Bug 33952

Closing, if this is still affecting you on a newer kernel please reopen.

Reprodused on my system with kernel 3.9.0 and rootfs on btrfs-raid0:
egorov-ey ~ # btrfs filesystem show --all-devices /dev/sdd
Label: 'BTRFS-RAID0'  uuid: da81a15f-c4d7-4252-8673-6504aadd074e
        Total devices 2 FS bytes used 25.94GB
        devid    1 size 232.89GB used 19.03GB path /dev/sdc
        devid    2 size 232.89GB used 19.01GB path /dev/sdd

Btrfs v0.20-rc1-253-g7854c8b

kernel can not mount fs without initramfs and with commandline
rootfstype=btrfs root/dev/sdc rootflags=device=/dev/sdd,subvol=/ROOT

>root=/dev/sdc
fix typo

Ok this is because you need to fix your initramfs to run btrfs device scan before it tries to mount the file system, then it will work fine.

(In reply to Josef Bacik from comment #4)
> Ok this is because you need to fix your initramfs to run btrfs device scan
> before it tries to mount the file system, then it will work fine.

Hello. What you think about patch from this thread?
https://forums.gentoo.org/viewtopic-t-923554-start-0.html

this patch solve the problem in my case

diff --git a/init/do_mounts.c b/init/do_mounts.c
index 82f2288..6d58f8d 100644
--- a/init/do_mounts.c
+++ b/init/do_mounts.c
@@ -484,7 +484,10 @@ void __init change_floppy(char *fmt, ...)
        va_start(args, fmt);
        vsprintf(buf, fmt, args);
        va_end(args);
-       fd = sys_open("/dev/root", O_RDWR | O_NDELAY, 0);
+       if (saved_root_name[0])
+               fd = sys_open(saved_root_name, O_RDWR | O_NDELAY, 0);
+       else
+               fd = sys_open("/dev/root", O_RDWR | O_NDELAY, 0);
        if (fd >= 0) {
                sys_ioctl(fd, FDEJECT, 0);
                sys_close(fd);
@@ -528,7 +531,14 @@ void __init mount_root(void)
 #endif
 #ifdef CONFIG_BLOCK
        create_dev("/dev/root", ROOT_DEV);
-       mount_block_root("/dev/root", root_mountflags);
+       if (saved_root_name[0]) {
+               create_dev(saved_root_name, ROOT_DEV);
+               devtmpfs_mount("dev");
+               mount_block_root(saved_root_name, root_mountflags);
+       } else {
+               create_dev("/dev/root", ROOT_DEV);
+               mount_block_root("/dev/root", root_mountflags);
+       }
 #endif
 }

Egor, thanks for the tips but I myself have serious doubts about your patch. Could you document a bit what you did and why was it necessary? I'm not a kernel expert but the following looks suspicious to me:

1. in the 'else' branch, create_dev("/dev/root", ROOT_DEV) is called second time with the same arguments as before the 'if'... is that intentional?

2. What purpose does using saved_root_name serve? I don't see how it is relevant here -- and AFAIU mounting devtmpfs will result in the device being created as well.

3. Can devtmpfs_mount() at this point have some side effects?


That said, I was able to get my multi-device btrfs mounted using a simpler patch:

diff --git a/init/do_mounts.c b/init/do_mounts.c
index e35fb52..434a54e 100644
--- a/init/do_mounts.c
+++ b/init/do_mounts.c
@@ -528,6 +528,7 @@ void __init mount_root(void)
        }
 #endif
 #ifdef CONFIG_BLOCK
+       devtmpfs_mount("dev");
        create_dev("/dev/root", ROOT_DEV);
        mount_block_root("/dev/root", root_mountflags);
 #endif

(In reply to Michał Górny from comment #7)
> Egor, thanks for the tips but I myself have serious doubts about your patch.
> Could you document a bit what you did and why was it necessary? I'm not a
> kernel expert but the following looks suspicious to me:
> 
> 1. in the 'else' branch, create_dev("/dev/root", ROOT_DEV) is called second
> time with the same arguments as before the 'if'... is that intentional?
> 2. What purpose does using saved_root_name serve? I don't see how it is
> relevant here -- and AFAIU mounting devtmpfs will result in the device being
> created as well.

I'm not a kernel expert too :) This is part of genpatches included in gentoo-sources. 
http://sources.gentoo.org/cgi-bin/viewvc.cgi/linux-patches/genpatches-2.6/trunk/3.15/2900_dev-root-proc-mount-fix.patch?revision=2771&view=markup

You can ask about this mpagano. 
And here: https://bugs.gentoo.org/show_bug.cgi?id=438380

> 3. Can devtmpfs_mount() at this point have some side effects?
> That said, I was able to get my multi-device btrfs mounted using a simpler
> patch:
> 
> diff --git a/init/do_mounts.c b/init/do_mounts.c
> index e35fb52..434a54e 100644
> --- a/init/do_mounts.c
> +++ b/init/do_mounts.c
> @@ -528,6 +528,7 @@ void __init mount_root(void)
>         }
>  #endif
>  #ifdef CONFIG_BLOCK
> +       devtmpfs_mount("dev");
>         create_dev("/dev/root", ROOT_DEV);
>         mount_block_root("/dev/root", root_mountflags);
>  #endif
Yes, this path is working as well.

The very fact of the mount I think it is unlikely to have a negative effect. But in some cases this can cause twice mounted devtmpfs: before mount root and after. So this patch is not ideal. May be should introduce new kernel-option: CONFIG_DEVTMPFS_PREMOUNT by analogy with CONFIG_DEVTMPFS_MOUNT and rework logic of mounting.

> i'm looking for Egor Y. Egorov - any chance you know where to find him? :)
Do not hesitate to write me email or xmmp: eegorov@jabber.ru ;)

Well, you definitely don't want to be mixing two patches together here. As far as I understand, the goal of genpatches patch is to not have /dev/root and use the root directly.

On top of that patch, I think it's enough to replace create_dev(saved_root_name, ROOT_DEV) with the mount -- devtmpfs will have the proper device anyway.