[PATCH v4 4/5] kexec: add option to fall back to KEXEC_LOAD when KEXEC_FILE_LOAD is not supported.

[Michal Suchánek]

On Wed, 14 Mar 2018 11:21:59 +0800
Dave Young <dyoung at redhat.com> wrote:

> On 03/06/18 at 02:15pm, Michal Suchanek wrote:
> > Not all architectures implement KEXEC_FILE_LOAD. However, on some
> > archiectures KEXEC_FILE_LOAD is required when secure boot is
> > enabled in locked-down mode. Previously users had to select the
> > KEXEC_FILE_LOAD syscall with undocumented -s option. However, if
> > they did pass the option kexec would fail on architectures that do
> > not support it.
> > 
> > So add an -a option that tries KEXEC_FILE_LOAD and when it is not
> > supported tries KEXEC_LOAD.
> > 
> > Signed-off-by: Michal Suchanek <msuchanek at suse.de>
> > ---
> > v3: instead of changing the deafult add extra option
> > v4: actually check -ENOSYS as well
> > ---
> >  kexec/kexec.c | 52
> > ++++++++++++++++++++++++++++++++++++++++++++++++---- kexec/kexec.h
> > |  4 +++- 2 files changed, 51 insertions(+), 5 deletions(-)
> > 
> > diff --git a/kexec/kexec.c b/kexec/kexec.c
> > index a95cfb473d6b..5c5aee344b41 100644
> > --- a/kexec/kexec.c
> > +++ b/kexec/kexec.c
> > @@ -1243,6 +1243,7 @@ int main(int argc, char *argv[])
> >  	int do_unload = 0;
> >  	int do_reuse_initrd = 0;
> >  	int do_kexec_file_syscall = 0;
> > +	int do_kexec_fallback = 0;
> >  	int do_status = 0;
> >  	void *entry = 0;
> >  	char *type = 0;
> > @@ -1367,10 +1368,15 @@ int main(int argc, char *argv[])
> >  			break;
> >  		case OPT_KEXEC_FILE_SYSCALL:
> >  			do_kexec_file_syscall = 1;
> > +			do_kexec_fallback = 0;
> >  			break;
> >  		case OPT_KEXEC_SYSCALL:
> >  			do_kexec_file_syscall = 0;
> > +			do_kexec_fallback = 0;
> >  			break;
> > +		case OPT_KEXEC_SYSCALL_AUTO:
> > +			do_kexec_file_syscall = 1;
> > +			do_kexec_fallback = 1;  
> 
> need a break here

Indeed

> 
> >  		case OPT_STATUS:
> >  			do_status = 1;
> >  			break;
> > @@ -1442,16 +1448,54 @@ int main(int argc, char *argv[])
> >  		result = k_status(kexec_flags);
> >  	}
> >  	if (do_unload) {
> > -		if (do_kexec_file_syscall)
> > +		if (do_kexec_file_syscall) {
> >  			result =
> > kexec_file_unload(kexec_file_flags);
> > -		else
> > +			if ((result == -ENOSYS) &&
> > do_kexec_fallback)
> > +				do_kexec_file_syscall = 0;
> > +		}
> > +		if (!do_kexec_file_syscall)
> >  			result = k_unload(kexec_flags);
> >  	}
> >  	if (do_load && (result == 0)) {
> > -		if (do_kexec_file_syscall)
> > +		if (do_kexec_file_syscall) {
> >  			result = do_kexec_file_load(fileind, argc,
> > argv, kexec_file_flags);
> > -		else
> > +			if (do_kexec_fallback) switch (result) {
> > +				/*
> > +				 * Something failed with signature
> > verification.
> > +				 * Reject the image.
> > +				 */
> > +				case -ELIBBAD:
> > +				case -EKEYREJECTED:
> > +				case -ENOPKG:
> > +				case -ENOKEY:
> > +				case -EBADMSG:
> > +				case -EMSGSIZE:
> > +					/*
> > +					 * By default reject or do
> > nothing if
> > +					 * succeded
> > +					 */
> > +				default: break;
> > +				case -ENOSYS: /* not implemented */
> > +					/*
> > +					 * Parsing image or other
> > options failed
> > +					 * The image may be
> > invalid or image
> > +					 * type may not supported
> > by kernel so
> > +					 * retry parsing in
> > kexec-tools.
> > +					 */
> > +				case -EINVAL:
> > +				case -ENOEXEC:
> > +					 /*
> > +					  * ENOTSUPP can be
> > unsupported image
> > +					  * type or unsupported PE
> > signature
> > +					  * wrapper type, duh
> > +					  */
> > +				case -ENOTSUP:
> > +					do_kexec_file_syscall = 0;
> > +					break;  
> 
> It looks to me it is enough only checking -ENOSYS maybe also
> -ENOTSUPP and then set do_kexec_file_syscall = 0;
> 
> EINVAL and ENOEXEC are real errors, I do not understand why still 
> fallback.  

If you pass an image type that the kernel does not understand (eg.
multiboot or uImage) then the kernel will return a real error because
it does not understand the image. However, kexec-tools should still be
able to load it, automatically. That's what the -auto stands for.

> Also thos signature verification errors are not needed
> in this code as well.

Yes, they are not needed. They are here so it's obvious which errors
are signature verification errors.

Thanks

Michal