src/libsystemd/sd-bus/bus-socket.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2 /***
   3   This file is part of systemd.
   4
   5   Copyright 2013 Lennart Poettering
   6
   7   systemd is free software; you can redistribute it and/or modify it
   8   under the terms of the GNU Lesser General Public License as published by
   9   the Free Software Foundation; either version 2.1 of the License, or
  10   (at your option) any later version.
  11
  12   systemd is distributed in the hope that it will be useful, but
  13   WITHOUT ANY WARRANTY; without even the implied warranty of
  14   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  15   Lesser General Public License for more details.
  16
  17   You should have received a copy of the GNU Lesser General Public License
  18   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  19 ***/
  20
  21 #include <endian.h>
  22 #include <poll.h>
  23 #include <stdlib.h>
  24 #include <unistd.h>
  25
  26 #include "sd-bus.h"
  27 #include "sd-daemon.h"
  28
  29 #include "alloc-util.h"
  30 #include "bus-internal.h"
  31 #include "bus-message.h"
  32 #include "bus-socket.h"
  33 #include "fd-util.h"
  34 #include "format-util.h"
  35 #include "fs-util.h"
  36 #include "hexdecoct.h"
  37 #include "io-util.h"
  38 #include "macro.h"
  39 #include "missing.h"
  40 #include "path-util.h"
  41 #include "process-util.h"
  42 #include "selinux-util.h"
  43 #include "signal-util.h"
  44 #include "stdio-util.h"
  45 #include "string-util.h"
  46 #include "user-util.h"
  47 #include "utf8.h"
  48 #include "util.h"
  49
  50 #define SNDBUF_SIZE (8*1024*1024)
  51
  52 static void iovec_advance(struct iovec iov[], unsigned *idx, size_t size) {
  53
  54         while (size > 0) {
  55                 struct iovec *i = iov + *idx;
  56
  57                 if (i->iov_len > size) {
  58                         i->iov_base = (uint8_t*) i->iov_base + size;
  59                         i->iov_len -= size;
  60                         return;
  61                 }
  62
  63                 size -= i->iov_len;
  64
  65                 i->iov_base = NULL;
  66                 i->iov_len = 0;
  67
  68                 (*idx)++;
  69         }
  70 }
  71
  72 static int append_iovec(sd_bus_message *m, const void *p, size_t sz) {
  73         assert(m);
  74         assert(p);
  75         assert(sz > 0);
  76
  77         m->iovec[m->n_iovec].iov_base = (void*) p;
  78         m->iovec[m->n_iovec].iov_len = sz;
  79         m->n_iovec++;
  80
  81         return 0;
  82 }
  83
  84 static int bus_message_setup_iovec(sd_bus_message *m) {
  85         struct bus_body_part *part;
  86         unsigned n, i;
  87         int r;
  88
  89         assert(m);
  90         assert(m->sealed);
  91
  92         if (m->n_iovec > 0)
  93                 return 0;
  94
  95         assert(!m->iovec);
  96
  97         n = 1 + m->n_body_parts;
  98         if (n < ELEMENTSOF(m->iovec_fixed))
  99                 m->iovec = m->iovec_fixed;
 100         else {
 101                 m->iovec = new(struct iovec, n);
 102                 if (!m->iovec) {
 103                         r = -ENOMEM;
 104                         goto fail;
 105                 }
 106         }
 107
 108         r = append_iovec(m, m->header, BUS_MESSAGE_BODY_BEGIN(m));
 109         if (r < 0)
 110                 goto fail;
 111
 112         MESSAGE_FOREACH_PART(part, i, m)  {
 113                 r = bus_body_part_map(part);
 114                 if (r < 0)
 115                         goto fail;
 116
 117                 r = append_iovec(m, part->data, part->size);
 118                 if (r < 0)
 119                         goto fail;
 120         }
 121
 122         assert(n == m->n_iovec);
 123
 124         return 0;
 125
 126 fail:
 127         m->poisoned = true;
 128         return r;
 129 }
 130
 131 bool bus_socket_auth_needs_write(sd_bus *b) {
 132
 133         unsigned i;
 134
 135         if (b->auth_index >= ELEMENTSOF(b->auth_iovec))
 136                 return false;
 137
 138         for (i = b->auth_index; i < ELEMENTSOF(b->auth_iovec); i++) {
 139                 struct iovec *j = b->auth_iovec + i;
 140
 141                 if (j->iov_len > 0)
 142                         return true;
 143         }
 144
 145         return false;
 146 }
 147
 148 static int bus_socket_write_auth(sd_bus *b) {
 149         ssize_t k;
 150
 151         assert(b);
 152         assert(b->state == BUS_AUTHENTICATING);
 153
 154         if (!bus_socket_auth_needs_write(b))
 155                 return 0;
 156
 157         if (b->prefer_writev)
 158                 k = writev(b->output_fd, b->auth_iovec + b->auth_index, ELEMENTSOF(b->auth_iovec) - b->auth_index);
 159         else {
 160                 struct msghdr mh;
 161                 zero(mh);
 162
 163                 mh.msg_iov = b->auth_iovec + b->auth_index;
 164                 mh.msg_iovlen = ELEMENTSOF(b->auth_iovec) - b->auth_index;
 165
 166                 k = sendmsg(b->output_fd, &mh, MSG_DONTWAIT|MSG_NOSIGNAL);
 167                 if (k < 0 && errno == ENOTSOCK) {
 168                         b->prefer_writev = true;
 169                         k = writev(b->output_fd, b->auth_iovec + b->auth_index, ELEMENTSOF(b->auth_iovec) - b->auth_index);
 170                 }
 171         }
 172
 173         if (k < 0)
 174                 return errno == EAGAIN ? 0 : -errno;
 175
 176         iovec_advance(b->auth_iovec, &b->auth_index, (size_t) k);
 177         return 1;
 178 }
 179
 180 static int bus_socket_auth_verify_client(sd_bus *b) {
 181         char *e, *f, *start;
 182         sd_id128_t peer;
 183         unsigned i;
 184         int r;
 185
 186         assert(b);
 187
 188         /* We expect two response lines: "OK" and possibly
 189          * "AGREE_UNIX_FD" */
 190
 191         e = memmem_safe(b->rbuffer, b->rbuffer_size, "\r\n", 2);
 192         if (!e)
 193                 return 0;
 194
 195         if (b->accept_fd) {
 196                 f = memmem(e + 2, b->rbuffer_size - (e - (char*) b->rbuffer) - 2, "\r\n", 2);
 197                 if (!f)
 198                         return 0;
 199
 200                 start = f + 2;
 201         } else {
 202                 f = NULL;
 203                 start = e + 2;
 204         }
 205
 206         /* Nice! We got all the lines we need. First check the OK
 207          * line */
 208
 209         if (e - (char*) b->rbuffer != 3 + 32)
 210                 return -EPERM;
 211
 212         if (memcmp(b->rbuffer, "OK ", 3))
 213                 return -EPERM;
 214
 215         b->auth = b->anonymous_auth ? BUS_AUTH_ANONYMOUS : BUS_AUTH_EXTERNAL;
 216
 217         for (i = 0; i < 32; i += 2) {
 218                 int x, y;
 219
 220                 x = unhexchar(((char*) b->rbuffer)[3 + i]);
 221                 y = unhexchar(((char*) b->rbuffer)[3 + i + 1]);
 222
 223                 if (x < 0 || y < 0)
 224                         return -EINVAL;
 225
 226                 peer.bytes[i/2] = ((uint8_t) x << 4 | (uint8_t) y);
 227         }
 228
 229         if (!sd_id128_is_null(b->server_id) &&
 230             !sd_id128_equal(b->server_id, peer))
 231                 return -EPERM;
 232
 233         b->server_id = peer;
 234
 235         /* And possibly check the second line, too */
 236
 237         if (f)
 238                 b->can_fds =
 239                         (f - e == STRLEN("\r\nAGREE_UNIX_FD")) &&
 240                         memcmp(e + 2, "AGREE_UNIX_FD",
 241                                STRLEN("AGREE_UNIX_FD")) == 0;
 242
 243         b->rbuffer_size -= (start - (char*) b->rbuffer);
 244         memmove(b->rbuffer, start, b->rbuffer_size);
 245
 246         r = bus_start_running(b);
 247         if (r < 0)
 248                 return r;
 249
 250         return 1;
 251 }
 252
 253 static bool line_equals(const char *s, size_t m, const char *line) {
 254         size_t l;
 255
 256         l = strlen(line);
 257         if (l != m)
 258                 return false;
 259
 260         return memcmp(s, line, l) == 0;
 261 }
 262
 263 static bool line_begins(const char *s, size_t m, const char *word) {
 264         size_t l;
 265
 266         l = strlen(word);
 267         if (m < l)
 268                 return false;
 269
 270         if (memcmp(s, word, l) != 0)
 271                 return false;
 272
 273         return m == l || (m > l && s[l] == ' ');
 274 }
 275
 276 static int verify_anonymous_token(sd_bus *b, const char *p, size_t l) {
 277         _cleanup_free_ char *token = NULL;
 278         size_t len;
 279         int r;
 280
 281         if (!b->anonymous_auth)
 282                 return 0;
 283
 284         if (l <= 0)
 285                 return 1;
 286
 287         assert(p[0] == ' ');
 288         p++; l--;
 289
 290         if (l % 2 != 0)
 291                 return 0;
 292
 293         r = unhexmem(p, l, (void **) &token, &len);
 294         if (r < 0)
 295                 return 0;
 296
 297         if (memchr(token, 0, len))
 298                 return 0;
 299
 300         return !!utf8_is_valid(token);
 301 }
 302
 303 static int verify_external_token(sd_bus *b, const char *p, size_t l) {
 304         _cleanup_free_ char *token = NULL;
 305         size_t len;
 306         uid_t u;
 307         int r;
 308
 309         /* We don't do any real authentication here. Instead, we if
 310          * the owner of this bus wanted authentication he should have
 311          * checked SO_PEERCRED before even creating the bus object. */
 312
 313         if (!b->anonymous_auth && !b->ucred_valid)
 314                 return 0;
 315
 316         if (l <= 0)
 317                 return 1;
 318
 319         assert(p[0] == ' ');
 320         p++; l--;
 321
 322         if (l % 2 != 0)
 323                 return 0;
 324
 325         r = unhexmem(p, l, (void**) &token, &len);
 326         if (r < 0)
 327                 return 0;
 328
 329         if (memchr(token, 0, len))
 330                 return 0;
 331
 332         r = parse_uid(token, &u);
 333         if (r < 0)
 334                 return 0;
 335
 336         /* We ignore the passed value if anonymous authentication is
 337          * on anyway. */
 338         if (!b->anonymous_auth && u != b->ucred.uid)
 339                 return 0;
 340
 341         return 1;
 342 }
 343
 344 static int bus_socket_auth_write(sd_bus *b, const char *t) {
 345         char *p;
 346         size_t l;
 347
 348         assert(b);
 349         assert(t);
 350
 351         /* We only make use of the first iovec */
 352         assert(IN_SET(b->auth_index, 0, 1));
 353
 354         l = strlen(t);
 355         p = malloc(b->auth_iovec[0].iov_len + l);
 356         if (!p)
 357                 return -ENOMEM;
 358
 359         memcpy_safe(p, b->auth_iovec[0].iov_base, b->auth_iovec[0].iov_len);
 360         memcpy(p + b->auth_iovec[0].iov_len, t, l);
 361
 362         b->auth_iovec[0].iov_base = p;
 363         b->auth_iovec[0].iov_len += l;
 364
 365         free(b->auth_buffer);
 366         b->auth_buffer = p;
 367         b->auth_index = 0;
 368         return 0;
 369 }
 370
 371 static int bus_socket_auth_write_ok(sd_bus *b) {
 372         char t[3 + 32 + 2 + 1];
 373
 374         assert(b);
 375
 376         xsprintf(t, "OK " SD_ID128_FORMAT_STR "\r\n", SD_ID128_FORMAT_VAL(b->server_id));
 377
 378         return bus_socket_auth_write(b, t);
 379 }
 380
 381 static int bus_socket_auth_verify_server(sd_bus *b) {
 382         char *e;
 383         const char *line;
 384         size_t l;
 385         bool processed = false;
 386         int r;
 387
 388         assert(b);
 389
 390         if (b->rbuffer_size < 1)
 391                 return 0;
 392
 393         /* First char must be a NUL byte */
 394         if (*(char*) b->rbuffer != 0)
 395                 return -EIO;
 396
 397         if (b->rbuffer_size < 3)
 398                 return 0;
 399
 400         /* Begin with the first line */
 401         if (b->auth_rbegin <= 0)
 402                 b->auth_rbegin = 1;
 403
 404         for (;;) {
 405                 /* Check if line is complete */
 406                 line = (char*) b->rbuffer + b->auth_rbegin;
 407                 e = memmem(line, b->rbuffer_size - b->auth_rbegin, "\r\n", 2);
 408                 if (!e)
 409                         return processed;
 410
 411                 l = e - line;
 412
 413                 if (line_begins(line, l, "AUTH ANONYMOUS")) {
 414
 415                         r = verify_anonymous_token(b, line + 14, l - 14);
 416                         if (r < 0)
 417                                 return r;
 418                         if (r == 0)
 419                                 r = bus_socket_auth_write(b, "REJECTED\r\n");
 420                         else {
 421                                 b->auth = BUS_AUTH_ANONYMOUS;
 422                                 r = bus_socket_auth_write_ok(b);
 423                         }
 424
 425                 } else if (line_begins(line, l, "AUTH EXTERNAL")) {
 426
 427                         r = verify_external_token(b, line + 13, l - 13);
 428                         if (r < 0)
 429                                 return r;
 430                         if (r == 0)
 431                                 r = bus_socket_auth_write(b, "REJECTED\r\n");
 432                         else {
 433                                 b->auth = BUS_AUTH_EXTERNAL;
 434                                 r = bus_socket_auth_write_ok(b);
 435                         }
 436
 437                 } else if (line_begins(line, l, "AUTH"))
 438                         r = bus_socket_auth_write(b, "REJECTED EXTERNAL ANONYMOUS\r\n");
 439                 else if (line_equals(line, l, "CANCEL") ||
 440                          line_begins(line, l, "ERROR")) {
 441
 442                         b->auth = _BUS_AUTH_INVALID;
 443                         r = bus_socket_auth_write(b, "REJECTED\r\n");
 444
 445                 } else if (line_equals(line, l, "BEGIN")) {
 446
 447                         if (b->auth == _BUS_AUTH_INVALID)
 448                                 r = bus_socket_auth_write(b, "ERROR\r\n");
 449                         else {
 450                                 /* We can't leave from the auth phase
 451                                  * before we haven't written
 452                                  * everything queued, so let's check
 453                                  * that */
 454
 455                                 if (bus_socket_auth_needs_write(b))
 456                                         return 1;
 457
 458                                 b->rbuffer_size -= (e + 2 - (char*) b->rbuffer);
 459                                 memmove(b->rbuffer, e + 2, b->rbuffer_size);
 460                                 return bus_start_running(b);
 461                         }
 462
 463                 } else if (line_begins(line, l, "DATA")) {
 464
 465                         if (b->auth == _BUS_AUTH_INVALID)
 466                                 r = bus_socket_auth_write(b, "ERROR\r\n");
 467                         else {
 468                                 if (b->auth == BUS_AUTH_ANONYMOUS)
 469                                         r = verify_anonymous_token(b, line + 4, l - 4);
 470                                 else
 471                                         r = verify_external_token(b, line + 4, l - 4);
 472
 473                                 if (r < 0)
 474                                         return r;
 475                                 if (r == 0) {
 476                                         b->auth = _BUS_AUTH_INVALID;
 477                                         r = bus_socket_auth_write(b, "REJECTED\r\n");
 478                                 } else
 479                                         r = bus_socket_auth_write_ok(b);
 480                         }
 481                 } else if (line_equals(line, l, "NEGOTIATE_UNIX_FD")) {
 482                         if (b->auth == _BUS_AUTH_INVALID || !b->accept_fd)
 483                                 r = bus_socket_auth_write(b, "ERROR\r\n");
 484                         else {
 485                                 b->can_fds = true;
 486                                 r = bus_socket_auth_write(b, "AGREE_UNIX_FD\r\n");
 487                         }
 488                 } else
 489                         r = bus_socket_auth_write(b, "ERROR\r\n");
 490
 491                 if (r < 0)
 492                         return r;
 493
 494                 b->auth_rbegin = e + 2 - (char*) b->rbuffer;
 495
 496                 processed = true;
 497         }
 498 }
 499
 500 static int bus_socket_auth_verify(sd_bus *b) {
 501         assert(b);
 502
 503         if (b->is_server)
 504                 return bus_socket_auth_verify_server(b);
 505         else
 506                 return bus_socket_auth_verify_client(b);
 507 }
 508
 509 static int bus_socket_read_auth(sd_bus *b) {
 510         struct msghdr mh;
 511         struct iovec iov = {};
 512         size_t n;
 513         ssize_t k;
 514         int r;
 515         void *p;
 516         union {
 517                 struct cmsghdr cmsghdr;
 518                 uint8_t buf[CMSG_SPACE(sizeof(int) * BUS_FDS_MAX)];
 519         } control;
 520         bool handle_cmsg = false;
 521
 522         assert(b);
 523         assert(b->state == BUS_AUTHENTICATING);
 524
 525         r = bus_socket_auth_verify(b);
 526         if (r != 0)
 527                 return r;
 528
 529         n = MAX(256u, b->rbuffer_size * 2);
 530
 531         if (n > BUS_AUTH_SIZE_MAX)
 532                 n = BUS_AUTH_SIZE_MAX;
 533
 534         if (b->rbuffer_size >= n)
 535                 return -ENOBUFS;
 536
 537         p = realloc(b->rbuffer, n);
 538         if (!p)
 539                 return -ENOMEM;
 540
 541         b->rbuffer = p;
 542
 543         iov.iov_base = (uint8_t*) b->rbuffer + b->rbuffer_size;
 544         iov.iov_len = n - b->rbuffer_size;
 545
 546         if (b->prefer_readv)
 547                 k = readv(b->input_fd, &iov, 1);
 548         else {
 549                 zero(mh);
 550                 mh.msg_iov = &iov;
 551                 mh.msg_iovlen = 1;
 552                 mh.msg_control = &control;
 553                 mh.msg_controllen = sizeof(control);
 554
 555                 k = recvmsg(b->input_fd, &mh, MSG_DONTWAIT|MSG_NOSIGNAL|MSG_CMSG_CLOEXEC);
 556                 if (k < 0 && errno == ENOTSOCK) {
 557                         b->prefer_readv = true;
 558                         k = readv(b->input_fd, &iov, 1);
 559                 } else
 560                         handle_cmsg = true;
 561         }
 562         if (k < 0)
 563                 return errno == EAGAIN ? 0 : -errno;
 564         if (k == 0)
 565                 return -ECONNRESET;
 566
 567         b->rbuffer_size += k;
 568
 569         if (handle_cmsg) {
 570                 struct cmsghdr *cmsg;
 571
 572                 CMSG_FOREACH(cmsg, &mh)
 573                         if (cmsg->cmsg_level == SOL_SOCKET &&
 574                             cmsg->cmsg_type == SCM_RIGHTS) {
 575                                 int j;
 576
 577                                 /* Whut? We received fds during the auth
 578                                  * protocol? Somebody is playing games with
 579                                  * us. Close them all, and fail */
 580                                 j = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
 581                                 close_many((int*) CMSG_DATA(cmsg), j);
 582                                 return -EIO;
 583                         } else
 584                                 log_debug("Got unexpected auxiliary data with level=%d and type=%d",
 585                                           cmsg->cmsg_level, cmsg->cmsg_type);
 586         }
 587
 588         r = bus_socket_auth_verify(b);
 589         if (r != 0)
 590                 return r;
 591
 592         return 1;
 593 }
 594
 595 void bus_socket_setup(sd_bus *b) {
 596         assert(b);
 597
 598         /* Increase the buffers to 8 MB */
 599         (void) fd_inc_rcvbuf(b->input_fd, SNDBUF_SIZE);
 600         (void) fd_inc_sndbuf(b->output_fd, SNDBUF_SIZE);
 601
 602         b->message_version = 1;
 603         b->message_endian = 0;
 604 }
 605
 606 static void bus_get_peercred(sd_bus *b) {
 607         int r;
 608
 609         assert(b);
 610         assert(!b->ucred_valid);
 611         assert(!b->label);
 612         assert(b->n_groups == (size_t) -1);
 613
 614         /* Get the peer for socketpair() sockets */
 615         b->ucred_valid = getpeercred(b->input_fd, &b->ucred) >= 0;
 616
 617         /* Get the SELinux context of the peer */
 618         r = getpeersec(b->input_fd, &b->label);
 619         if (r < 0 && !IN_SET(r, -EOPNOTSUPP, -ENOPROTOOPT))
 620                 log_debug_errno(r, "Failed to determine peer security context: %m");
 621
 622         /* Get the list of auxiliary groups of the peer */
 623         r = getpeergroups(b->input_fd, &b->groups);
 624         if (r >= 0)
 625                 b->n_groups = (size_t) r;
 626         else if (!IN_SET(r, -EOPNOTSUPP, -ENOPROTOOPT))
 627                 log_debug_errno(r, "Failed to determine peer's group list: %m");
 628 }
 629
 630 static int bus_socket_start_auth_client(sd_bus *b) {
 631         size_t l;
 632         const char *auth_suffix, *auth_prefix;
 633
 634         assert(b);
 635
 636         if (b->anonymous_auth) {
 637                 auth_prefix = "\0AUTH ANONYMOUS ";
 638
 639                 /* For ANONYMOUS auth we send some arbitrary "trace" string */
 640                 l = 9;
 641                 b->auth_buffer = hexmem("anonymous", l);
 642         } else {
 643                 char text[DECIMAL_STR_MAX(uid_t) + 1];
 644
 645                 auth_prefix = "\0AUTH EXTERNAL ";
 646
 647                 xsprintf(text, UID_FMT, geteuid());
 648
 649                 l = strlen(text);
 650                 b->auth_buffer = hexmem(text, l);
 651         }
 652
 653         if (!b->auth_buffer)
 654                 return -ENOMEM;
 655
 656         if (b->accept_fd)
 657                 auth_suffix = "\r\nNEGOTIATE_UNIX_FD\r\nBEGIN\r\n";
 658         else
 659                 auth_suffix = "\r\nBEGIN\r\n";
 660
 661         b->auth_iovec[0].iov_base = (void*) auth_prefix;
 662         b->auth_iovec[0].iov_len = 1 + strlen(auth_prefix + 1);
 663         b->auth_iovec[1].iov_base = (void*) b->auth_buffer;
 664         b->auth_iovec[1].iov_len = l * 2;
 665         b->auth_iovec[2].iov_base = (void*) auth_suffix;
 666         b->auth_iovec[2].iov_len = strlen(auth_suffix);
 667
 668         return bus_socket_write_auth(b);
 669 }
 670
 671 int bus_socket_start_auth(sd_bus *b) {
 672         assert(b);
 673
 674         bus_get_peercred(b);
 675
 676         bus_set_state(b, BUS_AUTHENTICATING);
 677         b->auth_timeout = now(CLOCK_MONOTONIC) + BUS_AUTH_TIMEOUT;
 678
 679         if (sd_is_socket(b->input_fd, AF_UNIX, 0, 0) <= 0)
 680                 b->accept_fd = false;
 681
 682         if (b->output_fd != b->input_fd)
 683                 if (sd_is_socket(b->output_fd, AF_UNIX, 0, 0) <= 0)
 684                         b->accept_fd = false;
 685
 686         if (b->is_server)
 687                 return bus_socket_read_auth(b);
 688         else
 689                 return bus_socket_start_auth_client(b);
 690 }
 691
 692 static int bus_socket_inotify_setup(sd_bus *b) {
 693         _cleanup_free_ int *new_watches = NULL;
 694         _cleanup_free_ char *absolute = NULL;
 695         size_t n_allocated = 0, n = 0, done = 0, i;
 696         unsigned max_follow = 32;
 697         const char *p;
 698         int wd, r;
 699
 700         assert(b);
 701         assert(b->watch_bind);
 702         assert(b->sockaddr.sa.sa_family == AF_UNIX);
 703         assert(b->sockaddr.un.sun_path[0] != 0);
 704
 705         /* Sets up an inotify fd in case watch_bind is enabled: wait until the configured AF_UNIX file system socket
 706          * appears before connecting to it. The implemented is pretty simplistic: we just subscribe to relevant changes
 707          * to all prefix components of the path, and every time we get an event for that we try to reconnect again,
 708          * without actually caring what precisely the event we got told us. If we still can't connect we re-subscribe
 709          * to all relevant changes of anything in the path, so that our watches include any possibly newly created path
 710          * components. */
 711
 712         if (b->inotify_fd < 0) {
 713                 b->inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
 714                 if (b->inotify_fd < 0)
 715                         return -errno;
 716         }
 717
 718         /* Make sure the path is NUL terminated */
 719         p = strndupa(b->sockaddr.un.sun_path, sizeof(b->sockaddr.un.sun_path));
 720
 721         /* Make sure the path is absolute */
 722         r = path_make_absolute_cwd(p, &absolute);
 723         if (r < 0)
 724                 goto fail;
 725
 726         /* Watch all parent directories, and don't mind any prefix that doesn't exist yet. For the innermost directory
 727          * that exists we want to know when files are created or moved into it. For all parents of it we just care if
 728          * they are removed or renamed. */
 729
 730         if (!GREEDY_REALLOC(new_watches, n_allocated, n + 1)) {
 731                 r = -ENOMEM;
 732                 goto fail;
 733         }
 734
 735         /* Start with the top-level directory, which is a bit simpler than the rest, since it can't be a symlink, and
 736          * always exists */
 737         wd = inotify_add_watch(b->inotify_fd, "/", IN_CREATE|IN_MOVED_TO);
 738         if (wd < 0) {
 739                 r = log_debug_errno(errno, "Failed to add inotify watch on /: %m");
 740                 goto fail;
 741         } else
 742                 new_watches[n++] = wd;
 743
 744         for (;;) {
 745                 _cleanup_free_ char *component = NULL, *prefix = NULL, *destination = NULL;
 746                 size_t n_slashes, n_component;
 747                 char *c = NULL;
 748
 749                 n_slashes = strspn(absolute + done, "/");
 750                 n_component = n_slashes + strcspn(absolute + done + n_slashes, "/");
 751
 752                 if (n_component == 0) /* The end */
 753                         break;
 754
 755                 component = strndup(absolute + done, n_component);
 756                 if (!component) {
 757                         r = -ENOMEM;
 758                         goto fail;
 759                 }
 760
 761                 /* A trailing slash? That's a directory, and not a socket then */
 762                 if (path_equal(component, "/")) {
 763                         r = -EISDIR;
 764                         goto fail;
 765                 }
 766
 767                 /* A single dot? Let's eat this up */
 768                 if (path_equal(component, "/.")) {
 769                         done += n_component;
 770                         continue;
 771                 }
 772
 773                 prefix = strndup(absolute, done + n_component);
 774                 if (!prefix) {
 775                         r = -ENOMEM;
 776                         goto fail;
 777                 }
 778
 779                 if (!GREEDY_REALLOC(new_watches, n_allocated, n + 1)) {
 780                         r = -ENOMEM;
 781                         goto fail;
 782                 }
 783
 784                 wd = inotify_add_watch(b->inotify_fd, prefix, IN_DELETE_SELF|IN_MOVE_SELF|IN_ATTRIB|IN_CREATE|IN_MOVED_TO|IN_DONT_FOLLOW);
 785                 log_debug("Added inotify watch for %s on bus %s: %i", prefix, strna(b->description), wd);
 786
 787                 if (wd < 0) {
 788                         if (IN_SET(errno, ENOENT, ELOOP))
 789                                 break; /* This component doesn't exist yet, or the path contains a cyclic symlink right now */
 790
 791                         r = log_debug_errno(errno, "Failed to add inotify watch on %s: %m", isempty(prefix) ? "/" : prefix);
 792                         goto fail;
 793                 } else
 794                         new_watches[n++] = wd;
 795
 796                 /* Check if this is possibly a symlink. If so, let's follow it and watch it too. */
 797                 r = readlink_malloc(prefix, &destination);
 798                 if (r == -EINVAL) { /* not a symlink */
 799                         done += n_component;
 800                         continue;
 801                 }
 802                 if (r < 0)
 803                         goto fail;
 804
 805                 if (isempty(destination)) { /* Empty symlink target? Yuck! */
 806                         r = -EINVAL;
 807                         goto fail;
 808                 }
 809
 810                 if (max_follow <= 0) { /* Let's make sure we don't follow symlinks forever */
 811                         r = -ELOOP;
 812                         goto fail;
 813                 }
 814
 815                 if (path_is_absolute(destination)) {
 816                         /* For absolute symlinks we build the new path and start anew */
 817                         c = strjoin(destination, absolute + done + n_component);
 818                         done = 0;
 819                 } else {
 820                         _cleanup_free_ char *t = NULL;
 821
 822                         /* For relative symlinks we replace the last component, and try again */
 823                         t = strndup(absolute, done);
 824                         if (!t)
 825                                 return -ENOMEM;
 826
 827                         c = strjoin(t, "/", destination, absolute + done + n_component);
 828                 }
 829                 if (!c) {
 830                         r = -ENOMEM;
 831                         goto fail;
 832                 }
 833
 834                 free(absolute);
 835                 absolute = c;
 836
 837                 max_follow--;
 838         }
 839
 840         /* And now, let's remove all watches from the previous iteration we don't need anymore */
 841         for (i = 0; i < b->n_inotify_watches; i++) {
 842                 bool found = false;
 843                 size_t j;
 844
 845                 for (j = 0; j < n; j++)
 846                         if (new_watches[j] == b->inotify_watches[i]) {
 847                                 found = true;
 848                                 break;
 849                         }
 850
 851                 if (found)
 852                         continue;
 853
 854                 (void) inotify_rm_watch(b->inotify_fd, b->inotify_watches[i]);
 855         }
 856
 857         free_and_replace(b->inotify_watches, new_watches);
 858         b->n_inotify_watches = n;
 859
 860         return 0;
 861
 862 fail:
 863         bus_close_inotify_fd(b);
 864         return r;
 865 }
 866
 867 int bus_socket_connect(sd_bus *b) {
 868         bool inotify_done = false;
 869         int r;
 870
 871         assert(b);
 872
 873         for (;;) {
 874                 assert(b->input_fd < 0);
 875                 assert(b->output_fd < 0);
 876                 assert(b->sockaddr.sa.sa_family != AF_UNSPEC);
 877
 878                 b->input_fd = socket(b->sockaddr.sa.sa_family, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
 879                 if (b->input_fd < 0)
 880                         return -errno;
 881
 882                 b->output_fd = b->input_fd;
 883                 bus_socket_setup(b);
 884
 885                 if (connect(b->input_fd, &b->sockaddr.sa, b->sockaddr_size) < 0) {
 886                         if (errno == EINPROGRESS) {
 887
 888                                 /* If we have any inotify watches open, close them now, we don't need them anymore, as
 889                                  * we have successfully initiated a connection */
 890                                 bus_close_inotify_fd(b);
 891
 892                                 /* Note that very likely we are already in BUS_OPENING state here, as we enter it when
 893                                  * we start parsing the address string. The only reason we set the state explicitly
 894                                  * here, is to undo BUS_WATCH_BIND, in case we did the inotify magic. */
 895                                 bus_set_state(b, BUS_OPENING);
 896                                 return 1;
 897                         }
 898
 899                         if (IN_SET(errno, ENOENT, ECONNREFUSED) &&  /* ENOENT → unix socket doesn't exist at all; ECONNREFUSED → unix socket stale */
 900                             b->watch_bind &&
 901                             b->sockaddr.sa.sa_family == AF_UNIX &&
 902                             b->sockaddr.un.sun_path[0] != 0) {
 903
 904                                 /* This connection attempt failed, let's release the socket for now, and start with a
 905                                  * fresh one when reconnecting. */
 906                                 bus_close_io_fds(b);
 907
 908                                 if (inotify_done) {
 909                                         /* inotify set up already, don't do it again, just return now, and remember
 910                                          * that we are waiting for inotify events now. */
 911                                         bus_set_state(b, BUS_WATCH_BIND);
 912                                         return 1;
 913                                 }
 914
 915                                 /* This is a file system socket, and the inotify logic is enabled. Let's create the necessary inotify fd. */
 916                                 r = bus_socket_inotify_setup(b);
 917                                 if (r < 0)
 918                                         return r;
 919
 920                                 /* Let's now try to connect a second time, because in theory there's otherwise a race
 921                                  * here: the socket might have been created in the time between our first connect() and
 922                                  * the time we set up the inotify logic. But let's remember that we set up inotify now,
 923                                  * so that we don't do the connect() more than twice. */
 924                                 inotify_done = true;
 925
 926                         } else
 927                                 return -errno;
 928                 } else
 929                         break;
 930         }
 931
 932         /* Yay, established, we don't need no inotify anymore! */
 933         bus_close_inotify_fd(b);
 934
 935         return bus_socket_start_auth(b);
 936 }
 937
 938 int bus_socket_exec(sd_bus *b) {
 939         int s[2], r;
 940         pid_t pid;
 941
 942         assert(b);
 943         assert(b->input_fd < 0);
 944         assert(b->output_fd < 0);
 945         assert(b->exec_path);
 946
 947         r = socketpair(AF_UNIX, SOCK_STREAM|SOCK_NONBLOCK|SOCK_CLOEXEC, 0, s);
 948         if (r < 0)
 949                 return -errno;
 950
 951         r = safe_fork_full("(sd-busexec)", s+1, 1, FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS, &pid);
 952         if (r < 0) {
 953                 safe_close_pair(s);
 954                 return r;
 955         }
 956         if (r == 0) {
 957                 /* Child */
 958
 959                 assert_se(dup3(s[1], STDIN_FILENO, 0) == STDIN_FILENO);
 960                 assert_se(dup3(s[1], STDOUT_FILENO, 0) == STDOUT_FILENO);
 961
 962                 if (!IN_SET(s[1], STDIN_FILENO, STDOUT_FILENO))
 963                         safe_close(s[1]);
 964
 965                 (void) fd_cloexec(STDIN_FILENO, false);
 966                 (void) fd_cloexec(STDOUT_FILENO, false);
 967                 (void) fd_nonblock(STDIN_FILENO, false);
 968                 (void) fd_nonblock(STDOUT_FILENO, false);
 969
 970                 if (b->exec_argv)
 971                         execvp(b->exec_path, b->exec_argv);
 972                 else {
 973                         const char *argv[] = { b->exec_path, NULL };
 974                         execvp(b->exec_path, (char**) argv);
 975                 }
 976
 977                 _exit(EXIT_FAILURE);
 978         }
 979
 980         safe_close(s[1]);
 981         b->output_fd = b->input_fd = s[0];
 982
 983         bus_socket_setup(b);
 984
 985         return bus_socket_start_auth(b);
 986 }
 987
 988 int bus_socket_take_fd(sd_bus *b) {
 989         assert(b);
 990
 991         bus_socket_setup(b);
 992
 993         return bus_socket_start_auth(b);
 994 }
 995
 996 int bus_socket_write_message(sd_bus *bus, sd_bus_message *m, size_t *idx) {
 997         struct iovec *iov;
 998         ssize_t k;
 999         size_t n;
1000         unsigned j;
1001         int r;
1002
1003         assert(bus);
1004         assert(m);
1005         assert(idx);
1006         assert(IN_SET(bus->state, BUS_RUNNING, BUS_HELLO));
1007
1008         if (*idx >= BUS_MESSAGE_SIZE(m))
1009                 return 0;
1010
1011         r = bus_message_setup_iovec(m);
1012         if (r < 0)
1013                 return r;
1014
1015         n = m->n_iovec * sizeof(struct iovec);
1016         iov = alloca(n);
1017         memcpy_safe(iov, m->iovec, n);
1018
1019         j = 0;
1020         iovec_advance(iov, &j, *idx);
1021
1022         if (bus->prefer_writev)
1023                 k = writev(bus->output_fd, iov, m->n_iovec);
1024         else {
1025                 struct msghdr mh = {
1026                         .msg_iov = iov,
1027                         .msg_iovlen = m->n_iovec,
1028                 };
1029
1030                 if (m->n_fds > 0 && *idx == 0) {
1031                         struct cmsghdr *control;
1032
1033                         mh.msg_control = control = alloca(CMSG_SPACE(sizeof(int) * m->n_fds));
1034                         mh.msg_controllen = control->cmsg_len = CMSG_LEN(sizeof(int) * m->n_fds);
1035                         control->cmsg_level = SOL_SOCKET;
1036                         control->cmsg_type = SCM_RIGHTS;
1037                         memcpy(CMSG_DATA(control), m->fds, sizeof(int) * m->n_fds);
1038                 }
1039
1040                 k = sendmsg(bus->output_fd, &mh, MSG_DONTWAIT|MSG_NOSIGNAL);
1041                 if (k < 0 && errno == ENOTSOCK) {
1042                         bus->prefer_writev = true;
1043                         k = writev(bus->output_fd, iov, m->n_iovec);
1044                 }
1045         }
1046
1047         if (k < 0)
1048                 return errno == EAGAIN ? 0 : -errno;
1049
1050         *idx += (size_t) k;
1051         return 1;
1052 }
1053
1054 static int bus_socket_read_message_need(sd_bus *bus, size_t *need) {
1055         uint32_t a, b;
1056         uint8_t e;
1057         uint64_t sum;
1058
1059         assert(bus);
1060         assert(need);
1061         assert(IN_SET(bus->state, BUS_RUNNING, BUS_HELLO));
1062
1063         if (bus->rbuffer_size < sizeof(struct bus_header)) {
1064                 *need = sizeof(struct bus_header) + 8;
1065
1066                 /* Minimum message size:
1067                  *
1068                  * Header +
1069                  *
1070                  *  Method Call: +2 string headers
1071                  *       Signal: +3 string headers
1072                  * Method Error: +1 string headers
1073                  *               +1 uint32 headers
1074                  * Method Reply: +1 uint32 headers
1075                  *
1076                  * A string header is at least 9 bytes
1077                  * A uint32 header is at least 8 bytes
1078                  *
1079                  * Hence the minimum message size of a valid message
1080                  * is header + 8 bytes */
1081
1082                 return 0;
1083         }
1084
1085         a = ((const uint32_t*) bus->rbuffer)[1];
1086         b = ((const uint32_t*) bus->rbuffer)[3];
1087
1088         e = ((const uint8_t*) bus->rbuffer)[0];
1089         if (e == BUS_LITTLE_ENDIAN) {
1090                 a = le32toh(a);
1091                 b = le32toh(b);
1092         } else if (e == BUS_BIG_ENDIAN) {
1093                 a = be32toh(a);
1094                 b = be32toh(b);
1095         } else
1096                 return -EBADMSG;
1097
1098         sum = (uint64_t) sizeof(struct bus_header) + (uint64_t) ALIGN_TO(b, 8) + (uint64_t) a;
1099         if (sum >= BUS_MESSAGE_SIZE_MAX)
1100                 return -ENOBUFS;
1101
1102         *need = (size_t) sum;
1103         return 0;
1104 }
1105
1106 static int bus_socket_make_message(sd_bus *bus, size_t size) {
1107         sd_bus_message *t;
1108         void *b;
1109         int r;
1110
1111         assert(bus);
1112         assert(bus->rbuffer_size >= size);
1113         assert(IN_SET(bus->state, BUS_RUNNING, BUS_HELLO));
1114
1115         r = bus_rqueue_make_room(bus);
1116         if (r < 0)
1117                 return r;
1118
1119         if (bus->rbuffer_size > size) {
1120                 b = memdup((const uint8_t*) bus->rbuffer + size,
1121                            bus->rbuffer_size - size);
1122                 if (!b)
1123                         return -ENOMEM;
1124         } else
1125                 b = NULL;
1126
1127         r = bus_message_from_malloc(bus,
1128                                     bus->rbuffer, size,
1129                                     bus->fds, bus->n_fds,
1130                                     NULL,
1131                                     &t);
1132         if (r < 0) {
1133                 free(b);
1134                 return r;
1135         }
1136
1137         bus->rbuffer = b;
1138         bus->rbuffer_size -= size;
1139
1140         bus->fds = NULL;
1141         bus->n_fds = 0;
1142
1143         bus->rqueue[bus->rqueue_size++] = t;
1144
1145         return 1;
1146 }
1147
1148 int bus_socket_read_message(sd_bus *bus) {
1149         struct msghdr mh;
1150         struct iovec iov = {};
1151         ssize_t k;
1152         size_t need;
1153         int r;
1154         void *b;
1155         union {
1156                 struct cmsghdr cmsghdr;
1157                 uint8_t buf[CMSG_SPACE(sizeof(int) * BUS_FDS_MAX)];
1158         } control;
1159         bool handle_cmsg = false;
1160
1161         assert(bus);
1162         assert(IN_SET(bus->state, BUS_RUNNING, BUS_HELLO));
1163
1164         r = bus_socket_read_message_need(bus, &need);
1165         if (r < 0)
1166                 return r;
1167
1168         if (bus->rbuffer_size >= need)
1169                 return bus_socket_make_message(bus, need);
1170
1171         b = realloc(bus->rbuffer, need);
1172         if (!b)
1173                 return -ENOMEM;
1174
1175         bus->rbuffer = b;
1176
1177         iov.iov_base = (uint8_t*) bus->rbuffer + bus->rbuffer_size;
1178         iov.iov_len = need - bus->rbuffer_size;
1179
1180         if (bus->prefer_readv)
1181                 k = readv(bus->input_fd, &iov, 1);
1182         else {
1183                 zero(mh);
1184                 mh.msg_iov = &iov;
1185                 mh.msg_iovlen = 1;
1186                 mh.msg_control = &control;
1187                 mh.msg_controllen = sizeof(control);
1188
1189                 k = recvmsg(bus->input_fd, &mh, MSG_DONTWAIT|MSG_NOSIGNAL|MSG_CMSG_CLOEXEC);
1190                 if (k < 0 && errno == ENOTSOCK) {
1191                         bus->prefer_readv = true;
1192                         k = readv(bus->input_fd, &iov, 1);
1193                 } else
1194                         handle_cmsg = true;
1195         }
1196         if (k < 0)
1197                 return errno == EAGAIN ? 0 : -errno;
1198         if (k == 0)
1199                 return -ECONNRESET;
1200
1201         bus->rbuffer_size += k;
1202
1203         if (handle_cmsg) {
1204                 struct cmsghdr *cmsg;
1205
1206                 CMSG_FOREACH(cmsg, &mh)
1207                         if (cmsg->cmsg_level == SOL_SOCKET &&
1208                             cmsg->cmsg_type == SCM_RIGHTS) {
1209                                 int n, *f;
1210
1211                                 n = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
1212
1213                                 if (!bus->can_fds) {
1214                                         /* Whut? We received fds but this
1215                                          * isn't actually enabled? Close them,
1216                                          * and fail */
1217
1218                                         close_many((int*) CMSG_DATA(cmsg), n);
1219                                         return -EIO;
1220                                 }
1221
1222                                 f = realloc(bus->fds, sizeof(int) * (bus->n_fds + n));
1223                                 if (!f) {
1224                                         close_many((int*) CMSG_DATA(cmsg), n);
1225                                         return -ENOMEM;
1226                                 }
1227
1228                                 memcpy_safe(f + bus->n_fds, CMSG_DATA(cmsg), n * sizeof(int));
1229                                 bus->fds = f;
1230                                 bus->n_fds += n;
1231                         } else
1232                                 log_debug("Got unexpected auxiliary data with level=%d and type=%d",
1233                                           cmsg->cmsg_level, cmsg->cmsg_type);
1234         }
1235
1236         r = bus_socket_read_message_need(bus, &need);
1237         if (r < 0)
1238                 return r;
1239
1240         if (bus->rbuffer_size >= need)
1241                 return bus_socket_make_message(bus, need);
1242
1243         return 1;
1244 }
1245
1246 int bus_socket_process_opening(sd_bus *b) {
1247         int error = 0;
1248         socklen_t slen = sizeof(error);
1249         struct pollfd p = {
1250                 .fd = b->output_fd,
1251                 .events = POLLOUT,
1252         };
1253         int r;
1254
1255         assert(b->state == BUS_OPENING);
1256
1257         r = poll(&p, 1, 0);
1258         if (r < 0)
1259                 return -errno;
1260
1261         if (!(p.revents & (POLLOUT|POLLERR|POLLHUP)))
1262                 return 0;
1263
1264         r = getsockopt(b->output_fd, SOL_SOCKET, SO_ERROR, &error, &slen);
1265         if (r < 0)
1266                 b->last_connect_error = errno;
1267         else if (error != 0)
1268                 b->last_connect_error = error;
1269         else if (p.revents & (POLLERR|POLLHUP))
1270                 b->last_connect_error = ECONNREFUSED;
1271         else
1272                 return bus_socket_start_auth(b);
1273
1274         return bus_next_address(b);
1275 }
1276
1277 int bus_socket_process_authenticating(sd_bus *b) {
1278         int r;
1279
1280         assert(b);
1281         assert(b->state == BUS_AUTHENTICATING);
1282
1283         if (now(CLOCK_MONOTONIC) >= b->auth_timeout)
1284                 return -ETIMEDOUT;
1285
1286         r = bus_socket_write_auth(b);
1287         if (r != 0)
1288                 return r;
1289
1290         return bus_socket_read_auth(b);
1291 }
1292
1293 int bus_socket_process_watch_bind(sd_bus *b) {
1294         int r, q;
1295
1296         assert(b);
1297         assert(b->state == BUS_WATCH_BIND);
1298         assert(b->inotify_fd >= 0);
1299
1300         r = flush_fd(b->inotify_fd);
1301         if (r <= 0)
1302                 return r;
1303
1304         log_debug("Got inotify event on bus %s.", strna(b->description));
1305
1306         /* We flushed events out of the inotify fd. In that case, maybe the socket is valid now? Let's try to connect
1307          * to it again */
1308
1309         r = bus_socket_connect(b);
1310         if (r < 0)
1311                 return r;
1312
1313         q = bus_attach_io_events(b);
1314         if (q < 0)
1315                 return q;
1316
1317         q = bus_attach_inotify_event(b);
1318         if (q < 0)
1319                 return q;
1320
1321         return r;
1322 }