| Message ID | 1610099389-28329-1-git-send-email-pnagar@codeaurora.org (mailing list archive) |
|---|---|
| State | Changes Requested |
| Delegated to: | Paul Moore |
| Headers | show |
| Series | [RFC,v2] selinux: security: Move selinux_state to a separate page | expand |
On Fri, Jan 8, 2021 at 10:52 AM Preeti Nagar <pnagar@codeaurora.org> wrote: > > We want to seek your suggestions and comments on the idea and the changes > in the patch. Not sure why I was Cc'd, but I have a quick comment nevertheless. > +#ifdef CONFIG_SECURITY_RTIC > +struct selinux_state selinux_state __rticdata; > +#else > struct selinux_state selinux_state; > +#endif If you define an empty __rticdata for the !CONFIG case, then we don't need #ifdefs for uses like this. Cheers, Miguel
On Fri, Jan 8, 2021 at 10:52 AM Preeti Nagar <pnagar@codeaurora.org> wrote: > > We want to seek your suggestions and comments on the idea and the changes > in the patch. Cc'ing Nick manually since his email domain was (again :-) incorrect. Cheers, Miguel
On 1/8/2021 1:49 AM, Preeti Nagar wrote: > The changes introduce a new security feature, RunTime Integrity Check > (RTIC), designed to protect Linux Kernel at runtime. The motivation > behind these changes is: > 1. The system protection offered by SE for Android relies on the > assumption of kernel integrity. If the kernel itself is compromised (by > a perhaps as yet unknown future vulnerability), SE for Android security > mechanisms could potentially be disabled and rendered ineffective. > 2. Qualcomm Snapdragon devices use Secure Boot, which adds cryptographic > checks to each stage of the boot-up process, to assert the authenticity > of all secure software images that the device executes. However, due to > various vulnerabilities in SW modules, the integrity of the system can be > compromised at any time after device boot-up, leading to un-authorized > SW executing. It would be helpful if you characterized the "various vulnerabilities" rather than simply asserting their existence. This would allow the reviewer to determine if the proposed patch addresses the issue. > > The feature's idea is to move some sensitive kernel structures to a > separate page and monitor further any unauthorized changes to these, > from higher Exception Levels using stage 2 MMU. Moving these to a > different page will help avoid getting page faults from un-related data. I've always been a little slow when it comes to understanding the details of advanced memory management facilities. That's part of why I work in access control. Could you expand this a bit, so that someone who doesn't already know how your stage 2 MMU works might be able to evaluate what you're doing here. > Using this mechanism, some sensitive variables of the kernel which are > initialized after init or are updated rarely can also be protected from > simple overwrites and attacks trying to modify these. How would this interact with or complement __read_mostly? > > Currently, the change moves selinux_state structure to a separate page. In > future we plan to move more security-related kernel assets to this page to > enhance protection. What's special about selinux_state? What about the SELinux policy? How would I, as maintainer of the Smack security module, know if some Smack data should be treated the same way? > > We want to seek your suggestions and comments on the idea and the changes > in the patch. > > Signed-off-by: Preeti Nagar <pnagar@codeaurora.org> > --- > include/asm-generic/vmlinux.lds.h | 10 ++++++++++ > include/linux/init.h | 4 ++++ > security/Kconfig | 10 ++++++++++ > security/selinux/hooks.c | 4 ++++ > 4 files changed, 28 insertions(+) > > diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h > index b2b3d81..158dbc2 100644 > --- a/include/asm-generic/vmlinux.lds.h > +++ b/include/asm-generic/vmlinux.lds.h > @@ -770,6 +770,15 @@ > *(.scommon) \ > } > > +#ifdef CONFIG_SECURITY_RTIC > +#define RTIC_BSS \ > + . = ALIGN(PAGE_SIZE); \ > + KEEP(*(.bss.rtic)) \ > + . = ALIGN(PAGE_SIZE); > +#else > +#define RTIC_BSS > +#endif > + > /* > * Allow archectures to redefine BSS_FIRST_SECTIONS to add extra > * sections to the front of bss. > @@ -782,6 +791,7 @@ > . = ALIGN(bss_align); \ > .bss : AT(ADDR(.bss) - LOAD_OFFSET) { \ > BSS_FIRST_SECTIONS \ > + RTIC_BSS \ > . = ALIGN(PAGE_SIZE); \ > *(.bss..page_aligned) \ > . = ALIGN(PAGE_SIZE); \ > diff --git a/include/linux/init.h b/include/linux/init.h > index 7b53cb3..617adcf 100644 > --- a/include/linux/init.h > +++ b/include/linux/init.h > @@ -300,6 +300,10 @@ void __init parse_early_options(char *cmdline); > /* Data marked not to be saved by software suspend */ > #define __nosavedata __section(".data..nosave") > > +#ifdef CONFIG_SECURITY_RTIC > +#define __rticdata __section(".bss.rtic") > +#endif > + > #ifdef MODULE > #define __exit_p(x) x > #else > diff --git a/security/Kconfig b/security/Kconfig > index 7561f6f..66b61b9 100644 > --- a/security/Kconfig > +++ b/security/Kconfig > @@ -291,5 +291,15 @@ config LSM > > source "security/Kconfig.hardening" > > +config SECURITY_RTIC > + bool "RunTime Integrity Check feature" Shouldn't this depend on the architecture(s) supporting the feature? > + help > + RTIC(RunTime Integrity Check) feature is to protect Linux kernel > + at runtime. This relocates some of the security sensitive kernel > + structures to a separate page aligned special section. > + > + This is to enable monitoring and protection of these kernel assets > + from a higher exception level(EL) against any unauthorized changes. "if you are unsure ..." > + > endmenu > > diff --git a/security/selinux/hooks.c b/security/selinux/hooks.c > index 6b1826f..7add17c 100644 > --- a/security/selinux/hooks.c > +++ b/security/selinux/hooks.c > @@ -104,7 +104,11 @@ > #include "audit.h" > #include "avc_ss.h" > > +#ifdef CONFIG_SECURITY_RTIC > +struct selinux_state selinux_state __rticdata; > +#else > struct selinux_state selinux_state; > +#endif Shouldn't the __rticdata tag be applied always, and its definition take care of the cases where it doesn't do anything? > > /* SECMARK reference count */ > static atomic_t selinux_secmark_refcount = ATOMIC_INIT(0);
Via: https://lore.kernel.org/lkml/1610099389-28329-1-git-send-email-pnagar@codeaurora.org/ > diff --git a/include/linux/init.h b/include/linux/init.h > index 7b53cb3..617adcf 100644 > --- a/include/linux/init.h > +++ b/include/linux/init.h > @@ -300,6 +300,10 @@ void __init parse_early_options(char *cmdline); > /* Data marked not to be saved by software suspend */ > #define __nosavedata __section(".data..nosave") > > +#ifdef CONFIG_SECURITY_RTIC > +#define __rticdata __section(".bss.rtic") if you put: #else #define __rticdata here, then you wouldn't need to label each datum you put in there. > +#endif > + > #ifdef MODULE > #define __exit_p(x) x > #else > --- a/security/selinux/hooks.c > +++ b/security/selinux/hooks.c > @@ -104,7 +104,11 @@ > #include "audit.h" > #include "avc_ss.h" > > +#ifdef CONFIG_SECURITY_RTIC > +struct selinux_state selinux_state __rticdata; > +#else > struct selinux_state selinux_state; > +#endif so you could then drop the if-def here. Happy to see this resolved when building with LLD+LTO, which has been a problem in the past. Disabling selinux is a common attack vector on Android devices, so happy to see some effort towards mitigation. You might want to communicate the feature more to existing OEMs that are using your chipsets that support this feature.
On 2021-01-08 20:55, Miguel Ojeda wrote: > On Fri, Jan 8, 2021 at 10:52 AM Preeti Nagar <pnagar@codeaurora.org> > wrote: >> >> We want to seek your suggestions and comments on the idea and the >> changes >> in the patch. > > Not sure why I was Cc'd, but I have a quick comment nevertheless. > >> +#ifdef CONFIG_SECURITY_RTIC >> +struct selinux_state selinux_state __rticdata; >> +#else >> struct selinux_state selinux_state; >> +#endif > > If you define an empty __rticdata for the !CONFIG case, then we don't > need #ifdefs for uses like this. > > Cheers, > Miguel Thank you for the review! Will update this change in the next version of the patch soon. Thanks, Preeti
On 2021-01-09 06:31, Nick Desaulniers wrote: > Via: > https://lore.kernel.org/lkml/1610099389-28329-1-git-send-email-pnagar@codeaurora.org/ > >> diff --git a/include/linux/init.h b/include/linux/init.h >> index 7b53cb3..617adcf 100644 >> --- a/include/linux/init.h >> +++ b/include/linux/init.h >> @@ -300,6 +300,10 @@ void __init parse_early_options(char *cmdline); >> /* Data marked not to be saved by software suspend */ >> #define __nosavedata __section(".data..nosave") >> >> +#ifdef CONFIG_SECURITY_RTIC >> +#define __rticdata __section(".bss.rtic") > > if you put: > > #else > #define __rticdata > > here, then you wouldn't need to label each datum you put in there. > >> +#endif >> + >> #ifdef MODULE >> #define __exit_p(x) x >> #else > >> --- a/security/selinux/hooks.c >> +++ b/security/selinux/hooks.c >> @@ -104,7 +104,11 @@ >> #include "audit.h" >> #include "avc_ss.h" >> >> +#ifdef CONFIG_SECURITY_RTIC >> +struct selinux_state selinux_state __rticdata; >> +#else >> struct selinux_state selinux_state; >> +#endif > > so you could then drop the if-def here. Will update this in next version, thank you for the suggestion. > Happy to see this resolved when building with LLD+LTO, which has been a > problem in the past. Yes, downstream we have this verified with LTO configs enabled. Let us know if you are suggesting to check anything additionally here. > Disabling selinux is a common attack vector on Android devices, so > happy > to see some effort towards mitigation. You might want to communicate > the feature more to existing OEMs that are using your chipsets that > support this feature. Glad to know the idea looks good! Yes, we will work on that, will communicate internally as well, thank you.
On 2021-01-08 22:41, Casey Schaufler wrote: > On 1/8/2021 1:49 AM, Preeti Nagar wrote: >> The changes introduce a new security feature, RunTime Integrity Check >> (RTIC), designed to protect Linux Kernel at runtime. The motivation >> behind these changes is: >> 1. The system protection offered by SE for Android relies on the >> assumption of kernel integrity. If the kernel itself is compromised >> (by >> a perhaps as yet unknown future vulnerability), SE for Android >> security >> mechanisms could potentially be disabled and rendered ineffective. >> 2. Qualcomm Snapdragon devices use Secure Boot, which adds >> cryptographic >> checks to each stage of the boot-up process, to assert the >> authenticity >> of all secure software images that the device executes. However, due >> to >> various vulnerabilities in SW modules, the integrity of the system can >> be >> compromised at any time after device boot-up, leading to un-authorized >> SW executing. > > It would be helpful if you characterized the "various vulnerabilities" > rather than simply asserting their existence. This would allow the > reviewer > to determine if the proposed patch addresses the issue. > There might not currently be vulnerabilities, but the system is meant more specifically to harden valuable assets against future compromises. The key value add is a third party independent entity keeping a watch on crucial kernel assets. >> >> The feature's idea is to move some sensitive kernel structures to a >> separate page and monitor further any unauthorized changes to these, >> from higher Exception Levels using stage 2 MMU. Moving these to a >> different page will help avoid getting page faults from un-related >> data. > > I've always been a little slow when it comes to understanding the > details of advanced memory management facilities. That's part of > why I work in access control. Could you expand this a bit, so that > someone who doesn't already know how your stage 2 MMU works might > be able to evaluate what you're doing here. > Sure, will include more details. The mechanism we have been working on removes the write permissions for HLOS in the stage 2 page tables for the regions to be monitored, such that any modification attempts to these will lead to faults being generated and handled by handlers. If the protected assets are moved to a separate page, faults will be generated corresponding to change attempts to these assets only. If not moved to a separate page, write attempts to un-related data which is present on the monitored pages will also be generated. >> Using this mechanism, some sensitive variables of the kernel which are >> initialized after init or are updated rarely can also be protected >> from >> simple overwrites and attacks trying to modify these. > > How would this interact with or complement __read_mostly? > Currently, the mechanism we are working on developing is independent of __read_mostly. This is something we can look more into while working further on the mechanism. >> >> Currently, the change moves selinux_state structure to a separate >> page. In >> future we plan to move more security-related kernel assets to this >> page to >> enhance protection. > > What's special about selinux_state? What about the SELinux policy? > How would I, as maintainer of the Smack security module, know if > some Smack data should be treated the same way? > We are investigating more of the SELinux related and other kernel assets which can be included in the protection. The basis of selinux_state is because disabling of SELinux is one of the common attack vectors in Android. We understand any kernel assets, unauthorized changes to which can give way to security or any other type of attack can be considered to be a potential asset to be added to the protection. >> >> We want to seek your suggestions and comments on the idea and the >> changes >> in the patch. >> >> Signed-off-by: Preeti Nagar <pnagar@codeaurora.org> >> --- >> include/asm-generic/vmlinux.lds.h | 10 ++++++++++ >> include/linux/init.h | 4 ++++ >> security/Kconfig | 10 ++++++++++ >> security/selinux/hooks.c | 4 ++++ >> 4 files changed, 28 insertions(+) >> >> diff --git a/include/asm-generic/vmlinux.lds.h >> b/include/asm-generic/vmlinux.lds.h >> index b2b3d81..158dbc2 100644 >> --- a/include/asm-generic/vmlinux.lds.h >> +++ b/include/asm-generic/vmlinux.lds.h >> @@ -770,6 +770,15 @@ >> *(.scommon) \ >> } >> >> +#ifdef CONFIG_SECURITY_RTIC >> +#define RTIC_BSS \ >> + . = ALIGN(PAGE_SIZE); \ >> + KEEP(*(.bss.rtic)) \ >> + . = ALIGN(PAGE_SIZE); >> +#else >> +#define RTIC_BSS >> +#endif >> + >> /* >> * Allow archectures to redefine BSS_FIRST_SECTIONS to add extra >> * sections to the front of bss. >> @@ -782,6 +791,7 @@ >> . = ALIGN(bss_align); \ >> .bss : AT(ADDR(.bss) - LOAD_OFFSET) { \ >> BSS_FIRST_SECTIONS \ >> + RTIC_BSS \ >> . = ALIGN(PAGE_SIZE); \ >> *(.bss..page_aligned) \ >> . = ALIGN(PAGE_SIZE); \ >> diff --git a/include/linux/init.h b/include/linux/init.h >> index 7b53cb3..617adcf 100644 >> --- a/include/linux/init.h >> +++ b/include/linux/init.h >> @@ -300,6 +300,10 @@ void __init parse_early_options(char *cmdline); >> /* Data marked not to be saved by software suspend */ >> #define __nosavedata __section(".data..nosave") >> >> +#ifdef CONFIG_SECURITY_RTIC >> +#define __rticdata __section(".bss.rtic") >> +#endif >> + >> #ifdef MODULE >> #define __exit_p(x) x >> #else >> diff --git a/security/Kconfig b/security/Kconfig >> index 7561f6f..66b61b9 100644 >> --- a/security/Kconfig >> +++ b/security/Kconfig >> @@ -291,5 +291,15 @@ config LSM >> >> source "security/Kconfig.hardening" >> >> +config SECURITY_RTIC >> + bool "RunTime Integrity Check feature" > > Shouldn't this depend on the architecture(s) supporting the > feature? > >> + help >> + RTIC(RunTime Integrity Check) feature is to protect Linux kernel >> + at runtime. This relocates some of the security sensitive kernel >> + structures to a separate page aligned special section. >> + >> + This is to enable monitoring and protection of these kernel assets >> + from a higher exception level(EL) against any unauthorized >> changes. > > "if you are unsure ..." > We just thought keeping it generic might be a better idea, thus, moved the changes to generic files from arch-specific files and thus, kept config also independent of the arch. Can surely make this config arch dependent if that is a better approach? >> + >> endmenu >> >> diff --git a/security/selinux/hooks.c b/security/selinux/hooks.c >> index 6b1826f..7add17c 100644 >> --- a/security/selinux/hooks.c >> +++ b/security/selinux/hooks.c >> @@ -104,7 +104,11 @@ >> #include "audit.h" >> #include "avc_ss.h" >> >> +#ifdef CONFIG_SECURITY_RTIC >> +struct selinux_state selinux_state __rticdata; >> +#else >> struct selinux_state selinux_state; >> +#endif > > Shouldn't the __rticdata tag be applied always, and its > definition take care of the cases where it doesn't do anything? > Will update this change in the next version of the patch. Thank you. >> >> /* SECMARK reference count */ >> static atomic_t selinux_secmark_refcount = ATOMIC_INIT(0);
On 1/12/2021 1:36 AM, pnagar@codeaurora.org wrote: > On 2021-01-08 22:41, Casey Schaufler wrote: >> On 1/8/2021 1:49 AM, Preeti Nagar wrote: >>> The changes introduce a new security feature, RunTime Integrity Check >>> (RTIC), designed to protect Linux Kernel at runtime. The motivation >>> behind these changes is: >>> 1. The system protection offered by SE for Android relies on the >>> assumption of kernel integrity. If the kernel itself is compromised (by >>> a perhaps as yet unknown future vulnerability), SE for Android security >>> mechanisms could potentially be disabled and rendered ineffective. >>> 2. Qualcomm Snapdragon devices use Secure Boot, which adds cryptographic >>> checks to each stage of the boot-up process, to assert the authenticity >>> of all secure software images that the device executes. However, due to >>> various vulnerabilities in SW modules, the integrity of the system can be >>> compromised at any time after device boot-up, leading to un-authorized >>> SW executing. >> >> It would be helpful if you characterized the "various vulnerabilities" >> rather than simply asserting their existence. This would allow the reviewer >> to determine if the proposed patch addresses the issue. >> > There might not currently be vulnerabilities, but the system is meant more > specifically to harden valuable assets against future compromises. The key > value add is a third party independent entity keeping a watch on crucial > kernel assets. Could you characterize the potential vulnerabilities, then? Seriously, there's a gazillion ways data integrity can be compromised. Which of those are addressed? > >>> >>> The feature's idea is to move some sensitive kernel structures to a >>> separate page and monitor further any unauthorized changes to these, >>> from higher Exception Levels using stage 2 MMU. Moving these to a >>> different page will help avoid getting page faults from un-related data. >> >> I've always been a little slow when it comes to understanding the >> details of advanced memory management facilities. That's part of >> why I work in access control. Could you expand this a bit, so that >> someone who doesn't already know how your stage 2 MMU works might >> be able to evaluate what you're doing here. >> > Sure, will include more details. The mechanism we have been working on > removes the write permissions for HLOS in the stage 2 page tables for > the regions to be monitored, such that any modification attempts to these > will lead to faults being generated and handled by handlers. If the > protected assets are moved to a separate page, faults will be generated > corresponding to change attempts to these assets only. If not moved to a > separate page, write attempts to un-related data which is present on the > monitored pages will also be generated. Thanks. > >>> Using this mechanism, some sensitive variables of the kernel which are >>> initialized after init or are updated rarely can also be protected from >>> simple overwrites and attacks trying to modify these. >> >> How would this interact with or complement __read_mostly? >> > Currently, the mechanism we are working on developing is > independent of __read_mostly. This is something we can look more into > while working further on the mechanism. Please either integrate the two or explain how they differ. It appears that you haven't considered how you might exploit or expand the existing mechanism. > >>> >>> Currently, the change moves selinux_state structure to a separate page. In >>> future we plan to move more security-related kernel assets to this page to >>> enhance protection. >> >> What's special about selinux_state? What about the SELinux policy? >> How would I, as maintainer of the Smack security module, know if >> some Smack data should be treated the same way? >> > We are investigating more of the SELinux related and other kernel assets > which can be included in the protection. The basis of selinux_state is > because disabling of SELinux is one of the common attack vectors in > Android. We understand any kernel assets, unauthorized changes to which > can give way to security or any other type of attack can be considered to > be a potential asset to be added to the protection. Yeah, I get that. It looks like this could be a useful mechanism beyond SELinux. No point in hoarding it. > >>> >>> We want to seek your suggestions and comments on the idea and the changes >>> in the patch. >>> >>> Signed-off-by: Preeti Nagar <pnagar@codeaurora.org> >>> --- >>> include/asm-generic/vmlinux.lds.h | 10 ++++++++++ >>> include/linux/init.h | 4 ++++ >>> security/Kconfig | 10 ++++++++++ >>> security/selinux/hooks.c | 4 ++++ >>> 4 files changed, 28 insertions(+) >>> >>> diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h >>> index b2b3d81..158dbc2 100644 >>> --- a/include/asm-generic/vmlinux.lds.h >>> +++ b/include/asm-generic/vmlinux.lds.h >>> @@ -770,6 +770,15 @@ >>> *(.scommon) \ >>> } >>> >>> +#ifdef CONFIG_SECURITY_RTIC >>> +#define RTIC_BSS \ >>> + . = ALIGN(PAGE_SIZE); \ >>> + KEEP(*(.bss.rtic)) \ >>> + . = ALIGN(PAGE_SIZE); >>> +#else >>> +#define RTIC_BSS >>> +#endif >>> + >>> /* >>> * Allow archectures to redefine BSS_FIRST_SECTIONS to add extra >>> * sections to the front of bss. >>> @@ -782,6 +791,7 @@ >>> . = ALIGN(bss_align); \ >>> .bss : AT(ADDR(.bss) - LOAD_OFFSET) { \ >>> BSS_FIRST_SECTIONS \ >>> + RTIC_BSS \ >>> . = ALIGN(PAGE_SIZE); \ >>> *(.bss..page_aligned) \ >>> . = ALIGN(PAGE_SIZE); \ >>> diff --git a/include/linux/init.h b/include/linux/init.h >>> index 7b53cb3..617adcf 100644 >>> --- a/include/linux/init.h >>> +++ b/include/linux/init.h >>> @@ -300,6 +300,10 @@ void __init parse_early_options(char *cmdline); >>> /* Data marked not to be saved by software suspend */ >>> #define __nosavedata __section(".data..nosave") >>> >>> +#ifdef CONFIG_SECURITY_RTIC >>> +#define __rticdata __section(".bss.rtic") >>> +#endif >>> + >>> #ifdef MODULE >>> #define __exit_p(x) x >>> #else >>> diff --git a/security/Kconfig b/security/Kconfig >>> index 7561f6f..66b61b9 100644 >>> --- a/security/Kconfig >>> +++ b/security/Kconfig >>> @@ -291,5 +291,15 @@ config LSM >>> >>> source "security/Kconfig.hardening" >>> >>> +config SECURITY_RTIC >>> + bool "RunTime Integrity Check feature" >> >> Shouldn't this depend on the architecture(s) supporting the >> feature? >> >>> + help >>> + RTIC(RunTime Integrity Check) feature is to protect Linux kernel >>> + at runtime. This relocates some of the security sensitive kernel >>> + structures to a separate page aligned special section. >>> + >>> + This is to enable monitoring and protection of these kernel assets >>> + from a higher exception level(EL) against any unauthorized changes. >> >> "if you are unsure ..." >> > We just thought keeping it generic might be a better idea, thus, moved the > changes to generic files from arch-specific files and thus, kept config also > independent of the arch. Can surely make this config arch dependent if that is > a better approach? It's kind of silly to enable this if the hardware doesn't support it, isn't it? > >>> + >>> endmenu >>> >>> diff --git a/security/selinux/hooks.c b/security/selinux/hooks.c >>> index 6b1826f..7add17c 100644 >>> --- a/security/selinux/hooks.c >>> +++ b/security/selinux/hooks.c >>> @@ -104,7 +104,11 @@ >>> #include "audit.h" >>> #include "avc_ss.h" >>> >>> +#ifdef CONFIG_SECURITY_RTIC >>> +struct selinux_state selinux_state __rticdata; >>> +#else >>> struct selinux_state selinux_state; >>> +#endif >> >> Shouldn't the __rticdata tag be applied always, and its >> definition take care of the cases where it doesn't do anything? >> > Will update this change in the next version of the patch. Thank you. I saw that several other people had the same comment. > >>> >>> /* SECMARK reference count */ >>> static atomic_t selinux_secmark_refcount = ATOMIC_INIT(0);
Casey Schaufler <casey@schaufler-ca.com> wrote: > >> How would this interact with or complement __read_mostly? > >> > > Currently, the mechanism we are working on developing is > > independent of __read_mostly. This is something we can look more into > > while working further on the mechanism. > > Please either integrate the two or explain how they differ. > It appears that you haven't considered how you might exploit > or expand the existing mechanism. I think __read_mostly is about grouping stuff together that's rarely going to be read to make the CPU's data cache more efficient. It doesn't stop people writing to such a variable. David
On 2021-01-13 16:26, David Howells wrote: > Casey Schaufler <casey@schaufler-ca.com> wrote: > >> >> How would this interact with or complement __read_mostly? >> >> >> > Currently, the mechanism we are working on developing is >> > independent of __read_mostly. This is something we can look more into >> > while working further on the mechanism. >> >> Please either integrate the two or explain how they differ. >> It appears that you haven't considered how you might exploit >> or expand the existing mechanism. > > I think __read_mostly is about grouping stuff together that's rarely > going to > be read to make the CPU's data cache more efficient. It doesn't stop > people > writing to such a variable. > > David Thank you for sharing!
diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h index b2b3d81..158dbc2 100644 --- a/include/asm-generic/vmlinux.lds.h +++ b/include/asm-generic/vmlinux.lds.h @@ -770,6 +770,15 @@ *(.scommon) \ } +#ifdef CONFIG_SECURITY_RTIC +#define RTIC_BSS \ + . = ALIGN(PAGE_SIZE); \ + KEEP(*(.bss.rtic)) \ + . = ALIGN(PAGE_SIZE); +#else +#define RTIC_BSS +#endif + /* * Allow archectures to redefine BSS_FIRST_SECTIONS to add extra * sections to the front of bss. @@ -782,6 +791,7 @@ . = ALIGN(bss_align); \ .bss : AT(ADDR(.bss) - LOAD_OFFSET) { \ BSS_FIRST_SECTIONS \ + RTIC_BSS \ . = ALIGN(PAGE_SIZE); \ *(.bss..page_aligned) \ . = ALIGN(PAGE_SIZE); \ diff --git a/include/linux/init.h b/include/linux/init.h index 7b53cb3..617adcf 100644 --- a/include/linux/init.h +++ b/include/linux/init.h @@ -300,6 +300,10 @@ void __init parse_early_options(char *cmdline); /* Data marked not to be saved by software suspend */ #define __nosavedata __section(".data..nosave") +#ifdef CONFIG_SECURITY_RTIC +#define __rticdata __section(".bss.rtic") +#endif + #ifdef MODULE #define __exit_p(x) x #else diff --git a/security/Kconfig b/security/Kconfig index 7561f6f..66b61b9 100644 --- a/security/Kconfig +++ b/security/Kconfig @@ -291,5 +291,15 @@ config LSM source "security/Kconfig.hardening" +config SECURITY_RTIC + bool "RunTime Integrity Check feature" + help + RTIC(RunTime Integrity Check) feature is to protect Linux kernel + at runtime. This relocates some of the security sensitive kernel + structures to a separate page aligned special section. + + This is to enable monitoring and protection of these kernel assets + from a higher exception level(EL) against any unauthorized changes. + endmenu diff --git a/security/selinux/hooks.c b/security/selinux/hooks.c index 6b1826f..7add17c 100644 --- a/security/selinux/hooks.c +++ b/security/selinux/hooks.c @@ -104,7 +104,11 @@ #include "audit.h" #include "avc_ss.h" +#ifdef CONFIG_SECURITY_RTIC +struct selinux_state selinux_state __rticdata; +#else struct selinux_state selinux_state; +#endif /* SECMARK reference count */ static atomic_t selinux_secmark_refcount = ATOMIC_INIT(0);
The changes introduce a new security feature, RunTime Integrity Check (RTIC), designed to protect Linux Kernel at runtime. The motivation behind these changes is: 1. The system protection offered by SE for Android relies on the assumption of kernel integrity. If the kernel itself is compromised (by a perhaps as yet unknown future vulnerability), SE for Android security mechanisms could potentially be disabled and rendered ineffective. 2. Qualcomm Snapdragon devices use Secure Boot, which adds cryptographic checks to each stage of the boot-up process, to assert the authenticity of all secure software images that the device executes. However, due to various vulnerabilities in SW modules, the integrity of the system can be compromised at any time after device boot-up, leading to un-authorized SW executing. The feature's idea is to move some sensitive kernel structures to a separate page and monitor further any unauthorized changes to these, from higher Exception Levels using stage 2 MMU. Moving these to a different page will help avoid getting page faults from un-related data. Using this mechanism, some sensitive variables of the kernel which are initialized after init or are updated rarely can also be protected from simple overwrites and attacks trying to modify these. Currently, the change moves selinux_state structure to a separate page. In future we plan to move more security-related kernel assets to this page to enhance protection. We want to seek your suggestions and comments on the idea and the changes in the patch. Signed-off-by: Preeti Nagar <pnagar@codeaurora.org> --- include/asm-generic/vmlinux.lds.h | 10 ++++++++++ include/linux/init.h | 4 ++++ security/Kconfig | 10 ++++++++++ security/selinux/hooks.c | 4 ++++ 4 files changed, 28 insertions(+)