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

   1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
   2
   3 /***
   4   This file is part of systemd.
   5
   6   Copyright 2013 Lennart Poettering
   7
   8   systemd is free software; you can redistribute it and/or modify it
   9   under the terms of the GNU Lesser General Public License as published by
  10   the Free Software Foundation; either version 2.1 of the License, or
  11   (at your option) any later version.
  12
  13   systemd is distributed in the hope that it will be useful, but
  14   WITHOUT ANY WARRANTY; without even the implied warranty of
  15   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  16   Lesser General Public License for more details.
  17
  18   You should have received a copy of the GNU Lesser General Public License
  19   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  20 ***/
  21
  22 #include <fcntl.h>
  23 #include <unistd.h>
  24
  25 #include "bus-container.h"
  26 #include "bus-internal.h"
  27 #include "bus-socket.h"
  28 #include "fd-util.h"
  29 #include "process-util.h"
  30 #include "util.h"
  31
  32 int bus_container_connect_socket(sd_bus *b) {
  33         _cleanup_close_pair_ int pair[2] = { -1, -1 };
  34         _cleanup_close_ int pidnsfd = -1, mntnsfd = -1, usernsfd = -1, rootfd = -1;
  35         pid_t child;
  36         siginfo_t si;
  37         int r, error_buf = 0;
  38         ssize_t n;
  39
  40         assert(b);
  41         assert(b->input_fd < 0);
  42         assert(b->output_fd < 0);
  43         assert(b->nspid > 0 || b->machine);
  44
  45         if (b->nspid <= 0) {
  46                 r = container_get_leader(b->machine, &b->nspid);
  47                 if (r < 0)
  48                         return r;
  49         }
  50
  51         r = namespace_open(b->nspid, &pidnsfd, &mntnsfd, NULL, &usernsfd, &rootfd);
  52         if (r < 0)
  53                 return r;
  54
  55         b->input_fd = socket(b->sockaddr.sa.sa_family, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
  56         if (b->input_fd < 0)
  57                 return -errno;
  58
  59         b->output_fd = b->input_fd;
  60
  61         bus_socket_setup(b);
  62
  63         if (socketpair(AF_UNIX, SOCK_SEQPACKET, 0, pair) < 0)
  64                 return -errno;
  65
  66         child = fork();
  67         if (child < 0)
  68                 return -errno;
  69
  70         if (child == 0) {
  71                 pid_t grandchild;
  72
  73                 pair[0] = safe_close(pair[0]);
  74
  75                 r = namespace_enter(pidnsfd, mntnsfd, -1, usernsfd, rootfd);
  76                 if (r < 0)
  77                         _exit(EXIT_FAILURE);
  78
  79                 /* We just changed PID namespace, however it will only
  80                  * take effect on the children we now fork. Hence,
  81                  * let's fork another time, and connect from this
  82                  * grandchild, so that SO_PEERCRED of our connection
  83                  * comes from a process from within the container, and
  84                  * not outside of it */
  85
  86                 grandchild = fork();
  87                 if (grandchild < 0)
  88                         _exit(EXIT_FAILURE);
  89
  90                 if (grandchild == 0) {
  91
  92                         r = connect(b->input_fd, &b->sockaddr.sa, b->sockaddr_size);
  93                         if (r < 0) {
  94                                 /* Try to send error up */
  95                                 error_buf = errno;
  96                                 (void) write(pair[1], &error_buf, sizeof(error_buf));
  97                                 _exit(EXIT_FAILURE);
  98                         }
  99
 100                         _exit(EXIT_SUCCESS);
 101                 }
 102
 103                 r = wait_for_terminate(grandchild, &si);
 104                 if (r < 0)
 105                         _exit(EXIT_FAILURE);
 106
 107                 if (si.si_code != CLD_EXITED)
 108                         _exit(EXIT_FAILURE);
 109
 110                 _exit(si.si_status);
 111         }
 112
 113         pair[1] = safe_close(pair[1]);
 114
 115         r = wait_for_terminate(child, &si);
 116         if (r < 0)
 117                 return r;
 118
 119         n = read(pair[0], &error_buf, sizeof(error_buf));
 120         if (n < 0)
 121                 return -errno;
 122
 123         if (n > 0) {
 124                 if (n != sizeof(error_buf))
 125                         return -EIO;
 126
 127                 if (error_buf < 0)
 128                         return -EIO;
 129
 130                 if (error_buf == EINPROGRESS)
 131                         return 1;
 132
 133                 if (error_buf > 0)
 134                         return -error_buf;
 135         }
 136
 137         if (si.si_code != CLD_EXITED)
 138                 return -EIO;
 139
 140         if (si.si_status != EXIT_SUCCESS)
 141                 return -EIO;
 142
 143         return bus_socket_start_auth(b);
 144 }
 145
 146 int bus_container_connect_kernel(sd_bus *b) {
 147         _cleanup_close_pair_ int pair[2] = { -1, -1 };
 148         _cleanup_close_ int pidnsfd = -1, mntnsfd = -1, usernsfd = -1, rootfd = -1;
 149         union {
 150                 struct cmsghdr cmsghdr;
 151                 uint8_t buf[CMSG_SPACE(sizeof(int))];
 152         } control = {};
 153         int error_buf = 0;
 154         struct iovec iov = {
 155                 .iov_base = &error_buf,
 156                 .iov_len = sizeof(error_buf),
 157         };
 158         struct msghdr mh = {
 159                 .msg_control = &control,
 160                 .msg_controllen = sizeof(control),
 161                 .msg_iov = &iov,
 162                 .msg_iovlen = 1,
 163         };
 164         struct cmsghdr *cmsg;
 165         pid_t child;
 166         siginfo_t si;
 167         int r, fd = -1;
 168         ssize_t n;
 169
 170         assert(b);
 171         assert(b->input_fd < 0);
 172         assert(b->output_fd < 0);
 173         assert(b->nspid > 0 || b->machine);
 174
 175         if (b->nspid <= 0) {
 176                 r = container_get_leader(b->machine, &b->nspid);
 177                 if (r < 0)
 178                         return r;
 179         }
 180
 181         r = namespace_open(b->nspid, &pidnsfd, &mntnsfd, NULL, &usernsfd, &rootfd);
 182         if (r < 0)
 183                 return r;
 184
 185         if (socketpair(AF_UNIX, SOCK_SEQPACKET, 0, pair) < 0)
 186                 return -errno;
 187
 188         child = fork();
 189         if (child < 0)
 190                 return -errno;
 191
 192         if (child == 0) {
 193                 pid_t grandchild;
 194
 195                 pair[0] = safe_close(pair[0]);
 196
 197                 r = namespace_enter(pidnsfd, mntnsfd, -1, usernsfd, rootfd);
 198                 if (r < 0)
 199                         _exit(EXIT_FAILURE);
 200
 201                 /* We just changed PID namespace, however it will only
 202                  * take effect on the children we now fork. Hence,
 203                  * let's fork another time, and connect from this
 204                  * grandchild, so that kdbus only sees the credentials
 205                  * of this process which comes from within the
 206                  * container, and not outside of it */
 207
 208                 grandchild = fork();
 209                 if (grandchild < 0)
 210                         _exit(EXIT_FAILURE);
 211
 212                 if (grandchild == 0) {
 213                         fd = open(b->kernel, O_RDWR|O_NOCTTY|O_CLOEXEC);
 214                         if (fd < 0) {
 215                                 /* Try to send error up */
 216                                 error_buf = errno;
 217                                 (void) write(pair[1], &error_buf, sizeof(error_buf));
 218                                 _exit(EXIT_FAILURE);
 219                         }
 220
 221                         r = send_one_fd(pair[1], fd, 0);
 222                         if (r < 0)
 223                                 _exit(EXIT_FAILURE);
 224
 225                         _exit(EXIT_SUCCESS);
 226                 }
 227
 228                 r = wait_for_terminate(grandchild, &si);
 229                 if (r < 0)
 230                         _exit(EXIT_FAILURE);
 231
 232                 if (si.si_code != CLD_EXITED)
 233                         _exit(EXIT_FAILURE);
 234
 235                 _exit(si.si_status);
 236         }
 237
 238         pair[1] = safe_close(pair[1]);
 239
 240         r = wait_for_terminate(child, &si);
 241         if (r < 0)
 242                 return r;
 243
 244         n = recvmsg(pair[0], &mh, MSG_NOSIGNAL|MSG_CMSG_CLOEXEC);
 245         if (n < 0)
 246                 return -errno;
 247
 248         CMSG_FOREACH(cmsg, &mh) {
 249                 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
 250                         int *fds;
 251                         unsigned n_fds;
 252
 253                         assert(fd < 0);
 254
 255                         fds = (int*) CMSG_DATA(cmsg);
 256                         n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
 257
 258                         if (n_fds != 1) {
 259                                 close_many(fds, n_fds);
 260                                 return -EIO;
 261                         }
 262
 263                         fd = fds[0];
 264                 }
 265         }
 266
 267         /* If there's an fd passed, we are good. */
 268         if (fd >= 0) {
 269                 b->input_fd = b->output_fd = fd;
 270                 return bus_kernel_take_fd(b);
 271         }
 272
 273         /* If there's an error passed, use it */
 274         if (n == sizeof(error_buf) && error_buf > 0)
 275                 return -error_buf;
 276
 277         /* Otherwise, we have no clue */
 278         return -EIO;
 279 }