[RFC PATCH] Introduce a 'recovery' command line option

[RFC PATCH] Introduce a 'recovery' command line option

[Janne Karhunen]

Recovery option can be used to define a secondary rootfs
in case mounting of the primary root fails. This allows
the kernel to automatically switch to a recovery
filesystem without the initrd or the bootloader support
for the switch.

Signed-off-by: Janne Karhunen <Janne.Karhunen@gmail.com>
---
 Documentation/kernel-parameters.txt |  4 ++++
 init/do_mounts.c                    | 47 +++++++++++++++++++++++++++++++++----
 2 files changed, 46 insertions(+), 5 deletions(-)

diff --git a/Documentation/kernel-parameters.txt b/Documentation/kernel-parameters.txt
index 46c030a..febbd3e 100644
--- a/Documentation/kernel-parameters.txt
+++ b/Documentation/kernel-parameters.txt
@@ -3587,6 +3587,10 @@ bytes respectively. Such letter suffixes can also be entirely omitted.
 			      reboot_cpu is s[mp]#### with #### being the processor
 					to be used for rebooting.
 
+	recovery= 	[KNL]
+			Recovery root filesystem. This partition is attempted as
+			root in case default root filesystem does not mount.
+
 	relax_domain_level=
 			[KNL, SMP] Set scheduler's default relax_domain_level.
 			See Documentation/cgroup-v1/cpusets.txt.
diff --git a/init/do_mounts.c b/init/do_mounts.c
index dea5de9..c746dce 100644
--- a/init/do_mounts.c
+++ b/init/do_mounts.c
@@ -39,8 +39,11 @@ int __initdata rd_doload;	/* 1 = load RAM disk, 0 = don't load */
 
 int root_mountflags = MS_RDONLY | MS_SILENT;
 static char * __initdata root_device_name;
+static char * __initdata recovery_device_name;
 static char __initdata saved_root_name[64];
+static char __initdata saved_recovery_name[64];
 static int root_wait;
+static int recovery_attempt;
 
 dev_t ROOT_DEV;
 
@@ -298,6 +301,15 @@ static int __init root_dev_setup(char *line)
 
 __setup("root=", root_dev_setup);
 
+static int __init recovery_setup(char *line)
+{
+	strlcpy(saved_recovery_name, line, sizeof(saved_recovery_name));
+	recovery_attempt = 1;
+	return 1;
+}
+
+__setup("recovery=", recovery_setup);
+
 static int __init rootwait_setup(char *str)
 {
 	if (*str)
@@ -384,6 +396,7 @@ void __init mount_block_root(char *name, int flags)
 					__GFP_NOTRACK_FALSE_POSITIVE);
 	char *fs_names = page_address(page);
 	char *p;
+	int err;
 #ifdef CONFIG_BLOCK
 	char b[BDEVNAME_SIZE];
 #else
@@ -393,7 +406,7 @@ void __init mount_block_root(char *name, int flags)
 	get_fs_names(fs_names);
 retry:
 	for (p = fs_names; *p; p += strlen(p)+1) {
-		int err = do_mount_root(name, p, flags, root_mount_data);
+		err = do_mount_root(name, p, flags, root_mount_data);
 		switch (err) {
 			case 0:
 				goto out;
@@ -401,6 +414,31 @@ retry:
 			case -EINVAL:
 				continue;
 		}
+		if (!(flags & MS_RDONLY)) {
+			pr_warn("Retrying rootfs mount as read-only.\n");
+			flags |= MS_RDONLY;
+			goto retry;
+		}
+		if (recovery_device_name && recovery_attempt) {
+			recovery_attempt = 0;
+
+			ROOT_DEV = name_to_dev_t(recovery_device_name);
+			if (strncmp(recovery_device_name, "/dev/", 5) == 0)
+				recovery_device_name += 5;
+
+			pr_warn("Unable to mount rootfs at %s, error %d\n",
+				root_device_name, err);
+			pr_warn("Attempting %s for recovery as requested.\n",
+				recovery_device_name);
+
+			err = create_dev("/dev/root", ROOT_DEV);
+			if (err < 0)
+				pr_emerg("Failed to create /dev/root: %d\n", err);
+
+			root_device_name = recovery_device_name;
+			goto retry;
+		}
+
 	        /*
 		 * Allow the user to distinguish between failed sys_open
 		 * and bad superblock on root device.
@@ -420,10 +458,6 @@ retry:
 #endif
 		panic("VFS: Unable to mount root fs on %s", b);
 	}
-	if (!(flags & MS_RDONLY)) {
-		flags |= MS_RDONLY;
-		goto retry;
-	}
 
 	printk("List of all partitions:\n");
 	printk_all_partitions();
@@ -567,6 +601,9 @@ void __init prepare_namespace(void)
 
 	md_run_setup();
 
+	if (saved_recovery_name[0])
+		recovery_device_name = saved_recovery_name;
+
 	if (saved_root_name[0]) {
 		root_device_name = saved_root_name;
 		if (!strncmp(root_device_name, "mtd", 3) ||
-- 
1.9.1

Re: [RFC PATCH] Introduce a 'recovery' command line option

[Richard Weinberger]

Hi!

On Tue, Aug 16, 2016 at 8:36 AM, Janne Karhunen
<Janne.Karhunen@gmail.com> wrote:
> Recovery option can be used to define a secondary rootfs
> in case mounting of the primary root fails. This allows
> the kernel to automatically switch to a recovery
> filesystem without the initrd or the bootloader support
> for the switch.

Why are you moving this feature into the kernel?
To support such advanced features we have the initramfs.

dracut has already rootfallback=.

-- 
Thanks,
//richard

Re: [RFC PATCH] Introduce a 'recovery' command line option

[Janne Karhunen]

On Tue, Aug 16, 2016 at 10:04 AM, Richard Weinberger
<richard.weinberger@gmail.com> wrote:

> Why are you moving this feature into the kernel?
> To support such advanced features we have the initramfs.
>
> dracut has already rootfallback=.

For saving some precious boot-up time (my systems run without initrd)
and to unify the solutions. If kernel does this bootloaders and
initrds don't have to care.


--
Janne

Re: [RFC PATCH] Introduce a 'recovery' command line option

[Richard Weinberger]

On 16.08.2016 09:15, Janne Karhunen wrote:
> On Tue, Aug 16, 2016 at 10:04 AM, Richard Weinberger
> <richard.weinberger@gmail.com> wrote:
> 
>> Why are you moving this feature into the kernel?
>> To support such advanced features we have the initramfs.
>>
>> dracut has already rootfallback=.
> 
> For saving some precious boot-up time (my systems run without initrd)
> and to unify the solutions. If kernel does this bootloaders and
> initrds don't have to care.

Features - collect them all? ;-)

I my opinion it is not wise to move feature into the kernel which can
be solved perfectly fine in userspace (initramfs).

How much slows an initramfs your boot time down?

Did you try using a handmade minimal initramfs? Not compressed, single c program,
statically linked..., etc.

Thanks,
//richard

Re: [RFC PATCH] Introduce a 'recovery' command line option

[Janne Karhunen]

On Tue, Aug 16, 2016 at 11:59 AM, Richard Weinberger <richard@nod.at> wrote:

>> For saving some precious boot-up time (my systems run without initrd)
>> and to unify the solutions. If kernel does this bootloaders and
>> initrds don't have to care.
>
> Features - collect them all? ;-)

Well if every vendor ends up implementing something different for this
purpose and it can be solved with just a few lines of common kernel
code, IMHO it might just be worth it.


> I my opinion it is not wise to move feature into the kernel which can
> be solved perfectly fine in userspace (initramfs).
>
> How much slows an initramfs your boot time down?
>
> Did you try using a handmade minimal initramfs? Not compressed, single c program,
> statically linked..., etc.

Depends on the initrd and the hardware, of course. Certainly you can
implement the same on initd, but you can never make it as fast as this
one. Moreover, this solution would have a common 'user interface' for
the functionality if anyone cares about the userspace unification
efforts.


-- 
Janne