[PATCH] ima: define an init_module critical data record

[PATCH] ima: define an init_module critical data record

[Mimi Zohar]

The init_module syscall loads an ELF image into kernel space without
measuring the buffer containing the ELF image.  To close this kernel
module integrity gap, define a new critical-data record which includes
the hash of the ELF image.

Instead of including the buffer data in the IMA measurement list,
include the hash of the buffer data to avoid large IMA measurement
list records.  The buffer data hash would be the same value as the
finit_module syscall file hash.

To enable measuring the init_module buffer and other critical data from
boot, define "ima_policy=critical_data" on the boot command line.  Since
builtin policies are not persistent, a custom IMA policy must include
the rule as well: measure func=CRITICAL_DATA label=modules

To verify the template data hash value, first convert the buffer data
hash to binary:
grep "init_module" \
	/sys/kernel/security/integrity/ima/ascii_runtime_measurements | \
	tail -1 | cut -d' ' -f 6 | xxd -r -p | sha256sum

Reported-by: Ken Goldman <kgold@linux.ibm.com>
Signed-off-by: Mimi Zohar <zohar@linux.ibm.com>
---
 security/integrity/ima/ima_main.c | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/security/integrity/ima/ima_main.c b/security/integrity/ima/ima_main.c
index c84e8c55333d..4b4348d681a6 100644
--- a/security/integrity/ima/ima_main.c
+++ b/security/integrity/ima/ima_main.c
@@ -902,6 +902,13 @@ static int ima_post_load_data(char *buf, loff_t size,
 		return 0;
 	}
 
+	/*
+	 * Measure the init_module syscall buffer containing the ELF image.
+	 */
+	if (load_id == LOADING_MODULE)
+		ima_measure_critical_data("modules", "init_module",
+					  buf, size, true, NULL, 0);
+
 	return 0;
 }
 
-- 
2.39.3

Re: [PATCH] ima: define an init_module critical data record

[Jarkko Sakkinen]

On Wed Mar 27, 2024 at 5:00 PM EET, Mimi Zohar wrote:
> The init_module syscall loads an ELF image into kernel space without
> measuring the buffer containing the ELF image.  To close this kernel
> module integrity gap, define a new critical-data record which includes
> the hash of the ELF image.
>
> Instead of including the buffer data in the IMA measurement list,
> include the hash of the buffer data to avoid large IMA measurement
> list records.  The buffer data hash would be the same value as the
> finit_module syscall file hash.
>
> To enable measuring the init_module buffer and other critical data from
> boot, define "ima_policy=critical_data" on the boot command line.  Since
> builtin policies are not persistent, a custom IMA policy must include
> the rule as well: measure func=CRITICAL_DATA label=modules
>
> To verify the template data hash value, first convert the buffer data
> hash to binary:
> grep "init_module" \
> 	/sys/kernel/security/integrity/ima/ascii_runtime_measurements | \
> 	tail -1 | cut -d' ' -f 6 | xxd -r -p | sha256sum
>
> Reported-by: Ken Goldman <kgold@linux.ibm.com>
> Signed-off-by: Mimi Zohar <zohar@linux.ibm.com>
> ---
>  security/integrity/ima/ima_main.c | 7 +++++++
>  1 file changed, 7 insertions(+)
>
> diff --git a/security/integrity/ima/ima_main.c b/security/integrity/ima/ima_main.c
> index c84e8c55333d..4b4348d681a6 100644
> --- a/security/integrity/ima/ima_main.c
> +++ b/security/integrity/ima/ima_main.c
> @@ -902,6 +902,13 @@ static int ima_post_load_data(char *buf, loff_t size,
>  		return 0;
>  	}
>  
> +	/*
> +	 * Measure the init_module syscall buffer containing the ELF image.
> +	 */
> +	if (load_id == LOADING_MODULE)
> +		ima_measure_critical_data("modules", "init_module",
> +					  buf, size, true, NULL, 0);

No reason not to ack but could be just as well (passing checkpatch):

	if (load_id == LOADING_MODULE)
		ima_measure_critical_data("modules", "init_module", buf, size, true, NULL, 0);

< 100 characters

> +
>  	return 0;
>  }
>  

Reviewed-by: Jarkko Sakkinen <jarkko@kernel.org>

BR, Jarkko

Re: [PATCH] ima: define an init_module critical data record

[Mimi Zohar]

On Wed, 2024-03-27 at 18:54 +0200, Jarkko Sakkinen wrote:
> On Wed Mar 27, 2024 at 5:00 PM EET, Mimi Zohar wrote:
> > The init_module syscall loads an ELF image into kernel space without
> > measuring the buffer containing the ELF image.  To close this kernel
> > module integrity gap, define a new critical-data record which includes
> > the hash of the ELF image.
> > 
> > Instead of including the buffer data in the IMA measurement list,
> > include the hash of the buffer data to avoid large IMA measurement
> > list records.  The buffer data hash would be the same value as the
> > finit_module syscall file hash.
> > 
> > To enable measuring the init_module buffer and other critical data from
> > boot, define "ima_policy=critical_data" on the boot command line.  Since
> > builtin policies are not persistent, a custom IMA policy must include
> > the rule as well: measure func=CRITICAL_DATA label=modules
> > 
> > To verify the template data hash value, first convert the buffer data
> > hash to binary:
> > grep "init_module" \
> > 	/sys/kernel/security/integrity/ima/ascii_runtime_measurements | \
> > 	tail -1 | cut -d' ' -f 6 | xxd -r -p | sha256sum
> > 
> > Reported-by: Ken Goldman <kgold@linux.ibm.com>
> > Signed-off-by: Mimi Zohar <zohar@linux.ibm.com>
> > ---
> >  security/integrity/ima/ima_main.c | 7 +++++++
> >  1 file changed, 7 insertions(+)
> > 
> > diff --git a/security/integrity/ima/ima_main.c
> > b/security/integrity/ima/ima_main.c
> > index c84e8c55333d..4b4348d681a6 100644
> > --- a/security/integrity/ima/ima_main.c
> > +++ b/security/integrity/ima/ima_main.c
> > @@ -902,6 +902,13 @@ static int ima_post_load_data(char *buf, loff_t size,
> >  		return 0;
> >  	}
> >  
> > +	/*
> > +	 * Measure the init_module syscall buffer containing the ELF image.
> > +	 */
> > +	if (load_id == LOADING_MODULE)
> > +		ima_measure_critical_data("modules", "init_module",
> > +					  buf, size, true, NULL, 0);
> 
> No reason not to ack but could be just as well (passing checkpatch):

Please review the tag usage as defined in 
https://docs.kernel.org/process/submitting-patches.html.

> 
> 	if (load_id == LOADING_MODULE)
> 		ima_measure_critical_data("modules", "init_module", buf, size,
> true, NULL, 0);
> 
> < 100 characters

From what I understand, it's still preferable to stay under the 80 character
limit, but checkpatch.pl will not complain.  From 
https://www.kernel.org/doc/Documentation/process/maintainer-tip.rst:

"The 80 character rule is not a strict rule, so please use common sense when
breaking lines. Especially format strings should never be broken up."

> 
> > +
> >  	return 0;
> >  }
> >  
> 
> Reviewed-by: Jarkko Sakkinen <jarkko@kernel.org>

Thanks!

Mimi

Re: [PATCH] ima: define an init_module critical data record

[Jarkko Sakkinen]

On Wed Mar 27, 2024 at 11:37 PM EET, Mimi Zohar wrote:
> On Wed, 2024-03-27 at 18:54 +0200, Jarkko Sakkinen wrote:
> > On Wed Mar 27, 2024 at 5:00 PM EET, Mimi Zohar wrote:
> > > The init_module syscall loads an ELF image into kernel space without
> > > measuring the buffer containing the ELF image.  To close this kernel
> > > module integrity gap, define a new critical-data record which includes
> > > the hash of the ELF image.
> > > 
> > > Instead of including the buffer data in the IMA measurement list,
> > > include the hash of the buffer data to avoid large IMA measurement
> > > list records.  The buffer data hash would be the same value as the
> > > finit_module syscall file hash.
> > > 
> > > To enable measuring the init_module buffer and other critical data from
> > > boot, define "ima_policy=critical_data" on the boot command line.  Since
> > > builtin policies are not persistent, a custom IMA policy must include
> > > the rule as well: measure func=CRITICAL_DATA label=modules
> > > 
> > > To verify the template data hash value, first convert the buffer data
> > > hash to binary:
> > > grep "init_module" \
> > > 	/sys/kernel/security/integrity/ima/ascii_runtime_measurements | \
> > > 	tail -1 | cut -d' ' -f 6 | xxd -r -p | sha256sum
> > > 
> > > Reported-by: Ken Goldman <kgold@linux.ibm.com>
> > > Signed-off-by: Mimi Zohar <zohar@linux.ibm.com>
> > > ---
> > >  security/integrity/ima/ima_main.c | 7 +++++++
> > >  1 file changed, 7 insertions(+)
> > > 
> > > diff --git a/security/integrity/ima/ima_main.c
> > > b/security/integrity/ima/ima_main.c
> > > index c84e8c55333d..4b4348d681a6 100644
> > > --- a/security/integrity/ima/ima_main.c
> > > +++ b/security/integrity/ima/ima_main.c
> > > @@ -902,6 +902,13 @@ static int ima_post_load_data(char *buf, loff_t size,
> > >  		return 0;
> > >  	}
> > >  
> > > +	/*
> > > +	 * Measure the init_module syscall buffer containing the ELF image.
> > > +	 */
> > > +	if (load_id == LOADING_MODULE)
> > > +		ima_measure_critical_data("modules", "init_module",
> > > +					  buf, size, true, NULL, 0);
> > 
> > No reason not to ack but could be just as well (passing checkpatch):
>
> Please review the tag usage as defined in 
> https://docs.kernel.org/process/submitting-patches.html.
>
> > 
> > 	if (load_id == LOADING_MODULE)
> > 		ima_measure_critical_data("modules", "init_module", buf, size,
> > true, NULL, 0);
> > 
> > < 100 characters
>
> From what I understand, it's still preferable to stay under the 80 character
> limit, but checkpatch.pl will not complain.  From 
> https://www.kernel.org/doc/Documentation/process/maintainer-tip.rst:
>
> "The 80 character rule is not a strict rule, so please use common sense when
> breaking lines. Especially format strings should never be broken up."

I agree with that is your decision to make (as maintainer of
IMA)!

BTW, I guess process/coding-style.rst would be better ref
here because maintainer-tip.rst is meant only for arch/x86
tree (aka tip.git).

That said, both maintainer-tip.rst and coding-style.rst recommend
the same 80 character length. I have no idea whether 80 or 100
character should be considered as "recommended".

The formatted string example is somewhat weird too because you
should not break that even if it was 150 characters so it does
not have any relation to 100 character limit...

R, Jarkko