Re: [PATCH] Make CONFIG_DEVTMPFS_MOUNT apply to initramfs/initmpfs.

From: Rich Felker
Date: Thu Jun 23 2016 - 15:39:19 EST


On Wed, Jun 22, 2016 at 02:06:45PM -0500, Rob Landley wrote:
> From: Rob Landley <rob@xxxxxxxxxxx>
>
> Make CONFIG_DEVTMPFS_MOUNT apply to initramfs/initmpfs.
>
> Update help text, slightly improve error reporting, move /dev/console open
> down after devtmpfs mount, don't check IS_ENABLED(CONFIG_TMPFS) before
> mounting devtmpfs (it's always there, even if just a ramfs alias), and
> report whether we think we're using tmpfs or ramfs for rootfs.
>
> Signed-off-by: Rob Landley <rob@xxxxxxxxxxx>

Comments below:

> ---
>
> drivers/base/Kconfig | 10 ++++++----
> drivers/base/devtmpfs.c | 3 ++-
> init/do_mounts.c | 7 +++----
> init/main.c | 17 ++++++++++-------
> 4 files changed, 21 insertions(+), 16 deletions(-)
>
> diff --git a/drivers/base/Kconfig b/drivers/base/Kconfig
> index 98504ec..3fb07c1 100644
> --- a/drivers/base/Kconfig
> +++ b/drivers/base/Kconfig
> @@ -45,16 +45,18 @@ config DEVTMPFS
> file system will be used instead.
>
> config DEVTMPFS_MOUNT
> - bool "Automount devtmpfs at /dev, after the kernel mounted the rootfs"
> + bool "Automount devtmpfs at /dev"
> depends on DEVTMPFS
> help
> This will instruct the kernel to automatically mount the
> devtmpfs filesystem at /dev, directly after the kernel has
> mounted the root filesystem. The behavior can be overridden
> with the commandline parameter: devtmpfs.mount=0|1.
> - This option does not affect initramfs based booting, here
> - the devtmpfs filesystem always needs to be mounted manually
> - after the rootfs is mounted.
> +
> + In an initramfs based system, this can create the /dev directory
> + as well. Other root filesystems require a /dev directory to exist
> + to act as a mount point.
> +
> With this option enabled, it allows to bring up a system in
> rescue mode with init=/bin/sh, even when the /dev directory
> on the rootfs is completely empty.
> diff --git a/drivers/base/devtmpfs.c b/drivers/base/devtmpfs.c
> index 44a74cf..eaf8532 100644
> --- a/drivers/base/devtmpfs.c
> +++ b/drivers/base/devtmpfs.c
> @@ -356,7 +356,8 @@ int devtmpfs_mount(const char *mntdir)
>
> err = sys_mount("devtmpfs", (char *)mntdir, "devtmpfs", MS_SILENT, NULL);
> if (err)
> - printk(KERN_INFO "devtmpfs: error mounting %i\n", err);
> + printk(KERN_ERR "devtmpfs: error %i mounting on %s\n",
> + err, mntdir);

Could stand on its own as:

"Fix loglevel for devtmpfs mounting error message."

> else
> printk(KERN_INFO "devtmpfs: mounted\n");
> return err;
> diff --git a/init/do_mounts.c b/init/do_mounts.c
> index dea5de9..6daf63e 100644
> --- a/init/do_mounts.c
> +++ b/init/do_mounts.c
> @@ -599,7 +599,6 @@ void __init prepare_namespace(void)
>
> mount_root();
> out:
> - devtmpfs_mount("dev");
> sys_mount(".", "/", NULL, MS_MOVE, NULL);
> sys_chroot(".");
> }
> @@ -614,8 +613,9 @@ static struct dentry *rootfs_mount(struct file_system_type *fs_type,
> if (test_and_set_bit(0, &once))
> return ERR_PTR(-ENODEV);
>
> - if (IS_ENABLED(CONFIG_TMPFS) && is_tmpfs)
> + if (is_tmpfs)
> fill = shmem_fill_super;

Perhaps independent, as:

"Fix incorrect use of CONFIG_TMPFS"?

Can you describe the problem this change is fixing? I believe it's
correct (since tmpfs exists with or without CONFIG_TMPFS, and
CONFIG_TMPFS is always disabled for nommu) but it would be nice to
understand the motivation.

> + printk(KERN_INFO "rootfs is %s\n", is_tmpfs ? "tmpfs" : "ramfs");
>
> return mount_nodev(fs_type, flags, data, fill);
> }
> @@ -637,9 +637,8 @@ int __init init_rootfs(void)
> (!root_fs_names || strstr(root_fs_names, "tmpfs"))) {
> err = shmem_init();
> is_tmpfs = true;
> - } else {
> + } else
> err = init_ramfs_fs();
> - }

This is gratuitous and could be dropped.

> if (err)
> unregister_filesystem(&rootfs_fs_type);
> diff --git a/init/main.c b/init/main.c
> index b3c6e36..e6bf551 100644
> --- a/init/main.c
> +++ b/init/main.c
> @@ -1007,12 +1007,6 @@ static noinline void __init kernel_init_freeable(void)
>
> do_basic_setup();
>
> - /* Open the /dev/console on the rootfs, this should never fail */
> - if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
> - pr_err("Warning: unable to open an initial console.\n");
> -
> - (void) sys_dup(0);
> - (void) sys_dup(0);
> /*
> * check if there is an early userspace init. If yes, let it do all
> * the work
> @@ -1024,7 +1018,16 @@ static noinline void __init kernel_init_freeable(void)
> if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
> ramdisk_execute_command = NULL;
> prepare_namespace();
> - }
> + } else if (config_enabled(CONFIG_DEVTMPFS_MOUNT))
> + sys_mkdir("/dev", 0755);

This could probably be a separate change or even omitted. There's no
reason the dir entry can't be in the initramfs cpio archive. (On the
other hand there _are_ good reasons not to put device nodes in the
cpio archive.)

> + devtmpfs_mount("dev");
> +
> + /* Open the /dev/console on the rootfs, this should never fail */
> + if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
> + pr_err("Warning: unable to open an initial console.\n");
> +
> + (void) sys_dup(0);
> + (void) sys_dup(0);

Moving this breaks use of a ramdisk_execute_command (see above) -- it
runs with no stdin/out/err. It would probably be preferable to leave
the console opening where it is, but find some way to open the console
device directly without the need for an actual device node in the
filesystem.

Rich