Skip to contentSkip to conversation

char: tpm: Protect tpm_pm_suspend with locks

PATCH Applied Jason A. Donenfeld 5 participants 11 messages Nov 28, 2022 – Dec 10, 2022

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH v3 1/1 char: tpm: Protect tpm_pm_suspend with locks

From: Jan Dabros <[email protected]>

Currently tpm transactions are executed unconditionally in tpm_pm_suspend() function, which may lead to races with other tpm accessors in the system. Specifically, the hw_random tpm driver makes use of tpm_get_random(), and this function is called in a loop from a kthread, which means it's not frozen alongside userspace, and so can race with the work done during system suspend:

[    3.277834] tpm tpm0: tpm_transmit: tpm_recv: error -52
[    3.278437] tpm tpm0: invalid TPM_STS.x 0xff, dumping stack for forensics
[    3.278445] CPU: 0 PID: 1 Comm: init Not tainted 6.1.0-rc5+ #135
[    3.278450] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.16.0-20220807_005459-localhost 04/01/2014
[    3.278453] Call Trace:
[    3.278458]  <TASK>
[    3.278460]  dump_stack_lvl+0x34/0x44
[    3.278471]  tpm_tis_status.cold+0x19/0x20
[    3.278479]  tpm_transmit+0x13b/0x390
[    3.278489]  tpm_transmit_cmd+0x20/0x80
[    3.278496]  tpm1_pm_suspend+0xa6/0x110
[    3.278503]  tpm_pm_suspend+0x53/0x80
[    3.278510]  __pnp_bus_suspend+0x35/0xe0
[    3.278515]  ? pnp_bus_freeze+0x10/0x10
[    3.278519]  __device_suspend+0x10f/0x350

Fix this by calling tpm_try_get_ops(), which itself is a wrapper around tpm_chip_start(), but takes the appropriate mutex.

Fixes e891db1a18bf ("tpm: turn on TPM on suspend for TPM 1.x")
[Jason: reworked commit message, added metadata]
---
 drivers/char/tpm/tpm-interface.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
drivers/char/tpm/tpm-interface.c +3 -2
index 1621ce818705..d69905233aff 100644
@@ -401,13 +401,14 @@ int tpm_pm_suspend(struct device *dev)
 	    !pm_suspend_via_firmware())
 		goto suspended;
 
-	if (!tpm_chip_start(chip)) {
+	rc = tpm_try_get_ops(chip);
+	if (!rc) {
 		if (chip->flags & TPM_CHIP_FLAG_TPM2)
 			tpm2_shutdown(chip, TPM2_SU_STATE);
 		else
 			rc = tpm1_pm_suspend(chip, tpm_suspend_pcr);
 
-		tpm_chip_stop(chip);
+		tpm_put_ops(chip);
 	}
 
 suspended:
Signature · 1 line
2.38.1
#1 lore raw

Hi, this is your Linux kernel regression tracker.

On 28.11.22 20:56, Jason A. Donenfeld wrote:

BTW, many thx for taking care of this Jason!

Quoted 8 lines

From: Jan Dabros <[email protected]>

Currently tpm transactions are executed unconditionally in tpm_pm_suspend() function, which may lead to races with other tpm accessors in the system. Specifically, the hw_random tpm driver makes use of tpm_get_random(), and this function is called in a loop from a kthread, which means it's not frozen alongside userspace, and so can race with the work done during system suspend:

Peter, Jarkko, did you look at this patch or even applied it already to send it to Linus soon? Doesn't look like it from here, but maybe I missed something.

Thing is: the linked regression afaics is overdue fixing (for details see "Prioritize work on fixing regressions" in
https://www.kernel.org/doc/html/latest/process/handling-regressions.html
). Hence if this doesn't make any progress I'll likely have to point Linus to this patch and suggest to apply it directly if it looks okay from his perspective.

Ciao, Thorsten (wearing his 'the Linux kernel's regression tracker' hat)

P.S.: As the Linux kernel's regression tracker I deal with a lot of reports and sometimes miss something important when writing mails like this. If that's the case here, don't hesitate to tell me in a public reply, it's in everyone's interest to set the public record straight.

Quoted 53 lines
[    3.277834] tpm tpm0: tpm_transmit: tpm_recv: error -52
[    3.278437] tpm tpm0: invalid TPM_STS.x 0xff, dumping stack for forensics
[    3.278445] CPU: 0 PID: 1 Comm: init Not tainted 6.1.0-rc5+ #135
[    3.278450] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.16.0-20220807_005459-localhost 04/01/2014
[    3.278453] Call Trace:
[    3.278458]  <TASK>
[    3.278460]  dump_stack_lvl+0x34/0x44
[    3.278471]  tpm_tis_status.cold+0x19/0x20
[    3.278479]  tpm_transmit+0x13b/0x390
[    3.278489]  tpm_transmit_cmd+0x20/0x80
[    3.278496]  tpm1_pm_suspend+0xa6/0x110
[    3.278503]  tpm_pm_suspend+0x53/0x80
[    3.278510]  __pnp_bus_suspend+0x35/0xe0
[    3.278515]  ? pnp_bus_freeze+0x10/0x10
[    3.278519]  __device_suspend+0x10f/0x350

Fix this by calling tpm_try_get_ops(), which itself is a wrapper around tpm_chip_start(), but takes the appropriate mutex.

Fixes e891db1a18bf ("tpm: turn on TPM on suspend for TPM 1.x")
[Jason: reworked commit message, added metadata]
---
 drivers/char/tpm/tpm-interface.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
drivers/char/tpm/tpm-interface.c +3 -2
index 1621ce818705..d69905233aff 100644
@@ -401,13 +401,14 @@ int tpm_pm_suspend(struct device *dev)
 	    !pm_suspend_via_firmware())
 		goto suspended;
 
-	if (!tpm_chip_start(chip)) {
+	rc = tpm_try_get_ops(chip);
+	if (!rc) {
 		if (chip->flags & TPM_CHIP_FLAG_TPM2)
 			tpm2_shutdown(chip, TPM2_SU_STATE);
 		else
 			rc = tpm1_pm_suspend(chip, tpm_suspend_pcr);
 
-		tpm_chip_stop(chip);
+		tpm_put_ops(chip);
 	}
 
 suspended:
#2 lore raw

Thanks for handling this, Thorsten. I had poked Jarkko about this earlier this week, but he didn't respond. So I'm glad you're on the case now getting this in somewhere. Probably this should make it to rc8, so there's still one week left of testing it.

Jason

#3 lore raw

Hi Jason,

Thanks for taking care of this!

Re 2/3 and 3/3 as you mentioned earlier, will get back to this when I have some bandwidth and send it separately.

Best Regards,
Jan

pt., 2 gru 2022 o 10:46 Jason A. Donenfeld <[email protected]> napisał(a):

Quoted 7 lines

Thanks for handling this, Thorsten. I had poked Jarkko about this earlier this week, but he didn't respond. So I'm glad you're on the case now getting this in somewhere. Probably this should make it to rc8, so there's still one week left of testing it.

Jason

#4 lore raw
In reply to Thorsten Leemhuis

Hi Thorsten / Linus,

Quoted 19 lines · Thorsten Leemhuis

Hi, this is your Linux kernel regression tracker.

On 28.11.22 20:56, Jason A. Donenfeld wrote:

BTW, many thx for taking care of this Jason!

Quoted 8 lines

From: Jan Dabros <[email protected]>

Currently tpm transactions are executed unconditionally in tpm_pm_suspend() function, which may lead to races with other tpm accessors in the system. Specifically, the hw_random tpm driver makes use of tpm_get_random(), and this function is called in a loop from a kthread, which means it's not frozen alongside userspace, and so can race with the work done during system suspend:

Peter, Jarkko, did you look at this patch or even applied it already to send it to Linus soon? Doesn't look like it from here, but maybe I missed something.

Quoted 6 lines

Thing is: the linked regression afaics is overdue fixing (for details see "Prioritize work on fixing regressions" in
https://www.kernel.org/doc/html/latest/process/handling-regressions.html
). Hence if this doesn't make any progress I'll likely have to point Linus to this patch and suggest to apply it directly if it looks okay from his perspective.

I'm very concerned about this. Jan posted the original fix a month ago, and then it fizzled out. Then I got word of the bug last week and revived the fix [1], while also figuring out how to reproduce it together with the reporter. I emailed the tpm maintainers offlist to poke them, and nobody woke up. And tomorrow is rc8 day. Given that this patch is pretty simple, has been tested to fix an annoying regression, and that neither of the three maintainers has popped up this week to get things rolling, I think we should just commit this now anyway, to make sure it gets in for rc8. This way there's still a solid week of testing. I'm in general not a big fan of the "nuclear option" of not waiting for out to lunch maintainers, but given that it is now December 3, it seems like the right decision.

[1] https://lore.kernel.org/all/[email protected]/

Jason

maintainer Reviewed-by on v3 1/1
#5 lore raw
In reply to Jason A. Donenfeld
Quoted 65 lines · Jason A. Donenfeld

From: Jan Dabros <[email protected]>

Currently tpm transactions are executed unconditionally in tpm_pm_suspend() function, which may lead to races with other tpm accessors in the system. Specifically, the hw_random tpm driver makes use of tpm_get_random(), and this function is called in a loop from a kthread, which means it's not frozen alongside userspace, and so can race with the work done during system suspend:

[    3.277834] tpm tpm0: tpm_transmit: tpm_recv: error -52
[    3.278437] tpm tpm0: invalid TPM_STS.x 0xff, dumping stack for forensics
[    3.278445] CPU: 0 PID: 1 Comm: init Not tainted 6.1.0-rc5+ #135
[    3.278450] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.16.0-20220807_005459-localhost 04/01/2014
[    3.278453] Call Trace:
[    3.278458]  <TASK>
[    3.278460]  dump_stack_lvl+0x34/0x44
[    3.278471]  tpm_tis_status.cold+0x19/0x20
[    3.278479]  tpm_transmit+0x13b/0x390
[    3.278489]  tpm_transmit_cmd+0x20/0x80
[    3.278496]  tpm1_pm_suspend+0xa6/0x110
[    3.278503]  tpm_pm_suspend+0x53/0x80
[    3.278510]  __pnp_bus_suspend+0x35/0xe0
[    3.278515]  ? pnp_bus_freeze+0x10/0x10
[    3.278519]  __device_suspend+0x10f/0x350

Fix this by calling tpm_try_get_ops(), which itself is a wrapper around tpm_chip_start(), but takes the appropriate mutex.

Fixes e891db1a18bf ("tpm: turn on TPM on suspend for TPM 1.x")
[Jason: reworked commit message, added metadata]
---
 drivers/char/tpm/tpm-interface.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
drivers/char/tpm/tpm-interface.c +3 -2
index 1621ce818705..d69905233aff 100644
@@ -401,13 +401,14 @@ int tpm_pm_suspend(struct device *dev)
 	    !pm_suspend_via_firmware())
 		goto suspended;
 
-	if (!tpm_chip_start(chip)) {
+	rc = tpm_try_get_ops(chip);
+	if (!rc) {
 		if (chip->flags & TPM_CHIP_FLAG_TPM2)
 			tpm2_shutdown(chip, TPM2_SU_STATE);
 		else
 			rc = tpm1_pm_suspend(chip, tpm_suspend_pcr);
 
-		tpm_chip_stop(chip);
+		tpm_put_ops(chip);
 	}
 
 suspended:
Signature · 1 line
2.38.1

Hi, sorry for the latency.

BR, Jarkko

maintainer Applied
#6 lore raw
Quoted 62 lines · Jarkko Sakkinen
Quoted 61 lines · Jason A. Donenfeld

From: Jan Dabros <[email protected]>

Currently tpm transactions are executed unconditionally in tpm_pm_suspend() function, which may lead to races with other tpm accessors in the system. Specifically, the hw_random tpm driver makes use of tpm_get_random(), and this function is called in a loop from a kthread, which means it's not frozen alongside userspace, and so can race with the work done during system suspend:

[    3.277834] tpm tpm0: tpm_transmit: tpm_recv: error -52
[    3.278437] tpm tpm0: invalid TPM_STS.x 0xff, dumping stack for forensics
[    3.278445] CPU: 0 PID: 1 Comm: init Not tainted 6.1.0-rc5+ #135
[    3.278450] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.16.0-20220807_005459-localhost 04/01/2014
[    3.278453] Call Trace:
[    3.278458]  <TASK>
[    3.278460]  dump_stack_lvl+0x34/0x44
[    3.278471]  tpm_tis_status.cold+0x19/0x20
[    3.278479]  tpm_transmit+0x13b/0x390
[    3.278489]  tpm_transmit_cmd+0x20/0x80
[    3.278496]  tpm1_pm_suspend+0xa6/0x110
[    3.278503]  tpm_pm_suspend+0x53/0x80
[    3.278510]  __pnp_bus_suspend+0x35/0xe0
[    3.278515]  ? pnp_bus_freeze+0x10/0x10
[    3.278519]  __device_suspend+0x10f/0x350

Fix this by calling tpm_try_get_ops(), which itself is a wrapper around tpm_chip_start(), but takes the appropriate mutex.

Fixes e891db1a18bf ("tpm: turn on TPM on suspend for TPM 1.x")
[Jason: reworked commit message, added metadata]
---
 drivers/char/tpm/tpm-interface.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
drivers/char/tpm/tpm-interface.c +3 -2
index 1621ce818705..d69905233aff 100644
@@ -401,13 +401,14 @@ int tpm_pm_suspend(struct device *dev)
 	    !pm_suspend_via_firmware())
 		goto suspended;
 
-	if (!tpm_chip_start(chip)) {
+	rc = tpm_try_get_ops(chip);
+	if (!rc) {
 		if (chip->flags & TPM_CHIP_FLAG_TPM2)
 			tpm2_shutdown(chip, TPM2_SU_STATE);
 		else
 			rc = tpm1_pm_suspend(chip, tpm_suspend_pcr);
 
-		tpm_chip_stop(chip);
+		tpm_put_ops(chip);
 	}
 
Quoted 8 lines
Quoted 4 lines
drivers/char/tpm/tpm-interface.c +0 -0
 suspended:
Signature · 1 line
2.38.1

Hi, sorry for the latency.

Applied to git://git.kernel.org/pub/scm/linux/kernel/git/jarkko/linux-tpmdd.git

BR, Jarkko

#7 lore raw
Quoted 70 lines · Jarkko Sakkinen
Quoted 69 lines · Jarkko Sakkinen
Quoted 65 lines · Jason A. Donenfeld

From: Jan Dabros <[email protected]>

Currently tpm transactions are executed unconditionally in tpm_pm_suspend() function, which may lead to races with other tpm accessors in the system. Specifically, the hw_random tpm driver makes use of tpm_get_random(), and this function is called in a loop from a kthread, which means it's not frozen alongside userspace, and so can race with the work done during system suspend:

[    3.277834] tpm tpm0: tpm_transmit: tpm_recv: error -52
[    3.278437] tpm tpm0: invalid TPM_STS.x 0xff, dumping stack for forensics
[    3.278445] CPU: 0 PID: 1 Comm: init Not tainted 6.1.0-rc5+ #135
[    3.278450] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.16.0-20220807_005459-localhost 04/01/2014
[    3.278453] Call Trace:
[    3.278458]  <TASK>
[    3.278460]  dump_stack_lvl+0x34/0x44
[    3.278471]  tpm_tis_status.cold+0x19/0x20
[    3.278479]  tpm_transmit+0x13b/0x390
[    3.278489]  tpm_transmit_cmd+0x20/0x80
[    3.278496]  tpm1_pm_suspend+0xa6/0x110
[    3.278503]  tpm_pm_suspend+0x53/0x80
[    3.278510]  __pnp_bus_suspend+0x35/0xe0
[    3.278515]  ? pnp_bus_freeze+0x10/0x10
[    3.278519]  __device_suspend+0x10f/0x350

Fix this by calling tpm_try_get_ops(), which itself is a wrapper around tpm_chip_start(), but takes the appropriate mutex.

Fixes e891db1a18bf ("tpm: turn on TPM on suspend for TPM 1.x")
[Jason: reworked commit message, added metadata]
---
 drivers/char/tpm/tpm-interface.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
drivers/char/tpm/tpm-interface.c +3 -2
index 1621ce818705..d69905233aff 100644
@@ -401,13 +401,14 @@ int tpm_pm_suspend(struct device *dev)
 	    !pm_suspend_via_firmware())
 		goto suspended;
 
-	if (!tpm_chip_start(chip)) {
+	rc = tpm_try_get_ops(chip);
+	if (!rc) {
 		if (chip->flags & TPM_CHIP_FLAG_TPM2)
 			tpm2_shutdown(chip, TPM2_SU_STATE);
 		else
 			rc = tpm1_pm_suspend(chip, tpm_suspend_pcr);
 
-		tpm_chip_stop(chip);
+		tpm_put_ops(chip);
 	}
 
 suspended:
Signature · 1 line
2.38.1

Hi, sorry for the latency.

Quoted 3 lines
Quoted 1 line

Applied to git://git.kernel.org/pub/scm/linux/kernel/git/jarkko/linux-tpmdd.git

Oh thank goodness. You'll send this in for rc8 today?

Jason

maintainer
#8 lore raw
Quoted 71 lines · Jason A. Donenfeld
Quoted 70 lines · Jarkko Sakkinen
Quoted 69 lines · Jarkko Sakkinen
Quoted 65 lines · Jason A. Donenfeld

From: Jan Dabros <[email protected]>

Currently tpm transactions are executed unconditionally in tpm_pm_suspend() function, which may lead to races with other tpm accessors in the system. Specifically, the hw_random tpm driver makes use of tpm_get_random(), and this function is called in a loop from a kthread, which means it's not frozen alongside userspace, and so can race with the work done during system suspend:

[    3.277834] tpm tpm0: tpm_transmit: tpm_recv: error -52
[    3.278437] tpm tpm0: invalid TPM_STS.x 0xff, dumping stack for forensics
[    3.278445] CPU: 0 PID: 1 Comm: init Not tainted 6.1.0-rc5+ #135
[    3.278450] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.16.0-20220807_005459-localhost 04/01/2014
[    3.278453] Call Trace:
[    3.278458]  <TASK>
[    3.278460]  dump_stack_lvl+0x34/0x44
[    3.278471]  tpm_tis_status.cold+0x19/0x20
[    3.278479]  tpm_transmit+0x13b/0x390
[    3.278489]  tpm_transmit_cmd+0x20/0x80
[    3.278496]  tpm1_pm_suspend+0xa6/0x110
[    3.278503]  tpm_pm_suspend+0x53/0x80
[    3.278510]  __pnp_bus_suspend+0x35/0xe0
[    3.278515]  ? pnp_bus_freeze+0x10/0x10
[    3.278519]  __device_suspend+0x10f/0x350

Fix this by calling tpm_try_get_ops(), which itself is a wrapper around tpm_chip_start(), but takes the appropriate mutex.

Fixes e891db1a18bf ("tpm: turn on TPM on suspend for TPM 1.x")
[Jason: reworked commit message, added metadata]
---
 drivers/char/tpm/tpm-interface.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
drivers/char/tpm/tpm-interface.c +3 -2
index 1621ce818705..d69905233aff 100644
@@ -401,13 +401,14 @@ int tpm_pm_suspend(struct device *dev)
 	    !pm_suspend_via_firmware())
 		goto suspended;
 
-	if (!tpm_chip_start(chip)) {
+	rc = tpm_try_get_ops(chip);
+	if (!rc) {
 		if (chip->flags & TPM_CHIP_FLAG_TPM2)
 			tpm2_shutdown(chip, TPM2_SU_STATE);
 		else
 			rc = tpm1_pm_suspend(chip, tpm_suspend_pcr);
 
-		tpm_chip_stop(chip);
+		tpm_put_ops(chip);
 	}
 
 suspended:
Signature · 1 line
2.38.1

Hi, sorry for the latency.

Quoted 5 lines
Quoted 3 lines
Quoted 1 line

Applied to git://git.kernel.org/pub/scm/linux/kernel/git/jarkko/linux-tpmdd.git

Oh thank goodness. You'll send this in for rc8 today?

for 6.2-rc1

BR, Jarkko

maintainer
#10 lore raw
Quoted 11 lines · Vlastimil Babka
Quoted 7 lines · Jarkko Sakkinen
Quoted 4 lines · Jason A. Donenfeld
Quoted 2 lines

Applied to git://git.kernel.org/pub/scm/linux/kernel/git/jarkko/linux-tpmdd.git

Oh thank goodness. You'll send this in for rc8 today?

for 6.2-rc1

Linus took it directly to rc8, so it would conflict now.
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?h=v6.1-rc8&id=23393c6461422df5bf8084a086ada9a7e17dc2ba

Thanks for the remark!

BR, Jarkko