src/nspawn/nspawn-mount.c

   1 /***
   2   This file is part of systemd.
   3
   4   Copyright 2015 Lennart Poettering
   5
   6   systemd is free software; you can redistribute it and/or modify it
   7   under the terms of the GNU Lesser General Public License as published by
   8   the Free Software Foundation; either version 2.1 of the License, or
   9   (at your option) any later version.
  10
  11   systemd is distributed in the hope that it will be useful, but
  12   WITHOUT ANY WARRANTY; without even the implied warranty of
  13   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  14   Lesser General Public License for more details.
  15
  16   You should have received a copy of the GNU Lesser General Public License
  17   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  18 ***/
  19
  20 #include <sys/mount.h>
  21 #include <linux/magic.h>
  22
  23 #include "alloc-util.h"
  24 #include "escape.h"
  25 #include "fd-util.h"
  26 #include "fileio.h"
  27 #include "fs-util.h"
  28 #include "label.h"
  29 #include "mkdir.h"
  30 #include "mount-util.h"
  31 #include "nspawn-mount.h"
  32 #include "parse-util.h"
  33 #include "path-util.h"
  34 #include "rm-rf.h"
  35 #include "set.h"
  36 #include "stat-util.h"
  37 #include "string-util.h"
  38 #include "strv.h"
  39 #include "user-util.h"
  40 #include "util.h"
  41
  42 CustomMount* custom_mount_add(CustomMount **l, unsigned *n, CustomMountType t) {
  43         CustomMount *c, *ret;
  44
  45         assert(l);
  46         assert(n);
  47         assert(t >= 0);
  48         assert(t < _CUSTOM_MOUNT_TYPE_MAX);
  49
  50         c = realloc_multiply(*l, (*n + 1), sizeof(CustomMount));
  51         if (!c)
  52                 return NULL;
  53
  54         *l = c;
  55         ret = *l + *n;
  56         (*n)++;
  57
  58         *ret = (CustomMount) { .type = t };
  59
  60         return ret;
  61 }
  62
  63 void custom_mount_free_all(CustomMount *l, unsigned n) {
  64         unsigned i;
  65
  66         for (i = 0; i < n; i++) {
  67                 CustomMount *m = l + i;
  68
  69                 free(m->source);
  70                 free(m->destination);
  71                 free(m->options);
  72
  73                 if (m->work_dir) {
  74                         (void) rm_rf(m->work_dir, REMOVE_ROOT|REMOVE_PHYSICAL);
  75                         free(m->work_dir);
  76                 }
  77
  78                 strv_free(m->lower);
  79         }
  80
  81         free(l);
  82 }
  83
  84 int custom_mount_compare(const void *a, const void *b) {
  85         const CustomMount *x = a, *y = b;
  86         int r;
  87
  88         r = path_compare(x->destination, y->destination);
  89         if (r != 0)
  90                 return r;
  91
  92         if (x->type < y->type)
  93                 return -1;
  94         if (x->type > y->type)
  95                 return 1;
  96
  97         return 0;
  98 }
  99
 100 int bind_mount_parse(CustomMount **l, unsigned *n, const char *s, bool read_only) {
 101         _cleanup_free_ char *source = NULL, *destination = NULL, *opts = NULL;
 102         const char *p = s;
 103         CustomMount *m;
 104         int r;
 105
 106         assert(l);
 107         assert(n);
 108
 109         r = extract_many_words(&p, ":", EXTRACT_DONT_COALESCE_SEPARATORS, &source, &destination, NULL);
 110         if (r < 0)
 111                 return r;
 112         if (r == 0)
 113                 return -EINVAL;
 114
 115         if (r == 1) {
 116                 destination = strdup(source);
 117                 if (!destination)
 118                         return -ENOMEM;
 119         }
 120
 121         if (r == 2 && !isempty(p)) {
 122                 opts = strdup(p);
 123                 if (!opts)
 124                         return -ENOMEM;
 125         }
 126
 127         if (!path_is_absolute(source))
 128                 return -EINVAL;
 129
 130         if (!path_is_absolute(destination))
 131                 return -EINVAL;
 132
 133         m = custom_mount_add(l, n, CUSTOM_MOUNT_BIND);
 134         if (!m)
 135                 return -ENOMEM;
 136
 137         m->source = source;
 138         m->destination = destination;
 139         m->read_only = read_only;
 140         m->options = opts;
 141
 142         source = destination = opts = NULL;
 143         return 0;
 144 }
 145
 146 int tmpfs_mount_parse(CustomMount **l, unsigned *n, const char *s) {
 147         _cleanup_free_ char *path = NULL, *opts = NULL;
 148         const char *p = s;
 149         CustomMount *m;
 150         int r;
 151
 152         assert(l);
 153         assert(n);
 154         assert(s);
 155
 156         r = extract_first_word(&p, &path, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
 157         if (r < 0)
 158                 return r;
 159         if (r == 0)
 160                 return -EINVAL;
 161
 162         if (isempty(p))
 163                 opts = strdup("mode=0755");
 164         else
 165                 opts = strdup(p);
 166         if (!opts)
 167                 return -ENOMEM;
 168
 169         if (!path_is_absolute(path))
 170                 return -EINVAL;
 171
 172         m = custom_mount_add(l, n, CUSTOM_MOUNT_TMPFS);
 173         if (!m)
 174                 return -ENOMEM;
 175
 176         m->destination = path;
 177         m->options = opts;
 178
 179         path = opts = NULL;
 180         return 0;
 181 }
 182
 183 int overlay_mount_parse(CustomMount **l, unsigned *n, const char *s, bool read_only) {
 184         _cleanup_free_ char *upper = NULL, *destination = NULL;
 185         _cleanup_strv_free_ char **lower = NULL;
 186         CustomMount *m;
 187         unsigned k = 0;
 188         char **i;
 189         int r;
 190
 191         r = strv_split_extract(&lower, s, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
 192         if (r < 0)
 193                 return r;
 194
 195         STRV_FOREACH(i, lower) {
 196                 if (!path_is_absolute(*i))
 197                         return -EINVAL;
 198
 199                 k++;
 200         }
 201
 202         if (k < 2)
 203                 return -EADDRNOTAVAIL;
 204         if (k == 2) {
 205                 /* If two parameters are specified,
 206                  * the first one is the lower, the
 207                  * second one the upper directory. And
 208                  * we'll also define the destination
 209                  * mount point the same as the upper. */
 210                 upper = lower[1];
 211                 lower[1] = NULL;
 212
 213                 destination = strdup(upper);
 214                 if (!destination)
 215                         return -ENOMEM;
 216
 217         } else {
 218                 upper = lower[k - 2];
 219                 destination = lower[k - 1];
 220                 lower[k - 2] = NULL;
 221         }
 222
 223         m = custom_mount_add(l, n, CUSTOM_MOUNT_OVERLAY);
 224         if (!m)
 225                 return -ENOMEM;
 226
 227         m->destination = destination;
 228         m->source = upper;
 229         m->lower = lower;
 230         m->read_only = read_only;
 231
 232         upper = destination = NULL;
 233         lower = NULL;
 234
 235         return 0;
 236 }
 237
 238 static int tmpfs_patch_options(
 239                 const char *options,
 240                 bool userns,
 241                 uid_t uid_shift, uid_t uid_range,
 242                 bool patch_ids,
 243                 const char *selinux_apifs_context,
 244                 char **ret) {
 245
 246         char *buf = NULL;
 247
 248         if ((userns && uid_shift != 0) || patch_ids) {
 249                 assert(uid_shift != UID_INVALID);
 250
 251                 if (asprintf(&buf, "%s%suid=" UID_FMT ",gid=" UID_FMT,
 252                              options ?: "", options ? "," : "",
 253                              uid_shift, uid_shift) < 0)
 254                         return -ENOMEM;
 255
 256                 options = buf;
 257         }
 258
 259 #ifdef HAVE_SELINUX
 260         if (selinux_apifs_context) {
 261                 char *t;
 262
 263                 t = strjoin(options ?: "", options ? "," : "",
 264                             "context=\"", selinux_apifs_context, "\"");
 265                 free(buf);
 266                 if (!t)
 267                         return -ENOMEM;
 268
 269                 buf = t;
 270         }
 271 #endif
 272
 273         if (!buf && options) {
 274                 buf = strdup(options);
 275                 if (!buf)
 276                         return -ENOMEM;
 277         }
 278         *ret = buf;
 279
 280         return !!buf;
 281 }
 282
 283 int mount_sysfs(const char *dest, MountSettingsMask mount_settings) {
 284         const char *full, *top, *x;
 285         int r;
 286         unsigned long extra_flags = 0;
 287
 288         top = prefix_roota(dest, "/sys");
 289         r = path_check_fstype(top, SYSFS_MAGIC);
 290         if (r < 0)
 291                 return log_error_errno(r, "Failed to determine filesystem type of %s: %m", top);
 292         /* /sys might already be mounted as sysfs by the outer child in the
 293          * !netns case. In this case, it's all good. Don't touch it because we
 294          * don't have the right to do so, see https://github.com/systemd/systemd/issues/1555.
 295          */
 296         if (r > 0)
 297                 return 0;
 298
 299         full = prefix_roota(top, "/full");
 300
 301         (void) mkdir(full, 0755);
 302
 303         if (mount_settings & MOUNT_APPLY_APIVFS_RO)
 304                 extra_flags |= MS_RDONLY;
 305
 306         r = mount_verbose(LOG_ERR, "sysfs", full, "sysfs",
 307                           MS_NOSUID|MS_NOEXEC|MS_NODEV|extra_flags, NULL);
 308         if (r < 0)
 309                 return r;
 310
 311         FOREACH_STRING(x, "block", "bus", "class", "dev", "devices", "kernel") {
 312                 _cleanup_free_ char *from = NULL, *to = NULL;
 313
 314                 from = prefix_root(full, x);
 315                 if (!from)
 316                         return log_oom();
 317
 318                 to = prefix_root(top, x);
 319                 if (!to)
 320                         return log_oom();
 321
 322                 (void) mkdir(to, 0755);
 323
 324                 r = mount_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
 325                 if (r < 0)
 326                         return r;
 327
 328                 r = mount_verbose(LOG_ERR, NULL, to, NULL,
 329                                   MS_BIND|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT|extra_flags, NULL);
 330                 if (r < 0)
 331                         return r;
 332         }
 333
 334         r = umount_verbose(full);
 335         if (r < 0)
 336                 return r;
 337
 338         if (rmdir(full) < 0)
 339                 return log_error_errno(errno, "Failed to remove %s: %m", full);
 340
 341         x = prefix_roota(top, "/fs/kdbus");
 342         (void) mkdir_p(x, 0755);
 343
 344         /* Create mountpoint for cgroups. Otherwise we are not allowed since we
 345          * remount /sys read-only.
 346          */
 347         if (cg_ns_supported()) {
 348                 x = prefix_roota(top, "/fs/cgroup");
 349                 (void) mkdir_p(x, 0755);
 350         }
 351
 352         return mount_verbose(LOG_ERR, NULL, top, NULL,
 353                              MS_BIND|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT|extra_flags, NULL);
 354 }
 355
 356 static int mkdir_userns(const char *path, mode_t mode, MountSettingsMask mask, uid_t uid_shift) {
 357         int r;
 358
 359         assert(path);
 360
 361         r = mkdir(path, mode);
 362         if (r < 0 && errno != EEXIST)
 363                 return -errno;
 364
 365         if ((mask & MOUNT_USE_USERNS) == 0)
 366                 return 0;
 367
 368         if (mask & MOUNT_IN_USERNS)
 369                 return 0;
 370
 371         r = lchown(path, uid_shift, uid_shift);
 372         if (r < 0)
 373                 return -errno;
 374
 375         return 0;
 376 }
 377
 378 static int mkdir_userns_p(const char *prefix, const char *path, mode_t mode, MountSettingsMask mask, uid_t uid_shift) {
 379         const char *p, *e;
 380         int r;
 381
 382         assert(path);
 383
 384         if (prefix && !path_startswith(path, prefix))
 385                 return -ENOTDIR;
 386
 387         /* create every parent directory in the path, except the last component */
 388         p = path + strspn(path, "/");
 389         for (;;) {
 390                 char t[strlen(path) + 1];
 391
 392                 e = p + strcspn(p, "/");
 393                 p = e + strspn(e, "/");
 394
 395                 /* Is this the last component? If so, then we're done */
 396                 if (*p == 0)
 397                         break;
 398
 399                 memcpy(t, path, e - path);
 400                 t[e-path] = 0;
 401
 402                 if (prefix && path_startswith(prefix, t))
 403                         continue;
 404
 405                 r = mkdir_userns(t, mode, mask, uid_shift);
 406                 if (r < 0)
 407                         return r;
 408         }
 409
 410         return mkdir_userns(path, mode, mask, uid_shift);
 411 }
 412
 413 int mount_all(const char *dest,
 414               MountSettingsMask mount_settings,
 415               uid_t uid_shift, uid_t uid_range,
 416               const char *selinux_apifs_context) {
 417
 418         typedef struct MountPoint {
 419                 const char *what;
 420                 const char *where;
 421                 const char *type;
 422                 const char *options;
 423                 unsigned long flags;
 424                 MountSettingsMask mount_settings;
 425         } MountPoint;
 426
 427         static const MountPoint mount_table[] = {
 428                 /* inner child mounts */
 429                 { "proc",                "/proc",               "proc",  NULL,        MS_NOSUID|MS_NOEXEC|MS_NODEV,                              MOUNT_FATAL|MOUNT_IN_USERNS },
 430                 { "/proc/sys",           "/proc/sys",           NULL,    NULL,        MS_BIND,                                                   MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO },                          /* Bind mount first ...*/
 431                 { "/proc/sys/net",       "/proc/sys/net",       NULL,    NULL,        MS_BIND,                                                   MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_APIVFS_NETNS }, /* (except for this) */
 432                 { NULL,                  "/proc/sys",           NULL,    NULL,        MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO },                          /* ... then, make it r/o */
 433                 { "/proc/sysrq-trigger", "/proc/sysrq-trigger", NULL,    NULL,        MS_BIND,                                                               MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO },                          /* Bind mount first ...*/
 434                 { NULL,                  "/proc/sysrq-trigger", NULL,    NULL,        MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT,             MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO },                          /* ... then, make it r/o */
 435                 { "tmpfs",               "/tmp",                "tmpfs", "mode=1777", MS_STRICTATIME,                                            MOUNT_FATAL|MOUNT_IN_USERNS },
 436
 437                 /* outer child mounts */
 438                 { "tmpfs",               "/sys",                "tmpfs", "mode=755",  MS_NOSUID|MS_NOEXEC|MS_NODEV,                              MOUNT_FATAL|MOUNT_APPLY_APIVFS_NETNS },
 439                 { "sysfs",               "/sys",                "sysfs", NULL,        MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV,                    MOUNT_FATAL|MOUNT_APPLY_APIVFS_RO },    /* skipped if above was mounted */
 440                 { "sysfs",               "/sys",                "sysfs", NULL,                  MS_NOSUID|MS_NOEXEC|MS_NODEV,                    MOUNT_FATAL },                          /* skipped if above was mounted */
 441
 442                 { "tmpfs",               "/dev",                "tmpfs", "mode=755",  MS_NOSUID|MS_STRICTATIME,                                  MOUNT_FATAL },
 443                 { "tmpfs",               "/dev/shm",            "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME,                         MOUNT_FATAL },
 444                 { "tmpfs",               "/run",                "tmpfs", "mode=755",  MS_NOSUID|MS_NODEV|MS_STRICTATIME,                         MOUNT_FATAL },
 445 #ifdef HAVE_SELINUX
 446                 { "/sys/fs/selinux",     "/sys/fs/selinux",     NULL,     NULL,       MS_BIND,                                                   0 },  /* Bind mount first */
 447                 { NULL,                  "/sys/fs/selinux",     NULL,     NULL,       MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, 0 },  /* Then, make it r/o */
 448 #endif
 449         };
 450
 451         unsigned k;
 452         int r;
 453         bool use_userns = (mount_settings & MOUNT_USE_USERNS);
 454         bool netns = (mount_settings & MOUNT_APPLY_APIVFS_NETNS);
 455         bool ro = (mount_settings & MOUNT_APPLY_APIVFS_RO);
 456         bool in_userns = (mount_settings & MOUNT_IN_USERNS);
 457
 458         for (k = 0; k < ELEMENTSOF(mount_table); k++) {
 459                 _cleanup_free_ char *where = NULL, *options = NULL;
 460                 const char *o;
 461                 bool fatal = (mount_table[k].mount_settings & MOUNT_FATAL);
 462
 463                 if (in_userns != (bool)(mount_table[k].mount_settings & MOUNT_IN_USERNS))
 464                         continue;
 465
 466                 if (!netns && (bool)(mount_table[k].mount_settings & MOUNT_APPLY_APIVFS_NETNS))
 467                         continue;
 468
 469                 if (!ro && (bool)(mount_table[k].mount_settings & MOUNT_APPLY_APIVFS_RO))
 470                         continue;
 471
 472                 r = chase_symlinks(mount_table[k].where, dest, CHASE_NON_EXISTING|CHASE_PREFIX_ROOT, &where);
 473                 if (r < 0)
 474                         return log_error_errno(r, "Failed to resolve %s: %m", mount_table[k].where);
 475
 476                 r = path_is_mount_point(where, NULL, 0);
 477                 if (r < 0 && r != -ENOENT)
 478                         return log_error_errno(r, "Failed to detect whether %s is a mount point: %m", where);
 479
 480                 /* Skip this entry if it is not a remount. */
 481                 if (mount_table[k].what && r > 0)
 482                         continue;
 483
 484                 r = mkdir_userns_p(dest, where, 0755, mount_settings, uid_shift);
 485                 if (r < 0 && r != -EEXIST) {
 486                         if (fatal)
 487                                 return log_error_errno(r, "Failed to create directory %s: %m", where);
 488
 489                         log_debug_errno(r, "Failed to create directory %s: %m", where);
 490                         continue;
 491                 }
 492
 493                 o = mount_table[k].options;
 494                 if (streq_ptr(mount_table[k].type, "tmpfs")) {
 495                         if (in_userns)
 496                                 r = tmpfs_patch_options(o, use_userns, 0, uid_range, true, selinux_apifs_context, &options);
 497                         else
 498                                 r = tmpfs_patch_options(o, use_userns, uid_shift, uid_range, false, selinux_apifs_context, &options);
 499                         if (r < 0)
 500                                 return log_oom();
 501                         if (r > 0)
 502                                 o = options;
 503                 }
 504
 505                 r = mount_verbose(fatal ? LOG_ERR : LOG_DEBUG,
 506                                   mount_table[k].what,
 507                                   where,
 508                                   mount_table[k].type,
 509                                   mount_table[k].flags,
 510                                   o);
 511                 if (r < 0 && fatal)
 512                         return r;
 513         }
 514
 515         return 0;
 516 }
 517
 518 static int parse_mount_bind_options(const char *options, unsigned long *mount_flags, char **mount_opts) {
 519         const char *p = options;
 520         unsigned long flags = *mount_flags;
 521         char *opts = NULL;
 522         int r;
 523
 524         assert(options);
 525
 526         for (;;) {
 527                 _cleanup_free_ char *word = NULL;
 528
 529                 r = extract_first_word(&p, &word, ",", 0);
 530                 if (r < 0)
 531                         return log_error_errno(r, "Failed to extract mount option: %m");
 532                 if (r == 0)
 533                         break;
 534
 535                 if (streq(word, "rbind"))
 536                         flags |= MS_REC;
 537                 else if (streq(word, "norbind"))
 538                         flags &= ~MS_REC;
 539                 else {
 540                         log_error("Invalid bind mount option: %s", word);
 541                         return -EINVAL;
 542                 }
 543         }
 544
 545         *mount_flags = flags;
 546         /* in the future mount_opts will hold string options for mount(2) */
 547         *mount_opts = opts;
 548
 549         return 0;
 550 }
 551
 552 static int mount_bind(const char *dest, CustomMount *m) {
 553
 554         _cleanup_free_ char *mount_opts = NULL, *where = NULL;
 555         unsigned long mount_flags = MS_BIND | MS_REC;
 556         struct stat source_st, dest_st;
 557         int r;
 558
 559         assert(m);
 560
 561         if (m->options) {
 562                 r = parse_mount_bind_options(m->options, &mount_flags, &mount_opts);
 563                 if (r < 0)
 564                         return r;
 565         }
 566
 567         if (stat(m->source, &source_st) < 0)
 568                 return log_error_errno(errno, "Failed to stat %s: %m", m->source);
 569
 570         r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NON_EXISTING, &where);
 571         if (r < 0)
 572                 return log_error_errno(r, "Failed to resolve %s: %m", m->destination);
 573         if (r > 0) { /* Path exists already? */
 574
 575                 if (stat(where, &dest_st) < 0)
 576                         return log_error_errno(errno, "Failed to stat %s: %m", where);
 577
 578                 if (S_ISDIR(source_st.st_mode) && !S_ISDIR(dest_st.st_mode)) {
 579                         log_error("Cannot bind mount directory %s on file %s.", m->source, where);
 580                         return -EINVAL;
 581                 }
 582
 583                 if (!S_ISDIR(source_st.st_mode) && S_ISDIR(dest_st.st_mode)) {
 584                         log_error("Cannot bind mount file %s on directory %s.", m->source, where);
 585                         return -EINVAL;
 586                 }
 587
 588         } else { /* Path doesn't exist yet? */
 589                 r = mkdir_parents_label(where, 0755);
 590                 if (r < 0)
 591                         return log_error_errno(r, "Failed to make parents of %s: %m", where);
 592
 593                 /* Create the mount point. Any non-directory file can be
 594                 * mounted on any non-directory file (regular, fifo, socket,
 595                 * char, block).
 596                 */
 597                 if (S_ISDIR(source_st.st_mode))
 598                         r = mkdir_label(where, 0755);
 599                 else
 600                         r = touch(where);
 601                 if (r < 0)
 602                         return log_error_errno(r, "Failed to create mount point %s: %m", where);
 603
 604         }
 605
 606         r = mount_verbose(LOG_ERR, m->source, where, NULL, mount_flags, mount_opts);
 607         if (r < 0)
 608                 return r;
 609
 610         if (m->read_only) {
 611                 r = bind_remount_recursive(where, true, NULL);
 612                 if (r < 0)
 613                         return log_error_errno(r, "Read-only bind mount failed: %m");
 614         }
 615
 616         return 0;
 617 }
 618
 619 static int mount_tmpfs(
 620                 const char *dest,
 621                 CustomMount *m,
 622                 bool userns, uid_t uid_shift, uid_t uid_range,
 623                 const char *selinux_apifs_context) {
 624
 625         const char *options;
 626         _cleanup_free_ char *buf = NULL, *where = NULL;
 627         int r;
 628
 629         assert(dest);
 630         assert(m);
 631
 632         r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NON_EXISTING, &where);
 633         if (r < 0)
 634                 return log_error_errno(r, "Failed to resolve %s: %m", m->destination);
 635         if (r == 0) { /* Doesn't exist yet? */
 636                 r = mkdir_p_label(where, 0755);
 637                 if (r < 0)
 638                         return log_error_errno(r, "Creating mount point for tmpfs %s failed: %m", where);
 639         }
 640
 641         r = tmpfs_patch_options(m->options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
 642         if (r < 0)
 643                 return log_oom();
 644         options = r > 0 ? buf : m->options;
 645
 646         return mount_verbose(LOG_ERR, "tmpfs", where, "tmpfs", MS_NODEV|MS_STRICTATIME, options);
 647 }
 648
 649 static char *joined_and_escaped_lower_dirs(char * const *lower) {
 650         _cleanup_strv_free_ char **sv = NULL;
 651
 652         sv = strv_copy(lower);
 653         if (!sv)
 654                 return NULL;
 655
 656         strv_reverse(sv);
 657
 658         if (!strv_shell_escape(sv, ",:"))
 659                 return NULL;
 660
 661         return strv_join(sv, ":");
 662 }
 663
 664 static int mount_overlay(const char *dest, CustomMount *m) {
 665
 666         _cleanup_free_ char *lower = NULL, *where = NULL;
 667         const char *options;
 668         int r;
 669
 670         assert(dest);
 671         assert(m);
 672
 673         r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NON_EXISTING, &where);
 674         if (r < 0)
 675                 return log_error_errno(r, "Failed to resolve %s: %m", m->destination);
 676         if (r == 0) { /* Doesn't exist yet? */
 677                 r = mkdir_label(where, 0755);
 678                 if (r < 0)
 679                         return log_error_errno(r, "Creating mount point for overlay %s failed: %m", where);
 680         }
 681
 682         (void) mkdir_p_label(m->source, 0755);
 683
 684         lower = joined_and_escaped_lower_dirs(m->lower);
 685         if (!lower)
 686                 return log_oom();
 687
 688         if (m->read_only) {
 689                 _cleanup_free_ char *escaped_source = NULL;
 690
 691                 escaped_source = shell_escape(m->source, ",:");
 692                 if (!escaped_source)
 693                         return log_oom();
 694
 695                 options = strjoina("lowerdir=", escaped_source, ":", lower);
 696         } else {
 697                 _cleanup_free_ char *escaped_source = NULL, *escaped_work_dir = NULL;
 698
 699                 assert(m->work_dir);
 700                 (void) mkdir_label(m->work_dir, 0700);
 701
 702                 escaped_source = shell_escape(m->source, ",:");
 703                 if (!escaped_source)
 704                         return log_oom();
 705                 escaped_work_dir = shell_escape(m->work_dir, ",:");
 706                 if (!escaped_work_dir)
 707                         return log_oom();
 708
 709                 options = strjoina("lowerdir=", lower, ",upperdir=", escaped_source, ",workdir=", escaped_work_dir);
 710         }
 711
 712         return mount_verbose(LOG_ERR, "overlay", where, "overlay", m->read_only ? MS_RDONLY : 0, options);
 713 }
 714
 715 int mount_custom(
 716                 const char *dest,
 717                 CustomMount *mounts, unsigned n,
 718                 bool userns, uid_t uid_shift, uid_t uid_range,
 719                 const char *selinux_apifs_context) {
 720
 721         unsigned i;
 722         int r;
 723
 724         assert(dest);
 725
 726         for (i = 0; i < n; i++) {
 727                 CustomMount *m = mounts + i;
 728
 729                 switch (m->type) {
 730
 731                 case CUSTOM_MOUNT_BIND:
 732                         r = mount_bind(dest, m);
 733                         break;
 734
 735                 case CUSTOM_MOUNT_TMPFS:
 736                         r = mount_tmpfs(dest, m, userns, uid_shift, uid_range, selinux_apifs_context);
 737                         break;
 738
 739                 case CUSTOM_MOUNT_OVERLAY:
 740                         r = mount_overlay(dest, m);
 741                         break;
 742
 743                 default:
 744                         assert_not_reached("Unknown custom mount type");
 745                 }
 746
 747                 if (r < 0)
 748                         return r;
 749         }
 750
 751         return 0;
 752 }
 753
 754 /* Retrieve existing subsystems. This function is called in a new cgroup
 755  * namespace.
 756  */
 757 static int get_controllers(Set *subsystems) {
 758         _cleanup_fclose_ FILE *f = NULL;
 759         char line[LINE_MAX];
 760
 761         assert(subsystems);
 762
 763         f = fopen("/proc/self/cgroup", "re");
 764         if (!f)
 765                 return errno == ENOENT ? -ESRCH : -errno;
 766
 767         FOREACH_LINE(line, f, return -errno) {
 768                 int r;
 769                 char *e, *l, *p;
 770
 771                 l = strchr(line, ':');
 772                 if (!l)
 773                         continue;
 774
 775                 l++;
 776                 e = strchr(l, ':');
 777                 if (!e)
 778                         continue;
 779
 780                 *e = 0;
 781
 782                 if (STR_IN_SET(l, "", "name=systemd"))
 783                         continue;
 784
 785                 p = strdup(l);
 786                 if (!p)
 787                         return -ENOMEM;
 788
 789                 r = set_consume(subsystems, p);
 790                 if (r < 0)
 791                         return r;
 792         }
 793
 794         return 0;
 795 }
 796
 797 static int mount_legacy_cgroup_hierarchy(
 798                 const char *dest,
 799                 const char *controller,
 800                 const char *hierarchy,
 801                 CGroupUnified unified_requested,
 802                 bool read_only) {
 803
 804         const char *to, *fstype, *opts;
 805         int r;
 806
 807         to = strjoina(strempty(dest), "/sys/fs/cgroup/", hierarchy);
 808
 809         r = path_is_mount_point(to, dest, 0);
 810         if (r < 0 && r != -ENOENT)
 811                 return log_error_errno(r, "Failed to determine if %s is mounted already: %m", to);
 812         if (r > 0)
 813                 return 0;
 814
 815         mkdir_p(to, 0755);
 816
 817         /* The superblock mount options of the mount point need to be
 818          * identical to the hosts', and hence writable... */
 819         if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
 820                 if (unified_requested >= CGROUP_UNIFIED_SYSTEMD) {
 821                         fstype = "cgroup2";
 822                         opts = NULL;
 823                 } else {
 824                         fstype = "cgroup";
 825                         opts = "none,name=systemd,xattr";
 826                 }
 827         } else {
 828                 fstype = "cgroup";
 829                 opts = controller;
 830         }
 831
 832         r = mount_verbose(LOG_ERR, "cgroup", to, fstype, MS_NOSUID|MS_NOEXEC|MS_NODEV, opts);
 833         if (r < 0)
 834                 return r;
 835
 836         /* ... hence let's only make the bind mount read-only, not the superblock. */
 837         if (read_only) {
 838                 r = mount_verbose(LOG_ERR, NULL, to, NULL,
 839                                   MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL);
 840                 if (r < 0)
 841                         return r;
 842         }
 843
 844         return 1;
 845 }
 846
 847 /* Mount a legacy cgroup hierarchy when cgroup namespaces are supported. */
 848 static int mount_legacy_cgns_supported(
 849                 const char *dest,
 850                 CGroupUnified unified_requested,
 851                 bool userns,
 852                 uid_t uid_shift,
 853                 uid_t uid_range,
 854                 const char *selinux_apifs_context) {
 855
 856         _cleanup_set_free_free_ Set *controllers = NULL;
 857         const char *cgroup_root = "/sys/fs/cgroup", *c;
 858         int r;
 859
 860         (void) mkdir_p(cgroup_root, 0755);
 861
 862         /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
 863         r = path_is_mount_point(cgroup_root, dest, AT_SYMLINK_FOLLOW);
 864         if (r < 0)
 865                 return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
 866         if (r == 0) {
 867                 _cleanup_free_ char *options = NULL;
 868
 869                 /* When cgroup namespaces are enabled and user namespaces are
 870                  * used then the mount of the cgroupfs is done *inside* the new
 871                  * user namespace. We're root in the new user namespace and the
 872                  * kernel will happily translate our uid/gid to the correct
 873                  * uid/gid as seen from e.g. /proc/1/mountinfo. So we simply
 874                  * pass uid 0 and not uid_shift to tmpfs_patch_options().
 875                  */
 876                 r = tmpfs_patch_options("mode=755", userns, 0, uid_range, true, selinux_apifs_context, &options);
 877                 if (r < 0)
 878                         return log_oom();
 879
 880                 r = mount_verbose(LOG_ERR, "tmpfs", cgroup_root, "tmpfs",
 881                                   MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options);
 882                 if (r < 0)
 883                         return r;
 884         }
 885
 886         if (cg_all_unified() > 0)
 887                 goto skip_controllers;
 888
 889         controllers = set_new(&string_hash_ops);
 890         if (!controllers)
 891                 return log_oom();
 892
 893         r = get_controllers(controllers);
 894         if (r < 0)
 895                 return log_error_errno(r, "Failed to determine cgroup controllers: %m");
 896
 897         for (;;) {
 898                 _cleanup_free_ const char *controller = NULL;
 899
 900                 controller = set_steal_first(controllers);
 901                 if (!controller)
 902                         break;
 903
 904                 r = mount_legacy_cgroup_hierarchy("", controller, controller, unified_requested, !userns);
 905                 if (r < 0)
 906                         return r;
 907
 908                 /* When multiple hierarchies are co-mounted, make their
 909                  * constituting individual hierarchies a symlink to the
 910                  * co-mount.
 911                  */
 912                 c = controller;
 913                 for (;;) {
 914                         _cleanup_free_ char *target = NULL, *tok = NULL;
 915
 916                         r = extract_first_word(&c, &tok, ",", 0);
 917                         if (r < 0)
 918                                 return log_error_errno(r, "Failed to extract co-mounted cgroup controller: %m");
 919                         if (r == 0)
 920                                 break;
 921
 922                         target = prefix_root("/sys/fs/cgroup", tok);
 923                         if (!target)
 924                                 return log_oom();
 925
 926                         if (streq(controller, tok))
 927                                 break;
 928
 929                         r = symlink_idempotent(controller, target);
 930                         if (r == -EINVAL)
 931                                 return log_error_errno(r, "Invalid existing symlink for combined hierarchy: %m");
 932                         if (r < 0)
 933                                 return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m");
 934                 }
 935         }
 936
 937 skip_controllers:
 938         r = mount_legacy_cgroup_hierarchy("", SYSTEMD_CGROUP_CONTROLLER, "systemd", unified_requested, false);
 939         if (r < 0)
 940                 return r;
 941
 942         if (!userns)
 943                 return mount_verbose(LOG_ERR, NULL, cgroup_root, NULL,
 944                                      MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
 945
 946         return 0;
 947 }
 948
 949 /* Mount legacy cgroup hierarchy when cgroup namespaces are unsupported. */
 950 static int mount_legacy_cgns_unsupported(
 951                 const char *dest,
 952                 CGroupUnified unified_requested,
 953                 bool userns,
 954                 uid_t uid_shift,
 955                 uid_t uid_range,
 956                 const char *selinux_apifs_context) {
 957
 958         _cleanup_set_free_free_ Set *controllers = NULL;
 959         const char *cgroup_root;
 960         int r;
 961
 962         cgroup_root = prefix_roota(dest, "/sys/fs/cgroup");
 963
 964         (void) mkdir_p(cgroup_root, 0755);
 965
 966         /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
 967         r = path_is_mount_point(cgroup_root, dest, AT_SYMLINK_FOLLOW);
 968         if (r < 0)
 969                 return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
 970         if (r == 0) {
 971                 _cleanup_free_ char *options = NULL;
 972
 973                 r = tmpfs_patch_options("mode=755", userns, uid_shift, uid_range, false, selinux_apifs_context, &options);
 974                 if (r < 0)
 975                         return log_oom();
 976
 977                 r = mount_verbose(LOG_ERR, "tmpfs", cgroup_root, "tmpfs",
 978                                   MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options);
 979                 if (r < 0)
 980                         return r;
 981         }
 982
 983         if (cg_all_unified() > 0)
 984                 goto skip_controllers;
 985
 986         controllers = set_new(&string_hash_ops);
 987         if (!controllers)
 988                 return log_oom();
 989
 990         r = cg_kernel_controllers(controllers);
 991         if (r < 0)
 992                 return log_error_errno(r, "Failed to determine cgroup controllers: %m");
 993
 994         for (;;) {
 995                 _cleanup_free_ char *controller = NULL, *origin = NULL, *combined = NULL;
 996
 997                 controller = set_steal_first(controllers);
 998                 if (!controller)
 999                         break;
1000
1001                 origin = prefix_root("/sys/fs/cgroup/", controller);
1002                 if (!origin)
1003                         return log_oom();
1004
1005                 r = readlink_malloc(origin, &combined);
1006                 if (r == -EINVAL) {
1007                         /* Not a symbolic link, but directly a single cgroup hierarchy */
1008
1009                         r = mount_legacy_cgroup_hierarchy(dest, controller, controller, unified_requested, true);
1010                         if (r < 0)
1011                                 return r;
1012
1013                 } else if (r < 0)
1014                         return log_error_errno(r, "Failed to read link %s: %m", origin);
1015                 else {
1016                         _cleanup_free_ char *target = NULL;
1017
1018                         target = prefix_root(dest, origin);
1019                         if (!target)
1020                                 return log_oom();
1021
1022                         /* A symbolic link, a combination of controllers in one hierarchy */
1023
1024                         if (!filename_is_valid(combined)) {
1025                                 log_warning("Ignoring invalid combined hierarchy %s.", combined);
1026                                 continue;
1027                         }
1028
1029                         r = mount_legacy_cgroup_hierarchy(dest, combined, combined, unified_requested, true);
1030                         if (r < 0)
1031                                 return r;
1032
1033                         r = symlink_idempotent(combined, target);
1034                         if (r == -EINVAL)
1035                                 return log_error_errno(r, "Invalid existing symlink for combined hierarchy: %m");
1036                         if (r < 0)
1037                                 return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m");
1038                 }
1039         }
1040
1041 skip_controllers:
1042         r = mount_legacy_cgroup_hierarchy(dest, SYSTEMD_CGROUP_CONTROLLER, "systemd", unified_requested, false);
1043         if (r < 0)
1044                 return r;
1045
1046         return mount_verbose(LOG_ERR, NULL, cgroup_root, NULL,
1047                              MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
1048 }
1049
1050 static int mount_unified_cgroups(const char *dest) {
1051         const char *p;
1052         int r;
1053
1054         assert(dest);
1055
1056         p = prefix_roota(dest, "/sys/fs/cgroup");
1057
1058         (void) mkdir_p(p, 0755);
1059
1060         r = path_is_mount_point(p, dest, AT_SYMLINK_FOLLOW);
1061         if (r < 0)
1062                 return log_error_errno(r, "Failed to determine if %s is mounted already: %m", p);
1063         if (r > 0) {
1064                 p = prefix_roota(dest, "/sys/fs/cgroup/cgroup.procs");
1065                 if (access(p, F_OK) >= 0)
1066                         return 0;
1067                 if (errno != ENOENT)
1068                         return log_error_errno(errno, "Failed to determine if mount point %s contains the unified cgroup hierarchy: %m", p);
1069
1070                 log_error("%s is already mounted but not a unified cgroup hierarchy. Refusing.", p);
1071                 return -EINVAL;
1072         }
1073
1074         return mount_verbose(LOG_ERR, "cgroup", p, "cgroup2", MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
1075 }
1076
1077 int mount_cgroups(
1078                 const char *dest,
1079                 CGroupUnified unified_requested,
1080                 bool userns,
1081                 uid_t uid_shift,
1082                 uid_t uid_range,
1083                 const char *selinux_apifs_context,
1084                 bool use_cgns) {
1085
1086         if (unified_requested >= CGROUP_UNIFIED_ALL)
1087                 return mount_unified_cgroups(dest);
1088         else if (use_cgns)
1089                 return mount_legacy_cgns_supported(dest, unified_requested, userns, uid_shift, uid_range, selinux_apifs_context);
1090
1091         return mount_legacy_cgns_unsupported(dest, unified_requested, userns, uid_shift, uid_range, selinux_apifs_context);
1092 }
1093
1094 int mount_systemd_cgroup_writable(
1095                 const char *dest,
1096                 CGroupUnified unified_requested) {
1097
1098         _cleanup_free_ char *own_cgroup_path = NULL;
1099         const char *systemd_root, *systemd_own;
1100         int r;
1101
1102         assert(dest);
1103
1104         r = cg_pid_get_path(NULL, 0, &own_cgroup_path);
1105         if (r < 0)
1106                 return log_error_errno(r, "Failed to determine our own cgroup path: %m");
1107
1108         /* If we are living in the top-level, then there's nothing to do... */
1109         if (path_equal(own_cgroup_path, "/"))
1110                 return 0;
1111
1112         if (unified_requested >= CGROUP_UNIFIED_ALL) {
1113                 systemd_own = strjoina(dest, "/sys/fs/cgroup", own_cgroup_path);
1114                 systemd_root = prefix_roota(dest, "/sys/fs/cgroup");
1115         } else {
1116                 systemd_own = strjoina(dest, "/sys/fs/cgroup/systemd", own_cgroup_path);
1117                 systemd_root = prefix_roota(dest, "/sys/fs/cgroup/systemd");
1118         }
1119
1120         /* Make our own cgroup a (writable) bind mount */
1121         r = mount_verbose(LOG_ERR, systemd_own, systemd_own,  NULL, MS_BIND, NULL);
1122         if (r < 0)
1123                 return r;
1124
1125         /* And then remount the systemd cgroup root read-only */
1126         return mount_verbose(LOG_ERR, NULL, systemd_root, NULL,
1127                              MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL);
1128 }
1129
1130 int setup_volatile_state(
1131                 const char *directory,
1132                 VolatileMode mode,
1133                 bool userns, uid_t uid_shift, uid_t uid_range,
1134                 const char *selinux_apifs_context) {
1135
1136         _cleanup_free_ char *buf = NULL;
1137         const char *p, *options;
1138         int r;
1139
1140         assert(directory);
1141
1142         if (mode != VOLATILE_STATE)
1143                 return 0;
1144
1145         /* --volatile=state means we simply overmount /var
1146            with a tmpfs, and the rest read-only. */
1147
1148         r = bind_remount_recursive(directory, true, NULL);
1149         if (r < 0)
1150                 return log_error_errno(r, "Failed to remount %s read-only: %m", directory);
1151
1152         p = prefix_roota(directory, "/var");
1153         r = mkdir(p, 0755);
1154         if (r < 0 && errno != EEXIST)
1155                 return log_error_errno(errno, "Failed to create %s: %m", directory);
1156
1157         options = "mode=755";
1158         r = tmpfs_patch_options(options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
1159         if (r < 0)
1160                 return log_oom();
1161         if (r > 0)
1162                 options = buf;
1163
1164         return mount_verbose(LOG_ERR, "tmpfs", p, "tmpfs", MS_STRICTATIME, options);
1165 }
1166
1167 int setup_volatile(
1168                 const char *directory,
1169                 VolatileMode mode,
1170                 bool userns, uid_t uid_shift, uid_t uid_range,
1171                 const char *selinux_apifs_context) {
1172
1173         bool tmpfs_mounted = false, bind_mounted = false;
1174         char template[] = "/tmp/nspawn-volatile-XXXXXX";
1175         _cleanup_free_ char *buf = NULL;
1176         const char *f, *t, *options;
1177         int r;
1178
1179         assert(directory);
1180
1181         if (mode != VOLATILE_YES)
1182                 return 0;
1183
1184         /* --volatile=yes means we mount a tmpfs to the root dir, and
1185            the original /usr to use inside it, and that read-only. */
1186
1187         if (!mkdtemp(template))
1188                 return log_error_errno(errno, "Failed to create temporary directory: %m");
1189
1190         options = "mode=755";
1191         r = tmpfs_patch_options(options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
1192         if (r < 0)
1193                 return log_oom();
1194         if (r > 0)
1195                 options = buf;
1196
1197         r = mount_verbose(LOG_ERR, "tmpfs", template, "tmpfs", MS_STRICTATIME, options);
1198         if (r < 0)
1199                 goto fail;
1200
1201         tmpfs_mounted = true;
1202
1203         f = prefix_roota(directory, "/usr");
1204         t = prefix_roota(template, "/usr");
1205
1206         r = mkdir(t, 0755);
1207         if (r < 0 && errno != EEXIST) {
1208                 r = log_error_errno(errno, "Failed to create %s: %m", t);
1209                 goto fail;
1210         }
1211
1212         r = mount_verbose(LOG_ERR, f, t, NULL, MS_BIND|MS_REC, NULL);
1213         if (r < 0)
1214                 goto fail;
1215
1216         bind_mounted = true;
1217
1218         r = bind_remount_recursive(t, true, NULL);
1219         if (r < 0) {
1220                 log_error_errno(r, "Failed to remount %s read-only: %m", t);
1221                 goto fail;
1222         }
1223
1224         r = mount_verbose(LOG_ERR, template, directory, NULL, MS_MOVE, NULL);
1225         if (r < 0)
1226                 goto fail;
1227
1228         (void) rmdir(template);
1229
1230         return 0;
1231
1232 fail:
1233         if (bind_mounted)
1234                 (void) umount_verbose(t);
1235
1236         if (tmpfs_mounted)
1237                 (void) umount_verbose(template);
1238         (void) rmdir(template);
1239         return r;
1240 }
1241
1242 VolatileMode volatile_mode_from_string(const char *s) {
1243         int b;
1244
1245         if (isempty(s))
1246                 return _VOLATILE_MODE_INVALID;
1247
1248         b = parse_boolean(s);
1249         if (b > 0)
1250                 return VOLATILE_YES;
1251         if (b == 0)
1252                 return VOLATILE_NO;
1253
1254         if (streq(s, "state"))
1255                 return VOLATILE_STATE;
1256
1257         return _VOLATILE_MODE_INVALID;
1258 }