Ilias Apalodimas ilias.apalodimas@linaro.org writes:
Hi Jon,
On Wed, 10 Apr 2024 at 20:35, Jon Humphreys j-humphreys@ti.com wrote:
Ilias Apalodimas ilias.apalodimas@linaro.org writes:
On Tue, 9 Apr 2024 at 23:14, Andrew Davis afd@ti.com wrote:
On 4/9/24 2:26 PM, Heinrich Schuchardt wrote:
On 4/9/24 14:14, Andrew Davis wrote:
On 4/8/24 10:34 PM, Heinrich Schuchardt wrote: > On 4/8/24 23:33, Jonathan Humphreys wrote: >> EFI signature list using TI dummy keys. > > Adding vendor public keys into the code base to lock down generated > binaries to the vendors unpublished private key does not match well with > the intent of the GNU public license. >
The matching private keys are already published in this same repo/directory (arch/arm/mach-k3/keys).
Andrew
Why should we create signed capsules which are already compromised by publishing the private key?
If you buy these devices you have two options, you can burn real keys, or you can burn these dummy keys. If you burn dummy keys then these images will boot and so will any image you or anyone else wants to boot on the device. (since the keys are published anyone can make images for them, that is how we do GP (general purpose) devices these days)
If you burn your own keys, then you switch out these keys here and your device will only boot images that you permit by signing with your keys.
I am not sure I am following you here. We don't burn anything in the case of EFI keys. They are placed in an elf section and we assume the device will have a chain of trust enabled, naturally verifying those keys along with the u-boot binary.
You'll find plenty of open source projects do the same and give out example keys to show how to use real keys, even official GNU projects.
Yes, but the keys defined here are useless unless you have a default defconfig that uses them and embeds them in the binary. I am not cc'ed in all the patches of the series, is that added somewhere? And if you
Yes, they are part of this series https://lore.kernel.org/r/20240408213349.96610-1-j-humphreys@ti.com. Thanks for the reviews.
unconditionally enable secure boot It would be far more interesting to embed the MS SHIM key along with that special key you are trying to define, so that firmware can boot COTS distros as well
Yes, we should consider. But since that is outside of the EFI capsule use case, I would rather take it up in a separate patch.
Ok, the commit message wasn't clear, and based on Andrews's initial response I thought you wanted to use those for UEFI secure boot, not capsule updates. Those are your boards so I won't NAK this, but I'd strongly advise *not* to add this. I assume you want capsule auth by default because SystemReady-IR >=2.0 mandates it?
In that case, it would be a far better idea to document the process of creating signed capsules clearly either in U-Boots EFI docs and/or your board docs. I am pretty confident that if we merge this now we will have future products using the keys above
Thanks Ilias.
If I understand you correctly, I don't agree with the approach of not having a working implementation so that developers are forced to think through their support. Not having a feature enabled in upstream leads to latent bugs, bit rot, lack of coordination and openess, etc. It worries me that there are so many claims of authenticated capsule support but nothing in upstream on those devices.
But I absolutely acknowledge your concern that if we make this 'just work' then developers will overlook the details and not properly secure their solutions.
What I suggest to mitigate this: 1) as you say, add documentation, including a 'porting guide' section so developers know what steps they need to take, and 2) Add a developer config that is set by default. With this config set, during capsule updates, emit a warning message that instructs them to read the porting section of the doc to ensure they have secured their solution. To remove the warning, a developer would follow the porting guidelines and then unset the "developer config" configuration to signify they have secured the solution. This is what optee does, emitting the below during boot, until the config is unset:
I/TC: WARNING: This OP-TEE configuration might be insecure! I/TC: WARNING: Please check https://optee.readthedocs.io/en/latest/architecture/porting_guidelines.html
Last, I am pondering the idea of not including the actual .esl certificate, but rather adding a make step that would generate the certificate from a given keypair. For the TI upstream implementation, we would point to the open developer keys already being used to demonstrate secure boot. As part of porting, the developer would point to their own keys. That way the certificate has the developer's info, even if they are using our open keys for development purposes. Thoughts?
Outside of the capsule authentication aspect, the porting section would be a good place to remind the developer to define his own capsule GUIDs as well.
Jon
Thanks /Ilias
Thanks /Ilias
https://github.com/gpg/gnupg/tree/master/tests/openpgp/samplekeys
Andrew
Best regards
Heinrich
> Best regards > > Heinrich > >> >> Signed-off-by: Jonathan Humphreys j-humphreys@ti.com >> --- >> arch/arm/mach-k3/keys/custMpk.esl | Bin 0 -> 1523 bytes >> 1 file changed, 0 insertions(+), 0 deletions(-) >> create mode 100644 arch/arm/mach-k3/keys/custMpk.esl >> >> diff --git a/arch/arm/mach-k3/keys/custMpk.esl >> b/arch/arm/mach-k3/keys/custMpk.esl >> new file mode 100644 >> index >> 0000000000000000000000000000000000000000..2feb704e0a5fd126410de451d3c0fa4d3edccc52 >> GIT binary patch >> literal 1523 >> zcmZ1&d0^?2Da*aux2_hA(f&~MnUw(yu0v@E4?-F=u^u*PVqVQ8QZ((-^A*$m*Kg7c >> z&78AJODc2mtxpELY@Awc9&O)w85y}*84Mcd8gd(OvN4CUun9AT2E#ZUJWL@GhWtR) >> zKpA!(HkZVloWx>7bput902hy3NNPo5v4Uq_aY<2WZfaf$h@G5YRFGekSdyAzC~P1I >> zQpnB26;PC)oLXF*UsMbeWai-t@l*&dEdVMmF_blshP#N9QH-w`BJNO<sh6CeYal1i >> zYh-L-W?*PwYGi0=7A4MWYz$;tLb-$9{Y^|t$U)A?%D~*j#Lr;R#Kgta#Kg$3Uu2!< >> zjryX?*~({Md+?>+QS$x7=il`0?bc6sZ`Vxxl^6N{>i2E;SY*4-T$+0G;)5dxe+2CR >> z@4+)sDPWdQb@%6KTpDVdm)v}?GSpG(w_UV)&i+#e3fJowDZO)JR83lIcbw(hMu}}Y >> z2ZZwYAI-LVx@^G;HdkgxaX&Hnl_l3&{H|3l7uX@Vl5di{>fQQ{pDynFlySp2(z~g) >> z{LIBUzm&K9j_CMw_SIFfPdcT#zmg6g<ji}(R`6geJLk-#o7bK^&&fT}#2zsD`=c9g >> zFUCK<Fz@{2kel&$W6zl<d|WNk#ZsNRd{_N_SJxWvh0*K$j!m)c@oT>{#b(Lp`M3Uj >> zGOKycyEe+n{G(Rmg}jB!)0ySk-!kkj_R7#OT+}pcG0VXh?f+ftRvnyw#hUea^Iyfn >> ze|zgKPKrqe@jYWU?v<50X(n^lZ*G%j$JyCh`*Px|H*K=2WXP)hx>jng+}Q}N^KoDN >> z8dh8T-~Dmrp2?yk3O6Gqbz7O@<TEz<^zIa7d#PKtHKHeAg?V0DMSin^o3F|IEfQWk >> zcmJwBy6&2hKub%G{j3IK(?7m@uI43#1e~wSZJ5sTtDjrp@7@{O3(faN{`Gp}x{$M5 >> z{A7`c@pjfYq1Z=JvgZ^-zCC<(HFTBwYhTX$k`7IJX`SM!H}f`Mv+(Op6uVY(<(^o4 >> zpyXAj9nF_c-1A<UIel9%6Eh<NBXSA>W=dcRVPvS;*B%(4`P|iK>Vg$XDgN9sr}Df{ >> z7X0es=RPHr8RB+*)}q}h%gn?x9PO4y*Qog};x<<LS+lxk$@$kYlG_hXu6p%jvB<%l >> zmcsdI9w!^rFPt^&c~{1?L~DJ4TRPv>t%rn8xi;KBE9A!Dppb9yru|>RCb9PcXWpE> >> zKlQ}fzw*izXI|}|r!O*nb&cP9#VhHRn;B<SRflN2Jl(*;W4e0LD$ORRIdjjhURZH+ >> zXWR0Vllb2@>`1LC^xvIctvLCYhRA_6yCS~2&!0SH1xwv(O~<l(HQxHJxzF!T_>+5t >> z^|E$S{MM^8j9J5`sQ6pud{2Lz?k`zncbjvHj%eutjusUol}8;%cbPLCO|e;ZJ^tXe >> z_N{pmM}uCi3UWO3=hMc<s}m1Jx4GS4F(<_N`R|o+)eAK3Yx{o$ygRe!;<_EoF&UhP >> zrslJ=2XA9^$j#UDYwo;ZvZwb!|L%YP%v|ie|7-1PP+q3DZ&vEWgHHrjHv|NzEVjO? >> zKFeRbXv>iTPl?N16Xv@buq_d@TU<MB;uD_jX^$J`&*C>`uX0_s&g9M2C6cKx4E;{? >> zt`1&)Tk-yb?sKMPI~!}xt*d*!tMat!r1`}jul#i@lDB8rnu>ba_-^4!iQ5{|tb3TX >> z>fTMIw2!Me3{Dw*WZotC<4@h<H`zaL+~Es<{Ccj5yS7zyNU!YsTG`^JqA6NkU%vnV >> D66<<J >> >> literal 0 >> HcmV?d00001 >> >