Skip to contentSkip to conversation

tpm: fixes uninitialized allocated banks for IBM vtpm driver

PATCH In review Nayna Jain 7 participants 13 messages Jul 4, 2019 – Jul 8, 2019

Subsystems (2)
lore
Share & tools
lore

Nayna Jain

author
#0 lore raw

PATCH 1/1 tpm: fixes uninitialized allocated banks for IBM vtpm driver

The nr_allocated_banks and allocated banks are initialized as part of tpm_chip_register. Currently, this is done as part of auto startup function. However, some drivers, like the ibm vtpm driver, do not run auto startup during initialization. This results in uninitialized memory issue and causes a kernel panic during boot.

This patch moves the pcr allocation outside the auto startup function into tpm_chip_register. This ensures that allocated banks are initialized in any case.

Fixes 879b589210a9 ("tpm: retrieve digest size of unknown algorithms with

PCR read")

---
 drivers/char/tpm/tpm-chip.c | 37 +++++++++++++++++++++++++++++++++++++
 drivers/char/tpm/tpm.h      |  1 +
 drivers/char/tpm/tpm1-cmd.c | 12 ------------
 drivers/char/tpm/tpm2-cmd.c |  6 +-----
 4 files changed, 39 insertions(+), 17 deletions(-)
drivers/char/tpm/tpm-chip.c +37 -0
index 8804c9e916fd..958508bb8379 100644
@@ -550,6 +550,39 @@ static int tpm_add_hwrng(struct tpm_chip *chip)
 	return hwrng_register(&chip->hwrng);
 }
 
+/*
+ * tpm_pcr_allocation() - initializes the chip allocated banks for PCRs
+ */
+static int tpm_pcr_allocation(struct tpm_chip *chip)
+{
+	int rc = 0;
+
+	if (chip->flags & TPM_CHIP_FLAG_TPM2) {
+		rc = tpm2_get_pcr_allocation(chip);
+		if (rc)
+			goto out;
+	}
+
+	/* Initialize TPM 1.2 */
+	chip->allocated_banks = kcalloc(1, sizeof(*chip->allocated_banks),
+			GFP_KERNEL);
+	if (!chip->allocated_banks) {
+		rc = -ENOMEM;
+		goto out;
+	}
+
+	chip->allocated_banks[0].alg_id = TPM_ALG_SHA1;
+	chip->allocated_banks[0].digest_size = hash_digest_size[HASH_ALGO_SHA1];
+	chip->allocated_banks[0].crypto_id = HASH_ALGO_SHA1;
+	chip->nr_allocated_banks = 1;
+
+	return 0;
+out:
+	if (rc < 0)
+		rc = -ENODEV;
+	return rc;
+}
+
 /*
  * tpm_chip_register() - create a character device for the TPM chip
  * @chip: TPM chip to use.
@@ -573,6 +606,10 @@ int tpm_chip_register(struct tpm_chip *chip)
 	if (rc)
 		return rc;
 
+	rc = tpm_pcr_allocation(chip);
+	if (rc)
+		return rc;
+
 	tpm_sysfs_add_device(chip);
 
 	rc = tpm_bios_log_setup(chip);
drivers/char/tpm/tpm.h +1 -0
index 2cce072f25b5..eabe6b755fa6 100644
@@ -454,6 +454,7 @@ int tpm2_unseal_trusted(struct tpm_chip *chip,
 ssize_t tpm2_get_tpm_pt(struct tpm_chip *chip, u32 property_id,
 			u32 *value, const char *desc);
 
+ssize_t tpm2_get_pcr_allocation(struct tpm_chip *chip);
 int tpm2_auto_startup(struct tpm_chip *chip);
 void tpm2_shutdown(struct tpm_chip *chip, u16 shutdown_type);
 unsigned long tpm2_calc_ordinal_duration(struct tpm_chip *chip, u32 ordinal);
drivers/char/tpm/tpm1-cmd.c +0 -12
index 85dcf2654d11..ec5f3693c096 100644
@@ -696,18 +696,6 @@ int tpm1_auto_startup(struct tpm_chip *chip)
 		goto out;
 	}
 
-	chip->allocated_banks = kcalloc(1, sizeof(*chip->allocated_banks),
-					GFP_KERNEL);
-	if (!chip->allocated_banks) {
-		rc = -ENOMEM;
-		goto out;
-	}
-
-	chip->allocated_banks[0].alg_id = TPM_ALG_SHA1;
-	chip->allocated_banks[0].digest_size = hash_digest_size[HASH_ALGO_SHA1];
-	chip->allocated_banks[0].crypto_id = HASH_ALGO_SHA1;
-	chip->nr_allocated_banks = 1;
-
 	return rc;
 out:
 	if (rc > 0)
drivers/char/tpm/tpm2-cmd.c +1 -5
index e74c5b7b64bf..b4384d0e3741 100644
@@ -841,7 +841,7 @@ struct tpm2_pcr_selection {
 	u8  pcr_select[3];
 } __packed;
 
-static ssize_t tpm2_get_pcr_allocation(struct tpm_chip *chip)
+ssize_t tpm2_get_pcr_allocation(struct tpm_chip *chip)
 {
 	struct tpm2_pcr_selection pcr_selection;
 	struct tpm_buf buf;
@@ -1041,10 +1041,6 @@ int tpm2_auto_startup(struct tpm_chip *chip)
 			goto out;
 	}
 
-	rc = tpm2_get_pcr_allocation(chip);
-	if (rc)
-		goto out;
-
 	rc = tpm2_get_cc_attrs_tbl(chip);
 
 out:
Signature · 1 line
2.20.1
maintainer Reviewed-by on v1 1/1
#1 lore raw
Quoted 6 lines · Nayna Jain

The nr_allocated_banks and allocated banks are initialized as part of tpm_chip_register. Currently, this is done as part of auto startup function. However, some drivers, like the ibm vtpm driver, do not run auto startup during initialization. This results in uninitialized memory issue and causes a kernel panic during boot.

Quoted 7 lines

This patch moves the pcr allocation outside the auto startup function into tpm_chip_register. This ensures that allocated banks are initialized in any case.

Fixes 879b589210a9 ("tpm: retrieve digest size of unknown algorithms with

PCR read")

Tested-by
#2 lore raw
Quoted 10 lines · Mimi Zohar
Quoted 6 lines · Nayna Jain

The nr_allocated_banks and allocated banks are initialized as part of tpm_chip_register. Currently, this is done as part of auto startup function. However, some drivers, like the ibm vtpm driver, do not run auto startup during initialization. This results in uninitialized memory issue and causes a kernel panic during boot.

Quoted 8 lines
Quoted 7 lines

This patch moves the pcr allocation outside the auto startup function into tpm_chip_register. This ensures that allocated banks are initialized in any case.

Fixes 879b589210a9 ("tpm: retrieve digest size of unknown algorithms with

PCR read")

Thanks for the fix. Kernel boots fine with this fix.

Thanks
-Sachin

#3 lore raw
Quoted 13 lines · On Thu, 4 Jul 2019 19:26:36 +0530 Sachin Sant <[email protected]>…
Quoted 13 lines · Mimi Zohar
Quoted 10 lines · Nayna Jain

The nr_allocated_banks and allocated banks are initialized as part of tpm_chip_register. Currently, this is done as part of auto startup function. However, some drivers, like the ibm vtpm driver, do not run auto startup during initialization. This results in uninitialized memory issue and causes a kernel panic during boot.

This patch moves the pcr allocation outside the auto startup function into tpm_chip_register. This ensures that allocated banks are initialized in any case.

Quoted 9 lines
Quoted 4 lines
Quoted 3 lines
Fixes 879b589210a9 ("tpm: retrieve digest size of unknown algorithms with

PCR read")

Thanks for the fix. Kernel boots fine with this fix.

Thanks

Michal

#4 lore raw
In reply to Nayna Jain
Quoted 6 lines · Nayna Jain

The nr_allocated_banks and allocated banks are initialized as part of tpm_chip_register. Currently, this is done as part of auto startup function. However, some drivers, like the ibm vtpm driver, do not run auto startup during initialization. This results in uninitialized memory issue and causes a kernel panic during boot.

Quoted 7 lines

This patch moves the pcr allocation outside the auto startup function into tpm_chip_register. This ensures that allocated banks are initialized in any case.

Fixes 879b589210a9 ("tpm: retrieve digest size of unknown algorithms with

PCR read")

Please add

It is missing. Michal is there a chance you could try this out once Nayna send a new version?

Quoted 12 lines
---
 drivers/char/tpm/tpm-chip.c | 37 +++++++++++++++++++++++++++++++++++++
 drivers/char/tpm/tpm.h      |  1 +
 drivers/char/tpm/tpm1-cmd.c | 12 ------------
 drivers/char/tpm/tpm2-cmd.c |  6 +-----
 4 files changed, 39 insertions(+), 17 deletions(-)
drivers/char/tpm/tpm-chip.c +0 -0
index 8804c9e916fd..958508bb8379 100644

drivers/char/tpm/tpm-chip.c static int tpm_add_hwrng(struct tpm_chip *chip) · line 556

Quoted 7 lines
drivers/char/tpm/tpm-chip.c +4 -0
@@ -550,6 +550,39 @@ static int tpm_add_hwrng(struct tpm_chip *chip)
 	return hwrng_register(&chip->hwrng);
 }
 
+/*
+ * tpm_pcr_allocation() - initializes the chip allocated banks for PCRs
+ */
+static int tpm_pcr_allocation(struct tpm_chip *chip)

Why that name and not tpm_get_pcr_allocation()? Do not get why "get_" has been dropped. Please add it back.

Would be senseful to create tpm1_get_pcr_allocation() to tpm1-cmd.c now that a new function needs to be introduced anyway. Please do it for that for TPM 1.x part.

/Jarkko

#5 lore raw
Quoted 15 lines · On Fri, 05 Jul 2019 13:42:18 +0300 Jarkko Sakkinen…
Quoted 6 lines · Nayna Jain

The nr_allocated_banks and allocated banks are initialized as part of tpm_chip_register. Currently, this is done as part of auto startup function. However, some drivers, like the ibm vtpm driver, do not run auto startup during initialization. This results in uninitialized memory issue and causes a kernel panic during boot.

Quoted 7 lines

This patch moves the pcr allocation outside the auto startup function into tpm_chip_register. This ensures that allocated banks are initialized in any case.

Fixes 879b589210a9 ("tpm: retrieve digest size of unknown algorithms with

PCR read")

Quoted 6 lines

Please add

It is missing. Michal is there a chance you could try this out once Nayna send a new version?

Should be easy to test. Without the patch the machine crashes on boot so it is quite obvious case. I have a few VMs with the vTPM available.

Thanks

Michal

#6 lore raw
In reply to Jarkko Sakkinen
Quoted 15 lines · Jarkko Sakkinen
Quoted 6 lines · Nayna Jain

The nr_allocated_banks and allocated banks are initialized as part of tpm_chip_register. Currently, this is done as part of auto startup function. However, some drivers, like the ibm vtpm driver, do not run auto startup during initialization. This results in uninitialized memory issue and causes a kernel panic during boot.

Quoted 7 lines

This patch moves the pcr allocation outside the auto startup function into tpm_chip_register. This ensures that allocated banks are initialized in any case.

Fixes 879b589210a9 ("tpm: retrieve digest size of unknown algorithms with

PCR read")

Quoted 6 lines

Please add

It is missing. Michal is there a chance you could try this out once Nayna send a new version?

Hey, I saw Michal's tested-by. I can do that minor reorg cosmetic bits myeslf and add Micha's tag.

Some issue with the network but I'll push a commit soonish.

/Jarkko

#7 lore raw
In reply to Nayna Jain
Quoted 48 lines · Nayna Jain

The nr_allocated_banks and allocated banks are initialized as part of tpm_chip_register. Currently, this is done as part of auto startup function. However, some drivers, like the ibm vtpm driver, do not run auto startup during initialization. This results in uninitialized memory issue and causes a kernel panic during boot.

This patch moves the pcr allocation outside the auto startup function into tpm_chip_register. This ensures that allocated banks are initialized in any case.

Fixes 879b589210a9 ("tpm: retrieve digest size of unknown algorithms with

PCR read")

---
  drivers/char/tpm/tpm-chip.c | 37 +++++++++++++++++++++++++++++++++++++
  drivers/char/tpm/tpm.h      |  1 +
  drivers/char/tpm/tpm1-cmd.c | 12 ------------
  drivers/char/tpm/tpm2-cmd.c |  6 +-----
  4 files changed, 39 insertions(+), 17 deletions(-)
drivers/char/tpm/tpm-chip.c +20 -0
index 8804c9e916fd..958508bb8379 100644
@@ -550,6 +550,39 @@ static int tpm_add_hwrng(struct tpm_chip *chip)
  	return hwrng_register(&chip->hwrng);
  }
 
+/*
+ * tpm_pcr_allocation() - initializes the chip allocated banks for PCRs
+ */
+static int tpm_pcr_allocation(struct tpm_chip *chip)
+{
+	int rc = 0;
+
+	if (chip->flags & TPM_CHIP_FLAG_TPM2) {
+		rc = tpm2_get_pcr_allocation(chip);
+		if (rc)
+			goto out;
+	}
+
+	/* Initialize TPM 1.2 */
+	chip->allocated_banks = kcalloc(1, sizeof(*chip->allocated_banks),
+			GFP_KERNEL);
+	if (!chip->allocated_banks) {
+		rc = -ENOMEM;
+		goto out;
+	}

drivers/char/tpm/tpm-chip.c static int tpm_add_hwrng(struct tpm_chip *chip) · line 582

Quoted 10 lines
drivers/char/tpm/tpm-chip.c +10 -0
+
+	chip->allocated_banks[0].alg_id = TPM_ALG_SHA1;
+	chip->allocated_banks[0].digest_size = hash_digest_size[HASH_ALGO_SHA1];
+	chip->allocated_banks[0].crypto_id = HASH_ALGO_SHA1;
+	chip->nr_allocated_banks = 1;
+
+	return 0;
+out:
+	if (rc < 0)
+		rc = -ENODEV;

The old code where you lifted this from said:

out:
    if (rc > 0)
        rc = -ENODEV;
    return rc;

It would not overwrite -ENOMEM with -ENODEV but yours does.

I think the correct fix would be to use:

if (rc > 0)

    rc = -ENODEV;

drivers/char/tpm/tpm-chip.c int tpm_chip_register(struct tpm_chip *chip) · line 607

Quoted 9 lines
drivers/char/tpm/tpm-chip.c +3 -0
+	return rc;
+}
+
  /*
   * tpm_chip_register() - create a character device for the TPM chip
   * @chip: TPM chip to use.
@@ -573,6 +606,10 @@ int tpm_chip_register(struct tpm_chip *chip)
  	if (rc)
  		return rc;

Above this is tpm_chip_stop(chip) because (afaik) none of the following function calls in tpm_chip_register() needed the TPM, but now with tpm_pcr_allocation() you will need to send a command to the TPM. So I would say you should move the tpm_chip_stop() into the error branch visible above and also after the tpm_pcr_allocation().

drivers/char/tpm/tpm-chip.c int tpm_chip_register(struct tpm_chip *chip) · line 609

Quoted 1 line
drivers/char/tpm/tpm-chip.c +1 -0
+	rc = tpm_pcr_allocation(chip);

tpm_chip_stop(chip);

drivers/char/tpm/tpm-chip.c int tpm_chip_register(struct tpm_chip *chip) · line 615

Quoted 6 lines
drivers/char/tpm/tpm-chip.c +3 -0
+	if (rc)
+		return rc;
+
  	tpm_sysfs_add_device(chip);
 
  	rc = tpm_bios_log_setup(chip);

  Stefan

#8 lore raw
Quoted 69 lines · Stefan Berger
Quoted 49 lines · Nayna Jain

The nr_allocated_banks and allocated banks are initialized as part of tpm_chip_register. Currently, this is done as part of auto startup function. However, some drivers, like the ibm vtpm driver, do not run auto startup during initialization. This results in uninitialized memory issue and causes a kernel panic during boot.

This patch moves the pcr allocation outside the auto startup function into tpm_chip_register. This ensures that allocated banks are initialized in any case.

Fixes 879b589210a9 ("tpm: retrieve digest size of unknown algorithms with

PCR read")

---
  drivers/char/tpm/tpm-chip.c | 37 +++++++++++++++++++++++++++++++++++++ drivers/char/tpm/tpm.h      |  1 + drivers/char/tpm/tpm1-cmd.c | 12 ------------ drivers/char/tpm/tpm2-cmd.c |  6 +-----

  4 files changed, 39 insertions(+), 17 deletions(-)
drivers/char/tpm/tpm-chip.c +21 -0
index 8804c9e916fd..958508bb8379 100644
@@ -550,6 +550,39 @@ static int tpm_add_hwrng(struct tpm_chip *chip)
       return hwrng_register(&chip->hwrng);
   }
 
+/*
+ * tpm_pcr_allocation() - initializes the chip allocated banks for PCRs
+ */
+static int tpm_pcr_allocation(struct tpm_chip *chip)
+{
+    int rc = 0;
+
+    if (chip->flags & TPM_CHIP_FLAG_TPM2) {
+        rc = tpm2_get_pcr_allocation(chip);
+        if (rc)
+            goto out;
+    }
+
+    /* Initialize TPM 1.2 */
+    chip->allocated_banks = kcalloc(1, sizeof(*chip->allocated_banks),
+            GFP_KERNEL);
+    if (!chip->allocated_banks) {
+        rc = -ENOMEM;
+        goto out;
+    }
+

drivers/char/tpm/tpm-chip.c static int tpm_add_hwrng(struct tpm_chip *chip) · line 583

Quoted 10 lines
drivers/char/tpm/tpm-chip.c +9 -0
+    chip->allocated_banks[0].alg_id = TPM_ALG_SHA1;
+    chip->allocated_banks[0].digest_size = 
 hash_digest_size[HASH_ALGO_SHA1];
+    chip->allocated_banks[0].crypto_id = HASH_ALGO_SHA1;
+    chip->nr_allocated_banks = 1;
+
+    return 0;
+out:
+    if (rc < 0)
+        rc = -ENODEV;

The old code where you lifted this from said:

out:
    if (rc > 0)
        rc = -ENODEV;
    return rc;

Quoted 8 lines

It would not overwrite -ENOMEM with -ENODEV but yours does.

I think the correct fix would be to use:

if (rc > 0)

    rc = -ENODEV;

Yes. I think I misread it. Thanks Stefan. Will fix this..

Quoted 17 lines

drivers/char/tpm/tpm-chip.c int tpm_chip_register(struct tpm_chip *chip) · line 607

Quoted 13 lines
drivers/char/tpm/tpm-chip.c +3 -0
+    return rc;
+}
+
   /*
    * tpm_chip_register() - create a character device for the TPM chip

   * @chip: TPM chip to use.

drivers/char/tpm/tpm-chip.c +0 -0
@@ -573,6 +606,10 @@ int tpm_chip_register(struct tpm_chip *chip)
       if (rc)
           return rc;

Above this is tpm_chip_stop(chip) because (afaik) none of the following function calls in tpm_chip_register() needed the TPM, but now with tpm_pcr_allocation() you will need to send a command to the TPM. So I would say you should move the tpm_chip_stop() into the error branch visible above and also after the tpm_pcr_allocation().

Quoted 2 lines

drivers/char/tpm/tpm-chip.c int tpm_chip_register(struct tpm_chip *chip) · line 609

Quoted 1 line
drivers/char/tpm/tpm-chip.c +1 -0
+    rc = tpm_pcr_allocation(chip);

tpm_chip_stop(chip);

I am not sure of the purpose of tpm_stop_chip(), so I have left it as it is. Jarkko, what do you think about the change ?

Thanks & Regards,
         - Nayna

#9 lore raw
Quoted 2 lines · Nayna

I am not sure of the purpose of tpm_stop_chip(), so I have left it as it is. Jarkko, what do you think about the change ?

Stefan right. Your does not work, or will randomly work or not work depending on the chip.

You need to turn the TPM on with tpm_chip_start() and turn it off with tpm_chip_stop() once you are done. This is done in tpm_chip_register() before calling tpm_auto_startup().

TPM power management was once in tpm_transmit() but not anymore after my patch set that removed nested tpm_transmit() calls.

While you're on it please take into account my earlier feedback.

Also, short summary could be "tpm: tpm_ibm_vtpm: Fix unallocated banks"

Some oddballs in your patch that I have to ask.

if (chip->flags & TPM_CHIP_FLAG_TPM2) {
	rc = tpm2_get_pcr_allocation(chip);
	if (rc)
		goto out;
}

chip->allocated_banks = kcalloc(1, sizeof(*chip->allocated_banks),

		GFP_KERNEL);
if (!chip->allocated_banks) {
	rc = -ENOMEM;
	goto out;
}

Why you don't return on site and instead jump somewhere? Also the 2nd line for kcalloc() is misaligned.

out:

	if (rc < 0)
		rc = -ENODEV;

This will cause a new regression i.e. you let TPM error codes through.

To summarize this patch fixes one regression and introduces two completely new ones...

/Jarkko`

#10 lore raw
Quoted 2 lines · Jarkko Sakkinen

To summarize this patch fixes one regression and introduces two completely new ones...

Anyway, take the time and update it. The principle is right anyway. I'll merge it once it is in a better shape...

/Jarkko

#11 lore raw
In reply to Jarkko Sakkinen
Quoted 36 lines · Jarkko Sakkinen
Quoted 2 lines · Nayna

I am not sure of the purpose of tpm_stop_chip(), so I have left it as it is. Jarkko, what do you think about the change ?

Stefan right. Your does not work, or will randomly work or not work depending on the chip.

You need to turn the TPM on with tpm_chip_start() and turn it off with tpm_chip_stop() once you are done. This is done in tpm_chip_register() before calling tpm_auto_startup().

TPM power management was once in tpm_transmit() but not anymore after my patch set that removed nested tpm_transmit() calls.

While you're on it please take into account my earlier feedback.

Also, short summary could be "tpm: tpm_ibm_vtpm: Fix unallocated banks"

Some oddballs in your patch that I have to ask.

if (chip->flags & TPM_CHIP_FLAG_TPM2) {
	rc = tpm2_get_pcr_allocation(chip);
	if (rc)
		goto out;
}

chip->allocated_banks = kcalloc(1, sizeof(*chip->allocated_banks),

		GFP_KERNEL);
if (!chip->allocated_banks) {
	rc = -ENOMEM;
	goto out;
}

Why you don't return on site and instead jump somewhere? Also the 2nd line for kcalloc() is misaligned.

out:

Quoted 8 lines
	if (rc < 0)
		rc = -ENODEV;

This will cause a new regression i.e. you let TPM error codes through.

To summarize this patch fixes one regression and introduces two completely new ones...

Thanks Jarkko. I just now posted the v2 version that includes your and Stefan's feedbacks.

Thanks & Regards,
       - Nayna

#12 lore raw
Quoted 2 lines · Nayna

Thanks Jarkko. I just now posted the v2 version that includes your and Stefan's feedbacks.

Looks almost legit :-)

/Jarkko