Linux cryptographic layer development
 help / color / mirror / Atom feed
* [PATCH v2] crypto: ccp - unwind PSP sub-devices on partial psp_init() failure
@ 2026-11-16 20:36 Kim Phillips
  2026-11-16 21:24 ` Mario Limonciello
  0 siblings, 1 reply; 2+ messages in thread
From: Kim Phillips @ 2026-11-16 20:36 UTC (permalink / raw)
  To: kim.phillips, Tom Lendacky, Herbert Xu, David S . Miller,
	Rijo Thomas, Mario Limonciello
  Cc: linux-crypto, linux-kernel, Sashiko, stable

psp_init() initializes sub-devices sequentially but returns immediately
on the first failure, leaving earlier sub-devices registered. Its caller
psp_dev_init() then sets sp->psp_data = NULL, so the later
psp_dev_destroy() on unbind bails out early and never unregisters them.

Non-devm resources such as the /dev/sev and /dev/dbc character devices
stay globally visible while their devm-allocated backing structures are
freed by the driver core, giving a use-after-free if userspace opens the
orphaned device.

Add reverse-order unwind labels to psp_init() so each sub-device is torn
down before the error is propagated. Every *_dev_destroy() is already
NULL-guarded on its own psp->*_data pointer, so skipped or unreached
sub-devices are no-ops.

sp_init() continues to ignore the psp_dev_init() return value on
purpose: a PSP failure must not fail the probe, as the CCP can still
initialize successfully.

Fixes: 33960acccfbd ("crypto: ccp - add TEE support for Raven Ridge")
Reported-by: Sashiko <sashiko-bot@kernel.org>
Closes: https://sashiko.dev/#/bug/linux-0de87f0c-0f5a-4dad-8ef1-f49844eb4cce
Cc: Tom Lendacky <thomas.lendacky@amd.com>
Cc: Rijo Thomas <Rijo-john.Thomas@amd.com>
Cc: Mario Limonciello <mario.limonciello@amd.com>
Cc: stable@vger.kernel.org
Signed-off-by: Kim Phillips <kim.phillips@amd.com>
---
v1: https://lore.kernel.org/linux-crypto/20261021215716.1076732-1-kim.phillips@amd.com/

 drivers/crypto/ccp/psp-dev.c | 23 ++++++++++++++++++-----
 1 file changed, 18 insertions(+), 5 deletions(-)

diff --git a/drivers/crypto/ccp/psp-dev.c b/drivers/crypto/ccp/psp-dev.c
index b14ce51065d5..3d9341c2876d 100644
--- a/drivers/crypto/ccp/psp-dev.c
+++ b/drivers/crypto/ccp/psp-dev.c
@@ -207,19 +207,19 @@ static int psp_init(struct psp_device *psp)
 	if (!psp_check_tee_support(psp)) {
 		ret = tee_dev_init(psp);
 		if (ret)
-			return ret;
+			goto e_sev;
 	}
 
 	if (!psp_check_sfs_support(psp)) {
 		ret = sfs_dev_init(psp);
 		if (ret)
-			return ret;
+			goto e_tee;
 	}
 
 	if (psp->vdata->platform_access) {
 		ret = platform_access_dev_init(psp);
 		if (ret)
-			return ret;
+			goto e_sfs;
 	}
 
 	/* dbc must come after platform access as it tests the feature */
@@ -227,15 +227,28 @@ static int psp_init(struct psp_device *psp)
 	    psp->capability.dbc_thru_ext) {
 		ret = dbc_dev_init(psp);
 		if (ret)
-			return ret;
+			goto e_platform_access;
 	}
 
 	/* HSTI uses platform access on some systems. */
 	ret = psp_init_hsti(psp);
 	if (ret)
-		return ret;
+		goto e_dbc;
 
 	return 0;
+
+e_dbc:
+	dbc_dev_destroy(psp);
+e_platform_access:
+	platform_access_dev_destroy(psp);
+e_sfs:
+	sfs_dev_destroy(psp);
+e_tee:
+	tee_dev_destroy(psp);
+e_sev:
+	sev_dev_destroy(psp);
+
+	return ret;
 }
 
 int psp_dev_init(struct sp_device *sp)

base-commit: 26ca5f34841f0a86d0883664a1c13be89a01d351
-- 
2.43.0


^ permalink raw reply related	[flat|nested] 2+ messages in thread

* Re: [PATCH v2] crypto: ccp - unwind PSP sub-devices on partial psp_init() failure
  2026-11-16 20:36 [PATCH v2] crypto: ccp - unwind PSP sub-devices on partial psp_init() failure Kim Phillips
@ 2026-11-16 21:24 ` Mario Limonciello
  0 siblings, 0 replies; 2+ messages in thread
From: Mario Limonciello @ 2026-11-16 21:24 UTC (permalink / raw)
  To: Kim Phillips, Tom Lendacky, Herbert Xu, David S . Miller,
	Rijo Thomas
  Cc: linux-crypto, linux-kernel, Sashiko, stable



On 11/16/26 14:36, Kim Phillips wrote:
> psp_init() initializes sub-devices sequentially but returns immediately
> on the first failure, leaving earlier sub-devices registered. Its caller
> psp_dev_init() then sets sp->psp_data = NULL, so the later
> psp_dev_destroy() on unbind bails out early and never unregisters them.
> 
> Non-devm resources such as the /dev/sev and /dev/dbc character devices
> stay globally visible while their devm-allocated backing structures are
> freed by the driver core, giving a use-after-free if userspace opens the
> orphaned device.
> 
> Add reverse-order unwind labels to psp_init() so each sub-device is torn
> down before the error is propagated. Every *_dev_destroy() is already
> NULL-guarded on its own psp->*_data pointer, so skipped or unreached
> sub-devices are no-ops.
> 
> sp_init() continues to ignore the psp_dev_init() return value on
> purpose: a PSP failure must not fail the probe, as the CCP can still
> initialize successfully.
> 
> Fixes: 33960acccfbd ("crypto: ccp - add TEE support for Raven Ridge")
> Reported-by: Sashiko <sashiko-bot@kernel.org>
> Closes: https://sashiko.dev/#/bug/linux-0de87f0c-0f5a-4dad-8ef1-f49844eb4cce
> Cc: Tom Lendacky <thomas.lendacky@amd.com>
> Cc: Rijo Thomas <Rijo-john.Thomas@amd.com>
> Cc: Mario Limonciello <mario.limonciello@amd.com>
> Cc: stable@vger.kernel.org
> Signed-off-by: Kim Phillips <kim.phillips@amd.com>
> ---
> v1: https://lore.kernel.org/linux-crypto/20261021215716.1076732-1-kim.phillips@amd.com/

Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>

> 
>   drivers/crypto/ccp/psp-dev.c | 23 ++++++++++++++++++-----
>   1 file changed, 18 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/crypto/ccp/psp-dev.c b/drivers/crypto/ccp/psp-dev.c
> index b14ce51065d5..3d9341c2876d 100644
> --- a/drivers/crypto/ccp/psp-dev.c
> +++ b/drivers/crypto/ccp/psp-dev.c
> @@ -207,19 +207,19 @@ static int psp_init(struct psp_device *psp)
>   	if (!psp_check_tee_support(psp)) {
>   		ret = tee_dev_init(psp);
>   		if (ret)
> -			return ret;
> +			goto e_sev;
>   	}
>   
>   	if (!psp_check_sfs_support(psp)) {
>   		ret = sfs_dev_init(psp);
>   		if (ret)
> -			return ret;
> +			goto e_tee;
>   	}
>   
>   	if (psp->vdata->platform_access) {
>   		ret = platform_access_dev_init(psp);
>   		if (ret)
> -			return ret;
> +			goto e_sfs;
>   	}
>   
>   	/* dbc must come after platform access as it tests the feature */
> @@ -227,15 +227,28 @@ static int psp_init(struct psp_device *psp)
>   	    psp->capability.dbc_thru_ext) {
>   		ret = dbc_dev_init(psp);
>   		if (ret)
> -			return ret;
> +			goto e_platform_access;
>   	}
>   
>   	/* HSTI uses platform access on some systems. */
>   	ret = psp_init_hsti(psp);
>   	if (ret)
> -		return ret;
> +		goto e_dbc;
>   
>   	return 0;
> +
> +e_dbc:
> +	dbc_dev_destroy(psp);
> +e_platform_access:
> +	platform_access_dev_destroy(psp);
> +e_sfs:
> +	sfs_dev_destroy(psp);
> +e_tee:
> +	tee_dev_destroy(psp);
> +e_sev:
> +	sev_dev_destroy(psp);
> +
> +	return ret;
>   }
>   
>   int psp_dev_init(struct sp_device *sp)
> 
> base-commit: 26ca5f34841f0a86d0883664a1c13be89a01d351


^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2026-11-16 21:24 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-16 20:36 [PATCH v2] crypto: ccp - unwind PSP sub-devices on partial psp_init() failure Kim Phillips
2026-11-16 21:24 ` Mario Limonciello

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox