U-Boot ECDSA Implementation Question
Hello,
I'm a current intern at Microsoft, and one of my priorities is to enable ECDSA for U-Boot image signing/verification. Simon mentioned someone is already working on ECC, it would be great to get synced up with related progress. For signing, I will likely replicate the existing approach of using the openssl library. I'm aware that signing happens on a host machine and verification happens during boot, which implies verification should have a custom implementation to avoid the openssl overhead in the U-Boot binary. My thoughts are to copy an ECC verification implementation from a well-tested widely-used open source project. I was wondering, is U-Boot's current RSA verification copied from another project? If so, how are security patches between the two copies of code usually handled? I'm thinking of deriving from the ECDSA implementation currently in the Linux kernel, though I'd also appreciate suggestions if there's a better/more widely tested & used implementation.
All the best, Tim
Hi Tim
+Alexandru Gagniuc
On Thu, 4 Feb 2021 at 15:01, Tim Romanski t-tromanski@microsoft.com wrote:
Hello,
I’m a current intern at Microsoft, and one of my priorities is to enable ECDSA for U-Boot image signing/verification. Simon mentioned someone is already working on ECC, it would be great to get synced up with related progress. For signing, I will likely replicate the existing approach of using the openssl library. I’m aware that signing happens on a host machine and verification happens during boot, which implies verification should have a custom implementation to avoid the openssl overhead in the U-Boot binary. My thoughts are to copy an ECC verification implementation from a well-tested widely-used open source project. I was wondering, is U-Boot’s current RSA verification copied from another project? If so, how are security patches between the two copies of code usually handled? I’m thinking of deriving from the ECDSA implementation currently in the Linux kernel, though I’d also appreciate suggestions if there’s a better/more widely tested & used implementation.
U-Boot's RSA came originally from Android I think and was modified for use in Chrome OS. However the implementation in U-Boot of the verification part is quite small - mostly in rsa-verify.c with some maths in rsa-mod-exp.c and U-Boot has added various new features over the years. We don't synchronous security patches formally although of course they are published. I think pulling in something from Linux makes sense if it is not too large, as the projects are fairly close in coding style, contributors, etc.
Alexandru Gagniuc, on cc, has been looking at implementing the signing side of this recently and has sent some patches that you could look at.
I hope you have a nice internship!
Regards, Simon
Hi Tim,
On 2/5/21 8:35 AM, Simon Glass wrote:
I’m a current intern at Microsoft, and one of my priorities is to enable ECDSA for U-Boot image signing/verification. Simon mentioned someone is already working on ECC, it would be great to get synced up with related progress. For signing, I will likely replicate the existing approach of using the openssl library. I’m aware that signing happens on a host machine and verification happens during boot, which implies verification should have a custom implementation to avoid the openssl overhead in the U-Boot binary. My thoughts are to copy an ECC verification implementation from a well-tested widely-used open source project. I was wondering, is U-Boot’s current RSA verification copied from another project? If so, how are security patches between the two copies of code usually handled? I’m thinking of deriving from the ECDSA implementation currently in the Linux kernel, though I’d also appreciate suggestions if there’s a better/more widely tested & used implementation.
[snip]
Alexandru Gagniuc, on cc, has been looking at implementing the signing side of this recently and has sent some patches that you could look at.
I hope I can save you some effort on the signing side. Generally, you have two types of signed images. The first is the signed bootloader (BL2 or FSBL in ARM terms). The other one is the signed Flattened Image Tree (FIT) that we use in u-boot. The first one is vendor-specific, so you'd usually use vendor tools or write your own. We use mkimage to deal with the latter.
I've implemented the signing part [1] for mkimage. mkimage has the ability to use hardware signing via the PKCS11 engine of openssl, which I did not implement. You can read more about it here [3].
The verification part is still being defined [4][5]. The idea is to define a UCLASS which abstracts the underlying implementation. For RSA, it's defined here [6].
My goal with ECDSA verification was to use the ROM API of the STM32MP1. This meant I don't have to write a software implementation of ECDSA. This would be useful in two ways. It would enable ECDSA verification on devices that don't support it in hardware, and would also allow us to add some unit tests for ECDSA.
I suspect what you could do from here, is try to build my branch with ECDSA signing, play around with mkimage, and let us know how we can point you to the correct documentation. There's a lot of it in doc/, but it's not always easy to find.
Alex
[1] https://github.com/mrnuke/u-boot/commits/patch-mkimage-keyfile-v1 [2] https://github.com/mrnuke/u-boot/commit/a2ae016f2f80579962d4ab058137c8e1a4f4... [3] https://github.com/mrnuke/u-boot/blob/3f447efcf8ad98d0eea349994810a66b453ac1... [4] https://github.com/mrnuke/u-boot/commit/31caceb0e28959881e96ea49a0b28fd44d13... [5] https://github.com/mrnuke/u-boot/commits/patch-stm32-ecdsa-v1 [6] https://github.com/mrnuke/u-boot/blob/7d7ce8d70287568071a5d24acb6dc74b923fe7...
Hi Alex,
Thanks for the context! What are your plans for upstreaming your ECDSA signing implementation? I've currently dedicated the next four weeks to getting signing+verification implemented, so if you'd like a helping hand either with any leftover signing work or to get verification started I'm happy to collaborate.
All the best, Tim
-----Original Message----- From: Alex G. mr.nuke.me@gmail.com Sent: February 5, 2021 11:09 AM To: Simon Glass sjg@chromium.org; Tim Romanski t-tromanski@microsoft.com Cc: u-boot@lists.denx.de; Deskin Miller deskinm@microsoft.com; Dylan D'Silva ddsilva@microsoft.com Subject: [EXTERNAL] Re: U-Boot ECDSA Implementation Question
Hi Tim,
On 2/5/21 8:35 AM, Simon Glass wrote:
I'm a current intern at Microsoft, and one of my priorities is to enable ECDSA for U-Boot image signing/verification. Simon mentioned someone is already working on ECC, it would be great to get synced up with related progress. For signing, I will likely replicate the existing approach of using the openssl library. I'm aware that signing happens on a host machine and verification happens during boot, which implies verification should have a custom implementation to avoid the openssl overhead in the U-Boot binary. My thoughts are to copy an ECC verification implementation from a well-tested widely-used open source project. I was wondering, is U-Boot's current RSA verification copied from another project? If so, how are security patches between the two copies of code usually handled? I'm thinking of deriving from the ECDSA implementation currently in the Linux kernel, though I'd also appreciate suggestions if there's a better/more widely tested & used implementation.
[snip]
Alexandru Gagniuc, on cc, has been looking at implementing the signing side of this recently and has sent some patches that you could look at.
I hope I can save you some effort on the signing side. Generally, you have two types of signed images. The first is the signed bootloader (BL2 or FSBL in ARM terms). The other one is the signed Flattened Image Tree (FIT) that we use in u-boot. The first one is vendor-specific, so you'd usually use vendor tools or write your own. We use mkimage to deal with the latter.
I've implemented the signing part [1] for mkimage. mkimage has the ability to use hardware signing via the PKCS11 engine of openssl, which I did not implement. You can read more about it here [3].
The verification part is still being defined [4][5]. The idea is to define a UCLASS which abstracts the underlying implementation. For RSA, it's defined here [6].
My goal with ECDSA verification was to use the ROM API of the STM32MP1. This meant I don't have to write a software implementation of ECDSA. This would be useful in two ways. It would enable ECDSA verification on devices that don't support it in hardware, and would also allow us to add some unit tests for ECDSA.
I suspect what you could do from here, is try to build my branch with ECDSA signing, play around with mkimage, and let us know how we can point you to the correct documentation. There's a lot of it in doc/, but it's not always easy to find.
Alex
[1] https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com... [2] https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com... [3] https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com... [4] https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com... [5] https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com... [6] https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com...
On 2/10/21 1:45 PM, Tim Romanski wrote:
Hi Alex,
Thanks for the context! What are your plans for upstreaming your ECDSA signing implementation?
I expect that the ECDSA signing series will get merged soon.
I've currently dedicated the next four weeks to getting signing+verification implemented, so if you'd like a helping hand either with any leftover signing work or to get verification started I'm happy to collaborate.
Verification on the target would be great. My implementation is platform-specific. It would make sense to also have a software implementation of ECDSA (as we do for RSA). Once that is in place, it opens the gates for unit-testing. Currently, we're only testing the host signing part, but it would be awesome to have a test for ECDSA_UCLASS.
I think getting started on a software implementation of ECDSA_UCLASS would be most beneficial. Is that something you'd like to take on?
Alex
All the best, Tim
-----Original Message----- From: Alex G. mr.nuke.me@gmail.com Sent: February 5, 2021 11:09 AM To: Simon Glass sjg@chromium.org; Tim Romanski t-tromanski@microsoft.com Cc: u-boot@lists.denx.de; Deskin Miller deskinm@microsoft.com; Dylan D'Silva ddsilva@microsoft.com Subject: [EXTERNAL] Re: U-Boot ECDSA Implementation Question
Hi Tim,
On 2/5/21 8:35 AM, Simon Glass wrote:
I'm a current intern at Microsoft, and one of my priorities is to enable ECDSA for U-Boot image signing/verification. Simon mentioned someone is already working on ECC, it would be great to get synced up with related progress. For signing, I will likely replicate the existing approach of using the openssl library. I'm aware that signing happens on a host machine and verification happens during boot, which implies verification should have a custom implementation to avoid the openssl overhead in the U-Boot binary. My thoughts are to copy an ECC verification implementation from a well-tested widely-used open source project. I was wondering, is U-Boot's current RSA verification copied from another project? If so, how are security patches between the two copies of code usually handled? I'm thinking of deriving from the ECDSA implementation currently in the Linux kernel, though I'd also appreciate suggestions if there's a better/more widely tested & used implementation.
[snip]
Alexandru Gagniuc, on cc, has been looking at implementing the signing side of this recently and has sent some patches that you could look at.
I hope I can save you some effort on the signing side. Generally, you have two types of signed images. The first is the signed bootloader (BL2 or FSBL in ARM terms). The other one is the signed Flattened Image Tree (FIT) that we use in u-boot. The first one is vendor-specific, so you'd usually use vendor tools or write your own. We use mkimage to deal with the latter.
I've implemented the signing part [1] for mkimage. mkimage has the ability to use hardware signing via the PKCS11 engine of openssl, which I did not implement. You can read more about it here [3].
The verification part is still being defined [4][5]. The idea is to define a UCLASS which abstracts the underlying implementation. For RSA, it's defined here [6].
My goal with ECDSA verification was to use the ROM API of the STM32MP1. This meant I don't have to write a software implementation of ECDSA. This would be useful in two ways. It would enable ECDSA verification on devices that don't support it in hardware, and would also allow us to add some unit tests for ECDSA.
I suspect what you could do from here, is try to build my branch with ECDSA signing, play around with mkimage, and let us know how we can point you to the correct documentation. There's a lot of it in doc/, but it's not always easy to find.
Alex
[1] https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com... [2] https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com... [3] https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com... [4] https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com... [5] https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com... [6] https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com...
On 2/22/21 6:33 PM, Alex G. wrote:
Verification on the target would be great. My implementation is platform-specific. It would make sense to also have a software implementation of ECDSA (as we do for RSA). Once that is in place, it opens the gates for unit-testing. Currently, we're only testing the host signing part, but it would be awesome to have a test for ECDSA_UCLASS.
That sounds good! It aligns with our though process on the topic too.
I think getting started on a software implementation of ECDSA_UCLASS would be most beneficial. Is that something you'd like to take on?
Yessir, happy to get that started! I've been looking at the Linux and OpenSSL implementations, turns out Linux's ECRDSA is slightly different than what OpenSSL uses so I'm going ahead with isolating the OpenSSL ECDSA implementation right now and will fit it into a UCLASS after. My intuition is this will align the best with your implementation, though if you have different thoughts please let me know.
CC'd our @linux.microsoft.com emails, will be using those from now on.
All the best, Tim
-----Original Message----- From: Alex G. mr.nuke.me@gmail.com Sent: February 22, 2021 6:33 PM To: Tim Romanski t-tromanski@microsoft.com; Simon Glass sjg@chromium.org Cc: u-boot@lists.denx.de; Deskin Miller deskinm@microsoft.com; Dylan D'Silva ddsilva@microsoft.com Subject: Re: [EXTERNAL] Re: U-Boot ECDSA Implementation Question
On 2/10/21 1:45 PM, Tim Romanski wrote:
Hi Alex,
Thanks for the context! What are your plans for upstreaming your ECDSA signing implementation?
I expect that the ECDSA signing series will get merged soon.
I've currently dedicated the next four weeks to getting signing+verification implemented, so if you'd like a helping hand either with any leftover signing work or to get verification started I'm happy to collaborate.
Verification on the target would be great. My implementation is platform-specific. It would make sense to also have a software implementation of ECDSA (as we do for RSA). Once that is in place, it opens the gates for unit-testing. Currently, we're only testing the host signing part, but it would be awesome to have a test for ECDSA_UCLASS.
I think getting started on a software implementation of ECDSA_UCLASS would be most beneficial. Is that something you'd like to take on?
Alex
All the best, Tim
-----Original Message----- From: Alex G. mr.nuke.me@gmail.com Sent: February 5, 2021 11:09 AM To: Simon Glass sjg@chromium.org; Tim Romanski t-tromanski@microsoft.com Cc: u-boot@lists.denx.de; Deskin Miller deskinm@microsoft.com; Dylan D'Silva ddsilva@microsoft.com Subject: [EXTERNAL] Re: U-Boot ECDSA Implementation Question
Hi Tim,
On 2/5/21 8:35 AM, Simon Glass wrote:
I'm a current intern at Microsoft, and one of my priorities is to enable ECDSA for U-Boot image signing/verification. Simon mentioned someone is already working on ECC, it would be great to get synced up with related progress. For signing, I will likely replicate the existing approach of using the openssl library. I'm aware that signing happens on a host machine and verification happens during boot, which implies verification should have a custom implementation to avoid the openssl overhead in the U-Boot binary. My thoughts are to copy an ECC verification implementation from a well-tested widely-used open source project. I was wondering, is U-Boot's current RSA verification copied from another project? If so, how are security patches between the two copies of code usually handled? I'm thinking of deriving from the ECDSA implementation currently in the Linux kernel, though I'd also appreciate suggestions if there's a better/more widely tested & used implementation.
[snip]
Alexandru Gagniuc, on cc, has been looking at implementing the signing side of this recently and has sent some patches that you could look at.
I hope I can save you some effort on the signing side. Generally, you have two types of signed images. The first is the signed bootloader (BL2 or FSBL in ARM terms). The other one is the signed Flattened Image Tree (FIT) that we use in u-boot. The first one is vendor-specific, so you'd usually use vendor tools or write your own. We use mkimage to deal with the latter.
I've implemented the signing part [1] for mkimage. mkimage has the ability to use hardware signing via the PKCS11 engine of openssl, which I did not implement. You can read more about it here [3].
The verification part is still being defined [4][5]. The idea is to define a UCLASS which abstracts the underlying implementation. For RSA, it's defined here [6].
My goal with ECDSA verification was to use the ROM API of the STM32MP1. This meant I don't have to write a software implementation of ECDSA. This would be useful in two ways. It would enable ECDSA verification on devices that don't support it in hardware, and would also allow us to add some unit tests for ECDSA.
I suspect what you could do from here, is try to build my branch with ECDSA signing, play around with mkimage, and let us know how we can point you to the correct documentation. There's a lot of it in doc/, but it's not always easy to find.
Alex
[1] https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgith ub.com%2Fmrnuke%2Fu-boot%2Fcommits%2Fpatch-mkimage-keyfile-v1&data =04%7C01%7Ct-tromanski%40microsoft.com%7Cdf9b77d80e0a407b118408d8d78a4 148%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C637496336104405466%7C Unknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1h aWwiLCJXVCI6Mn0%3D%7C1000&sdata=DaeOfna1Judsb9nMxHis68hqh2bM45ojlh kqoorEfTQ%3D&reserved=0 [2] https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgith ub.com%2Fmrnuke%2Fu-boot%2Fcommit%2Fa2ae016f2f80579962d4ab058137c8e1a4 f4741f&data=04%7C01%7Ct-tromanski%40microsoft.com%7Cdf9b77d80e0a40 7b118408d8d78a4148%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C637496 336104405466%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luM zIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=x%2FnhL7GVyi2yL34kU Lzn3O5eOQMzkLZUpLolBIrg9wM%3D&reserved=0 [3] https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgith ub.com%2Fmrnuke%2Fu-boot%2Fblob%2F3f447efcf8ad98d0eea349994810a66b453a c188%2Fdoc%2FuImage.FIT%2Fsignature.txt%23L488&data=04%7C01%7Ct-tr omanski%40microsoft.com%7Cdf9b77d80e0a407b118408d8d78a4148%7C72f988bf8 6f141af91ab2d7cd011db47%7C1%7C0%7C637496336104405466%7CUnknown%7CTWFpb GZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0 %3D%7C1000&sdata=2TCDph4oYNyhKsjS8gc7MVKnzff42EXcrQC%2Flt5qILw%3D& amp;reserved=0 [4] https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgith ub.com%2Fmrnuke%2Fu-boot%2Fcommit%2F31caceb0e28959881e96ea49a0b28fd44d 13a947&data=04%7C01%7Ct-tromanski%40microsoft.com%7Cdf9b77d80e0a40 7b118408d8d78a4148%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C637496 336104405466%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luM zIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=bICF7YOUPu4Hnhi0XlQ BKnktnGAqJvj1auE3GUrRb44%3D&reserved=0 [5] https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgith ub.com%2Fmrnuke%2Fu-boot%2Fcommits%2Fpatch-stm32-ecdsa-v1&data=04% 7C01%7Ct-tromanski%40microsoft.com%7Cdf9b77d80e0a407b118408d8d78a4148% 7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C637496336104405466%7CUnkn own%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwi LCJXVCI6Mn0%3D%7C1000&sdata=PCskH6TTPETw3Hzfreytuaf33xRaeRn4WpHf4n 9wn9k%3D&reserved=0 [6] https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgith ub.com%2Fmrnuke%2Fu-boot%2Fblob%2F7d7ce8d70287568071a5d24acb6dc74b923f e7e0%2Finclude%2Fu-boot%2Frsa-mod-exp.h%23L79&data=04%7C01%7Ct-tro manski%40microsoft.com%7Cdf9b77d80e0a407b118408d8d78a4148%7C72f988bf86 f141af91ab2d7cd011db47%7C1%7C0%7C637496336104415461%7CUnknown%7CTWFpbG Zsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0% 3D%7C1000&sdata=WrQeFYIFOA3DWuztq7%2BbLUwWCQmYpnFRdc80Tyg%2FoIs%3D &reserved=0
participants (3)
-
Alex G. -
Simon Glass -
Tim Romanski