1 /* vi: set sw=4 ts=4: */
2 /* Copyright 2005 Rob Landley <rob@landley.net>
4 * Switch from rootfs to another filesystem as the root of the mount tree.
6 * Licensed under GPLv2, see file LICENSE in this source tree.
8 //config:config SWITCH_ROOT
9 //config: bool "switch_root"
11 //config: select PLATFORM_LINUX
13 //config: The switch_root utility is used from initramfs to select a new
14 //config: root device. Under initramfs, you have to use this instead of
15 //config: pivot_root. (Stop reading here if you don't care why.)
17 //config: Booting with initramfs extracts a gzipped cpio archive into rootfs
18 //config: (which is a variant of ramfs/tmpfs). Because rootfs can't be moved
19 //config: or unmounted*, pivot_root will not work from initramfs. Instead,
20 //config: switch_root deletes everything out of rootfs (including itself),
21 //config: does a mount --move that overmounts rootfs with the new root, and
22 //config: then execs the specified init program.
24 //config: * Because the Linux kernel uses rootfs internally as the starting
25 //config: and ending point for searching through the kernel's doubly linked
26 //config: list of active mount points. That's why.
28 //applet:IF_SWITCH_ROOT(APPLET(switch_root, BB_DIR_SBIN, BB_SUID_DROP))
30 //kbuild:lib-$(CONFIG_SWITCH_ROOT) += switch_root.o
32 //usage:#define switch_root_trivial_usage
33 //usage: "[-c /dev/console] NEW_ROOT NEW_INIT [ARGS]"
34 //usage:#define switch_root_full_usage "\n\n"
35 //usage: "Free initramfs and switch to another root fs:\n"
36 //usage: "chroot to NEW_ROOT, delete all in /, move NEW_ROOT to /,\n"
37 //usage: "execute NEW_INIT. PID must be 1. NEW_ROOT must be a mountpoint.\n"
38 //usage: "\n -c DEV Reopen stdio to DEV after switch"
41 #include <sys/mount.h>
43 // Make up for header deficiencies
45 # define RAMFS_MAGIC ((unsigned)0x858458f6)
48 # define TMPFS_MAGIC ((unsigned)0x01021994)
54 // Recursively delete contents of rootfs
55 static void delete_contents(const char *directory, dev_t rootdev)
61 // Don't descend into other filesystems
62 if (lstat(directory, &st) || st.st_dev != rootdev)
65 // Recursively delete the contents of directories
66 if (S_ISDIR(st.st_mode)) {
67 dir = opendir(directory);
69 while ((d = readdir(dir))) {
70 char *newdir = d->d_name;
73 if (DOT_OR_DOTDOT(newdir))
76 // Recurse to delete contents
77 newdir = concat_path_file(directory, newdir);
78 delete_contents(newdir, rootdev);
83 // Directory should now be empty, zap it
87 // It wasn't a directory, zap it
92 int switch_root_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
93 int switch_root_main(int argc UNUSED_PARAM, char **argv)
95 char *newroot, *console = NULL;
100 // Parse args (-c console)
101 opt_complementary = "-2"; // minimum 2 params
102 getopt32(argv, "+c:", &console); // '+': stop at first non-option
106 // Change to new root directory and verify it's a different fs
111 if (st.st_dev == rootdev || getpid() != 1) {
112 // Show usage, it says new root must be a mountpoint
113 // and we must be PID 1
117 // Additional sanity checks: we're about to rm -rf /, so be REALLY SURE
118 // we mean it. I could make this a CONFIG option, but I would get email
119 // from all the people who WILL destroy their filesystems.
120 if (stat("/init", &st) != 0 || !S_ISREG(st.st_mode)) {
121 bb_error_msg_and_die("/init is not a regular file");
123 statfs("/", &stfs); // this never fails
124 if ((unsigned)stfs.f_type != RAMFS_MAGIC
125 && (unsigned)stfs.f_type != TMPFS_MAGIC
127 bb_error_msg_and_die("root filesystem is not ramfs/tmpfs");
130 // Zap everything out of rootdev
131 delete_contents("/", rootdev);
133 // Overmount / with newdir and chroot into it
134 if (mount(".", "/", NULL, MS_MOVE, NULL)) {
135 // For example, fails when newroot is not a mountpoint
136 bb_perror_msg_and_die("error moving root");
139 // The chdir is needed to recalculate "." and ".." links
140 /*xchdir("/"); - done in xchroot */
142 // If a new console specified, redirect stdin/stdout/stderr to it
144 int fd = open_or_warn(console, O_RDWR);
153 execv(argv[0], argv);
154 bb_perror_msg_and_die("can't execute '%s'", argv[0]);
158 From: Rob Landley <rob@landley.net>
159 Date: Tue, Jun 16, 2009 at 7:47 PM
160 Subject: Re: switch_root...
166 If you're _not_ running out of init_ramfs (if for example you're using initrd
167 instead), you probably shouldn't use switch_root because it's the wrong tool.
169 Basically what the sucker does is something like the following shell script:
171 find / -xdev | xargs rm -rf
177 There are a couple reasons that won't work as a shell script:
179 1) If you delete the commands out of your $PATH, your shell scripts can't run
180 more commands, but you can't start using dynamically linked _new_ commands
181 until after you do the chroot because the path to the dynamic linker is wrong.
182 So there's a step that needs to be sort of atomic but can't be as a shell
183 script. (You can work around this with static linking or very carefully laid
184 out paths and sequencing, but it's brittle, ugly, and non-obvious.)
186 2) The "find | rm" bit will acually delete everything because the mount points
187 still show up (even if their contents don't), and rm -rf will then happily zap
188 that. So the first line is an oversimplification of what you need to do _not_
189 to descend into other filesystems and delete their contents.
191 The reason we do this is to free up memory, by the way. Since initramfs is a
192 ramfs, deleting its contents frees up the memory it uses. (We leave it with
193 one remaining dentry for the new mount point, but that's ok.)
195 Note that you cannot ever umount rootfs, for approximately the same reason you
196 can't kill PID 1. The kernel tracks mount points as a doubly linked list, and
197 the pointer to the start/end of that list always points to an entry that's
198 known to be there (rootfs), so it never has to worry about moving that pointer
199 and it never has to worry about the list being empty. (Back around 2.6.13
200 there _was_ a bug that let you umount rootfs, and the system locked hard the
201 instant you did so endlessly looping to find the end of the mount list and
202 never stopping. They fixed it.)
204 Oh, and the reason we mount --move _and_ do the chroot is due to the way "/"
205 works. Each process has two special symlinks, ".", and "/". Each of them
206 points to the dentry of a directory, and give you a location paths can start
207 from. (Historically ".." was also special, because you could enter a
208 directory via a symlink so backing out to the directory you came from doesn't
209 necessarily mean the one physically above where "." points to. These days I
210 think it's just handed off to the filesystem.)
212 Anyway, path resolution starts with "." or "/" (although the "./" at the start
213 of the path may be implicit), meaning it's relative to one of those two
214 directories. Your current directory, and your current root directory. The
215 chdir() syscall changes where "." points to, and the chroot() syscall changes
216 where "/" points to. (Again, both are per-process which is why chroot only
217 affects your current process and its child processes.)
219 Note that chroot() does _not_ change where "." points to, and back before they
220 put crazy security checks into the kernel your current directory could be
221 somewhere you could no longer access after the chroot. (The command line
222 chroot does a cd as well, the chroot _syscall_ is what I'm talking about.)
224 The reason mounting something new over / has no obvious effect is the same
225 reason mounting something over your current directory has no obvious effect:
226 the . and / links aren't recalculated after a mount, so they still point to
227 the same dentry they did before, even if that dentry is no longer accessible
228 by other means. Note that "cd ." is a NOP, and "chroot /" is a nop; both look
229 up the cached dentry and set it right back. They don't re-parse any paths,
230 because they're what all paths your process uses would be relative to.
232 That's why the careful sequencing above: we cd into the new mount point before
233 we do the mount --move. Moving the mount point would otherwise make it
234 totally inaccessible to us because cd-ing to the old path wouldn't give it to
235 us anymore, and cd "/" just gives us the cached dentry from when the process
236 was created (in this case the old initramfs one). But the "." symlink gives
237 us the dentry of the filesystem we just moved, so we can then "chroot ." to
238 copy that dentry to "/" and get the new filesystem. If we _didn't_ save that
239 dentry in "." we couldn't get it back after the mount --move.
241 (Yes, this is all screwy and I had to email questions to Linus Torvalds to get
242 it straight myself. I keep meaning to write up a "how mount actually works"