Re-establish ability for exclusive TPM access to userspace

From: Jonathan McDowell <[email protected]>

There is an is_open lock on /dev/tpm<n> that dates back to at least 2013, but it only prevents multiple accesses via *this* interface. It is perfectly possible for userspace to use /dev/tpmrm<n>, or the kernel to use the internal interfaces, to access the TPM.

This can cause problems with userspace expecting exclusive access and something changing state underneath it, for example while performing a TPM firmware upgrade.

Close the userspace loophole by changing the simple bit lock to a full read/write mutex. Direct /dev/tpm<n> access needs an exclusive write lock, the resource broker continues to allow concurrent access *except* when /dev/tpm<n> is open.

Signed-off-by Jonathan McDowell <[email protected]>

---
 drivers/char/tpm/tpm-chip.c  |  1 +
 drivers/char/tpm/tpm-dev.c   | 14 ++++++++------
 drivers/char/tpm/tpmrm-dev.c | 20 ++++++++++++++++++--
 include/linux/tpm.h          |  3 ++-
 4 files changed, 29 insertions(+), 9 deletions(-)

drivers/char/tpm/tpm-chip.c +1 -0

index e25daf2396d3..8c8e9054762a 100644
@@ -338,6 +338,7 @@ struct tpm_chip *tpm_chip_alloc(struct device *pdev,
 
 	mutex_init(&chip->tpm_mutex);
 	init_rwsem(&chip->ops_sem);
+	init_rwsem(&chip->open_lock);
 
 	chip->ops = ops;
 

drivers/char/tpm/tpm-dev.c +8 -6

index 97c94b5e9340..80c4b3f3ad18 100644
@@ -22,10 +22,12 @@ static int tpm_open(struct inode *inode, struct file *file)
 
 	chip = container_of(inode->i_cdev, struct tpm_chip, cdev);
 
-	/* It's assured that the chip will be opened just once,
-	 * by the check of is_open variable, which is protected
-	 * by driver_lock. */
-	if (test_and_set_bit(0, &chip->is_open)) {
+	/*
+	 * Only one client is allowed to have /dev/tpm0 open at a time, so we
+	 * treat it as a write lock. The shared /dev/tpmrm0 is treated as a
+	 * read lock.
+	 */
+	if (!down_write_trylock(&chip->open_lock)) {
 		dev_dbg(&chip->dev, "Another process owns this TPM\n");
 		return -EBUSY;
 	}

@@ -39,7 +41,7 @@ static int tpm_open(struct inode *inode, struct file *file)
 	return 0;
 
  out:
-	clear_bit(0, &chip->is_open);
+	up_write(&chip->open_lock);
 	return -ENOMEM;
 }
 

@@ -51,7 +53,7 @@ static int tpm_release(struct inode *inode, struct file *file)
 	struct file_priv *priv = file->private_data;
 
 	tpm_common_release(file, priv);
-	clear_bit(0, &priv->chip->is_open);
+	up_write(&priv->chip->open_lock);
 	kfree(priv);
 
 	return 0;

drivers/char/tpm/tpmrm-dev.c +12 -1

index c25df7ea064e..40c139a080b6 100644
@@ -17,19 +17,34 @@ static int tpmrm_open(struct inode *inode, struct file *file)
 	int rc;
 
 	chip = container_of(inode->i_cdev, struct tpm_chip, cdevs);
+
+	/*
+	 * Only one client is allowed to have /dev/tpm0 open at a time, so we
+	 * treat it as a write lock. The shared /dev/tpmrm0 is treated as a
+	 * read lock.
+	 */
+	if (!down_read_trylock(&chip->open_lock)) {
+		dev_dbg(&chip->dev, "Another process owns this TPM\n");
+		return -EBUSY;
+	}
+
 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
 	if (priv == NULL)
-		return -ENOMEM;
+		goto out;
 
 	rc = tpm2_init_space(&priv->space, TPM2_SPACE_BUFFER_SIZE);
 	if (rc) {

drivers/char/tpm/tpmrm-dev.c +3 -0

+	up_read(&chip->open_lock);
+	return -ENOMEM;
 }
 
 static int tpmrm_release(struct inode *inode, struct file *file)

@@ -40,6 +55,7 @@ static int tpmrm_release(struct inode *inode, struct file *file)
 	tpm_common_release(file, fpriv);
 	tpm2_del_space(fpriv->chip, &priv->space);
 	kfree(priv);
+	up_read(&fpriv->chip->open_lock);
 
 	return 0;
 }

include/linux/tpm.h +2 -1

index b0e9eb5ef022..548362d20b32 100644
@@ -22,6 +22,7 @@
 #include <linux/cdev.h>
 #include <linux/fs.h>
 #include <linux/highmem.h>
+#include <linux/rwsem.h>
 #include <crypto/hash_info.h>
 #include <crypto/aes.h>
 

@@ -168,7 +169,7 @@ struct tpm_chip {
 	unsigned int flags;
 
 	int dev_num;		/* /dev/tpm# */
-	unsigned long is_open;	/* only one allowed */
+	struct rw_semaphore open_lock;
 
 	char hwrng_name[64];
 	struct hwrng hwrng;

Signature · 1 line

2.51.0

From: Jonathan McDowell <[email protected]>

tpm_find_get_ops() looks for the first valid TPM if the caller passes in NULL. All internal users have been converted to either associate themselves with a TPM directly, or call tpm_default_chip() as part of their setup. Remove the no longer necessary tpm_find_get_ops().

Signed-off-by Jonathan McDowell <[email protected]>

---
 drivers/char/tpm/tpm-chip.c      | 36 --------------------------------
 drivers/char/tpm/tpm-interface.c | 20 ++++++++++++++----
 drivers/char/tpm/tpm.h           |  1 -
 drivers/char/tpm/tpm_tis_core.c  |  3 +--
 4 files changed, 17 insertions(+), 43 deletions(-)

drivers/char/tpm/tpm-chip.c +0 -36

index 8c8e9054762a..ba906966721a 100644
@@ -230,42 +230,6 @@ struct tpm_chip *tpm_default_chip(void)
 }
 EXPORT_SYMBOL_GPL(tpm_default_chip);
 
-/**
- * tpm_find_get_ops() - find and reserve a TPM chip
- * @chip:	a &struct tpm_chip instance, %NULL for the default chip
- *
- * Finds a TPM chip and reserves its class device and operations. The chip must
- * be released with tpm_put_ops() after use.
- * This function is for internal use only. It supports existing TPM callers
- * by accepting NULL, but those callers should be converted to pass in a chip
- * directly.
- *
- * Return:
- * A reserved &struct tpm_chip instance.
- * %NULL if a chip is not found.
- * %NULL if the chip is not available.
- */
-struct tpm_chip *tpm_find_get_ops(struct tpm_chip *chip)
-{
-	int rc;
-
-	if (chip) {
-		if (!tpm_try_get_ops(chip))
-			return chip;
-		return NULL;
-	}
-
-	chip = tpm_default_chip();
-	if (!chip)
-		return NULL;
-	rc = tpm_try_get_ops(chip);
-	/* release additional reference we got from tpm_default_chip() */
-	put_device(&chip->dev);
-	if (rc)
-		return NULL;
-	return chip;
-}
-
 /**
  * tpm_dev_release() - free chip memory and the device number
  * @dev: the character device for the TPM chip

drivers/char/tpm/tpm-interface.c +16 -4

index b71725827743..8f65dc06a157 100644
@@ -313,10 +313,13 @@ int tpm_is_tpm2(struct tpm_chip *chip)
 {
 	int rc;
 
-	chip = tpm_find_get_ops(chip);
 	if (!chip)
 		return -ENODEV;
 
+	rc = tpm_try_get_ops(chip);
+	if (rc)
+		return rc;
+
 	rc = (chip->flags & TPM_CHIP_FLAG_TPM2) != 0;
 
 	tpm_put_ops(chip);

@@ -338,10 +341,13 @@ int tpm_pcr_read(struct tpm_chip *chip, u32 pcr_idx,
 {
 	int rc;
 
-	chip = tpm_find_get_ops(chip);
 	if (!chip)
 		return -ENODEV;
 
+	rc = tpm_try_get_ops(chip);
+	if (rc)
+		return rc;
+
 	if (chip->flags & TPM_CHIP_FLAG_TPM2)
 		rc = tpm2_pcr_read(chip, pcr_idx, digest, NULL);
 	else

@@ -369,10 +375,13 @@ int tpm_pcr_extend(struct tpm_chip *chip, u32 pcr_idx,
 	int rc;
 	int i;
 
-	chip = tpm_find_get_ops(chip);
 	if (!chip)
 		return -ENODEV;
 
+	rc = tpm_try_get_ops(chip);
+	if (rc)
+		return rc;
+
 	for (i = 0; i < chip->nr_allocated_banks; i++) {
 		if (digests[i].alg_id != chip->allocated_banks[i].alg_id) {
 			rc = -EINVAL;

@@ -492,10 +501,13 @@ int tpm_get_random(struct tpm_chip *chip, u8 *out, size_t max)
 	if (!out || max > TPM_MAX_RNG_DATA)
 		return -EINVAL;
 
-	chip = tpm_find_get_ops(chip);
 	if (!chip)
 		return -ENODEV;
 
+	rc = tpm_try_get_ops(chip);
+	if (rc)
+		return rc;
+
 	if (chip->flags & TPM_CHIP_FLAG_TPM2)
 		rc = tpm2_get_random(chip, out, max);
 	else

drivers/char/tpm/tpm.h +0 -1

index 7bb87fa5f7a1..9c158c55c05f 100644
@@ -267,7 +267,6 @@ static inline void tpm_msleep(unsigned int delay_msec)
 int tpm_chip_bootstrap(struct tpm_chip *chip);
 int tpm_chip_start(struct tpm_chip *chip);
 void tpm_chip_stop(struct tpm_chip *chip);
-struct tpm_chip *tpm_find_get_ops(struct tpm_chip *chip);
 
 struct tpm_chip *tpm_chip_alloc(struct device *dev,
 				const struct tpm_class_ops *ops);

drivers/char/tpm/tpm_tis_core.c +0 -0

index 4b12c4b9da8b..73b94f4daf4b 100644

From: Jonathan McDowell <[email protected]>

There are situations where userspace might reasonably desire exclusive access to the TPM, or the kernel's internal context saving + flushing may cause issues, for example when performing firmware upgrades. Extend the locking already used for avoiding concurrent userspace access to prevent internal users of the TPM when /dev/tpm<n> is in use.

The few internal users who already hold the open_lock are changed to use tpm_internal_(try_get|put)_ops, with the old tpm_(try_get|put)_ops functions changing to obtain read access to the open_lock. We return -EBUSY when another user has exclusive access, rather than adding waits.

Signed-off-by Jonathan McDowell <[email protected]>

---
 drivers/char/tpm/tpm-chip.c       | 53 +++++++++++++++++++++++++------
 drivers/char/tpm/tpm-dev-common.c |  8 ++---
 drivers/char/tpm/tpm.h            |  2 ++
 drivers/char/tpm/tpm2-space.c     |  5 ++-
 4 files changed, 52 insertions(+), 16 deletions(-)

drivers/char/tpm/tpm-chip.c +1 -1

index ba906966721a..3d69ccff4c2a 100644
@@ -144,7 +144,7 @@ void tpm_chip_stop(struct tpm_chip *chip)
 EXPORT_SYMBOL_GPL(tpm_chip_stop);
 
 /**
- * tpm_try_get_ops() - Get a ref to the tpm_chip
+ * tpm_internal_try_get_ops() - Get a ref to the tpm_chip
  * @chip: Chip to ref
  *
  * The caller must already have some kind of locking to ensure that chip is

From: Jonathan McDowell <[email protected]>

Given that /dev/tpm has not had exclusive access to the TPM since the existence of the kernel resource broker and other internal users, stop defaulted to exclusive access to the first client that opens the device. Continue to support exclusive access, but only with the use of the O_EXCL flag on device open.

Signed-off-by Jonathan McDowell <[email protected]>

---
 drivers/char/tpm/tpm-dev.c | 25 +++++++++++++++++++------
 drivers/char/tpm/tpm-dev.h |  1 +
 2 files changed, 20 insertions(+), 6 deletions(-)

drivers/char/tpm/tpm-dev.c +19 -6

index 80c4b3f3ad18..8921bbb541c1 100644
@@ -19,15 +19,21 @@ static int tpm_open(struct inode *inode, struct file *file)
 {
 	struct tpm_chip *chip;
 	struct file_priv *priv;
+	int rc;
 
 	chip = container_of(inode->i_cdev, struct tpm_chip, cdev);
 
 	/*
-	 * Only one client is allowed to have /dev/tpm0 open at a time, so we
-	 * treat it as a write lock. The shared /dev/tpmrm0 is treated as a
-	 * read lock.
+	 * If a client uses the O_EXCL flag then it expects to be the only TPM
+	 * user, so we treat it as a write lock. Otherwise we do as /dev/tpmrm
+	 * and use a read lock.
 	 */
-	if (!down_write_trylock(&chip->open_lock)) {
+	if (file->f_flags & O_EXCL)
+		rc = down_write_trylock(&chip->open_lock);
+	else
+		rc = down_read_trylock(&chip->open_lock);
+
+	if (!rc) {
 		dev_dbg(&chip->dev, "Another process owns this TPM\n");
 		return -EBUSY;
 	}

@@ -35,13 +41,17 @@ static int tpm_open(struct inode *inode, struct file *file)
 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
 	if (priv == NULL)
 		goto out;
+	priv->exclusive = (file->f_flags & O_EXCL);
 
 	tpm_common_open(file, chip, priv, NULL);
 
 	return 0;
 
  out:
-	up_write(&chip->open_lock);
+	if (file->f_flags & O_EXCL)
+		up_write(&chip->open_lock);
+	else
+		up_read(&chip->open_lock);
 	return -ENOMEM;
 }
 

@@ -53,7 +63,10 @@ static int tpm_release(struct inode *inode, struct file *file)
 	struct file_priv *priv = file->private_data;
 
 	tpm_common_release(file, priv);
-	up_write(&priv->chip->open_lock);
+	if (priv->exclusive)
+		up_write(&priv->chip->open_lock);
+	else
+		up_read(&priv->chip->open_lock);
 	kfree(priv);
 
 	return 0;

drivers/char/tpm/tpm-dev.h +1 -0

index f3742bcc73e3..0ad8504c73e4 100644
@@ -17,6 +17,7 @@ struct file_priv {
 	ssize_t response_length;
 	bool response_read;
 	bool command_enqueued;
+	bool exclusive;
 
 	u8 data_buffer[TPM_BUFSIZE];
 };

Signature · 1 line

2.51.0

From: Jonathan McDowell <[email protected]>

There is an is_open lock on /dev/tpm<n> that dates back to at least 2013, but it only prevents multiple accesses via *this* interface. It is perfectly possible for userspace to use /dev/tpmrm<n>, or the kernel to use the internal interfaces, to access the TPM. For tooling expecting exclusive access, such as firmware updates, this can cause issues.

Close the userspace loophole by changing the simple bit lock to a full read/write mutex. Direct /dev/tpm<n> access needs an exclusive write lock, the resource broker continues to allow concurrent access *except* when /dev/tpm<n> is open.

Signed-off-by Jonathan McDowell <[email protected]>

---
v2: Change error path label to err instead of out. Rework commit
message.

 drivers/char/tpm/tpm-chip.c  |  1 +
 drivers/char/tpm/tpm-dev.c   | 14 ++++++++------
 drivers/char/tpm/tpmrm-dev.c | 20 ++++++++++++++++++--
 include/linux/tpm.h          |  3 ++-
 4 files changed, 29 insertions(+), 9 deletions(-)

drivers/char/tpm/tpm-chip.c +1 -0

index e25daf2396d3..8c8e9054762a 100644
@@ -338,6 +338,7 @@ struct tpm_chip *tpm_chip_alloc(struct device *pdev,
 
 	mutex_init(&chip->tpm_mutex);
 	init_rwsem(&chip->ops_sem);
+	init_rwsem(&chip->open_lock);
 
 	chip->ops = ops;
 

drivers/char/tpm/tpm-dev.c +8 -6

index 97c94b5e9340..80c4b3f3ad18 100644
@@ -22,10 +22,12 @@ static int tpm_open(struct inode *inode, struct file *file)
 
 	chip = container_of(inode->i_cdev, struct tpm_chip, cdev);
 
-	/* It's assured that the chip will be opened just once,
-	 * by the check of is_open variable, which is protected
-	 * by driver_lock. */
-	if (test_and_set_bit(0, &chip->is_open)) {
+	/*
+	 * Only one client is allowed to have /dev/tpm0 open at a time, so we
+	 * treat it as a write lock. The shared /dev/tpmrm0 is treated as a
+	 * read lock.
+	 */
+	if (!down_write_trylock(&chip->open_lock)) {
 		dev_dbg(&chip->dev, "Another process owns this TPM\n");
 		return -EBUSY;
 	}

@@ -39,7 +41,7 @@ static int tpm_open(struct inode *inode, struct file *file)
 	return 0;
 
  out:
-	clear_bit(0, &chip->is_open);
+	up_write(&chip->open_lock);
 	return -ENOMEM;
 }
 

@@ -51,7 +53,7 @@ static int tpm_release(struct inode *inode, struct file *file)
 	struct file_priv *priv = file->private_data;
 
 	tpm_common_release(file, priv);
-	clear_bit(0, &priv->chip->is_open);
+	up_write(&priv->chip->open_lock);
 	kfree(priv);
 
 	return 0;

drivers/char/tpm/tpmrm-dev.c +18 -2

index c25df7ea064e..13322dd9ac9e 100644
@@ -17,19 +17,34 @@ static int tpmrm_open(struct inode *inode, struct file *file)
 	int rc;
 
 	chip = container_of(inode->i_cdev, struct tpm_chip, cdevs);
+
+	/*
+	 * Only one client is allowed to have /dev/tpm0 open at a time, so we
+	 * treat it as a write lock. The shared /dev/tpmrm0 is treated as a
+	 * read lock.
+	 */
+	if (!down_read_trylock(&chip->open_lock)) {
+		dev_dbg(&chip->dev, "Another process owns this TPM\n");
+		return -EBUSY;
+	}
+
 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
 	if (priv == NULL)
-		return -ENOMEM;
+		goto err;
 
 	rc = tpm2_init_space(&priv->space, TPM2_SPACE_BUFFER_SIZE);
 	if (rc) {
 		kfree(priv);
-		return -ENOMEM;
+		goto err;
 	}
 
 	tpm_common_open(file, chip, &priv->priv, &priv->space);
 
 	return 0;
+
+err:
+	up_read(&chip->open_lock);
+	return -ENOMEM;
 }
 
 static int tpmrm_release(struct inode *inode, struct file *file)

@@ -40,6 +55,7 @@ static int tpmrm_release(struct inode *inode, struct file *file)
 	tpm_common_release(file, fpriv);
 	tpm2_del_space(fpriv->chip, &priv->space);
 	kfree(priv);
+	up_read(&fpriv->chip->open_lock);
 
 	return 0;
 }

include/linux/tpm.h +2 -1

index b0e9eb5ef022..548362d20b32 100644
@@ -22,6 +22,7 @@
 #include <linux/cdev.h>
 #include <linux/fs.h>
 #include <linux/highmem.h>
+#include <linux/rwsem.h>
 #include <crypto/hash_info.h>
 #include <crypto/aes.h>
 

@@ -168,7 +169,7 @@ struct tpm_chip {
 	unsigned int flags;
 
 	int dev_num;		/* /dev/tpm# */
-	unsigned long is_open;	/* only one allowed */
+	struct rw_semaphore open_lock;
 
 	char hwrng_name[64];
 	struct hwrng hwrng;

Signature · 1 line

2.51.0

From: Jonathan McDowell <[email protected]>

tpm_find_get_ops() looks for the first valid TPM if the caller passes in NULL. All internal users have been converted to either associate themselves with a TPM directly, or call tpm_default_chip() as part of their setup. Remove the no longer necessary tpm_find_get_ops().

Signed-off-by Jonathan McDowell <[email protected]>

---
 drivers/char/tpm/tpm-chip.c      | 36 --------------------------------
 drivers/char/tpm/tpm-interface.c | 20 ++++++++++++++----
 drivers/char/tpm/tpm.h           |  1 -
 drivers/char/tpm/tpm_tis_core.c  |  3 +--
 4 files changed, 17 insertions(+), 43 deletions(-)

drivers/char/tpm/tpm-chip.c +0 -36

index 8c8e9054762a..ba906966721a 100644
@@ -230,42 +230,6 @@ struct tpm_chip *tpm_default_chip(void)
 }
 EXPORT_SYMBOL_GPL(tpm_default_chip);
 
-/**
- * tpm_find_get_ops() - find and reserve a TPM chip
- * @chip:	a &struct tpm_chip instance, %NULL for the default chip
- *
- * Finds a TPM chip and reserves its class device and operations. The chip must
- * be released with tpm_put_ops() after use.
- * This function is for internal use only. It supports existing TPM callers
- * by accepting NULL, but those callers should be converted to pass in a chip
- * directly.
- *
- * Return:
- * A reserved &struct tpm_chip instance.
- * %NULL if a chip is not found.
- * %NULL if the chip is not available.
- */
-struct tpm_chip *tpm_find_get_ops(struct tpm_chip *chip)
-{
-	int rc;
-
-	if (chip) {
-		if (!tpm_try_get_ops(chip))
-			return chip;
-		return NULL;
-	}
-
-	chip = tpm_default_chip();
-	if (!chip)
-		return NULL;
-	rc = tpm_try_get_ops(chip);
-	/* release additional reference we got from tpm_default_chip() */
-	put_device(&chip->dev);
-	if (rc)
-		return NULL;
-	return chip;
-}
-
 /**
  * tpm_dev_release() - free chip memory and the device number
  * @dev: the character device for the TPM chip

drivers/char/tpm/tpm-interface.c +16 -4

index b71725827743..8f65dc06a157 100644
@@ -313,10 +313,13 @@ int tpm_is_tpm2(struct tpm_chip *chip)
 {
 	int rc;
 
-	chip = tpm_find_get_ops(chip);
 	if (!chip)
 		return -ENODEV;
 
+	rc = tpm_try_get_ops(chip);
+	if (rc)
+		return rc;
+
 	rc = (chip->flags & TPM_CHIP_FLAG_TPM2) != 0;
 
 	tpm_put_ops(chip);

@@ -338,10 +341,13 @@ int tpm_pcr_read(struct tpm_chip *chip, u32 pcr_idx,
 {
 	int rc;
 
-	chip = tpm_find_get_ops(chip);
 	if (!chip)
 		return -ENODEV;
 
+	rc = tpm_try_get_ops(chip);
+	if (rc)
+		return rc;
+
 	if (chip->flags & TPM_CHIP_FLAG_TPM2)
 		rc = tpm2_pcr_read(chip, pcr_idx, digest, NULL);
 	else

@@ -369,10 +375,13 @@ int tpm_pcr_extend(struct tpm_chip *chip, u32 pcr_idx,
 	int rc;
 	int i;
 
-	chip = tpm_find_get_ops(chip);
 	if (!chip)
 		return -ENODEV;
 
+	rc = tpm_try_get_ops(chip);
+	if (rc)
+		return rc;
+
 	for (i = 0; i < chip->nr_allocated_banks; i++) {
 		if (digests[i].alg_id != chip->allocated_banks[i].alg_id) {
 			rc = -EINVAL;

@@ -492,10 +501,13 @@ int tpm_get_random(struct tpm_chip *chip, u8 *out, size_t max)
 	if (!out || max > TPM_MAX_RNG_DATA)
 		return -EINVAL;
 
-	chip = tpm_find_get_ops(chip);
 	if (!chip)
 		return -ENODEV;
 
+	rc = tpm_try_get_ops(chip);
+	if (rc)
+		return rc;
+
 	if (chip->flags & TPM_CHIP_FLAG_TPM2)
 		rc = tpm2_get_random(chip, out, max);
 	else

drivers/char/tpm/tpm.h +0 -1

index 7bb87fa5f7a1..9c158c55c05f 100644
@@ -267,7 +267,6 @@ static inline void tpm_msleep(unsigned int delay_msec)
 int tpm_chip_bootstrap(struct tpm_chip *chip);
 int tpm_chip_start(struct tpm_chip *chip);
 void tpm_chip_stop(struct tpm_chip *chip);
-struct tpm_chip *tpm_find_get_ops(struct tpm_chip *chip);
 
 struct tpm_chip *tpm_chip_alloc(struct device *dev,
 				const struct tpm_class_ops *ops);

drivers/char/tpm/tpm_tis_core.c +1 -2

index 4b12c4b9da8b..73b94f4daf4b 100644
@@ -265,8 +265,7 @@ static u8 tpm_tis_status(struct tpm_chip *chip)
 
 			/*
 			 * Dump stack for forensics, as invalid TPM_STS.x could be
-			 * potentially triggered by impaired tpm_try_get_ops() or
-			 * tpm_find_get_ops().
+			 * potentially triggered by impaired tpm_try_get_ops().
 			 */
 			dump_stack();
 		}

Signature · 1 line

2.51.0

From: Jonathan McDowell <[email protected]>

There are situations where userspace might reasonably desire exclusive access to the TPM, or the kernel's internal context saving + flushing may cause issues, for example when performing firmware upgrades. Extend the locking already used for avoiding concurrent userspace access to prevent internal users of the TPM when /dev/tpm<n> is in use.

The few internal users who already hold the open_lock are changed to use tpm_(try_get|put)_ops_locked, with the old tpm_(try_get|put)_ops functions changing to obtain read access to the open_lock. We return -EBUSY when another user has exclusive access, rather than adding waits.

Signed-off-by Jonathan McDowell <[email protected]>

---
v2: Switch to _locked instead of _internal_ for function names.

 drivers/char/tpm/tpm-chip.c       | 53 +++++++++++++++++++++++++------
 drivers/char/tpm/tpm-dev-common.c |  8 ++---
 drivers/char/tpm/tpm.h            |  2 ++
 drivers/char/tpm/tpm2-space.c     |  5 ++-
 4 files changed, 52 insertions(+), 16 deletions(-)

drivers/char/tpm/tpm-chip.c +44 -9

index ba906966721a..687f6d8cd601 100644
@@ -144,7 +144,7 @@ void tpm_chip_stop(struct tpm_chip *chip)
 EXPORT_SYMBOL_GPL(tpm_chip_stop);
 
 /**
- * tpm_try_get_ops() - Get a ref to the tpm_chip
+ * tpm_try_get_ops_locked() - Get a ref to the tpm_chip
  * @chip: Chip to ref
  *
  * The caller must already have some kind of locking to ensure that chip is

@@ -154,7 +154,7 @@ EXPORT_SYMBOL_GPL(tpm_chip_stop);
  *
  * Returns -ERRNO if the chip could not be got.
  */
-int tpm_try_get_ops(struct tpm_chip *chip)
+int tpm_try_get_ops_locked(struct tpm_chip *chip)
 {
 	int rc = -EIO;
 

@@ -185,22 +185,57 @@ int tpm_try_get_ops(struct tpm_chip *chip)
 	put_device(&chip->dev);
 	return rc;
 }
-EXPORT_SYMBOL_GPL(tpm_try_get_ops);
 
 /**
- * tpm_put_ops() - Release a ref to the tpm_chip
+ * tpm_put_ops_locked() - Release a ref to the tpm_chip
  * @chip: Chip to put
  *
- * This is the opposite pair to tpm_try_get_ops(). After this returns chip may
- * be kfree'd.
+ * This is the opposite pair to tpm_try_get_ops_locked(). After this returns
+ * chip may be kfree'd.
  */
-void tpm_put_ops(struct tpm_chip *chip)
+void tpm_put_ops_locked(struct tpm_chip *chip)
 {
 	tpm_chip_stop(chip);
 	mutex_unlock(&chip->tpm_mutex);
 	up_read(&chip->ops_sem);
 	put_device(&chip->dev);
 }
+
+/**
+ * tpm_try_get_ops() - Get a ref to the tpm_chip
+ * @chip: Chip to ref
+ *
+ * The caller must already have some kind of locking to ensure that chip is
+ * valid. This function will attempt to get the open_lock for the chip,
+ * ensuring no other user is expecting exclusive access, before locking the
+ * chip so that the ops member can be accessed safely. The locking prevents
+ * tpm_chip_unregister from completing, so it should not be held for long
+ * periods.
+ *
+ * Returns -ERRNO if the chip could not be got.
+ */
+int tpm_try_get_ops(struct tpm_chip *chip)
+{
+	if (!down_read_trylock(&chip->open_lock))
+		return -EBUSY;
+
+	return tpm_try_get_ops_locked(chip);
+}
+EXPORT_SYMBOL_GPL(tpm_try_get_ops);
+
+/**
+ * tpm_put_ops() - Release a ref to the tpm_chip
+ * @chip: Chip to put
+ *
+ * This is the opposite pair to tpm_try_get_ops(). After this returns
+ * chip may be kfree'd.
+ */
+void tpm_put_ops(struct tpm_chip *chip)
+{
+	tpm_put_ops_locked(chip);
+
+	up_read(&chip->open_lock);
+}
 EXPORT_SYMBOL_GPL(tpm_put_ops);
 
 /**

@@ -644,10 +679,10 @@ void tpm_chip_unregister(struct tpm_chip *chip)
 #ifdef CONFIG_TCG_TPM2_HMAC
 	int rc;
 
-	rc = tpm_try_get_ops(chip);
+	rc = tpm_try_get_ops_locked(chip);
 	if (!rc) {
 		tpm2_end_auth_session(chip);
-		tpm_put_ops(chip);
+		tpm_put_ops_locked(chip);
 	}
 #endif
 

drivers/char/tpm/tpm-dev-common.c +4 -4

index f2a5e09257dd..0f5bc63411aa 100644
@@ -65,7 +65,7 @@ static void tpm_dev_async_work(struct work_struct *work)
 
 	mutex_lock(&priv->buffer_mutex);
 	priv->command_enqueued = false;
-	ret = tpm_try_get_ops(priv->chip);
+	ret = tpm_try_get_ops_locked(priv->chip);
 	if (ret) {
 		priv->response_length = ret;
 		goto out;

@@ -73,7 +73,7 @@ static void tpm_dev_async_work(struct work_struct *work)
 
 	ret = tpm_dev_transmit(priv->chip, priv->space, priv->data_buffer,
 			       sizeof(priv->data_buffer));
-	tpm_put_ops(priv->chip);
+	tpm_put_ops_locked(priv->chip);
 
 	/*
 	 * If ret is > 0 then tpm_dev_transmit returned the size of the

@@ -220,14 +220,14 @@ ssize_t tpm_common_write(struct file *file, const char __user *buf,
 	 * lock during this period so that the tpm can be unregistered even if
 	 * the char dev is held open.
 	 */
-	if (tpm_try_get_ops(priv->chip)) {
+	if (tpm_try_get_ops_locked(priv->chip)) {
 		ret = -EPIPE;
 		goto out;
 	}
 
 	ret = tpm_dev_transmit(priv->chip, priv->space, priv->data_buffer,
 			       sizeof(priv->data_buffer));
-	tpm_put_ops(priv->chip);
+	tpm_put_ops_locked(priv->chip);
 
 	if (ret > 0) {
 		priv->response_length = ret;

drivers/char/tpm/tpm.h +2 -0

index 9c158c55c05f..c2ec56e2cd24 100644
@@ -272,6 +272,8 @@ struct tpm_chip *tpm_chip_alloc(struct device *dev,
 				const struct tpm_class_ops *ops);
 struct tpm_chip *tpmm_chip_alloc(struct device *pdev,
 				 const struct tpm_class_ops *ops);
+int tpm_try_get_ops_locked(struct tpm_chip *chip);
+void tpm_put_ops_locked(struct tpm_chip *chip);
 int tpm_chip_register(struct tpm_chip *chip);
 void tpm_chip_unregister(struct tpm_chip *chip);
 

drivers/char/tpm/tpm2-space.c +2 -3

index 60354cd53b5c..0ad5e18355e0 100644
@@ -58,10 +58,9 @@ int tpm2_init_space(struct tpm_space *space, unsigned int buf_size)
 
 void tpm2_del_space(struct tpm_chip *chip, struct tpm_space *space)
 {
-
-	if (tpm_try_get_ops(chip) == 0) {
+	if (tpm_try_get_ops_locked(chip) == 0) {
 		tpm2_flush_sessions(chip, space);
-		tpm_put_ops(chip);
+		tpm_put_ops_locked(chip);
 	}
 
 	kfree(space->context_buf);

Signature · 1 line

2.51.0

From: Jonathan McDowell <[email protected]>

Given that /dev/tpm has not had exclusive access to the TPM since the existence of the kernel resource broker and other internal users, stop defaulted to exclusive access to the first client that opens the device. Continue to support exclusive access, but only with the use of the O_EXCL flag on device open.

Signed-off-by Jonathan McDowell <[email protected]>

---
 drivers/char/tpm/tpm-dev.c | 25 +++++++++++++++++++------
 drivers/char/tpm/tpm-dev.h |  1 +
 2 files changed, 20 insertions(+), 6 deletions(-)

drivers/char/tpm/tpm-dev.c +19 -6

index 80c4b3f3ad18..8921bbb541c1 100644
@@ -19,15 +19,21 @@ static int tpm_open(struct inode *inode, struct file *file)
 {
 	struct tpm_chip *chip;
 	struct file_priv *priv;
+	int rc;
 
 	chip = container_of(inode->i_cdev, struct tpm_chip, cdev);
 
 	/*
-	 * Only one client is allowed to have /dev/tpm0 open at a time, so we
-	 * treat it as a write lock. The shared /dev/tpmrm0 is treated as a
-	 * read lock.
+	 * If a client uses the O_EXCL flag then it expects to be the only TPM
+	 * user, so we treat it as a write lock. Otherwise we do as /dev/tpmrm
+	 * and use a read lock.
 	 */
-	if (!down_write_trylock(&chip->open_lock)) {
+	if (file->f_flags & O_EXCL)
+		rc = down_write_trylock(&chip->open_lock);
+	else
+		rc = down_read_trylock(&chip->open_lock);
+
+	if (!rc) {
 		dev_dbg(&chip->dev, "Another process owns this TPM\n");
 		return -EBUSY;
 	}

@@ -35,13 +41,17 @@ static int tpm_open(struct inode *inode, struct file *file)
 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
 	if (priv == NULL)
 		goto out;
+	priv->exclusive = (file->f_flags & O_EXCL);
 
 	tpm_common_open(file, chip, priv, NULL);
 
 	return 0;
 
  out:
-	up_write(&chip->open_lock);
+	if (file->f_flags & O_EXCL)
+		up_write(&chip->open_lock);
+	else
+		up_read(&chip->open_lock);
 	return -ENOMEM;
 }
 

@@ -53,7 +63,10 @@ static int tpm_release(struct inode *inode, struct file *file)
 	struct file_priv *priv = file->private_data;
 
 	tpm_common_release(file, priv);
-	up_write(&priv->chip->open_lock);
+	if (priv->exclusive)
+		up_write(&priv->chip->open_lock);
+	else
+		up_read(&priv->chip->open_lock);
 	kfree(priv);
 
 	return 0;

drivers/char/tpm/tpm-dev.h +1 -0

index f3742bcc73e3..0ad8504c73e4 100644
@@ -17,6 +17,7 @@ struct file_priv {
 	ssize_t response_length;
 	bool response_read;
 	bool command_enqueued;
+	bool exclusive;
 
 	u8 data_buffer[TPM_BUFSIZE];
 };

Signature · 1 line

2.51.0

From: Jonathan McDowell <[email protected]>

There are situations where userspace might reasonably desire exclusive access to the TPM, or the kernel's internal context saving + flushing may cause issues, for example when performing firmware upgrades. Extend the locking already used for avoiding concurrent userspace access to prevent internal users of the TPM when /dev/tpm<n> is in use.

The few internal users who already hold the open_lock are changed to use tpm_internal_(try_get|put)_ops, with the old tpm_(try_get|put)_ops functions changing to obtain read access to the open_lock. We return -EBUSY when another user has exclusive access, rather than adding waits.

Signed-off-by Jonathan McDowell <[email protected]>

---
v2: Switch to _locked instead of _internal_ for function names.
v3: Move to end of patch series.

 drivers/char/tpm/tpm-chip.c       | 53 +++++++++++++++++++++++++------
 drivers/char/tpm/tpm-dev-common.c |  8 ++---
 drivers/char/tpm/tpm.h            |  2 ++
 drivers/char/tpm/tpm2-space.c     |  5 ++-
 4 files changed, 52 insertions(+), 16 deletions(-)

drivers/char/tpm/tpm-chip.c +14 -7

index ba906966721a..687f6d8cd601 100644
@@ -144,7 +144,7 @@ void tpm_chip_stop(struct tpm_chip *chip)
 EXPORT_SYMBOL_GPL(tpm_chip_stop);
 
 /**
- * tpm_try_get_ops() - Get a ref to the tpm_chip
+ * tpm_try_get_ops_locked() - Get a ref to the tpm_chip
  * @chip: Chip to ref
  *
  * The caller must already have some kind of locking to ensure that chip is

@@ -154,7 +154,7 @@ EXPORT_SYMBOL_GPL(tpm_chip_stop);
  *
  * Returns -ERRNO if the chip could not be got.
  */
-int tpm_try_get_ops(struct tpm_chip *chip)
+int tpm_try_get_ops_locked(struct tpm_chip *chip)
 {
 	int rc = -EIO;
 

@@ -185,22 +185,57 @@ int tpm_try_get_ops(struct tpm_chip *chip)
 	put_device(&chip->dev);
 	return rc;
 }
-EXPORT_SYMBOL_GPL(tpm_try_get_ops);
 
 /**
- * tpm_put_ops() - Release a ref to the tpm_chip
+ * tpm_put_ops_locked() - Release a ref to the tpm_chip
  * @chip: Chip to put
  *
- * This is the opposite pair to tpm_try_get_ops(). After this returns chip may
- * be kfree'd.
+ * This is the opposite pair to tpm_try_get_ops_locked(). After this returns
+ * chip may be kfree'd.
  */
-void tpm_put_ops(struct tpm_chip *chip)
+void tpm_put_ops_locked(struct tpm_chip *chip)
 {
 	tpm_chip_stop(chip);
 	mutex_unlock(&chip->tpm_mutex);
 	up_read(&chip->ops_sem);
 	put_device(&chip->dev);
 }
+
+/**
+ * tpm_try_get_ops() - Get a ref to the tpm_chip
+ * @chip: Chip to ref
+ *
+ * The caller must already have some kind of locking to ensure that chip is
+ * valid. This function will attempt to get the open_lock for the chip,
+ * ensuring no other user is expecting exclusive access, before locking the

drivers/char/tpm/tpm-chip.c +20 -2

+
+	return tpm_try_get_ops_locked(chip);
+}
+EXPORT_SYMBOL_GPL(tpm_try_get_ops);
+
+/**
+ * tpm_put_ops() - Release a ref to the tpm_chip
+ * @chip: Chip to put
+ *
+ * This is the opposite pair to tpm_try_get_ops(). After this returns
+ * chip may be kfree'd.
+ */
+void tpm_put_ops(struct tpm_chip *chip)
+{
+	tpm_put_ops_locked(chip);
+
+	up_read(&chip->open_lock);
+}
 EXPORT_SYMBOL_GPL(tpm_put_ops);
 
 /**

@@ -644,10 +679,10 @@ void tpm_chip_unregister(struct tpm_chip *chip)
 #ifdef CONFIG_TCG_TPM2_HMAC
 	int rc;
 
-	rc = tpm_try_get_ops(chip);
+	rc = tpm_try_get_ops_locked(chip);
 	if (!rc) {
 		tpm2_end_auth_session(chip);
-		tpm_put_ops(chip);
+		tpm_put_ops_locked(chip);
 	}
 #endif
 

drivers/char/tpm/tpm-dev-common.c +4 -4

index f2a5e09257dd..0f5bc63411aa 100644
@@ -65,7 +65,7 @@ static void tpm_dev_async_work(struct work_struct *work)
 
 	mutex_lock(&priv->buffer_mutex);
 	priv->command_enqueued = false;
-	ret = tpm_try_get_ops(priv->chip);
+	ret = tpm_try_get_ops_locked(priv->chip);
 	if (ret) {
 		priv->response_length = ret;
 		goto out;

@@ -73,7 +73,7 @@ static void tpm_dev_async_work(struct work_struct *work)
 
 	ret = tpm_dev_transmit(priv->chip, priv->space, priv->data_buffer,
 			       sizeof(priv->data_buffer));
-	tpm_put_ops(priv->chip);
+	tpm_put_ops_locked(priv->chip);
 
 	/*
 	 * If ret is > 0 then tpm_dev_transmit returned the size of the

@@ -220,14 +220,14 @@ ssize_t tpm_common_write(struct file *file, const char __user *buf,
 	 * lock during this period so that the tpm can be unregistered even if
 	 * the char dev is held open.
 	 */
-	if (tpm_try_get_ops(priv->chip)) {
+	if (tpm_try_get_ops_locked(priv->chip)) {
 		ret = -EPIPE;
 		goto out;
 	}
 
 	ret = tpm_dev_transmit(priv->chip, priv->space, priv->data_buffer,
 			       sizeof(priv->data_buffer));
-	tpm_put_ops(priv->chip);
+	tpm_put_ops_locked(priv->chip);
 
 	if (ret > 0) {
 		priv->response_length = ret;

drivers/char/tpm/tpm.h +2 -0

index 02c07fef41ba..57ef8589f5f5 100644
@@ -272,6 +272,8 @@ struct tpm_chip *tpm_chip_alloc(struct device *dev,
 				const struct tpm_class_ops *ops);
 struct tpm_chip *tpmm_chip_alloc(struct device *pdev,
 				 const struct tpm_class_ops *ops);
+int tpm_try_get_ops_locked(struct tpm_chip *chip);
+void tpm_put_ops_locked(struct tpm_chip *chip);
 int tpm_chip_register(struct tpm_chip *chip);
 void tpm_chip_unregister(struct tpm_chip *chip);
 

drivers/char/tpm/tpm2-space.c +2 -3

index 60354cd53b5c..0ad5e18355e0 100644
@@ -58,10 +58,9 @@ int tpm2_init_space(struct tpm_space *space, unsigned int buf_size)
 
 void tpm2_del_space(struct tpm_chip *chip, struct tpm_space *space)
 {
-
-	if (tpm_try_get_ops(chip) == 0) {
+	if (tpm_try_get_ops_locked(chip) == 0) {
 		tpm2_flush_sessions(chip, space);
-		tpm_put_ops(chip);
+		tpm_put_ops_locked(chip);
 	}
 
 	kfree(space->context_buf);

Quoted 78 lines · Jonathan McDowell

From: Jonathan McDowell <[email protected]>

There are situations where userspace might reasonably desire exclusive access to the TPM, or the kernel's internal context saving + flushing may cause issues, for example when performing firmware upgrades. Extend the locking already used for avoiding concurrent userspace access to prevent internal users of the TPM when /dev/tpm<n> is in use.

The few internal users who already hold the open_lock are changed to use tpm_internal_(try_get|put)_ops, with the old tpm_(try_get|put)_ops functions changing to obtain read access to the open_lock. We return -EBUSY when another user has exclusive access, rather than adding waits.

Signed-off-by Jonathan McDowell <[email protected]>

---
v2: Switch to _locked instead of _internal_ for function names.
v3: Move to end of patch series.

 drivers/char/tpm/tpm-chip.c       | 53 +++++++++++++++++++++++++------
 drivers/char/tpm/tpm-dev-common.c |  8 ++---
 drivers/char/tpm/tpm.h            |  2 ++
 drivers/char/tpm/tpm2-space.c     |  5 ++-
 4 files changed, 52 insertions(+), 16 deletions(-)

drivers/char/tpm/tpm-chip.c +14 -7

index ba906966721a..687f6d8cd601 100644
@@ -144,7 +144,7 @@ void tpm_chip_stop(struct tpm_chip *chip)
 EXPORT_SYMBOL_GPL(tpm_chip_stop);
 
 /**
- * tpm_try_get_ops() - Get a ref to the tpm_chip
+ * tpm_try_get_ops_locked() - Get a ref to the tpm_chip
  * @chip: Chip to ref
  *
  * The caller must already have some kind of locking to ensure that chip is

@@ -154,7 +154,7 @@ EXPORT_SYMBOL_GPL(tpm_chip_stop);
  *
  * Returns -ERRNO if the chip could not be got.
  */
-int tpm_try_get_ops(struct tpm_chip *chip)
+int tpm_try_get_ops_locked(struct tpm_chip *chip)
 {
 	int rc = -EIO;
 

@@ -185,22 +185,57 @@ int tpm_try_get_ops(struct tpm_chip *chip)
 	put_device(&chip->dev);
 	return rc;
 }
-EXPORT_SYMBOL_GPL(tpm_try_get_ops);
 
 /**
- * tpm_put_ops() - Release a ref to the tpm_chip
+ * tpm_put_ops_locked() - Release a ref to the tpm_chip
  * @chip: Chip to put
  *
- * This is the opposite pair to tpm_try_get_ops(). After this returns chip may
- * be kfree'd.
+ * This is the opposite pair to tpm_try_get_ops_locked(). After this returns
+ * chip may be kfree'd.
  */
-void tpm_put_ops(struct tpm_chip *chip)
+void tpm_put_ops_locked(struct tpm_chip *chip)
 {
 	tpm_chip_stop(chip);
 	mutex_unlock(&chip->tpm_mutex);
 	up_read(&chip->ops_sem);
 	put_device(&chip->dev);
 }
+
+/**
+ * tpm_try_get_ops() - Get a ref to the tpm_chip
+ * @chip: Chip to ref
+ *
+ * The caller must already have some kind of locking to ensure that chip is
+ * valid. This function will attempt to get the open_lock for the chip,
+ * ensuring no other user is expecting exclusive access, before locking the

drivers/char/tpm/tpm-chip.c int tpm_try_get_ops(struct tpm_chip *chip) · line 220

Quoted 10 lines

drivers/char/tpm/tpm-chip.c +10 -0

+ * chip so that the ops member can be accessed safely. The locking prevents
+ * tpm_chip_unregister from completing, so it should not be held for long
+ * periods.
+ *
+ * Returns -ERRNO if the chip could not be got.
+ */
+int tpm_try_get_ops(struct tpm_chip *chip)
+{
+	if (!down_read_trylock(&chip->open_lock))
+		return -EBUSY;

Hi Jonathan

do I understand it correctly, that a process might open the TPM with O_EXCL, and this will prevent IMA from extending a PCR until that process closes the file descriptor?

If yes, this might be a concern, and I think an additional API to prevent such behavior would be needed (for example when IMA is active, i.e. there is a measurement policy loaded).

Quoted 102 lines

drivers/char/tpm/tpm-chip.c +20 -2

+
+	return tpm_try_get_ops_locked(chip);
+}
+EXPORT_SYMBOL_GPL(tpm_try_get_ops);
+
+/**
+ * tpm_put_ops() - Release a ref to the tpm_chip
+ * @chip: Chip to put
+ *
+ * This is the opposite pair to tpm_try_get_ops(). After this returns
+ * chip may be kfree'd.
+ */
+void tpm_put_ops(struct tpm_chip *chip)
+{
+	tpm_put_ops_locked(chip);
+
+	up_read(&chip->open_lock);
+}
 EXPORT_SYMBOL_GPL(tpm_put_ops);
 
 /**

@@ -644,10 +679,10 @@ void tpm_chip_unregister(struct tpm_chip *chip)
 #ifdef CONFIG_TCG_TPM2_HMAC
 	int rc;
 
-	rc = tpm_try_get_ops(chip);
+	rc = tpm_try_get_ops_locked(chip);
 	if (!rc) {
 		tpm2_end_auth_session(chip);
-		tpm_put_ops(chip);
+		tpm_put_ops_locked(chip);
 	}
 #endif
 

drivers/char/tpm/tpm-dev-common.c +4 -4

index f2a5e09257dd..0f5bc63411aa 100644
@@ -65,7 +65,7 @@ static void tpm_dev_async_work(struct work_struct *work)
 
 	mutex_lock(&priv->buffer_mutex);
 	priv->command_enqueued = false;
-	ret = tpm_try_get_ops(priv->chip);
+	ret = tpm_try_get_ops_locked(priv->chip);
 	if (ret) {
 		priv->response_length = ret;
 		goto out;

@@ -73,7 +73,7 @@ static void tpm_dev_async_work(struct work_struct *work)
 
 	ret = tpm_dev_transmit(priv->chip, priv->space, priv->data_buffer,
 			       sizeof(priv->data_buffer));
-	tpm_put_ops(priv->chip);
+	tpm_put_ops_locked(priv->chip);
 
 	/*
 	 * If ret is > 0 then tpm_dev_transmit returned the size of the

@@ -220,14 +220,14 @@ ssize_t tpm_common_write(struct file *file, const char __user *buf,
 	 * lock during this period so that the tpm can be unregistered even if
 	 * the char dev is held open.
 	 */
-	if (tpm_try_get_ops(priv->chip)) {
+	if (tpm_try_get_ops_locked(priv->chip)) {
 		ret = -EPIPE;
 		goto out;
 	}
 
 	ret = tpm_dev_transmit(priv->chip, priv->space, priv->data_buffer,
 			       sizeof(priv->data_buffer));
-	tpm_put_ops(priv->chip);
+	tpm_put_ops_locked(priv->chip);
 
 	if (ret > 0) {
 		priv->response_length = ret;

drivers/char/tpm/tpm.h +2 -0

index 02c07fef41ba..57ef8589f5f5 100644
@@ -272,6 +272,8 @@ struct tpm_chip *tpm_chip_alloc(struct device *dev,
 				const struct tpm_class_ops *ops);
 struct tpm_chip *tpmm_chip_alloc(struct device *pdev,
 				 const struct tpm_class_ops *ops);
+int tpm_try_get_ops_locked(struct tpm_chip *chip);
+void tpm_put_ops_locked(struct tpm_chip *chip);
 int tpm_chip_register(struct tpm_chip *chip);
 void tpm_chip_unregister(struct tpm_chip *chip);
 

drivers/char/tpm/tpm2-space.c +2 -3

index 60354cd53b5c..0ad5e18355e0 100644
@@ -58,10 +58,9 @@ int tpm2_init_space(struct tpm_space *space, unsigned int buf_size)
 
 void tpm2_del_space(struct tpm_chip *chip, struct tpm_space *space)
 {
-
-	if (tpm_try_get_ops(chip) == 0) {
+	if (tpm_try_get_ops_locked(chip) == 0) {
 		tpm2_flush_sessions(chip, space);
-		tpm_put_ops(chip);
+		tpm_put_ops_locked(chip);
 	}
 

drivers/char/tpm/tpm2-space.c int tpm2_init_space(struct tpm_space *space, unsigned int buf_size) · line 66

Quoted 1 line

drivers/char/tpm/tpm2-space.c +0 -0

 	kfree(space->context_buf);

I hit a problem where ~ 1% of TPM firmware upgrades were failing. Investigating revealed the issue was that although the upgrade tool uses /dev/tpm0 this does not actually prevent access via /dev/tpmrm0, nor internal kernel users. It *does* prevent access to others via /dev/tpm0

So the upgrade process started, the HW RNG came in to get some randomness in the middle, did the HMAC context dance, and confused everything to the point the TPM was no longer visible to the OS even after a reboot.

Thankfully I've been able to recover those devices, but really what I'd like is the ability for a userspace tool to exclusively access the TPM without something coming in behind it. Given the lightweight attempt at locking that already exists I think this was the original intention.

I've reworked this series based on feedback received.

Firstly, it's been reordered TPM sharing functionality doesn't break during bisection.

Secondly, the O_EXCL check has been tightend up to ensure the caller is also opening the device O_RDWR. Callers shouldn't really be opening the TPM except for reading + writing, but this should help guard against unexpected flags usage a bit more.

Finally, this revision keeps the prohibition on more than one user of /dev/tpm#, to avoid unexpected breakages for clients that expect this to guard against multiple invocations. A client only then needs to use O_EXCL if it wants to prevent *all* other access, even with ContextSaves, such as the firmware upgrade case.

(I've sent a separate standalone patch that allows the TPM HW RNG to be disabled at run time, and it's now in -next, but even with that I think something like this is a good idea as well.)

Jonathan McDowell (4):
tpm: Remove tpm_find_get_ops
tpm: Add O_EXCL for exclusive /dev/tpm access
tpm: Include /dev/tpmrm<n> when checking exclusive userspace TPM
access
tpm: Allow for exclusive TPM access when using /dev/tpm<n>

 drivers/char/tpm/tpm-chip.c       | 90 +++++++++++++++----------------
 drivers/char/tpm/tpm-dev-common.c |  8 +--
 drivers/char/tpm/tpm-dev.c        | 35 ++++++++++--
 drivers/char/tpm/tpm-dev.h        |  1 +
 drivers/char/tpm/tpm-interface.c  | 20 +++++--
 drivers/char/tpm/tpm.h            |  3 +-
 drivers/char/tpm/tpm2-space.c     |  5 +-
 drivers/char/tpm/tpm_tis_core.c   |  3 +-
 drivers/char/tpm/tpmrm-dev.c      | 20 ++++++-
 include/linux/tpm.h               |  4 +-
 10 files changed, 124 insertions(+), 65 deletions(-)

Signature · 1 line

2.51.0

Quoted 42 lines · Jonathan McDowell

I hit a problem where ~ 1% of TPM firmware upgrades were failing. Investigating revealed the issue was that although the upgrade tool uses /dev/tpm0 this does not actually prevent access via /dev/tpmrm0, nor internal kernel users. It *does* prevent access to others via /dev/tpm0

So the upgrade process started, the HW RNG came in to get some randomness in the middle, did the HMAC context dance, and confused everything to the point the TPM was no longer visible to the OS even after a reboot.

Thankfully I've been able to recover those devices, but really what I'd like is the ability for a userspace tool to exclusively access the TPM without something coming in behind it. Given the lightweight attempt at locking that already exists I think this was the original intention.

I've reworked this series based on feedback received.

Firstly, it's been reordered TPM sharing functionality doesn't break during bisection.

Secondly, the O_EXCL check has been tightend up to ensure the caller is also opening the device O_RDWR. Callers shouldn't really be opening the TPM except for reading + writing, but this should help guard against unexpected flags usage a bit more.

Finally, this revision keeps the prohibition on more than one user of /dev/tpm#, to avoid unexpected breakages for clients that expect this to guard against multiple invocations. A client only then needs to use O_EXCL if it wants to prevent *all* other access, even with ContextSaves, such as the firmware upgrade case.

(I've sent a separate standalone patch that allows the TPM HW RNG to be disabled at run time, and it's now in -next, but even with that I think something like this is a good idea as well.)

Jonathan McDowell (4):
tpm: Remove tpm_find_get_ops
tpm: Add O_EXCL for exclusive /dev/tpm access
tpm: Include /dev/tpmrm<n> when checking exclusive userspace TPM
access
tpm: Allow for exclusive TPM access when using /dev/tpm<n>

Quoted 15 lines

 drivers/char/tpm/tpm-chip.c       | 90 +++++++++++++++----------------
 drivers/char/tpm/tpm-dev-common.c |  8 +--
 drivers/char/tpm/tpm-dev.c        | 35 ++++++++++--
 drivers/char/tpm/tpm-dev.h        |  1 +
 drivers/char/tpm/tpm-interface.c  | 20 +++++--
 drivers/char/tpm/tpm.h            |  3 +-
 drivers/char/tpm/tpm2-space.c     |  5 +-
 drivers/char/tpm/tpm_tis_core.c   |  3 +-
 drivers/char/tpm/tpmrm-dev.c      | 20 ++++++-
 include/linux/tpm.h               |  4 +-
 10 files changed, 124 insertions(+), 65 deletions(-)

Signature · 1 line

2.51.0

I will put to queue with my tags but I just want to make first sure that we do not break anything.

I'll upgrade my test suite first to have TPM 1.2 tests (which is also needed for my own series) and run it in bunch of configurations. And on TPM2 I check the behavior with TSS2 daemon on / off.

I have no doubts on the code changes, and it is most importantly a security improvement, given that "who has the access and how long" can be deduced for a system configuration. I just feel that with this code change it is better to check and verify everything :-)

Quoted 78 lines · Jonathan McDowell

From: Jonathan McDowell <[email protected]>

There are situations where userspace might reasonably desire exclusive access to the TPM, or the kernel's internal context saving + flushing may cause issues, for example when performing firmware upgrades. Extend the locking already used for avoiding concurrent userspace access to prevent internal users of the TPM when /dev/tpm<n> is in use.

The few internal users who already hold the open_lock are changed to use tpm_internal_(try_get|put)_ops, with the old tpm_(try_get|put)_ops functions changing to obtain read access to the open_lock. We return -EBUSY when another user has exclusive access, rather than adding waits.

Signed-off-by Jonathan McDowell <[email protected]>

---
v2: Switch to _locked instead of _internal_ for function names.
v3: Move to end of patch series.

 drivers/char/tpm/tpm-chip.c       | 53 +++++++++++++++++++++++++------
 drivers/char/tpm/tpm-dev-common.c |  8 ++---
 drivers/char/tpm/tpm.h            |  2 ++
 drivers/char/tpm/tpm2-space.c     |  5 ++-
 4 files changed, 52 insertions(+), 16 deletions(-)

drivers/char/tpm/tpm-chip.c +14 -7

index ba906966721a..687f6d8cd601 100644
@@ -144,7 +144,7 @@ void tpm_chip_stop(struct tpm_chip *chip)
 EXPORT_SYMBOL_GPL(tpm_chip_stop);
 
 /**
- * tpm_try_get_ops() - Get a ref to the tpm_chip
+ * tpm_try_get_ops_locked() - Get a ref to the tpm_chip
  * @chip: Chip to ref
  *
  * The caller must already have some kind of locking to ensure that chip is

@@ -154,7 +154,7 @@ EXPORT_SYMBOL_GPL(tpm_chip_stop);
  *
  * Returns -ERRNO if the chip could not be got.
  */
-int tpm_try_get_ops(struct tpm_chip *chip)
+int tpm_try_get_ops_locked(struct tpm_chip *chip)
 {
 	int rc = -EIO;
 

@@ -185,22 +185,57 @@ int tpm_try_get_ops(struct tpm_chip *chip)
 	put_device(&chip->dev);
 	return rc;
 }
-EXPORT_SYMBOL_GPL(tpm_try_get_ops);
 
 /**
- * tpm_put_ops() - Release a ref to the tpm_chip
+ * tpm_put_ops_locked() - Release a ref to the tpm_chip
  * @chip: Chip to put
  *
- * This is the opposite pair to tpm_try_get_ops(). After this returns chip may
- * be kfree'd.
+ * This is the opposite pair to tpm_try_get_ops_locked(). After this returns
+ * chip may be kfree'd.
  */
-void tpm_put_ops(struct tpm_chip *chip)
+void tpm_put_ops_locked(struct tpm_chip *chip)
 {
 	tpm_chip_stop(chip);
 	mutex_unlock(&chip->tpm_mutex);
 	up_read(&chip->ops_sem);
 	put_device(&chip->dev);
 }
+
+/**
+ * tpm_try_get_ops() - Get a ref to the tpm_chip
+ * @chip: Chip to ref
+ *
+ * The caller must already have some kind of locking to ensure that chip is
+ * valid. This function will attempt to get the open_lock for the chip,
+ * ensuring no other user is expecting exclusive access, before locking the

drivers/char/tpm/tpm-chip.c int tpm_try_get_ops(struct tpm_chip *chip) · line 220

Quoted 10 lines

drivers/char/tpm/tpm-chip.c +10 -0

+ * chip so that the ops member can be accessed safely. The locking prevents
+ * tpm_chip_unregister from completing, so it should not be held for long
+ * periods.
+ *
+ * Returns -ERRNO if the chip could not be got.
+ */
+int tpm_try_get_ops(struct tpm_chip *chip)
+{
+	if (!down_read_trylock(&chip->open_lock))
+		return -EBUSY;

Hi Jonathan

do I understand it correctly, that a process might open the TPM with O_EXCL, and this will prevent IMA from extending a PCR until that process closes the file descriptor?

If yes, this might be a concern, and I think an additional API to prevent such behavior would be needed (for example when IMA is active, i.e. there is a measurement policy loaded).

Thanks

Roberto

Quoted 66 lines · Jarkko Sakkinen

Quoted 42 lines · Jonathan McDowell

I hit a problem where ~ 1% of TPM firmware upgrades were failing. Investigating revealed the issue was that although the upgrade tool uses /dev/tpm0 this does not actually prevent access via /dev/tpmrm0, nor internal kernel users. It *does* prevent access to others via /dev/tpm0

So the upgrade process started, the HW RNG came in to get some randomness in the middle, did the HMAC context dance, and confused everything to the point the TPM was no longer visible to the OS even after a reboot.

Thankfully I've been able to recover those devices, but really what I'd like is the ability for a userspace tool to exclusively access the TPM without something coming in behind it. Given the lightweight attempt at locking that already exists I think this was the original intention.

I've reworked this series based on feedback received.

Firstly, it's been reordered TPM sharing functionality doesn't break during bisection.

Secondly, the O_EXCL check has been tightend up to ensure the caller is also opening the device O_RDWR. Callers shouldn't really be opening the TPM except for reading + writing, but this should help guard against unexpected flags usage a bit more.

Finally, this revision keeps the prohibition on more than one user of /dev/tpm#, to avoid unexpected breakages for clients that expect this to guard against multiple invocations. A client only then needs to use O_EXCL if it wants to prevent *all* other access, even with ContextSaves, such as the firmware upgrade case.

(I've sent a separate standalone patch that allows the TPM HW RNG to be disabled at run time, and it's now in -next, but even with that I think something like this is a good idea as well.)

Jonathan McDowell (4):
tpm: Remove tpm_find_get_ops
tpm: Add O_EXCL for exclusive /dev/tpm access
tpm: Include /dev/tpmrm<n> when checking exclusive userspace TPM
access
tpm: Allow for exclusive TPM access when using /dev/tpm<n>

Quoted 15 lines

 drivers/char/tpm/tpm-chip.c       | 90 +++++++++++++++----------------
 drivers/char/tpm/tpm-dev-common.c |  8 +--
 drivers/char/tpm/tpm-dev.c        | 35 ++++++++++--
 drivers/char/tpm/tpm-dev.h        |  1 +
 drivers/char/tpm/tpm-interface.c  | 20 +++++--
 drivers/char/tpm/tpm.h            |  3 +-
 drivers/char/tpm/tpm2-space.c     |  5 +-
 drivers/char/tpm/tpm_tis_core.c   |  3 +-
 drivers/char/tpm/tpmrm-dev.c      | 20 ++++++-
 include/linux/tpm.h               |  4 +-
 10 files changed, 124 insertions(+), 65 deletions(-)

Signature · 1 line

2.51.0

I will put to queue with my tags but I just want to make first sure that we do not break anything.

I'll upgrade my test suite first to have TPM 1.2 tests (which is also needed for my own series) and run it in bunch of configurations. And on TPM2 I check the behavior with TSS2 daemon on / off.

Quoted 4 lines

I have no doubts on the code changes, and it is most importantly a security improvement, given that "who has the access and how long" can be deduced for a system configuration. I just feel that with this code change it is better to check and verify everything :-)

Roberto has already commented on this patch set, saying it would affect IMA[1]. I still need to look at the patch set, but please don't break IMA.

[1]https://lore.kernel.org/linux-integrity/[email protected]/

Signature · 3 lines

thanks,

Mimi

On Mon, Oct 20, 2025 at 01:53:30PM +0200, Roberto Sassu wrote:

Quoted 9 lines

Hi Jonathan

do I understand it correctly, that a process might open the TPM with O_EXCL, and this will prevent IMA from extending a PCR until that process closes the file descriptor?

If yes, this might be a concern, and I think an additional API to prevent such behavior would be needed (for example when IMA is active, i.e. there is a measurement policy loaded).

Also this would be a problem with hwrng.

This probably needs to be refined somehow. I don't have a solution at hand but "invariant" is that in-kernel caller should override user space exclusion, even when O_EXCL is used.

Quoted 1 line · Jarkko Sakkinen

On Mon, Oct 20, 2025 at 01:53:30PM +0200, Roberto Sassu wrote:

[...]

Quoted 10 lines

Quoted 9 lines

Hi Jonathan

do I understand it correctly, that a process might open the TPM with O_EXCL, and this will prevent IMA from extending a PCR until that process closes the file descriptor?

If yes, this might be a concern, and I think an additional API to prevent such behavior would be needed (for example when IMA is active, i.e. there is a measurement policy loaded).

Quoted 5 lines

Also this would be a problem with hwrng.

This probably needs to be refined somehow. I don't have a solution at hand but "invariant" is that in-kernel caller should override user space exclusion, even when O_EXCL is used.

Also, are we sure we need O_EXCL in the first place? A well functioning TPM is supposed to be able to cope with field upgrade while it receives other commands. When it's in this state, it's supposed to return TPM_RC_UPGRADE to inappropriate commands, so if we made sure we can correctly handle that in the kernel, that might be enough to get all this to work correctly without needing an exclusive open.

Of course, Field Upgrade is likely to be the least well tested of any TPM capability, so there's a good chance at least one TPM out there isn't going to behave as the standard says it should.

Regards,

James

Quoted 92 lines · Roberto Sassu

Quoted 78 lines · Jonathan McDowell

From: Jonathan McDowell <[email protected]>

There are situations where userspace might reasonably desire exclusive access to the TPM, or the kernel's internal context saving + flushing may cause issues, for example when performing firmware upgrades. Extend the locking already used for avoiding concurrent userspace access to prevent internal users of the TPM when /dev/tpm<n> is in use.

The few internal users who already hold the open_lock are changed to use tpm_internal_(try_get|put)_ops, with the old tpm_(try_get|put)_ops functions changing to obtain read access to the open_lock. We return -EBUSY when another user has exclusive access, rather than adding waits.

Signed-off-by Jonathan McDowell <[email protected]>

---
v2: Switch to _locked instead of _internal_ for function names.
v3: Move to end of patch series.

 drivers/char/tpm/tpm-chip.c       | 53 +++++++++++++++++++++++++------
 drivers/char/tpm/tpm-dev-common.c |  8 ++---
 drivers/char/tpm/tpm.h            |  2 ++
 drivers/char/tpm/tpm2-space.c     |  5 ++-
 4 files changed, 52 insertions(+), 16 deletions(-)

drivers/char/tpm/tpm-chip.c +14 -7

index ba906966721a..687f6d8cd601 100644
@@ -144,7 +144,7 @@ void tpm_chip_stop(struct tpm_chip *chip)
 EXPORT_SYMBOL_GPL(tpm_chip_stop);
 
 /**
- * tpm_try_get_ops() - Get a ref to the tpm_chip
+ * tpm_try_get_ops_locked() - Get a ref to the tpm_chip
  * @chip: Chip to ref
  *
  * The caller must already have some kind of locking to ensure that chip is

@@ -154,7 +154,7 @@ EXPORT_SYMBOL_GPL(tpm_chip_stop);
  *
  * Returns -ERRNO if the chip could not be got.
  */
-int tpm_try_get_ops(struct tpm_chip *chip)
+int tpm_try_get_ops_locked(struct tpm_chip *chip)
 {
 	int rc = -EIO;
 

@@ -185,22 +185,57 @@ int tpm_try_get_ops(struct tpm_chip *chip)
 	put_device(&chip->dev);
 	return rc;
 }
-EXPORT_SYMBOL_GPL(tpm_try_get_ops);
 
 /**
- * tpm_put_ops() - Release a ref to the tpm_chip
+ * tpm_put_ops_locked() - Release a ref to the tpm_chip
  * @chip: Chip to put
  *
- * This is the opposite pair to tpm_try_get_ops(). After this returns chip may
- * be kfree'd.
+ * This is the opposite pair to tpm_try_get_ops_locked(). After this returns
+ * chip may be kfree'd.
  */
-void tpm_put_ops(struct tpm_chip *chip)
+void tpm_put_ops_locked(struct tpm_chip *chip)
 {
 	tpm_chip_stop(chip);
 	mutex_unlock(&chip->tpm_mutex);
 	up_read(&chip->ops_sem);
 	put_device(&chip->dev);
 }
+
+/**
+ * tpm_try_get_ops() - Get a ref to the tpm_chip
+ * @chip: Chip to ref
+ *
+ * The caller must already have some kind of locking to ensure that chip is
+ * valid. This function will attempt to get the open_lock for the chip,
+ * ensuring no other user is expecting exclusive access, before locking the

drivers/char/tpm/tpm-chip.c int tpm_try_get_ops(struct tpm_chip *chip) · line 220

Quoted 10 lines

drivers/char/tpm/tpm-chip.c +10 -0

+ * chip so that the ops member can be accessed safely. The locking prevents
+ * tpm_chip_unregister from completing, so it should not be held for long
+ * periods.
+ *
+ * Returns -ERRNO if the chip could not be got.
+ */
+int tpm_try_get_ops(struct tpm_chip *chip)
+{
+	if (!down_read_trylock(&chip->open_lock))
+		return -EBUSY;

Hi Jonathan

Quoted 7 lines

do I understand it correctly, that a process might open the TPM with O_EXCL, and this will prevent IMA from extending a PCR until that process closes the file descriptor?

If yes, this might be a concern, and I think an additional API to prevent such behavior would be needed (for example when IMA is active, i.e. there is a measurement policy loaded).

Also this would be a problem with hwrng.

This probably needs to be refined somehow. I don't have a solution at hand but "invariant" is that in-kernel caller should override user space exclusion, even when O_EXCL is used.

Quoted 101 lines · Jarkko Sakkinen

Quoted 92 lines · Roberto Sassu

Quoted 78 lines · Jonathan McDowell

From: Jonathan McDowell <[email protected]>

There are situations where userspace might reasonably desire exclusive access to the TPM, or the kernel's internal context saving + flushing may cause issues, for example when performing firmware upgrades. Extend the locking already used for avoiding concurrent userspace access to prevent internal users of the TPM when /dev/tpm<n> is in use.

The few internal users who already hold the open_lock are changed to use tpm_internal_(try_get|put)_ops, with the old tpm_(try_get|put)_ops functions changing to obtain read access to the open_lock. We return -EBUSY when another user has exclusive access, rather than adding waits.

Signed-off-by Jonathan McDowell <[email protected]>

---
v2: Switch to _locked instead of _internal_ for function names.
v3: Move to end of patch series.

 drivers/char/tpm/tpm-chip.c       | 53 +++++++++++++++++++++++++------
 drivers/char/tpm/tpm-dev-common.c |  8 ++---
 drivers/char/tpm/tpm.h            |  2 ++
 drivers/char/tpm/tpm2-space.c     |  5 ++-
 4 files changed, 52 insertions(+), 16 deletions(-)

drivers/char/tpm/tpm-chip.c +14 -7

index ba906966721a..687f6d8cd601 100644
@@ -144,7 +144,7 @@ void tpm_chip_stop(struct tpm_chip *chip)
 EXPORT_SYMBOL_GPL(tpm_chip_stop);
 
 /**
- * tpm_try_get_ops() - Get a ref to the tpm_chip
+ * tpm_try_get_ops_locked() - Get a ref to the tpm_chip
  * @chip: Chip to ref
  *
  * The caller must already have some kind of locking to ensure that chip is

@@ -154,7 +154,7 @@ EXPORT_SYMBOL_GPL(tpm_chip_stop);
  *
  * Returns -ERRNO if the chip could not be got.
  */
-int tpm_try_get_ops(struct tpm_chip *chip)
+int tpm_try_get_ops_locked(struct tpm_chip *chip)
 {
 	int rc = -EIO;
 

@@ -185,22 +185,57 @@ int tpm_try_get_ops(struct tpm_chip *chip)
 	put_device(&chip->dev);
 	return rc;
 }
-EXPORT_SYMBOL_GPL(tpm_try_get_ops);
 
 /**
- * tpm_put_ops() - Release a ref to the tpm_chip
+ * tpm_put_ops_locked() - Release a ref to the tpm_chip
  * @chip: Chip to put
  *
- * This is the opposite pair to tpm_try_get_ops(). After this returns chip may
- * be kfree'd.
+ * This is the opposite pair to tpm_try_get_ops_locked(). After this returns
+ * chip may be kfree'd.
  */
-void tpm_put_ops(struct tpm_chip *chip)
+void tpm_put_ops_locked(struct tpm_chip *chip)
 {
 	tpm_chip_stop(chip);
 	mutex_unlock(&chip->tpm_mutex);
 	up_read(&chip->ops_sem);
 	put_device(&chip->dev);
 }
+
+/**
+ * tpm_try_get_ops() - Get a ref to the tpm_chip
+ * @chip: Chip to ref
+ *
+ * The caller must already have some kind of locking to ensure that chip is
+ * valid. This function will attempt to get the open_lock for the chip,
+ * ensuring no other user is expecting exclusive access, before locking the

drivers/char/tpm/tpm-chip.c int tpm_try_get_ops(struct tpm_chip *chip) · line 220

Quoted 10 lines

drivers/char/tpm/tpm-chip.c +10 -0

+ * chip so that the ops member can be accessed safely. The locking prevents
+ * tpm_chip_unregister from completing, so it should not be held for long
+ * periods.
+ *
+ * Returns -ERRNO if the chip could not be got.
+ */
+int tpm_try_get_ops(struct tpm_chip *chip)
+{
+	if (!down_read_trylock(&chip->open_lock))
+		return -EBUSY;

Hi Jonathan

Quoted 7 lines

do I understand it correctly, that a process might open the TPM with O_EXCL, and this will prevent IMA from extending a PCR until that process closes the file descriptor?

If yes, this might be a concern, and I think an additional API to prevent such behavior would be needed (for example when IMA is active, i.e. there is a measurement policy loaded).

Quoted 5 lines

Also this would be a problem with hwrng.

This probably needs to be refined somehow. I don't have a solution at hand but "invariant" is that in-kernel caller should override user space exclusion, even when O_EXCL is used.

Kernel access is exactly what caused the issue for me, in particular the HW RNG access during a firmware upgrade. My patch to be able to disable the HW RNG at runtime has landed in -next, which helps a lot, but it really would be nice to be able to say "Hands off, I'm busy with this", which is what led to this patch set.

To James' query about the fact the upgrade process should be properly handled, I think the issue is probably that the HMAC context saving around HW RNG access hit errors that were not gracefully handled, and we marked the TPM as disabled in tpm2_load_null, causing failure mid-upgrade.

J.

Signature · 1 line

What have you got in your pocket?