Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
fs: use nullfs unconditionally as the real rootfs
Remove the "nullfs_rootfs" boot parameter and simply always use nullfs.
The mutable rootfs will be mounted on top of it. Systems that don't use
pivot_root() to pivot away from the real rootfs will have an additional
mount stick around but that shouldn't be a problem at all. If it is
we'll rever this commit.
This also simplifies the boot process and removes the need for the
traditional switch_root workarounds.
Suggested-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Christian Brauner <brauner@kernel.org>
+32
-77
+4
-20
Documentation/filesystems/ramfs-rootfs-initramfs.rst
+4
-20
Documentation/filesystems/ramfs-rootfs-initramfs.rst
···
76
76
---------------
77
77
78
78
Rootfs is a special instance of ramfs (or tmpfs, if that's enabled), which is
79
-
always present in 2.6 systems. Traditionally, you can't unmount rootfs for
80
-
approximately the same reason you can't kill the init process; rather than
81
-
having special code to check for and handle an empty list, it's smaller and
82
-
simpler for the kernel to just make sure certain lists can't become empty.
83
-
84
-
However, if the kernel is booted with "nullfs_rootfs", an immutable empty
85
-
filesystem called nullfs is used as the true root, with the mutable rootfs
79
+
always present in Linux systems. The kernel uses an immutable empty filesystem
80
+
called nullfs as the true root of the VFS hierarchy, with the mutable rootfs
86
81
(tmpfs/ramfs) mounted on top of it. This allows pivot_root() and unmounting
87
82
of the initramfs to work normally.
88
83
···
121
126
program. See the switch_root utility, below.)
122
127
123
128
- When switching another root device, initrd would pivot_root and then
124
-
umount the ramdisk. Traditionally, initramfs is rootfs: you can neither
125
-
pivot_root rootfs, nor unmount it. Instead delete everything out of
126
-
rootfs to free up the space (find -xdev / -exec rm '{}' ';'), overmount
127
-
rootfs with the new root (cd /newmount; mount --move . /; chroot .),
128
-
attach stdin/stdout/stderr to the new /dev/console, and exec the new init.
129
-
130
-
Since this is a remarkably persnickety process (and involves deleting
131
-
commands before you can run them), the klibc package introduced a helper
132
-
program (utils/run_init.c) to do all this for you. Most other packages
133
-
(such as busybox) have named this command "switch_root".
134
-
135
-
However, if the kernel is booted with "nullfs_rootfs", pivot_root() works
129
+
umount the ramdisk. With nullfs as the true root, pivot_root() works
136
130
normally from the initramfs. Userspace can simply do::
137
131
138
132
chdir(new_root);
139
133
pivot_root(".", ".");
140
134
umount2(".", MNT_DETACH);
141
135
142
-
This is the preferred method when nullfs_rootfs is enabled.
136
+
This is the preferred method for switching root filesystems.
143
137
144
138
Populating initramfs:
145
139
---------------------
+21
-43
fs/namespace.c
+21
-43
fs/namespace.c
···
75
75
76
76
__setup("initramfs_options=", initramfs_options_setup);
77
77
78
-
bool nullfs_rootfs = false;
79
-
80
-
static int __init nullfs_rootfs_setup(char *str)
81
-
{
82
-
if (*str)
83
-
return 0;
84
-
nullfs_rootfs = true;
85
-
return 1;
86
-
}
87
-
__setup("nullfs_rootfs", nullfs_rootfs_setup);
88
-
89
78
static u64 event;
90
79
static DEFINE_XARRAY_FLAGS(mnt_id_xa, XA_FLAGS_ALLOC);
91
80
static DEFINE_IDA(mnt_group_ida);
···
4582
4593
* pointed to by put_old must yield the same directory as new_root. No other
4583
4594
* file system may be mounted on put_old. After all, new_root is a mountpoint.
4584
4595
*
4585
-
* Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem
4586
-
* unless the kernel was booted with "nullfs_rootfs". See
4587
-
* Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives
4588
-
* in this situation.
4596
+
* The immutable nullfs filesystem is mounted as the true root of the VFS
4597
+
* hierarchy. The mutable rootfs (tmpfs/ramfs) is layered on top of this,
4598
+
* allowing pivot_root() to work normally from initramfs.
4589
4599
*
4590
4600
* Notes:
4591
4601
* - we don't move root/cwd if they are not at the root (reason: if something
···
5981
5993
struct path root;
5982
5994
5983
5995
/*
5984
-
* When nullfs is used, we create two mounts:
5996
+
* We create two mounts:
5985
5997
*
5986
5998
* (1) nullfs with mount id 1
5987
5999
* (2) mutable rootfs with mount id 2
5988
6000
*
5989
6001
* with (2) mounted on top of (1).
5990
6002
*/
5991
-
if (nullfs_rootfs) {
5992
-
nullfs_mnt = vfs_kern_mount(&nullfs_fs_type, 0, "nullfs", NULL);
5993
-
if (IS_ERR(nullfs_mnt))
5994
-
panic("VFS: Failed to create nullfs");
5995
-
}
6003
+
nullfs_mnt = vfs_kern_mount(&nullfs_fs_type, 0, "nullfs", NULL);
6004
+
if (IS_ERR(nullfs_mnt))
6005
+
panic("VFS: Failed to create nullfs");
5996
6006
5997
6007
mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", initramfs_options);
5998
6008
if (IS_ERR(mnt))
5999
6009
panic("Can't create rootfs");
6000
6010
6001
-
if (nullfs_rootfs) {
6002
-
VFS_WARN_ON_ONCE(real_mount(nullfs_mnt)->mnt_id != 1);
6003
-
VFS_WARN_ON_ONCE(real_mount(mnt)->mnt_id != 2);
6011
+
VFS_WARN_ON_ONCE(real_mount(nullfs_mnt)->mnt_id != 1);
6012
+
VFS_WARN_ON_ONCE(real_mount(mnt)->mnt_id != 2);
6004
6013
6005
-
/* The namespace root is the nullfs mnt. */
6006
-
mnt_root = real_mount(nullfs_mnt);
6007
-
init_mnt_ns.root = mnt_root;
6014
+
/* The namespace root is the nullfs mnt. */
6015
+
mnt_root = real_mount(nullfs_mnt);
6016
+
init_mnt_ns.root = mnt_root;
6008
6017
6009
-
/* Mount mutable rootfs on top of nullfs. */
6010
-
root.mnt = nullfs_mnt;
6011
-
root.dentry = nullfs_mnt->mnt_root;
6018
+
/* Mount mutable rootfs on top of nullfs. */
6019
+
root.mnt = nullfs_mnt;
6020
+
root.dentry = nullfs_mnt->mnt_root;
6012
6021
6013
-
LOCK_MOUNT_EXACT(mp, &root);
6014
-
if (unlikely(IS_ERR(mp.parent)))
6015
-
panic("VFS: Failed to mount rootfs on nullfs");
6016
-
scoped_guard(mount_writer)
6017
-
attach_mnt(real_mount(mnt), mp.parent, mp.mp);
6022
+
LOCK_MOUNT_EXACT(mp, &root);
6023
+
if (unlikely(IS_ERR(mp.parent)))
6024
+
panic("VFS: Failed to mount rootfs on nullfs");
6025
+
scoped_guard(mount_writer)
6026
+
attach_mnt(real_mount(mnt), mp.parent, mp.mp);
6018
6027
6019
-
pr_info("VFS: Finished mounting rootfs on nullfs\n");
6020
-
} else {
6021
-
VFS_WARN_ON_ONCE(real_mount(mnt)->mnt_id != 1);
6022
-
6023
-
/* The namespace root is the mutable rootfs. */
6024
-
mnt_root = real_mount(mnt);
6025
-
init_mnt_ns.root = mnt_root;
6026
-
}
6028
+
pr_info("VFS: Finished mounting rootfs on nullfs\n");
6027
6029
6028
6030
/*
6029
6031
* We've dropped all locks here but that's fine. Not just are we
+7
-13
init/do_mounts.c
+7
-13
init/do_mounts.c
···
493
493
out:
494
494
devtmpfs_mount();
495
495
496
-
if (nullfs_rootfs) {
497
-
if (init_pivot_root(".", ".")) {
498
-
pr_err("VFS: Failed to pivot into new rootfs\n");
499
-
return;
500
-
}
501
-
if (init_umount(".", MNT_DETACH)) {
502
-
pr_err("VFS: Failed to unmount old rootfs\n");
503
-
return;
504
-
}
505
-
pr_info("VFS: Pivoted into new rootfs\n");
496
+
if (init_pivot_root(".", ".")) {
497
+
pr_err("VFS: Failed to pivot into new rootfs\n");
506
498
return;
507
499
}
508
-
509
-
init_mount(".", "/", NULL, MS_MOVE, NULL);
510
-
init_chroot(".");
500
+
if (init_umount(".", MNT_DETACH)) {
501
+
pr_err("VFS: Failed to unmount old rootfs\n");
502
+
return;
503
+
}
504
+
pr_info("VFS: Pivoted into new rootfs\n");
511
505
}
512
506
513
507
static bool is_tmpfs;