Merge remote-tracking branch 'remotes/gonglei/tags/bootdevice-next-20141222' into...
[qemu/qmp-unstable.git] / net / l2tpv3.c
blob8c598b09bc2fccb3044be434532f076ebee29854
1 /*
2 * QEMU System Emulator
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
23 * THE SOFTWARE.
26 #include <linux/ip.h>
27 #include <netdb.h>
28 #include "config-host.h"
29 #include "net/net.h"
30 #include "clients.h"
31 #include "monitor/monitor.h"
32 #include "qemu-common.h"
33 #include "qemu/error-report.h"
34 #include "qemu/option.h"
35 #include "qemu/sockets.h"
36 #include "qemu/iov.h"
37 #include "qemu/main-loop.h"
40 /* The buffer size needs to be investigated for optimum numbers and
41 * optimum means of paging in on different systems. This size is
42 * chosen to be sufficient to accommodate one packet with some headers
45 #define BUFFER_ALIGN sysconf(_SC_PAGESIZE)
46 #define BUFFER_SIZE 2048
47 #define IOVSIZE 2
48 #define MAX_L2TPV3_MSGCNT 64
49 #define MAX_L2TPV3_IOVCNT (MAX_L2TPV3_MSGCNT * IOVSIZE)
51 /* Header set to 0x30000 signifies a data packet */
53 #define L2TPV3_DATA_PACKET 0x30000
55 /* IANA-assigned IP protocol ID for L2TPv3 */
57 #ifndef IPPROTO_L2TP
58 #define IPPROTO_L2TP 0x73
59 #endif
61 typedef struct NetL2TPV3State {
62 NetClientState nc;
63 int fd;
66 * these are used for xmit - that happens packet a time
67 * and for first sign of life packet (easier to parse that once)
70 uint8_t *header_buf;
71 struct iovec *vec;
74 * these are used for receive - try to "eat" up to 32 packets at a time
77 struct mmsghdr *msgvec;
80 * peer address
83 struct sockaddr_storage *dgram_dst;
84 uint32_t dst_size;
87 * L2TPv3 parameters
90 uint64_t rx_cookie;
91 uint64_t tx_cookie;
92 uint32_t rx_session;
93 uint32_t tx_session;
94 uint32_t header_size;
95 uint32_t counter;
98 * DOS avoidance in error handling
101 bool header_mismatch;
104 * Ring buffer handling
107 int queue_head;
108 int queue_tail;
109 int queue_depth;
112 * Precomputed offsets
115 uint32_t offset;
116 uint32_t cookie_offset;
117 uint32_t counter_offset;
118 uint32_t session_offset;
120 /* Poll Control */
122 bool read_poll;
123 bool write_poll;
125 /* Flags */
127 bool ipv6;
128 bool udp;
129 bool has_counter;
130 bool pin_counter;
131 bool cookie;
132 bool cookie_is_64;
134 } NetL2TPV3State;
136 static int l2tpv3_can_send(void *opaque);
137 static void net_l2tpv3_send(void *opaque);
138 static void l2tpv3_writable(void *opaque);
140 static void l2tpv3_update_fd_handler(NetL2TPV3State *s)
142 qemu_set_fd_handler2(s->fd,
143 s->read_poll ? l2tpv3_can_send : NULL,
144 s->read_poll ? net_l2tpv3_send : NULL,
145 s->write_poll ? l2tpv3_writable : NULL,
149 static void l2tpv3_read_poll(NetL2TPV3State *s, bool enable)
151 if (s->read_poll != enable) {
152 s->read_poll = enable;
153 l2tpv3_update_fd_handler(s);
157 static void l2tpv3_write_poll(NetL2TPV3State *s, bool enable)
159 if (s->write_poll != enable) {
160 s->write_poll = enable;
161 l2tpv3_update_fd_handler(s);
165 static void l2tpv3_writable(void *opaque)
167 NetL2TPV3State *s = opaque;
168 l2tpv3_write_poll(s, false);
169 qemu_flush_queued_packets(&s->nc);
172 static int l2tpv3_can_send(void *opaque)
174 NetL2TPV3State *s = opaque;
176 return qemu_can_send_packet(&s->nc);
179 static void l2tpv3_send_completed(NetClientState *nc, ssize_t len)
181 NetL2TPV3State *s = DO_UPCAST(NetL2TPV3State, nc, nc);
182 l2tpv3_read_poll(s, true);
185 static void l2tpv3_poll(NetClientState *nc, bool enable)
187 NetL2TPV3State *s = DO_UPCAST(NetL2TPV3State, nc, nc);
188 l2tpv3_write_poll(s, enable);
189 l2tpv3_read_poll(s, enable);
192 static void l2tpv3_form_header(NetL2TPV3State *s)
194 uint32_t *counter;
196 if (s->udp) {
197 stl_be_p((uint32_t *) s->header_buf, L2TPV3_DATA_PACKET);
199 stl_be_p(
200 (uint32_t *) (s->header_buf + s->session_offset),
201 s->tx_session
203 if (s->cookie) {
204 if (s->cookie_is_64) {
205 stq_be_p(
206 (uint64_t *)(s->header_buf + s->cookie_offset),
207 s->tx_cookie
209 } else {
210 stl_be_p(
211 (uint32_t *) (s->header_buf + s->cookie_offset),
212 s->tx_cookie
216 if (s->has_counter) {
217 counter = (uint32_t *)(s->header_buf + s->counter_offset);
218 if (s->pin_counter) {
219 *counter = 0;
220 } else {
221 stl_be_p(counter, ++s->counter);
226 static ssize_t net_l2tpv3_receive_dgram_iov(NetClientState *nc,
227 const struct iovec *iov,
228 int iovcnt)
230 NetL2TPV3State *s = DO_UPCAST(NetL2TPV3State, nc, nc);
232 struct msghdr message;
233 int ret;
235 if (iovcnt > MAX_L2TPV3_IOVCNT - 1) {
236 error_report(
237 "iovec too long %d > %d, change l2tpv3.h",
238 iovcnt, MAX_L2TPV3_IOVCNT
240 return -1;
242 l2tpv3_form_header(s);
243 memcpy(s->vec + 1, iov, iovcnt * sizeof(struct iovec));
244 s->vec->iov_base = s->header_buf;
245 s->vec->iov_len = s->offset;
246 message.msg_name = s->dgram_dst;
247 message.msg_namelen = s->dst_size;
248 message.msg_iov = s->vec;
249 message.msg_iovlen = iovcnt + 1;
250 message.msg_control = NULL;
251 message.msg_controllen = 0;
252 message.msg_flags = 0;
253 do {
254 ret = sendmsg(s->fd, &message, 0);
255 } while ((ret == -1) && (errno == EINTR));
256 if (ret > 0) {
257 ret -= s->offset;
258 } else if (ret == 0) {
259 /* belt and braces - should not occur on DGRAM
260 * we should get an error and never a 0 send
262 ret = iov_size(iov, iovcnt);
263 } else {
264 /* signal upper layer that socket buffer is full */
265 ret = -errno;
266 if (ret == -EAGAIN || ret == -ENOBUFS) {
267 l2tpv3_write_poll(s, true);
268 ret = 0;
271 return ret;
274 static ssize_t net_l2tpv3_receive_dgram(NetClientState *nc,
275 const uint8_t *buf,
276 size_t size)
278 NetL2TPV3State *s = DO_UPCAST(NetL2TPV3State, nc, nc);
280 struct iovec *vec;
281 struct msghdr message;
282 ssize_t ret = 0;
284 l2tpv3_form_header(s);
285 vec = s->vec;
286 vec->iov_base = s->header_buf;
287 vec->iov_len = s->offset;
288 vec++;
289 vec->iov_base = (void *) buf;
290 vec->iov_len = size;
291 message.msg_name = s->dgram_dst;
292 message.msg_namelen = s->dst_size;
293 message.msg_iov = s->vec;
294 message.msg_iovlen = 2;
295 message.msg_control = NULL;
296 message.msg_controllen = 0;
297 message.msg_flags = 0;
298 do {
299 ret = sendmsg(s->fd, &message, 0);
300 } while ((ret == -1) && (errno == EINTR));
301 if (ret > 0) {
302 ret -= s->offset;
303 } else if (ret == 0) {
304 /* belt and braces - should not occur on DGRAM
305 * we should get an error and never a 0 send
307 ret = size;
308 } else {
309 ret = -errno;
310 if (ret == -EAGAIN || ret == -ENOBUFS) {
311 /* signal upper layer that socket buffer is full */
312 l2tpv3_write_poll(s, true);
313 ret = 0;
316 return ret;
319 static int l2tpv3_verify_header(NetL2TPV3State *s, uint8_t *buf)
322 uint32_t *session;
323 uint64_t cookie;
325 if ((!s->udp) && (!s->ipv6)) {
326 buf += sizeof(struct iphdr) /* fix for ipv4 raw */;
329 /* we do not do a strict check for "data" packets as per
330 * the RFC spec because the pure IP spec does not have
331 * that anyway.
334 if (s->cookie) {
335 if (s->cookie_is_64) {
336 cookie = ldq_be_p(buf + s->cookie_offset);
337 } else {
338 cookie = ldl_be_p(buf + s->cookie_offset);
340 if (cookie != s->rx_cookie) {
341 if (!s->header_mismatch) {
342 error_report("unknown cookie id");
344 return -1;
347 session = (uint32_t *) (buf + s->session_offset);
348 if (ldl_be_p(session) != s->rx_session) {
349 if (!s->header_mismatch) {
350 error_report("session mismatch");
352 return -1;
354 return 0;
357 static void net_l2tpv3_process_queue(NetL2TPV3State *s)
359 int size = 0;
360 struct iovec *vec;
361 bool bad_read;
362 int data_size;
363 struct mmsghdr *msgvec;
365 /* go into ring mode only if there is a "pending" tail */
366 if (s->queue_depth > 0) {
367 do {
368 msgvec = s->msgvec + s->queue_tail;
369 if (msgvec->msg_len > 0) {
370 data_size = msgvec->msg_len - s->header_size;
371 vec = msgvec->msg_hdr.msg_iov;
372 if ((data_size > 0) &&
373 (l2tpv3_verify_header(s, vec->iov_base) == 0)) {
374 vec++;
375 /* Use the legacy delivery for now, we will
376 * switch to using our own ring as a queueing mechanism
377 * at a later date
379 size = qemu_send_packet_async(
380 &s->nc,
381 vec->iov_base,
382 data_size,
383 l2tpv3_send_completed
385 if (size == 0) {
386 l2tpv3_read_poll(s, false);
388 bad_read = false;
389 } else {
390 bad_read = true;
391 if (!s->header_mismatch) {
392 /* report error only once */
393 error_report("l2tpv3 header verification failed");
394 s->header_mismatch = true;
397 } else {
398 bad_read = true;
400 s->queue_tail = (s->queue_tail + 1) % MAX_L2TPV3_MSGCNT;
401 s->queue_depth--;
402 } while (
403 (s->queue_depth > 0) &&
404 qemu_can_send_packet(&s->nc) &&
405 ((size > 0) || bad_read)
410 static void net_l2tpv3_send(void *opaque)
412 NetL2TPV3State *s = opaque;
413 int target_count, count;
414 struct mmsghdr *msgvec;
416 /* go into ring mode only if there is a "pending" tail */
418 if (s->queue_depth) {
420 /* The ring buffer we use has variable intake
421 * count of how much we can read varies - adjust accordingly
424 target_count = MAX_L2TPV3_MSGCNT - s->queue_depth;
426 /* Ensure we do not overrun the ring when we have
427 * a lot of enqueued packets
430 if (s->queue_head + target_count > MAX_L2TPV3_MSGCNT) {
431 target_count = MAX_L2TPV3_MSGCNT - s->queue_head;
433 } else {
435 /* we do not have any pending packets - we can use
436 * the whole message vector linearly instead of using
437 * it as a ring
440 s->queue_head = 0;
441 s->queue_tail = 0;
442 target_count = MAX_L2TPV3_MSGCNT;
445 msgvec = s->msgvec + s->queue_head;
446 if (target_count > 0) {
447 do {
448 count = recvmmsg(
449 s->fd,
450 msgvec,
451 target_count, MSG_DONTWAIT, NULL);
452 } while ((count == -1) && (errno == EINTR));
453 if (count < 0) {
454 /* Recv error - we still need to flush packets here,
455 * (re)set queue head to current position
457 count = 0;
459 s->queue_head = (s->queue_head + count) % MAX_L2TPV3_MSGCNT;
460 s->queue_depth += count;
462 net_l2tpv3_process_queue(s);
465 static void destroy_vector(struct mmsghdr *msgvec, int count, int iovcount)
467 int i, j;
468 struct iovec *iov;
469 struct mmsghdr *cleanup = msgvec;
470 if (cleanup) {
471 for (i = 0; i < count; i++) {
472 if (cleanup->msg_hdr.msg_iov) {
473 iov = cleanup->msg_hdr.msg_iov;
474 for (j = 0; j < iovcount; j++) {
475 g_free(iov->iov_base);
476 iov++;
478 g_free(cleanup->msg_hdr.msg_iov);
480 cleanup++;
482 g_free(msgvec);
486 static struct mmsghdr *build_l2tpv3_vector(NetL2TPV3State *s, int count)
488 int i;
489 struct iovec *iov;
490 struct mmsghdr *msgvec, *result;
492 msgvec = g_new(struct mmsghdr, count);
493 result = msgvec;
494 for (i = 0; i < count ; i++) {
495 msgvec->msg_hdr.msg_name = NULL;
496 msgvec->msg_hdr.msg_namelen = 0;
497 iov = g_new(struct iovec, IOVSIZE);
498 msgvec->msg_hdr.msg_iov = iov;
499 iov->iov_base = g_malloc(s->header_size);
500 iov->iov_len = s->header_size;
501 iov++ ;
502 iov->iov_base = qemu_memalign(BUFFER_ALIGN, BUFFER_SIZE);
503 iov->iov_len = BUFFER_SIZE;
504 msgvec->msg_hdr.msg_iovlen = 2;
505 msgvec->msg_hdr.msg_control = NULL;
506 msgvec->msg_hdr.msg_controllen = 0;
507 msgvec->msg_hdr.msg_flags = 0;
508 msgvec++;
510 return result;
513 static void net_l2tpv3_cleanup(NetClientState *nc)
515 NetL2TPV3State *s = DO_UPCAST(NetL2TPV3State, nc, nc);
516 qemu_purge_queued_packets(nc);
517 l2tpv3_read_poll(s, false);
518 l2tpv3_write_poll(s, false);
519 if (s->fd >= 0) {
520 close(s->fd);
522 destroy_vector(s->msgvec, MAX_L2TPV3_MSGCNT, IOVSIZE);
523 g_free(s->vec);
524 g_free(s->header_buf);
525 g_free(s->dgram_dst);
528 static NetClientInfo net_l2tpv3_info = {
529 .type = NET_CLIENT_OPTIONS_KIND_L2TPV3,
530 .size = sizeof(NetL2TPV3State),
531 .receive = net_l2tpv3_receive_dgram,
532 .receive_iov = net_l2tpv3_receive_dgram_iov,
533 .poll = l2tpv3_poll,
534 .cleanup = net_l2tpv3_cleanup,
537 int net_init_l2tpv3(const NetClientOptions *opts,
538 const char *name,
539 NetClientState *peer)
543 const NetdevL2TPv3Options *l2tpv3;
544 NetL2TPV3State *s;
545 NetClientState *nc;
546 int fd = -1, gairet;
547 struct addrinfo hints;
548 struct addrinfo *result = NULL;
549 char *srcport, *dstport;
551 nc = qemu_new_net_client(&net_l2tpv3_info, peer, "l2tpv3", name);
553 s = DO_UPCAST(NetL2TPV3State, nc, nc);
555 s->queue_head = 0;
556 s->queue_tail = 0;
557 s->header_mismatch = false;
559 assert(opts->kind == NET_CLIENT_OPTIONS_KIND_L2TPV3);
560 l2tpv3 = opts->l2tpv3;
562 if (l2tpv3->has_ipv6 && l2tpv3->ipv6) {
563 s->ipv6 = l2tpv3->ipv6;
564 } else {
565 s->ipv6 = false;
568 if ((l2tpv3->has_offset) && (l2tpv3->offset > 256)) {
569 error_report("l2tpv3_open : offset must be less than 256 bytes");
570 goto outerr;
573 if (l2tpv3->has_rxcookie || l2tpv3->has_txcookie) {
574 if (l2tpv3->has_rxcookie && l2tpv3->has_txcookie) {
575 s->cookie = true;
576 } else {
577 goto outerr;
579 } else {
580 s->cookie = false;
583 if (l2tpv3->has_cookie64 || l2tpv3->cookie64) {
584 s->cookie_is_64 = true;
585 } else {
586 s->cookie_is_64 = false;
589 if (l2tpv3->has_udp && l2tpv3->udp) {
590 s->udp = true;
591 if (!(l2tpv3->has_srcport && l2tpv3->has_dstport)) {
592 error_report("l2tpv3_open : need both src and dst port for udp");
593 goto outerr;
594 } else {
595 srcport = l2tpv3->srcport;
596 dstport = l2tpv3->dstport;
598 } else {
599 s->udp = false;
600 srcport = NULL;
601 dstport = NULL;
605 s->offset = 4;
606 s->session_offset = 0;
607 s->cookie_offset = 4;
608 s->counter_offset = 4;
610 s->tx_session = l2tpv3->txsession;
611 if (l2tpv3->has_rxsession) {
612 s->rx_session = l2tpv3->rxsession;
613 } else {
614 s->rx_session = s->tx_session;
617 if (s->cookie) {
618 s->rx_cookie = l2tpv3->rxcookie;
619 s->tx_cookie = l2tpv3->txcookie;
620 if (s->cookie_is_64 == true) {
621 /* 64 bit cookie */
622 s->offset += 8;
623 s->counter_offset += 8;
624 } else {
625 /* 32 bit cookie */
626 s->offset += 4;
627 s->counter_offset += 4;
631 memset(&hints, 0, sizeof(hints));
633 if (s->ipv6) {
634 hints.ai_family = AF_INET6;
635 } else {
636 hints.ai_family = AF_INET;
638 if (s->udp) {
639 hints.ai_socktype = SOCK_DGRAM;
640 hints.ai_protocol = 0;
641 s->offset += 4;
642 s->counter_offset += 4;
643 s->session_offset += 4;
644 s->cookie_offset += 4;
645 } else {
646 hints.ai_socktype = SOCK_RAW;
647 hints.ai_protocol = IPPROTO_L2TP;
650 gairet = getaddrinfo(l2tpv3->src, srcport, &hints, &result);
652 if ((gairet != 0) || (result == NULL)) {
653 error_report(
654 "l2tpv3_open : could not resolve src, errno = %s",
655 gai_strerror(gairet)
657 goto outerr;
659 fd = socket(result->ai_family, result->ai_socktype, result->ai_protocol);
660 if (fd == -1) {
661 fd = -errno;
662 error_report("l2tpv3_open : socket creation failed, errno = %d", -fd);
663 goto outerr;
665 if (bind(fd, (struct sockaddr *) result->ai_addr, result->ai_addrlen)) {
666 error_report("l2tpv3_open : could not bind socket err=%i", errno);
667 goto outerr;
669 if (result) {
670 freeaddrinfo(result);
673 memset(&hints, 0, sizeof(hints));
675 if (s->ipv6) {
676 hints.ai_family = AF_INET6;
677 } else {
678 hints.ai_family = AF_INET;
680 if (s->udp) {
681 hints.ai_socktype = SOCK_DGRAM;
682 hints.ai_protocol = 0;
683 } else {
684 hints.ai_socktype = SOCK_RAW;
685 hints.ai_protocol = IPPROTO_L2TP;
688 result = NULL;
689 gairet = getaddrinfo(l2tpv3->dst, dstport, &hints, &result);
690 if ((gairet != 0) || (result == NULL)) {
691 error_report(
692 "l2tpv3_open : could not resolve dst, error = %s",
693 gai_strerror(gairet)
695 goto outerr;
698 s->dgram_dst = g_new0(struct sockaddr_storage, 1);
699 memcpy(s->dgram_dst, result->ai_addr, result->ai_addrlen);
700 s->dst_size = result->ai_addrlen;
702 if (result) {
703 freeaddrinfo(result);
706 if (l2tpv3->has_counter && l2tpv3->counter) {
707 s->has_counter = true;
708 s->offset += 4;
709 } else {
710 s->has_counter = false;
713 if (l2tpv3->has_pincounter && l2tpv3->pincounter) {
714 s->has_counter = true; /* pin counter implies that there is counter */
715 s->pin_counter = true;
716 } else {
717 s->pin_counter = false;
720 if (l2tpv3->has_offset) {
721 /* extra offset */
722 s->offset += l2tpv3->offset;
725 if ((s->ipv6) || (s->udp)) {
726 s->header_size = s->offset;
727 } else {
728 s->header_size = s->offset + sizeof(struct iphdr);
731 s->msgvec = build_l2tpv3_vector(s, MAX_L2TPV3_MSGCNT);
732 s->vec = g_new(struct iovec, MAX_L2TPV3_IOVCNT);
733 s->header_buf = g_malloc(s->header_size);
735 qemu_set_nonblock(fd);
737 s->fd = fd;
738 s->counter = 0;
740 l2tpv3_read_poll(s, true);
742 snprintf(s->nc.info_str, sizeof(s->nc.info_str),
743 "l2tpv3: connected");
744 return 0;
745 outerr:
746 qemu_del_net_client(nc);
747 if (fd >= 0) {
748 close(fd);
750 if (result) {
751 freeaddrinfo(result);
753 return -1;