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(*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 log_oom();
 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 static int tmpfs_patch_options(
 184                 const char *options,
 185                 bool userns,
 186                 uid_t uid_shift, uid_t uid_range,
 187                 bool patch_ids,
 188                 const char *selinux_apifs_context,
 189                 char **ret) {
 190
 191         char *buf = NULL;
 192
 193         if ((userns && uid_shift != 0) || patch_ids) {
 194                 assert(uid_shift != UID_INVALID);
 195
 196                 if (options)
 197                         (void) asprintf(&buf, "%s,uid=" UID_FMT ",gid=" UID_FMT, options, uid_shift, uid_shift);
 198                 else
 199                         (void) asprintf(&buf, "uid=" UID_FMT ",gid=" UID_FMT, uid_shift, uid_shift);
 200                 if (!buf)
 201                         return -ENOMEM;
 202
 203                 options = buf;
 204         }
 205
 206 #ifdef HAVE_SELINUX
 207         if (selinux_apifs_context) {
 208                 char *t;
 209
 210                 if (options)
 211                         t = strjoin(options, ",context=\"", selinux_apifs_context, "\"", NULL);
 212                 else
 213                         t = strjoin("context=\"", selinux_apifs_context, "\"", NULL);
 214                 if (!t) {
 215                         free(buf);
 216                         return -ENOMEM;
 217                 }
 218
 219                 free(buf);
 220                 buf = t;
 221         }
 222 #endif
 223
 224         if (!buf && options) {
 225                 buf = strdup(options);
 226                 if (!buf)
 227                         return -ENOMEM;
 228         }
 229         *ret = buf;
 230
 231         return !!buf;
 232 }
 233
 234 int mount_sysfs(const char *dest) {
 235         const char *full, *top, *x;
 236         int r;
 237
 238         top = prefix_roota(dest, "/sys");
 239         r = path_check_fstype(top, SYSFS_MAGIC);
 240         if (r < 0)
 241                 return log_error_errno(r, "Failed to determine filesystem type of %s: %m", top);
 242         /* /sys might already be mounted as sysfs by the outer child in the
 243          * !netns case. In this case, it's all good. Don't touch it because we
 244          * don't have the right to do so, see https://github.com/systemd/systemd/issues/1555.
 245          */
 246         if (r > 0)
 247                 return 0;
 248
 249         full = prefix_roota(top, "/full");
 250
 251         (void) mkdir(full, 0755);
 252
 253         r = mount_verbose(LOG_ERR, "sysfs", full, "sysfs",
 254                           MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
 255         if (r < 0)
 256                 return r;
 257
 258         FOREACH_STRING(x, "block", "bus", "class", "dev", "devices", "kernel") {
 259                 _cleanup_free_ char *from = NULL, *to = NULL;
 260
 261                 from = prefix_root(full, x);
 262                 if (!from)
 263                         return log_oom();
 264
 265                 to = prefix_root(top, x);
 266                 if (!to)
 267                         return log_oom();
 268
 269                 (void) mkdir(to, 0755);
 270
 271                 r = mount_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
 272                 if (r < 0)
 273                         return r;
 274
 275                 r = mount_verbose(LOG_ERR, NULL, to, NULL,
 276                                   MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, NULL);
 277                 if (r < 0)
 278                         return r;
 279         }
 280
 281         r = umount_verbose(full);
 282         if (r < 0)
 283                 return r;
 284
 285         if (rmdir(full) < 0)
 286                 return log_error_errno(errno, "Failed to remove %s: %m", full);
 287
 288         x = prefix_roota(top, "/fs/kdbus");
 289         (void) mkdir_p(x, 0755);
 290
 291         /* Create mountpoint for cgroups. Otherwise we are not allowed since we
 292          * remount /sys read-only.
 293          */
 294         if (cg_ns_supported()) {
 295                 x = prefix_roota(top, "/fs/cgroup");
 296                 (void) mkdir_p(x, 0755);
 297         }
 298
 299         return mount_verbose(LOG_ERR, NULL, top, NULL,
 300                              MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, NULL);
 301 }
 302
 303 static int mkdir_userns(const char *path, mode_t mode, bool in_userns, uid_t uid_shift) {
 304         int r;
 305
 306         assert(path);
 307
 308         r = mkdir(path, mode);
 309         if (r < 0 && errno != EEXIST)
 310                 return -errno;
 311
 312         if (!in_userns) {
 313                 r = lchown(path, uid_shift, uid_shift);
 314                 if (r < 0)
 315                         return -errno;
 316         }
 317
 318         return 0;
 319 }
 320
 321 static int mkdir_userns_p(const char *prefix, const char *path, mode_t mode, bool in_userns, uid_t uid_shift) {
 322         const char *p, *e;
 323         int r;
 324
 325         assert(path);
 326
 327         if (prefix && !path_startswith(path, prefix))
 328                 return -ENOTDIR;
 329
 330         /* create every parent directory in the path, except the last component */
 331         p = path + strspn(path, "/");
 332         for (;;) {
 333                 char t[strlen(path) + 1];
 334
 335                 e = p + strcspn(p, "/");
 336                 p = e + strspn(e, "/");
 337
 338                 /* Is this the last component? If so, then we're done */
 339                 if (*p == 0)
 340                         break;
 341
 342                 memcpy(t, path, e - path);
 343                 t[e-path] = 0;
 344
 345                 if (prefix && path_startswith(prefix, t))
 346                         continue;
 347
 348                 r = mkdir_userns(t, mode, in_userns, uid_shift);
 349                 if (r < 0)
 350                         return r;
 351         }
 352
 353         return mkdir_userns(path, mode, in_userns, uid_shift);
 354 }
 355
 356 int mount_all(const char *dest,
 357               bool use_userns, bool in_userns,
 358               bool use_netns,
 359               uid_t uid_shift, uid_t uid_range,
 360               const char *selinux_apifs_context) {
 361
 362         typedef struct MountPoint {
 363                 const char *what;
 364                 const char *where;
 365                 const char *type;
 366                 const char *options;
 367                 unsigned long flags;
 368                 bool fatal;
 369                 bool in_userns;
 370                 bool use_netns;
 371         } MountPoint;
 372
 373         static const MountPoint mount_table[] = {
 374                 { "proc",                "/proc",               "proc",  NULL,        MS_NOSUID|MS_NOEXEC|MS_NODEV,                              true,  true,  false },
 375                 { "/proc/sys",           "/proc/sys",           NULL,    NULL,        MS_BIND,                                                   true,  true,  false },   /* Bind mount first ...*/
 376                 { "/proc/sys/net",       "/proc/sys/net",       NULL,    NULL,        MS_BIND,                                                   true,  true,  true  },   /* (except for this) */
 377                 { 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 */
 378                 { "/proc/sysrq-trigger", "/proc/sysrq-trigger", NULL,    NULL,        MS_BIND,                                                   false, true,  false },   /* Bind mount first ...*/
 379                 { NULL,                  "/proc/sysrq-trigger", NULL,    NULL,        MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, false, true,  false },   /* ... then, make it r/o */
 380                 { "tmpfs",               "/sys",                "tmpfs", "mode=755",  MS_NOSUID|MS_NOEXEC|MS_NODEV,                              true,  false, true  },
 381                 { "sysfs",               "/sys",                "sysfs", NULL,        MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV,                    true,  false, false },
 382                 { "tmpfs",               "/dev",                "tmpfs", "mode=755",  MS_NOSUID|MS_STRICTATIME,                                  true,  false, false },
 383                 { "tmpfs",               "/dev/shm",            "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME,                         true,  false, false },
 384                 { "tmpfs",               "/run",                "tmpfs", "mode=755",  MS_NOSUID|MS_NODEV|MS_STRICTATIME,                         true,  false, false },
 385                 { "tmpfs",               "/tmp",                "tmpfs", "mode=1777", MS_STRICTATIME,                                            true,  true,  false },
 386 #ifdef HAVE_SELINUX
 387                 { "/sys/fs/selinux",     "/sys/fs/selinux",     NULL,     NULL,       MS_BIND,                                                   false, false, false },  /* Bind mount first */
 388                 { 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 */
 389 #endif
 390         };
 391
 392         unsigned k;
 393         int r;
 394
 395         for (k = 0; k < ELEMENTSOF(mount_table); k++) {
 396                 _cleanup_free_ char *where = NULL, *options = NULL;
 397                 const char *o;
 398
 399                 if (in_userns != mount_table[k].in_userns)
 400                         continue;
 401
 402                 if (!use_netns && mount_table[k].use_netns)
 403                         continue;
 404
 405                 where = prefix_root(dest, mount_table[k].where);
 406                 if (!where)
 407                         return log_oom();
 408
 409                 r = path_is_mount_point(where, AT_SYMLINK_FOLLOW);
 410                 if (r < 0 && r != -ENOENT)
 411                         return log_error_errno(r, "Failed to detect whether %s is a mount point: %m", where);
 412
 413                 /* Skip this entry if it is not a remount. */
 414                 if (mount_table[k].what && r > 0)
 415                         continue;
 416
 417                 r = mkdir_userns_p(dest, where, 0755, in_userns, uid_shift);
 418                 if (r < 0 && r != -EEXIST) {
 419                         if (mount_table[k].fatal)
 420                                 return log_error_errno(r, "Failed to create directory %s: %m", where);
 421
 422                         log_debug_errno(r, "Failed to create directory %s: %m", where);
 423                         continue;
 424                 }
 425
 426                 o = mount_table[k].options;
 427                 if (streq_ptr(mount_table[k].type, "tmpfs")) {
 428                         if (in_userns)
 429                                 r = tmpfs_patch_options(o, use_userns, 0, uid_range, true, selinux_apifs_context, &options);
 430                         else
 431                                 r = tmpfs_patch_options(o, use_userns, uid_shift, uid_range, false, selinux_apifs_context, &options);
 432                         if (r < 0)
 433                                 return log_oom();
 434                         if (r > 0)
 435                                 o = options;
 436                 }
 437
 438                 r = mount_verbose(mount_table[k].fatal ? LOG_ERR : LOG_WARNING,
 439                                   mount_table[k].what,
 440                                   where,
 441                                   mount_table[k].type,
 442                                   mount_table[k].flags,
 443                                   o);
 444                 if (r < 0 && mount_table[k].fatal)
 445                         return r;
 446         }
 447
 448         return 0;
 449 }
 450
 451 static int parse_mount_bind_options(const char *options, unsigned long *mount_flags, char **mount_opts) {
 452         const char *p = options;
 453         unsigned long flags = *mount_flags;
 454         char *opts = NULL;
 455
 456         assert(options);
 457
 458         for (;;) {
 459                 _cleanup_free_ char *word = NULL;
 460                 int r = extract_first_word(&p, &word, ",", 0);
 461                 if (r < 0)
 462                         return log_error_errno(r, "Failed to extract mount option: %m");
 463                 if (r == 0)
 464                         break;
 465
 466                 if (streq(word, "rbind"))
 467                         flags |= MS_REC;
 468                 else if (streq(word, "norbind"))
 469                         flags &= ~MS_REC;
 470                 else {
 471                         log_error("Invalid bind mount option: %s", word);
 472                         return -EINVAL;
 473                 }
 474         }
 475
 476         *mount_flags = flags;
 477         /* in the future mount_opts will hold string options for mount(2) */
 478         *mount_opts = opts;
 479
 480         return 0;
 481 }
 482
 483 static int mount_bind(const char *dest, CustomMount *m) {
 484         struct stat source_st, dest_st;
 485         const char *where;
 486         unsigned long mount_flags = MS_BIND | MS_REC;
 487         _cleanup_free_ char *mount_opts = NULL;
 488         int r;
 489
 490         assert(m);
 491
 492         if (m->options) {
 493                 r = parse_mount_bind_options(m->options, &mount_flags, &mount_opts);
 494                 if (r < 0)
 495                         return r;
 496         }
 497
 498         if (stat(m->source, &source_st) < 0)
 499                 return log_error_errno(errno, "Failed to stat %s: %m", m->source);
 500
 501         where = prefix_roota(dest, m->destination);
 502
 503         if (stat(where, &dest_st) >= 0) {
 504                 if (S_ISDIR(source_st.st_mode) && !S_ISDIR(dest_st.st_mode)) {
 505                         log_error("Cannot bind mount directory %s on file %s.", m->source, where);
 506                         return -EINVAL;
 507                 }
 508
 509                 if (!S_ISDIR(source_st.st_mode) && S_ISDIR(dest_st.st_mode)) {
 510                         log_error("Cannot bind mount file %s on directory %s.", m->source, where);
 511                         return -EINVAL;
 512                 }
 513
 514         } else if (errno == ENOENT) {
 515                 r = mkdir_parents_label(where, 0755);
 516                 if (r < 0)
 517                         return log_error_errno(r, "Failed to make parents of %s: %m", where);
 518
 519                 /* Create the mount point. Any non-directory file can be
 520                 * mounted on any non-directory file (regular, fifo, socket,
 521                 * char, block).
 522                 */
 523                 if (S_ISDIR(source_st.st_mode))
 524                         r = mkdir_label(where, 0755);
 525                 else
 526                         r = touch(where);
 527                 if (r < 0)
 528                         return log_error_errno(r, "Failed to create mount point %s: %m", where);
 529
 530         } else
 531                 return log_error_errno(errno, "Failed to stat %s: %m", where);
 532
 533         r = mount_verbose(LOG_ERR, m->source, where, NULL, mount_flags, mount_opts);
 534         if (r < 0)
 535                 return r;
 536
 537         if (m->read_only) {
 538                 r = bind_remount_recursive(where, true, NULL);
 539                 if (r < 0)
 540                         return log_error_errno(r, "Read-only bind mount failed: %m");
 541         }
 542
 543         return 0;
 544 }
 545
 546 static int mount_tmpfs(
 547                 const char *dest,
 548                 CustomMount *m,
 549                 bool userns, uid_t uid_shift, uid_t uid_range,
 550                 const char *selinux_apifs_context) {
 551
 552         const char *where, *options;
 553         _cleanup_free_ char *buf = NULL;
 554         int r;
 555
 556         assert(dest);
 557         assert(m);
 558
 559         where = prefix_roota(dest, m->destination);
 560
 561         r = mkdir_p_label(where, 0755);
 562         if (r < 0 && r != -EEXIST)
 563                 return log_error_errno(r, "Creating mount point for tmpfs %s failed: %m", where);
 564
 565         r = tmpfs_patch_options(m->options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
 566         if (r < 0)
 567                 return log_oom();
 568         options = r > 0 ? buf : m->options;
 569
 570         return mount_verbose(LOG_ERR, "tmpfs", where, "tmpfs", MS_NODEV|MS_STRICTATIME, options);
 571 }
 572
 573 static char *joined_and_escaped_lower_dirs(char * const *lower) {
 574         _cleanup_strv_free_ char **sv = NULL;
 575
 576         sv = strv_copy(lower);
 577         if (!sv)
 578                 return NULL;
 579
 580         strv_reverse(sv);
 581
 582         if (!strv_shell_escape(sv, ",:"))
 583                 return NULL;
 584
 585         return strv_join(sv, ":");
 586 }
 587
 588 static int mount_overlay(const char *dest, CustomMount *m) {
 589         _cleanup_free_ char *lower = NULL;
 590         const char *where, *options;
 591         int r;
 592
 593         assert(dest);
 594         assert(m);
 595
 596         where = prefix_roota(dest, m->destination);
 597
 598         r = mkdir_label(where, 0755);
 599         if (r < 0 && r != -EEXIST)
 600                 return log_error_errno(r, "Creating mount point for overlay %s failed: %m", where);
 601
 602         (void) mkdir_p_label(m->source, 0755);
 603
 604         lower = joined_and_escaped_lower_dirs(m->lower);
 605         if (!lower)
 606                 return log_oom();
 607
 608         if (m->read_only) {
 609                 _cleanup_free_ char *escaped_source = NULL;
 610
 611                 escaped_source = shell_escape(m->source, ",:");
 612                 if (!escaped_source)
 613                         return log_oom();
 614
 615                 options = strjoina("lowerdir=", escaped_source, ":", lower);
 616         } else {
 617                 _cleanup_free_ char *escaped_source = NULL, *escaped_work_dir = NULL;
 618
 619                 assert(m->work_dir);
 620                 (void) mkdir_label(m->work_dir, 0700);
 621
 622                 escaped_source = shell_escape(m->source, ",:");
 623                 if (!escaped_source)
 624                         return log_oom();
 625                 escaped_work_dir = shell_escape(m->work_dir, ",:");
 626                 if (!escaped_work_dir)
 627                         return log_oom();
 628
 629                 options = strjoina("lowerdir=", lower, ",upperdir=", escaped_source, ",workdir=", escaped_work_dir);
 630         }
 631
 632         return mount_verbose(LOG_ERR, "overlay", where, "overlay", m->read_only ? MS_RDONLY : 0, options);
 633 }
 634
 635 int mount_custom(
 636                 const char *dest,
 637                 CustomMount *mounts, unsigned n,
 638                 bool userns, uid_t uid_shift, uid_t uid_range,
 639                 const char *selinux_apifs_context) {
 640
 641         unsigned i;
 642         int r;
 643
 644         assert(dest);
 645
 646         for (i = 0; i < n; i++) {
 647                 CustomMount *m = mounts + i;
 648
 649                 switch (m->type) {
 650
 651                 case CUSTOM_MOUNT_BIND:
 652                         r = mount_bind(dest, m);
 653                         break;
 654
 655                 case CUSTOM_MOUNT_TMPFS:
 656                         r = mount_tmpfs(dest, m, userns, uid_shift, uid_range, selinux_apifs_context);
 657                         break;
 658
 659                 case CUSTOM_MOUNT_OVERLAY:
 660                         r = mount_overlay(dest, m);
 661                         break;
 662
 663                 default:
 664                         assert_not_reached("Unknown custom mount type");
 665                 }
 666
 667                 if (r < 0)
 668                         return r;
 669         }
 670
 671         return 0;
 672 }
 673
 674 /* Retrieve existing subsystems. This function is called in a new cgroup
 675  * namespace.
 676  */
 677 static int get_controllers(Set *subsystems) {
 678         _cleanup_fclose_ FILE *f = NULL;
 679         char line[LINE_MAX];
 680
 681         assert(subsystems);
 682
 683         f = fopen("/proc/self/cgroup", "re");
 684         if (!f)
 685                 return errno == ENOENT ? -ESRCH : -errno;
 686
 687         FOREACH_LINE(line, f, return -errno) {
 688                 int r;
 689                 char *e, *l, *p;
 690
 691                 l = strchr(line, ':');
 692                 if (!l)
 693                         continue;
 694
 695                 l++;
 696                 e = strchr(l, ':');
 697                 if (!e)
 698                         continue;
 699
 700                 *e = 0;
 701
 702                 if (STR_IN_SET(l, "", "name=systemd"))
 703                         continue;
 704
 705                 p = strdup(l);
 706                 if (!p)
 707                         return -ENOMEM;
 708
 709                 r = set_consume(subsystems, p);
 710                 if (r < 0)
 711                         return r;
 712         }
 713
 714         return 0;
 715 }
 716
 717 static int mount_legacy_cgroup_hierarchy(const char *dest, const char *controller, const char *hierarchy,
 718                                          CGroupUnified unified_requested, bool read_only) {
 719         const char *to, *fstype, *opts;
 720         int r;
 721
 722         to = strjoina(strempty(dest), "/sys/fs/cgroup/", hierarchy);
 723
 724         r = path_is_mount_point(to, 0);
 725         if (r < 0 && r != -ENOENT)
 726                 return log_error_errno(r, "Failed to determine if %s is mounted already: %m", to);
 727         if (r > 0)
 728                 return 0;
 729
 730         mkdir_p(to, 0755);
 731
 732         /* The superblock mount options of the mount point need to be
 733          * identical to the hosts', and hence writable... */
 734         if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
 735                 if (unified_requested >= CGROUP_UNIFIED_SYSTEMD) {
 736                         fstype = "cgroup2";
 737                         opts = NULL;
 738                 } else {
 739                         fstype = "cgroup";
 740                         opts = "none,name=systemd,xattr";
 741                 }
 742         } else {
 743                 fstype = "cgroup";
 744                 opts = controller;
 745         }
 746
 747         r = mount_verbose(LOG_ERR, "cgroup", to, fstype, MS_NOSUID|MS_NOEXEC|MS_NODEV, opts);
 748         if (r < 0)
 749                 return r;
 750
 751         /* ... hence let's only make the bind mount read-only, not the superblock. */
 752         if (read_only) {
 753                 r = mount_verbose(LOG_ERR, NULL, to, NULL,
 754                                   MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL);
 755                 if (r < 0)
 756                         return r;
 757         }
 758
 759         return 1;
 760 }
 761
 762 /* Mount a legacy cgroup hierarchy when cgroup namespaces are supported. */
 763 static int mount_legacy_cgns_supported(
 764                 CGroupUnified unified_requested, bool userns, uid_t uid_shift,
 765                 uid_t uid_range, const char *selinux_apifs_context) {
 766         _cleanup_set_free_free_ Set *controllers = NULL;
 767         const char *cgroup_root = "/sys/fs/cgroup", *c;
 768         int r;
 769
 770         (void) mkdir_p(cgroup_root, 0755);
 771
 772         /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
 773         r = path_is_mount_point(cgroup_root, AT_SYMLINK_FOLLOW);
 774         if (r < 0)
 775                 return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
 776         if (r == 0) {
 777                 _cleanup_free_ char *options = NULL;
 778
 779                 /* When cgroup namespaces are enabled and user namespaces are
 780                  * used then the mount of the cgroupfs is done *inside* the new
 781                  * user namespace. We're root in the new user namespace and the
 782                  * kernel will happily translate our uid/gid to the correct
 783                  * uid/gid as seen from e.g. /proc/1/mountinfo. So we simply
 784                  * pass uid 0 and not uid_shift to tmpfs_patch_options().
 785                  */
 786                 r = tmpfs_patch_options("mode=755", userns, 0, uid_range, true, selinux_apifs_context, &options);
 787                 if (r < 0)
 788                         return log_oom();
 789
 790                 r = mount_verbose(LOG_ERR, "tmpfs", cgroup_root, "tmpfs",
 791                                   MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options);
 792                 if (r < 0)
 793                         return r;
 794         }
 795
 796         if (cg_all_unified() > 0)
 797                 goto skip_controllers;
 798
 799         controllers = set_new(&string_hash_ops);
 800         if (!controllers)
 801                 return log_oom();
 802
 803         r = get_controllers(controllers);
 804         if (r < 0)
 805                 return log_error_errno(r, "Failed to determine cgroup controllers: %m");
 806
 807         for (;;) {
 808                 _cleanup_free_ const char *controller = NULL;
 809
 810                 controller = set_steal_first(controllers);
 811                 if (!controller)
 812                         break;
 813
 814                 r = mount_legacy_cgroup_hierarchy("", controller, controller, unified_requested, !userns);
 815                 if (r < 0)
 816                         return r;
 817
 818                 /* When multiple hierarchies are co-mounted, make their
 819                  * constituting individual hierarchies a symlink to the
 820                  * co-mount.
 821                  */
 822                 c = controller;
 823                 for (;;) {
 824                         _cleanup_free_ char *target = NULL, *tok = NULL;
 825
 826                         r = extract_first_word(&c, &tok, ",", 0);
 827                         if (r < 0)
 828                                 return log_error_errno(r, "Failed to extract co-mounted cgroup controller: %m");
 829                         if (r == 0)
 830                                 break;
 831
 832                         target = prefix_root("/sys/fs/cgroup", tok);
 833                         if (!target)
 834                                 return log_oom();
 835
 836                         if (streq(controller, tok))
 837                                 break;
 838
 839                         r = symlink_idempotent(controller, target);
 840                         if (r == -EINVAL)
 841                                 return log_error_errno(r, "Invalid existing symlink for combined hierarchy: %m");
 842                         if (r < 0)
 843                                 return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m");
 844                 }
 845         }
 846
 847 skip_controllers:
 848         r = mount_legacy_cgroup_hierarchy("", SYSTEMD_CGROUP_CONTROLLER, "systemd", unified_requested, false);
 849         if (r < 0)
 850                 return r;
 851
 852         if (!userns)
 853                 return mount_verbose(LOG_ERR, NULL, cgroup_root, NULL,
 854                                      MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
 855
 856         return 0;
 857 }
 858
 859 /* Mount legacy cgroup hierarchy when cgroup namespaces are unsupported. */
 860 static int mount_legacy_cgns_unsupported(
 861                 const char *dest,
 862                 CGroupUnified unified_requested, bool userns, uid_t uid_shift, uid_t uid_range,
 863                 const char *selinux_apifs_context) {
 864         _cleanup_set_free_free_ Set *controllers = NULL;
 865         const char *cgroup_root;
 866         int r;
 867
 868         cgroup_root = prefix_roota(dest, "/sys/fs/cgroup");
 869
 870         (void) mkdir_p(cgroup_root, 0755);
 871
 872         /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
 873         r = path_is_mount_point(cgroup_root, AT_SYMLINK_FOLLOW);
 874         if (r < 0)
 875                 return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
 876         if (r == 0) {
 877                 _cleanup_free_ char *options = NULL;
 878
 879                 r = tmpfs_patch_options("mode=755", userns, uid_shift, uid_range, false, selinux_apifs_context, &options);
 880                 if (r < 0)
 881                         return log_oom();
 882
 883                 r = mount_verbose(LOG_ERR, "tmpfs", cgroup_root, "tmpfs",
 884                                   MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options);
 885                 if (r < 0)
 886                         return r;
 887         }
 888
 889         if (cg_all_unified() > 0)
 890                 goto skip_controllers;
 891
 892         controllers = set_new(&string_hash_ops);
 893         if (!controllers)
 894                 return log_oom();
 895
 896         r = cg_kernel_controllers(controllers);
 897         if (r < 0)
 898                 return log_error_errno(r, "Failed to determine cgroup controllers: %m");
 899
 900         for (;;) {
 901                 _cleanup_free_ char *controller = NULL, *origin = NULL, *combined = NULL;
 902
 903                 controller = set_steal_first(controllers);
 904                 if (!controller)
 905                         break;
 906
 907                 origin = prefix_root("/sys/fs/cgroup/", controller);
 908                 if (!origin)
 909                         return log_oom();
 910
 911                 r = readlink_malloc(origin, &combined);
 912                 if (r == -EINVAL) {
 913                         /* Not a symbolic link, but directly a single cgroup hierarchy */
 914
 915                         r = mount_legacy_cgroup_hierarchy(dest, controller, controller, unified_requested, true);
 916                         if (r < 0)
 917                                 return r;
 918
 919                 } else if (r < 0)
 920                         return log_error_errno(r, "Failed to read link %s: %m", origin);
 921                 else {
 922                         _cleanup_free_ char *target = NULL;
 923
 924                         target = prefix_root(dest, origin);
 925                         if (!target)
 926                                 return log_oom();
 927
 928                         /* A symbolic link, a combination of controllers in one hierarchy */
 929
 930                         if (!filename_is_valid(combined)) {
 931                                 log_warning("Ignoring invalid combined hierarchy %s.", combined);
 932                                 continue;
 933                         }
 934
 935                         r = mount_legacy_cgroup_hierarchy(dest, combined, combined, unified_requested, true);
 936                         if (r < 0)
 937                                 return r;
 938
 939                         r = symlink_idempotent(combined, target);
 940                         if (r == -EINVAL)
 941                                 return log_error_errno(r, "Invalid existing symlink for combined hierarchy: %m");
 942                         if (r < 0)
 943                                 return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m");
 944                 }
 945         }
 946
 947 skip_controllers:
 948         r = mount_legacy_cgroup_hierarchy(dest, SYSTEMD_CGROUP_CONTROLLER, "systemd", unified_requested, false);
 949         if (r < 0)
 950                 return r;
 951
 952         return mount_verbose(LOG_ERR, NULL, cgroup_root, NULL,
 953                              MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
 954 }
 955
 956 static int mount_unified_cgroups(const char *dest) {
 957         const char *p;
 958         int r;
 959
 960         assert(dest);
 961
 962         p = prefix_roota(dest, "/sys/fs/cgroup");
 963
 964         (void) mkdir_p(p, 0755);
 965
 966         r = path_is_mount_point(p, AT_SYMLINK_FOLLOW);
 967         if (r < 0)
 968                 return log_error_errno(r, "Failed to determine if %s is mounted already: %m", p);
 969         if (r > 0) {
 970                 p = prefix_roota(dest, "/sys/fs/cgroup/cgroup.procs");
 971                 if (access(p, F_OK) >= 0)
 972                         return 0;
 973                 if (errno != ENOENT)
 974                         return log_error_errno(errno, "Failed to determine if mount point %s contains the unified cgroup hierarchy: %m", p);
 975
 976                 log_error("%s is already mounted but not a unified cgroup hierarchy. Refusing.", p);
 977                 return -EINVAL;
 978         }
 979
 980         return mount_verbose(LOG_ERR, "cgroup", p, "cgroup2", MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
 981 }
 982
 983 int mount_cgroups(
 984                 const char *dest,
 985                 CGroupUnified unified_requested,
 986                 bool userns, uid_t uid_shift, uid_t uid_range,
 987                 const char *selinux_apifs_context,
 988                 bool use_cgns) {
 989
 990         if (unified_requested >= CGROUP_UNIFIED_ALL)
 991                 return mount_unified_cgroups(dest);
 992         else if (use_cgns)
 993                 return mount_legacy_cgns_supported(unified_requested, userns, uid_shift, uid_range, selinux_apifs_context);
 994
 995         return mount_legacy_cgns_unsupported(dest, unified_requested, userns, uid_shift, uid_range, selinux_apifs_context);
 996 }
 997
 998 int mount_systemd_cgroup_writable(
 999                 const char *dest,
1000                 CGroupUnified unified_requested) {
1001
1002         _cleanup_free_ char *own_cgroup_path = NULL;
1003         const char *systemd_root, *systemd_own;
1004         int r;
1005
1006         assert(dest);
1007
1008         r = cg_pid_get_path(NULL, 0, &own_cgroup_path);
1009         if (r < 0)
1010                 return log_error_errno(r, "Failed to determine our own cgroup path: %m");
1011
1012         /* If we are living in the top-level, then there's nothing to do... */
1013         if (path_equal(own_cgroup_path, "/"))
1014                 return 0;
1015
1016         if (unified_requested >= CGROUP_UNIFIED_ALL) {
1017                 systemd_own = strjoina(dest, "/sys/fs/cgroup", own_cgroup_path);
1018                 systemd_root = prefix_roota(dest, "/sys/fs/cgroup");
1019         } else {
1020                 systemd_own = strjoina(dest, "/sys/fs/cgroup/systemd", own_cgroup_path);
1021                 systemd_root = prefix_roota(dest, "/sys/fs/cgroup/systemd");
1022         }
1023
1024         /* Make our own cgroup a (writable) bind mount */
1025         r = mount_verbose(LOG_ERR, systemd_own, systemd_own,  NULL, MS_BIND, NULL);
1026         if (r < 0)
1027                 return r;
1028
1029         /* And then remount the systemd cgroup root read-only */
1030         return mount_verbose(LOG_ERR, NULL, systemd_root, NULL,
1031                              MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL);
1032 }
1033
1034 int setup_volatile_state(
1035                 const char *directory,
1036                 VolatileMode mode,
1037                 bool userns, uid_t uid_shift, uid_t uid_range,
1038                 const char *selinux_apifs_context) {
1039
1040         _cleanup_free_ char *buf = NULL;
1041         const char *p, *options;
1042         int r;
1043
1044         assert(directory);
1045
1046         if (mode != VOLATILE_STATE)
1047                 return 0;
1048
1049         /* --volatile=state means we simply overmount /var
1050            with a tmpfs, and the rest read-only. */
1051
1052         r = bind_remount_recursive(directory, true, NULL);
1053         if (r < 0)
1054                 return log_error_errno(r, "Failed to remount %s read-only: %m", directory);
1055
1056         p = prefix_roota(directory, "/var");
1057         r = mkdir(p, 0755);
1058         if (r < 0 && errno != EEXIST)
1059                 return log_error_errno(errno, "Failed to create %s: %m", directory);
1060
1061         options = "mode=755";
1062         r = tmpfs_patch_options(options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
1063         if (r < 0)
1064                 return log_oom();
1065         if (r > 0)
1066                 options = buf;
1067
1068         return mount_verbose(LOG_ERR, "tmpfs", p, "tmpfs", MS_STRICTATIME, options);
1069 }
1070
1071 int setup_volatile(
1072                 const char *directory,
1073                 VolatileMode mode,
1074                 bool userns, uid_t uid_shift, uid_t uid_range,
1075                 const char *selinux_apifs_context) {
1076
1077         bool tmpfs_mounted = false, bind_mounted = false;
1078         char template[] = "/tmp/nspawn-volatile-XXXXXX";
1079         _cleanup_free_ char *buf = NULL;
1080         const char *f, *t, *options;
1081         int r;
1082
1083         assert(directory);
1084
1085         if (mode != VOLATILE_YES)
1086                 return 0;
1087
1088         /* --volatile=yes means we mount a tmpfs to the root dir, and
1089            the original /usr to use inside it, and that read-only. */
1090
1091         if (!mkdtemp(template))
1092                 return log_error_errno(errno, "Failed to create temporary directory: %m");
1093
1094         options = "mode=755";
1095         r = tmpfs_patch_options(options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
1096         if (r < 0)
1097                 return log_oom();
1098         if (r > 0)
1099                 options = buf;
1100
1101         r = mount_verbose(LOG_ERR, "tmpfs", template, "tmpfs", MS_STRICTATIME, options);
1102         if (r < 0)
1103                 goto fail;
1104
1105         tmpfs_mounted = true;
1106
1107         f = prefix_roota(directory, "/usr");
1108         t = prefix_roota(template, "/usr");
1109
1110         r = mkdir(t, 0755);
1111         if (r < 0 && errno != EEXIST) {
1112                 r = log_error_errno(errno, "Failed to create %s: %m", t);
1113                 goto fail;
1114         }
1115
1116         r = mount_verbose(LOG_ERR, f, t, NULL, MS_BIND|MS_REC, NULL);
1117         if (r < 0)
1118                 goto fail;
1119
1120         bind_mounted = true;
1121
1122         r = bind_remount_recursive(t, true, NULL);
1123         if (r < 0) {
1124                 log_error_errno(r, "Failed to remount %s read-only: %m", t);
1125                 goto fail;
1126         }
1127
1128         r = mount_verbose(LOG_ERR, template, directory, NULL, MS_MOVE, NULL);
1129         if (r < 0)
1130                 goto fail;
1131
1132         (void) rmdir(template);
1133
1134         return 0;
1135
1136 fail:
1137         if (bind_mounted)
1138                 (void) umount_verbose(t);
1139
1140         if (tmpfs_mounted)
1141                 (void) umount_verbose(template);
1142         (void) rmdir(template);
1143         return r;
1144 }
1145
1146 VolatileMode volatile_mode_from_string(const char *s) {
1147         int b;
1148
1149         if (isempty(s))
1150                 return _VOLATILE_MODE_INVALID;
1151
1152         b = parse_boolean(s);
1153         if (b > 0)
1154                 return VOLATILE_YES;
1155         if (b == 0)
1156                 return VOLATILE_NO;
1157
1158         if (streq(s, "state"))
1159                 return VOLATILE_STATE;
1160
1161         return _VOLATILE_MODE_INVALID;
1162 }