src/basic/socket-util.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2
   3 #include <arpa/inet.h>
   4 #include <errno.h>
   5 #include <limits.h>
   6 #include <net/if.h>
   7 #include <netdb.h>
   8 #include <netinet/ip.h>
   9 #include <poll.h>
  10 #include <stddef.h>
  11 #include <stdint.h>
  12 #include <stdio.h>
  13 #include <stdlib.h>
  14 #include <string.h>
  15 #include <unistd.h>
  16
  17 #include "alloc-util.h"
  18 #include "escape.h"
  19 #include "fd-util.h"
  20 #include "fileio.h"
  21 #include "format-util.h"
  22 #include "log.h"
  23 #include "macro.h"
  24 #include "missing.h"
  25 #include "parse-util.h"
  26 #include "path-util.h"
  27 #include "process-util.h"
  28 #include "socket-util.h"
  29 #include "string-table.h"
  30 #include "string-util.h"
  31 #include "strv.h"
  32 #include "user-util.h"
  33 #include "utf8.h"
  34 #include "util.h"
  35
  36 #if ENABLE_IDN
  37 #  define IDN_FLAGS NI_IDN
  38 #else
  39 #  define IDN_FLAGS 0
  40 #endif
  41
  42 static const char* const socket_address_type_table[] = {
  43         [SOCK_STREAM] = "Stream",
  44         [SOCK_DGRAM] = "Datagram",
  45         [SOCK_RAW] = "Raw",
  46         [SOCK_RDM] = "ReliableDatagram",
  47         [SOCK_SEQPACKET] = "SequentialPacket",
  48         [SOCK_DCCP] = "DatagramCongestionControl",
  49 };
  50
  51 DEFINE_STRING_TABLE_LOOKUP(socket_address_type, int);
  52
  53 int socket_address_parse(SocketAddress *a, const char *s) {
  54         _cleanup_free_ char *n = NULL;
  55         char *e;
  56         int r;
  57
  58         assert(a);
  59         assert(s);
  60
  61         *a = (SocketAddress) {
  62                 .type = SOCK_STREAM,
  63         };
  64
  65         if (*s == '[') {
  66                 uint16_t port;
  67
  68                 /* IPv6 in [x:.....:z]:p notation */
  69
  70                 e = strchr(s+1, ']');
  71                 if (!e)
  72                         return -EINVAL;
  73
  74                 n = strndup(s+1, e-s-1);
  75                 if (!n)
  76                         return -ENOMEM;
  77
  78                 errno = 0;
  79                 if (inet_pton(AF_INET6, n, &a->sockaddr.in6.sin6_addr) <= 0)
  80                         return errno > 0 ? -errno : -EINVAL;
  81
  82                 e++;
  83                 if (*e != ':')
  84                         return -EINVAL;
  85
  86                 e++;
  87                 r = parse_ip_port(e, &port);
  88                 if (r < 0)
  89                         return r;
  90
  91                 a->sockaddr.in6.sin6_family = AF_INET6;
  92                 a->sockaddr.in6.sin6_port = htobe16(port);
  93                 a->size = sizeof(struct sockaddr_in6);
  94
  95         } else if (*s == '/') {
  96                 /* AF_UNIX socket */
  97
  98                 size_t l;
  99
 100                 l = strlen(s);
 101                 if (l >= sizeof(a->sockaddr.un.sun_path)) /* Note that we refuse non-NUL-terminated sockets when
 102                                                            * parsing (the kernel itself is less strict here in what it
 103                                                            * accepts) */
 104                         return -EINVAL;
 105
 106                 a->sockaddr.un.sun_family = AF_UNIX;
 107                 memcpy(a->sockaddr.un.sun_path, s, l);
 108                 a->size = offsetof(struct sockaddr_un, sun_path) + l + 1;
 109
 110         } else if (*s == '@') {
 111                 /* Abstract AF_UNIX socket */
 112                 size_t l;
 113
 114                 l = strlen(s+1);
 115                 if (l >= sizeof(a->sockaddr.un.sun_path) - 1) /* Note that we refuse non-NUL-terminated sockets here
 116                                                                * when parsing, even though abstract namespace sockets
 117                                                                * explicitly allow embedded NUL bytes and don't consider
 118                                                                * them special. But it's simply annoying to debug such
 119                                                                * sockets. */
 120                         return -EINVAL;
 121
 122                 a->sockaddr.un.sun_family = AF_UNIX;
 123                 memcpy(a->sockaddr.un.sun_path+1, s+1, l);
 124                 a->size = offsetof(struct sockaddr_un, sun_path) + 1 + l;
 125
 126         } else if (startswith(s, "vsock:")) {
 127                 /* AF_VSOCK socket in vsock:cid:port notation */
 128                 const char *cid_start = s + STRLEN("vsock:");
 129                 unsigned port;
 130
 131                 e = strchr(cid_start, ':');
 132                 if (!e)
 133                         return -EINVAL;
 134
 135                 r = safe_atou(e+1, &port);
 136                 if (r < 0)
 137                         return r;
 138
 139                 n = strndup(cid_start, e - cid_start);
 140                 if (!n)
 141                         return -ENOMEM;
 142
 143                 if (!isempty(n)) {
 144                         r = safe_atou(n, &a->sockaddr.vm.svm_cid);
 145                         if (r < 0)
 146                                 return r;
 147                 } else
 148                         a->sockaddr.vm.svm_cid = VMADDR_CID_ANY;
 149
 150                 a->sockaddr.vm.svm_family = AF_VSOCK;
 151                 a->sockaddr.vm.svm_port = port;
 152                 a->size = sizeof(struct sockaddr_vm);
 153
 154         } else {
 155                 uint16_t port;
 156
 157                 e = strchr(s, ':');
 158                 if (e) {
 159                         r = parse_ip_port(e + 1, &port);
 160                         if (r < 0)
 161                                 return r;
 162
 163                         n = strndup(s, e-s);
 164                         if (!n)
 165                                 return -ENOMEM;
 166
 167                         /* IPv4 in w.x.y.z:p notation? */
 168                         r = inet_pton(AF_INET, n, &a->sockaddr.in.sin_addr);
 169                         if (r < 0)
 170                                 return -errno;
 171
 172                         if (r > 0) {
 173                                 /* Gotcha, it's a traditional IPv4 address */
 174                                 a->sockaddr.in.sin_family = AF_INET;
 175                                 a->sockaddr.in.sin_port = htobe16(port);
 176                                 a->size = sizeof(struct sockaddr_in);
 177                         } else {
 178                                 unsigned idx;
 179
 180                                 if (strlen(n) > IF_NAMESIZE-1)
 181                                         return -EINVAL;
 182
 183                                 /* Uh, our last resort, an interface name */
 184                                 idx = if_nametoindex(n);
 185                                 if (idx == 0)
 186                                         return -EINVAL;
 187
 188                                 a->sockaddr.in6.sin6_family = AF_INET6;
 189                                 a->sockaddr.in6.sin6_port = htobe16(port);
 190                                 a->sockaddr.in6.sin6_scope_id = idx;
 191                                 a->sockaddr.in6.sin6_addr = in6addr_any;
 192                                 a->size = sizeof(struct sockaddr_in6);
 193                         }
 194                 } else {
 195
 196                         /* Just a port */
 197                         r = parse_ip_port(s, &port);
 198                         if (r < 0)
 199                                 return r;
 200
 201                         if (socket_ipv6_is_supported()) {
 202                                 a->sockaddr.in6.sin6_family = AF_INET6;
 203                                 a->sockaddr.in6.sin6_port = htobe16(port);
 204                                 a->sockaddr.in6.sin6_addr = in6addr_any;
 205                                 a->size = sizeof(struct sockaddr_in6);
 206                         } else {
 207                                 a->sockaddr.in.sin_family = AF_INET;
 208                                 a->sockaddr.in.sin_port = htobe16(port);
 209                                 a->sockaddr.in.sin_addr.s_addr = INADDR_ANY;
 210                                 a->size = sizeof(struct sockaddr_in);
 211                         }
 212                 }
 213         }
 214
 215         return 0;
 216 }
 217
 218 int socket_address_parse_and_warn(SocketAddress *a, const char *s) {
 219         SocketAddress b;
 220         int r;
 221
 222         /* Similar to socket_address_parse() but warns for IPv6 sockets when we don't support them. */
 223
 224         r = socket_address_parse(&b, s);
 225         if (r < 0)
 226                 return r;
 227
 228         if (!socket_ipv6_is_supported() && b.sockaddr.sa.sa_family == AF_INET6) {
 229                 log_warning("Binding to IPv6 address not available since kernel does not support IPv6.");
 230                 return -EAFNOSUPPORT;
 231         }
 232
 233         *a = b;
 234         return 0;
 235 }
 236
 237 int socket_address_parse_netlink(SocketAddress *a, const char *s) {
 238         int family;
 239         unsigned group = 0;
 240         _cleanup_free_ char *sfamily = NULL;
 241         assert(a);
 242         assert(s);
 243
 244         zero(*a);
 245         a->type = SOCK_RAW;
 246
 247         errno = 0;
 248         if (sscanf(s, "%ms %u", &sfamily, &group) < 1)
 249                 return errno > 0 ? -errno : -EINVAL;
 250
 251         family = netlink_family_from_string(sfamily);
 252         if (family < 0)
 253                 return -EINVAL;
 254
 255         a->sockaddr.nl.nl_family = AF_NETLINK;
 256         a->sockaddr.nl.nl_groups = group;
 257
 258         a->type = SOCK_RAW;
 259         a->size = sizeof(struct sockaddr_nl);
 260         a->protocol = family;
 261
 262         return 0;
 263 }
 264
 265 int socket_address_verify(const SocketAddress *a, bool strict) {
 266         assert(a);
 267
 268         /* With 'strict' we enforce additional sanity constraints which are not set by the standard,
 269          * but should only apply to sockets we create ourselves. */
 270
 271         switch (socket_address_family(a)) {
 272
 273         case AF_INET:
 274                 if (a->size != sizeof(struct sockaddr_in))
 275                         return -EINVAL;
 276
 277                 if (a->sockaddr.in.sin_port == 0)
 278                         return -EINVAL;
 279
 280                 if (!IN_SET(a->type, SOCK_STREAM, SOCK_DGRAM))
 281                         return -EINVAL;
 282
 283                 return 0;
 284
 285         case AF_INET6:
 286                 if (a->size != sizeof(struct sockaddr_in6))
 287                         return -EINVAL;
 288
 289                 if (a->sockaddr.in6.sin6_port == 0)
 290                         return -EINVAL;
 291
 292                 if (!IN_SET(a->type, SOCK_STREAM, SOCK_DGRAM))
 293                         return -EINVAL;
 294
 295                 return 0;
 296
 297         case AF_UNIX:
 298                 if (a->size < offsetof(struct sockaddr_un, sun_path))
 299                         return -EINVAL;
 300                 if (a->size > sizeof(struct sockaddr_un) + !strict)
 301                         /* If !strict, allow one extra byte, since getsockname() on Linux will append
 302                          * a NUL byte if we have path sockets that are above sun_path's full size. */
 303                         return -EINVAL;
 304
 305                 if (a->size > offsetof(struct sockaddr_un, sun_path) &&
 306                     a->sockaddr.un.sun_path[0] != 0 &&
 307                     strict) {
 308                         /* Only validate file system sockets here, and only in strict mode */
 309                         const char *e;
 310
 311                         e = memchr(a->sockaddr.un.sun_path, 0, sizeof(a->sockaddr.un.sun_path));
 312                         if (e) {
 313                                 /* If there's an embedded NUL byte, make sure the size of the socket address matches it */
 314                                 if (a->size != offsetof(struct sockaddr_un, sun_path) + (e - a->sockaddr.un.sun_path) + 1)
 315                                         return -EINVAL;
 316                         } else {
 317                                 /* If there's no embedded NUL byte, then then the size needs to match the whole
 318                                  * structure or the structure with one extra NUL byte suffixed. (Yeah, Linux is awful,
 319                                  * and considers both equivalent: getsockname() even extends sockaddr_un beyond its
 320                                  * size if the path is non NUL terminated.)*/
 321                                 if (!IN_SET(a->size, sizeof(a->sockaddr.un.sun_path), sizeof(a->sockaddr.un.sun_path)+1))
 322                                         return -EINVAL;
 323                         }
 324                 }
 325
 326                 if (!IN_SET(a->type, SOCK_STREAM, SOCK_DGRAM, SOCK_SEQPACKET))
 327                         return -EINVAL;
 328
 329                 return 0;
 330
 331         case AF_NETLINK:
 332
 333                 if (a->size != sizeof(struct sockaddr_nl))
 334                         return -EINVAL;
 335
 336                 if (!IN_SET(a->type, SOCK_RAW, SOCK_DGRAM))
 337                         return -EINVAL;
 338
 339                 return 0;
 340
 341         case AF_VSOCK:
 342                 if (a->size != sizeof(struct sockaddr_vm))
 343                         return -EINVAL;
 344
 345                 if (!IN_SET(a->type, SOCK_STREAM, SOCK_DGRAM))
 346                         return -EINVAL;
 347
 348                 return 0;
 349
 350         default:
 351                 return -EAFNOSUPPORT;
 352         }
 353 }
 354
 355 int socket_address_print(const SocketAddress *a, char **ret) {
 356         int r;
 357
 358         assert(a);
 359         assert(ret);
 360
 361         r = socket_address_verify(a, false); /* We do non-strict validation, because we want to be
 362                                               * able to pretty-print any socket the kernel considers
 363                                               * valid. We still need to do validation to know if we
 364                                               * can meaningfully print the address. */
 365         if (r < 0)
 366                 return r;
 367
 368         if (socket_address_family(a) == AF_NETLINK) {
 369                 _cleanup_free_ char *sfamily = NULL;
 370
 371                 r = netlink_family_to_string_alloc(a->protocol, &sfamily);
 372                 if (r < 0)
 373                         return r;
 374
 375                 r = asprintf(ret, "%s %u", sfamily, a->sockaddr.nl.nl_groups);
 376                 if (r < 0)
 377                         return -ENOMEM;
 378
 379                 return 0;
 380         }
 381
 382         return sockaddr_pretty(&a->sockaddr.sa, a->size, false, true, ret);
 383 }
 384
 385 bool socket_address_can_accept(const SocketAddress *a) {
 386         assert(a);
 387
 388         return
 389                 IN_SET(a->type, SOCK_STREAM, SOCK_SEQPACKET);
 390 }
 391
 392 bool socket_address_equal(const SocketAddress *a, const SocketAddress *b) {
 393         assert(a);
 394         assert(b);
 395
 396         /* Invalid addresses are unequal to all */
 397         if (socket_address_verify(a, false) < 0 ||
 398             socket_address_verify(b, false) < 0)
 399                 return false;
 400
 401         if (a->type != b->type)
 402                 return false;
 403
 404         if (socket_address_family(a) != socket_address_family(b))
 405                 return false;
 406
 407         switch (socket_address_family(a)) {
 408
 409         case AF_INET:
 410                 if (a->sockaddr.in.sin_addr.s_addr != b->sockaddr.in.sin_addr.s_addr)
 411                         return false;
 412
 413                 if (a->sockaddr.in.sin_port != b->sockaddr.in.sin_port)
 414                         return false;
 415
 416                 break;
 417
 418         case AF_INET6:
 419                 if (memcmp(&a->sockaddr.in6.sin6_addr, &b->sockaddr.in6.sin6_addr, sizeof(a->sockaddr.in6.sin6_addr)) != 0)
 420                         return false;
 421
 422                 if (a->sockaddr.in6.sin6_port != b->sockaddr.in6.sin6_port)
 423                         return false;
 424
 425                 break;
 426
 427         case AF_UNIX:
 428                 if (a->size <= offsetof(struct sockaddr_un, sun_path) ||
 429                     b->size <= offsetof(struct sockaddr_un, sun_path))
 430                         return false;
 431
 432                 if ((a->sockaddr.un.sun_path[0] == 0) != (b->sockaddr.un.sun_path[0] == 0))
 433                         return false;
 434
 435                 if (a->sockaddr.un.sun_path[0]) {
 436                         if (!path_equal_or_files_same(a->sockaddr.un.sun_path, b->sockaddr.un.sun_path, 0))
 437                                 return false;
 438                 } else {
 439                         if (a->size != b->size)
 440                                 return false;
 441
 442                         if (memcmp(a->sockaddr.un.sun_path, b->sockaddr.un.sun_path, a->size) != 0)
 443                                 return false;
 444                 }
 445
 446                 break;
 447
 448         case AF_NETLINK:
 449                 if (a->protocol != b->protocol)
 450                         return false;
 451
 452                 if (a->sockaddr.nl.nl_groups != b->sockaddr.nl.nl_groups)
 453                         return false;
 454
 455                 break;
 456
 457         case AF_VSOCK:
 458                 if (a->sockaddr.vm.svm_cid != b->sockaddr.vm.svm_cid)
 459                         return false;
 460
 461                 if (a->sockaddr.vm.svm_port != b->sockaddr.vm.svm_port)
 462                         return false;
 463
 464                 break;
 465
 466         default:
 467                 /* Cannot compare, so we assume the addresses are different */
 468                 return false;
 469         }
 470
 471         return true;
 472 }
 473
 474 bool socket_address_is(const SocketAddress *a, const char *s, int type) {
 475         struct SocketAddress b;
 476
 477         assert(a);
 478         assert(s);
 479
 480         if (socket_address_parse(&b, s) < 0)
 481                 return false;
 482
 483         b.type = type;
 484
 485         return socket_address_equal(a, &b);
 486 }
 487
 488 bool socket_address_is_netlink(const SocketAddress *a, const char *s) {
 489         struct SocketAddress b;
 490
 491         assert(a);
 492         assert(s);
 493
 494         if (socket_address_parse_netlink(&b, s) < 0)
 495                 return false;
 496
 497         return socket_address_equal(a, &b);
 498 }
 499
 500 const char* socket_address_get_path(const SocketAddress *a) {
 501         assert(a);
 502
 503         if (socket_address_family(a) != AF_UNIX)
 504                 return NULL;
 505
 506         if (a->sockaddr.un.sun_path[0] == 0)
 507                 return NULL;
 508
 509         /* Note that this is only safe because we know that there's an extra NUL byte after the sockaddr_un
 510          * structure. On Linux AF_UNIX file system socket addresses don't have to be NUL terminated if they take up the
 511          * full sun_path space. */
 512         assert_cc(sizeof(union sockaddr_union) >= sizeof(struct sockaddr_un)+1);
 513         return a->sockaddr.un.sun_path;
 514 }
 515
 516 bool socket_ipv6_is_supported(void) {
 517         if (access("/proc/net/if_inet6", F_OK) != 0)
 518                 return false;
 519
 520         return true;
 521 }
 522
 523 bool socket_address_matches_fd(const SocketAddress *a, int fd) {
 524         SocketAddress b;
 525         socklen_t solen;
 526
 527         assert(a);
 528         assert(fd >= 0);
 529
 530         b.size = sizeof(b.sockaddr);
 531         if (getsockname(fd, &b.sockaddr.sa, &b.size) < 0)
 532                 return false;
 533
 534         if (b.sockaddr.sa.sa_family != a->sockaddr.sa.sa_family)
 535                 return false;
 536
 537         solen = sizeof(b.type);
 538         if (getsockopt(fd, SOL_SOCKET, SO_TYPE, &b.type, &solen) < 0)
 539                 return false;
 540
 541         if (b.type != a->type)
 542                 return false;
 543
 544         if (a->protocol != 0)  {
 545                 solen = sizeof(b.protocol);
 546                 if (getsockopt(fd, SOL_SOCKET, SO_PROTOCOL, &b.protocol, &solen) < 0)
 547                         return false;
 548
 549                 if (b.protocol != a->protocol)
 550                         return false;
 551         }
 552
 553         return socket_address_equal(a, &b);
 554 }
 555
 556 int sockaddr_port(const struct sockaddr *_sa, unsigned *ret_port) {
 557         union sockaddr_union *sa = (union sockaddr_union*) _sa;
 558
 559         /* Note, this returns the port as 'unsigned' rather than 'uint16_t', as AF_VSOCK knows larger ports */
 560
 561         assert(sa);
 562
 563         switch (sa->sa.sa_family) {
 564
 565         case AF_INET:
 566                 *ret_port = be16toh(sa->in.sin_port);
 567                 return 0;
 568
 569         case AF_INET6:
 570                 *ret_port = be16toh(sa->in6.sin6_port);
 571                 return 0;
 572
 573         case AF_VSOCK:
 574                 *ret_port = sa->vm.svm_port;
 575                 return 0;
 576
 577         default:
 578                 return -EAFNOSUPPORT;
 579         }
 580 }
 581
 582 int sockaddr_pretty(
 583                 const struct sockaddr *_sa,
 584                 socklen_t salen,
 585                 bool translate_ipv6,
 586                 bool include_port,
 587                 char **ret) {
 588
 589         union sockaddr_union *sa = (union sockaddr_union*) _sa;
 590         char *p;
 591         int r;
 592
 593         assert(sa);
 594         assert(salen >= sizeof(sa->sa.sa_family));
 595
 596         switch (sa->sa.sa_family) {
 597
 598         case AF_INET: {
 599                 uint32_t a;
 600
 601                 a = be32toh(sa->in.sin_addr.s_addr);
 602
 603                 if (include_port)
 604                         r = asprintf(&p,
 605                                      "%u.%u.%u.%u:%u",
 606                                      a >> 24, (a >> 16) & 0xFF, (a >> 8) & 0xFF, a & 0xFF,
 607                                      be16toh(sa->in.sin_port));
 608                 else
 609                         r = asprintf(&p,
 610                                      "%u.%u.%u.%u",
 611                                      a >> 24, (a >> 16) & 0xFF, (a >> 8) & 0xFF, a & 0xFF);
 612                 if (r < 0)
 613                         return -ENOMEM;
 614                 break;
 615         }
 616
 617         case AF_INET6: {
 618                 static const unsigned char ipv4_prefix[] = {
 619                         0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0xFF, 0xFF
 620                 };
 621
 622                 if (translate_ipv6 &&
 623                     memcmp(&sa->in6.sin6_addr, ipv4_prefix, sizeof(ipv4_prefix)) == 0) {
 624                         const uint8_t *a = sa->in6.sin6_addr.s6_addr+12;
 625                         if (include_port)
 626                                 r = asprintf(&p,
 627                                              "%u.%u.%u.%u:%u",
 628                                              a[0], a[1], a[2], a[3],
 629                                              be16toh(sa->in6.sin6_port));
 630                         else
 631                                 r = asprintf(&p,
 632                                              "%u.%u.%u.%u",
 633                                              a[0], a[1], a[2], a[3]);
 634                         if (r < 0)
 635                                 return -ENOMEM;
 636                 } else {
 637                         char a[INET6_ADDRSTRLEN];
 638
 639                         inet_ntop(AF_INET6, &sa->in6.sin6_addr, a, sizeof(a));
 640
 641                         if (include_port) {
 642                                 r = asprintf(&p,
 643                                              "[%s]:%u",
 644                                              a,
 645                                              be16toh(sa->in6.sin6_port));
 646                                 if (r < 0)
 647                                         return -ENOMEM;
 648                         } else {
 649                                 p = strdup(a);
 650                                 if (!p)
 651                                         return -ENOMEM;
 652                         }
 653                 }
 654
 655                 break;
 656         }
 657
 658         case AF_UNIX:
 659                 if (salen <= offsetof(struct sockaddr_un, sun_path) ||
 660                     (sa->un.sun_path[0] == 0 && salen == offsetof(struct sockaddr_un, sun_path) + 1))
 661                         /* The name must have at least one character (and the leading NUL does not count) */
 662                         p = strdup("<unnamed>");
 663                 else {
 664                         /* Note that we calculate the path pointer here through the .un_buffer[] field, in order to
 665                          * outtrick bounds checking tools such as ubsan, which are too smart for their own good: on
 666                          * Linux the kernel may return sun_path[] data one byte longer than the declared size of the
 667                          * field. */
 668                         char *path = (char*) sa->un_buffer + offsetof(struct sockaddr_un, sun_path);
 669                         size_t path_len = salen - offsetof(struct sockaddr_un, sun_path);
 670
 671                         if (path[0] == 0) {
 672                                 /* Abstract socket. When parsing address information from, we
 673                                  * explicitly reject overly long paths and paths with embedded NULs.
 674                                  * But we might get such a socket from the outside. Let's return
 675                                  * something meaningful and printable in this case. */
 676
 677                                 _cleanup_free_ char *e = NULL;
 678
 679                                 e = cescape_length(path + 1, path_len - 1);
 680                                 if (!e)
 681                                         return -ENOMEM;
 682
 683                                 p = strjoin("@", e);
 684                         } else {
 685                                 if (path[path_len - 1] == '\0')
 686                                         /* We expect a terminating NUL and don't print it */
 687                                         path_len --;
 688
 689                                 p = cescape_length(path, path_len);
 690                         }
 691                 }
 692                 if (!p)
 693                         return -ENOMEM;
 694
 695                 break;
 696
 697         case AF_VSOCK:
 698                 if (include_port) {
 699                         if (sa->vm.svm_cid == VMADDR_CID_ANY)
 700                                 r = asprintf(&p, "vsock::%u", sa->vm.svm_port);
 701                         else
 702                                 r = asprintf(&p, "vsock:%u:%u", sa->vm.svm_cid, sa->vm.svm_port);
 703                 } else
 704                         r = asprintf(&p, "vsock:%u", sa->vm.svm_cid);
 705                 if (r < 0)
 706                         return -ENOMEM;
 707                 break;
 708
 709         default:
 710                 return -EOPNOTSUPP;
 711         }
 712
 713         *ret = p;
 714         return 0;
 715 }
 716
 717 int getpeername_pretty(int fd, bool include_port, char **ret) {
 718         union sockaddr_union sa;
 719         socklen_t salen = sizeof(sa);
 720         int r;
 721
 722         assert(fd >= 0);
 723         assert(ret);
 724
 725         if (getpeername(fd, &sa.sa, &salen) < 0)
 726                 return -errno;
 727
 728         if (sa.sa.sa_family == AF_UNIX) {
 729                 struct ucred ucred = {};
 730
 731                 /* UNIX connection sockets are anonymous, so let's use
 732                  * PID/UID as pretty credentials instead */
 733
 734                 r = getpeercred(fd, &ucred);
 735                 if (r < 0)
 736                         return r;
 737
 738                 if (asprintf(ret, "PID "PID_FMT"/UID "UID_FMT, ucred.pid, ucred.uid) < 0)
 739                         return -ENOMEM;
 740
 741                 return 0;
 742         }
 743
 744         /* For remote sockets we translate IPv6 addresses back to IPv4
 745          * if applicable, since that's nicer. */
 746
 747         return sockaddr_pretty(&sa.sa, salen, true, include_port, ret);
 748 }
 749
 750 int getsockname_pretty(int fd, char **ret) {
 751         union sockaddr_union sa;
 752         socklen_t salen = sizeof(sa);
 753
 754         assert(fd >= 0);
 755         assert(ret);
 756
 757         if (getsockname(fd, &sa.sa, &salen) < 0)
 758                 return -errno;
 759
 760         /* For local sockets we do not translate IPv6 addresses back
 761          * to IPv6 if applicable, since this is usually used for
 762          * listening sockets where the difference between IPv4 and
 763          * IPv6 matters. */
 764
 765         return sockaddr_pretty(&sa.sa, salen, false, true, ret);
 766 }
 767
 768 int socknameinfo_pretty(union sockaddr_union *sa, socklen_t salen, char **_ret) {
 769         int r;
 770         char host[NI_MAXHOST], *ret;
 771
 772         assert(_ret);
 773
 774         r = getnameinfo(&sa->sa, salen, host, sizeof(host), NULL, 0, IDN_FLAGS);
 775         if (r != 0) {
 776                 int saved_errno = errno;
 777
 778                 r = sockaddr_pretty(&sa->sa, salen, true, true, &ret);
 779                 if (r < 0)
 780                         return r;
 781
 782                 log_debug_errno(saved_errno, "getnameinfo(%s) failed: %m", ret);
 783         } else {
 784                 ret = strdup(host);
 785                 if (!ret)
 786                         return -ENOMEM;
 787         }
 788
 789         *_ret = ret;
 790         return 0;
 791 }
 792
 793 static const char* const netlink_family_table[] = {
 794         [NETLINK_ROUTE] = "route",
 795         [NETLINK_FIREWALL] = "firewall",
 796         [NETLINK_INET_DIAG] = "inet-diag",
 797         [NETLINK_NFLOG] = "nflog",
 798         [NETLINK_XFRM] = "xfrm",
 799         [NETLINK_SELINUX] = "selinux",
 800         [NETLINK_ISCSI] = "iscsi",
 801         [NETLINK_AUDIT] = "audit",
 802         [NETLINK_FIB_LOOKUP] = "fib-lookup",
 803         [NETLINK_CONNECTOR] = "connector",
 804         [NETLINK_NETFILTER] = "netfilter",
 805         [NETLINK_IP6_FW] = "ip6-fw",
 806         [NETLINK_DNRTMSG] = "dnrtmsg",
 807         [NETLINK_KOBJECT_UEVENT] = "kobject-uevent",
 808         [NETLINK_GENERIC] = "generic",
 809         [NETLINK_SCSITRANSPORT] = "scsitransport",
 810         [NETLINK_ECRYPTFS] = "ecryptfs",
 811         [NETLINK_RDMA] = "rdma",
 812 };
 813
 814 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(netlink_family, int, INT_MAX);
 815
 816 static const char* const socket_address_bind_ipv6_only_table[_SOCKET_ADDRESS_BIND_IPV6_ONLY_MAX] = {
 817         [SOCKET_ADDRESS_DEFAULT] = "default",
 818         [SOCKET_ADDRESS_BOTH] = "both",
 819         [SOCKET_ADDRESS_IPV6_ONLY] = "ipv6-only"
 820 };
 821
 822 DEFINE_STRING_TABLE_LOOKUP(socket_address_bind_ipv6_only, SocketAddressBindIPv6Only);
 823
 824 SocketAddressBindIPv6Only socket_address_bind_ipv6_only_or_bool_from_string(const char *n) {
 825         int r;
 826
 827         r = parse_boolean(n);
 828         if (r > 0)
 829                 return SOCKET_ADDRESS_IPV6_ONLY;
 830         if (r == 0)
 831                 return SOCKET_ADDRESS_BOTH;
 832
 833         return socket_address_bind_ipv6_only_from_string(n);
 834 }
 835
 836 bool sockaddr_equal(const union sockaddr_union *a, const union sockaddr_union *b) {
 837         assert(a);
 838         assert(b);
 839
 840         if (a->sa.sa_family != b->sa.sa_family)
 841                 return false;
 842
 843         if (a->sa.sa_family == AF_INET)
 844                 return a->in.sin_addr.s_addr == b->in.sin_addr.s_addr;
 845
 846         if (a->sa.sa_family == AF_INET6)
 847                 return memcmp(&a->in6.sin6_addr, &b->in6.sin6_addr, sizeof(a->in6.sin6_addr)) == 0;
 848
 849         if (a->sa.sa_family == AF_VSOCK)
 850                 return a->vm.svm_cid == b->vm.svm_cid;
 851
 852         return false;
 853 }
 854
 855 int fd_inc_sndbuf(int fd, size_t n) {
 856         int r, value;
 857         socklen_t l = sizeof(value);
 858
 859         r = getsockopt(fd, SOL_SOCKET, SO_SNDBUF, &value, &l);
 860         if (r >= 0 && l == sizeof(value) && (size_t) value >= n*2)
 861                 return 0;
 862
 863         /* If we have the privileges we will ignore the kernel limit. */
 864
 865         if (setsockopt_int(fd, SOL_SOCKET, SO_SNDBUF, n) < 0) {
 866                 r = setsockopt_int(fd, SOL_SOCKET, SO_SNDBUFFORCE, n);
 867                 if (r < 0)
 868                         return r;
 869         }
 870
 871         return 1;
 872 }
 873
 874 int fd_inc_rcvbuf(int fd, size_t n) {
 875         int r, value;
 876         socklen_t l = sizeof(value);
 877
 878         r = getsockopt(fd, SOL_SOCKET, SO_RCVBUF, &value, &l);
 879         if (r >= 0 && l == sizeof(value) && (size_t) value >= n*2)
 880                 return 0;
 881
 882         /* If we have the privileges we will ignore the kernel limit. */
 883
 884         if (setsockopt_int(fd, SOL_SOCKET, SO_RCVBUF, n) < 0) {
 885                 r = setsockopt_int(fd, SOL_SOCKET, SO_RCVBUFFORCE, n);
 886                 if (r < 0)
 887                         return r;
 888         }
 889
 890         return 1;
 891 }
 892
 893 static const char* const ip_tos_table[] = {
 894         [IPTOS_LOWDELAY] = "low-delay",
 895         [IPTOS_THROUGHPUT] = "throughput",
 896         [IPTOS_RELIABILITY] = "reliability",
 897         [IPTOS_LOWCOST] = "low-cost",
 898 };
 899
 900 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ip_tos, int, 0xff);
 901
 902 bool ifname_valid(const char *p) {
 903         bool numeric = true;
 904
 905         /* Checks whether a network interface name is valid. This is inspired by dev_valid_name() in the kernel sources
 906          * but slightly stricter, as we only allow non-control, non-space ASCII characters in the interface name. We
 907          * also don't permit names that only container numbers, to avoid confusion with numeric interface indexes. */
 908
 909         if (isempty(p))
 910                 return false;
 911
 912         if (strlen(p) >= IFNAMSIZ)
 913                 return false;
 914
 915         if (dot_or_dot_dot(p))
 916                 return false;
 917
 918         while (*p) {
 919                 if ((unsigned char) *p >= 127U)
 920                         return false;
 921
 922                 if ((unsigned char) *p <= 32U)
 923                         return false;
 924
 925                 if (IN_SET(*p, ':', '/'))
 926                         return false;
 927
 928                 numeric = numeric && (*p >= '0' && *p <= '9');
 929                 p++;
 930         }
 931
 932         if (numeric)
 933                 return false;
 934
 935         return true;
 936 }
 937
 938 bool address_label_valid(const char *p) {
 939
 940         if (isempty(p))
 941                 return false;
 942
 943         if (strlen(p) >= IFNAMSIZ)
 944                 return false;
 945
 946         while (*p) {
 947                 if ((uint8_t) *p >= 127U)
 948                         return false;
 949
 950                 if ((uint8_t) *p <= 31U)
 951                         return false;
 952                 p++;
 953         }
 954
 955         return true;
 956 }
 957
 958 int getpeercred(int fd, struct ucred *ucred) {
 959         socklen_t n = sizeof(struct ucred);
 960         struct ucred u;
 961         int r;
 962
 963         assert(fd >= 0);
 964         assert(ucred);
 965
 966         r = getsockopt(fd, SOL_SOCKET, SO_PEERCRED, &u, &n);
 967         if (r < 0)
 968                 return -errno;
 969
 970         if (n != sizeof(struct ucred))
 971                 return -EIO;
 972
 973         /* Check if the data is actually useful and not suppressed due to namespacing issues */
 974         if (!pid_is_valid(u.pid))
 975                 return -ENODATA;
 976
 977         /* Note that we don't check UID/GID here, as namespace translation works differently there: instead of
 978          * receiving in "invalid" user/group we get the overflow UID/GID. */
 979
 980         *ucred = u;
 981         return 0;
 982 }
 983
 984 int getpeersec(int fd, char **ret) {
 985         _cleanup_free_ char *s = NULL;
 986         socklen_t n = 64;
 987
 988         assert(fd >= 0);
 989         assert(ret);
 990
 991         for (;;) {
 992                 s = new0(char, n+1);
 993                 if (!s)
 994                         return -ENOMEM;
 995
 996                 if (getsockopt(fd, SOL_SOCKET, SO_PEERSEC, s, &n) >= 0)
 997                         break;
 998
 999                 if (errno != ERANGE)
1000                         return -errno;
1001
1002                 s = mfree(s);
1003         }
1004
1005         if (isempty(s))
1006                 return -EOPNOTSUPP;
1007
1008         *ret = TAKE_PTR(s);
1009
1010         return 0;
1011 }
1012
1013 int getpeergroups(int fd, gid_t **ret) {
1014         socklen_t n = sizeof(gid_t) * 64;
1015         _cleanup_free_ gid_t *d = NULL;
1016
1017         assert(fd >= 0);
1018         assert(ret);
1019
1020         for (;;) {
1021                 d = malloc(n);
1022                 if (!d)
1023                         return -ENOMEM;
1024
1025                 if (getsockopt(fd, SOL_SOCKET, SO_PEERGROUPS, d, &n) >= 0)
1026                         break;
1027
1028                 if (errno != ERANGE)
1029                         return -errno;
1030
1031                 d = mfree(d);
1032         }
1033
1034         assert_se(n % sizeof(gid_t) == 0);
1035         n /= sizeof(gid_t);
1036
1037         if ((socklen_t) (int) n != n)
1038                 return -E2BIG;
1039
1040         *ret = TAKE_PTR(d);
1041
1042         return (int) n;
1043 }
1044
1045 ssize_t send_one_fd_iov_sa(
1046                 int transport_fd,
1047                 int fd,
1048                 struct iovec *iov, size_t iovlen,
1049                 const struct sockaddr *sa, socklen_t len,
1050                 int flags) {
1051
1052         union {
1053                 struct cmsghdr cmsghdr;
1054                 uint8_t buf[CMSG_SPACE(sizeof(int))];
1055         } control = {};
1056         struct msghdr mh = {
1057                 .msg_name = (struct sockaddr*) sa,
1058                 .msg_namelen = len,
1059                 .msg_iov = iov,
1060                 .msg_iovlen = iovlen,
1061         };
1062         ssize_t k;
1063
1064         assert(transport_fd >= 0);
1065
1066         /*
1067          * We need either an FD or data to send.
1068          * If there's nothing, return an error.
1069          */
1070         if (fd < 0 && !iov)
1071                 return -EINVAL;
1072
1073         if (fd >= 0) {
1074                 struct cmsghdr *cmsg;
1075
1076                 mh.msg_control = &control;
1077                 mh.msg_controllen = sizeof(control);
1078
1079                 cmsg = CMSG_FIRSTHDR(&mh);
1080                 cmsg->cmsg_level = SOL_SOCKET;
1081                 cmsg->cmsg_type = SCM_RIGHTS;
1082                 cmsg->cmsg_len = CMSG_LEN(sizeof(int));
1083                 memcpy(CMSG_DATA(cmsg), &fd, sizeof(int));
1084
1085                 mh.msg_controllen = CMSG_SPACE(sizeof(int));
1086         }
1087         k = sendmsg(transport_fd, &mh, MSG_NOSIGNAL | flags);
1088         if (k < 0)
1089                 return (ssize_t) -errno;
1090
1091         return k;
1092 }
1093
1094 int send_one_fd_sa(
1095                 int transport_fd,
1096                 int fd,
1097                 const struct sockaddr *sa, socklen_t len,
1098                 int flags) {
1099
1100         assert(fd >= 0);
1101
1102         return (int) send_one_fd_iov_sa(transport_fd, fd, NULL, 0, sa, len, flags);
1103 }
1104
1105 ssize_t receive_one_fd_iov(
1106                 int transport_fd,
1107                 struct iovec *iov, size_t iovlen,
1108                 int flags,
1109                 int *ret_fd) {
1110
1111         union {
1112                 struct cmsghdr cmsghdr;
1113                 uint8_t buf[CMSG_SPACE(sizeof(int))];
1114         } control = {};
1115         struct msghdr mh = {
1116                 .msg_control = &control,
1117                 .msg_controllen = sizeof(control),
1118                 .msg_iov = iov,
1119                 .msg_iovlen = iovlen,
1120         };
1121         struct cmsghdr *cmsg, *found = NULL;
1122         ssize_t k;
1123
1124         assert(transport_fd >= 0);
1125         assert(ret_fd);
1126
1127         /*
1128          * Receive a single FD via @transport_fd. We don't care for
1129          * the transport-type. We retrieve a single FD at most, so for
1130          * packet-based transports, the caller must ensure to send
1131          * only a single FD per packet.  This is best used in
1132          * combination with send_one_fd().
1133          */
1134
1135         k = recvmsg(transport_fd, &mh, MSG_CMSG_CLOEXEC | flags);
1136         if (k < 0)
1137                 return (ssize_t) -errno;
1138
1139         CMSG_FOREACH(cmsg, &mh) {
1140                 if (cmsg->cmsg_level == SOL_SOCKET &&
1141                     cmsg->cmsg_type == SCM_RIGHTS &&
1142                     cmsg->cmsg_len == CMSG_LEN(sizeof(int))) {
1143                         assert(!found);
1144                         found = cmsg;
1145                         break;
1146                 }
1147         }
1148
1149         if (!found)
1150                 cmsg_close_all(&mh);
1151
1152         /* If didn't receive an FD or any data, return an error. */
1153         if (k == 0 && !found)
1154                 return -EIO;
1155
1156         if (found)
1157                 *ret_fd = *(int*) CMSG_DATA(found);
1158         else
1159                 *ret_fd = -1;
1160
1161         return k;
1162 }
1163
1164 int receive_one_fd(int transport_fd, int flags) {
1165         int fd;
1166         ssize_t k;
1167
1168         k = receive_one_fd_iov(transport_fd, NULL, 0, flags, &fd);
1169         if (k == 0)
1170                 return fd;
1171
1172         /* k must be negative, since receive_one_fd_iov() only returns
1173          * a positive value if data was received through the iov. */
1174         assert(k < 0);
1175         return (int) k;
1176 }
1177
1178 ssize_t next_datagram_size_fd(int fd) {
1179         ssize_t l;
1180         int k;
1181
1182         /* This is a bit like FIONREAD/SIOCINQ, however a bit more powerful. The difference being: recv(MSG_PEEK) will
1183          * actually cause the next datagram in the queue to be validated regarding checksums, which FIONREAD doesn't
1184          * do. This difference is actually of major importance as we need to be sure that the size returned here
1185          * actually matches what we will read with recvmsg() next, as otherwise we might end up allocating a buffer of
1186          * the wrong size. */
1187
1188         l = recv(fd, NULL, 0, MSG_PEEK|MSG_TRUNC);
1189         if (l < 0) {
1190                 if (IN_SET(errno, EOPNOTSUPP, EFAULT))
1191                         goto fallback;
1192
1193                 return -errno;
1194         }
1195         if (l == 0)
1196                 goto fallback;
1197
1198         return l;
1199
1200 fallback:
1201         k = 0;
1202
1203         /* Some sockets (AF_PACKET) do not support null-sized recv() with MSG_TRUNC set, let's fall back to FIONREAD
1204          * for them. Checksums don't matter for raw sockets anyway, hence this should be fine. */
1205
1206         if (ioctl(fd, FIONREAD, &k) < 0)
1207                 return -errno;
1208
1209         return (ssize_t) k;
1210 }
1211
1212 int flush_accept(int fd) {
1213
1214         struct pollfd pollfd = {
1215                 .fd = fd,
1216                 .events = POLLIN,
1217         };
1218         int r;
1219
1220         /* Similar to flush_fd() but flushes all incoming connection by accepting them and immediately closing them. */
1221
1222         for (;;) {
1223                 int cfd;
1224
1225                 r = poll(&pollfd, 1, 0);
1226                 if (r < 0) {
1227                         if (errno == EINTR)
1228                                 continue;
1229
1230                         return -errno;
1231
1232                 } else if (r == 0)
1233                         return 0;
1234
1235                 cfd = accept4(fd, NULL, NULL, SOCK_NONBLOCK|SOCK_CLOEXEC);
1236                 if (cfd < 0) {
1237                         if (errno == EINTR)
1238                                 continue;
1239
1240                         if (errno == EAGAIN)
1241                                 return 0;
1242
1243                         return -errno;
1244                 }
1245
1246                 close(cfd);
1247         }
1248 }
1249
1250 struct cmsghdr* cmsg_find(struct msghdr *mh, int level, int type, socklen_t length) {
1251         struct cmsghdr *cmsg;
1252
1253         assert(mh);
1254
1255         CMSG_FOREACH(cmsg, mh)
1256                 if (cmsg->cmsg_level == level &&
1257                     cmsg->cmsg_type == type &&
1258                     (length == (socklen_t) -1 || length == cmsg->cmsg_len))
1259                         return cmsg;
1260
1261         return NULL;
1262 }
1263
1264 int socket_ioctl_fd(void) {
1265         int fd;
1266
1267         /* Create a socket to invoke the various network interface ioctl()s on. Traditionally only AF_INET was good for
1268          * that. Since kernel 4.6 AF_NETLINK works for this too. We first try to use AF_INET hence, but if that's not
1269          * available (for example, because it is made unavailable via SECCOMP or such), we'll fall back to the more
1270          * generic AF_NETLINK. */
1271
1272         fd = socket(AF_INET, SOCK_DGRAM|SOCK_CLOEXEC, 0);
1273         if (fd < 0)
1274                 fd = socket(AF_NETLINK, SOCK_RAW|SOCK_CLOEXEC, NETLINK_GENERIC);
1275         if (fd < 0)
1276                 return -errno;
1277
1278         return fd;
1279 }
1280
1281 int sockaddr_un_unlink(const struct sockaddr_un *sa) {
1282         const char *p, * nul;
1283
1284         assert(sa);
1285
1286         if (sa->sun_family != AF_UNIX)
1287                 return -EPROTOTYPE;
1288
1289         if (sa->sun_path[0] == 0) /* Nothing to do for abstract sockets */
1290                 return 0;
1291
1292         /* The path in .sun_path is not necessarily NUL terminated. Let's fix that. */
1293         nul = memchr(sa->sun_path, 0, sizeof(sa->sun_path));
1294         if (nul)
1295                 p = sa->sun_path;
1296         else
1297                 p = memdupa_suffix0(sa->sun_path, sizeof(sa->sun_path));
1298
1299         if (unlink(p) < 0)
1300                 return -errno;
1301
1302         return 1;
1303 }
1304
1305 int sockaddr_un_set_path(struct sockaddr_un *ret, const char *path) {
1306         size_t l;
1307
1308         assert(ret);
1309         assert(path);
1310
1311         /* Initialize ret->sun_path from the specified argument. This will interpret paths starting with '@' as
1312          * abstract namespace sockets, and those starting with '/' as regular filesystem sockets. It won't accept
1313          * anything else (i.e. no relative paths), to avoid ambiguities. Note that this function cannot be used to
1314          * reference paths in the abstract namespace that include NUL bytes in the name. */
1315
1316         l = strlen(path);
1317         if (l == 0)
1318                 return -EINVAL;
1319         if (!IN_SET(path[0], '/', '@'))
1320                 return -EINVAL;
1321         if (path[1] == 0)
1322                 return -EINVAL;
1323
1324         /* Don't allow paths larger than the space in sockaddr_un. Note that we are a tiny bit more restrictive than
1325          * the kernel is: we insist on NUL termination (both for abstract namespace and regular file system socket
1326          * addresses!), which the kernel doesn't. We do this to reduce chance of incompatibility with other apps that
1327          * do not expect non-NUL terminated file system path*/
1328         if (l+1 > sizeof(ret->sun_path))
1329                 return -EINVAL;
1330
1331         *ret = (struct sockaddr_un) {
1332                 .sun_family = AF_UNIX,
1333         };
1334
1335         if (path[0] == '@') {
1336                 /* Abstract namespace socket */
1337                 memcpy(ret->sun_path + 1, path + 1, l); /* copy *with* trailing NUL byte */
1338                 return (int) (offsetof(struct sockaddr_un, sun_path) + l); /* 🔥 *don't* 🔥 include trailing NUL in size */
1339
1340         } else {
1341                 assert(path[0] == '/');
1342
1343                 /* File system socket */
1344                 memcpy(ret->sun_path, path, l + 1); /* copy *with* trailing NUL byte */
1345                 return (int) (offsetof(struct sockaddr_un, sun_path) + l + 1); /* include trailing NUL in size */
1346         }
1347 }