src/core/namespace.c

   1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
   2
   3 /***
   4   This file is part of systemd.
   5
   6   Copyright 2010 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 <errno.h>
  23 #include <sys/mount.h>
  24 #include <string.h>
  25 #include <stdio.h>
  26 #include <unistd.h>
  27 #include <sys/stat.h>
  28 #include <sys/types.h>
  29 #include <sched.h>
  30 #include <sys/syscall.h>
  31 #include <limits.h>
  32 #include <linux/fs.h>
  33 #include <sys/file.h>
  34
  35 #include "strv.h"
  36 #include "util.h"
  37 #include "path-util.h"
  38 #include "namespace.h"
  39 #include "missing.h"
  40 #include "execute.h"
  41 #include "loopback-setup.h"
  42
  43 typedef enum MountMode {
  44         /* This is ordered by priority! */
  45         INACCESSIBLE,
  46         READONLY,
  47         PRIVATE_TMP,
  48         PRIVATE_VAR_TMP,
  49         READWRITE
  50 } MountMode;
  51
  52 typedef struct BindMount {
  53         const char *path;
  54         MountMode mode;
  55         bool done;
  56         bool ignore;
  57 } BindMount;
  58
  59 static int append_mounts(BindMount **p, char **strv, MountMode mode) {
  60         char **i;
  61
  62         assert(p);
  63
  64         STRV_FOREACH(i, strv) {
  65
  66                 (*p)->ignore = false;
  67
  68                 if ((mode == INACCESSIBLE || mode == READONLY) && (*i)[0] == '-') {
  69                         (*p)->ignore = true;
  70                         (*i)++;
  71                 }
  72
  73                 if (!path_is_absolute(*i))
  74                         return -EINVAL;
  75
  76                 (*p)->path = *i;
  77                 (*p)->mode = mode;
  78                 (*p)++;
  79         }
  80
  81         return 0;
  82 }
  83
  84 static int mount_path_compare(const void *a, const void *b) {
  85         const BindMount *p = a, *q = b;
  86
  87         if (path_equal(p->path, q->path)) {
  88
  89                 /* If the paths are equal, check the mode */
  90                 if (p->mode < q->mode)
  91                         return -1;
  92
  93                 if (p->mode > q->mode)
  94                         return 1;
  95
  96                 return 0;
  97         }
  98
  99         /* If the paths are not equal, then order prefixes first */
 100         if (path_startswith(p->path, q->path))
 101                 return 1;
 102
 103         if (path_startswith(q->path, p->path))
 104                 return -1;
 105
 106         return 0;
 107 }
 108
 109 static void drop_duplicates(BindMount *m, unsigned *n) {
 110         BindMount *f, *t, *previous;
 111
 112         assert(m);
 113         assert(n);
 114
 115         for (f = m, t = m, previous = NULL; f < m+*n; f++) {
 116
 117                 /* The first one wins */
 118                 if (previous && path_equal(f->path, previous->path))
 119                         continue;
 120
 121                 t->path = f->path;
 122                 t->mode = f->mode;
 123
 124                 previous = t;
 125
 126                 t++;
 127         }
 128
 129         *n = t - m;
 130 }
 131
 132 static int apply_mount(
 133                 BindMount *m,
 134                 const char *tmp_dir,
 135                 const char *var_tmp_dir) {
 136
 137         const char *what;
 138         int r;
 139
 140         assert(m);
 141
 142         switch (m->mode) {
 143
 144         case INACCESSIBLE:
 145                 what = "/run/systemd/inaccessible";
 146                 break;
 147
 148         case READONLY:
 149         case READWRITE:
 150                 what = m->path;
 151                 break;
 152
 153         case PRIVATE_TMP:
 154                 what = tmp_dir;
 155                 break;
 156
 157         case PRIVATE_VAR_TMP:
 158                 what = var_tmp_dir;
 159                 break;
 160
 161         default:
 162                 assert_not_reached("Unknown mode");
 163         }
 164
 165         assert(what);
 166
 167         r = mount(what, m->path, NULL, MS_BIND|MS_REC, NULL);
 168         if (r >= 0)
 169                 log_debug("Successfully mounted %s to %s", what, m->path);
 170         else if (m->ignore && errno == ENOENT)
 171                 r = 0;
 172
 173         return r;
 174 }
 175
 176 static int make_read_only(BindMount *m) {
 177         int r;
 178
 179         assert(m);
 180
 181         if (m->mode != INACCESSIBLE && m->mode != READONLY)
 182                 return 0;
 183
 184         r = mount(NULL, m->path, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_REC, NULL);
 185         if (r < 0 && !(m->ignore && errno == ENOENT))
 186                 return -errno;
 187
 188         return 0;
 189 }
 190
 191 int setup_namespace(
 192                 char** read_write_dirs,
 193                 char** read_only_dirs,
 194                 char** inaccessible_dirs,
 195                 char* tmp_dir,
 196                 char* var_tmp_dir,
 197                 unsigned mount_flags) {
 198
 199         BindMount *m, *mounts = NULL;
 200         unsigned n;
 201         int r = 0;
 202
 203         if (mount_flags == 0)
 204                 mount_flags = MS_SHARED;
 205
 206         if (unshare(CLONE_NEWNS) < 0)
 207                 return -errno;
 208
 209         n = !!tmp_dir + !!var_tmp_dir +
 210                 strv_length(read_write_dirs) +
 211                 strv_length(read_only_dirs) +
 212                 strv_length(inaccessible_dirs);
 213
 214         if (n > 0) {
 215                 m = mounts = (BindMount *) alloca(n * sizeof(BindMount));
 216                 r = append_mounts(&m, read_write_dirs, READWRITE);
 217                 if (r < 0)
 218                         return r;
 219
 220                 r = append_mounts(&m, read_only_dirs, READONLY);
 221                 if (r < 0)
 222                         return r;
 223
 224                 r = append_mounts(&m, inaccessible_dirs, INACCESSIBLE);
 225                 if (r < 0)
 226                         return r;
 227
 228                 if (tmp_dir) {
 229                         m->path = "/tmp";
 230                         m->mode = PRIVATE_TMP;
 231                         m++;
 232                 }
 233
 234                 if (var_tmp_dir) {
 235                         m->path = "/var/tmp";
 236                         m->mode = PRIVATE_VAR_TMP;
 237                         m++;
 238                 }
 239
 240                 assert(mounts + n == m);
 241
 242                 qsort(mounts, n, sizeof(BindMount), mount_path_compare);
 243                 drop_duplicates(mounts, &n);
 244         }
 245
 246         /* Remount / as SLAVE so that nothing now mounted in the namespace
 247            shows up in the parent */
 248         if (mount(NULL, "/", NULL, MS_SLAVE|MS_REC, NULL) < 0)
 249                 return -errno;
 250
 251         for (m = mounts; m < mounts + n; ++m) {
 252                 r = apply_mount(m, tmp_dir, var_tmp_dir);
 253                 if (r < 0)
 254                         goto fail;
 255         }
 256
 257         for (m = mounts; m < mounts + n; ++m) {
 258                 r = make_read_only(m);
 259                 if (r < 0)
 260                         goto fail;
 261         }
 262
 263         /* Remount / as the desired mode */
 264         if (mount(NULL, "/", NULL, mount_flags | MS_REC, NULL) < 0) {
 265                 r = -errno;
 266                 goto fail;
 267         }
 268
 269         return 0;
 270
 271 fail:
 272         for (m = mounts; m < mounts + n; ++m)
 273                 if (m->done)
 274                         umount2(m->path, MNT_DETACH);
 275
 276         return r;
 277 }
 278
 279 static int setup_one_tmp_dir(const char *id, const char *prefix, char **path) {
 280         _cleanup_free_ char *x = NULL;
 281
 282         assert(id);
 283         assert(prefix);
 284         assert(path);
 285
 286         x = strjoin(prefix, "/systemd-", id, "-XXXXXX", NULL);
 287         if (!x)
 288                 return -ENOMEM;
 289
 290         RUN_WITH_UMASK(0077)
 291                 if (!mkdtemp(x))
 292                         return -errno;
 293
 294         RUN_WITH_UMASK(0000) {
 295                 char *y;
 296
 297                 y = strappenda(x, "/tmp");
 298
 299                 if (mkdir(y, 0777 | S_ISVTX) < 0)
 300                         return -errno;
 301         }
 302
 303         *path = x;
 304         x = NULL;
 305
 306         return 0;
 307 }
 308
 309 int setup_tmp_dirs(const char *id, char **tmp_dir, char **var_tmp_dir) {
 310         char *a, *b;
 311         int r;
 312
 313         assert(id);
 314         assert(tmp_dir);
 315         assert(var_tmp_dir);
 316
 317         r = setup_one_tmp_dir(id, "/tmp", &a);
 318         if (r < 0)
 319                 return r;
 320
 321         r = setup_one_tmp_dir(id, "/var/tmp", &b);
 322         if (r < 0) {
 323                 char *t;
 324
 325                 t = strappenda(a, "/tmp");
 326                 rmdir(t);
 327                 rmdir(a);
 328
 329                 free(a);
 330                 return r;
 331         }
 332
 333         *tmp_dir = a;
 334         *var_tmp_dir = b;
 335
 336         return 0;
 337 }
 338
 339 int setup_netns(int netns_storage_socket[2]) {
 340         _cleanup_close_ int netns = -1;
 341         union {
 342                 struct cmsghdr cmsghdr;
 343                 uint8_t buf[CMSG_SPACE(sizeof(int))];
 344         } control = {};
 345         struct msghdr mh = {
 346                 .msg_control = &control,
 347                 .msg_controllen = sizeof(control),
 348         };
 349         struct cmsghdr *cmsg;
 350         int r;
 351
 352         assert(netns_storage_socket);
 353         assert(netns_storage_socket[0] >= 0);
 354         assert(netns_storage_socket[1] >= 0);
 355
 356         /* We use the passed socketpair as a storage buffer for our
 357          * namespace socket. Whatever process runs this first shall
 358          * create a new namespace, all others should just join it. To
 359          * serialize that we use a file lock on the socket pair.
 360          *
 361          * It's a bit crazy, but hey, works great! */
 362
 363         if (lockf(netns_storage_socket[0], F_LOCK, 0) < 0)
 364                 return -errno;
 365
 366         if (recvmsg(netns_storage_socket[0], &mh, MSG_DONTWAIT|MSG_CMSG_CLOEXEC) < 0) {
 367                 if (errno != EAGAIN) {
 368                         r = -errno;
 369                         goto fail;
 370                 }
 371
 372                 /* Nothing stored yet, so let's create a new namespace */
 373
 374                 if (unshare(CLONE_NEWNET) < 0) {
 375                         r = -errno;
 376                         goto fail;
 377                 }
 378
 379                 loopback_setup();
 380
 381                 netns = open("/proc/self/ns/net", O_RDONLY|O_CLOEXEC|O_NOCTTY);
 382                 if (netns < 0) {
 383                         r = -errno;
 384                         goto fail;
 385                 }
 386
 387                 r = 1;
 388         } else {
 389                 /* Yay, found something, so let's join the namespace */
 390
 391                 for (cmsg = CMSG_FIRSTHDR(&mh); cmsg; cmsg = CMSG_NXTHDR(&mh, cmsg)) {
 392                         if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
 393                                 assert(cmsg->cmsg_len == CMSG_LEN(sizeof(int)));
 394                                 netns = *(int*) CMSG_DATA(cmsg);
 395                         }
 396                 }
 397
 398                 if (setns(netns, CLONE_NEWNET) < 0) {
 399                         r = -errno;
 400                         goto fail;
 401                 }
 402
 403                 r = 0;
 404         }
 405
 406         cmsg = CMSG_FIRSTHDR(&mh);
 407         cmsg->cmsg_level = SOL_SOCKET;
 408         cmsg->cmsg_type = SCM_RIGHTS;
 409         cmsg->cmsg_len = CMSG_LEN(sizeof(int));
 410         memcpy(CMSG_DATA(cmsg), &netns, sizeof(int));
 411         mh.msg_controllen = cmsg->cmsg_len;
 412
 413         if (sendmsg(netns_storage_socket[1], &mh, MSG_DONTWAIT|MSG_NOSIGNAL) < 0) {
 414                 r = -errno;
 415                 goto fail;
 416         }
 417
 418 fail:
 419         lockf(netns_storage_socket[0], F_ULOCK, 0);
 420
 421         return r;
 422 }