2 * userv service (or standalone program) for per-user IP subranges.
4 * When invoked appropriately, it creates a point-to-point network
5 * interface with specified parameters. It arranges for packets sent out
6 * via that interface by the kernel to appear on its own stdout in SLIP or
7 * CSLIP encoding, and packets injected into its own stdin to be given to
8 * the kernel as if received on that interface. Optionally, additional
9 * routes can be set up to arrange for traffic for other address ranges to
10 * be routed through the new interface.
12 * This is the service program, which is invoked as root from userv (or may
13 * be invoked firectly).
15 * Its arguments are supposed to be, in order, as follows:
17 * The first two arguments are usually supplied by the userv
18 * configuration. See the file `ipif/ipif' in the source tree, which
19 * is installed in /etc/userv/services.d/ipif by `make install':
23 * Specifies address ranges and gids which own them. The default
24 * configuration supplies /etc/userv/ipif-networks, which is then read
25 * for a list of entries, one per line.
28 * Serves to separate the user-supplied and therefore untrusted
29 * arguments from the trusted first argument.
31 * The remaining arguments are supplied by the (untrusted) caller:
33 * <local-addr>,<peer-addr>,<mtu>,<proto>
35 * As for slattach. The only supported protocol is slip.
36 * Alternatively, set to `debug' to print debugging info and
37 * exit. <local-addr> is address of the interface to be created
38 * on the local system; <peer-addr> is the address of the
39 * point-to-point peer. They must be actual addresses (not
42 * <prefix>/<mask>,<prefix>/<mask>,...
44 * List of additional routes to add for this interface. routes will
45 * be set up on the local system arranging for packets for those
46 * networks to be sent via the created interface. <prefix> must be an
47 * IPv4 address, and mask must be an integer (dotted-quad masks are
48 * not supported). If no additional routes are to be set up, use `-'
49 * or supply an empty argument.
51 * Each <config> item - whether a line in a file such as
52 * /etc/userv/ipif-networks, or the single trusted argument supplied
53 * on the service program command line - is one of:
56 * ./<config-file-name>
57 * ../<config-file-name>
59 * Reads a file which contains lines which are each <config>
62 * <gid>,[=][-|+]<prefix>/<len>(-|+<prefix>/<len>...)[,<junk>]
64 * Indicates that <gid> may allocate addresses in the relevant address
65 * range (<junk> is ignored). <gid> must be numeric. To specify a
66 * single host address, you must specify a mask of /32. If `=' is
67 * specified then the specific subrange is only allowed for the local
68 * endpoint address, but not for remote addresses.
70 * More than one range may be given, with each range prefixed
71 * by + or -. In this case each address range in the rule will
72 * scanned in order, and the first range in the rule that matches
73 * any desired rule will count: if that first matching range is
74 * prefixed by `+' (or nothing) then the rule applies, if it
75 * is prefixed by `-' (or nothing matches), the rule does not.
78 * Means that anything is to be permitted. This should not appear in
79 * /etc/userv/ipif-networks, as that would permit any user on the
80 * system to create any interfaces with any addresses and routes
81 * attached. It is provided so that root can usefully invoke the ipif
82 * service program directly (not via userv), without needing to set up
83 * permissions in /etc/userv/ipif-networks.
87 * Comment. Blank lines are also ignored.
89 * NB: Permission is granted if _any_ config entry matches the request.
91 * The service program should be run from userv with no-disconnect-hup.
94 * Copyright (C) 1999-2000,2003 Ian Jackson
95 * This file is part of ipif, part of userv-utils
97 * This is free software; you can redistribute it and/or modify it
98 * under the terms of the GNU General Public License as published by
99 * the Free Software Foundation; either version 2 of the License, or
100 * (at your option) any later version.
102 * This program is distributed in the hope that it will be useful, but
103 * WITHOUT ANY WARRANTY; without even the implied warranty of
104 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
105 * General Public License for more details.
107 * You should have received a copy of the GNU General Public License
108 * along with userv-utils; if not, write to the Free Software
109 * Foundation, 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
127 #include <sys/types.h>
128 #include <sys/wait.h>
129 #include <sys/stat.h>
131 #include <sys/types.h>
132 #include <sys/ioctl.h>
133 #include <sys/socket.h>
135 #include <sys/stat.h>
138 #include <linux/if.h>
139 #include <linux/if_tun.h>
142 #define MAXEXROUTES 50
145 static const unsigned long gidmaxval= (unsigned long)((gid_t)-2);
146 static const char *const protos_ok[]= { "slip", 0 };
147 static const int signals[]= { SIGHUP, SIGINT, SIGTERM, 0 };
149 static const char *configstr, *proto;
150 static unsigned long localaddr, peeraddr, mtu;
151 static int localpming, peerpming;
152 static int localallow, peerallow, allallow;
153 static int nexroutes;
154 static struct exroute {
155 unsigned long prefix, mask;
157 char prefixtxt[ATXTLEN], masktxt[ATXTLEN];
158 } exroutes[MAXEXROUTES];
160 static char localtxt[ATXTLEN];
161 static char peertxt[ATXTLEN];
163 static struct pplace {
164 struct pplace *parent;
165 const char *filename;
174 static void terminate(int estatus) {
179 static void fatal(const char *fmt, ...)
180 __attribute__((format(printf,1,2)));
181 static void fatal(const char *fmt, ...) {
185 fputs("userv-ipif service: fatal error: ",stderr);
186 vfprintf(stderr, fmt, al);
191 static void sysfatal(const char *fmt, ...)
192 __attribute__((format(printf,1,2)));
193 static void sysfatal(const char *fmt, ...) {
200 fputs("userv-ipif service: fatal system error: ",stderr);
201 vfprintf(stderr, fmt, al);
202 fprintf(stderr,": %s\n", strerror(e));
207 static void badusage(const char *fmt, ...)
208 __attribute__((format(printf,1,2)));
209 static void badusage(const char *fmt, ...) {
215 "userv-ipif service: %s:%d: ",
216 cpplace->filename, cpplace->lineno);
218 fputs("userv-ipif service: invalid usage: ",stderr);
221 vfprintf(stderr, fmt, al);
225 for (cpp=cpplace->parent; cpp; cpp=cpp->parent) {
227 "userv-ipif service: %s:%d: ... in file included from here\n",
228 cpp->filename, cpp->lineno);
234 static char *ip2txt(unsigned long addr, char *buf) {
235 sprintf(buf, "%lu.%lu.%lu.%lu",
243 static unsigned long eat_number(const char **argp, const char *what,
244 unsigned long min, unsigned long max,
245 const char *endchars, int *endchar_r) {
246 /* If !endchars then the endchar must be a nul, otherwise it may be
247 * a nul (resulting in *argp set to 0) or something else (*argp set
248 * to point to after delim, *endchar_r set to delim).
249 * *endchar_r may be 0.
255 if (!*argp) { badusage("missing number %s",what); }
256 rv= strtoul(*argp,&ep,0);
257 if ((endchar= *ep)) {
258 if (!endchars) badusage("junk after number %s",what);
259 if (!strchr(endchars,endchar))
260 badusage("invalid character or delimiter `%c' in or after number, %s:"
261 " expected %s (or none?)", endchar,what,endchars);
266 if (endchar_r) *endchar_r= endchar;
267 if (rv < min || rv > max) badusage("number %s value %lu out of range %lu..%lu",
272 static int addrnet_overlap(unsigned long p1, unsigned long m1,
273 unsigned long p2, unsigned long m2) {
277 return (p1 & mask) == (p2 & mask);
280 static void addrnet_mustdiffer(const char *w1, unsigned long p1, unsigned long m1,
281 const char *w2, unsigned long p2, unsigned long m2) {
282 if (!addrnet_overlap(p1,m1,p2,m2)) return;
283 badusage("%s %08lx/%08lx overlaps/clashes with %s %08lx/%08lx",
287 static unsigned long eat_addr(const char **argp, const char *what,
288 const char *endchars, int *endchar_r) {
297 sprintf(whatbuf,"%s byte #%d",what,i);
298 rv |= eat_number(argp,whatbuf, 0,255, i<3 ? "." : endchars, endchar_r);
304 static void eat_prefixmask(const char **argp, const char *what,
305 const char *endchars, int *endchar_r,
306 unsigned long *prefix_r, unsigned long *mask_r, int *len_r) {
307 /* mask_r and len_r may be 0 */
310 unsigned long prefix, mask;
312 prefix= eat_addr(argp,what, "/",0);
313 sprintf(whatbuf,"%s length",what);
314 len= eat_number(argp,whatbuf, 0,32, endchars,endchar_r);
316 mask= len ? (~0UL << (32-len)) : 0UL;
317 if (prefix & ~mask) badusage("%s prefix %08lx not fully contained in mask %08lx",
320 if (mask_r) *mask_r= mask;
321 if (len_r) *len_r= len;
324 static int addrnet_isin(unsigned long prefix, unsigned long mask,
325 unsigned long mprefix, unsigned long mmask) {
326 return !(~mask & mmask) && (prefix & mmask) == mprefix;
329 /* Totally hideous algorithm for parsing the config file lines.
330 * For each config file line, we first see if its gid applies. If not
331 * we skip it. Otherwise, we do
333 * which sets <foo>pming to 1
334 * for each range. <foo>pming may be 0 if we've determined that
335 * this line does not apply to <foo>.
337 * which calls permit_range_thing for each <foo>
338 * which checks to see if <foo> is inside the relevant
339 * range (for +) or overlaps it (for -) and updates
340 * <foo>allow and <foo>pming.
343 static void permit_begin(void) {
346 localpming= peerpming= 1;
347 for (i=0; i<nexroutes; i++) exroutes[i].pming= 1;
350 static void permit_range_thing(unsigned long tprefix, unsigned long tmask,
351 const char *what, int *tallow, int *tpming,
352 unsigned long pprefix, unsigned long pmask,
353 int plus, int *any) {
355 if (!addrnet_isin(tprefix,tmask, pprefix,pmask)) return;
356 if (*tpming) *tallow= 1;
358 if (!addrnet_overlap(tprefix,tmask, pprefix,pmask)) return;
361 if (!proto) printf(" %c%s", plus?'+':'-', what);
365 static void permit_range(unsigned long prefix, unsigned long mask,
366 int plus, int localonly) {
370 assert(!(prefix & ~mask));
373 permit_range_thing(localaddr,~0UL,"local", &localallow,&localpming,
374 prefix,mask, plus,&any);
377 permit_range_thing(peeraddr,~0UL, "peer-addr", &peerallow,&peerpming,
378 prefix,mask, plus,&any);
379 for (i=0; i<nexroutes; i++) {
380 sprintf(idbuf,"route#%d",i);
381 permit_range_thing(exroutes[i].prefix,exroutes[i].mask, idbuf,
382 &exroutes[i].allow,&exroutes[i].pming,
383 prefix,mask, plus,&any);
387 if (!any) fputs(" nothing",stdout);
390 static void pconfig(const char *configstr, int truncated);
392 static void pfile(const char *filename) {
396 struct pplace npp, *cpp;
398 for (cpp=cpplace; cpp; cpp=cpp->parent) {
399 if (!strcmp(cpp->filename,filename))
400 badusage("recursive configuration file `%s'",filename);
403 file= fopen(filename,"r");
405 badusage("cannot open configuration file `%s': %s", filename, strerror(errno));
407 if (!proto) printf("config file `%s':\n",filename);
410 npp.filename= filename;
414 while (fgets(buf, sizeof(buf), file)) {
419 truncated= (buf[l-1] != '\n');
420 while (l>0 && isspace((unsigned char) buf[l-1])) l--;
425 while ((c= getc(file)) != EOF && c != '\n');
429 pconfig(buf,truncated);
432 badusage("failed while reading configuration file: %s", strerror(errno));
437 static void pconfig(const char *configstr, int truncated) {
438 unsigned long fgid, tgid, pprefix, pmask;
439 int plen, localonly, plus, rangeix, delim;
444 switch (configstr[0]) {
447 permit_range(0UL,0UL,1,0);
454 if (truncated) badusage("filename too long (`%.100s...')",configstr);
459 if (!isdigit((unsigned char)configstr[0]))
460 badusage("unknown configuration directive");
462 fgid= eat_number(&configstr,"gid", 0,gidmaxval, ",",0);
464 if (!proto) printf(" %5lu", fgid);
466 gidlist= getenv("USERV_GID");
467 if (!gidlist) fatal("USERV_GID not set");
470 if (!proto) printf(" no matching gid\n");
473 tgid= eat_number(&gidlist,"userv-gid", 0,gidmaxval, " ",0);
474 if (tgid == fgid) break;
477 if (configstr[0] == '=') {
488 switch (configstr[0]) {
489 case '-': plus= 0; /* fall through */
490 case '+': configstr++;
495 sprintf(whattxt, "%s-prefix#%d",
496 plus ? "permitted" : "notpermitted",
498 eat_prefixmask(&configstr,whattxt, ",+-",&delim,
499 &pprefix,&pmask,&plen);
500 if (!configstr && truncated)
501 badusage("gid,prefix/len,... spec too long");
504 printf(" %c%s/%d:", plus?'+':'-',ip2txt(pprefix,ptxt), plen);
506 permit_range(pprefix,pmask,plus,localonly);
507 if (delim==',') break;
509 plus= delim=='-' ? 0 : 1;
518 static void checkallow(int allow, const char *what,
519 const char *prefixtxt, const char *masktxt) {
521 fprintf(stderr,"userv-ipif service: access denied for %s, %s/%s\n",
522 what, prefixtxt, masktxt);
526 static void parseargs(int argc, const char *const *argv) {
527 unsigned long routeaddr, routemask;
529 const char *const *cprotop;
531 char erwhatbuf[100], erwhatbuf2[100];
533 if (argc < NARGS+1) { badusage("too few arguments"); }
534 if (argc > NARGS+1) { badusage("too many arguments"); }
539 if (strcmp(carg,"--")) badusage("separator argument `--' not found, got `%s'",carg);
542 localaddr= eat_addr(&carg,"local-addr", ",",0);
543 peeraddr= eat_addr(&carg,"peer-addr", ",",0);
544 mtu= eat_number(&carg,"mtu", 576,65536, ",",0);
545 localallow= peerallow= 0;
547 if (!strcmp(carg,"debug")) {
550 for (cprotop= protos_ok;
551 (proto= *cprotop) && strcmp(proto,carg);
553 if (!proto) fatal("invalid protocol");
556 addrnet_mustdiffer("local-addr",localaddr,~0UL, "peer-addr",peeraddr,~0UL);
559 if (strcmp(carg,"-")) {
563 if (nexroutes == MAXEXROUTES)
564 fatal("too many extra routes (only %d allowed)",MAXEXROUTES);
565 sprintf(erwhatbuf,"route#%d",nexroutes);
567 eat_prefixmask(&carg,erwhatbuf, ",",0, &routeaddr,&routemask,0);
568 if (routemask == ~0UL) {
569 addrnet_mustdiffer(erwhatbuf,routeaddr,routemask, "local-addr",localaddr,~0UL);
570 addrnet_mustdiffer(erwhatbuf,routeaddr,routemask, "peer-addr",peeraddr,~0UL);
572 for (i=0; i<nexroutes; i++) {
573 sprintf(erwhatbuf2,"route#%d",i);
574 addrnet_mustdiffer(erwhatbuf,routeaddr,routemask,
575 erwhatbuf2,exroutes[i].prefix,exroutes[i].mask);
577 exroutes[nexroutes].prefix= routeaddr;
578 exroutes[nexroutes].mask= routemask;
579 exroutes[nexroutes].allow= 0;
580 ip2txt(routeaddr,exroutes[nexroutes].prefixtxt);
581 ip2txt(routemask,exroutes[nexroutes].masktxt);
585 ip2txt(localaddr,localtxt);
586 ip2txt(peeraddr,peertxt);
589 static void checkpermit(void) {
594 checkallow(localallow,"local-addr", localtxt,"32");
595 checkallow(peerallow,"peer-addr", peertxt,"32");
596 for (i=0; i<nexroutes; i++) {
597 sprintf(erwhatbuf, "route#%d", i);
598 checkallow(exroutes[i].allow, erwhatbuf, exroutes[i].prefixtxt, exroutes[i].masktxt);
600 if (!allallow) fatal("access denied");
603 static void dumpdebug(void) __attribute__((noreturn));
604 static void dumpdebug(void) {
608 printf("protocol: debug\n"
609 "local: %08lx == %s\n"
610 "peer: %08lx == %s\n"
617 for (i=0; i<nexroutes; i++) {
618 sprintf(erwhatbuf, "route#%d:", i);
619 printf("%-9s %08lx/%08lx == %s/%s\n",
621 exroutes[i].prefix, exroutes[i].mask,
622 exroutes[i].prefixtxt, exroutes[i].masktxt);
624 if (ferror(stdout) || fclose(stdout)) sysfatal("flush stdout");
629 static int task(const char *desc) {
634 if (pid == (pid_t)-1) sysfatal("fork for task");
638 pidr= waitpid(pid,&status,0);
639 if (pidr!=(pid_t)-1) break;
640 if (errno==EINTR) continue;
641 sysfatal("waitpid for task");
645 if (WIFEXITED(status)) {
646 if (WEXITSTATUS(status))
647 fatal("userv-ipif service: %s exited with error exit status %d\n",
648 desc, WEXITSTATUS(status));
649 } else if (WIFSIGNALED(status)) {
650 fatal("userv-ipif service: %s died due to signal %s%s\n",
651 desc, strsignal(WTERMSIG(status)),
652 WCOREDUMP(status) ? " (core dumped)" : "");
654 fatal("userv-ipif service: %s unexpectedly terminated"
655 " with unknown status code %d\n", desc, status);
661 static void createif(void) {
662 static const char ifnamepat[]= "userv%d";
666 memset(&ifr,0,sizeof(ifr));
667 ifr.ifr_flags= IFF_TUN | IFF_NO_PI;
669 assert(sizeof(ifr.ifr_name) >= sizeof(ifnamepat));
670 strcpy(ifr.ifr_name, ifnamepat);
672 tunfd= open("/dev/net/tun", O_RDWR);
673 if (!tunfd) sysfatal("open /dev/net/tun");
675 r= fcntl(tunfd, F_GETFD);
676 if (r==-1) sysfatal("fcntl(tunfd,F_GETFD)");
677 r= fcntl(tunfd, F_SETFD, r|FD_CLOEXEC);
678 if (r==-1) sysfatal("fcntl(tunfd,F_SETFD,|FD_CLOEXEC)");
680 r= ioctl(tunfd, TUNSETIFF, (void*)&ifr);
681 if (r) sysfatal("ioctl TUNSETIFF");
683 /* ifr.ifr_name might not be null-terminated. crazy abi. */
684 ifname= malloc(sizeof(ifr.ifr_name)+1);
685 if (!ifname) sysfatal("malloc for interface name");
686 memcpy(ifname, ifr.ifr_name, sizeof(ifr.ifr_name));
687 ifname[sizeof(ifr.ifr_name)]= 0;
690 static void netconfigure(void) {
694 if (task("ifconfig")) {
695 sprintf(mtutxt,"%lu",mtu);
697 execlp("ifconfig", "ifconfig", ifname, localtxt,
698 "netmask","255.255.255.255", "pointopoint",peertxt, "-broadcast",
699 "mtu",mtutxt, "up", (char*)0);
700 sysfatal("cannot exec ifconfig");
703 for (i=0; i<nexroutes; i++) {
705 execlp("route","route", "add", "-net",exroutes[i].prefixtxt,
706 "netmask",exroutes[i].masktxt,
707 "gw",peertxt, "dev",ifname, (char*)0);
708 sysfatal("cannot exec route (for route)");
713 static void setnonblock(int fd) {
715 r= fcntl(fd,F_GETFL);
716 if (r==-1) sysfatal("fcntl F_GETFL");
717 r= fcntl(fd,F_SETFL, r|O_NONBLOCK);
718 if (r==-1) sysfatal("fcntl F_SETFL O_NONBLOCK");
721 static void rx_packet(const uint8_t *packet, int len) {
723 int r= write(tunfd, packet, len);
725 if (errno==EINTR) continue;
726 if (errno==EAGAIN) return; /* oh well */
727 sysfatal("error writing packet to tun (transmitting)");
734 static int output_waiting, input_waiting;
736 #define SLIP_END 0300
737 #define SLIP_ESC 0333
738 #define SLIP_ESC_END 0334
739 #define SLIP_ESC_ESC 0335
741 static void more_rx_data(uint8_t *input_buf, uint8_t *output_buf) {
742 /* we make slip_data never contain continuation of a packet */
743 /* input_buf is passed as a parameter since it's in copydata's stack frame */
745 static int output_len;
747 uint8_t *op= output_buf + output_len;
748 const uint8_t *ip= input_buf + scanned;
749 const uint8_t *ip_end= input_buf + input_waiting;
753 if (ip>=ip_end) break;
756 rx_packet(output_buf, op-output_buf);
758 eaten= ip - input_buf;
762 if (ip>=ip_end) { /* rescan this when there's more */ ip--; break; }
764 if (c==SLIP_ESC_END) c=SLIP_END;
765 else if (c==SLIP_ESC_ESC) c=SLIP_ESC;
766 else fatal("unexpected byte 0%o after SLIP_ESC",c);
768 if (op == output_buf+mtu)
769 fatal("SLIP packet exceeds mtu");
773 output_len= op - output_buf;
774 scanned= ip - input_buf;
776 input_waiting -= eaten;
777 memmove(input_buf, input_buf+eaten, input_waiting);
781 static void tx_packet(uint8_t *output_buf, const uint8_t *ip, int inlen) {
782 /* output_buf is passed as a parameter since it's in copydata's stack frame */
783 assert(!output_waiting);
784 uint8_t *op= output_buf;
789 if (c==SLIP_END) { *op++= SLIP_ESC; *op++= SLIP_ESC_END; }
790 else if (c==SLIP_ESC) { *op++= SLIP_ESC; *op++= SLIP_ESC_ESC; }
794 assert(op <= output_buf + mtu*2+2);
796 output_waiting= op - output_buf;
799 static void copydata(void) __attribute__((noreturn));
800 static void copydata(void) {
801 uint8_t output_buf[mtu*2+2];
802 uint8_t input_buf[mtu*2+2];
803 uint8_t rx_packet_buf[mtu];
807 struct pollfd polls[3];
808 memset(polls, 0, sizeof(polls));
810 polls[0].fd= 0; polls[0].events= POLLIN;
814 /* We don't do flow control on input packets; instead, we just throw
815 * away ones which the kernel doesn't accept. So we always poll for
818 * Output packets we buffer, so we poll only as appropriate for those.
821 /* Start by transmitting one END byte to say we're ready. */
822 output_buf[0]= SLIP_END;
826 if (output_waiting) {
827 r= write(1, output_buf, output_waiting);
829 if (errno==EINTR) continue;
831 sysfatal("error writing SLIP output (packets being received)");
835 memmove(output_buf, output_buf+r, output_waiting);
838 if (output_waiting) {
839 polls[1].events |= POLLOUT;
840 polls[2].events &= ~POLLIN;
842 polls[1].events &= ~POLLOUT;
843 polls[2].events |= POLLIN;
848 if (errno==EINTR) continue;
849 sysfatal("poll() failed");
851 assert(r>0); /* we used an infinite timeout */
853 for (i=0; i<sizeof(polls)/sizeof(polls[0]); i++)
854 if (polls[i].revents & ~polls[i].events)
855 fatal("unexpected revents 0x%x for fd=%d",
856 polls[i].revents, polls[i].fd);
858 if (polls[0].events & POLLIN) {
859 int want= sizeof(input_buf) - input_waiting;
860 if (want<0) fatal("incoming packet necessarily exceeds MTU");
861 r= read(0, input_buf + input_waiting, want);
864 assert(r < sizeof(input_buf));
865 more_rx_data(input_buf, rx_packet_buf);
869 if (!(errno==EINTR || errno==EAGAIN))
870 sysfatal("error reading input SLIP data (packets to transmit)");
874 /* We handle what would be (polls[1].events & POLLOUT) above,
875 * unconditionally. That eliminates the need to poll in the usual case */
877 if (polls[2].events & POLLIN) {
878 uint8_t packet_buf[mtu];
879 r= read(tunfd, packet_buf, mtu);
881 tx_packet(output_buf, packet_buf, r);
884 if (!(errno==EAGAIN || errno==EWOULDBLOCK))
885 sysfatal("error reading packet (being transmitted) from tun");
891 int main(int argc, const char *const *argv) {
892 parseargs(argc,argv);
893 pconfig(configstr,0);
895 if (!proto) dumpdebug();