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