switch_root.c 13 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384
  1. /* vi: set sw=4 ts=4: */
  2. /*
  3. * Copyright 2005 Rob Landley <rob@landley.net>
  4. *
  5. * Switch from rootfs to another filesystem as the root of the mount tree.
  6. *
  7. * Licensed under GPLv2, see file LICENSE in this source tree.
  8. */
  9. //config:config SWITCH_ROOT
  10. //config: bool "switch_root (5.2 kb)"
  11. //config: default y
  12. //config: select PLATFORM_LINUX
  13. //config: help
  14. //config: The switch_root utility is used from initramfs to select a new
  15. //config: root device. Under initramfs, you have to use this instead of
  16. //config: pivot_root. (Stop reading here if you don't care why.)
  17. //config:
  18. //config: Booting with initramfs extracts a gzipped cpio archive into rootfs
  19. //config: (which is a variant of ramfs/tmpfs). Because rootfs can't be moved
  20. //config: or unmounted*, pivot_root will not work from initramfs. Instead,
  21. //config: switch_root deletes everything out of rootfs (including itself),
  22. //config: does a mount --move that overmounts rootfs with the new root, and
  23. //config: then execs the specified init program.
  24. //config:
  25. //config: * Because the Linux kernel uses rootfs internally as the starting
  26. //config: and ending point for searching through the kernel's doubly linked
  27. //config: list of active mount points. That's why.
  28. //config:
  29. // RUN_INIT config item is in klibc-utils
  30. //applet:IF_SWITCH_ROOT(APPLET(switch_root, BB_DIR_SBIN, BB_SUID_DROP))
  31. // APPLET_ODDNAME:name main location suid_type help
  32. //applet:IF_RUN_INIT( APPLET_ODDNAME(run-init, switch_root, BB_DIR_SBIN, BB_SUID_DROP, run_init))
  33. //kbuild:lib-$(CONFIG_SWITCH_ROOT) += switch_root.o
  34. //kbuild:lib-$(CONFIG_RUN_INIT) += switch_root.o
  35. #include <sys/vfs.h>
  36. #include <sys/mount.h>
  37. #if ENABLE_RUN_INIT
  38. # include <sys/prctl.h>
  39. # ifndef PR_CAPBSET_READ
  40. # define PR_CAPBSET_READ 23
  41. # endif
  42. # ifndef PR_CAPBSET_DROP
  43. # define PR_CAPBSET_DROP 24
  44. # endif
  45. # include <linux/capability.h>
  46. // #include <sys/capability.h>
  47. // This header is in libcap, but the functions are in libc.
  48. // Comment in the header says this above capset/capget:
  49. /* system calls - look to libc for function to system call mapping */
  50. extern int capset(cap_user_header_t header, cap_user_data_t data);
  51. extern int capget(cap_user_header_t header, const cap_user_data_t data);
  52. // so for bbox, let's just repeat the declarations.
  53. // This way, libcap needs not be installed in build environment.
  54. #endif
  55. #include "libbb.h"
  56. // Make up for header deficiencies
  57. #ifndef RAMFS_MAGIC
  58. # define RAMFS_MAGIC ((unsigned)0x858458f6)
  59. #endif
  60. #ifndef TMPFS_MAGIC
  61. # define TMPFS_MAGIC ((unsigned)0x01021994)
  62. #endif
  63. #ifndef MS_MOVE
  64. # define MS_MOVE 8192
  65. #endif
  66. // Recursively delete contents of rootfs
  67. static void delete_contents(const char *directory, dev_t rootdev)
  68. {
  69. DIR *dir;
  70. struct dirent *d;
  71. struct stat st;
  72. // Don't descend into other filesystems
  73. if (lstat(directory, &st) || st.st_dev != rootdev)
  74. return;
  75. // Recursively delete the contents of directories
  76. if (S_ISDIR(st.st_mode)) {
  77. dir = opendir(directory);
  78. if (dir) {
  79. while ((d = readdir(dir))) {
  80. char *newdir = d->d_name;
  81. // Skip . and ..
  82. if (DOT_OR_DOTDOT(newdir))
  83. continue;
  84. // Recurse to delete contents
  85. newdir = concat_path_file(directory, newdir);
  86. delete_contents(newdir, rootdev);
  87. free(newdir);
  88. }
  89. closedir(dir);
  90. // Directory should now be empty, zap it
  91. rmdir(directory);
  92. }
  93. } else {
  94. // It wasn't a directory, zap it
  95. unlink(directory);
  96. }
  97. }
  98. #if ENABLE_RUN_INIT
  99. DEFINE_STRUCT_CAPS;
  100. static void drop_capset(int cap_idx)
  101. {
  102. struct caps caps;
  103. getcaps(&caps);
  104. caps.data[CAP_TO_INDEX(cap_idx)].inheritable &= ~CAP_TO_MASK(cap_idx);
  105. if (capset(&caps.header, caps.data) != 0)
  106. bb_perror_msg_and_die("capset");
  107. }
  108. static void drop_bounding_set(int cap_idx)
  109. {
  110. int ret;
  111. ret = prctl(PR_CAPBSET_READ, cap_idx, 0, 0, 0);
  112. if (ret < 0)
  113. bb_perror_msg_and_die("prctl: %s", "PR_CAPBSET_READ");
  114. if (ret == 1) {
  115. ret = prctl(PR_CAPBSET_DROP, cap_idx, 0, 0, 0);
  116. if (ret != 0)
  117. bb_perror_msg_and_die("prctl: %s", "PR_CAPBSET_DROP");
  118. }
  119. }
  120. static void drop_usermodehelper(const char *filename, int cap_idx)
  121. {
  122. unsigned lo, hi;
  123. char buf[sizeof(int)*3 * 2 + 8];
  124. int fd;
  125. int ret;
  126. ret = open_read_close(filename, buf, sizeof(buf) - 1);
  127. if (ret < 0)
  128. return; /* assuming files do not exist */
  129. buf[ret] = '\0';
  130. ret = sscanf(buf, "%u %u", &lo, &hi);
  131. if (ret != 2)
  132. bb_perror_msg_and_die("can't parse file '%s'", filename);
  133. if (cap_idx < 32)
  134. lo &= ~(1 << cap_idx);
  135. else
  136. hi &= ~(1 << (cap_idx - 32));
  137. fd = xopen(filename, O_WRONLY);
  138. fdprintf(fd, "%u %u", lo, hi);
  139. close(fd);
  140. }
  141. static void drop_capabilities(char *string)
  142. {
  143. char *cap;
  144. cap = strtok(string, ",");
  145. while (cap) {
  146. unsigned cap_idx;
  147. cap_idx = cap_name_to_number(cap);
  148. drop_usermodehelper("/proc/sys/kernel/usermodehelper/bset", cap_idx);
  149. drop_usermodehelper("/proc/sys/kernel/usermodehelper/inheritable", cap_idx);
  150. drop_bounding_set(cap_idx);
  151. drop_capset(cap_idx);
  152. bb_error_msg("dropped capability: %s", cap);
  153. cap = strtok(NULL, ",");
  154. }
  155. }
  156. #endif
  157. int switch_root_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
  158. int switch_root_main(int argc UNUSED_PARAM, char **argv)
  159. {
  160. char *newroot, *console = NULL;
  161. struct stat st;
  162. struct statfs stfs;
  163. unsigned dry_run = 0;
  164. dev_t rootdev;
  165. // Parse args. '+': stop at first non-option
  166. if (ENABLE_SWITCH_ROOT && (!ENABLE_RUN_INIT || applet_name[0] == 's')) {
  167. //usage:#define switch_root_trivial_usage
  168. //usage: "[-c CONSOLE_DEV] NEW_ROOT NEW_INIT [ARGS]"
  169. //usage:#define switch_root_full_usage "\n\n"
  170. //usage: "Free initramfs and switch to another root fs:\n"
  171. //usage: "chroot to NEW_ROOT, delete all in /, move NEW_ROOT to /,\n"
  172. //usage: "execute NEW_INIT. PID must be 1. NEW_ROOT must be a mountpoint.\n"
  173. //usage: "\n -c DEV Reopen stdio to DEV after switch"
  174. getopt32(argv, "^+"
  175. "c:"
  176. "\0" "-2" /* minimum 2 args */,
  177. &console
  178. );
  179. } else {
  180. #if ENABLE_RUN_INIT
  181. //usage:#define run_init_trivial_usage
  182. //usage: "[-d CAP,CAP...] [-n] [-c CONSOLE_DEV] NEW_ROOT NEW_INIT [ARGS]"
  183. //usage:#define run_init_full_usage "\n\n"
  184. //usage: "Free initramfs and switch to another root fs:\n"
  185. //usage: "chroot to NEW_ROOT, delete all in /, move NEW_ROOT to /,\n"
  186. //usage: "execute NEW_INIT. PID must be 1. NEW_ROOT must be a mountpoint.\n"
  187. //usage: "\n -c DEV Reopen stdio to DEV after switch"
  188. //usage: "\n -d CAPS Drop capabilities"
  189. //usage: "\n -n Dry run"
  190. char *cap_list = NULL;
  191. dry_run = getopt32(argv, "^+"
  192. "c:d:n"
  193. "\0" "-2" /* minimum 2 args */,
  194. &console,
  195. &cap_list
  196. );
  197. dry_run >>= 2; // -n
  198. if (cap_list)
  199. drop_capabilities(cap_list);
  200. #endif
  201. }
  202. argv += optind;
  203. newroot = *argv++;
  204. // Change to new root directory and verify it's a different fs
  205. xchdir(newroot);
  206. xstat("/", &st);
  207. rootdev = st.st_dev;
  208. xstat(".", &st);
  209. if (st.st_dev == rootdev) {
  210. // Show usage, it says new root must be a mountpoint
  211. bb_show_usage();
  212. }
  213. if (!dry_run && getpid() != 1) {
  214. // Show usage, it says we must be PID 1
  215. bb_show_usage();
  216. }
  217. // Additional sanity checks: we're about to rm -rf /, so be REALLY SURE
  218. // we mean it. I could make this a CONFIG option, but I would get email
  219. // from all the people who WILL destroy their filesystems.
  220. if (stat("/init", &st) != 0 || !S_ISREG(st.st_mode)) {
  221. bb_error_msg_and_die("'%s' is not a regular file", "/init");
  222. }
  223. statfs("/", &stfs); // this never fails
  224. if ((unsigned)stfs.f_type != RAMFS_MAGIC
  225. && (unsigned)stfs.f_type != TMPFS_MAGIC
  226. ) {
  227. bb_error_msg_and_die("root filesystem is not ramfs/tmpfs");
  228. }
  229. if (!dry_run) {
  230. // Zap everything out of rootdev
  231. delete_contents("/", rootdev);
  232. // Overmount / with newdir and chroot into it
  233. if (mount(".", "/", NULL, MS_MOVE, NULL)) {
  234. // For example, fails when newroot is not a mountpoint
  235. bb_perror_msg_and_die("error moving root");
  236. }
  237. }
  238. xchroot(".");
  239. // The chdir is needed to recalculate "." and ".." links
  240. /*xchdir("/"); - done in xchroot */
  241. // If a new console specified, redirect stdin/stdout/stderr to it
  242. if (console) {
  243. int fd = open_or_warn(console, O_RDWR);
  244. if (fd >= 0) {
  245. xmove_fd(fd, 0);
  246. xdup2(0, 1);
  247. xdup2(0, 2);
  248. }
  249. }
  250. if (dry_run) {
  251. // Does NEW_INIT look like it can be executed?
  252. //xstat(argv[0], &st);
  253. //if (!S_ISREG(st.st_mode))
  254. // bb_perror_msg_and_die("'%s' is not a regular file", argv[0]);
  255. if (access(argv[0], X_OK) == 0)
  256. return 0;
  257. } else {
  258. // Exec NEW_INIT
  259. execv(argv[0], argv);
  260. }
  261. bb_perror_msg_and_die("can't execute '%s'", argv[0]);
  262. }
  263. /*
  264. From: Rob Landley <rob@landley.net>
  265. Date: Tue, Jun 16, 2009 at 7:47 PM
  266. Subject: Re: switch_root...
  267. ...
  268. ...
  269. ...
  270. If you're _not_ running out of init_ramfs (if for example you're using initrd
  271. instead), you probably shouldn't use switch_root because it's the wrong tool.
  272. Basically what the sucker does is something like the following shell script:
  273. find / -xdev | xargs rm -rf
  274. cd "$1"
  275. shift
  276. mount --move . /
  277. exec chroot . "$@"
  278. There are a couple reasons that won't work as a shell script:
  279. 1) If you delete the commands out of your $PATH, your shell scripts can't run
  280. more commands, but you can't start using dynamically linked _new_ commands
  281. until after you do the chroot because the path to the dynamic linker is wrong.
  282. So there's a step that needs to be sort of atomic but can't be as a shell
  283. script. (You can work around this with static linking or very carefully laid
  284. out paths and sequencing, but it's brittle, ugly, and non-obvious.)
  285. 2) The "find | rm" bit will actually delete everything because the mount points
  286. still show up (even if their contents don't), and rm -rf will then happily zap
  287. that. So the first line is an oversimplification of what you need to do _not_
  288. to descend into other filesystems and delete their contents.
  289. The reason we do this is to free up memory, by the way. Since initramfs is a
  290. ramfs, deleting its contents frees up the memory it uses. (We leave it with
  291. one remaining dentry for the new mount point, but that's ok.)
  292. Note that you cannot ever umount rootfs, for approximately the same reason you
  293. can't kill PID 1. The kernel tracks mount points as a doubly linked list, and
  294. the pointer to the start/end of that list always points to an entry that's
  295. known to be there (rootfs), so it never has to worry about moving that pointer
  296. and it never has to worry about the list being empty. (Back around 2.6.13
  297. there _was_ a bug that let you umount rootfs, and the system locked hard the
  298. instant you did so endlessly looping to find the end of the mount list and
  299. never stopping. They fixed it.)
  300. Oh, and the reason we mount --move _and_ do the chroot is due to the way "/"
  301. works. Each process has two special symlinks, ".", and "/". Each of them
  302. points to the dentry of a directory, and give you a location paths can start
  303. from. (Historically ".." was also special, because you could enter a
  304. directory via a symlink so backing out to the directory you came from doesn't
  305. necessarily mean the one physically above where "." points to. These days I
  306. think it's just handed off to the filesystem.)
  307. Anyway, path resolution starts with "." or "/" (although the "./" at the start
  308. of the path may be implicit), meaning it's relative to one of those two
  309. directories. Your current directory, and your current root directory. The
  310. chdir() syscall changes where "." points to, and the chroot() syscall changes
  311. where "/" points to. (Again, both are per-process which is why chroot only
  312. affects your current process and its child processes.)
  313. Note that chroot() does _not_ change where "." points to, and back before they
  314. put crazy security checks into the kernel your current directory could be
  315. somewhere you could no longer access after the chroot. (The command line
  316. chroot does a cd as well, the chroot _syscall_ is what I'm talking about.)
  317. The reason mounting something new over / has no obvious effect is the same
  318. reason mounting something over your current directory has no obvious effect:
  319. the . and / links aren't recalculated after a mount, so they still point to
  320. the same dentry they did before, even if that dentry is no longer accessible
  321. by other means. Note that "cd ." is a NOP, and "chroot /" is a nop; both look
  322. up the cached dentry and set it right back. They don't re-parse any paths,
  323. because they're what all paths your process uses would be relative to.
  324. That's why the careful sequencing above: we cd into the new mount point before
  325. we do the mount --move. Moving the mount point would otherwise make it
  326. totally inaccessible to us because cd-ing to the old path wouldn't give it to
  327. us anymore, and cd "/" just gives us the cached dentry from when the process
  328. was created (in this case the old initramfs one). But the "." symlink gives
  329. us the dentry of the filesystem we just moved, so we can then "chroot ." to
  330. copy that dentry to "/" and get the new filesystem. If we _didn't_ save that
  331. dentry in "." we couldn't get it back after the mount --move.
  332. (Yes, this is all screwy and I had to email questions to Linus Torvalds to get
  333. it straight myself. I keep meaning to write up a "how mount actually works"
  334. document someday...)
  335. */