varies - adjust accordingly
414 target_count = MAX_L2TPV3_MSGCNT - s->queue_depth;
416 /* Ensure we do not overrun the ring when we have
417 * a lot of enqueued packets
420 if (s->queue_head + target_count > MAX_L2TPV3_MSGCNT) {
421 target_count = MAX_L2TPV3_MSGCNT - s->queue_head;
425 /* we do not have any pending packets - we can use
426 * the whole message vector linearly instead of using
432 target_count = MAX_L2TPV3_MSGCNT;
435 msgvec = s->msgvec + s->queue_head;
436 if (target_count > 0) {
441 target_count, MSG_DONTWAIT, NULL);
442 } while ((count == -1) && (errno == EINTR));
444 /* Recv error - we still need to flush packets here,
445 * (re)set queue head to current position
449 s->queue_head = (s->queue_head + count) % MAX_L2TPV3_MSGCNT;
450 s->queue_depth += count;
452 net_l2tpv3_process_queue(s);
455 static void destroy_vector(struct mmsghdr *msgvec, int count, int iovcount)
459 struct mmsghdr *cleanup = msgvec;
461 for (i = 0; i < count; i++) {
462 if (cleanup->msg_hdr.msg_iov) {
463 iov = cleanup->msg_hdr.msg_iov;
464 for (j = 0; j < iovcount; j++) {
465 g_free(iov->iov_base);
468 g_free(cleanup->msg_hdr.msg_iov);
476 static struct mmsghdr *build_l2tpv3_vector(NetL2TPV3State *s, int count)
480 struct mmsghdr *msgvec, *result;
482 msgvec = g_new(struct mmsghdr, count);
484 for (i = 0; i < count ; i++) {
485 msgvec->msg_hdr.msg_name = NULL;
486 msgvec->msg_hdr.msg_namelen = 0;
487 iov = g_new(struct iovec, IOVSIZE);
488 msgvec->msg_hdr.msg_iov = iov;
489 iov->iov_base = g_malloc(s->header_size);
490 iov->iov_len = s->header_size;
492 iov->iov_base = qemu_memalign(BUFFER_ALIGN, BUFFER_SIZE);
493 iov->iov_len = BUFFER_SIZE;
494 msgvec->msg_hdr.msg_iovlen = 2;
495 msgvec->msg_hdr.msg_control = NULL;
496 msgvec->msg_hdr.msg_controllen = 0;
497 msgvec->msg_hdr.msg_flags = 0;
503 static void net_l2tpv3_cleanup(NetClientState *nc)
505 NetL2TPV3State *s = DO_UPCAST(NetL2TPV3State, nc, nc);
506 qemu_purge_queued_packets(nc);
507 l2tpv3_read_poll(s, false);
508 l2tpv3_write_poll(s, false);
512 destroy_vector(s->msgvec, MAX_L2TPV3_MSGCNT, IOVSIZE);
514 g_free(s->header_buf);
515 g_free(s->dgram_dst);
518 static NetClientInfo net_l2tpv3_info = {
519 .type = NET_CLIENT_DRIVER_L2TPV3,
520 .size = sizeof(NetL2TPV3State),
521 .receive = net_l2tpv3_receive_dgram,
522 .receive_iov = net_l2tpv3_receive_dgram_iov,
524 .cleanup = net_l2tpv3_cleanup,
527 int net_init_l2tpv3(const Netdev *netdev,
529 NetClientState *peer, Error **errp)
531 const NetdevL2TPv3Options *l2tpv3;
535 struct addrinfo hints;
536 struct addrinfo *result = NULL;
537 char *srcport, *dstport;
539 nc = qemu_new_net_client(&net_l2tpv3_info, peer, "l2tpv3", name);
541 s = DO_UPCAST(NetL2TPV3State, nc, nc);
545 s->header_mismatch = false;
547 assert(netdev->type == NET_CLIENT_DRIVER_L2TPV3);
548 l2tpv3 = &netdev->u.l2tpv3;
550 if (l2tpv3->has_ipv6 && l2tpv3->ipv6) {
551 s->ipv6 = l2tpv3->ipv6;
556 if ((l2tpv3->has_offset) && (l2tpv3->offset > 256)) {
557 error_setg(errp, "offset must be less than 256 bytes");
561 if (l2tpv3->has_rxcookie || l2tpv3->has_txcookie) {
562 if (l2tpv3->has_rxcookie && l2tpv3->has_txcookie) {
566 "require both 'rxcookie' and 'txcookie' or neither");
573 if (l2tpv3->has_cookie64 || l2tpv3->cookie64) {
574 s->cookie_is_64 = true;
576 s->cookie_is_64 = false;
579 if (l2tpv3->has_udp && l2tpv3->udp) {
581 if (!(l2tpv3->srcport && l2tpv3->dstport)) {
582 error_setg(errp, "need both src and dst port for udp");
585 srcport = l2tpv3->srcport;
586 dstport = l2tpv3->dstport;
596 s->session_offset = 0;
597 s->cookie_offset = 4;
598 s->counter_offset = 4;
600 s->tx_session = l2tpv3->txsession;
601 if (l2tpv3->has_rxsession) {
602 s->rx_session = l2tpv3->rxsession;
604 s->rx_session = s->tx_session;
608 s->rx_cookie = l2tpv3->rxcookie;
609 s->tx_cookie = l2tpv3->txcookie;
610 if (s->cookie_is_64 == true) {
613 s->counter_offset += 8;
617 s->counter_offset += 4;
621 memset(&hints, 0, sizeof(hints));
624 hints.ai_family = AF_INET6;
626 hints.ai_family = AF_INET;
629 hints.ai_socktype = SOCK_DGRAM;
630 hints.ai_protocol = 0;
632 s->counter_offset += 4;
633 s->session_offset += 4;
634 s->cookie_offset += 4;
636 hints.ai_socktype = SOCK_RAW;
637 hints.ai_protocol = IPPROTO_L2TP;
640 gairet = getaddrinfo(l2tpv3->src, srcport, &hints, &result);
642 if ((gairet != 0) || (result == NULL)) {
643 error_setg(errp, "could not resolve src, errno = %s",
644 gai_strerror(gairet));
647 fd = socket(result->ai_family, result->ai_socktype, result->ai_protocol);
650 error_setg(errp, "socket creation failed, errno = %d",
654 if (bind(fd, (struct sockaddr *) result->ai_addr, result->ai_addrlen)) {
655 error_setg(errp, "could not bind socket err=%i", errno);
659 freeaddrinfo(result);
661 memset(&hints, 0, sizeof(hints));
664 hints.ai_family = AF_INET6;
666 hints.ai_family = AF_INET;
669 hints.ai_socktype = SOCK_DGRAM;
670 hints.ai_protocol = 0;
672 hints.ai_socktype = SOCK_RAW;
673 hints.ai_protocol = IPPROTO_L2TP;
677 gairet = getaddrinfo(l2tpv3->dst, dstport, &hints, &result);
678 if ((gairet != 0) || (result == NULL)) {
679 error_setg(errp, "could not resolve dst, error = %s",
680 gai_strerror(gairet));
684 s->dgram_dst = g_new0(struct sockaddr_storage, 1);
685 memcpy(s->dgram_dst, result->ai_addr, result->ai_addrlen);
686 s->dst_size = result->ai_addrlen;
688 freeaddrinfo(result);
690 if (l2tpv3->has_counter && l2tpv3->counter) {
691 s->has_counter = true;
694 s->has_counter = false;
697 if (l2tpv3->has_pincounter && l2tpv3->pincounter) {
698 s->has_counter = true; /* pin counter implies that there is counter */
699 s->pin_counter = true;
701 s->pin_counter = false;
704 if (l2tpv3->has_offset) {
706 s->offset += l2tpv3->offset;
709 if ((s->ipv6) || (s->udp)) {
710 s->header_size = s->offset;
712 s->header_size = s->offset + sizeof(struct iphdr);
715 s->msgvec = build_l2tpv3_vector(s, MAX_L2TPV3_MSGCNT);
716 s->vec = g_new(struct iovec, MAX_L2TPV3_IOVCNT);
717 s->header_buf = g_malloc(s->header_size);
719 qemu_socket_set_nonblock(fd);
724 l2tpv3_read_poll(s, true);
726 qemu_set_info_str(&s->nc, "l2tpv3: connected");
729 qemu_del_net_client(nc);
734 freeaddrinfo(result);