Re: [4.1-rc1][Hibernation failed] bisect result included

[Mike Snitzer <snitzer@redhat.com>]

On Mon, May 04 2015 at 10:08am -0400,
Rafael J. Wysocki <rjw@rjwysocki.net> wrote:

> On Sunday, May 03, 2015 03:54:39 PM Chen, Yu C wrote:
> > Hi,all,
> > When we are doing QA test, a hibernation failor was found on latest kernel 4.1-rc1,
> > resume from hibernation failed with the following error:
> > 
> > PM: Hibernation image not present or could not be loaded.
> > 
> > Bisect shows following commit might trigger this problem:
> > commit 283e7ad0241155710f99a9f39d13313a53336926
> > 
> > The  reason for failor is that, some distribution use klibc in initrd
> > to restore the image, and klibc will provide kernel with the device 
> > format of major:minor:offset,
> > for example, 8:3:0 represents a swap partition, and klibc will
> > echo 8:3:0 > /sys/power/resume to resume the system manually.
> > However in current implementation, format of 8:3:0 will be regarded 
> > as an invalid device, so restoring from hibernation failed.
> > 
> > Do we need to add support for device format like 8:3:0?
> > I'm happy to debug more specifically if you can let me know what 
> > would be of use, and I've also attached a temporary patch 
> > for this problem, not sure if it is suitable,  thanks.
> 
> It'd be better to send the patch inline so it is not lost in replies.
> Reproduced below.
> 
> Dan, Mike, any chance to look at this?  It is a functional regression for
> some people.

Sorry about this.  Wasn't aware that such a strange format was being
used.  The proposed patch looks good to me (formatting is a bit weird
though, the last "&dummy) == 3)) {" line shouldnt be needed if previous
line isn't indented so far.

Unless other more official init maintainer(s) want to pick this fix up:
since I committed the commit that caused this regression I can pick this
patch up and get it to Linus for 4.1-rc3 inclussion via linux-dm.git.
Please just let me know, thanks!

> ---
> From a65d76c143b315c322b5c7a9fb365703c9f32798 Mon Sep 17 00:00:00 2001
> From: Chen Yu <yu.c.chen@intel.com>
> Date: Sun, 3 May 2015 22:35:05 +0800
> Subject: [RFC] init: support device of major:minor:offset format
> 
> Distribution like Ubuntu uses klibc rather than uswsusp to resume
> system from hibernation, which will treat swap partition/file in
> the form of major:minor:offset. For example, 8:3:0 represents a
> swap partition in klibc, and klibc's resume process in initrd will
> finally echo 8:3:0 to /sys/power/resume for manually restoring.
> However in current implementation, 8:3:0 will be treated as an invalid
> device format, and it is found that manual resumming from hibernation
> will fail on lastest kernel.
> 
> This patch adds support for device with major:minor:offset format
> when resumming from hibernation.
> 
> Reported-by: Prigent, Christophe <christophe.prigent@intel.com>
> 
> Signed-off-by: Chen Yu <yu.c.chen@intel.com>
> ---
>  init/do_mounts.c | 6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
> 
> diff --git a/init/do_mounts.c b/init/do_mounts.c
> index 8369ffa..2d36bf1 100644
> --- a/init/do_mounts.c
> +++ b/init/do_mounts.c
> @@ -225,10 +225,12 @@ dev_t name_to_dev_t(const char *name)
>  #endif
>  
>  	if (strncmp(name, "/dev/", 5) != 0) {
> -		unsigned maj, min;
> +		unsigned maj, min, offset;
>  		char dummy;
>  
> -		if (sscanf(name, "%u:%u%c", &maj, &min, &dummy) == 2) {
> +		if ((sscanf(name, "%u:%u%c", &maj, &min, &dummy) == 2) ||
> +			(sscanf(name, "%u:%u:%u:%c", &maj, &min, &offset,
> +			&dummy) == 3)) {
>  			res = MKDEV(maj, min);
>  			if (maj != MAJOR(res) || min != MINOR(res))
>  				goto fail;
> -- 
> 1.9.1
> 
>