#include "netlink.h"
#include "process.h"
+#ifdef NETLINK_DEBUG
+#define MDEBUG(...) Message(M_DEBUG, __VA_ARGS__)
+#else /* !NETLINK_DEBUG */
+#define MDEBUG(...) ((void)0)
+#endif /* !NETLINK_DEBUG */
+
#define ICMP_TYPE_ECHO_REPLY 0
#define ICMP_TYPE_UNREACHABLE 3
#define ICMP_CODE_TTL_EXCEEDED 0
/* Generic IP checksum routine */
-static inline uint16_t ip_csum(uint8_t *iph,int32_t count)
+static inline uint16_t ip_csum(const uint8_t *iph,int32_t count)
{
register uint32_t sum=0;
* By Jorge Cwik <jorge@laser.satlink.net>, adapted for linux by
* Arnt Gulbrandsen.
*/
-static inline uint16_t ip_fast_csum(uint8_t *iph, int32_t ihl) {
+static inline uint16_t ip_fast_csum(const uint8_t *iph, int32_t ihl) {
uint32_t sum;
__asm__ __volatile__(
uint8_t tos;
uint16_t tot_len;
uint16_t id;
- uint16_t frag_off;
+ uint16_t frag;
+#define IPHDR_FRAG_OFF ((uint16_t)0x1fff)
+#define IPHDR_FRAG_MORE ((uint16_t)0x2000)
+#define IPHDR_FRAG_DONT ((uint16_t)0x4000)
+/* reserved 0x8000 */
uint8_t ttl;
uint8_t protocol;
uint16_t check;
uint8_t type;
uint8_t code;
uint16_t check;
- union {
+ union icmpinfofield {
uint32_t unused;
struct {
uint8_t pointer;
uint16_t id;
uint16_t seq;
} echo;
+ struct {
+ uint16_t unused;
+ uint16_t mtu;
+ } fragneeded;
} d;
};
+
+static const union icmpinfofield icmp_noinfo;
static void netlink_packet_deliver(struct netlink *st,
struct netlink_client *client,
h->iph.tos=0;
h->iph.tot_len=htons(len+(h->iph.ihl*4)+8);
h->iph.id=0;
- h->iph.frag_off=0;
+ h->iph.frag=0;
h->iph.ttl=255; /* XXX should be configurable */
h->iph.protocol=1;
h->iph.saddr=htonl(st->secnet_address);
icmph=(struct icmphdr *)buf->start;
if (iph->protocol==1) {
switch(icmph->type) {
- case 3: /* Destination unreachable */
- case 11: /* Time Exceeded */
- case 12: /* Parameter Problem */
+ /* Based on http://www.iana.org/assignments/icmp-parameters/icmp-parameters.xhtml#icmp-parameters-types
+ * as retrieved Thu, 20 Mar 2014 00:16:44 +0000.
+ * Deprecated, reserved, unassigned and experimental
+ * options are treated as not safe to reply to.
+ */
+ case 0: /* Echo Reply */
+ case 8: /* Echo */
+ case 13: /* Timestamp */
+ case 14: /* Timestamp Reply */
+ return True;
+ default:
return False;
}
}
/* How do we spot broadcast destination addresses? */
- if (ntohs(iph->frag_off)&0x1fff) return False; /* Non-initial fragment */
+ if (ntohs(iph->frag)&IPHDR_FRAG_OFF) return False;
source=ntohl(iph->saddr);
if (source==0) return False;
if ((source&0xff000000)==0x7f000000) return False;
/* client indicates where the packet we're constructing a response to
comes from. NULL indicates the host. */
static void netlink_icmp_simple(struct netlink *st, struct buffer_if *buf,
- struct netlink_client *client,
- uint8_t type, uint8_t code)
+ uint8_t type, uint8_t code,
+ union icmpinfofield info)
{
struct icmphdr *h;
uint16_t len;
struct iphdr *iph=(struct iphdr *)buf->start;
len=netlink_icmp_reply_len(buf);
h=netlink_icmp_tmpl(st,ntohl(iph->saddr),len);
- h->type=type; h->code=code;
+ h->type=type; h->code=code; h->d=info;
memcpy(buf_append(&st->icmp,len),buf->start,len);
netlink_icmp_csum(h);
netlink_packet_deliver(st,NULL,&st->icmp);
#undef BAD
}
+static const char *fragment_filter_header(uint8_t *base, long *hlp)
+{
+ const int fixedhl = sizeof(struct iphdr);
+ long hl = *hlp;
+ const uint8_t *ipend = base + hl;
+ uint8_t *op = base + fixedhl;
+ const uint8_t *ip = op;
+
+ while (ip < ipend) {
+ uint8_t opt = ip[0];
+ int remain = ipend - ip;
+ if (opt == 0x00) /* End of Options List */ break;
+ if (opt == 0x01) /* No Operation */ continue;
+ if (remain < 2) return "IPv4 options truncated at length";
+ int optlen = ip[1];
+ if (remain < optlen) return "IPv4 options truncated in option";
+ if (opt & 0x80) /* copy */ {
+ memmove(op, ip, optlen);
+ op += optlen;
+ }
+ ip += optlen;
+ }
+ while ((hl = (op - base)) & 0x3)
+ *op++ = 0x00 /* End of Option List */;
+ ((struct iphdr*)base)->ihl = hl >> 2;
+ *hlp = hl;
+
+ return 0;
+}
+
+/* Fragment or send ICMP Fragmentation Needed */
+static void netlink_maybe_fragment(struct netlink *st,
+ netlink_deliver_fn *deliver,
+ void *deliver_dst,
+ const char *delivery_name,
+ int32_t mtu,
+ uint32_t source, uint32_t dest,
+ struct buffer_if *buf)
+{
+ struct iphdr *iph=(struct iphdr*)buf->start;
+ long hl = iph->ihl*4;
+ const char *ssource = ipaddr_to_string(source);
+
+ if (buf->size <= mtu) {
+ deliver(deliver_dst, buf);
+ return;
+ }
+
+ MDEBUG("%s: fragmenting %s->%s org.size=%"PRId32"\n",
+ st->name, ssource, delivery_name, buf->size);
+
+#define BADFRAG(m, ...) \
+ Message(M_WARNING, \
+ "%s: fragmenting packet from source %s" \
+ " for transmission via %s: " m "\n", \
+ st->name, ssource, delivery_name, \
+ ## __VA_ARGS__);
+
+ unsigned orig_frag = ntohs(iph->frag);
+
+ if (orig_frag&IPHDR_FRAG_DONT) {
+ union icmpinfofield info =
+ { .fragneeded = { .unused = 0, .mtu = htons(mtu) } };
+ netlink_icmp_simple(st,buf,
+ ICMP_TYPE_UNREACHABLE,
+ ICMP_CODE_FRAGMENTATION_REQUIRED,
+ info);
+ BUF_FREE(buf);
+ return;
+ }
+ if (mtu < hl + 8) {
+ BADFRAG("mtu %"PRId32" too small", mtu);
+ BUF_FREE(buf);
+ return;
+ }
+
+ /* we (ab)use the icmp buffer to stash the original packet */
+ struct buffer_if *orig = &st->icmp;
+ BUF_ALLOC(orig,"netlink_client_deliver fragment orig");
+ buffer_copy(orig,buf);
+ BUF_FREE(buf);
+
+ const uint8_t *startindata = orig->start + hl;
+ const uint8_t *indata = startindata;
+ const uint8_t *endindata = orig->start + orig->size;
+ _Bool filtered = 0;
+
+ for (;;) {
+ /* compute our fragment offset */
+ long dataoffset = indata - startindata
+ + (orig_frag & IPHDR_FRAG_OFF)*8;
+ assert(!(dataoffset & 7));
+ if (dataoffset > IPHDR_FRAG_OFF*8) {
+ BADFRAG("ultimate fragment offset out of range");
+ break;
+ }
+
+ BUF_ALLOC(buf,"netlink_client_deliver fragment frag");
+ buffer_init(buf,calculate_max_start_pad());
+
+ /* copy header (possibly filtered); will adjust in a bit */
+ struct iphdr *fragh = buf_append(buf, hl);
+ memcpy(fragh, orig->start, hl);
+
+ /* decide how much payload to copy and copy it */
+ long avail = mtu - hl;
+ long remain = endindata - indata;
+ long use = avail < remain ? (avail & ~(long)7) : remain;
+ memcpy(buf_append(buf, use), indata, use);
+ indata += use;
+
+ _Bool last_frag = indata >= endindata;
+
+ /* adjust the header */
+ fragh->tot_len = htons(buf->size);
+ fragh->frag =
+ htons((orig_frag & ~IPHDR_FRAG_OFF) |
+ (last_frag ? 0 : IPHDR_FRAG_MORE) |
+ (dataoffset >> 3));
+ fragh->check = 0;
+ fragh->check = ip_fast_csum((const void*)fragh, fragh->ihl);
+
+ /* actually send it */
+ deliver(deliver_dst, buf);
+ if (last_frag)
+ break;
+
+ /* after copying the header for the first frag,
+ * we filter the header for the remaining frags */
+ if (!filtered++) {
+ const char *bad = fragment_filter_header(orig->start, &hl);
+ if (bad) { BADFRAG("%s", bad); break; }
+ }
+ }
+
+ BUF_FREE(orig);
+
+#undef BADFRAG
+}
+
/* Deliver a packet _to_ client; used after we have decided
* what to do with it (and just to check that the client has
* actually registered a delivery function with us). */
BUF_FREE(buf);
return;
}
- client->deliver(client->dst, buf);
+ netlink_maybe_fragment(st, client->deliver,client->dst,client->name,
+ client->mtu, source,dest,buf);
client->outcount++;
}
+/* Deliver a packet to the host; used after we have decided that that
+ * is what to do with it. */
+static void netlink_host_deliver(struct netlink *st,
+ uint32_t source, uint32_t dest,
+ struct buffer_if *buf)
+{
+ netlink_maybe_fragment(st, st->deliver_to_host,st->dst,"(host)",
+ st->mtu, source,dest,buf);
+ st->outcount++;
+}
+
/* Deliver a packet. "client" is the _origin_ of the packet, not its
destination, and is NULL for packets from the host and packets
generated internally in secnet. */
/* The packet's not going down a tunnel. It might (ought to)
be for the host. */
if (ipset_contains_addr(st->networks,dest)) {
- st->deliver_to_host(st->dst,buf);
- st->outcount++;
+ netlink_host_deliver(st,source,dest,buf);
BUF_ASSERT_FREE(buf);
} else {
string_t s,d;
Message(M_DEBUG,"%s: don't know where to deliver packet "
"(s=%s, d=%s)\n", st->name, s, d);
free(s); free(d);
- netlink_icmp_simple(st,buf,client,ICMP_TYPE_UNREACHABLE,
- ICMP_CODE_NET_UNREACHABLE);
+ netlink_icmp_simple(st,buf,ICMP_TYPE_UNREACHABLE,
+ ICMP_CODE_NET_UNREACHABLE, icmp_noinfo);
BUF_FREE(buf);
}
} else {
st->name,s,d);
free(s); free(d);
- netlink_icmp_simple(st,buf,client,ICMP_TYPE_UNREACHABLE,
- ICMP_CODE_NET_PROHIBITED);
+ netlink_icmp_simple(st,buf,ICMP_TYPE_UNREACHABLE,
+ ICMP_CODE_NET_PROHIBITED, icmp_noinfo);
BUF_FREE(buf);
} else {
if (best_quality>0) {
- /* XXX Fragment if required */
netlink_client_deliver(st,st->routes[best_match],
source,dest,buf);
BUF_ASSERT_FREE(buf);
} else {
/* Generate ICMP destination unreachable */
- netlink_icmp_simple(st,buf,client,ICMP_TYPE_UNREACHABLE,
- ICMP_CODE_NET_UNREACHABLE); /* client==NULL */
+ netlink_icmp_simple(st,buf,
+ ICMP_TYPE_UNREACHABLE,
+ ICMP_CODE_NET_UNREACHABLE,
+ icmp_noinfo);
BUF_FREE(buf);
}
}
/* Packet has already been checked */
if (iph->ttl<=1) {
/* Generate ICMP time exceeded */
- netlink_icmp_simple(st,buf,client,ICMP_TYPE_TIME_EXCEEDED,
- ICMP_CODE_TTL_EXCEEDED);
+ netlink_icmp_simple(st,buf,ICMP_TYPE_TIME_EXCEEDED,
+ ICMP_CODE_TTL_EXCEEDED,icmp_noinfo);
BUF_FREE(buf);
return;
}
}
h=(struct icmphdr *)buf->start;
- if ((ntohs(h->iph.frag_off)&0xbfff)!=0) {
- Message(M_WARNING,"%s: fragmented packet addressed to secnet; "
- "ignoring it\n",st->name);
+ unsigned fraginfo = ntohs(h->iph.frag);
+ if ((fraginfo&(IPHDR_FRAG_OFF|IPHDR_FRAG_MORE))!=0) {
+ if (!(fraginfo & IPHDR_FRAG_OFF))
+ /* report only for first fragment */
+ Message(M_WARNING,"%s: fragmented packet addressed to secnet; "
+ "ignoring it\n",st->name);
BUF_FREE(buf);
return;
}
Message(M_WARNING,"%s: unknown incoming ICMP\n",st->name);
} else {
/* Send ICMP protocol unreachable */
- netlink_icmp_simple(st,buf,client,ICMP_TYPE_UNREACHABLE,
- ICMP_CODE_PROTOCOL_UNREACHABLE);
+ netlink_icmp_simple(st,buf,ICMP_TYPE_UNREACHABLE,
+ ICMP_CODE_PROTOCOL_UNREACHABLE,icmp_noinfo);
BUF_FREE(buf);
return;
}
address validity and generate ICMP, etc. */
if (st->ptp) {
if (client) {
- st->deliver_to_host(st->dst,buf);
+ netlink_host_deliver(st,source,dest,buf);
} else {
netlink_client_deliver(st,st->clients,source,dest,buf);
}
if (requested) c=M_WARNING;
if (st->ptp) {
net=ipaddr_to_string(st->secnet_address);
- Message(c,"%s: point-to-point (remote end is %s); routes:\n",
+ Message(c,"%s: point-to-point (remote end is %s); routes: ",
st->name, net);
free(net);
netlink_output_subnets(st,c,st->clients->subnets);
though, and will make the route dump look complicated... */
st->subnets=ipset_to_subnet_list(st->networks);
st->mtu=dict_read_number(dict, "mtu", False, "netlink", loc, DEFAULT_MTU);
- buffer_new(&st->icmp,ICMP_BUFSIZE);
+ buffer_new(&st->icmp,MAX(ICMP_BUFSIZE,st->mtu));
st->outcount=0;
st->localcount=0;
~ianmdlvl / secnet.git / blobdiff