1 /* vi: set sw=4 ts=4: */
3 * Copyright 2005 Rob Landley <rob@landley.net>
5 * Switch from rootfs to another filesystem as the root of the mount tree.
7 * Licensed under GPLv2, see file LICENSE in this source tree.
9 //config:config SWITCH_ROOT
10 //config: bool "switch_root (5.5 kb)"
12 //config: select PLATFORM_LINUX
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.)
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.
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.
29 // RUN_INIT config item is in klibc-utils
31 //applet:IF_SWITCH_ROOT(APPLET(switch_root, BB_DIR_SBIN, BB_SUID_DROP))
32 // APPLET_ODDNAME:name main location suid_type help
33 //applet:IF_RUN_INIT( APPLET_ODDNAME(run-init, switch_root, BB_DIR_SBIN, BB_SUID_DROP, run_init))
35 //kbuild:lib-$(CONFIG_SWITCH_ROOT) += switch_root.o
36 //kbuild:lib-$(CONFIG_RUN_INIT) += switch_root.o
39 #include <sys/mount.h>
41 # include <sys/prctl.h>
42 # ifndef PR_CAPBSET_READ
43 # define PR_CAPBSET_READ 23
45 # ifndef PR_CAPBSET_DROP
46 # define PR_CAPBSET_DROP 24
48 # include <linux/capability.h>
49 // #include <sys/capability.h>
50 // This header is in libcap, but the functions are in libc.
51 // Comment in the header says this above capset/capget:
52 /* system calls - look to libc for function to system call mapping */
53 extern int capset(cap_user_header_t header, cap_user_data_t data);
54 extern int capget(cap_user_header_t header, const cap_user_data_t data);
55 // so for bbox, let's just repeat the declarations.
56 // This way, libcap needs not be installed in build environment.
61 // Make up for header deficiencies
63 # define RAMFS_MAGIC ((unsigned)0x858458f6)
66 # define TMPFS_MAGIC ((unsigned)0x01021994)
72 // Recursively delete contents of rootfs
73 static void delete_contents(const char *directory, dev_t rootdev)
79 // Don't descend into other filesystems
80 if (lstat(directory, &st) || st.st_dev != rootdev)
83 // Recursively delete the contents of directories
84 if (S_ISDIR(st.st_mode)) {
85 dir = opendir(directory);
87 while ((d = readdir(dir))) {
88 char *newdir = d->d_name;
91 if (DOT_OR_DOTDOT(newdir))
94 // Recurse to delete contents
95 newdir = concat_path_file(directory, newdir);
96 delete_contents(newdir, rootdev);
101 // Directory should now be empty, zap it
105 // It wasn't a directory, zap it
113 static void drop_capset(int cap_idx)
118 caps.data[CAP_TO_INDEX(cap_idx)].inheritable &= ~CAP_TO_MASK(cap_idx);
119 if (capset(&caps.header, caps.data) != 0)
120 bb_perror_msg_and_die("capset");
123 static void drop_bounding_set(int cap_idx)
127 ret = prctl(PR_CAPBSET_READ, cap_idx, 0, 0, 0);
129 bb_perror_msg_and_die("prctl: %s", "PR_CAPBSET_READ");
132 ret = prctl(PR_CAPBSET_DROP, cap_idx, 0, 0, 0);
134 bb_perror_msg_and_die("prctl: %s", "PR_CAPBSET_DROP");
138 static void drop_usermodehelper(const char *filename, int cap_idx)
141 char buf[sizeof(int)*3 * 2 + 8];
145 ret = open_read_close(filename, buf, sizeof(buf) - 1);
147 return; /* assuming files do not exist */
150 ret = sscanf(buf, "%u %u", &lo, &hi);
152 bb_perror_msg_and_die("can't parse file '%s'", filename);
155 lo &= ~(1 << cap_idx);
157 hi &= ~(1 << (cap_idx - 32));
159 fd = xopen(filename, O_WRONLY);
160 fdprintf(fd, "%u %u", lo, hi);
164 static void drop_capabilities(char *string)
168 cap = strtok(string, ",");
172 cap_idx = cap_name_to_number(cap);
173 drop_usermodehelper("/proc/sys/kernel/usermodehelper/bset", cap_idx);
174 drop_usermodehelper("/proc/sys/kernel/usermodehelper/inheritable", cap_idx);
175 drop_bounding_set(cap_idx);
176 drop_capset(cap_idx);
177 bb_error_msg("dropped capability: %s", cap);
178 cap = strtok(NULL, ",");
183 int switch_root_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
184 int switch_root_main(int argc UNUSED_PARAM, char **argv)
186 char *newroot, *console = NULL;
189 unsigned dry_run = 0;
192 // Parse args. '+': stop at first non-option
193 if (ENABLE_SWITCH_ROOT && (!ENABLE_RUN_INIT || applet_name[0] == 's')) {
194 //usage:#define switch_root_trivial_usage
195 //usage: "[-c CONSOLE_DEV] NEW_ROOT NEW_INIT [ARGS]"
196 //usage:#define switch_root_full_usage "\n\n"
197 //usage: "Free initramfs and switch to another root fs:\n"
198 //usage: "chroot to NEW_ROOT, delete all in /, move NEW_ROOT to /,\n"
199 //usage: "execute NEW_INIT. PID must be 1. NEW_ROOT must be a mountpoint.\n"
200 //usage: "\n -c DEV Reopen stdio to DEV after switch"
203 "\0" "-2" /* minimum 2 args */,
208 //usage:#define run_init_trivial_usage
209 //usage: "[-d CAP,CAP...] [-n] [-c CONSOLE_DEV] NEW_ROOT NEW_INIT [ARGS]"
210 //usage:#define run_init_full_usage "\n\n"
211 //usage: "Free initramfs and switch to another root fs:\n"
212 //usage: "chroot to NEW_ROOT, delete all in /, move NEW_ROOT to /,\n"
213 //usage: "execute NEW_INIT. PID must be 1. NEW_ROOT must be a mountpoint.\n"
214 //usage: "\n -c DEV Reopen stdio to DEV after switch"
215 //usage: "\n -d CAPS Drop capabilities"
216 //usage: "\n -n Dry run"
217 char *cap_list = NULL;
218 dry_run = getopt32(argv, "^+"
220 "\0" "-2" /* minimum 2 args */,
226 drop_capabilities(cap_list);
232 // Change to new root directory and verify it's a different fs
237 if (st.st_dev == rootdev) {
238 // Show usage, it says new root must be a mountpoint
241 if (!dry_run && getpid() != 1) {
242 // Show usage, it says we must be PID 1
246 // Additional sanity checks: we're about to rm -rf /, so be REALLY SURE
247 // we mean it. I could make this a CONFIG option, but I would get email
248 // from all the people who WILL destroy their filesystems.
249 if (stat("/init", &st) != 0 || !S_ISREG(st.st_mode)) {
250 bb_error_msg_and_die("'%s' is not a regular file", "/init");
252 statfs("/", &stfs); // this never fails
253 if ((unsigned)stfs.f_type != RAMFS_MAGIC
254 && (unsigned)stfs.f_type != TMPFS_MAGIC
256 bb_error_msg_and_die("root filesystem is not ramfs/tmpfs");
260 // Zap everything out of rootdev
261 delete_contents("/", rootdev);
263 // Overmount / with newdir and chroot into it
264 if (mount(".", "/", NULL, MS_MOVE, NULL)) {
265 // For example, fails when newroot is not a mountpoint
266 bb_perror_msg_and_die("error moving root");
270 // The chdir is needed to recalculate "." and ".." links
271 /*xchdir("/"); - done in xchroot */
273 // If a new console specified, redirect stdin/stdout/stderr to it
275 int fd = open_or_warn(console, O_RDWR);
284 // Does NEW_INIT look like it can be executed?
285 //xstat(argv[0], &st);
286 //if (!S_ISREG(st.st_mode))
287 // bb_perror_msg_and_die("'%s' is not a regular file", argv[0]);
288 if (access(argv[0], X_OK) == 0)
292 execv(argv[0], argv);
294 bb_perror_msg_and_die("can't execute '%s'", argv[0]);
298 From: Rob Landley <rob@landley.net>
299 Date: Tue, Jun 16, 2009 at 7:47 PM
300 Subject: Re: switch_root...
306 If you're _not_ running out of init_ramfs (if for example you're using initrd
307 instead), you probably shouldn't use switch_root because it's the wrong tool.
309 Basically what the sucker does is something like the following shell script:
311 find / -xdev | xargs rm -rf
317 There are a couple reasons that won't work as a shell script:
319 1) If you delete the commands out of your $PATH, your shell scripts can't run
320 more commands, but you can't start using dynamically linked _new_ commands
321 until after you do the chroot because the path to the dynamic linker is wrong.
322 So there's a step that needs to be sort of atomic but can't be as a shell
323 script. (You can work around this with static linking or very carefully laid
324 out paths and sequencing, but it's brittle, ugly, and non-obvious.)
326 2) The "find | rm" bit will actually delete everything because the mount points
327 still show up (even if their contents don't), and rm -rf will then happily zap
328 that. So the first line is an oversimplification of what you need to do _not_
329 to descend into other filesystems and delete their contents.
331 The reason we do this is to free up memory, by the way. Since initramfs is a
332 ramfs, deleting its contents frees up the memory it uses. (We leave it with
333 one remaining dentry for the new mount point, but that's ok.)
335 Note that you cannot ever umount rootfs, for approximately the same reason you
336 can't kill PID 1. The kernel tracks mount points as a doubly linked list, and
337 the pointer to the start/end of that list always points to an entry that's
338 known to be there (rootfs), so it never has to worry about moving that pointer
339 and it never has to worry about the list being empty. (Back around 2.6.13
340 there _was_ a bug that let you umount rootfs, and the system locked hard the
341 instant you did so endlessly looping to find the end of the mount list and
342 never stopping. They fixed it.)
344 Oh, and the reason we mount --move _and_ do the chroot is due to the way "/"
345 works. Each process has two special symlinks, ".", and "/". Each of them
346 points to the dentry of a directory, and give you a location paths can start
347 from. (Historically ".." was also special, because you could enter a
348 directory via a symlink so backing out to the directory you came from doesn't
349 necessarily mean the one physically above where "." points to. These days I
350 think it's just handed off to the filesystem.)
352 Anyway, path resolution starts with "." or "/" (although the "./" at the start
353 of the path may be implicit), meaning it's relative to one of those two
354 directories. Your current directory, and your current root directory. The
355 chdir() syscall changes where "." points to, and the chroot() syscall changes
356 where "/" points to. (Again, both are per-process which is why chroot only
357 affects your current process and its child processes.)
359 Note that chroot() does _not_ change where "." points to, and back before they
360 put crazy security checks into the kernel your current directory could be
361 somewhere you could no longer access after the chroot. (The command line
362 chroot does a cd as well, the chroot _syscall_ is what I'm talking about.)
364 The reason mounting something new over / has no obvious effect is the same
365 reason mounting something over your current directory has no obvious effect:
366 the . and / links aren't recalculated after a mount, so they still point to
367 the same dentry they did before, even if that dentry is no longer accessible
368 by other means. Note that "cd ." is a NOP, and "chroot /" is a nop; both look
369 up the cached dentry and set it right back. They don't re-parse any paths,
370 because they're what all paths your process uses would be relative to.
372 That's why the careful sequencing above: we cd into the new mount point before
373 we do the mount --move. Moving the mount point would otherwise make it
374 totally inaccessible to us because cd-ing to the old path wouldn't give it to
375 us anymore, and cd "/" just gives us the cached dentry from when the process
376 was created (in this case the old initramfs one). But the "." symlink gives
377 us the dentry of the filesystem we just moved, so we can then "chroot ." to
378 copy that dentry to "/" and get the new filesystem. If we _didn't_ save that
379 dentry in "." we couldn't get it back after the mount --move.
381 (Yes, this is all screwy and I had to email questions to Linus Torvalds to get
382 it straight myself. I keep meaning to write up a "how mount actually works"