Skip to contentSkip to conversation

TPM: Fixup pcrs sysfs file

PATCH In review Jason Gunthorpe 4 participants 9 messages Sep 2, 2009 – Sep 14, 2009 TPM device driver

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 TPM: Fixup pcrs sysfs file

I'm testing the tpm_tis low level driver with a winbond WPCT200:

$ cat caps
Manufacturer: 0x57454300
TCG version: 1.2
Firmware version: 2.16

and noted that tpm_pcr_read for the pcrs sysfile file does not function. tpm_tis_recv returned with an error because the expected reply size was set to 14 (the request size) and the chip returned 30 bytes.

The TCG spec says the reply size is supposed to be 30 bytes.

First, the BUILD_BUG_ON was surely never correct, testing a run time value in big endian with that macro is just wrong. I belive the intended test was to ensure that the cmd buffer has enough space to store the reply.

Second, the length input to transmit_cmd is the size of the reply, not of the request. It should be 30 for READ_PCR.

With this change my chip reports all 23 pcrs.

---
 drivers/char/tpm/tpm.c |    4 ++--
 1 files changed, 2 insertions(+), 2 deletions(-)
drivers/char/tpm/tpm.c +2 -2
index a6b52d6..8ba0187 100644
@@ -696,8 +696,8 @@ int __tpm_pcr_read(struct tpm_chip *chip, int pcr_idx, u8 *res_buf)
 
 	cmd.header.in = pcrread_header;
 	cmd.params.pcrread_in.pcr_idx = cpu_to_be32(pcr_idx);
-	BUILD_BUG_ON(cmd.header.in.length > READ_PCR_RESULT_SIZE);
-	rc = transmit_cmd(chip, &cmd, cmd.header.in.length,
+	BUILD_BUG_ON(sizeof(cmd) < READ_PCR_RESULT_SIZE);
+	rc = transmit_cmd(chip, &cmd, READ_PCR_RESULT_SIZE,
 			  "attempting to read a pcr value");
 
 	if (rc == 0)
Signature · 1 line
1.5.4.2
#1 lore raw
Quoted 32 lines · On Tue, 1 Sep 2009 21:16:13 -0600 Jason Gunthorpe…

I'm testing the tpm_tis low level driver with a winbond WPCT200:

$ cat caps
Manufacturer: 0x57454300
TCG version: 1.2
Firmware version: 2.16

and noted that tpm_pcr_read for the pcrs sysfile file does not function. tpm_tis_recv returned with an error because the expected reply size was set to 14 (the request size) and the chip returned 30 bytes.

The TCG spec says the reply size is supposed to be 30 bytes.

First, the BUILD_BUG_ON was surely never correct, testing a run time value in big endian with that macro is just wrong. I belive the intended test was to ensure that the cmd buffer has enough space to store the reply.

Second, the length input to transmit_cmd is the size of the reply, not of the request. It should be 30 for READ_PCR.

With this change my chip reports all 23 pcrs.

---
 drivers/char/tpm/tpm.c |    4 ++--
 1 files changed, 2 insertions(+), 2 deletions(-)
drivers/char/tpm/tpm.c +0 -0
index a6b52d6..8ba0187 100644

drivers/char/tpm/tpm.c int __tpm_pcr_read(struct tpm_chip *chip, int pcr_idx, u8 *res_buf) · line 703

Quoted 10 lines
drivers/char/tpm/tpm.c +2 -2
@@ -696,8 +696,8 @@ int __tpm_pcr_read(struct tpm_chip *chip, int pcr_idx, u8 *res_buf)
 
 	cmd.header.in = pcrread_header;
 	cmd.params.pcrread_in.pcr_idx = cpu_to_be32(pcr_idx);
-	BUILD_BUG_ON(cmd.header.in.length > READ_PCR_RESULT_SIZE);
-	rc = transmit_cmd(chip, &cmd, cmd.header.in.length,
+	BUILD_BUG_ON(sizeof(cmd) < READ_PCR_RESULT_SIZE);
+	rc = transmit_cmd(chip, &cmd, READ_PCR_RESULT_SIZE,
 			  "attempting to read a pcr value");
 
 	if (rc == 0)

That sounds like a fairly serious bug, and this looks like a 2.6.31 patch.

Jan's build_bug_on-fix-it-and-a-couple-of-bogus-uses-of-it.patch (in -mm) simply removes the bogus BUILD_BUG_ON(). I think we might as well do that within the context of your patch.

So I end up with the below, which I propose for 2.6.31:

From: Jason Gunthorpe <[email protected]>

I'm testing the tpm_tis low level driver with a winbond WPCT200:

$ cat caps
Manufacturer: 0x57454300
TCG version: 1.2
Firmware version: 2.16

and noted that tpm_pcr_read for the pcrs sysfile file does not function. tpm_tis_recv returned with an error because the expected reply size was set to 14 (the request size) and the chip returned 30 bytes.

The TCG spec says the reply size is supposed to be 30 bytes.

The length input to transmit_cmd is the size of the reply, not of the request. It should be 30 for READ_PCR.

With this change my chip reports all 23 pcrs.

Also, the BUILD_BUG_ON() is just wrong - it's testing a value which isn't a compile-time constant. Simply remove that assertion.

Cc Debora Velarde <[email protected]>
Cc Rajiv Andrade <[email protected]>
Cc Marcel Selhorst <[email protected]>
Cc James Morris <[email protected]>

---

 drivers/char/tpm/tpm.c |    3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
drivers/char/tpm/tpm.c renamed from drivers/char/tpm/tpm.c~tpm-fixup-pcrs-sysfs-file +1 -2
diff -puN drivers/char/tpm/tpm.c~tpm-fixup-pcrs-sysfs-file drivers/char/tpm/tpm.c
@@ -696,8 +696,7 @@ int __tpm_pcr_read(struct tpm_chip *chip
 
 	cmd.header.in = pcrread_header;
 	cmd.params.pcrread_in.pcr_idx = cpu_to_be32(pcr_idx);
-	BUILD_BUG_ON(cmd.header.in.length > READ_PCR_RESULT_SIZE);
-	rc = transmit_cmd(chip, &cmd, cmd.header.in.length,
+	rc = transmit_cmd(chip, &cmd, READ_PCR_RESULT_SIZE,,
 			  "attempting to read a pcr value");
 
 	if (rc == 0)

_

#2 lore raw
Quoted 13 lines · Andrew Morton
Quoted 13 lines
drivers/char/tpm/tpm.c +3 -2
index a6b52d6..8ba0187 100644
+++ b/drivers/char/tpm/tpm.c
@@ -696,8 +696,8 @@ int __tpm_pcr_read(struct tpm_chip *chip, int pcr_idx, u8 *res_buf)
 
 	cmd.header.in = pcrread_header;
 	cmd.params.pcrread_in.pcr_idx = cpu_to_be32(pcr_idx);
-	BUILD_BUG_ON(cmd.header.in.length > READ_PCR_RESULT_SIZE);
-	rc = transmit_cmd(chip, &cmd, cmd.header.in.length,
+	BUILD_BUG_ON(sizeof(cmd) < READ_PCR_RESULT_SIZE);
+	rc = transmit_cmd(chip, &cmd, READ_PCR_RESULT_SIZE,
 			  "attempting to read a pcr value");
 
Quoted 4 lines

drivers/char/tpm/tpm.c int __tpm_pcr_read(struct tpm_chip *chip, int pcr_idx, u8 *res_buf) · line 703

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

That sounds like a fairly serious bug, and this looks like a 2.6.31 patch.

To be fair, I'm not sure the pcrs sysfile provides anything terribly usefull.. None of the sysfs files in this driver seem to follow the standard one-value-one-file convention either. But, if it is going to be included it may as well work properly...

Quoted 5 lines

Jan's build_bug_on-fix-it-and-a-couple-of-bogus-uses-of-it.patch (in -mm) simply removes the bogus BUILD_BUG_ON(). I think we might as well do that within the context of your patch.

So I end up with the below, which I propose for 2.6.31:

OK. That is fair. The tpm_cmd_params union contains a tpm_pcrread_out which should 'by design' ensure there is enough space.

Jan's removal of the 2nd BUILD_BUG_ON is also good.

But I notice tpm_pcr_extend also has a mis-use of the transmit_cmd idiom. This one functions ok because the in/out RPC message size happen to be the same. But lets fix it too?

Thanks,
Jason

Quoted 1 line

From 25da64a0927088c766745763728c6bcd973d0f4e Mon Sep 17 00:00:00 2001

From: Jason Gunthorpe <[email protected]>
Date: Tue, 1 Sep 2009 21:08:55 -0600
Subject: [PATCH] TPM: Fixup pcrs sysfs file

I'm testing the tpm_tis low level driver with a winbond WPCT200:

$ cat caps
Manufacturer: 0x57454300
TCG version: 1.2
Firmware version: 2.16

and noted that tpm_pcr_read for the pcrs sysfile file does not function. tpm_tis_recv returned with an error because the expected reply size was set to 14 (the request size) and the chip returned 30 bytes.

The TCG spec says the reply size for READ_PCR is supposed to be 30 bytes.

The length input to transmit_cmd is the size of the reply, not of the request.

With this change my chip reports all 23 pcrs.

Also fix tpm_pcr_extend to match the idiom of the rest of the code to prevent future confusion.

Finally, the BUILD_BUG_ON() is just wrong - it's testing a value which isn't a compile-time constant. Simply remove that assertion, the buffer is large enough by design.

---
 drivers/char/tpm/tpm.c |    8 +++-----
 1 files changed, 3 insertions(+), 5 deletions(-)
drivers/char/tpm/tpm.c +3 -5
index a6b52d6..5d5b324 100644
@@ -696,8 +696,7 @@ int __tpm_pcr_read(struct tpm_chip *chip, int pcr_idx, u8 *res_buf)
 
 	cmd.header.in = pcrread_header;
 	cmd.params.pcrread_in.pcr_idx = cpu_to_be32(pcr_idx);
-	BUILD_BUG_ON(cmd.header.in.length > READ_PCR_RESULT_SIZE);
-	rc = transmit_cmd(chip, &cmd, cmd.header.in.length,
+	rc = transmit_cmd(chip, &cmd, READ_PCR_RESULT_SIZE,
 			  "attempting to read a pcr value");
 
 	if (rc == 0)
@@ -742,7 +741,7 @@ EXPORT_SYMBOL_GPL(tpm_pcr_read);
  * the module usage count.
  */
 #define TPM_ORD_PCR_EXTEND cpu_to_be32(20)
-#define EXTEND_PCR_SIZE 34
+#define EXTEND_PCR_RESULT_SIZE 34
 static struct tpm_input_header pcrextend_header = {
 	.tag = TPM_TAG_RQU_COMMAND,
 	.length = cpu_to_be32(34),
@@ -760,10 +759,9 @@ int tpm_pcr_extend(u32 chip_num, int pcr_idx, const u8 *hash)
 		return -ENODEV;
 
 	cmd.header.in = pcrextend_header;
-	BUILD_BUG_ON(be32_to_cpu(cmd.header.in.length) > EXTEND_PCR_SIZE);
 	cmd.params.pcrextend_in.pcr_idx = cpu_to_be32(pcr_idx);
 	memcpy(cmd.params.pcrextend_in.hash, hash, TPM_DIGEST_SIZE);
-	rc = transmit_cmd(chip, &cmd, cmd.header.in.length,
+	rc = transmit_cmd(chip, &cmd, EXTEND_PCR_RESULT_SIZE,
 			  "attempting extend a PCR value");
 
 	module_put(chip->dev->driver->owner);
Signature · 1 line
1.5.4.2
#3 lore raw
Quoted 3 lines · Jason Gunthorpe
Quoted 3 lines

That sounds like a fairly serious bug, and this looks like a 2.6.31 patch.

Any comments from the maintainers on this patch?

Quoted 95 lines

To be fair, I'm not sure the pcrs sysfile provides anything terribly usefull.. None of the sysfs files in this driver seem to follow the standard one-value-one-file convention either. But, if it is going to be included it may as well work properly...

Quoted 5 lines

Jan's build_bug_on-fix-it-and-a-couple-of-bogus-uses-of-it.patch (in -mm) simply removes the bogus BUILD_BUG_ON(). I think we might as well do that within the context of your patch.

So I end up with the below, which I propose for 2.6.31:

OK. That is fair. The tpm_cmd_params union contains a tpm_pcrread_out which should 'by design' ensure there is enough space.

Jan's removal of the 2nd BUILD_BUG_ON is also good.

But I notice tpm_pcr_extend also has a mis-use of the transmit_cmd idiom. This one functions ok because the in/out RPC message size happen to be the same. But lets fix it too?

Thanks,
Jason

Quoted 1 line

From 25da64a0927088c766745763728c6bcd973d0f4e Mon Sep 17 00:00:00 2001

From: Jason Gunthorpe <[email protected]>
Date: Tue, 1 Sep 2009 21:08:55 -0600
Subject: [PATCH] TPM: Fixup pcrs sysfs file

I'm testing the tpm_tis low level driver with a winbond WPCT200:

$ cat caps
Manufacturer: 0x57454300
TCG version: 1.2
Firmware version: 2.16

and noted that tpm_pcr_read for the pcrs sysfile file does not function. tpm_tis_recv returned with an error because the expected reply size was set to 14 (the request size) and the chip returned 30 bytes.

The TCG spec says the reply size for READ_PCR is supposed to be 30 bytes.

The length input to transmit_cmd is the size of the reply, not of the request.

With this change my chip reports all 23 pcrs.

Also fix tpm_pcr_extend to match the idiom of the rest of the code to prevent future confusion.

Finally, the BUILD_BUG_ON() is just wrong - it's testing a value which isn't a compile-time constant. Simply remove that assertion, the buffer is large enough by design.

---
 drivers/char/tpm/tpm.c |    8 +++-----
 1 files changed, 3 insertions(+), 5 deletions(-)
drivers/char/tpm/tpm.c +3 -5
index a6b52d6..5d5b324 100644
@@ -696,8 +696,7 @@ int __tpm_pcr_read(struct tpm_chip *chip, int pcr_idx, u8 *res_buf)
 
 	cmd.header.in = pcrread_header;
 	cmd.params.pcrread_in.pcr_idx = cpu_to_be32(pcr_idx);
-	BUILD_BUG_ON(cmd.header.in.length > READ_PCR_RESULT_SIZE);
-	rc = transmit_cmd(chip, &cmd, cmd.header.in.length,
+	rc = transmit_cmd(chip, &cmd, READ_PCR_RESULT_SIZE,
 			  "attempting to read a pcr value");
 
 	if (rc == 0)
@@ -742,7 +741,7 @@ EXPORT_SYMBOL_GPL(tpm_pcr_read);
  * the module usage count.
  */
 #define TPM_ORD_PCR_EXTEND cpu_to_be32(20)
-#define EXTEND_PCR_SIZE 34
+#define EXTEND_PCR_RESULT_SIZE 34
 static struct tpm_input_header pcrextend_header = {
 	.tag = TPM_TAG_RQU_COMMAND,
 	.length = cpu_to_be32(34),
@@ -760,10 +759,9 @@ int tpm_pcr_extend(u32 chip_num, int pcr_idx, const u8 *hash)
 		return -ENODEV;
 
 	cmd.header.in = pcrextend_header;
-	BUILD_BUG_ON(be32_to_cpu(cmd.header.in.length) > EXTEND_PCR_SIZE);
 	cmd.params.pcrextend_in.pcr_idx = cpu_to_be32(pcr_idx);
 	memcpy(cmd.params.pcrextend_in.hash, hash, TPM_DIGEST_SIZE);
-	rc = transmit_cmd(chip, &cmd, cmd.header.in.length,
+	rc = transmit_cmd(chip, &cmd, EXTEND_PCR_RESULT_SIZE,
 			  "attempting extend a PCR value");
 
 	module_put(chip->dev->driver->owner);
Signature · 1 line
1.5.4.2
Signature · 2 lines
James Morris
<[email protected]>
#4 lore raw
Quoted 6 lines · James Morris
Quoted 2 lines · Jason Gunthorpe
Quoted 2 lines

That sounds like a fairly serious bug, and this looks like a 2.6.31 patch.

Any comments from the maintainers on this patch?

FWIW, I didn't mention in the patch emails, but all the patches I sent fix regressions that have been introduced in the past couple years by clean up patches that never tested the code paths they alter.

What becomes of patches that end up in the mm tree? Do they still route through you for mainline inclusion?

Thanks for looking at this.

Jason

#5 lore raw
Quoted 6 lines · On Mon, 14 Sep 2009 00:25:20 -0600 Jason Gunthorpe…
Quoted 5 lines · James Morris
Quoted 2 lines · Jason Gunthorpe
Quoted 2 lines

That sounds like a fairly serious bug, and this looks like a 2.6.31 patch.

Quoted 8 lines
Quoted 1 line

Any comments from the maintainers on this patch?

FWIW, I didn't mention in the patch emails, but all the patches I sent fix regressions that have been introduced in the past couple years by clean up patches that never tested the code paths they alter.

What becomes of patches that end up in the mm tree? Do they still route through you for mainline inclusion?

I'll be sending tpm patches to James henceforth. I presently have

tpm-fixup-pcrs-sysfs-file.patch
tpm-fixup-pcrs-sysfs-file-update.patch
tpm-fix-up-pubek-sysfs-file.patch

#6 lore raw
In reply to Jason Gunthorpe
Quoted 8 lines · Jason Gunthorpe
Quoted 6 lines · James Morris
Quoted 2 lines · Jason Gunthorpe
Quoted 2 lines

That sounds like a fairly serious bug, and this looks like a 2.6.31 patch.

Any comments from the maintainers on this patch?

Sorry for the delay, just reviewed the patches and Jason is right, Jason's and Andrew's final work (last patch) is great.

Quoted 4 lines

FWIW, I didn't mention in the patch emails, but all the patches I sent fix regressions that have been introduced in the past couple years by clean up patches that never tested the code paths they alter.

Hold on, I tested these functions before when submitting them and they worked fine, and, I couldn't figure it out why yet (only guesses), they still work for me here, probably due the mishandling of the be32 formatted value when calling transmit_cmd().

Thanks,
Rajiv

#7 lore raw
Quoted 18 lines · Rajiv Andrade
Quoted 8 lines · Jason Gunthorpe
Quoted 6 lines · James Morris
Quoted 2 lines · Jason Gunthorpe
Quoted 2 lines

That sounds like a fairly serious bug, and this looks like a 2.6.31 patch.

Any comments from the maintainers on this patch?

Sorry for the delay, just reviewed the patches and Jason is right, Jason's and Andrew's final work (last patch) is great.

Quoted 4 lines

FWIW, I didn't mention in the patch emails, but all the patches I sent fix regressions that have been introduced in the past couple years by clean up patches that never tested the code paths they alter.

Quoted 4 lines

Hold on, I tested these functions before when submitting them and they worked fine, and, I couldn't figure it out why yet (only guesses), they still work for me here, probably due the mishandling of the be32 formatted value when calling transmit_cmd().

Regardless the testing and what's happening here in my machine, I agree with you both that these fixes should go to 2.6.31.

Rajiv

#8 lore raw
In reply to Rajiv Andrade
Quoted 2 lines · Rajiv Andrade

Sorry for the delay, just reviewed the patches and Jason is right, Jason's and Andrew's final work (last patch) is great.

Thnx

Quoted 8 lines
Quoted 3 lines

FWIW, I didn't mention in the patch emails, but all the patches I sent fix regressions that have been introduced in the past couple years by clean up patches that never tested the code paths they alter.

Hold on, I tested these functions before when submitting them and they worked fine, and, I couldn't figure it out why yet (only guesses), they still work for me here, probably due the mishandling of the be32 formatted value when calling transmit_cmd().

Ahh, yes, you are right, the pcr thing wouldn't blow up on a LE system, I'm testing on BE. Didn't notice that subtlety. Sorry

Also please look at the last patch Andrew has queued up:
http://patchwork.kernel.org/patch/45512/

Regards,
Jason