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                 b->inotify_fd = fd_move_above_stdio(b->inotify_fd);
 718         }
 719
 720         /* Make sure the path is NUL terminated */
 721         p = strndupa(b->sockaddr.un.sun_path, sizeof(b->sockaddr.un.sun_path));
 722
 723         /* Make sure the path is absolute */
 724         r = path_make_absolute_cwd(p, &absolute);
 725         if (r < 0)
 726                 goto fail;
 727
 728         /* Watch all parent directories, and don't mind any prefix that doesn't exist yet. For the innermost directory
 729          * that exists we want to know when files are created or moved into it. For all parents of it we just care if
 730          * they are removed or renamed. */
 731
 732         if (!GREEDY_REALLOC(new_watches, n_allocated, n + 1)) {
 733                 r = -ENOMEM;
 734                 goto fail;
 735         }
 736
 737         /* Start with the top-level directory, which is a bit simpler than the rest, since it can't be a symlink, and
 738          * always exists */
 739         wd = inotify_add_watch(b->inotify_fd, "/", IN_CREATE|IN_MOVED_TO);
 740         if (wd < 0) {
 741                 r = log_debug_errno(errno, "Failed to add inotify watch on /: %m");
 742                 goto fail;
 743         } else
 744                 new_watches[n++] = wd;
 745
 746         for (;;) {
 747                 _cleanup_free_ char *component = NULL, *prefix = NULL, *destination = NULL;
 748                 size_t n_slashes, n_component;
 749                 char *c = NULL;
 750
 751                 n_slashes = strspn(absolute + done, "/");
 752                 n_component = n_slashes + strcspn(absolute + done + n_slashes, "/");
 753
 754                 if (n_component == 0) /* The end */
 755                         break;
 756
 757                 component = strndup(absolute + done, n_component);
 758                 if (!component) {
 759                         r = -ENOMEM;
 760                         goto fail;
 761                 }
 762
 763                 /* A trailing slash? That's a directory, and not a socket then */
 764                 if (path_equal(component, "/")) {
 765                         r = -EISDIR;
 766                         goto fail;
 767                 }
 768
 769                 /* A single dot? Let's eat this up */
 770                 if (path_equal(component, "/.")) {
 771                         done += n_component;
 772                         continue;
 773                 }
 774
 775                 prefix = strndup(absolute, done + n_component);
 776                 if (!prefix) {
 777                         r = -ENOMEM;
 778                         goto fail;
 779                 }
 780
 781                 if (!GREEDY_REALLOC(new_watches, n_allocated, n + 1)) {
 782                         r = -ENOMEM;
 783                         goto fail;
 784                 }
 785
 786                 wd = inotify_add_watch(b->inotify_fd, prefix, IN_DELETE_SELF|IN_MOVE_SELF|IN_ATTRIB|IN_CREATE|IN_MOVED_TO|IN_DONT_FOLLOW);
 787                 log_debug("Added inotify watch for %s on bus %s: %i", prefix, strna(b->description), wd);
 788
 789                 if (wd < 0) {
 790                         if (IN_SET(errno, ENOENT, ELOOP))
 791                                 break; /* This component doesn't exist yet, or the path contains a cyclic symlink right now */
 792
 793                         r = log_debug_errno(errno, "Failed to add inotify watch on %s: %m", isempty(prefix) ? "/" : prefix);
 794                         goto fail;
 795                 } else
 796                         new_watches[n++] = wd;
 797
 798                 /* Check if this is possibly a symlink. If so, let's follow it and watch it too. */
 799                 r = readlink_malloc(prefix, &destination);
 800                 if (r == -EINVAL) { /* not a symlink */
 801                         done += n_component;
 802                         continue;
 803                 }
 804                 if (r < 0)
 805                         goto fail;
 806
 807                 if (isempty(destination)) { /* Empty symlink target? Yuck! */
 808                         r = -EINVAL;
 809                         goto fail;
 810                 }
 811
 812                 if (max_follow <= 0) { /* Let's make sure we don't follow symlinks forever */
 813                         r = -ELOOP;
 814                         goto fail;
 815                 }
 816
 817                 if (path_is_absolute(destination)) {
 818                         /* For absolute symlinks we build the new path and start anew */
 819                         c = strjoin(destination, absolute + done + n_component);
 820                         done = 0;
 821                 } else {
 822                         _cleanup_free_ char *t = NULL;
 823
 824                         /* For relative symlinks we replace the last component, and try again */
 825                         t = strndup(absolute, done);
 826                         if (!t)
 827                                 return -ENOMEM;
 828
 829                         c = strjoin(t, "/", destination, absolute + done + n_component);
 830                 }
 831                 if (!c) {
 832                         r = -ENOMEM;
 833                         goto fail;
 834                 }
 835
 836                 free(absolute);
 837                 absolute = c;
 838
 839                 max_follow--;
 840         }
 841
 842         /* And now, let's remove all watches from the previous iteration we don't need anymore */
 843         for (i = 0; i < b->n_inotify_watches; i++) {
 844                 bool found = false;
 845                 size_t j;
 846
 847                 for (j = 0; j < n; j++)
 848                         if (new_watches[j] == b->inotify_watches[i]) {
 849                                 found = true;
 850                                 break;
 851                         }
 852
 853                 if (found)
 854                         continue;
 855
 856                 (void) inotify_rm_watch(b->inotify_fd, b->inotify_watches[i]);
 857         }
 858
 859         free_and_replace(b->inotify_watches, new_watches);
 860         b->n_inotify_watches = n;
 861
 862         return 0;
 863
 864 fail:
 865         bus_close_inotify_fd(b);
 866         return r;
 867 }
 868
 869 int bus_socket_connect(sd_bus *b) {
 870         bool inotify_done = false;
 871         int r;
 872
 873         assert(b);
 874
 875         for (;;) {
 876                 assert(b->input_fd < 0);
 877                 assert(b->output_fd < 0);
 878                 assert(b->sockaddr.sa.sa_family != AF_UNSPEC);
 879
 880                 b->input_fd = socket(b->sockaddr.sa.sa_family, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
 881                 if (b->input_fd < 0)
 882                         return -errno;
 883
 884                 b->input_fd = fd_move_above_stdio(b->input_fd);
 885
 886                 b->output_fd = b->input_fd;
 887                 bus_socket_setup(b);
 888
 889                 if (connect(b->input_fd, &b->sockaddr.sa, b->sockaddr_size) < 0) {
 890                         if (errno == EINPROGRESS) {
 891
 892                                 /* If we have any inotify watches open, close them now, we don't need them anymore, as
 893                                  * we have successfully initiated a connection */
 894                                 bus_close_inotify_fd(b);
 895
 896                                 /* Note that very likely we are already in BUS_OPENING state here, as we enter it when
 897                                  * we start parsing the address string. The only reason we set the state explicitly
 898                                  * here, is to undo BUS_WATCH_BIND, in case we did the inotify magic. */
 899                                 bus_set_state(b, BUS_OPENING);
 900                                 return 1;
 901                         }
 902
 903                         if (IN_SET(errno, ENOENT, ECONNREFUSED) &&  /* ENOENT → unix socket doesn't exist at all; ECONNREFUSED → unix socket stale */
 904                             b->watch_bind &&
 905                             b->sockaddr.sa.sa_family == AF_UNIX &&
 906                             b->sockaddr.un.sun_path[0] != 0) {
 907
 908                                 /* This connection attempt failed, let's release the socket for now, and start with a
 909                                  * fresh one when reconnecting. */
 910                                 bus_close_io_fds(b);
 911
 912                                 if (inotify_done) {
 913                                         /* inotify set up already, don't do it again, just return now, and remember
 914                                          * that we are waiting for inotify events now. */
 915                                         bus_set_state(b, BUS_WATCH_BIND);
 916                                         return 1;
 917                                 }
 918
 919                                 /* This is a file system socket, and the inotify logic is enabled. Let's create the necessary inotify fd. */
 920                                 r = bus_socket_inotify_setup(b);
 921                                 if (r < 0)
 922                                         return r;
 923
 924                                 /* Let's now try to connect a second time, because in theory there's otherwise a race
 925                                  * here: the socket might have been created in the time between our first connect() and
 926                                  * the time we set up the inotify logic. But let's remember that we set up inotify now,
 927                                  * so that we don't do the connect() more than twice. */
 928                                 inotify_done = true;
 929
 930                         } else
 931                                 return -errno;
 932                 } else
 933                         break;
 934         }
 935
 936         /* Yay, established, we don't need no inotify anymore! */
 937         bus_close_inotify_fd(b);
 938
 939         return bus_socket_start_auth(b);
 940 }
 941
 942 int bus_socket_exec(sd_bus *b) {
 943         int s[2], r;
 944
 945         assert(b);
 946         assert(b->input_fd < 0);
 947         assert(b->output_fd < 0);
 948         assert(b->exec_path);
 949         assert(b->busexec_pid == 0);
 950
 951         r = socketpair(AF_UNIX, SOCK_STREAM|SOCK_NONBLOCK|SOCK_CLOEXEC, 0, s);
 952         if (r < 0)
 953                 return -errno;
 954
 955         r = safe_fork_full("(sd-busexec)", s+1, 1, FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS, &b->busexec_pid);
 956         if (r < 0) {
 957                 safe_close_pair(s);
 958                 return r;
 959         }
 960         if (r == 0) {
 961                 /* Child */
 962
 963                 if (rearrange_stdio(s[1], s[1], STDERR_FILENO) < 0)
 964                         _exit(EXIT_FAILURE);
 965
 966                 if (b->exec_argv)
 967                         execvp(b->exec_path, b->exec_argv);
 968                 else {
 969                         const char *argv[] = { b->exec_path, NULL };
 970                         execvp(b->exec_path, (char**) argv);
 971                 }
 972
 973                 _exit(EXIT_FAILURE);
 974         }
 975
 976         safe_close(s[1]);
 977         b->output_fd = b->input_fd = fd_move_above_stdio(s[0]);
 978
 979         bus_socket_setup(b);
 980
 981         return bus_socket_start_auth(b);
 982 }
 983
 984 int bus_socket_take_fd(sd_bus *b) {
 985         assert(b);
 986
 987         bus_socket_setup(b);
 988
 989         return bus_socket_start_auth(b);
 990 }
 991
 992 int bus_socket_write_message(sd_bus *bus, sd_bus_message *m, size_t *idx) {
 993         struct iovec *iov;
 994         ssize_t k;
 995         size_t n;
 996         unsigned j;
 997         int r;
 998
 999         assert(bus);
1000         assert(m);
1001         assert(idx);
1002         assert(IN_SET(bus->state, BUS_RUNNING, BUS_HELLO));
1003
1004         if (*idx >= BUS_MESSAGE_SIZE(m))
1005                 return 0;
1006
1007         r = bus_message_setup_iovec(m);
1008         if (r < 0)
1009                 return r;
1010
1011         n = m->n_iovec * sizeof(struct iovec);
1012         iov = alloca(n);
1013         memcpy_safe(iov, m->iovec, n);
1014
1015         j = 0;
1016         iovec_advance(iov, &j, *idx);
1017
1018         if (bus->prefer_writev)
1019                 k = writev(bus->output_fd, iov, m->n_iovec);
1020         else {
1021                 struct msghdr mh = {
1022                         .msg_iov = iov,
1023                         .msg_iovlen = m->n_iovec,
1024                 };
1025
1026                 if (m->n_fds > 0 && *idx == 0) {
1027                         struct cmsghdr *control;
1028
1029                         mh.msg_control = control = alloca(CMSG_SPACE(sizeof(int) * m->n_fds));
1030                         mh.msg_controllen = control->cmsg_len = CMSG_LEN(sizeof(int) * m->n_fds);
1031                         control->cmsg_level = SOL_SOCKET;
1032                         control->cmsg_type = SCM_RIGHTS;
1033                         memcpy(CMSG_DATA(control), m->fds, sizeof(int) * m->n_fds);
1034                 }
1035
1036                 k = sendmsg(bus->output_fd, &mh, MSG_DONTWAIT|MSG_NOSIGNAL);
1037                 if (k < 0 && errno == ENOTSOCK) {
1038                         bus->prefer_writev = true;
1039                         k = writev(bus->output_fd, iov, m->n_iovec);
1040                 }
1041         }
1042
1043         if (k < 0)
1044                 return errno == EAGAIN ? 0 : -errno;
1045
1046         *idx += (size_t) k;
1047         return 1;
1048 }
1049
1050 static int bus_socket_read_message_need(sd_bus *bus, size_t *need) {
1051         uint32_t a, b;
1052         uint8_t e;
1053         uint64_t sum;
1054
1055         assert(bus);
1056         assert(need);
1057         assert(IN_SET(bus->state, BUS_RUNNING, BUS_HELLO));
1058
1059         if (bus->rbuffer_size < sizeof(struct bus_header)) {
1060                 *need = sizeof(struct bus_header) + 8;
1061
1062                 /* Minimum message size:
1063                  *
1064                  * Header +
1065                  *
1066                  *  Method Call: +2 string headers
1067                  *       Signal: +3 string headers
1068                  * Method Error: +1 string headers
1069                  *               +1 uint32 headers
1070                  * Method Reply: +1 uint32 headers
1071                  *
1072                  * A string header is at least 9 bytes
1073                  * A uint32 header is at least 8 bytes
1074                  *
1075                  * Hence the minimum message size of a valid message
1076                  * is header + 8 bytes */
1077
1078                 return 0;
1079         }
1080
1081         a = ((const uint32_t*) bus->rbuffer)[1];
1082         b = ((const uint32_t*) bus->rbuffer)[3];
1083
1084         e = ((const uint8_t*) bus->rbuffer)[0];
1085         if (e == BUS_LITTLE_ENDIAN) {
1086                 a = le32toh(a);
1087                 b = le32toh(b);
1088         } else if (e == BUS_BIG_ENDIAN) {
1089                 a = be32toh(a);
1090                 b = be32toh(b);
1091         } else
1092                 return -EBADMSG;
1093
1094         sum = (uint64_t) sizeof(struct bus_header) + (uint64_t) ALIGN_TO(b, 8) + (uint64_t) a;
1095         if (sum >= BUS_MESSAGE_SIZE_MAX)
1096                 return -ENOBUFS;
1097
1098         *need = (size_t) sum;
1099         return 0;
1100 }
1101
1102 static int bus_socket_make_message(sd_bus *bus, size_t size) {
1103         sd_bus_message *t;
1104         void *b;
1105         int r;
1106
1107         assert(bus);
1108         assert(bus->rbuffer_size >= size);
1109         assert(IN_SET(bus->state, BUS_RUNNING, BUS_HELLO));
1110
1111         r = bus_rqueue_make_room(bus);
1112         if (r < 0)
1113                 return r;
1114
1115         if (bus->rbuffer_size > size) {
1116                 b = memdup((const uint8_t*) bus->rbuffer + size,
1117                            bus->rbuffer_size - size);
1118                 if (!b)
1119                         return -ENOMEM;
1120         } else
1121                 b = NULL;
1122
1123         r = bus_message_from_malloc(bus,
1124                                     bus->rbuffer, size,
1125                                     bus->fds, bus->n_fds,
1126                                     NULL,
1127                                     &t);
1128         if (r < 0) {
1129                 free(b);
1130                 return r;
1131         }
1132
1133         bus->rbuffer = b;
1134         bus->rbuffer_size -= size;
1135
1136         bus->fds = NULL;
1137         bus->n_fds = 0;
1138
1139         bus->rqueue[bus->rqueue_size++] = t;
1140
1141         return 1;
1142 }
1143
1144 int bus_socket_read_message(sd_bus *bus) {
1145         struct msghdr mh;
1146         struct iovec iov = {};
1147         ssize_t k;
1148         size_t need;
1149         int r;
1150         void *b;
1151         union {
1152                 struct cmsghdr cmsghdr;
1153                 uint8_t buf[CMSG_SPACE(sizeof(int) * BUS_FDS_MAX)];
1154         } control;
1155         bool handle_cmsg = false;
1156
1157         assert(bus);
1158         assert(IN_SET(bus->state, BUS_RUNNING, BUS_HELLO));
1159
1160         r = bus_socket_read_message_need(bus, &need);
1161         if (r < 0)
1162                 return r;
1163
1164         if (bus->rbuffer_size >= need)
1165                 return bus_socket_make_message(bus, need);
1166
1167         b = realloc(bus->rbuffer, need);
1168         if (!b)
1169                 return -ENOMEM;
1170
1171         bus->rbuffer = b;
1172
1173         iov.iov_base = (uint8_t*) bus->rbuffer + bus->rbuffer_size;
1174         iov.iov_len = need - bus->rbuffer_size;
1175
1176         if (bus->prefer_readv)
1177                 k = readv(bus->input_fd, &iov, 1);
1178         else {
1179                 zero(mh);
1180                 mh.msg_iov = &iov;
1181                 mh.msg_iovlen = 1;
1182                 mh.msg_control = &control;
1183                 mh.msg_controllen = sizeof(control);
1184
1185                 k = recvmsg(bus->input_fd, &mh, MSG_DONTWAIT|MSG_NOSIGNAL|MSG_CMSG_CLOEXEC);
1186                 if (k < 0 && errno == ENOTSOCK) {
1187                         bus->prefer_readv = true;
1188                         k = readv(bus->input_fd, &iov, 1);
1189                 } else
1190                         handle_cmsg = true;
1191         }
1192         if (k < 0)
1193                 return errno == EAGAIN ? 0 : -errno;
1194         if (k == 0)
1195                 return -ECONNRESET;
1196
1197         bus->rbuffer_size += k;
1198
1199         if (handle_cmsg) {
1200                 struct cmsghdr *cmsg;
1201
1202                 CMSG_FOREACH(cmsg, &mh)
1203                         if (cmsg->cmsg_level == SOL_SOCKET &&
1204                             cmsg->cmsg_type == SCM_RIGHTS) {
1205                                 int n, *f, i;
1206
1207                                 n = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
1208
1209                                 if (!bus->can_fds) {
1210                                         /* Whut? We received fds but this
1211                                          * isn't actually enabled? Close them,
1212                                          * and fail */
1213
1214                                         close_many((int*) CMSG_DATA(cmsg), n);
1215                                         return -EIO;
1216                                 }
1217
1218                                 f = reallocarray(bus->fds, bus->n_fds + n, sizeof(int));
1219                                 if (!f) {
1220                                         close_many((int*) CMSG_DATA(cmsg), n);
1221                                         return -ENOMEM;
1222                                 }
1223
1224                                 for (i = 0; i < n; i++)
1225                                         f[bus->n_fds++] = fd_move_above_stdio(((int*) CMSG_DATA(cmsg))[i]);
1226                                 bus->fds = f;
1227                         } else
1228                                 log_debug("Got unexpected auxiliary data with level=%d and type=%d",
1229                                           cmsg->cmsg_level, cmsg->cmsg_type);
1230         }
1231
1232         r = bus_socket_read_message_need(bus, &need);
1233         if (r < 0)
1234                 return r;
1235
1236         if (bus->rbuffer_size >= need)
1237                 return bus_socket_make_message(bus, need);
1238
1239         return 1;
1240 }
1241
1242 int bus_socket_process_opening(sd_bus *b) {
1243         int error = 0;
1244         socklen_t slen = sizeof(error);
1245         struct pollfd p = {
1246                 .fd = b->output_fd,
1247                 .events = POLLOUT,
1248         };
1249         int r;
1250
1251         assert(b->state == BUS_OPENING);
1252
1253         r = poll(&p, 1, 0);
1254         if (r < 0)
1255                 return -errno;
1256
1257         if (!(p.revents & (POLLOUT|POLLERR|POLLHUP)))
1258                 return 0;
1259
1260         r = getsockopt(b->output_fd, SOL_SOCKET, SO_ERROR, &error, &slen);
1261         if (r < 0)
1262                 b->last_connect_error = errno;
1263         else if (error != 0)
1264                 b->last_connect_error = error;
1265         else if (p.revents & (POLLERR|POLLHUP))
1266                 b->last_connect_error = ECONNREFUSED;
1267         else
1268                 return bus_socket_start_auth(b);
1269
1270         return bus_next_address(b);
1271 }
1272
1273 int bus_socket_process_authenticating(sd_bus *b) {
1274         int r;
1275
1276         assert(b);
1277         assert(b->state == BUS_AUTHENTICATING);
1278
1279         if (now(CLOCK_MONOTONIC) >= b->auth_timeout)
1280                 return -ETIMEDOUT;
1281
1282         r = bus_socket_write_auth(b);
1283         if (r != 0)
1284                 return r;
1285
1286         return bus_socket_read_auth(b);
1287 }
1288
1289 int bus_socket_process_watch_bind(sd_bus *b) {
1290         int r, q;
1291
1292         assert(b);
1293         assert(b->state == BUS_WATCH_BIND);
1294         assert(b->inotify_fd >= 0);
1295
1296         r = flush_fd(b->inotify_fd);
1297         if (r <= 0)
1298                 return r;
1299
1300         log_debug("Got inotify event on bus %s.", strna(b->description));
1301
1302         /* We flushed events out of the inotify fd. In that case, maybe the socket is valid now? Let's try to connect
1303          * to it again */
1304
1305         r = bus_socket_connect(b);
1306         if (r < 0)
1307                 return r;
1308
1309         q = bus_attach_io_events(b);
1310         if (q < 0)
1311                 return q;
1312
1313         q = bus_attach_inotify_event(b);
1314         if (q < 0)
1315                 return q;
1316
1317         return r;
1318 }