Re: initramfs: Always do fput() and load modules after rootfs populate

[Stafford Horne]

On Sat, May 06, 2017 at 05:17:04PM +0900, Stafford Horne wrote:
> On Sat, May 06, 2017 at 12:02:32AM -0700, Andrei Vagin wrote:
> > Hi,
> > 
> > I see the following error with this patch:
> > 
> > init/initramfs.c: In function 'populate_rootfs':
> > init/initramfs.c:644:2: error: label at end of compound statement
> >   done:
> >   ^
> > scripts/Makefile.build:294: recipe for target 'init/initramfs.o' failed
> > make[1]: *** [init/initramfs.o] Error 1
> 
> Ahh, I didnt notice this and none of the build robots reported to us.
> Sorry about it, the above patch is already merged, so this is not good, Ill
> get this fixed.
> 
> Could you give me some info on your compiler?

Nvermind, I was able to reproduce enabling:

  CONFIG_BLK_DEV_RAM=y

I should have run my allyesconfig testing, instead of just the defconfig!

-Stafford

> > probably we need something like this:
> > diff --git a/init/initramfs.c b/init/initramfs.c
> > index 3a68715..9966c0f 100644
> > --- a/init/initramfs.c
> > +++ b/init/initramfs.c
> > @@ -641,7 +641,6 @@ static int __init populate_rootfs(void)
> >                         sys_close(fd);
> >                         free_initrd();
> >                 }
> > -       done:
> >  #else
> >                 printk(KERN_INFO "Unpacking initramfs...\n");
> >                 err = unpack_to_rootfs((char *)initrd_start,
> > @@ -651,6 +650,7 @@ static int __init populate_rootfs(void)
> >                 free_initrd();
> >  #endif
> >         }
> > +done:
> >         flush_delayed_fput();
> >         /*
> >          * Try loading default modules from initramfs.  This gives
> > 
> 
> Right, this looks correct for the above error.
> 
> -Stafford