| 1 | /* polypath |
| 2 | * send/receive module for secnet |
| 3 | * for multi-route setups */ |
| 4 | /* |
| 5 | * This file is part of secnet. |
| 6 | * See README for full list of copyright holders. |
| 7 | * |
| 8 | * secnet is free software; you can redistribute it and/or modify it |
| 9 | * under the terms of the GNU General Public License as published by |
| 10 | * the Free Software Foundation; either version d of the License, or |
| 11 | * (at your option) any later version. |
| 12 | * |
| 13 | * secnet is distributed in the hope that it will be useful, but |
| 14 | * WITHOUT ANY WARRANTY; without even the implied warranty of |
| 15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 16 | * General Public License for more details. |
| 17 | * |
| 18 | * You should have received a copy of the GNU General Public License |
| 19 | * version 3 along with secnet; if not, see |
| 20 | * https://www.gnu.org/licenses/gpl.html. |
| 21 | */ |
| 22 | |
| 23 | #include "secnet.h" |
| 24 | #include "util.h" |
| 25 | #include "unaligned.h" |
| 26 | #include "comm-common.h" |
| 27 | |
| 28 | #include <adns.h> |
| 29 | #include <ctype.h> |
| 30 | |
| 31 | #ifdef CONFIG_IPV6 |
| 32 | |
| 33 | static comm_sendmsg_fn polypath_sendmsg; |
| 34 | |
| 35 | struct interf { |
| 36 | char *name; /* from malloc */ |
| 37 | struct udpsocks socks; |
| 38 | bool_t experienced_xmit_noaf[MAX_AF+1]; |
| 39 | LIST_ENTRY(interf) entry; |
| 40 | }; |
| 41 | |
| 42 | struct polypath { |
| 43 | struct udpcommon uc; |
| 44 | int max_interfs; |
| 45 | const char *const *ifname_pats; |
| 46 | const char *const *monitor_command; |
| 47 | bool_t permit_loopback; |
| 48 | LIST_HEAD(,interf) interfs; |
| 49 | struct buffer_if lbuf; |
| 50 | int monitor_fd; |
| 51 | pid_t monitor_pid; |
| 52 | int privsep_incoming_fd; |
| 53 | int privsep_ipcsock_fd; |
| 54 | }; |
| 55 | |
| 56 | static void polypath_phase_shutdown(void *sst, uint32_t newphase); |
| 57 | |
| 58 | #define LG 0, st->uc.cc.cl.description, &st->uc.cc.loc |
| 59 | |
| 60 | static const char *const default_loopback_ifname_pats[] = { |
| 61 | "!lo", 0 |
| 62 | }; |
| 63 | static const char *const default_ifname_pats[] = { |
| 64 | "!tun*","!tap*","!sl*","!userv*", "*", 0 |
| 65 | }; |
| 66 | |
| 67 | static const char *const default_monitor_command[] = { |
| 68 | #if __linux__ |
| 69 | DATAROOTDIR "/secnet/" "polypath-interface-monitor-linux", 0 |
| 70 | #else |
| 71 | 0 |
| 72 | #endif |
| 73 | }; |
| 74 | |
| 75 | static const char *polypath_addr_to_string(void *commst, |
| 76 | const struct comm_addr *ca) |
| 77 | { |
| 78 | static char sbuf[100]; |
| 79 | |
| 80 | snprintf(sbuf, sizeof(sbuf), "polypath:%s", |
| 81 | iaddr_to_string(&ca->ia)); |
| 82 | return sbuf; |
| 83 | } |
| 84 | |
| 85 | static bool_t ifname_search_pats(struct polypath *st, struct cloc loc, |
| 86 | const char *ifname, bool_t *want_io, |
| 87 | const char *const *pats) { |
| 88 | /* Returns True iff we found a list entry, in which case *want_io |
| 89 | * is set to the sense of that entry. Otherwise *want_io is set |
| 90 | * to the sense of the last entry, or unchanged if there were no pats. */ |
| 91 | if (!pats) |
| 92 | return False; |
| 93 | const char *const *pati; |
| 94 | for (pati=pats; *pati; pati++) { |
| 95 | const char *pat=*pati; |
| 96 | if (*pat=='!') { *want_io=False; pat++; } |
| 97 | else if (*pat=='+') { *want_io=True; pat++; } |
| 98 | else if (*pat=='*' || isalnum((unsigned char)*pat)) { *want_io=True; } |
| 99 | else cfgfatal(loc,"polypath","invalid interface name pattern `%s'",pat); |
| 100 | int match=fnmatch(pat,ifname,0); |
| 101 | if (match==0) return True; |
| 102 | if (match!=FNM_NOMATCH) |
| 103 | cfgfatal(loc,"polypath","fnmatch failed! (pattern `%s')",pat); |
| 104 | } |
| 105 | return False; |
| 106 | } |
| 107 | |
| 108 | static bool_t ifname_wanted(struct polypath *st, struct cloc loc, |
| 109 | const char *ifname) { |
| 110 | bool_t want=False; /* pretend an empty cfg ends with !<doesn'tmatch> */ |
| 111 | if (ifname_search_pats(st,loc,ifname,&want, st->ifname_pats)) |
| 112 | return want; |
| 113 | if (want) /* last pattern was positive, do not search default */ |
| 114 | return False; |
| 115 | if (!st->permit_loopback && |
| 116 | ifname_search_pats(st,loc,ifname,&want, default_loopback_ifname_pats)) |
| 117 | return want; |
| 118 | if (ifname_search_pats(st,loc,ifname,&want, default_ifname_pats)) |
| 119 | return want; |
| 120 | abort(); |
| 121 | } |
| 122 | |
| 123 | static int polypath_beforepoll(void *state, struct pollfd *fds, int *nfds_io, |
| 124 | int *timeout_io) |
| 125 | { |
| 126 | struct polypath *st=state; |
| 127 | BEFOREPOLL_WANT_FDS(1); |
| 128 | fds[0].fd=st->monitor_fd; |
| 129 | fds[0].events=POLLIN; |
| 130 | return 0; |
| 131 | } |
| 132 | |
| 133 | static inline bool_t matches32(uint32_t word, uint32_t prefix, int prefixlen) |
| 134 | { |
| 135 | assert(prefixlen>0); |
| 136 | assert(prefixlen<=32); |
| 137 | uint32_t mask = ~(((uint32_t)1 << (32-prefixlen)) - 1); |
| 138 | assert(!(prefix & ~mask)); |
| 139 | return (word & mask) == prefix; |
| 140 | } |
| 141 | |
| 142 | /* These macros expect |
| 143 | * bad_fn_type *const bad; |
| 144 | * void *badctx; |
| 145 | * and |
| 146 | * out: |
| 147 | */ |
| 148 | #define BAD(m) do{ bad(st,badctx,M_WARNING,m,0); goto out; }while(0) |
| 149 | #define BADE(m,ev) do{ bad(st,badctx,M_WARNING,m,ev); goto out; }while(0) |
| 150 | typedef void bad_fn_type(struct polypath *st, void *badctx, |
| 151 | int mclass, const char* m, int ev); |
| 152 | |
| 153 | typedef void polypath_ppml_callback_type(struct polypath *st, |
| 154 | bad_fn_type *bad, void *badctx, |
| 155 | bool_t add, const char *ifname, const char *ifaddr, |
| 156 | const union iaddr *ia, int fd /* -1 if none yet */); |
| 157 | |
| 158 | struct ppml_bad_ctx { |
| 159 | const char *orgl; |
| 160 | char *undospace; |
| 161 | }; |
| 162 | |
| 163 | static void ppml_bad(struct polypath *st, void *badctx, |
| 164 | int mclass, const char *m, int ev) |
| 165 | { |
| 166 | struct ppml_bad_ctx *bc=badctx; |
| 167 | if (bc->undospace) |
| 168 | *(bc->undospace)=' '; |
| 169 | lg_perror(LG,mclass,ev, |
| 170 | "error processing polypath state change: %s" |
| 171 | " (while processing `%s')", |
| 172 | m,bc->orgl); |
| 173 | } |
| 174 | |
| 175 | static void polypath_process_monitor_line(struct polypath *st, char *orgl, |
| 176 | polypath_ppml_callback_type *callback) |
| 177 | /* always calls callback with fd==-1 */ |
| 178 | { |
| 179 | struct udpcommon *uc=&st->uc; |
| 180 | char *l=orgl; |
| 181 | bad_fn_type (*const bad)=ppml_bad; |
| 182 | struct ppml_bad_ctx badctx[1]={{ |
| 183 | .orgl=orgl, |
| 184 | .undospace=0 |
| 185 | }}; |
| 186 | |
| 187 | bool_t add; |
| 188 | int c=*l++; |
| 189 | if (c=='+') add=True; |
| 190 | else if (c=='-') add=False; |
| 191 | else BAD("bad +/-"); |
| 192 | |
| 193 | int proto; |
| 194 | c=*l++; |
| 195 | if (c=='4') proto=AF_INET; |
| 196 | else if (c=='6') proto=AF_INET6; |
| 197 | else BAD("bad proto"); |
| 198 | |
| 199 | char *space=strchr(l,' '); |
| 200 | if (!space) BAD("no first space"); |
| 201 | const char *ifname=space+1; |
| 202 | |
| 203 | space=strchr(ifname,' '); |
| 204 | if (!space) BAD("no second space"); |
| 205 | const char *ifaddr=space+1; |
| 206 | *space=0; |
| 207 | badctx->undospace=space; |
| 208 | |
| 209 | union iaddr ia; |
| 210 | FILLZERO(ia); |
| 211 | socklen_t salen=sizeof(ia); |
| 212 | int r=adns_text2addr(ifaddr,uc->port, adns_qf_addrlit_ipv4_quadonly, |
| 213 | &ia.sa, &salen); |
| 214 | assert(r!=ENOSPC); |
| 215 | if (r) BADE("adns_text2addr",r); |
| 216 | if (ia.sa.sa_family!=proto) BAD("address family mismatch"); |
| 217 | |
| 218 | #define DONT(m) do{ \ |
| 219 | if (add) \ |
| 220 | lg_perror(LG,M_INFO,0,"ignoring %s [%s]: %s",ifname,ifaddr,m); \ |
| 221 | goto out; \ |
| 222 | }while(0) |
| 223 | |
| 224 | if (!ifname_wanted(st,st->uc.cc.loc,ifname)) |
| 225 | DONT("unwanted interface name"); |
| 226 | |
| 227 | switch (ia.sa.sa_family) { |
| 228 | case AF_INET6: { |
| 229 | const struct in6_addr *i6=&ia.sin6.sin6_addr; |
| 230 | #define DONTKIND(X,m) \ |
| 231 | if (IN6_IS_ADDR_##X(i6)) DONT("IPv6 address is " m) |
| 232 | DONTKIND(UNSPECIFIED, "unspecified"); |
| 233 | DONTKIND(MULTICAST , "multicast" ); |
| 234 | DONTKIND(LINKLOCAL , "link local" ); |
| 235 | DONTKIND(SITELOCAL , "site local" ); |
| 236 | DONTKIND(V4MAPPED , "v4-mapped" ); |
| 237 | if (!st->permit_loopback) |
| 238 | DONTKIND(LOOPBACK , "loopback" ); |
| 239 | #undef DONTKIND |
| 240 | #define DONTMASK(w7x,w6x,prefixlen,m) \ |
| 241 | if (matches32(get_uint32(i6->s6_addr), \ |
| 242 | ((uint32_t)0x##w7x << 16) | (uint32_t)0x##w6x, \ |
| 243 | prefixlen)) \ |
| 244 | DONT("IPv6 address is " m) |
| 245 | DONTMASK( 100, 0, 8, "Discard-Only (RFC6666)"); |
| 246 | DONTMASK(2001, 0, 23, "in IETF protocol block (RFC2928)"); |
| 247 | DONTMASK(fc00, 0, 7, "Uniqe Local unicast (RFC4193)"); |
| 248 | #undef DONTMASK |
| 249 | break; |
| 250 | } |
| 251 | case AF_INET: { |
| 252 | const uint32_t i4=htonl(ia.sin.sin_addr.s_addr); |
| 253 | if (i4==INADDR_ANY) DONT("IPv4 address is any/unspecified"); |
| 254 | if (i4==INADDR_BROADCAST) DONT("IPv4 address is all hosts broadcast"); |
| 255 | #define DONTMASK(b3,b2,b1,b0,prefixlen,m) do{ \ |
| 256 | const uint8_t prefixbytes[4] = { (b3),(b2),(b1),(b0) }; \ |
| 257 | if (matches32(i4,get_uint32(prefixbytes),prefixlen)) \ |
| 258 | DONT("IPv4 address is " m); \ |
| 259 | }while(0) |
| 260 | DONTMASK(169,254,0,0, 16, "link local"); |
| 261 | DONTMASK(224, 0,0,0, 4, "multicast"); |
| 262 | DONTMASK(192, 0,0,0, 24, "in IETF protocol block (RFC6890)"); |
| 263 | DONTMASK(240, 0,0,0, 4, "in reserved addressing block (RFC1112)"); |
| 264 | if (!st->permit_loopback) |
| 265 | DONTMASK(127, 0,0,0, 8, "loopback"); |
| 266 | #undef DONTMASK |
| 267 | break; |
| 268 | } |
| 269 | default: |
| 270 | abort(); |
| 271 | } |
| 272 | |
| 273 | #undef DONT |
| 274 | |
| 275 | /* OK, process it */ |
| 276 | callback(st, bad,badctx, add,ifname,ifaddr,&ia,-1); |
| 277 | |
| 278 | out:; |
| 279 | } |
| 280 | |
| 281 | static void dump_pria(struct polypath *st, const char *ifname) |
| 282 | { |
| 283 | #ifdef POLYPATH_DEBUG |
| 284 | struct interf *interf; |
| 285 | if (ifname) |
| 286 | lg_perror(LG,M_DEBUG,0, "polypath record ifaddr `%s'",ifname); |
| 287 | LIST_FOREACH(interf, &st->interfs, entry) { |
| 288 | lg_perror(LG,M_DEBUG,0, " polypath interface `%s', nsocks=%d", |
| 289 | interf->name, interf->socks.n_socks); |
| 290 | int i; |
| 291 | for (i=0; i<interf->socks.n_socks; i++) { |
| 292 | struct udpsock *us=&interf->socks.socks[i]; |
| 293 | lg_perror(LG,M_DEBUG,0, " polypath sock fd=%d addr=%s", |
| 294 | us->fd, iaddr_to_string(&us->addr)); |
| 295 | } |
| 296 | } |
| 297 | #endif |
| 298 | } |
| 299 | |
| 300 | static bool_t polypath_make_socket(struct polypath *st, |
| 301 | bad_fn_type *bad, void *badctx, |
| 302 | struct udpsock *us, const char *ifname) |
| 303 | /* on error exit has called bad; might leave us->fd as -1 */ |
| 304 | { |
| 305 | assert(us->fd==-1); |
| 306 | |
| 307 | bool_t ok=udp_make_socket(&st->uc,us,M_WARNING); |
| 308 | if (!ok) BAD("unable to set up socket"); |
| 309 | int r=setsockopt(us->fd,SOL_SOCKET,SO_BINDTODEVICE, |
| 310 | ifname,strlen(ifname)+1); |
| 311 | if (r) BADE("setsockopt(,,SO_BINDTODEVICE,)",errno); |
| 312 | return True; |
| 313 | |
| 314 | out: |
| 315 | return False; |
| 316 | } |
| 317 | |
| 318 | static void polypath_record_ifaddr(struct polypath *st, |
| 319 | bad_fn_type *bad, void *badctx, |
| 320 | bool_t add, const char *ifname, |
| 321 | const char *ifaddr, |
| 322 | const union iaddr *ia, int fd) |
| 323 | { |
| 324 | struct udpcommon *uc=&st->uc; |
| 325 | struct interf *interf=0; |
| 326 | struct udpsock *us=0; |
| 327 | |
| 328 | dump_pria(st,ifname); |
| 329 | |
| 330 | int n_ifs=0; |
| 331 | LIST_FOREACH(interf,&st->interfs,entry) { |
| 332 | if (!strcmp(interf->name,ifname)) |
| 333 | goto found_interf; |
| 334 | n_ifs++; |
| 335 | } |
| 336 | /* not found */ |
| 337 | if (n_ifs==st->max_interfs) BAD("too many interfaces"); |
| 338 | interf=malloc(sizeof(*interf)); |
| 339 | if (!interf) BADE("malloc for new interface",errno); |
| 340 | interf->name=0; |
| 341 | interf->socks.n_socks=0; |
| 342 | FILLZERO(interf->experienced_xmit_noaf); |
| 343 | LIST_INSERT_HEAD(&st->interfs,interf,entry); |
| 344 | interf->name=strdup(ifname); |
| 345 | udp_socks_register(&st->uc,&interf->socks,interf->name); |
| 346 | if (!interf->name) BADE("strdup interface name",errno); |
| 347 | found_interf: |
| 348 | |
| 349 | if (add) { |
| 350 | if (interf->socks.n_socks == UDP_MAX_SOCKETS) |
| 351 | BAD("too many addresses on this interface"); |
| 352 | struct udpsock *us=&interf->socks.socks[interf->socks.n_socks]; |
| 353 | us->fd=-1; |
| 354 | COPY_OBJ(us->addr,*ia); |
| 355 | if (fd<0) { |
| 356 | bool_t ok=polypath_make_socket(st,bad,badctx, us,ifname); |
| 357 | if (!ok) goto out; |
| 358 | } else { |
| 359 | bool_t ok=udp_import_socket(uc,us,M_WARNING,fd); |
| 360 | if (!ok) goto out; |
| 361 | fd=-1; |
| 362 | } |
| 363 | interf->socks.n_socks++; |
| 364 | lg_perror(LG,M_INFO,0,"using %s %s",ifname, |
| 365 | iaddr_to_string(&us->addr)); |
| 366 | us=0; /* do not destroy this socket during `out' */ |
| 367 | } else { |
| 368 | int i; |
| 369 | for (i=0; i<interf->socks.n_socks; i++) |
| 370 | if (iaddr_equal(&interf->socks.socks[i].addr,ia,True)) |
| 371 | goto address_remove_found; |
| 372 | bad(st,badctx,M_DEBUG,"address to remove not found",0); |
| 373 | goto out; |
| 374 | address_remove_found: |
| 375 | lg_perror(LG,M_INFO,0,"removed %s %s",ifname, |
| 376 | iaddr_to_string(&interf->socks.socks[i].addr)); |
| 377 | udp_destroy_socket(&st->uc,&interf->socks.socks[i]); |
| 378 | interf->socks.socks[i]= |
| 379 | interf->socks.socks[--interf->socks.n_socks]; |
| 380 | } |
| 381 | |
| 382 | out: |
| 383 | if (us) |
| 384 | udp_destroy_socket(uc,us); |
| 385 | if (fd>=0) |
| 386 | close(fd); |
| 387 | if (interf && !interf->socks.n_socks) { |
| 388 | udp_socks_deregister(&st->uc,&interf->socks); |
| 389 | LIST_REMOVE(interf,entry); |
| 390 | free(interf->name); |
| 391 | free(interf); |
| 392 | } |
| 393 | |
| 394 | dump_pria(st,0); |
| 395 | } |
| 396 | |
| 397 | static void subproc_problem(struct polypath *st, |
| 398 | enum async_linebuf_result alr, const char *emsg) |
| 399 | { |
| 400 | int status; |
| 401 | assert(st->monitor_pid); |
| 402 | |
| 403 | pid_t gotpid=waitpid(st->monitor_pid,&status,WNOHANG); |
| 404 | if (gotpid==st->monitor_pid) { |
| 405 | st->monitor_pid=0; |
| 406 | lg_exitstatus(LG,M_FATAL,status,"interface monitor"); |
| 407 | } else if (gotpid<0) |
| 408 | lg_perror(LG,M_ERR,errno,"unable to reap interface monitor"); |
| 409 | else |
| 410 | assert(gotpid==0); |
| 411 | |
| 412 | if (alr==async_linebuf_eof) |
| 413 | lg_perror(LG,M_FATAL,0,"unexpected EOF from interface monitor"); |
| 414 | else |
| 415 | lg_perror(LG,M_FATAL,0,"bad output from interface monitor: %s",emsg); |
| 416 | assert(!"not reached"); |
| 417 | } |
| 418 | |
| 419 | /* Used in non-privsep case, and in privsep child */ |
| 420 | static void afterpoll_monitor(struct polypath *st, struct pollfd *fd, |
| 421 | polypath_ppml_callback_type *callback) |
| 422 | { |
| 423 | enum async_linebuf_result alr; |
| 424 | const char *emsg; |
| 425 | |
| 426 | while ((alr=async_linebuf_read(fd,&st->lbuf,&emsg)) == async_linebuf_ok) |
| 427 | polypath_process_monitor_line(st,st->lbuf.base,callback); |
| 428 | |
| 429 | if (alr==async_linebuf_nothing) |
| 430 | return; |
| 431 | |
| 432 | subproc_problem(st,alr,emsg); |
| 433 | } |
| 434 | |
| 435 | /* Used in non-privsep case only - glue for secnet main loop */ |
| 436 | static void polypath_afterpoll_monitor(void *state, struct pollfd *fds, |
| 437 | int nfds) |
| 438 | { |
| 439 | struct polypath *st=state; |
| 440 | if (nfds<1) return; |
| 441 | afterpoll_monitor(st,fds,polypath_record_ifaddr); |
| 442 | } |
| 443 | |
| 444 | /* Actual udp packet sending work */ |
| 445 | static bool_t polypath_sendmsg(void *commst, struct buffer_if *buf, |
| 446 | const struct comm_addr *dest) |
| 447 | { |
| 448 | struct polypath *st=commst; |
| 449 | struct interf *interf; |
| 450 | bool_t allreasonable=True; |
| 451 | int af=dest->ia.sa.sa_family; |
| 452 | |
| 453 | LIST_FOREACH(interf,&st->interfs,entry) { |
| 454 | int i; |
| 455 | bool_t attempted=False, reasonable=False; |
| 456 | for (i=0; i<interf->socks.n_socks; i++) { |
| 457 | struct udpsock *us=&interf->socks.socks[i]; |
| 458 | if (af != us->addr.sa.sa_family) |
| 459 | continue; |
| 460 | attempted=True; |
| 461 | int r=sendto(us->fd,buf->start,buf->size, |
| 462 | 0,&dest->ia.sa,iaddr_socklen(&dest->ia)); |
| 463 | udp_sock_experienced(0,&st->uc,&interf->socks,us, |
| 464 | &dest->ia,af, r,errno); |
| 465 | if (r>=0) { |
| 466 | reasonable=True; |
| 467 | break; |
| 468 | } |
| 469 | if (!(errno==EAFNOSUPPORT || errno==ENETUNREACH)) |
| 470 | reasonable=True; |
| 471 | lg_perror(LG,M_DEBUG,errno,"%s [%s] xmit %"PRIu32" bytes to %s", |
| 472 | interf->name,iaddr_to_string(&us->addr), |
| 473 | buf->size,iaddr_to_string(&dest->ia)); |
| 474 | } |
| 475 | if (!attempted) |
| 476 | if (!interf->experienced_xmit_noaf[af]++) |
| 477 | lg_perror(LG,M_WARNING,0, |
| 478 | "%s has no suitable address to transmit %s", |
| 479 | interf->name, af_name(af)); |
| 480 | allreasonable *= reasonable; |
| 481 | } |
| 482 | return allreasonable; |
| 483 | } |
| 484 | |
| 485 | /* Non-privsep: called in (sole) child. Privsep: in grandchild. */ |
| 486 | static void child_monitor(struct polypath *st, int childfd) |
| 487 | { |
| 488 | dup2(childfd,1); |
| 489 | execvp(st->monitor_command[0],(char**)st->monitor_command); |
| 490 | fprintf(stderr,"secnet: cannot execute %s: %s\n", |
| 491 | st->monitor_command[0], strerror(errno)); |
| 492 | exit(-1); |
| 493 | } |
| 494 | |
| 495 | /* General utility function. */ |
| 496 | static void start_subproc(struct polypath *st, void (*make_fdpair)(int[2]), |
| 497 | void (*child)(struct polypath *st, int childfd), |
| 498 | const char *desc) |
| 499 | { |
| 500 | int pfds[2]; |
| 501 | |
| 502 | assert(!st->monitor_pid); |
| 503 | assert(st->monitor_fd<0); |
| 504 | |
| 505 | make_fdpair(pfds); |
| 506 | |
| 507 | pid_t pid=fork(); |
| 508 | if (!pid) { |
| 509 | afterfork(); |
| 510 | close(pfds[0]); |
| 511 | child(st,pfds[1]); |
| 512 | abort(); |
| 513 | } |
| 514 | if (pid<0) |
| 515 | fatal_perror("%s: failed to fork for interface monitoring", |
| 516 | st->uc.cc.cl.description); |
| 517 | |
| 518 | close(pfds[1]); |
| 519 | st->monitor_pid=pid; |
| 520 | st->monitor_fd=pfds[0]; |
| 521 | setnonblock(st->monitor_fd); |
| 522 | |
| 523 | lg_perror(LG,M_NOTICE,0, "%s: spawning %s [pid %ld]", |
| 524 | st->uc.cc.cl.description, desc, (long)st->monitor_pid); |
| 525 | } |
| 526 | |
| 527 | /* Non-privsep only: glue for forking the monitor, from the main loop */ |
| 528 | static void polypath_phase_startmonitor(void *sst, uint32_t newphase) |
| 529 | { |
| 530 | struct polypath *st=sst; |
| 531 | start_subproc(st,pipe_cloexec,child_monitor, |
| 532 | "interface monitor (no privsep)"); |
| 533 | register_for_poll(st,polypath_beforepoll, |
| 534 | polypath_afterpoll_monitor,"polypath"); |
| 535 | } |
| 536 | |
| 537 | /*----- Privsep-only: -----*/ |
| 538 | |
| 539 | /* |
| 540 | * We use two subprocesses, a child and a grandchild. These are |
| 541 | * forked before secnet drops privilege. |
| 542 | * |
| 543 | * The grandchild is the same interface monitor helper script as used |
| 544 | * in the non-privsep case. But its lines are read by the child |
| 545 | * instead of by the main secnet. The child is responsible for |
| 546 | * creating the actual socket (binding it, etc.). Each socket is |
| 547 | * passed to secnet proper via fd passing, along with a data message |
| 548 | * describing the interface name and address. The child does not |
| 549 | * retain a list of current interfaces and addresses - it trusts the |
| 550 | * interface monitor to get that right. secnet proper still maintains |
| 551 | * that data structure. |
| 552 | * |
| 553 | * The result is that much of the non-privsep code can be reused, but |
| 554 | * just plumbed together differently. |
| 555 | * |
| 556 | * The child does not retain the socket after passing it on. |
| 557 | * Interface removals are handled similarly but without any fd. |
| 558 | * |
| 559 | * The result is that the configuration's limits on which interfaces |
| 560 | * and ports secnet may use are enforced by the privileged child. |
| 561 | */ |
| 562 | |
| 563 | struct privsep_mdata { |
| 564 | bool_t add; |
| 565 | char ifname[100]; |
| 566 | union iaddr ia; |
| 567 | }; |
| 568 | |
| 569 | static void papp_bad(struct polypath *st, void *badctx, |
| 570 | int mclass, const char *m, int ev) |
| 571 | { |
| 572 | const struct privsep_mdata *mdata=(const void*)st->lbuf.start; |
| 573 | const char *addr_str=badctx; |
| 574 | |
| 575 | lg_perror(LG,mclass,ev, |
| 576 | "error processing polypath address change %s %s [%s]: %s", |
| 577 | mdata->add ? "+" : "-", |
| 578 | mdata->ifname, addr_str, m); |
| 579 | } |
| 580 | |
| 581 | static void polypath_afterpoll_privsep(void *state, struct pollfd *fds, |
| 582 | int nfds) |
| 583 | /* In secnet proper; receives messages from child. */ |
| 584 | { |
| 585 | struct polypath *st=state; |
| 586 | |
| 587 | if (nfds<1) return; |
| 588 | |
| 589 | int revents=fds[0].revents; |
| 590 | |
| 591 | const char *badbit=pollbadbit(revents); |
| 592 | if (badbit) subproc_problem(st,async_linebuf_broken,badbit); |
| 593 | |
| 594 | if (!(revents & POLLIN)) return; |
| 595 | |
| 596 | for (;;) { |
| 597 | if (st->lbuf.size==sizeof(struct privsep_mdata)) { |
| 598 | const struct privsep_mdata *mdata=(const void*)st->lbuf.start; |
| 599 | if (mdata->add && st->privsep_incoming_fd<0) |
| 600 | fatal("polypath (privsep): got add message data but no fd"); |
| 601 | if (!mdata->add && st->privsep_incoming_fd>=0) |
| 602 | fatal("polypath (privsep): got remove message data with fd"); |
| 603 | if (!memchr(mdata->ifname,0,sizeof(mdata->ifname))) |
| 604 | fatal("polypath (privsep): got ifname with no terminating nul"); |
| 605 | int af=mdata->ia.sa.sa_family; |
| 606 | if (!(af==AF_INET6 || af==AF_INET)) |
| 607 | fatal("polypath (privsep): got message data but bad AF %d",af); |
| 608 | const char *addr_str=iaddr_to_string(&mdata->ia); |
| 609 | polypath_record_ifaddr(st,papp_bad,(void*)addr_str, |
| 610 | mdata->add,mdata->ifname,addr_str, |
| 611 | &mdata->ia, st->privsep_incoming_fd); |
| 612 | st->privsep_incoming_fd=-1; |
| 613 | st->lbuf.size=0; |
| 614 | } |
| 615 | struct msghdr msg; |
| 616 | int fd; |
| 617 | size_t cmsgdatalen=sizeof(fd); |
| 618 | char cmsg_control_buf[CMSG_SPACE(cmsgdatalen)]; |
| 619 | struct iovec iov; |
| 620 | |
| 621 | FILLZERO(msg); |
| 622 | msg.msg_iov=&iov; |
| 623 | msg.msg_iovlen=1; |
| 624 | |
| 625 | iov.iov_base=st->lbuf.start+st->lbuf.size; |
| 626 | iov.iov_len=sizeof(struct privsep_mdata)-st->lbuf.size; |
| 627 | |
| 628 | if (st->privsep_incoming_fd<0) { |
| 629 | msg.msg_control=cmsg_control_buf; |
| 630 | msg.msg_controllen=sizeof(cmsg_control_buf); |
| 631 | } |
| 632 | |
| 633 | ssize_t got=recvmsg(st->monitor_fd,&msg,0); |
| 634 | if (got<0) { |
| 635 | if (errno==EINTR) continue; |
| 636 | if (iswouldblock(errno)) break; |
| 637 | fatal_perror("polypath (privsep): recvmsg failed"); |
| 638 | } |
| 639 | if (got==0) |
| 640 | subproc_problem(st,async_linebuf_eof,0); |
| 641 | |
| 642 | st->lbuf.size+=got; |
| 643 | |
| 644 | if (msg.msg_controllen) { |
| 645 | size_t cmsgdatalen=sizeof(st->privsep_incoming_fd); |
| 646 | struct cmsghdr *h=CMSG_FIRSTHDR(&msg); |
| 647 | if (!(st->privsep_incoming_fd==-1 && |
| 648 | h && |
| 649 | h->cmsg_level==SOL_SOCKET && |
| 650 | h->cmsg_type==SCM_RIGHTS && |
| 651 | h->cmsg_len==CMSG_LEN(cmsgdatalen) && |
| 652 | !CMSG_NXTHDR(&msg,h))) |
| 653 | subproc_problem(st,async_linebuf_broken,"bad cmsg"); |
| 654 | memcpy(&st->privsep_incoming_fd,CMSG_DATA(h),cmsgdatalen); |
| 655 | assert(st->privsep_incoming_fd>=0); |
| 656 | } |
| 657 | |
| 658 | } |
| 659 | } |
| 660 | |
| 661 | static void privsep_handle_ifaddr(struct polypath *st, |
| 662 | bad_fn_type *bad, void *badctx, |
| 663 | bool_t add, const char *ifname, |
| 664 | const char *ifaddr, |
| 665 | const union iaddr *ia, int fd_dummy) |
| 666 | /* In child: handles discovered wanted interfaces, making sockets |
| 667 | and sending them to secnet proper. */ |
| 668 | { |
| 669 | struct msghdr msg; |
| 670 | struct iovec iov; |
| 671 | struct udpsock us={ .fd=-1 }; |
| 672 | size_t cmsgdatalen=sizeof(us.fd); |
| 673 | char cmsg_control_buf[CMSG_SPACE(cmsgdatalen)]; |
| 674 | |
| 675 | assert(fd_dummy==-1); |
| 676 | |
| 677 | struct privsep_mdata mdata; |
| 678 | FILLZERO(mdata); |
| 679 | mdata.add=add; |
| 680 | |
| 681 | size_t l=strlen(ifname); |
| 682 | if (l>=sizeof(mdata.ifname)) BAD("interface name too long"); |
| 683 | strcpy(mdata.ifname,ifname); |
| 684 | |
| 685 | COPY_OBJ(mdata.ia,*ia); |
| 686 | |
| 687 | iov.iov_base=&mdata; |
| 688 | iov.iov_len =sizeof(mdata); |
| 689 | |
| 690 | FILLZERO(msg); |
| 691 | msg.msg_iov=&iov; |
| 692 | msg.msg_iovlen=1; |
| 693 | |
| 694 | if (add) { |
| 695 | COPY_OBJ(us.addr,*ia); |
| 696 | bool_t ok=polypath_make_socket(st,bad,badctx,&us,ifname); |
| 697 | if (!ok) goto out; |
| 698 | |
| 699 | msg.msg_control=cmsg_control_buf; |
| 700 | msg.msg_controllen=sizeof(cmsg_control_buf); |
| 701 | |
| 702 | struct cmsghdr *h=CMSG_FIRSTHDR(&msg); |
| 703 | h->cmsg_level=SOL_SOCKET; |
| 704 | h->cmsg_type =SCM_RIGHTS; |
| 705 | h->cmsg_len =CMSG_LEN(cmsgdatalen); |
| 706 | memcpy(CMSG_DATA(h),&us.fd,cmsgdatalen); |
| 707 | } |
| 708 | |
| 709 | while (iov.iov_len) { |
| 710 | ssize_t got=sendmsg(st->privsep_ipcsock_fd,&msg,0); |
| 711 | if (got<0) { |
| 712 | if (errno!=EINTR) fatal_perror("polypath privsep sendmsg"); |
| 713 | got=0; |
| 714 | } else { |
| 715 | assert(got>0); |
| 716 | assert((size_t)got<=iov.iov_len); |
| 717 | } |
| 718 | iov.iov_base=(char*)iov.iov_base+got; |
| 719 | iov.iov_len-=got; |
| 720 | msg.msg_control=0; |
| 721 | msg.msg_controllen=0; |
| 722 | } |
| 723 | |
| 724 | out: |
| 725 | if (us.fd>=0) close(us.fd); |
| 726 | } |
| 727 | |
| 728 | static void child_privsep(struct polypath *st, int ipcsockfd) |
| 729 | /* Privsep child main loop. */ |
| 730 | { |
| 731 | struct pollfd fds[2]; |
| 732 | |
| 733 | enter_phase(PHASE_CHILDPERSIST); |
| 734 | |
| 735 | st->privsep_ipcsock_fd=ipcsockfd; |
| 736 | start_subproc(st,pipe_cloexec,child_monitor, |
| 737 | "interface monitor (grandchild)"); |
| 738 | for (;;) { |
| 739 | int nfds=1; |
| 740 | int r=polypath_beforepoll(st,fds,&nfds,0); |
| 741 | assert(nfds==1); |
| 742 | assert(!r); |
| 743 | |
| 744 | fds[1].fd=st->privsep_ipcsock_fd; |
| 745 | fds[1].events=POLLIN; |
| 746 | |
| 747 | r=poll(fds,ARRAY_SIZE(fds),-1); |
| 748 | |
| 749 | if (r<0) { |
| 750 | if (errno==EINTR) continue; |
| 751 | fatal_perror("polypath privsep poll"); |
| 752 | } |
| 753 | if (fds[1].revents) { |
| 754 | if (fds[1].revents & (POLLHUP|POLLIN)) { |
| 755 | polypath_phase_shutdown(st,PHASE_SHUTDOWN); |
| 756 | exit(0); |
| 757 | } |
| 758 | fatal("polypath privsep poll parent socket revents=%#x", |
| 759 | fds[1].revents); |
| 760 | } |
| 761 | if (fds[0].revents & POLLNVAL) |
| 762 | fatal("polypath privsep poll child socket POLLNVAL"); |
| 763 | afterpoll_monitor(st,fds,privsep_handle_ifaddr); |
| 764 | } |
| 765 | } |
| 766 | |
| 767 | static void privsep_socketpair(int *fd) |
| 768 | { |
| 769 | int r=socketpair(AF_UNIX,SOCK_STREAM,0,fd); |
| 770 | if (r) fatal_perror("socketpair(AF_UNIX,SOCK_STREAM,,)"); |
| 771 | setcloexec(fd[0]); |
| 772 | setcloexec(fd[1]); |
| 773 | } |
| 774 | |
| 775 | static void polypath_phase_startprivsep(void *sst, uint32_t newphase) |
| 776 | { |
| 777 | struct polypath *st=sst; |
| 778 | |
| 779 | if (!will_droppriv()) { |
| 780 | add_hook(PHASE_RUN, polypath_phase_startmonitor,st); |
| 781 | return; |
| 782 | } |
| 783 | |
| 784 | start_subproc(st,privsep_socketpair,child_privsep, |
| 785 | "socket generator (privsep interface handler)"); |
| 786 | |
| 787 | BUF_FREE(&st->lbuf); |
| 788 | buffer_destroy(&st->lbuf); |
| 789 | buffer_new(&st->lbuf,sizeof(struct privsep_mdata)); |
| 790 | BUF_ALLOC(&st->lbuf,"polypath mdata buf"); |
| 791 | st->privsep_incoming_fd=-1; |
| 792 | |
| 793 | register_for_poll(st,polypath_beforepoll, |
| 794 | polypath_afterpoll_privsep,"polypath"); |
| 795 | } |
| 796 | |
| 797 | static void polypath_phase_shutdown(void *sst, uint32_t newphase) |
| 798 | { |
| 799 | struct polypath *st=sst; |
| 800 | if (st->monitor_pid) { |
| 801 | assert(st->monitor_pid>0); |
| 802 | kill(st->monitor_pid,SIGTERM); |
| 803 | } |
| 804 | } |
| 805 | |
| 806 | static void polypath_phase_childpersist(void *sst, uint32_t newphase) |
| 807 | { |
| 808 | struct polypath *st=sst; |
| 809 | struct interf *interf; |
| 810 | |
| 811 | LIST_FOREACH(interf,&st->interfs,entry) |
| 812 | udp_socks_childpersist(&st->uc,&interf->socks); |
| 813 | } |
| 814 | |
| 815 | #undef BAD |
| 816 | #undef BADE |
| 817 | |
| 818 | /*----- generic closure and module setup -----*/ |
| 819 | |
| 820 | static list_t *polypath_apply(closure_t *self, struct cloc loc, |
| 821 | dict_t *context, list_t *args) |
| 822 | { |
| 823 | struct polypath *st; |
| 824 | |
| 825 | COMM_APPLY(st,&st->uc.cc,polypath_,"polypath",loc); |
| 826 | COMM_APPLY_STANDARD(st,&st->uc.cc,"polypath",args); |
| 827 | UDP_APPLY_STANDARD(st,&st->uc,"polypath"); |
| 828 | |
| 829 | struct udpcommon *uc=&st->uc; |
| 830 | struct commcommon *cc=&uc->cc; |
| 831 | |
| 832 | st->max_interfs=dict_read_number(d,"max-interfaces",False,"polypath",loc,3); |
| 833 | |
| 834 | st->ifname_pats=dict_read_string_array(d,"interfaces",False,"polypath", |
| 835 | cc->loc,0); |
| 836 | st->permit_loopback=0; /* ifname_wanted reads this */ |
| 837 | ifname_wanted(st,st->uc.cc.loc," "); /* try to check each pattern */ |
| 838 | |
| 839 | st->monitor_command=dict_read_string_array(d,"monitor-command",False, |
| 840 | "polypath",cc->loc, default_monitor_command); |
| 841 | if (!st->monitor_command[0]) |
| 842 | cfgfatal(loc,"polypath","no polypath interface monitor-command" |
| 843 | " (polypath unsupported on this platform?)\n"); |
| 844 | |
| 845 | st->permit_loopback=dict_read_bool(d,"permit-loopback",False, |
| 846 | "polypath",cc->loc,False); |
| 847 | |
| 848 | LIST_INIT(&st->interfs); |
| 849 | buffer_new(&st->lbuf,ADNS_ADDR2TEXT_BUFLEN+100); |
| 850 | BUF_ALLOC(&st->lbuf,"polypath lbuf"); |
| 851 | |
| 852 | st->monitor_fd=-1; |
| 853 | st->monitor_pid=0; |
| 854 | |
| 855 | add_hook(PHASE_GETRESOURCES, polypath_phase_startprivsep,st); |
| 856 | |
| 857 | add_hook(PHASE_SHUTDOWN, polypath_phase_shutdown, st); |
| 858 | add_hook(PHASE_CHILDPERSIST,polypath_phase_childpersist,st); |
| 859 | |
| 860 | return new_closure(&cc->cl); |
| 861 | } |
| 862 | |
| 863 | #endif /* CONFIG_IPV6 */ |
| 864 | |
| 865 | void polypath_module(dict_t *dict) |
| 866 | { |
| 867 | #ifdef CONFIG_IPV6 |
| 868 | add_closure(dict,"polypath",polypath_apply); |
| 869 | #endif /* CONFIG_IPV6 */ |
| 870 | } |