src/shared/machine-image.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2 /***
   3   This file is part of systemd.
   4
   5   Copyright 2013 Lennart Poettering
   6 ***/
   7
   8 #include <dirent.h>
   9 #include <errno.h>
  10 #include <fcntl.h>
  11 #include <stdio.h>
  12 #include <stdlib.h>
  13 #include <string.h>
  14 #include <sys/file.h>
  15 #include <sys/stat.h>
  16 #include <unistd.h>
  17 #include <linux/fs.h>
  18
  19 #include "alloc-util.h"
  20 #include "btrfs-util.h"
  21 #include "chattr-util.h"
  22 #include "copy.h"
  23 #include "dirent-util.h"
  24 #include "dissect-image.h"
  25 #include "env-util.h"
  26 #include "fd-util.h"
  27 #include "fileio.h"
  28 #include "fs-util.h"
  29 #include "hashmap.h"
  30 #include "hostname-util.h"
  31 #include "id128-util.h"
  32 #include "lockfile-util.h"
  33 #include "log.h"
  34 #include "loop-util.h"
  35 #include "machine-image.h"
  36 #include "macro.h"
  37 #include "mkdir.h"
  38 #include "os-util.h"
  39 #include "path-util.h"
  40 #include "rm-rf.h"
  41 #include "string-table.h"
  42 #include "string-util.h"
  43 #include "strv.h"
  44 #include "time-util.h"
  45 #include "utf8.h"
  46 #include "util.h"
  47 #include "xattr-util.h"
  48
  49 static const char* const image_search_path[_IMAGE_CLASS_MAX] = {
  50         [IMAGE_MACHINE] =  "/var/lib/machines\0"
  51                            "/var/lib/container\0" /* legacy */
  52                            "/usr/local/lib/machines\0"
  53                            "/usr/lib/machines\0",
  54
  55         [IMAGE_PORTABLE] = "/var/lib/portables\0"
  56                            "/usr/local/lib/portables\0"
  57                            "/usr/lib/portables\0",
  58 };
  59
  60 Image *image_unref(Image *i) {
  61         if (!i)
  62                 return NULL;
  63
  64         assert(i->n_ref > 0);
  65         i->n_ref--;
  66
  67         if (i->n_ref > 0)
  68                 return NULL;
  69
  70         free(i->name);
  71         free(i->path);
  72
  73         free(i->hostname);
  74         strv_free(i->machine_info);
  75         strv_free(i->os_release);
  76
  77         return mfree(i);
  78 }
  79
  80 Image *image_ref(Image *i) {
  81         if (!i)
  82                 return NULL;
  83
  84         assert(i->n_ref > 0);
  85         i->n_ref++;
  86
  87         return i;
  88 }
  89
  90 static char **image_settings_path(Image *image) {
  91         _cleanup_strv_free_ char **l = NULL;
  92         const char *fn, *s;
  93         unsigned i = 0;
  94
  95         assert(image);
  96
  97         l = new0(char*, 4);
  98         if (!l)
  99                 return NULL;
 100
 101         fn = strjoina(image->name, ".nspawn");
 102
 103         FOREACH_STRING(s, "/etc/systemd/nspawn/", "/run/systemd/nspawn/") {
 104                 l[i] = strappend(s, fn);
 105                 if (!l[i])
 106                         return NULL;
 107
 108                 i++;
 109         }
 110
 111         l[i] = file_in_same_dir(image->path, fn);
 112         if (!l[i])
 113                 return NULL;
 114
 115         return TAKE_PTR(l);
 116 }
 117
 118 static char *image_roothash_path(Image *image) {
 119         const char *fn;
 120
 121         assert(image);
 122
 123         fn = strjoina(image->name, ".roothash");
 124
 125         return file_in_same_dir(image->path, fn);
 126 }
 127
 128 static int image_new(
 129                 ImageType t,
 130                 const char *pretty,
 131                 const char *path,
 132                 const char *filename,
 133                 bool read_only,
 134                 usec_t crtime,
 135                 usec_t mtime,
 136                 Image **ret) {
 137
 138         _cleanup_(image_unrefp) Image *i = NULL;
 139
 140         assert(t >= 0);
 141         assert(t < _IMAGE_TYPE_MAX);
 142         assert(pretty);
 143         assert(filename);
 144         assert(ret);
 145
 146         i = new0(Image, 1);
 147         if (!i)
 148                 return -ENOMEM;
 149
 150         i->n_ref = 1;
 151         i->type = t;
 152         i->read_only = read_only;
 153         i->crtime = crtime;
 154         i->mtime = mtime;
 155         i->usage = i->usage_exclusive = (uint64_t) -1;
 156         i->limit = i->limit_exclusive = (uint64_t) -1;
 157
 158         i->name = strdup(pretty);
 159         if (!i->name)
 160                 return -ENOMEM;
 161
 162         if (path)
 163                 i->path = strjoin(path, "/", filename);
 164         else
 165                 i->path = strdup(filename);
 166         if (!i->path)
 167                 return -ENOMEM;
 168
 169         path_kill_slashes(i->path);
 170
 171         *ret = TAKE_PTR(i);
 172
 173         return 0;
 174 }
 175
 176 static int extract_pretty(const char *path, const char *suffix, char **ret) {
 177         _cleanup_free_ char *name = NULL;
 178         const char *p;
 179         size_t n;
 180
 181         assert(path);
 182         assert(ret);
 183
 184         p = last_path_component(path);
 185         n = strcspn(p, "/");
 186
 187         name = strndup(p, n);
 188         if (!name)
 189                 return -ENOMEM;
 190
 191         if (suffix) {
 192                 char *e;
 193
 194                 e = endswith(name, suffix);
 195                 if (!e)
 196                         return -EINVAL;
 197
 198                 *e = 0;
 199         }
 200
 201         if (!image_name_is_valid(name))
 202                 return -EINVAL;
 203
 204         *ret = TAKE_PTR(name);
 205         return 0;
 206 }
 207
 208 static int image_make(
 209                 const char *pretty,
 210                 int dfd,
 211                 const char *path,
 212                 const char *filename,
 213                 const struct stat *st,
 214                 Image **ret) {
 215
 216         _cleanup_free_ char *pretty_buffer = NULL;
 217         struct stat stbuf;
 218         bool read_only;
 219         int r;
 220
 221         assert(dfd >= 0 || dfd == AT_FDCWD);
 222         assert(filename);
 223
 224         /* We explicitly *do* follow symlinks here, since we want to allow symlinking trees, raw files and block
 225          * devices into /var/lib/machines/, and treat them normally.
 226          *
 227          * This function returns -ENOENT if we can't find the image after all, and -EMEDIUMTYPE if it's not a file we
 228          * recognize. */
 229
 230         if (!st) {
 231                 if (fstatat(dfd, filename, &stbuf, 0) < 0)
 232                         return -errno;
 233
 234                 st = &stbuf;
 235         }
 236
 237         read_only =
 238                 (path && path_startswith(path, "/usr")) ||
 239                 (faccessat(dfd, filename, W_OK, AT_EACCESS) < 0 && errno == EROFS);
 240
 241         if (S_ISDIR(st->st_mode)) {
 242                 _cleanup_close_ int fd = -1;
 243                 unsigned file_attr = 0;
 244
 245                 if (!ret)
 246                         return 0;
 247
 248                 if (!pretty) {
 249                         r = extract_pretty(filename, NULL, &pretty_buffer);
 250                         if (r < 0)
 251                                 return r;
 252
 253                         pretty = pretty_buffer;
 254                 }
 255
 256                 fd = openat(dfd, filename, O_CLOEXEC|O_NOCTTY|O_DIRECTORY);
 257                 if (fd < 0)
 258                         return -errno;
 259
 260                 /* btrfs subvolumes have inode 256 */
 261                 if (st->st_ino == 256) {
 262
 263                         r = btrfs_is_filesystem(fd);
 264                         if (r < 0)
 265                                 return r;
 266                         if (r) {
 267                                 BtrfsSubvolInfo info;
 268
 269                                 /* It's a btrfs subvolume */
 270
 271                                 r = btrfs_subvol_get_info_fd(fd, 0, &info);
 272                                 if (r < 0)
 273                                         return r;
 274
 275                                 r = image_new(IMAGE_SUBVOLUME,
 276                                               pretty,
 277                                               path,
 278                                               filename,
 279                                               info.read_only || read_only,
 280                                               info.otime,
 281                                               0,
 282                                               ret);
 283                                 if (r < 0)
 284                                         return r;
 285
 286                                 if (btrfs_quota_scan_ongoing(fd) == 0) {
 287                                         BtrfsQuotaInfo quota;
 288
 289                                         r = btrfs_subvol_get_subtree_quota_fd(fd, 0, &quota);
 290                                         if (r >= 0) {
 291                                                 (*ret)->usage = quota.referenced;
 292                                                 (*ret)->usage_exclusive = quota.exclusive;
 293
 294                                                 (*ret)->limit = quota.referenced_max;
 295                                                 (*ret)->limit_exclusive = quota.exclusive_max;
 296                                         }
 297                                 }
 298
 299                                 return 0;
 300                         }
 301                 }
 302
 303                 /* If the IMMUTABLE bit is set, we consider the
 304                  * directory read-only. Since the ioctl is not
 305                  * supported everywhere we ignore failures. */
 306                 (void) read_attr_fd(fd, &file_attr);
 307
 308                 /* It's just a normal directory. */
 309                 r = image_new(IMAGE_DIRECTORY,
 310                               pretty,
 311                               path,
 312                               filename,
 313                               read_only || (file_attr & FS_IMMUTABLE_FL),
 314                               0,
 315                               0,
 316                               ret);
 317                 if (r < 0)
 318                         return r;
 319
 320                 return 0;
 321
 322         } else if (S_ISREG(st->st_mode) && endswith(filename, ".raw")) {
 323                 usec_t crtime = 0;
 324
 325                 /* It's a RAW disk image */
 326
 327                 if (!ret)
 328                         return 0;
 329
 330                 (void) fd_getcrtime_at(dfd, filename, &crtime, 0);
 331
 332                 if (!pretty) {
 333                         r = extract_pretty(filename, ".raw", &pretty_buffer);
 334                         if (r < 0)
 335                                 return r;
 336
 337                         pretty = pretty_buffer;
 338                 }
 339
 340                 r = image_new(IMAGE_RAW,
 341                               pretty,
 342                               path,
 343                               filename,
 344                               !(st->st_mode & 0222) || read_only,
 345                               crtime,
 346                               timespec_load(&st->st_mtim),
 347                               ret);
 348                 if (r < 0)
 349                         return r;
 350
 351                 (*ret)->usage = (*ret)->usage_exclusive = st->st_blocks * 512;
 352                 (*ret)->limit = (*ret)->limit_exclusive = st->st_size;
 353
 354                 return 0;
 355
 356         } else if (S_ISBLK(st->st_mode)) {
 357                 _cleanup_close_ int block_fd = -1;
 358                 uint64_t size = UINT64_MAX;
 359
 360                 /* A block device */
 361
 362                 if (!ret)
 363                         return 0;
 364
 365                 if (!pretty) {
 366                         r = extract_pretty(filename, NULL, &pretty_buffer);
 367                         if (r < 0)
 368                                 return r;
 369
 370                         pretty = pretty_buffer;
 371                 }
 372
 373                 block_fd = openat(dfd, filename, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY);
 374                 if (block_fd < 0)
 375                         log_debug_errno(errno, "Failed to open block device %s/%s, ignoring: %m", path, filename);
 376                 else {
 377                         /* Refresh stat data after opening the node */
 378                         if (fstat(block_fd, &stbuf) < 0)
 379                                 return -errno;
 380                         st = &stbuf;
 381
 382                         if (!S_ISBLK(st->st_mode)) /* Verify that what we opened is actually what we think it is */
 383                                 return -ENOTTY;
 384
 385                         if (!read_only) {
 386                                 int state = 0;
 387
 388                                 if (ioctl(block_fd, BLKROGET, &state) < 0)
 389                                         log_debug_errno(errno, "Failed to issue BLKROGET on device %s/%s, ignoring: %m", path, filename);
 390                                 else if (state)
 391                                         read_only = true;
 392                         }
 393
 394                         if (ioctl(block_fd, BLKGETSIZE64, &size) < 0)
 395                                 log_debug_errno(errno, "Failed to issue BLKGETSIZE64 on device %s/%s, ignoring: %m", path, filename);
 396
 397                         block_fd = safe_close(block_fd);
 398                 }
 399
 400                 r = image_new(IMAGE_BLOCK,
 401                               pretty,
 402                               path,
 403                               filename,
 404                               !(st->st_mode & 0222) || read_only,
 405                               0,
 406                               0,
 407                               ret);
 408                 if (r < 0)
 409                         return r;
 410
 411                 if (size != 0 && size != UINT64_MAX)
 412                         (*ret)->usage = (*ret)->usage_exclusive = (*ret)->limit = (*ret)->limit_exclusive = size;
 413
 414                 return 0;
 415         }
 416
 417         return -EMEDIUMTYPE;
 418 }
 419
 420 int image_find(ImageClass class, const char *name, Image **ret) {
 421         const char *path;
 422         int r;
 423
 424         assert(class >= 0);
 425         assert(class < _IMAGE_CLASS_MAX);
 426         assert(name);
 427
 428         /* There are no images with invalid names */
 429         if (!image_name_is_valid(name))
 430                 return -ENOENT;
 431
 432         NULSTR_FOREACH(path, image_search_path[class]) {
 433                 _cleanup_closedir_ DIR *d = NULL;
 434                 struct stat st;
 435
 436                 d = opendir(path);
 437                 if (!d) {
 438                         if (errno == ENOENT)
 439                                 continue;
 440
 441                         return -errno;
 442                 }
 443
 444                 /* As mentioned above, we follow symlinks on this fstatat(), because we want to permit people to
 445                  * symlink block devices into the search path */
 446                 if (fstatat(dirfd(d), name, &st, 0) < 0) {
 447                         _cleanup_free_ char *raw = NULL;
 448
 449                         if (errno != ENOENT)
 450                                 return -errno;
 451
 452                         raw = strappend(name, ".raw");
 453                         if (!raw)
 454                                 return -ENOMEM;
 455
 456                         if (fstatat(dirfd(d), raw, &st, 0) < 0) {
 457
 458                                 if (errno == ENOENT)
 459                                         continue;
 460
 461                                 return -errno;
 462                         }
 463
 464                         if (!S_ISREG(st.st_mode))
 465                                 continue;
 466
 467                         r = image_make(name, dirfd(d), path, raw, &st, ret);
 468
 469                 } else {
 470                         if (!S_ISDIR(st.st_mode) && !S_ISBLK(st.st_mode))
 471                                 continue;
 472
 473                         r = image_make(name, dirfd(d), path, name, &st, ret);
 474                 }
 475                 if (IN_SET(r, -ENOENT, -EMEDIUMTYPE))
 476                         continue;
 477                 if (r < 0)
 478                         return r;
 479
 480                 if (ret)
 481                         (*ret)->discoverable = true;
 482
 483                 return 1;
 484         }
 485
 486         if (class == IMAGE_MACHINE && streq(name, ".host")) {
 487                 r = image_make(".host", AT_FDCWD, NULL, "/", NULL, ret);
 488                 if (r < 0)
 489                         return r;
 490
 491                 if (ret)
 492                         (*ret)->discoverable = true;
 493
 494                 return r;
 495         }
 496
 497         return -ENOENT;
 498 };
 499
 500 int image_from_path(const char *path, Image **ret) {
 501
 502         /* Note that we don't set the 'discoverable' field of the returned object, because we don't check here whether
 503          * the image is in the image search path. And if it is we don't know if the path we used is actually not
 504          * overriden by another, different image earlier in the search path */
 505
 506         if (path_equal(path, "/"))
 507                 return image_make(".host", AT_FDCWD, NULL, "/", NULL, ret);
 508
 509         return image_make(NULL, AT_FDCWD, NULL, path, NULL, ret);
 510 }
 511
 512 int image_find_harder(ImageClass class, const char *name_or_path, Image **ret) {
 513         if (image_name_is_valid(name_or_path))
 514                 return image_find(class, name_or_path, ret);
 515
 516         return image_from_path(name_or_path, ret);
 517 }
 518
 519 int image_discover(ImageClass class, Hashmap *h) {
 520         const char *path;
 521         int r;
 522
 523         assert(class >= 0);
 524         assert(class < _IMAGE_CLASS_MAX);
 525         assert(h);
 526
 527         NULSTR_FOREACH(path, image_search_path[class]) {
 528                 _cleanup_closedir_ DIR *d = NULL;
 529                 struct dirent *de;
 530
 531                 d = opendir(path);
 532                 if (!d) {
 533                         if (errno == ENOENT)
 534                                 continue;
 535
 536                         return -errno;
 537                 }
 538
 539                 FOREACH_DIRENT_ALL(de, d, return -errno) {
 540                         _cleanup_(image_unrefp) Image *image = NULL;
 541                         _cleanup_free_ char *truncated = NULL;
 542                         const char *pretty;
 543                         struct stat st;
 544
 545                         if (dot_or_dot_dot(de->d_name))
 546                                 continue;
 547
 548                         /* As mentioned above, we follow symlinks on this fstatat(), because we want to permit people
 549                          * to symlink block devices into the search path */
 550                         if (fstatat(dirfd(d), de->d_name, &st, 0) < 0) {
 551                                 if (errno == ENOENT)
 552                                         continue;
 553
 554                                 return -errno;
 555                         }
 556
 557                         if (S_ISREG(st.st_mode)) {
 558                                 const char *e;
 559
 560                                 e = endswith(de->d_name, ".raw");
 561                                 if (!e)
 562                                         continue;
 563
 564                                 truncated = strndup(de->d_name, e - de->d_name);
 565                                 if (!truncated)
 566                                         return -ENOMEM;
 567
 568                                 pretty = truncated;
 569                         } else if (S_ISDIR(st.st_mode) || S_ISBLK(st.st_mode))
 570                                 pretty = de->d_name;
 571                         else
 572                                 continue;
 573
 574                         if (!image_name_is_valid(pretty))
 575                                 continue;
 576
 577                         if (hashmap_contains(h, pretty))
 578                                 continue;
 579
 580                         r = image_make(pretty, dirfd(d), path, de->d_name, &st, &image);
 581                         if (IN_SET(r, -ENOENT, -EMEDIUMTYPE))
 582                                 continue;
 583                         if (r < 0)
 584                                 return r;
 585
 586                         image->discoverable = true;
 587
 588                         r = hashmap_put(h, image->name, image);
 589                         if (r < 0)
 590                                 return r;
 591
 592                         image = NULL;
 593                 }
 594         }
 595
 596         if (class == IMAGE_MACHINE && !hashmap_contains(h, ".host")) {
 597                 _cleanup_(image_unrefp) Image *image = NULL;
 598
 599                 r = image_make(".host", AT_FDCWD, NULL, "/", NULL, &image);
 600                 if (r < 0)
 601                         return r;
 602
 603                 image->discoverable = true;
 604
 605                 r = hashmap_put(h, image->name, image);
 606                 if (r < 0)
 607                         return r;
 608
 609                 image = NULL;
 610         }
 611
 612         return 0;
 613 }
 614
 615 int image_remove(Image *i) {
 616         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
 617         _cleanup_strv_free_ char **settings = NULL;
 618         _cleanup_free_ char *roothash = NULL;
 619         char **j;
 620         int r;
 621
 622         assert(i);
 623
 624         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 625                 return -EROFS;
 626
 627         settings = image_settings_path(i);
 628         if (!settings)
 629                 return -ENOMEM;
 630
 631         roothash = image_roothash_path(i);
 632         if (!roothash)
 633                 return -ENOMEM;
 634
 635         /* Make sure we don't interfere with a running nspawn */
 636         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 637         if (r < 0)
 638                 return r;
 639
 640         switch (i->type) {
 641
 642         case IMAGE_SUBVOLUME:
 643
 644                 /* Let's unlink first, maybe it is a symlink? If that works we are happy. Otherwise, let's get out the
 645                  * big guns */
 646                 if (unlink(i->path) < 0) {
 647                         r = btrfs_subvol_remove(i->path, BTRFS_REMOVE_RECURSIVE|BTRFS_REMOVE_QUOTA);
 648                         if (r < 0)
 649                                 return r;
 650                 }
 651
 652                 break;
 653
 654         case IMAGE_DIRECTORY:
 655                 /* Allow deletion of read-only directories */
 656                 (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL);
 657                 r = rm_rf(i->path, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
 658                 if (r < 0)
 659                         return r;
 660
 661                 break;
 662
 663         case IMAGE_BLOCK:
 664
 665                 /* If this is inside of /dev, then it's a real block device, hence let's not touch the device node
 666                  * itself (but let's remove the stuff stored alongside it). If it's anywhere else, let's try to unlink
 667                  * the thing (it's most likely a symlink after all). */
 668
 669                 if (path_startswith(i->path, "/dev"))
 670                         break;
 671
 672                 _fallthrough_;
 673         case IMAGE_RAW:
 674                 if (unlink(i->path) < 0)
 675                         return -errno;
 676                 break;
 677
 678         default:
 679                 return -EOPNOTSUPP;
 680         }
 681
 682         STRV_FOREACH(j, settings) {
 683                 if (unlink(*j) < 0 && errno != ENOENT)
 684                         log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", *j);
 685         }
 686
 687         if (unlink(roothash) < 0 && errno != ENOENT)
 688                 log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", roothash);
 689
 690         return 0;
 691 }
 692
 693 static int rename_auxiliary_file(const char *path, const char *new_name, const char *suffix) {
 694         _cleanup_free_ char *rs = NULL;
 695         const char *fn;
 696
 697         fn = strjoina(new_name, suffix);
 698
 699         rs = file_in_same_dir(path, fn);
 700         if (!rs)
 701                 return -ENOMEM;
 702
 703         return rename_noreplace(AT_FDCWD, path, AT_FDCWD, rs);
 704 }
 705
 706 int image_rename(Image *i, const char *new_name) {
 707         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT, name_lock = LOCK_FILE_INIT;
 708         _cleanup_free_ char *new_path = NULL, *nn = NULL, *roothash = NULL;
 709         _cleanup_strv_free_ char **settings = NULL;
 710         unsigned file_attr = 0;
 711         char **j;
 712         int r;
 713
 714         assert(i);
 715
 716         if (!image_name_is_valid(new_name))
 717                 return -EINVAL;
 718
 719         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 720                 return -EROFS;
 721
 722         settings = image_settings_path(i);
 723         if (!settings)
 724                 return -ENOMEM;
 725
 726         roothash = image_roothash_path(i);
 727         if (!roothash)
 728                 return -ENOMEM;
 729
 730         /* Make sure we don't interfere with a running nspawn */
 731         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 732         if (r < 0)
 733                 return r;
 734
 735         /* Make sure nobody takes the new name, between the time we
 736          * checked it is currently unused in all search paths, and the
 737          * time we take possession of it */
 738         r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
 739         if (r < 0)
 740                 return r;
 741
 742         r = image_find(IMAGE_MACHINE, new_name, NULL);
 743         if (r >= 0)
 744                 return -EEXIST;
 745         if (r != -ENOENT)
 746                 return r;
 747
 748         switch (i->type) {
 749
 750         case IMAGE_DIRECTORY:
 751                 /* Turn of the immutable bit while we rename the image, so that we can rename it */
 752                 (void) read_attr_path(i->path, &file_attr);
 753
 754                 if (file_attr & FS_IMMUTABLE_FL)
 755                         (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL);
 756
 757                 _fallthrough_;
 758         case IMAGE_SUBVOLUME:
 759                 new_path = file_in_same_dir(i->path, new_name);
 760                 break;
 761
 762         case IMAGE_BLOCK:
 763
 764                 /* Refuse renaming raw block devices in /dev, the names are picked by udev after all. */
 765                 if (path_startswith(i->path, "/dev"))
 766                         return -EROFS;
 767
 768                 new_path = file_in_same_dir(i->path, new_name);
 769                 break;
 770
 771         case IMAGE_RAW: {
 772                 const char *fn;
 773
 774                 fn = strjoina(new_name, ".raw");
 775                 new_path = file_in_same_dir(i->path, fn);
 776                 break;
 777         }
 778
 779         default:
 780                 return -EOPNOTSUPP;
 781         }
 782
 783         if (!new_path)
 784                 return -ENOMEM;
 785
 786         nn = strdup(new_name);
 787         if (!nn)
 788                 return -ENOMEM;
 789
 790         r = rename_noreplace(AT_FDCWD, i->path, AT_FDCWD, new_path);
 791         if (r < 0)
 792                 return r;
 793
 794         /* Restore the immutable bit, if it was set before */
 795         if (file_attr & FS_IMMUTABLE_FL)
 796                 (void) chattr_path(new_path, FS_IMMUTABLE_FL, FS_IMMUTABLE_FL);
 797
 798         free_and_replace(i->path, new_path);
 799         free_and_replace(i->name, nn);
 800
 801         STRV_FOREACH(j, settings) {
 802                 r = rename_auxiliary_file(*j, new_name, ".nspawn");
 803                 if (r < 0 && r != -ENOENT)
 804                         log_debug_errno(r, "Failed to rename settings file %s, ignoring: %m", *j);
 805         }
 806
 807         r = rename_auxiliary_file(roothash, new_name, ".roothash");
 808         if (r < 0 && r != -ENOENT)
 809                 log_debug_errno(r, "Failed to rename roothash file %s, ignoring: %m", roothash);
 810
 811         return 0;
 812 }
 813
 814 static int clone_auxiliary_file(const char *path, const char *new_name, const char *suffix) {
 815         _cleanup_free_ char *rs = NULL;
 816         const char *fn;
 817
 818         fn = strjoina(new_name, suffix);
 819
 820         rs = file_in_same_dir(path, fn);
 821         if (!rs)
 822                 return -ENOMEM;
 823
 824         return copy_file_atomic(path, rs, 0664, 0, COPY_REFLINK);
 825 }
 826
 827 int image_clone(Image *i, const char *new_name, bool read_only) {
 828         _cleanup_(release_lock_file) LockFile name_lock = LOCK_FILE_INIT;
 829         _cleanup_strv_free_ char **settings = NULL;
 830         _cleanup_free_ char *roothash = NULL;
 831         const char *new_path;
 832         char **j;
 833         int r;
 834
 835         assert(i);
 836
 837         if (!image_name_is_valid(new_name))
 838                 return -EINVAL;
 839
 840         settings = image_settings_path(i);
 841         if (!settings)
 842                 return -ENOMEM;
 843
 844         roothash = image_roothash_path(i);
 845         if (!roothash)
 846                 return -ENOMEM;
 847
 848         /* Make sure nobody takes the new name, between the time we
 849          * checked it is currently unused in all search paths, and the
 850          * time we take possession of it */
 851         r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
 852         if (r < 0)
 853                 return r;
 854
 855         r = image_find(IMAGE_MACHINE, new_name, NULL);
 856         if (r >= 0)
 857                 return -EEXIST;
 858         if (r != -ENOENT)
 859                 return r;
 860
 861         switch (i->type) {
 862
 863         case IMAGE_SUBVOLUME:
 864         case IMAGE_DIRECTORY:
 865                 /* If we can we'll always try to create a new btrfs subvolume here, even if the source is a plain
 866                  * directory. */
 867
 868                 new_path = strjoina("/var/lib/machines/", new_name);
 869
 870                 r = btrfs_subvol_snapshot(i->path, new_path,
 871                                           (read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
 872                                           BTRFS_SNAPSHOT_FALLBACK_COPY |
 873                                           BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
 874                                           BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE |
 875                                           BTRFS_SNAPSHOT_RECURSIVE |
 876                                           BTRFS_SNAPSHOT_QUOTA);
 877                 if (r >= 0)
 878                         /* Enable "subtree" quotas for the copy, if we didn't copy any quota from the source. */
 879                         (void) btrfs_subvol_auto_qgroup(new_path, 0, true);
 880
 881                 break;
 882
 883         case IMAGE_RAW:
 884                 new_path = strjoina("/var/lib/machines/", new_name, ".raw");
 885
 886                 r = copy_file_atomic(i->path, new_path, read_only ? 0444 : 0644, FS_NOCOW_FL, COPY_REFLINK);
 887                 break;
 888
 889         case IMAGE_BLOCK:
 890         default:
 891                 return -EOPNOTSUPP;
 892         }
 893
 894         if (r < 0)
 895                 return r;
 896
 897         STRV_FOREACH(j, settings) {
 898                 r = clone_auxiliary_file(*j, new_name, ".nspawn");
 899                 if (r < 0 && r != -ENOENT)
 900                         log_debug_errno(r, "Failed to clone settings %s, ignoring: %m", *j);
 901         }
 902
 903         r = clone_auxiliary_file(roothash, new_name, ".roothash");
 904         if (r < 0 && r != -ENOENT)
 905                 log_debug_errno(r, "Failed to clone root hash file %s, ignoring: %m", roothash);
 906
 907         return 0;
 908 }
 909
 910 int image_read_only(Image *i, bool b) {
 911         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
 912         int r;
 913
 914         assert(i);
 915
 916         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 917                 return -EROFS;
 918
 919         /* Make sure we don't interfere with a running nspawn */
 920         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 921         if (r < 0)
 922                 return r;
 923
 924         switch (i->type) {
 925
 926         case IMAGE_SUBVOLUME:
 927
 928                 /* Note that we set the flag only on the top-level
 929                  * subvolume of the image. */
 930
 931                 r = btrfs_subvol_set_read_only(i->path, b);
 932                 if (r < 0)
 933                         return r;
 934
 935                 break;
 936
 937         case IMAGE_DIRECTORY:
 938                 /* For simple directory trees we cannot use the access
 939                    mode of the top-level directory, since it has an
 940                    effect on the container itself.  However, we can
 941                    use the "immutable" flag, to at least make the
 942                    top-level directory read-only. It's not as good as
 943                    a read-only subvolume, but at least something, and
 944                    we can read the value back. */
 945
 946                 r = chattr_path(i->path, b ? FS_IMMUTABLE_FL : 0, FS_IMMUTABLE_FL);
 947                 if (r < 0)
 948                         return r;
 949
 950                 break;
 951
 952         case IMAGE_RAW: {
 953                 struct stat st;
 954
 955                 if (stat(i->path, &st) < 0)
 956                         return -errno;
 957
 958                 if (chmod(i->path, (st.st_mode & 0444) | (b ? 0000 : 0200)) < 0)
 959                         return -errno;
 960
 961                 /* If the images is now read-only, it's a good time to
 962                  * defrag it, given that no write patterns will
 963                  * fragment it again. */
 964                 if (b)
 965                         (void) btrfs_defrag(i->path);
 966                 break;
 967         }
 968
 969         case IMAGE_BLOCK: {
 970                 _cleanup_close_ int fd = -1;
 971                 struct stat st;
 972                 int state = b;
 973
 974                 fd = open(i->path, O_CLOEXEC|O_RDONLY|O_NONBLOCK|O_NOCTTY);
 975                 if (fd < 0)
 976                         return -errno;
 977
 978                 if (fstat(fd, &st) < 0)
 979                         return -errno;
 980                 if (!S_ISBLK(st.st_mode))
 981                         return -ENOTTY;
 982
 983                 if (ioctl(fd, BLKROSET, &state) < 0)
 984                         return -errno;
 985
 986                 break;
 987         }
 988
 989         default:
 990                 return -EOPNOTSUPP;
 991         }
 992
 993         return 0;
 994 }
 995
 996 int image_path_lock(const char *path, int operation, LockFile *global, LockFile *local) {
 997         _cleanup_free_ char *p = NULL;
 998         LockFile t = LOCK_FILE_INIT;
 999         struct stat st;
1000         int r;
1001
1002         assert(path);
1003         assert(global);
1004         assert(local);
1005
1006         /* Locks an image path. This actually creates two locks: one
1007          * "local" one, next to the image path itself, which might be
1008          * shared via NFS. And another "global" one, in /run, that
1009          * uses the device/inode number. This has the benefit that we
1010          * can even lock a tree that is a mount point, correctly. */
1011
1012         if (!path_is_absolute(path))
1013                 return -EINVAL;
1014
1015         if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) {
1016                 *local = *global = (LockFile) LOCK_FILE_INIT;
1017                 return 0;
1018         }
1019
1020         if (path_equal(path, "/"))
1021                 return -EBUSY;
1022
1023         if (stat(path, &st) >= 0) {
1024                 if (S_ISBLK(st.st_mode))
1025                         r = asprintf(&p, "/run/systemd/nspawn/locks/block-%u:%u", major(st.st_rdev), minor(st.st_rdev));
1026                 else if (S_ISDIR(st.st_mode) || S_ISREG(st.st_mode))
1027                         r = asprintf(&p, "/run/systemd/nspawn/locks/inode-%lu:%lu", (unsigned long) st.st_dev, (unsigned long) st.st_ino);
1028                 else
1029                         return -ENOTTY;
1030
1031                 if (r < 0)
1032                         return -ENOMEM;
1033         }
1034
1035         /* For block devices we don't need the "local" lock, as the major/minor lock above should be sufficient, since
1036          * block devices are device local anyway. */
1037         if (!path_startswith(path, "/dev")) {
1038                 r = make_lock_file_for(path, operation, &t);
1039                 if (r < 0) {
1040                         if ((operation & LOCK_SH) && r == -EROFS)
1041                                 log_debug_errno(r, "Failed to create shared lock for '%s', ignoring: %m", path);
1042                         else
1043                                 return r;
1044                 }
1045         }
1046
1047         if (p) {
1048                 mkdir_p("/run/systemd/nspawn/locks", 0700);
1049
1050                 r = make_lock_file(p, operation, global);
1051                 if (r < 0) {
1052                         release_lock_file(&t);
1053                         return r;
1054                 }
1055         } else
1056                 *global = (LockFile) LOCK_FILE_INIT;
1057
1058         *local = t;
1059         return 0;
1060 }
1061
1062 int image_set_limit(Image *i, uint64_t referenced_max) {
1063         assert(i);
1064
1065         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
1066                 return -EROFS;
1067
1068         if (i->type != IMAGE_SUBVOLUME)
1069                 return -EOPNOTSUPP;
1070
1071         /* We set the quota both for the subvolume as well as for the
1072          * subtree. The latter is mostly for historical reasons, since
1073          * we didn't use to have a concept of subtree quota, and hence
1074          * only modified the subvolume quota. */
1075
1076         (void) btrfs_qgroup_set_limit(i->path, 0, referenced_max);
1077         (void) btrfs_subvol_auto_qgroup(i->path, 0, true);
1078         return btrfs_subvol_set_subtree_quota_limit(i->path, 0, referenced_max);
1079 }
1080
1081 int image_read_metadata(Image *i) {
1082         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
1083         int r;
1084
1085         assert(i);
1086
1087         r = image_path_lock(i->path, LOCK_SH|LOCK_NB, &global_lock, &local_lock);
1088         if (r < 0)
1089                 return r;
1090
1091         switch (i->type) {
1092
1093         case IMAGE_SUBVOLUME:
1094         case IMAGE_DIRECTORY: {
1095                 _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL;
1096                 sd_id128_t machine_id = SD_ID128_NULL;
1097                 _cleanup_free_ char *hostname = NULL;
1098                 _cleanup_free_ char *path = NULL;
1099                 _cleanup_fclose_ FILE *f = NULL;
1100
1101                 r = chase_symlinks("/etc/hostname", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path);
1102                 if (r < 0 && r != -ENOENT)
1103                         log_debug_errno(r, "Failed to chase /etc/hostname in image %s: %m", i->name);
1104                 else if (r >= 0) {
1105                         r = read_etc_hostname(path, &hostname);
1106                         if (r < 0)
1107                                 log_debug_errno(errno, "Failed to read /etc/hostname of image %s: %m", i->name);
1108                 }
1109
1110                 path = mfree(path);
1111
1112                 r = chase_symlinks("/etc/machine-id", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path);
1113                 if (r < 0 && r != -ENOENT)
1114                         log_debug_errno(r, "Failed to chase /etc/machine-id in image %s: %m", i->name);
1115                 else if (r >= 0) {
1116                         _cleanup_close_ int fd = -1;
1117
1118                         fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY);
1119                         if (fd < 0)
1120                                 log_debug_errno(errno, "Failed to open %s: %m", path);
1121                         else {
1122                                 r = id128_read_fd(fd, ID128_PLAIN, &machine_id);
1123                                 if (r < 0)
1124                                         log_debug_errno(r, "Image %s contains invalid machine ID.", i->name);
1125                         }
1126                 }
1127
1128                 path = mfree(path);
1129
1130                 r = chase_symlinks("/etc/machine-info", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path);
1131                 if (r < 0 && r != -ENOENT)
1132                         log_debug_errno(r, "Failed to chase /etc/machine-info in image %s: %m", i->name);
1133                 else if (r >= 0) {
1134                         r = load_env_file_pairs(NULL, path, NULL, &machine_info);
1135                         if (r < 0)
1136                                 log_debug_errno(r, "Failed to parse machine-info data of %s: %m", i->name);
1137                 }
1138
1139                 r = load_os_release_pairs(i->path, &os_release);
1140                 if (r < 0)
1141                         log_debug_errno(r, "Failed to read os-release in image, ignoring: %m");
1142
1143                 free_and_replace(i->hostname, hostname);
1144                 i->machine_id = machine_id;
1145                 strv_free_and_replace(i->machine_info, machine_info);
1146                 strv_free_and_replace(i->os_release, os_release);
1147
1148                 break;
1149         }
1150
1151         case IMAGE_RAW:
1152         case IMAGE_BLOCK: {
1153                 _cleanup_(loop_device_unrefp) LoopDevice *d = NULL;
1154                 _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL;
1155
1156                 r = loop_device_make_by_path(i->path, O_RDONLY, &d);
1157                 if (r < 0)
1158                         return r;
1159
1160                 r = dissect_image(d->fd, NULL, 0, DISSECT_IMAGE_REQUIRE_ROOT, &m);
1161                 if (r < 0)
1162                         return r;
1163
1164                 r = dissected_image_acquire_metadata(m);
1165                 if (r < 0)
1166                         return r;
1167
1168                 free_and_replace(i->hostname, m->hostname);
1169                 i->machine_id = m->machine_id;
1170                 strv_free_and_replace(i->machine_info, m->machine_info);
1171                 strv_free_and_replace(i->os_release, m->os_release);
1172
1173                 break;
1174         }
1175
1176         default:
1177                 return -EOPNOTSUPP;
1178         }
1179
1180         i->metadata_valid = true;
1181
1182         return 0;
1183 }
1184
1185 int image_name_lock(const char *name, int operation, LockFile *ret) {
1186         const char *p;
1187
1188         assert(name);
1189         assert(ret);
1190
1191         /* Locks an image name, regardless of the precise path used. */
1192
1193         if (!image_name_is_valid(name))
1194                 return -EINVAL;
1195
1196         if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) {
1197                 *ret = (LockFile) LOCK_FILE_INIT;
1198                 return 0;
1199         }
1200
1201         if (streq(name, ".host"))
1202                 return -EBUSY;
1203
1204         mkdir_p("/run/systemd/nspawn/locks", 0700);
1205         p = strjoina("/run/systemd/nspawn/locks/name-", name);
1206
1207         return make_lock_file(p, operation, ret);
1208 }
1209
1210 bool image_name_is_valid(const char *s) {
1211         if (!filename_is_valid(s))
1212                 return false;
1213
1214         if (string_has_cc(s, NULL))
1215                 return false;
1216
1217         if (!utf8_is_valid(s))
1218                 return false;
1219
1220         /* Temporary files for atomically creating new files */
1221         if (startswith(s, ".#"))
1222                 return false;
1223
1224         return true;
1225 }
1226
1227 static const char* const image_type_table[_IMAGE_TYPE_MAX] = {
1228         [IMAGE_DIRECTORY] = "directory",
1229         [IMAGE_SUBVOLUME] = "subvolume",
1230         [IMAGE_RAW] = "raw",
1231         [IMAGE_BLOCK] = "block",
1232 };
1233
1234 DEFINE_STRING_TABLE_LOOKUP(image_type, ImageType);