LKML Archive on lore.kernel.org
help / color / mirror / Atom feed
* [PATCH v2 0/4] Have IMA find and use a tpm_chip until system shutdown
@ 2018-06-20 20:42 Stefan Berger
  2018-06-20 20:42 ` [PATCH v2 1/4] tpm: Implement tpm_chip_find() and tpm_chip_put() for other subsystems Stefan Berger
                   ` (3 more replies)
  0 siblings, 4 replies; 21+ messages in thread
From: Stefan Berger @ 2018-06-20 20:42 UTC (permalink / raw)
  To: linux-integrity, jarkko.sakkinen, zohar; +Cc: jgg, linux-kernel, Stefan Berger

This series of patches converts IMA's usage of the tpm_chip to find a TPM
chip initially and use it until the machine is shut down. To do this we need
to introduce a kref for the tpm_chip that IMA and all other users of a
tpm_chip hold onto until they don't need the TPM chip anymore.

    Stefan

v1->v2:
  - use the kref of the device via get_device()/put_device()

Stefan Berger (4):
  tpm: Implement tpm_chip_find() and tpm_chip_put() for other subsystems
  ima: Implement ima_shutdown and register it as a reboot_notifier
  ima: Use tpm_chip_find() and access TPM functions using it
  ima: Get rid of ima_used_chip and use ima_tpm_chip != NULL instead

 drivers/char/tpm/tpm-chip.c         | 37 +++++++++++++++++++++++++++++++++++++
 include/linux/tpm.h                 |  9 +++++++++
 security/integrity/ima/ima.h        |  4 +++-
 security/integrity/ima/ima_crypto.c | 14 +++++++++++---
 security/integrity/ima/ima_init.c   | 36 ++++++++++++++++++++++++++----------
 security/integrity/ima/ima_queue.c  |  9 ++++++---
 6 files changed, 92 insertions(+), 17 deletions(-)

-- 
2.13.6


^ permalink raw reply	[flat|nested] 21+ messages in thread

* [PATCH v2 1/4] tpm: Implement tpm_chip_find() and tpm_chip_put() for other subsystems
  2018-06-20 20:42 [PATCH v2 0/4] Have IMA find and use a tpm_chip until system shutdown Stefan Berger
@ 2018-06-20 20:42 ` Stefan Berger
  2018-06-20 20:50   ` Jason Gunthorpe
  2018-06-21 17:15   ` Jarkko Sakkinen
  2018-06-20 20:42 ` [PATCH v2 2/4] ima: Implement ima_shutdown and register it as a reboot_notifier Stefan Berger
                   ` (2 subsequent siblings)
  3 siblings, 2 replies; 21+ messages in thread
From: Stefan Berger @ 2018-06-20 20:42 UTC (permalink / raw)
  To: linux-integrity, jarkko.sakkinen, zohar; +Cc: jgg, linux-kernel, Stefan Berger

Implement tpm_chip_find() for other subsystems to find a TPM chip and
get a reference to that chip. Once done with using the chip, the reference
is released using tpm_chip_put().

Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com>
---
 drivers/char/tpm/tpm-chip.c | 37 +++++++++++++++++++++++++++++++++++++
 include/linux/tpm.h         |  9 +++++++++
 2 files changed, 46 insertions(+)

diff --git a/drivers/char/tpm/tpm-chip.c b/drivers/char/tpm/tpm-chip.c
index 0a62c19937b6..eaaf41ce73d8 100644
--- a/drivers/char/tpm/tpm-chip.c
+++ b/drivers/char/tpm/tpm-chip.c
@@ -81,6 +81,43 @@ void tpm_put_ops(struct tpm_chip *chip)
 EXPORT_SYMBOL_GPL(tpm_put_ops);
 
 /**
+ * tpm_chip_put() - Release a ref to the tpm_chip
+ * @chip: Chip to put
+ */
+void tpm_chip_put(struct tpm_chip *chip)
+{
+	if (chip)
+		put_device(&chip->dev);
+}
+
+/**
+ * tpm_chip_find() - find a TPM chip and get a reference to it
+ */
+struct tpm_chip *tpm_chip_find(void)
+{
+	struct tpm_chip *chip, *res = NULL;
+	int chip_num = 0;
+	int chip_prev;
+
+	mutex_lock(&idr_lock);
+
+	do {
+		chip_prev = chip_num;
+		chip = idr_get_next(&dev_nums_idr, &chip_num);
+		if (chip) {
+			get_device(&chip->dev);
+			res = chip;
+			break;
+		}
+	} while (chip_prev != chip_num);
+
+	mutex_unlock(&idr_lock);
+
+	return res;
+}
+EXPORT_SYMBOL_GPL(tpm_chip_find);
+
+/**
  * tpm_chip_find_get() - find and reserve a TPM chip
  * @chip:	a &struct tpm_chip instance, %NULL for the default chip
  *
diff --git a/include/linux/tpm.h b/include/linux/tpm.h
index 06639fb6ab85..f0d1abcaf4af 100644
--- a/include/linux/tpm.h
+++ b/include/linux/tpm.h
@@ -61,6 +61,8 @@ extern int tpm_seal_trusted(struct tpm_chip *chip,
 extern int tpm_unseal_trusted(struct tpm_chip *chip,
 			      struct trusted_key_payload *payload,
 			      struct trusted_key_options *options);
+extern struct tpm_chip *tpm_chip_find(void);
+extern void tpm_chip_put(struct tpm_chip *chip);
 #else
 static inline int tpm_is_tpm2(struct tpm_chip *chip)
 {
@@ -96,5 +98,12 @@ static inline int tpm_unseal_trusted(struct tpm_chip *chip,
 {
 	return -ENODEV;
 }
+static inline struct tpm_chip *tpm_chip_find(void)
+{
+	return NULL;
+}
+static inline void tpm_chip_put(struct tpm_chip *chip)
+{
+}
 #endif
 #endif
-- 
2.13.6


^ permalink raw reply	[flat|nested] 21+ messages in thread

* [PATCH v2 2/4] ima: Implement ima_shutdown and register it as a reboot_notifier
  2018-06-20 20:42 [PATCH v2 0/4] Have IMA find and use a tpm_chip until system shutdown Stefan Berger
  2018-06-20 20:42 ` [PATCH v2 1/4] tpm: Implement tpm_chip_find() and tpm_chip_put() for other subsystems Stefan Berger
@ 2018-06-20 20:42 ` Stefan Berger
  2018-06-20 20:42 ` [PATCH v2 3/4] ima: Use tpm_chip_find() and access TPM functions using it Stefan Berger
  2018-06-20 20:42 ` [PATCH v2 4/4] ima: Get rid of ima_used_chip and use ima_tpm_chip != NULL instead Stefan Berger
  3 siblings, 0 replies; 21+ messages in thread
From: Stefan Berger @ 2018-06-20 20:42 UTC (permalink / raw)
  To: linux-integrity, jarkko.sakkinen, zohar; +Cc: jgg, linux-kernel, Stefan Berger

Implement ima_shutdown so that we can release the tpm_chip before
devices are shut down. Register it as a low-priority reboot_notifier.

Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com>
---
 security/integrity/ima/ima_init.c | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

diff --git a/security/integrity/ima/ima_init.c b/security/integrity/ima/ima_init.c
index 29b72cd2502e..8a5258eb32b6 100644
--- a/security/integrity/ima/ima_init.c
+++ b/security/integrity/ima/ima_init.c
@@ -21,6 +21,7 @@
 #include <linux/scatterlist.h>
 #include <linux/slab.h>
 #include <linux/err.h>
+#include <linux/reboot.h>
 
 #include "ima.h"
 
@@ -104,11 +105,24 @@ void __init ima_load_x509(void)
 }
 #endif
 
+static int ima_shutdown(struct notifier_block *this, unsigned long action,
+			void *data)
+{
+	return NOTIFY_DONE;
+}
+
+static struct notifier_block ima_reboot_notifier = {
+	.notifier_call = ima_shutdown,
+	.priority = 0,
+};
+
 int __init ima_init(void)
 {
 	u8 pcr_i[TPM_DIGEST_SIZE];
 	int rc;
 
+	register_reboot_notifier(&ima_reboot_notifier);
+
 	ima_used_chip = 0;
 	rc = tpm_pcr_read(NULL, 0, pcr_i);
 	if (rc == 0)
-- 
2.13.6


^ permalink raw reply	[flat|nested] 21+ messages in thread

* [PATCH v2 3/4] ima: Use tpm_chip_find() and access TPM functions using it
  2018-06-20 20:42 [PATCH v2 0/4] Have IMA find and use a tpm_chip until system shutdown Stefan Berger
  2018-06-20 20:42 ` [PATCH v2 1/4] tpm: Implement tpm_chip_find() and tpm_chip_put() for other subsystems Stefan Berger
  2018-06-20 20:42 ` [PATCH v2 2/4] ima: Implement ima_shutdown and register it as a reboot_notifier Stefan Berger
@ 2018-06-20 20:42 ` Stefan Berger
  2018-06-21 20:53   ` Mimi Zohar
  2018-06-20 20:42 ` [PATCH v2 4/4] ima: Get rid of ima_used_chip and use ima_tpm_chip != NULL instead Stefan Berger
  3 siblings, 1 reply; 21+ messages in thread
From: Stefan Berger @ 2018-06-20 20:42 UTC (permalink / raw)
  To: linux-integrity, jarkko.sakkinen, zohar; +Cc: jgg, linux-kernel, Stefan Berger

Rather than accessing the TPM functions using a NULL pointer, which
causes a lookup for a suitable chip every time, get a hold of a tpm_chip
and access the TPM functions using this chip. We call the tpm_chip
ima_tpm_chip and protect it, once initialization is done, using a
rw_semaphore called ima_tpm_chip_lock.

Use ima_shutdown to release the tpm_chip.

Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com>
---
 security/integrity/ima/ima.h        |  3 +++
 security/integrity/ima/ima_crypto.c | 12 ++++++++++--
 security/integrity/ima/ima_init.c   | 19 ++++++++++++-------
 security/integrity/ima/ima_queue.c  |  7 +++++--
 4 files changed, 30 insertions(+), 11 deletions(-)

diff --git a/security/integrity/ima/ima.h b/security/integrity/ima/ima.h
index 354bb5716ce3..53a88d578ca5 100644
--- a/security/integrity/ima/ima.h
+++ b/security/integrity/ima/ima.h
@@ -24,6 +24,7 @@
 #include <linux/hash.h>
 #include <linux/tpm.h>
 #include <linux/audit.h>
+#include <linux/rwsem.h>
 #include <crypto/hash_info.h>
 
 #include "../integrity.h"
@@ -56,6 +57,8 @@ extern int ima_policy_flag;
 extern int ima_used_chip;
 extern int ima_hash_algo;
 extern int ima_appraise;
+extern struct rw_semaphore ima_tpm_chip_lock;
+extern struct tpm_chip *ima_tpm_chip;
 
 /* IMA event related data */
 struct ima_event_data {
diff --git a/security/integrity/ima/ima_crypto.c b/security/integrity/ima/ima_crypto.c
index 4e085a17124f..da7715240476 100644
--- a/security/integrity/ima/ima_crypto.c
+++ b/security/integrity/ima/ima_crypto.c
@@ -631,10 +631,18 @@ int ima_calc_buffer_hash(const void *buf, loff_t len,
 
 static void __init ima_pcrread(int idx, u8 *pcr)
 {
+	int result = 0;
+
+	down_read(&ima_tpm_chip_lock);
+
 	if (!ima_used_chip)
-		return;
+		goto out;
+
+	result = tpm_pcr_read(ima_tpm_chip, idx, pcr);
+out:
+	up_read(&ima_tpm_chip_lock);
 
-	if (tpm_pcr_read(NULL, idx, pcr) != 0)
+	if (result != 0)
 		pr_err("Error Communicating to TPM chip\n");
 }
 
diff --git a/security/integrity/ima/ima_init.c b/security/integrity/ima/ima_init.c
index 8a5258eb32b6..24db06c4f463 100644
--- a/security/integrity/ima/ima_init.c
+++ b/security/integrity/ima/ima_init.c
@@ -28,6 +28,8 @@
 /* name for boot aggregate entry */
 static const char *boot_aggregate_name = "boot_aggregate";
 int ima_used_chip;
+struct rw_semaphore ima_tpm_chip_lock = __RWSEM_INITIALIZER(ima_tpm_chip_lock);
+struct tpm_chip *ima_tpm_chip;
 
 /* Add the boot aggregate to the IMA measurement list and extend
  * the PCR register.
@@ -108,6 +110,13 @@ void __init ima_load_x509(void)
 static int ima_shutdown(struct notifier_block *this, unsigned long action,
 			void *data)
 {
+	down_write(&ima_tpm_chip_lock);
+	if (ima_tpm_chip) {
+		tpm_chip_put(ima_tpm_chip);
+		ima_tpm_chip = NULL;
+		ima_used_chip = 0;
+	}
+	up_write(&ima_tpm_chip_lock);
 	return NOTIFY_DONE;
 }
 
@@ -118,19 +127,15 @@ static struct notifier_block ima_reboot_notifier = {
 
 int __init ima_init(void)
 {
-	u8 pcr_i[TPM_DIGEST_SIZE];
 	int rc;
 
 	register_reboot_notifier(&ima_reboot_notifier);
 
-	ima_used_chip = 0;
-	rc = tpm_pcr_read(NULL, 0, pcr_i);
-	if (rc == 0)
-		ima_used_chip = 1;
+	ima_tpm_chip = tpm_chip_find();
 
+	ima_used_chip = (ima_tpm_chip != NULL);
 	if (!ima_used_chip)
-		pr_info("No TPM chip found, activating TPM-bypass! (rc=%d)\n",
-			rc);
+		pr_info("No TPM chip found, activating TPM-bypass!\n");
 
 	rc = integrity_init_keyring(INTEGRITY_KEYRING_IMA);
 	if (rc)
diff --git a/security/integrity/ima/ima_queue.c b/security/integrity/ima/ima_queue.c
index 418f35e38015..6c9427939a28 100644
--- a/security/integrity/ima/ima_queue.c
+++ b/security/integrity/ima/ima_queue.c
@@ -142,10 +142,13 @@ static int ima_pcr_extend(const u8 *hash, int pcr)
 {
 	int result = 0;
 
+	down_read(&ima_tpm_chip_lock);
 	if (!ima_used_chip)
-		return result;
+		goto out;
 
-	result = tpm_pcr_extend(NULL, pcr, hash);
+	result = tpm_pcr_extend(ima_tpm_chip, pcr, hash);
+out:
+	up_read(&ima_tpm_chip_lock);
 	if (result != 0)
 		pr_err("Error Communicating to TPM chip, result: %d\n", result);
 	return result;
-- 
2.13.6


^ permalink raw reply	[flat|nested] 21+ messages in thread

* [PATCH v2 4/4] ima: Get rid of ima_used_chip and use ima_tpm_chip != NULL instead
  2018-06-20 20:42 [PATCH v2 0/4] Have IMA find and use a tpm_chip until system shutdown Stefan Berger
                   ` (2 preceding siblings ...)
  2018-06-20 20:42 ` [PATCH v2 3/4] ima: Use tpm_chip_find() and access TPM functions using it Stefan Berger
@ 2018-06-20 20:42 ` Stefan Berger
  3 siblings, 0 replies; 21+ messages in thread
From: Stefan Berger @ 2018-06-20 20:42 UTC (permalink / raw)
  To: linux-integrity, jarkko.sakkinen, zohar; +Cc: jgg, linux-kernel, Stefan Berger

Get rid of ima_used_chip and use ima_tpm_chip variable instead for
determining whether to use the TPM chip.

Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com>
---
 security/integrity/ima/ima.h        | 1 -
 security/integrity/ima/ima_crypto.c | 2 +-
 security/integrity/ima/ima_init.c   | 7 ++-----
 security/integrity/ima/ima_queue.c  | 2 +-
 4 files changed, 4 insertions(+), 8 deletions(-)

diff --git a/security/integrity/ima/ima.h b/security/integrity/ima/ima.h
index 53a88d578ca5..18491c0e9b97 100644
--- a/security/integrity/ima/ima.h
+++ b/security/integrity/ima/ima.h
@@ -54,7 +54,6 @@ enum tpm_pcrs { TPM_PCR0 = 0, TPM_PCR8 = 8 };
 extern int ima_policy_flag;
 
 /* set during initialization */
-extern int ima_used_chip;
 extern int ima_hash_algo;
 extern int ima_appraise;
 extern struct rw_semaphore ima_tpm_chip_lock;
diff --git a/security/integrity/ima/ima_crypto.c b/security/integrity/ima/ima_crypto.c
index da7715240476..bc6ad6b3243a 100644
--- a/security/integrity/ima/ima_crypto.c
+++ b/security/integrity/ima/ima_crypto.c
@@ -635,7 +635,7 @@ static void __init ima_pcrread(int idx, u8 *pcr)
 
 	down_read(&ima_tpm_chip_lock);
 
-	if (!ima_used_chip)
+	if (!ima_tpm_chip)
 		goto out;
 
 	result = tpm_pcr_read(ima_tpm_chip, idx, pcr);
diff --git a/security/integrity/ima/ima_init.c b/security/integrity/ima/ima_init.c
index 24db06c4f463..84fba44c4415 100644
--- a/security/integrity/ima/ima_init.c
+++ b/security/integrity/ima/ima_init.c
@@ -27,7 +27,6 @@
 
 /* name for boot aggregate entry */
 static const char *boot_aggregate_name = "boot_aggregate";
-int ima_used_chip;
 struct rw_semaphore ima_tpm_chip_lock = __RWSEM_INITIALIZER(ima_tpm_chip_lock);
 struct tpm_chip *ima_tpm_chip;
 
@@ -67,7 +66,7 @@ static int __init ima_add_boot_aggregate(void)
 	iint->ima_hash->algo = HASH_ALGO_SHA1;
 	iint->ima_hash->length = SHA1_DIGEST_SIZE;
 
-	if (ima_used_chip) {
+	if (ima_tpm_chip) {
 		result = ima_calc_boot_aggregate(&hash.hdr);
 		if (result < 0) {
 			audit_cause = "hashing_error";
@@ -114,7 +113,6 @@ static int ima_shutdown(struct notifier_block *this, unsigned long action,
 	if (ima_tpm_chip) {
 		tpm_chip_put(ima_tpm_chip);
 		ima_tpm_chip = NULL;
-		ima_used_chip = 0;
 	}
 	up_write(&ima_tpm_chip_lock);
 	return NOTIFY_DONE;
@@ -133,8 +131,7 @@ int __init ima_init(void)
 
 	ima_tpm_chip = tpm_chip_find();
 
-	ima_used_chip = (ima_tpm_chip != NULL);
-	if (!ima_used_chip)
+	if (!ima_tpm_chip)
 		pr_info("No TPM chip found, activating TPM-bypass!\n");
 
 	rc = integrity_init_keyring(INTEGRITY_KEYRING_IMA);
diff --git a/security/integrity/ima/ima_queue.c b/security/integrity/ima/ima_queue.c
index 6c9427939a28..591ff3d76fe6 100644
--- a/security/integrity/ima/ima_queue.c
+++ b/security/integrity/ima/ima_queue.c
@@ -143,7 +143,7 @@ static int ima_pcr_extend(const u8 *hash, int pcr)
 	int result = 0;
 
 	down_read(&ima_tpm_chip_lock);
-	if (!ima_used_chip)
+	if (!ima_tpm_chip)
 		goto out;
 
 	result = tpm_pcr_extend(ima_tpm_chip, pcr, hash);
-- 
2.13.6


^ permalink raw reply	[flat|nested] 21+ messages in thread

* Re: [PATCH v2 1/4] tpm: Implement tpm_chip_find() and tpm_chip_put() for other subsystems
  2018-06-20 20:42 ` [PATCH v2 1/4] tpm: Implement tpm_chip_find() and tpm_chip_put() for other subsystems Stefan Berger
@ 2018-06-20 20:50   ` Jason Gunthorpe
  2018-06-20 21:13     ` Stefan Berger
  2018-06-21 17:13     ` Jarkko Sakkinen
  2018-06-21 17:15   ` Jarkko Sakkinen
  1 sibling, 2 replies; 21+ messages in thread
From: Jason Gunthorpe @ 2018-06-20 20:50 UTC (permalink / raw)
  To: Stefan Berger; +Cc: linux-integrity, jarkko.sakkinen, zohar, linux-kernel

On Wed, Jun 20, 2018 at 04:42:33PM -0400, Stefan Berger wrote:
> Implement tpm_chip_find() for other subsystems to find a TPM chip and
> get a reference to that chip. Once done with using the chip, the reference
> is released using tpm_chip_put().
> 
> Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com>
>  drivers/char/tpm/tpm-chip.c | 37 +++++++++++++++++++++++++++++++++++++
>  include/linux/tpm.h         |  9 +++++++++
>  2 files changed, 46 insertions(+)
> 
> diff --git a/drivers/char/tpm/tpm-chip.c b/drivers/char/tpm/tpm-chip.c
> index 0a62c19937b6..eaaf41ce73d8 100644
> +++ b/drivers/char/tpm/tpm-chip.c
> @@ -81,6 +81,43 @@ void tpm_put_ops(struct tpm_chip *chip)
>  EXPORT_SYMBOL_GPL(tpm_put_ops);
>  
>  /**
> + * tpm_chip_put() - Release a ref to the tpm_chip
> + * @chip: Chip to put
> + */
> +void tpm_chip_put(struct tpm_chip *chip)
> +{
> +	if (chip)
> +		put_device(&chip->dev);

Rarely like to see those if's inside a put function.. Does it actually
help anything?

Jason

^ permalink raw reply	[flat|nested] 21+ messages in thread

* Re: [PATCH v2 1/4] tpm: Implement tpm_chip_find() and tpm_chip_put() for other subsystems
  2018-06-20 20:50   ` Jason Gunthorpe
@ 2018-06-20 21:13     ` Stefan Berger
  2018-06-21 17:13     ` Jarkko Sakkinen
  1 sibling, 0 replies; 21+ messages in thread
From: Stefan Berger @ 2018-06-20 21:13 UTC (permalink / raw)
  To: Jason Gunthorpe; +Cc: linux-integrity, jarkko.sakkinen, zohar, linux-kernel

On 06/20/2018 04:50 PM, Jason Gunthorpe wrote:
> On Wed, Jun 20, 2018 at 04:42:33PM -0400, Stefan Berger wrote:
>> Implement tpm_chip_find() for other subsystems to find a TPM chip and
>> get a reference to that chip. Once done with using the chip, the reference
>> is released using tpm_chip_put().
>>
>> Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com>
>>   drivers/char/tpm/tpm-chip.c | 37 +++++++++++++++++++++++++++++++++++++
>>   include/linux/tpm.h         |  9 +++++++++
>>   2 files changed, 46 insertions(+)
>>
>> diff --git a/drivers/char/tpm/tpm-chip.c b/drivers/char/tpm/tpm-chip.c
>> index 0a62c19937b6..eaaf41ce73d8 100644
>> +++ b/drivers/char/tpm/tpm-chip.c
>> @@ -81,6 +81,43 @@ void tpm_put_ops(struct tpm_chip *chip)
>>   EXPORT_SYMBOL_GPL(tpm_put_ops);
>>   
>>   /**
>> + * tpm_chip_put() - Release a ref to the tpm_chip
>> + * @chip: Chip to put
>> + */
>> +void tpm_chip_put(struct tpm_chip *chip)
>> +{
>> +	if (chip)
>> +		put_device(&chip->dev);
> Rarely like to see those if's inside a put function.. Does it actually
> help anything?
Following put_device() 'pattern' here which also checks using 'if (dev)'.

     Stefan


^ permalink raw reply	[flat|nested] 21+ messages in thread

* Re: [PATCH v2 1/4] tpm: Implement tpm_chip_find() and tpm_chip_put() for other subsystems
  2018-06-20 20:50   ` Jason Gunthorpe
  2018-06-20 21:13     ` Stefan Berger
@ 2018-06-21 17:13     ` Jarkko Sakkinen
  1 sibling, 0 replies; 21+ messages in thread
From: Jarkko Sakkinen @ 2018-06-21 17:13 UTC (permalink / raw)
  To: Jason Gunthorpe; +Cc: Stefan Berger, linux-integrity, zohar, linux-kernel

On Wed, Jun 20, 2018 at 02:50:04PM -0600, Jason Gunthorpe wrote:
> On Wed, Jun 20, 2018 at 04:42:33PM -0400, Stefan Berger wrote:
> > Implement tpm_chip_find() for other subsystems to find a TPM chip and
> > get a reference to that chip. Once done with using the chip, the reference
> > is released using tpm_chip_put().
> > 
> > Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com>
> >  drivers/char/tpm/tpm-chip.c | 37 +++++++++++++++++++++++++++++++++++++
> >  include/linux/tpm.h         |  9 +++++++++
> >  2 files changed, 46 insertions(+)
> > 
> > diff --git a/drivers/char/tpm/tpm-chip.c b/drivers/char/tpm/tpm-chip.c
> > index 0a62c19937b6..eaaf41ce73d8 100644
> > +++ b/drivers/char/tpm/tpm-chip.c
> > @@ -81,6 +81,43 @@ void tpm_put_ops(struct tpm_chip *chip)
> >  EXPORT_SYMBOL_GPL(tpm_put_ops);
> >  
> >  /**
> > + * tpm_chip_put() - Release a ref to the tpm_chip
> > + * @chip: Chip to put
> > + */
> > +void tpm_chip_put(struct tpm_chip *chip)
> > +{
> > +	if (chip)
> > +		put_device(&chip->dev);
> 
> Rarely like to see those if's inside a put function.. Does it actually
> help anything?
> 
> Jason

The check only helps to hidden regressions here...

/Jarkko

^ permalink raw reply	[flat|nested] 21+ messages in thread

* Re: [PATCH v2 1/4] tpm: Implement tpm_chip_find() and tpm_chip_put() for other subsystems
  2018-06-20 20:42 ` [PATCH v2 1/4] tpm: Implement tpm_chip_find() and tpm_chip_put() for other subsystems Stefan Berger
  2018-06-20 20:50   ` Jason Gunthorpe
@ 2018-06-21 17:15   ` Jarkko Sakkinen
  2018-06-21 17:27     ` Stefan Berger
  2018-06-21 17:45     ` Stefan Berger
  1 sibling, 2 replies; 21+ messages in thread
From: Jarkko Sakkinen @ 2018-06-21 17:15 UTC (permalink / raw)
  To: Stefan Berger; +Cc: linux-integrity, zohar, jgg, linux-kernel

On Wed, Jun 20, 2018 at 04:42:33PM -0400, Stefan Berger wrote:
> Implement tpm_chip_find() for other subsystems to find a TPM chip and
> get a reference to that chip. Once done with using the chip, the reference
> is released using tpm_chip_put().
> 
> Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com>

You should sort this out in a way that we don't end up with duplicate
functions.

/Jarkko

^ permalink raw reply	[flat|nested] 21+ messages in thread

* Re: [PATCH v2 1/4] tpm: Implement tpm_chip_find() and tpm_chip_put() for other subsystems
  2018-06-21 17:15   ` Jarkko Sakkinen
@ 2018-06-21 17:27     ` Stefan Berger
  2018-06-21 17:45     ` Stefan Berger
  1 sibling, 0 replies; 21+ messages in thread
From: Stefan Berger @ 2018-06-21 17:27 UTC (permalink / raw)
  To: Jarkko Sakkinen; +Cc: linux-integrity, zohar, jgg, linux-kernel

On 06/21/2018 01:15 PM, Jarkko Sakkinen wrote:
> On Wed, Jun 20, 2018 at 04:42:33PM -0400, Stefan Berger wrote:
>> Implement tpm_chip_find() for other subsystems to find a TPM chip and
>> get a reference to that chip. Once done with using the chip, the reference
>> is released using tpm_chip_put().
>>
>> Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com>
> You should sort this out in a way that we don't end up with duplicate
> functions.

We cannot hold the ops_sem semaphore for a long time. So we need a 
function that gets us a reference and doesn't hold the ops semaphore and 
that's where this tpm_chip_find() comes from.

    Stefan

> /Jarkko
>


^ permalink raw reply	[flat|nested] 21+ messages in thread

* Re: [PATCH v2 1/4] tpm: Implement tpm_chip_find() and tpm_chip_put() for other subsystems
  2018-06-21 17:15   ` Jarkko Sakkinen
  2018-06-21 17:27     ` Stefan Berger
@ 2018-06-21 17:45     ` Stefan Berger
  2018-06-21 17:56       ` Jason Gunthorpe
  1 sibling, 1 reply; 21+ messages in thread
From: Stefan Berger @ 2018-06-21 17:45 UTC (permalink / raw)
  To: Jarkko Sakkinen; +Cc: linux-integrity, zohar, jgg, linux-kernel

On 06/21/2018 01:15 PM, Jarkko Sakkinen wrote:
> On Wed, Jun 20, 2018 at 04:42:33PM -0400, Stefan Berger wrote:
>> Implement tpm_chip_find() for other subsystems to find a TPM chip and
>> get a reference to that chip. Once done with using the chip, the reference
>> is released using tpm_chip_put().
>>
>> Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com>
> You should sort this out in a way that we don't end up with duplicate
> functions.

Do you want me to create a function *like* tpm_chip_find_get() that 
takes an additional parameter whether to get the ops semaphore and have 
that function called by the existing tpm_chip_find_get() and the new 
tpm_chip_find(). The latter would then not get the ops semphore. I 
didn't want to do this since one time the function returns with a lock 
held and the other time not.

   Stefan

>
> /Jarkko
>


^ permalink raw reply	[flat|nested] 21+ messages in thread

* Re: [PATCH v2 1/4] tpm: Implement tpm_chip_find() and tpm_chip_put() for other subsystems
  2018-06-21 17:45     ` Stefan Berger
@ 2018-06-21 17:56       ` Jason Gunthorpe
  2018-06-21 18:19         ` Stefan Berger
  0 siblings, 1 reply; 21+ messages in thread
From: Jason Gunthorpe @ 2018-06-21 17:56 UTC (permalink / raw)
  To: Stefan Berger; +Cc: Jarkko Sakkinen, linux-integrity, zohar, linux-kernel

On Thu, Jun 21, 2018 at 01:45:03PM -0400, Stefan Berger wrote:
> On 06/21/2018 01:15 PM, Jarkko Sakkinen wrote:
> >On Wed, Jun 20, 2018 at 04:42:33PM -0400, Stefan Berger wrote:
> >>Implement tpm_chip_find() for other subsystems to find a TPM chip and
> >>get a reference to that chip. Once done with using the chip, the reference
> >>is released using tpm_chip_put().
> >>
> >>Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com>
> >You should sort this out in a way that we don't end up with duplicate
> >functions.
> 
> Do you want me to create a function *like* tpm_chip_find_get() that takes an
> additional parameter whether to get the ops semaphore and have that function
> called by the existing tpm_chip_find_get() and the new tpm_chip_find(). The
> latter would then not get the ops semphore. I didn't want to do this since
> one time the function returns with a lock held and the other time not.

Another option, and I haven't looked, is to revise the callers of
tpm_chip_find_get to not require it to hold the ops semaphore for
them.

Either by giving them an API to do it, or revising the TPM entry
points to do it.

I didn't look, but how did the ops semaphore get grabbed in your
revised patches? They do grab it, right?

Jason

^ permalink raw reply	[flat|nested] 21+ messages in thread

* Re: [PATCH v2 1/4] tpm: Implement tpm_chip_find() and tpm_chip_put() for other subsystems
  2018-06-21 17:56       ` Jason Gunthorpe
@ 2018-06-21 18:19         ` Stefan Berger
  2018-06-21 19:06           ` Jason Gunthorpe
  0 siblings, 1 reply; 21+ messages in thread
From: Stefan Berger @ 2018-06-21 18:19 UTC (permalink / raw)
  To: Jason Gunthorpe; +Cc: Jarkko Sakkinen, linux-integrity, zohar, linux-kernel

On 06/21/2018 01:56 PM, Jason Gunthorpe wrote:
> On Thu, Jun 21, 2018 at 01:45:03PM -0400, Stefan Berger wrote:
>> On 06/21/2018 01:15 PM, Jarkko Sakkinen wrote:
>>> On Wed, Jun 20, 2018 at 04:42:33PM -0400, Stefan Berger wrote:
>>>> Implement tpm_chip_find() for other subsystems to find a TPM chip and
>>>> get a reference to that chip. Once done with using the chip, the reference
>>>> is released using tpm_chip_put().
>>>>
>>>> Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com>
>>> You should sort this out in a way that we don't end up with duplicate
>>> functions.
>> Do you want me to create a function *like* tpm_chip_find_get() that takes an
>> additional parameter whether to get the ops semaphore and have that function
>> called by the existing tpm_chip_find_get() and the new tpm_chip_find(). The
>> latter would then not get the ops semphore. I didn't want to do this since
>> one time the function returns with a lock held and the other time not.
> Another option, and I haven't looked, is to revise the callers of
> tpm_chip_find_get to not require it to hold the ops semaphore for
> them.

We have tpm_chip_unregister calling tpm_del_char_device to set the ops 
to NULL once a chip is unregistered. All existing callers, if they pass 
in a tpm_chip != NULL, currently fail if the ops are NULL. (If they pass 
in tpm_chip = NULL, they shouldn't find a chip once ops are null and it 
has been removed from the IDR). I wouldn't change that since IMA will 
call in with a tpm_chip != NULL and we want to protect the ops. All 
existing code within the tpm subsystem does seem to call 
tpm_chip_find_get() with a NULL pointer, though. Also trusted keys seems 
to pass in a NULL pointer every time.

>
> Either by giving them an API to do it, or revising the TPM entry
> points to do it.
>
> I didn't look, but how did the ops semaphore get grabbed in your
> revised patches? They do grab it, right?

The revised patches do not touch the existing code much but will call 
tpm_chip_find_get() and get that semaphore every time before the ops are 
used. IMA is the only caller of tpm_chip_find() that now gets an 
additional reference to the tpm_chip and these APIs get called like this 
from IMA:

ima init: chip = tpm_chip_find()

ima::tpm: tpm_chip_find_get(chip) ... tpm_put_ops(chip)

ima::tpm: tpm_chip_find_get(chip) ... tpm_put_ops(chip)

[repeat]

ima shutdown: tpm_chip_put(chip)

     Stefan

>
> Jason
>


^ permalink raw reply	[flat|nested] 21+ messages in thread

* Re: [PATCH v2 1/4] tpm: Implement tpm_chip_find() and tpm_chip_put() for other subsystems
  2018-06-21 18:19         ` Stefan Berger
@ 2018-06-21 19:06           ` Jason Gunthorpe
  2018-06-21 20:14             ` Stefan Berger
  0 siblings, 1 reply; 21+ messages in thread
From: Jason Gunthorpe @ 2018-06-21 19:06 UTC (permalink / raw)
  To: Stefan Berger; +Cc: Jarkko Sakkinen, linux-integrity, zohar, linux-kernel

On Thu, Jun 21, 2018 at 02:19:44PM -0400, Stefan Berger wrote:
> On 06/21/2018 01:56 PM, Jason Gunthorpe wrote:
> >On Thu, Jun 21, 2018 at 01:45:03PM -0400, Stefan Berger wrote:
> >>On 06/21/2018 01:15 PM, Jarkko Sakkinen wrote:
> >>>On Wed, Jun 20, 2018 at 04:42:33PM -0400, Stefan Berger wrote:
> >>>>Implement tpm_chip_find() for other subsystems to find a TPM chip and
> >>>>get a reference to that chip. Once done with using the chip, the reference
> >>>>is released using tpm_chip_put().
> >>>>
> >>>>Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com>
> >>>You should sort this out in a way that we don't end up with duplicate
> >>>functions.
> >>Do you want me to create a function *like* tpm_chip_find_get() that takes an
> >>additional parameter whether to get the ops semaphore and have that function
> >>called by the existing tpm_chip_find_get() and the new tpm_chip_find(). The
> >>latter would then not get the ops semphore. I didn't want to do this since
> >>one time the function returns with a lock held and the other time not.
> >Another option, and I haven't looked, is to revise the callers of
> >tpm_chip_find_get to not require it to hold the ops semaphore for
> >them.
> 
> We have tpm_chip_unregister calling tpm_del_char_device to set the ops to
> NULL once a chip is unregistered. All existing callers, if they pass in a
> tpm_chip != NULL, currently fail if the ops are NULL. (If they pass in
> tpm_chip = NULL, they shouldn't find a chip once ops are null and it has
> been removed from the IDR). I wouldn't change that since IMA will call in
> with a tpm_chip != NULL and we want to protect the ops. All existing code
> within the tpm subsystem does seem to call tpm_chip_find_get() with a NULL
> pointer, though. Also trusted keys seems to pass in a NULL pointer every
> time.
> 
> >
> >Either by giving them an API to do it, or revising the TPM entry
> >points to do it.
> >
> >I didn't look, but how did the ops semaphore get grabbed in your
> >revised patches? They do grab it, right?
> 
> The revised patches do not touch the existing code much but will call
> tpm_chip_find_get() and get that semaphore every time before the ops are
> used. IMA is the only caller of tpm_chip_find() that now gets an additional
> reference to the tpm_chip and these APIs get called like this from IMA:
> 
> ima init: chip = tpm_chip_find()
> 
> ima::tpm: tpm_chip_find_get(chip) ... tpm_put_ops(chip)
> 
> ima::tpm: tpm_chip_find_get(chip) ... tpm_put_ops(chip)
> 
> [repeat]
> 
> ima shutdown: tpm_chip_put(chip)

Maybe just change tpm_chip_find_get() into tpm_get_ops(chip) and
convert all callers?

Jason

^ permalink raw reply	[flat|nested] 21+ messages in thread

* Re: [PATCH v2 1/4] tpm: Implement tpm_chip_find() and tpm_chip_put() for other subsystems
  2018-06-21 19:06           ` Jason Gunthorpe
@ 2018-06-21 20:14             ` Stefan Berger
  2018-06-21 20:51               ` Jason Gunthorpe
  0 siblings, 1 reply; 21+ messages in thread
From: Stefan Berger @ 2018-06-21 20:14 UTC (permalink / raw)
  To: Jason Gunthorpe; +Cc: Jarkko Sakkinen, linux-integrity, zohar, linux-kernel

On 06/21/2018 03:06 PM, Jason Gunthorpe wrote:
> On Thu, Jun 21, 2018 at 02:19:44PM -0400, Stefan Berger wrote:
>> On 06/21/2018 01:56 PM, Jason Gunthorpe wrote:
>>> On Thu, Jun 21, 2018 at 01:45:03PM -0400, Stefan Berger wrote:
>>>> On 06/21/2018 01:15 PM, Jarkko Sakkinen wrote:
>>>>> On Wed, Jun 20, 2018 at 04:42:33PM -0400, Stefan Berger wrote:
>>>>>> Implement tpm_chip_find() for other subsystems to find a TPM chip and
>>>>>> get a reference to that chip. Once done with using the chip, the reference
>>>>>> is released using tpm_chip_put().
>>>>>>
>>>>>> Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com>
>>>>> You should sort this out in a way that we don't end up with duplicate
>>>>> functions.
>>>> Do you want me to create a function *like* tpm_chip_find_get() that takes an
>>>> additional parameter whether to get the ops semaphore and have that function
>>>> called by the existing tpm_chip_find_get() and the new tpm_chip_find(). The
>>>> latter would then not get the ops semphore. I didn't want to do this since
>>>> one time the function returns with a lock held and the other time not.
>>> Another option, and I haven't looked, is to revise the callers of
>>> tpm_chip_find_get to not require it to hold the ops semaphore for
>>> them.
>> We have tpm_chip_unregister calling tpm_del_char_device to set the ops to
>> NULL once a chip is unregistered. All existing callers, if they pass in a
>> tpm_chip != NULL, currently fail if the ops are NULL. (If they pass in
>> tpm_chip = NULL, they shouldn't find a chip once ops are null and it has
>> been removed from the IDR). I wouldn't change that since IMA will call in
>> with a tpm_chip != NULL and we want to protect the ops. All existing code
>> within the tpm subsystem does seem to call tpm_chip_find_get() with a NULL
>> pointer, though. Also trusted keys seems to pass in a NULL pointer every
>> time.
>>
>>> Either by giving them an API to do it, or revising the TPM entry
>>> points to do it.
>>>
>>> I didn't look, but how did the ops semaphore get grabbed in your
>>> revised patches? They do grab it, right?
>> The revised patches do not touch the existing code much but will call
>> tpm_chip_find_get() and get that semaphore every time before the ops are
>> used. IMA is the only caller of tpm_chip_find() that now gets an additional
>> reference to the tpm_chip and these APIs get called like this from IMA:
>>
>> ima init: chip = tpm_chip_find()
>>
>> ima::tpm: tpm_chip_find_get(chip) ... tpm_put_ops(chip)
>>
>> ima::tpm: tpm_chip_find_get(chip) ... tpm_put_ops(chip)
>>
>> [repeat]
>>
>> ima shutdown: tpm_chip_put(chip)
> Maybe just change tpm_chip_find_get() into tpm_get_ops(chip) and
> convert all callers?

And then re-introduce tpm_chip_find_get() for IMA to call ?


^ permalink raw reply	[flat|nested] 21+ messages in thread

* Re: [PATCH v2 1/4] tpm: Implement tpm_chip_find() and tpm_chip_put() for other subsystems
  2018-06-21 20:14             ` Stefan Berger
@ 2018-06-21 20:51               ` Jason Gunthorpe
  0 siblings, 0 replies; 21+ messages in thread
From: Jason Gunthorpe @ 2018-06-21 20:51 UTC (permalink / raw)
  To: Stefan Berger; +Cc: Jarkko Sakkinen, linux-integrity, zohar, linux-kernel

On Thu, Jun 21, 2018 at 04:14:46PM -0400, Stefan Berger wrote:
> On 06/21/2018 03:06 PM, Jason Gunthorpe wrote:
> >On Thu, Jun 21, 2018 at 02:19:44PM -0400, Stefan Berger wrote:
> >>On 06/21/2018 01:56 PM, Jason Gunthorpe wrote:
> >>>On Thu, Jun 21, 2018 at 01:45:03PM -0400, Stefan Berger wrote:
> >>>>On 06/21/2018 01:15 PM, Jarkko Sakkinen wrote:
> >>>>>On Wed, Jun 20, 2018 at 04:42:33PM -0400, Stefan Berger wrote:
> >>>>>>Implement tpm_chip_find() for other subsystems to find a TPM chip and
> >>>>>>get a reference to that chip. Once done with using the chip, the reference
> >>>>>>is released using tpm_chip_put().
> >>>>>>
> >>>>>>Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com>
> >>>>>You should sort this out in a way that we don't end up with duplicate
> >>>>>functions.
> >>>>Do you want me to create a function *like* tpm_chip_find_get() that takes an
> >>>>additional parameter whether to get the ops semaphore and have that function
> >>>>called by the existing tpm_chip_find_get() and the new tpm_chip_find(). The
> >>>>latter would then not get the ops semphore. I didn't want to do this since
> >>>>one time the function returns with a lock held and the other time not.
> >>>Another option, and I haven't looked, is to revise the callers of
> >>>tpm_chip_find_get to not require it to hold the ops semaphore for
> >>>them.
> >>We have tpm_chip_unregister calling tpm_del_char_device to set the ops to
> >>NULL once a chip is unregistered. All existing callers, if they pass in a
> >>tpm_chip != NULL, currently fail if the ops are NULL. (If they pass in
> >>tpm_chip = NULL, they shouldn't find a chip once ops are null and it has
> >>been removed from the IDR). I wouldn't change that since IMA will call in
> >>with a tpm_chip != NULL and we want to protect the ops. All existing code
> >>within the tpm subsystem does seem to call tpm_chip_find_get() with a NULL
> >>pointer, though. Also trusted keys seems to pass in a NULL pointer every
> >>time.
> >>
> >>>Either by giving them an API to do it, or revising the TPM entry
> >>>points to do it.
> >>>
> >>>I didn't look, but how did the ops semaphore get grabbed in your
> >>>revised patches? They do grab it, right?
> >>The revised patches do not touch the existing code much but will call
> >>tpm_chip_find_get() and get that semaphore every time before the ops are
> >>used. IMA is the only caller of tpm_chip_find() that now gets an additional
> >>reference to the tpm_chip and these APIs get called like this from IMA:
> >>
> >>ima init: chip = tpm_chip_find()
> >>
> >>ima::tpm: tpm_chip_find_get(chip) ... tpm_put_ops(chip)
> >>
> >>ima::tpm: tpm_chip_find_get(chip) ... tpm_put_ops(chip)
> >>
> >>[repeat]
> >>
> >>ima shutdown: tpm_chip_put(chip)
> >Maybe just change tpm_chip_find_get() into tpm_get_ops(chip) and
> >convert all callers?
> 
> And then re-introduce tpm_chip_find_get() for IMA to call ?

You could keep it as 'tpm_chip_find', that seems like a fine name to
me

Jason

^ permalink raw reply	[flat|nested] 21+ messages in thread

* Re: [PATCH v2 3/4] ima: Use tpm_chip_find() and access TPM functions using it
  2018-06-20 20:42 ` [PATCH v2 3/4] ima: Use tpm_chip_find() and access TPM functions using it Stefan Berger
@ 2018-06-21 20:53   ` Mimi Zohar
  2018-06-21 20:59     ` Stefan Berger
  0 siblings, 1 reply; 21+ messages in thread
From: Mimi Zohar @ 2018-06-21 20:53 UTC (permalink / raw)
  To: Stefan Berger, linux-integrity, jarkko.sakkinen; +Cc: jgg, linux-kernel

On Wed, 2018-06-20 at 16:42 -0400, Stefan Berger wrote:
> Rather than accessing the TPM functions using a NULL pointer, which
> causes a lookup for a suitable chip every time, get a hold of a tpm_chip
> and access the TPM functions using this chip. We call the tpm_chip
> ima_tpm_chip and protect it, once initialization is done, using a
> rw_semaphore called ima_tpm_chip_lock.
> 
> Use ima_shutdown to release the tpm_chip.
> 
> Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com>
> ---
>  security/integrity/ima/ima.h        |  3 +++
>  security/integrity/ima/ima_crypto.c | 12 ++++++++++--
>  security/integrity/ima/ima_init.c   | 19 ++++++++++++-------
>  security/integrity/ima/ima_queue.c  |  7 +++++--
>  4 files changed, 30 insertions(+), 11 deletions(-)
> 
> diff --git a/security/integrity/ima/ima.h b/security/integrity/ima/ima.h
> index 354bb5716ce3..53a88d578ca5 100644
> --- a/security/integrity/ima/ima.h
> +++ b/security/integrity/ima/ima.h
> @@ -24,6 +24,7 @@
>  #include <linux/hash.h>
>  #include <linux/tpm.h>
>  #include <linux/audit.h>
> +#include <linux/rwsem.h>
>  #include <crypto/hash_info.h>
> 
>  #include "../integrity.h"
> @@ -56,6 +57,8 @@ extern int ima_policy_flag;
>  extern int ima_used_chip;
>  extern int ima_hash_algo;
>  extern int ima_appraise;
> +extern struct rw_semaphore ima_tpm_chip_lock;
> +extern struct tpm_chip *ima_tpm_chip;


ima_add_templatE_entry() synchronizes appending a measurement to the
measurement list and extending the TPM by taking a lock.  Do we really
need to introduce another lock?

Mimi

> 
>  /* IMA event related data */
>  struct ima_event_data {
> diff --git a/security/integrity/ima/ima_crypto.c b/security/integrity/ima/ima_crypto.c
> index 4e085a17124f..da7715240476 100644
> --- a/security/integrity/ima/ima_crypto.c
> +++ b/security/integrity/ima/ima_crypto.c
> @@ -631,10 +631,18 @@ int ima_calc_buffer_hash(const void *buf, loff_t len,
> 
>  static void __init ima_pcrread(int idx, u8 *pcr)
>  {
> +	int result = 0;
> +
> +	down_read(&ima_tpm_chip_lock);
> +
>  	if (!ima_used_chip)
> -		return;
> +		goto out;
> +
> +	result = tpm_pcr_read(ima_tpm_chip, idx, pcr);
> +out:
> +	up_read(&ima_tpm_chip_lock);
> 
> -	if (tpm_pcr_read(NULL, idx, pcr) != 0)
> +	if (result != 0)
>  		pr_err("Error Communicating to TPM chip\n");
>  }
> 
> diff --git a/security/integrity/ima/ima_init.c b/security/integrity/ima/ima_init.c
> index 8a5258eb32b6..24db06c4f463 100644
> --- a/security/integrity/ima/ima_init.c
> +++ b/security/integrity/ima/ima_init.c
> @@ -28,6 +28,8 @@
>  /* name for boot aggregate entry */
>  static const char *boot_aggregate_name = "boot_aggregate";
>  int ima_used_chip;
> +struct rw_semaphore ima_tpm_chip_lock = __RWSEM_INITIALIZER(ima_tpm_chip_lock);
> +struct tpm_chip *ima_tpm_chip;
> 
>  /* Add the boot aggregate to the IMA measurement list and extend
>   * the PCR register.
> @@ -108,6 +110,13 @@ void __init ima_load_x509(void)
>  static int ima_shutdown(struct notifier_block *this, unsigned long action,
>  			void *data)
>  {
> +	down_write(&ima_tpm_chip_lock);
> +	if (ima_tpm_chip) {
> +		tpm_chip_put(ima_tpm_chip);
> +		ima_tpm_chip = NULL;
> +		ima_used_chip = 0;
> +	}
> +	up_write(&ima_tpm_chip_lock);
>  	return NOTIFY_DONE;
>  }
> 
> @@ -118,19 +127,15 @@ static struct notifier_block ima_reboot_notifier = {
> 
>  int __init ima_init(void)
>  {
> -	u8 pcr_i[TPM_DIGEST_SIZE];
>  	int rc;
> 
>  	register_reboot_notifier(&ima_reboot_notifier);
> 
> -	ima_used_chip = 0;
> -	rc = tpm_pcr_read(NULL, 0, pcr_i);
> -	if (rc == 0)
> -		ima_used_chip = 1;
> +	ima_tpm_chip = tpm_chip_find();
> 
> +	ima_used_chip = (ima_tpm_chip != NULL);
>  	if (!ima_used_chip)
> -		pr_info("No TPM chip found, activating TPM-bypass! (rc=%d)\n",
> -			rc);
> +		pr_info("No TPM chip found, activating TPM-bypass!\n");
> 
>  	rc = integrity_init_keyring(INTEGRITY_KEYRING_IMA);
>  	if (rc)
> diff --git a/security/integrity/ima/ima_queue.c b/security/integrity/ima/ima_queue.c
> index 418f35e38015..6c9427939a28 100644
> --- a/security/integrity/ima/ima_queue.c
> +++ b/security/integrity/ima/ima_queue.c
> @@ -142,10 +142,13 @@ static int ima_pcr_extend(const u8 *hash, int pcr)
>  {
>  	int result = 0;
> 
> +	down_read(&ima_tpm_chip_lock);
>  	if (!ima_used_chip)
> -		return result;
> +		goto out;
> 
> -	result = tpm_pcr_extend(NULL, pcr, hash);
> +	result = tpm_pcr_extend(ima_tpm_chip, pcr, hash);
> +out:
> +	up_read(&ima_tpm_chip_lock);
>  	if (result != 0)
>  		pr_err("Error Communicating to TPM chip, result: %d\n", result);
>  	return result;


^ permalink raw reply	[flat|nested] 21+ messages in thread

* Re: [PATCH v2 3/4] ima: Use tpm_chip_find() and access TPM functions using it
  2018-06-21 20:53   ` Mimi Zohar
@ 2018-06-21 20:59     ` Stefan Berger
  2018-06-22  3:25       ` Jason Gunthorpe
  0 siblings, 1 reply; 21+ messages in thread
From: Stefan Berger @ 2018-06-21 20:59 UTC (permalink / raw)
  To: Mimi Zohar, linux-integrity, jarkko.sakkinen; +Cc: jgg, linux-kernel

On 06/21/2018 04:53 PM, Mimi Zohar wrote:
> On Wed, 2018-06-20 at 16:42 -0400, Stefan Berger wrote:
>> Rather than accessing the TPM functions using a NULL pointer, which
>> causes a lookup for a suitable chip every time, get a hold of a tpm_chip
>> and access the TPM functions using this chip. We call the tpm_chip
>> ima_tpm_chip and protect it, once initialization is done, using a
>> rw_semaphore called ima_tpm_chip_lock.
>>
>> Use ima_shutdown to release the tpm_chip.
>>
>> Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com>
>> ---
>>   security/integrity/ima/ima.h        |  3 +++
>>   security/integrity/ima/ima_crypto.c | 12 ++++++++++--
>>   security/integrity/ima/ima_init.c   | 19 ++++++++++++-------
>>   security/integrity/ima/ima_queue.c  |  7 +++++--
>>   4 files changed, 30 insertions(+), 11 deletions(-)
>>
>> diff --git a/security/integrity/ima/ima.h b/security/integrity/ima/ima.h
>> index 354bb5716ce3..53a88d578ca5 100644
>> --- a/security/integrity/ima/ima.h
>> +++ b/security/integrity/ima/ima.h
>> @@ -24,6 +24,7 @@
>>   #include <linux/hash.h>
>>   #include <linux/tpm.h>
>>   #include <linux/audit.h>
>> +#include <linux/rwsem.h>
>>   #include <crypto/hash_info.h>
>>
>>   #include "../integrity.h"
>> @@ -56,6 +57,8 @@ extern int ima_policy_flag;
>>   extern int ima_used_chip;
>>   extern int ima_hash_algo;
>>   extern int ima_appraise;
>> +extern struct rw_semaphore ima_tpm_chip_lock;
>> +extern struct tpm_chip *ima_tpm_chip;
>
> ima_add_templatE_entry() synchronizes appending a measurement to the
> measurement list and extending the TPM by taking a lock.  Do we really
> need to introduce another lock?

This lock protects the ima_tpm_chip from going from != NULL to NULL in 
the ima_shutdown function. Basically, a global pointer accessed by 
concurrent threads should be protected if its value can change. However, 
in this case ima_shutdown would be called so late that there shouldn't 
be concurrency anymore. Though, I found it better to protect it. Maybe 
someone else has an opinion?

     Stefan
>
> Mimi
>
>>   /* IMA event related data */
>>   struct ima_event_data {
>> diff --git a/security/integrity/ima/ima_crypto.c b/security/integrity/ima/ima_crypto.c
>> index 4e085a17124f..da7715240476 100644
>> --- a/security/integrity/ima/ima_crypto.c
>> +++ b/security/integrity/ima/ima_crypto.c
>> @@ -631,10 +631,18 @@ int ima_calc_buffer_hash(const void *buf, loff_t len,
>>
>>   static void __init ima_pcrread(int idx, u8 *pcr)
>>   {
>> +	int result = 0;
>> +
>> +	down_read(&ima_tpm_chip_lock);
>> +
>>   	if (!ima_used_chip)
>> -		return;
>> +		goto out;
>> +
>> +	result = tpm_pcr_read(ima_tpm_chip, idx, pcr);
>> +out:
>> +	up_read(&ima_tpm_chip_lock);
>>
>> -	if (tpm_pcr_read(NULL, idx, pcr) != 0)
>> +	if (result != 0)
>>   		pr_err("Error Communicating to TPM chip\n");
>>   }
>>
>> diff --git a/security/integrity/ima/ima_init.c b/security/integrity/ima/ima_init.c
>> index 8a5258eb32b6..24db06c4f463 100644
>> --- a/security/integrity/ima/ima_init.c
>> +++ b/security/integrity/ima/ima_init.c
>> @@ -28,6 +28,8 @@
>>   /* name for boot aggregate entry */
>>   static const char *boot_aggregate_name = "boot_aggregate";
>>   int ima_used_chip;
>> +struct rw_semaphore ima_tpm_chip_lock = __RWSEM_INITIALIZER(ima_tpm_chip_lock);
>> +struct tpm_chip *ima_tpm_chip;
>>
>>   /* Add the boot aggregate to the IMA measurement list and extend
>>    * the PCR register.
>> @@ -108,6 +110,13 @@ void __init ima_load_x509(void)
>>   static int ima_shutdown(struct notifier_block *this, unsigned long action,
>>   			void *data)
>>   {
>> +	down_write(&ima_tpm_chip_lock);
>> +	if (ima_tpm_chip) {
>> +		tpm_chip_put(ima_tpm_chip);
>> +		ima_tpm_chip = NULL;
>> +		ima_used_chip = 0;
>> +	}
>> +	up_write(&ima_tpm_chip_lock);
>>   	return NOTIFY_DONE;
>>   }
>>
>> @@ -118,19 +127,15 @@ static struct notifier_block ima_reboot_notifier = {
>>
>>   int __init ima_init(void)
>>   {
>> -	u8 pcr_i[TPM_DIGEST_SIZE];
>>   	int rc;
>>
>>   	register_reboot_notifier(&ima_reboot_notifier);
>>
>> -	ima_used_chip = 0;
>> -	rc = tpm_pcr_read(NULL, 0, pcr_i);
>> -	if (rc == 0)
>> -		ima_used_chip = 1;
>> +	ima_tpm_chip = tpm_chip_find();
>>
>> +	ima_used_chip = (ima_tpm_chip != NULL);
>>   	if (!ima_used_chip)
>> -		pr_info("No TPM chip found, activating TPM-bypass! (rc=%d)\n",
>> -			rc);
>> +		pr_info("No TPM chip found, activating TPM-bypass!\n");
>>
>>   	rc = integrity_init_keyring(INTEGRITY_KEYRING_IMA);
>>   	if (rc)
>> diff --git a/security/integrity/ima/ima_queue.c b/security/integrity/ima/ima_queue.c
>> index 418f35e38015..6c9427939a28 100644
>> --- a/security/integrity/ima/ima_queue.c
>> +++ b/security/integrity/ima/ima_queue.c
>> @@ -142,10 +142,13 @@ static int ima_pcr_extend(const u8 *hash, int pcr)
>>   {
>>   	int result = 0;
>>
>> +	down_read(&ima_tpm_chip_lock);
>>   	if (!ima_used_chip)
>> -		return result;
>> +		goto out;
>>
>> -	result = tpm_pcr_extend(NULL, pcr, hash);
>> +	result = tpm_pcr_extend(ima_tpm_chip, pcr, hash);
>> +out:
>> +	up_read(&ima_tpm_chip_lock);
>>   	if (result != 0)
>>   		pr_err("Error Communicating to TPM chip, result: %d\n", result);
>>   	return result;



^ permalink raw reply	[flat|nested] 21+ messages in thread

* Re: [PATCH v2 3/4] ima: Use tpm_chip_find() and access TPM functions using it
  2018-06-21 20:59     ` Stefan Berger
@ 2018-06-22  3:25       ` Jason Gunthorpe
  2018-06-22 11:40         ` Stefan Berger
  0 siblings, 1 reply; 21+ messages in thread
From: Jason Gunthorpe @ 2018-06-22  3:25 UTC (permalink / raw)
  To: Stefan Berger; +Cc: Mimi Zohar, linux-integrity, jarkko.sakkinen, linux-kernel

On Thu, Jun 21, 2018 at 04:59:55PM -0400, Stefan Berger wrote:
> On 06/21/2018 04:53 PM, Mimi Zohar wrote:
> >On Wed, 2018-06-20 at 16:42 -0400, Stefan Berger wrote:
> >>Rather than accessing the TPM functions using a NULL pointer, which
> >>causes a lookup for a suitable chip every time, get a hold of a tpm_chip
> >>and access the TPM functions using this chip. We call the tpm_chip
> >>ima_tpm_chip and protect it, once initialization is done, using a
> >>rw_semaphore called ima_tpm_chip_lock.
> >>
> >>Use ima_shutdown to release the tpm_chip.
> >>
> >>Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com>
> >>  security/integrity/ima/ima.h        |  3 +++
> >>  security/integrity/ima/ima_crypto.c | 12 ++++++++++--
> >>  security/integrity/ima/ima_init.c   | 19 ++++++++++++-------
> >>  security/integrity/ima/ima_queue.c  |  7 +++++--
> >>  4 files changed, 30 insertions(+), 11 deletions(-)
> >>
> >>diff --git a/security/integrity/ima/ima.h b/security/integrity/ima/ima.h
> >>index 354bb5716ce3..53a88d578ca5 100644
> >>+++ b/security/integrity/ima/ima.h
> >>@@ -24,6 +24,7 @@
> >>  #include <linux/hash.h>
> >>  #include <linux/tpm.h>
> >>  #include <linux/audit.h>
> >>+#include <linux/rwsem.h>
> >>  #include <crypto/hash_info.h>
> >>
> >>  #include "../integrity.h"
> >>@@ -56,6 +57,8 @@ extern int ima_policy_flag;
> >>  extern int ima_used_chip;
> >>  extern int ima_hash_algo;
> >>  extern int ima_appraise;
> >>+extern struct rw_semaphore ima_tpm_chip_lock;
> >>+extern struct tpm_chip *ima_tpm_chip;
> >
> >ima_add_templatE_entry() synchronizes appending a measurement to the
> >measurement list and extending the TPM by taking a lock.  Do we really
> >need to introduce another lock?
> 
> This lock protects the ima_tpm_chip from going from != NULL to NULL in the
> ima_shutdown function. Basically, a global pointer accessed by concurrent
> threads should be protected if its value can change. However, in this case
> ima_shutdown would be called so late that there shouldn't be concurrency
> anymore. Though, I found it better to protect it. Maybe someone else has an
> opinion?

Why have a shutdown block? There is no harm in holding a kref if the
machine is shutting down.

Jason

^ permalink raw reply	[flat|nested] 21+ messages in thread

* Re: [PATCH v2 3/4] ima: Use tpm_chip_find() and access TPM functions using it
  2018-06-22  3:25       ` Jason Gunthorpe
@ 2018-06-22 11:40         ` Stefan Berger
  2018-06-22 14:19           ` Jason Gunthorpe
  0 siblings, 1 reply; 21+ messages in thread
From: Stefan Berger @ 2018-06-22 11:40 UTC (permalink / raw)
  To: Jason Gunthorpe
  Cc: Mimi Zohar, linux-integrity, jarkko.sakkinen, linux-kernel

On 06/21/2018 11:25 PM, Jason Gunthorpe wrote:
> On Thu, Jun 21, 2018 at 04:59:55PM -0400, Stefan Berger wrote:
>> On 06/21/2018 04:53 PM, Mimi Zohar wrote:
>>> On Wed, 2018-06-20 at 16:42 -0400, Stefan Berger wrote:
>>>> Rather than accessing the TPM functions using a NULL pointer, which
>>>> causes a lookup for a suitable chip every time, get a hold of a tpm_chip
>>>> and access the TPM functions using this chip. We call the tpm_chip
>>>> ima_tpm_chip and protect it, once initialization is done, using a
>>>> rw_semaphore called ima_tpm_chip_lock.
>>>>
>>>> Use ima_shutdown to release the tpm_chip.
>>>>
>>>> Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com>
>>>>   security/integrity/ima/ima.h        |  3 +++
>>>>   security/integrity/ima/ima_crypto.c | 12 ++++++++++--
>>>>   security/integrity/ima/ima_init.c   | 19 ++++++++++++-------
>>>>   security/integrity/ima/ima_queue.c  |  7 +++++--
>>>>   4 files changed, 30 insertions(+), 11 deletions(-)
>>>>
>>>> diff --git a/security/integrity/ima/ima.h b/security/integrity/ima/ima.h
>>>> index 354bb5716ce3..53a88d578ca5 100644
>>>> +++ b/security/integrity/ima/ima.h
>>>> @@ -24,6 +24,7 @@
>>>>   #include <linux/hash.h>
>>>>   #include <linux/tpm.h>
>>>>   #include <linux/audit.h>
>>>> +#include <linux/rwsem.h>
>>>>   #include <crypto/hash_info.h>
>>>>
>>>>   #include "../integrity.h"
>>>> @@ -56,6 +57,8 @@ extern int ima_policy_flag;
>>>>   extern int ima_used_chip;
>>>>   extern int ima_hash_algo;
>>>>   extern int ima_appraise;
>>>> +extern struct rw_semaphore ima_tpm_chip_lock;
>>>> +extern struct tpm_chip *ima_tpm_chip;
>>> ima_add_templatE_entry() synchronizes appending a measurement to the
>>> measurement list and extending the TPM by taking a lock.  Do we really
>>> need to introduce another lock?
>> This lock protects the ima_tpm_chip from going from != NULL to NULL in the
>> ima_shutdown function. Basically, a global pointer accessed by concurrent
>> threads should be protected if its value can change. However, in this case
>> ima_shutdown would be called so late that there shouldn't be concurrency
>> anymore. Though, I found it better to protect it. Maybe someone else has an
>> opinion?
> Why have a shutdown block? There is no harm in holding a kref if the
> machine is shutting down.

Looking around at other drivers' usage of the reboot notifier, I find 
other drivers as well that use spinlocks or mutexes during the shutdown. 
Besides that, we do have the shutdown block already when device_shutdown 
calls tpm_class_shutdown() and we get the ops_sem.

     Stefan
>
> Jason
>


^ permalink raw reply	[flat|nested] 21+ messages in thread

* Re: [PATCH v2 3/4] ima: Use tpm_chip_find() and access TPM functions using it
  2018-06-22 11:40         ` Stefan Berger
@ 2018-06-22 14:19           ` Jason Gunthorpe
  0 siblings, 0 replies; 21+ messages in thread
From: Jason Gunthorpe @ 2018-06-22 14:19 UTC (permalink / raw)
  To: Stefan Berger; +Cc: Mimi Zohar, linux-integrity, jarkko.sakkinen, linux-kernel

On Fri, Jun 22, 2018 at 07:40:37AM -0400, Stefan Berger wrote:
> On 06/21/2018 11:25 PM, Jason Gunthorpe wrote:
> >On Thu, Jun 21, 2018 at 04:59:55PM -0400, Stefan Berger wrote:
> >>On 06/21/2018 04:53 PM, Mimi Zohar wrote:
> >>>On Wed, 2018-06-20 at 16:42 -0400, Stefan Berger wrote:
> >>>>Rather than accessing the TPM functions using a NULL pointer, which
> >>>>causes a lookup for a suitable chip every time, get a hold of a tpm_chip
> >>>>and access the TPM functions using this chip. We call the tpm_chip
> >>>>ima_tpm_chip and protect it, once initialization is done, using a
> >>>>rw_semaphore called ima_tpm_chip_lock.
> >>>>
> >>>>Use ima_shutdown to release the tpm_chip.
> >>>>
> >>>>Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com>
> >>>>  security/integrity/ima/ima.h        |  3 +++
> >>>>  security/integrity/ima/ima_crypto.c | 12 ++++++++++--
> >>>>  security/integrity/ima/ima_init.c   | 19 ++++++++++++-------
> >>>>  security/integrity/ima/ima_queue.c  |  7 +++++--
> >>>>  4 files changed, 30 insertions(+), 11 deletions(-)
> >>>>
> >>>>diff --git a/security/integrity/ima/ima.h b/security/integrity/ima/ima.h
> >>>>index 354bb5716ce3..53a88d578ca5 100644
> >>>>+++ b/security/integrity/ima/ima.h
> >>>>@@ -24,6 +24,7 @@
> >>>>  #include <linux/hash.h>
> >>>>  #include <linux/tpm.h>
> >>>>  #include <linux/audit.h>
> >>>>+#include <linux/rwsem.h>
> >>>>  #include <crypto/hash_info.h>
> >>>>
> >>>>  #include "../integrity.h"
> >>>>@@ -56,6 +57,8 @@ extern int ima_policy_flag;
> >>>>  extern int ima_used_chip;
> >>>>  extern int ima_hash_algo;
> >>>>  extern int ima_appraise;
> >>>>+extern struct rw_semaphore ima_tpm_chip_lock;
> >>>>+extern struct tpm_chip *ima_tpm_chip;
> >>>ima_add_templatE_entry() synchronizes appending a measurement to the
> >>>measurement list and extending the TPM by taking a lock.  Do we really
> >>>need to introduce another lock?
> >>This lock protects the ima_tpm_chip from going from != NULL to NULL in the
> >>ima_shutdown function. Basically, a global pointer accessed by concurrent
> >>threads should be protected if its value can change. However, in this case
> >>ima_shutdown would be called so late that there shouldn't be concurrency
> >>anymore. Though, I found it better to protect it. Maybe someone else has an
> >>opinion?
> >Why have a shutdown block? There is no harm in holding a kref if the
> >machine is shutting down.
> 
> Looking around at other drivers' usage of the reboot notifier, I find other
> drivers as well that use spinlocks or mutexes during the shutdown. Besides
> that, we do have the shutdown block already when device_shutdown calls
> tpm_class_shutdown() and we get the ops_sem.

But the shutdown handler in TPM an actual purpose, we are doing
something to the persistent state in the TPM itself during shutdown.

I can't see why IMA needs a shutdown handler. You shouldn't add one
'just because'

Jason

^ permalink raw reply	[flat|nested] 21+ messages in thread

end of thread, other threads:[~2018-06-22 14:20 UTC | newest]

Thread overview: 21+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2018-06-20 20:42 [PATCH v2 0/4] Have IMA find and use a tpm_chip until system shutdown Stefan Berger
2018-06-20 20:42 ` [PATCH v2 1/4] tpm: Implement tpm_chip_find() and tpm_chip_put() for other subsystems Stefan Berger
2018-06-20 20:50   ` Jason Gunthorpe
2018-06-20 21:13     ` Stefan Berger
2018-06-21 17:13     ` Jarkko Sakkinen
2018-06-21 17:15   ` Jarkko Sakkinen
2018-06-21 17:27     ` Stefan Berger
2018-06-21 17:45     ` Stefan Berger
2018-06-21 17:56       ` Jason Gunthorpe
2018-06-21 18:19         ` Stefan Berger
2018-06-21 19:06           ` Jason Gunthorpe
2018-06-21 20:14             ` Stefan Berger
2018-06-21 20:51               ` Jason Gunthorpe
2018-06-20 20:42 ` [PATCH v2 2/4] ima: Implement ima_shutdown and register it as a reboot_notifier Stefan Berger
2018-06-20 20:42 ` [PATCH v2 3/4] ima: Use tpm_chip_find() and access TPM functions using it Stefan Berger
2018-06-21 20:53   ` Mimi Zohar
2018-06-21 20:59     ` Stefan Berger
2018-06-22  3:25       ` Jason Gunthorpe
2018-06-22 11:40         ` Stefan Berger
2018-06-22 14:19           ` Jason Gunthorpe
2018-06-20 20:42 ` [PATCH v2 4/4] ima: Get rid of ima_used_chip and use ima_tpm_chip != NULL instead Stefan Berger

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).