Hi, Michael It's indeed a problem that NET_RANDOM (NetRandomInitSeed ()) may return same value if the time service is not accuracy enough on some platform. And as you said it's used to give unique but not necessarily to be real random number, so I think we could just add some salt into the seed. I think the platform monotonic counter may be a good choose, we can call gBS->GetNextMonotonicCount() and add the low 32 bits of MC to the seed. This could make the returned seed unique even GetTime() return all zero in the worst case. What do you think about it?
For your recommendation #3, the IaId is configured by Ip6Dxe and PXE driver, and there are no interface to exchange the index between them. DHCP6 driver just use the IaId configured by caller and it couldn't modify it. And for 4, one of the design intention of PXE boot is "platform doesn't need any pre-configuration", so it's also not acceptable. Best Regards, Siyuan -----Original Message----- From: Michael Chang [mailto:mch...@suse.com] Sent: Thursday, July 23, 2015 3:33 PM To: edk2-devel@lists.sourceforge.net Subject: [edk2] DHCPv6 IAID and NetRandomInitSeed problem in OVMF Hi, I have problem in IPv6 PXE booting under OVMF that the process terminated immediately right after selecting the UEFI IPv6 boot entry. I have to change the IPv6 config policy from automatic to manual to get things working again for me. I looked a bit into this problem, it was terminated by the error in IaId check in NetworkPkg/Dhcp6Dxe/Dhcp6Impl.c::EfiDhcp6Configure function // // Make sure the (IaId, IaType) is unique over all the instances. // NET_LIST_FOR_EACH (Entry, &Service->Child) { Other = NET_LIST_USER_STRUCT (Entry, DHCP6_INSTANCE, Link); if (Other->IaCb.Ia != NULL && Other->IaCb.Ia->Descriptor.Type == Dhcp6CfgData->IaDescriptor.Type && Other->IaCb.Ia->Descriptor.IaId == Dhcp6CfgData->IaDescriptor.IaId ) { return EFI_INVALID_PARAMETER; } } Which would be saying clearly it was not "unique" among all the running DHCPv6 instances. At that time there should be only two: 1. The IPv6 config instance, while the policy is set to automatic would exchange information with server with the IaId initialized at NetworkPkg/Ip6Dxe/Ip6ConfigImpl.c::Ip6ConfigInitInstance Instance->IaId = NET_RANDOM (NetRandomInitSeed ()); for (Index = 0; Index < IpSb->SnpMode.HwAddressSize; Index++) { Instance->IaId |= (IpSb->SnpMode.CurrentAddress.Addr[Index] << ((Index << 3) & 31)); } 2. The IPv6 PXE Base code driver would start another DHCPv6 session with the IaId initialized at NetworkPkg/UefiPxeBcDxe/PxeBcDriver.c::PxeBcCreateIp6Children // // Generate a random IAID for the Dhcp6 assigned address. // Private->IaId = NET_RANDOM (NetRandomInitSeed ()); if (Private->Snp != NULL) { for (Index = 0; Index < Private->Snp->Mode->HwAddressSize; Index++) { Private->IaId |= (Private->Snp->Mode->CurrentAddress.Addr[Index] << ((Index << 3) & 31)); } } Unfortunately both IaId collides, and quite surprisingly it's NET_RANDOM (NetRandomInitSeed ()), or NetRandomInitSeed () generates the same number for them. UINT32 EFIAPI NetRandomInitSeed ( VOID ) { EFI_TIME Time; UINT32 Seed; gRT->GetTime (&Time, NULL); Seed = (~Time.Hour << 24 | Time.Day << 16 | Time.Minute << 8 | Time.Second); Seed ^= Time.Nanosecond; Seed ^= Time.Year << 7; return Seed; } The returned Time.Nanosecond is always zero as the timer resolution is l Hz for PC-AT CMOS RTC, that seems to be default real time clock driver used by OVMF. At least from .dsc file it says so .. PcAtChipsetPkg/PcatRealTimeClockRuntimeDxe/PcatRealTimeClockRuntimeDxe.inf The NetRandomInitSeed is not random enough in this case, and the NET_RANDOM marco is even more confusing as it's not a random number generator either (so what is good for always feeding random "seed" to a deterministic mathematic fucntion ??) I don't have good idea how to fix the problem, but here are of my thoughs. 1. Use real timer clock driver which could provide higher resolution for time service. Qemu is able to provide HPET timer but I don't know how to configure OVMF to use it. 2. Use GetPerformanceCounter(), but that would increase image size by linking TimerLib. 3. The IaId has to be "unique" thus not necessarily to be random, probably it should be derived by some properties like MAC address and an assgined index (staring from zero) for the interface to be confiured. 4. Use manual as default IPv6 policy, but that just hide the problem. Hope that my information and findings helps. :) Thanks, Michael ------------------------------------------------------------------------------ _______________________________________________ edk2-devel mailing list edk2-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/edk2-devel ------------------------------------------------------------------------------ _______________________________________________ edk2-devel mailing list edk2-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/edk2-devel
