initmpfs v2: use tmpfs instead of ramfs for rootfs

PATCH In review Rob Landley 2 participants 8 messages Jul 16, 2013 – Jul 26, 2013

Subsystems (4)

Memory management Documentation VFS tmpfs (shmem filesystem)

lore

Share & tools

lore

Rob Landley

author

2013-07-16 23:45 UTC #0 lore raw

cover initmpfs v2: use tmpfs instead of ramfs for rootfs

Use tmpfs for rootfs when CONFIG_TMPFS=y and there's no root=. Specify rootfstype=ramfs to get the old initramfs behavior.

The previous initramfs code provided a fairly crappy root filesystem:
didn't let you --bind mount directories out of it, reported zero size/usage so it didn't show up in "df" and couldn't run things like rpm that query available space before proceeding, would fill up all available memory and panic the system if you wrote too much to it...

Using tmpfs instead provides a much better root filesystem.

Changes from v1: use test_and_set_bit() for "once" logic.

Changes from this morning's send: none, just hopefully not screwing up the message-id this time trying to make it a reply to another message via cut and paste...

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

Rob Landley

author

2013-07-16 23:45 UTC #1 lore raw

PATCH 1/5 initmpfs v2: replace MS_NOUSER in initramfs

From: Rob Landley <[email protected]>

Mounting MS_NOUSER prevents --bind mounts from rootfs. Prevent new rootfs mounts with a different mechanism that doesn't affect bind mounts.

Signed-off-by Rob Landley <[email protected]>

---

 fs/ramfs/inode.c |    7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

fs/ramfs/inode.c +6 -1

index c24f1e1..14b9c35 100644
@@ -247,7 +247,12 @@ struct dentry *ramfs_mount(struct file_system_type *fs_type,
 static struct dentry *rootfs_mount(struct file_system_type *fs_type,
 	int flags, const char *dev_name, void *data)
 {
-	return mount_nodev(fs_type, flags|MS_NOUSER, data, ramfs_fill_super);
+	static unsigned long once;
+
+	if (test_and_set_bit(1, &once))
+		return ERR_PTR(-ENODEV);
+
+	return mount_nodev(fs_type, flags, data, ramfs_fill_super);
 }
 
 static void ramfs_kill_sb(struct super_block *sb)

Rob Landley

author

2013-07-16 23:45 UTC #2 lore raw

PATCH 2/5 initmpfs v2: Move bdi setup from init_rootfs to init_ramfs

In reply to Rob Landley

From: Rob Landley <[email protected]>

Even though ramfs hasn't got a backing device, commit e0bf68ddec4f added one anyway, and put the initialization in init_rootfs() since that's the first user, leaving it out of init_ramfs() to avoid duplication.

But initmpfs uses init_tmpfs() instead, so move the init into the filesystem's init function, add a "once" guard to prevent duplicate initialization, and call the filesystem init from rootfs init.

This goes part of the way to allowing ramfs to be built as a module.

Signed-off-by Rob Landley <[email protected]>

---

 fs/ramfs/inode.c |   23 ++++++++++++++++++-----
 1 file changed, 18 insertions(+), 5 deletions(-)

fs/ramfs/inode.c +18 -5

@@ -275,21 +275,34 @@
 
 static int __init init_ramfs_fs(void)
 {
-	return register_filesystem(&ramfs_fs_type);
+	static unsigned long once;
+	int err;
+
+	if (test_and_set_bit(1, &once))
+		return 0;
+
+	err = bdi_init(&ramfs_backing_dev_info);
+	if (err)
+		return err;
+
+	err = register_filesystem(&ramfs_fs_type);
+	if (err)
+		bdi_destroy(&ramfs_backing_dev_info);
+
+	return err;
 }
 module_init(init_ramfs_fs)
 
 int __init init_rootfs(void)
 {
-	int err;
+	int err = register_filesystem(&rootfs_fs_type);
 
-	err = bdi_init(&ramfs_backing_dev_info);
 	if (err)
 		return err;
 
-	err = register_filesystem(&rootfs_fs_type);
+	err = init_ramfs_fs();
 	if (err)
-		bdi_destroy(&ramfs_backing_dev_info);
+		unregister_filesystem(&rootfs_fs_type);
 
 	return err;
 }

Rob Landley

author

2013-07-16 23:45 UTC #3 lore raw

PATCH 3/5 initmpfs v2: Move rootfs code from fs/ramfs/ to init/

In reply to Rob Landley

From: Rob Landley <[email protected]>

When the rootfs code was a wrapper around ramfs, having them in the same file made sense. Now that it can wrap another filesystem type, move it in with the init code instead.

This also allows a subsequent patch to access rootfstype= command line arg.

Signed-off-by Rob Landley <[email protected]>

---

 fs/namespace.c        |    2 +-
 fs/ramfs/inode.c      |   32 +-------------------------------
 include/linux/init.h  |    1 +
 include/linux/ramfs.h |    2 +-
 init/do_mounts.c      |   32 ++++++++++++++++++++++++++++++++
 5 files changed, 36 insertions(+), 33 deletions(-)

fs/namespace.c +1 -1

@@ -17,7 +17,7 @@
 #include <linux/security.h>
 #include <linux/idr.h>
 #include <linux/acct.h>		/* acct_auto_close_mnt */
-#include <linux/ramfs.h>	/* init_rootfs */
+#include <linux/init.h>		/* init_rootfs */
 #include <linux/fs_struct.h>	/* get_fs_root et.al. */
 #include <linux/fsnotify.h>	/* fsnotify_vfsmount_delete */
 #include <linux/uaccess.h>

fs/ramfs/inode.c +1 ~~-31~~

@@ -244,17 +244,6 @@
 	return mount_nodev(fs_type, flags, data, ramfs_fill_super);
 }
 
-static struct dentry *rootfs_mount(struct file_system_type *fs_type,
-	int flags, const char *dev_name, void *data)
-{
-	static unsigned long once;
-
-	if (test_and_set_bit(1, &once))
-		return ERR_PTR(-ENODEV);
-
-	return mount_nodev(fs_type, flags, data, ramfs_fill_super);
-}
-
 static void ramfs_kill_sb(struct super_block *sb)
 {
 	kfree(sb->s_fs_info);

@@ -267,13 +256,8 @@
 	.kill_sb	= ramfs_kill_sb,
 	.fs_flags	= FS_USERNS_MOUNT,
 };
-static struct file_system_type rootfs_fs_type = {
-	.name		= "rootfs",
-	.mount		= rootfs_mount,
-	.kill_sb	= kill_litter_super,
-};
 
-static int __init init_ramfs_fs(void)
+int __init init_ramfs_fs(void)
 {
 	static unsigned long once;
 	int err;

@@ -292,17 +276,3 @@
 	return err;
 }
 module_init(init_ramfs_fs)
-
-int __init init_rootfs(void)
-{
-	int err = register_filesystem(&rootfs_fs_type);
-
-	if (err)
-		return err;
-
-	err = init_ramfs_fs();
-	if (err)
-		unregister_filesystem(&rootfs_fs_type);
-
-	return err;
-}

include/linux/init.h +1 -0

@@ -153,6 +153,7 @@
 void setup_arch(char **);
 void prepare_namespace(void);
 void __init load_default_modules(void);
+int __init init_rootfs(void);
 
 extern void (*late_time_init)(void);

include/linux/ramfs.h +1 -1

@@ -25,7 +25,7 @@
 
 extern const struct file_operations ramfs_file_operations;
 extern const struct vm_operations_struct generic_file_vm_ops;
-extern int __init init_rootfs(void);
+extern int __init init_ramfs_fs(void);
 
 int ramfs_fill_super(struct super_block *sb, void *data, int silent);

init/do_mounts.c +32 -0

@@ -26,6 +26,7 @@
 #include <linux/async.h>
 #include <linux/fs_struct.h>
 #include <linux/slab.h>
+#include <linux/ramfs.h>
 
 #include <linux/nfs_fs.h>
 #include <linux/nfs_fs_sb.h>

@@ -588,3 +589,34 @@
 	sys_mount(".", "/", NULL, MS_MOVE, NULL);
 	sys_chroot(".");
 }
+
+static struct dentry *rootfs_mount(struct file_system_type *fs_type,
+	int flags, const char *dev_name, void *data)
+{
+	static unsigned long once;
+
+	if (test_and_set_bit(1, &once))
+		return ERR_PTR(-ENODEV);
+
+	return mount_nodev(fs_type, flags, data, ramfs_fill_super);
+}
+
+static struct file_system_type rootfs_fs_type = {
+	.name		= "rootfs",
+	.mount		= rootfs_mount,
+	.kill_sb	= kill_litter_super,
+};
+
+int __init init_rootfs(void)
+{
+	int err = register_filesystem(&rootfs_fs_type);
+
+	if (err)
+		return err;
+
+	err = init_ramfs_fs();
+	if (err)
+		unregister_filesystem(&rootfs_fs_type);
+
+	return err;
+}

fs/ramfs/inode.c +0 -0

index c24f1e1..3b9f114 100644

Rob Landley

author

2013-07-16 23:45 UTC #4 lore raw

PATCH 4/5 initmpfs v2: Make rootfs use tmpfs when CONFIG_TMPFS enabled.

In reply to Rob Landley

From: Rob Landley <[email protected]>

Conditionally call the appropriate fs_init function and fill_super functions. Add a use once guard to shmem_init() to simply succeed on a second call.

(Note that IS_ENABLED() is a compile time constant so dead code elimination removes unused function calls when CONFIG_TMPFS is disabled.)

Signed-off-by Rob Landley <[email protected]>

---

 init/do_mounts.c |   10 ++++++++--
 mm/shmem.c       |    4 ++++
 2 files changed, 12 insertions(+), 2 deletions(-)

init/do_mounts.c +8 -2

@@ -27,6 +27,7 @@
 #include <linux/fs_struct.h>
 #include <linux/slab.h>
 #include <linux/ramfs.h>
+#include <linux/shmem_fs.h>
 
 #include <linux/nfs_fs.h>
 #include <linux/nfs_fs_sb.h>

@@ -598,7 +597,8 @@
 	if (test_and_set_bit(1, &once))
 		return ERR_PTR(-ENODEV);
 
-	return mount_nodev(fs_type, flags, data, ramfs_fill_super);
+	return mount_nodev(fs_type, flags, data,
+		IS_ENABLED(CONFIG_TMPFS) ? shmem_fill_super : ramfs_fill_super);
 }
 
 static struct file_system_type rootfs_fs_type = {

@@ -614,7 +614,11 @@
 	if (err)
 		return err;
 
-	err = init_ramfs_fs();
+	if (IS_ENABLED(CONFIG_TMPFS))
+		err = shmem_init();
+	else
+		err = init_ramfs_fs();
+
 	if (err)
 		unregister_filesystem(&rootfs_fs_type);

mm/shmem.c +4 -0

@@ -2816,6 +2816,10 @@
 {
 	int error;
 
+	/* If rootfs called this, don't re-init */
+	if (shmem_inode_cachep)
+		return 0;
+
 	error = bdi_init(&shmem_backing_dev_info);
 	if (error)
 		goto out4;

Rob Landley

author

2013-07-16 23:45 UTC #5 lore raw

PATCH 5/5 initmpfs v2: Use initramfs if rootfstype= or root= specified.

In reply to Rob Landley

From: Rob Landley <[email protected]>

Command line option rootfstype=ramfs to obtain old initramfs behavior, and use ramfs instead of tmpfs for stub when root= defined (for cosmetic reasons).

Signed-off-by Rob Landley <[email protected]>

---

 init/do_mounts.c |   15 +++++++++++----
 1 file changed, 11 insertions(+), 4 deletions(-)

init/do_mounts.c +11 -4

@@ -591,16 +591,20 @@
 	sys_chroot(".");
 }
 
+static bool is_tmpfs;
 static struct dentry *rootfs_mount(struct file_system_type *fs_type,
 	int flags, const char *dev_name, void *data)
 {
 	static unsigned long once;
+	void *fill = ramfs_fill_super;
 
 	if (test_and_set_bit(1, &once))
 		return ERR_PTR(-ENODEV);
 
-	return mount_nodev(fs_type, flags, data,
-		IS_ENABLED(CONFIG_TMPFS) ? shmem_fill_super : ramfs_fill_super);
+	if (IS_ENABLED(CONFIG_TMPFS) && is_tmpfs)
+		fill = shmem_fill_super;
+
+	return mount_nodev(fs_type, flags, data, fill);
 }
 
 static struct file_system_type rootfs_fs_type = {

@@ -616,9 +620,12 @@
 	if (err)
 		return err;
 
-	if (IS_ENABLED(CONFIG_TMPFS))
+	if (IS_ENABLED(CONFIG_TMPFS) && !saved_root_name[0] &&
+		(!root_fs_names || strstr(root_fs_names, "tmpfs")))
+	{
 		err = shmem_init();
-	else
+		is_tmpfs = true;
+	} else
 		err = init_ramfs_fs();
 
 	if (err)

Andrew Morton

maintainer

2013-07-19 19:57 UTC #6 lore raw

Re: [PATCH 5/5] initmpfs v2: Use initramfs if rootfstype= or root= specified.

Quoted 3 lines · On Tue, 16 Jul 2013 16:45:39 -0700 (PDT) Rob Landley <[email protected]> wrote:

Command line option rootfstype=ramfs to obtain old initramfs behavior, and use ramfs instead of tmpfs for stub when root= defined (for cosmetic reasons).

Could we get a Documentation/kernel-parameters.txt update please?

Rob Landley

author

2013-07-26 04:12 UTC #7 lore raw

Re: [PATCH 5/5] initmpfs v2: Use initramfs if rootfstype= or root= specified.

Quoted 9 lines · Andrew Morton

Quoted 1 line · On Tue, 16 Jul 2013 16:45:39 -0700 (PDT) Rob Landley <[email protected]> wrote:
Command line option rootfstype=ramfs to obtain old initramfs
behavior,
Quoted 1 line
and use ramfs instead of tmpfs for stub when root= defined (for
cosmetic
Quoted 1 line
reasons).
Could we get a Documentation/kernel-parameters.txt update please?

Sorry for the delay, traveling. Behind on email...

"rootfstype" is already documented in kernel parameters:

rootfstype= [KNL] Set root filesystem type

I just applied the existing definition to initramfs now that it has more than one filesystem option there. Do you want me to add a special case here to say that rootfstype= still works when the root filesystem is initramfs? Or would it instead make more sense to add:

Signed-off-by Rob Landley <[email protected]>

Document that rootfstype= applies to initmpfs too.

Documentation/filesystems/ramfs-rootfs-initramfs.txt +0 -0

@@ -79,6 +79,9 @@ to just make sure certain lists can't become empty.
 Most systems just mount another filesystem over rootfs and ignore

it. The
amount of space an empty instance of ramfs takes up is tiny.

+If CONFIG_TMPFS is enabled, rootfs will use tmpfs instead of ramfs by default. +To force ramfs, add "rootfstype=ramfs" to the kernel command line. +
What is initramfs?
------------------

(What I _didn't_ do yet is hook up rootflags= to initmpfs, so you can specify size= as something other than 50%. Partly because if you have an empty cpio archive and the rootflags= applies to ext3 or something, those flags could potentially confuse tmpfs with unknown options and throw errors. And partly because mount -o remount,size=20% works fine after the fact, so it's not time critical and I easily can do a follow-up patch. The one potential downside is if you want to have a cpio archive eat more than 50% of the kernel's memory, it'll fail to extract into tmpfs with the size limits. But the rootfstype=ramfs downgrade also works around that for now...)

There are a number of follow up patches I could do, but the basic functionality doesn't depend on them...

Rob