Re: [PATCH 2/3] mm/filemap: initiate readahead even if IOCB_NOWAIT is set for the I/O
On 1/31/19 1:08 PM, Jiri Kosina wrote: > On Thu, 31 Jan 2019, Daniel Gruss wrote: > >> If I understood it correctly, this patch just removes the advantages of >> preadv2 over mmmap+access for the attacker. > > Which is the desired effect. We are not trying to solve the timing aspect, > as I don't think there is a reasonable way to do it, is there? There are two building blocks to cache attacks, bringing the cache into a state, and observing a state change, you can mitigate them by breaking either of these building blocks. For most attacks the attacker would be interested in observing *when* a specific victim page is loaded into the page cache rather than observing whether it is in the page cache right now (it could be there for ages if the system was not under memory pressure). So, one could try to prevent interference in the page cache between attacker and victim -> working set algorithms do that to some extent. Simpler idea (with more side effects) would be limiting the maximum share of the page cache per user (or per process, depending on the threat model)... Cheers, Daniel
Re: [PATCH 2/3] mm/filemap: initiate readahead even if IOCB_NOWAIT is set for the I/O
On 1/30/19 1:44 PM, Vlastimil Babka wrote: > Close that sidechannel by always initiating readahead on the cache if we > encounter a cache miss for preadv2(RWF_NOWAIT); with that in place, probing > the pagecache residency itself will actually populate the cache, making the > sidechannel useless. I fear this does not really close the side channel. You can time the preadv2 function and infer which path it took, so you just bring it down to the same as using mmap and timing accesses. If I understood it correctly, this patch just removes the advantages of preadv2 over mmmap+access for the attacker. Cheers, Daniel
Re: [PATCH] mm/mincore: allow for making sys_mincore() privileged
On 1/7/19 12:08 PM, Dominique Martinet wrote: >> That's my bigger concern here. In [1] there's described a remote attack >> (on webserver) using the page fault timing differences for present/not >> present page cache pages. Noisy but works, and I expect locally it to be >> much less noisy. Yet the countermeasures section only mentions >> restricting mincore() as if it was sufficient (and also how to make >> evictions harder, but that's secondary IMHO). > > I'd suggest making clock rougher for non-root users but javascript tried > that and it wasn't enough... :) > Honestly won't be of much help there, good luck? Restricting mincore() is sufficient to fix the hardware-agnostic part. If the attack is not hardware-agnostic anymore, an attacker could also just use a hardware cache attack, which has a higher temporal and spatial resolution, so there's no reason why the attacker would use page cache attacks instead then. Cheers, Daniel
Re: [kernel-hardening] [RFC, PATCH] x86_64: KAISER - do not map kernel in user mode
On 08.05.2017 16:09, Thomas Garnier wrote: Just to correct my answer here as well: Although we experimented with fixed mappings for per-cpu addresses, the current patch does not incorporate this yet, so it indeed still leaks. However, it is not a severe problem. The mapping of the required (per-cpu) variables would be at a fixed location in the user CR3, instead of the ones that are used in the kernel. Why do you think it should be at a fixed location in the user CR3? I see that you just mirror the entries. You also mirror __entry_text_start / __entry_text_end which is part of the binary so will leak the base address of the kernel. Maybe I am missing something. As I said, the current patch does not incorporate this yet, so yes, this part currently still leaks because we did not implement it yet.
Re: [kernel-hardening] [RFC, PATCH] x86_64: KAISER - do not map kernel in user mode
On 08.05.2017 16:09, Thomas Garnier wrote: Just to correct my answer here as well: Although we experimented with fixed mappings for per-cpu addresses, the current patch does not incorporate this yet, so it indeed still leaks. However, it is not a severe problem. The mapping of the required (per-cpu) variables would be at a fixed location in the user CR3, instead of the ones that are used in the kernel. Why do you think it should be at a fixed location in the user CR3? I see that you just mirror the entries. You also mirror __entry_text_start / __entry_text_end which is part of the binary so will leak the base address of the kernel. Maybe I am missing something. As I said, the current patch does not incorporate this yet, so yes, this part currently still leaks because we did not implement it yet.
Re: [kernel-hardening] [RFC, PATCH] x86_64: KAISER - do not map kernel in user mode
On 06.05.2017 10:38, Daniel Gruss wrote: On 2017-05-06 06:02, David Gens wrote: Assuming that their patch indeed leaks per-cpu addresses.. it might not necessarily be required to change it. I think we're not leaking them (unless we still have some bug in our code). Just to correct my answer here as well: Although we experimented with fixed mappings for per-cpu addresses, the current patch does not incorporate this yet, so it indeed still leaks. However, it is not a severe problem. The mapping of the required (per-cpu) variables would be at a fixed location in the user CR3, instead of the ones that are used in the kernel.
Re: [kernel-hardening] [RFC, PATCH] x86_64: KAISER - do not map kernel in user mode
On 06.05.2017 10:38, Daniel Gruss wrote: On 2017-05-06 06:02, David Gens wrote: Assuming that their patch indeed leaks per-cpu addresses.. it might not necessarily be required to change it. I think we're not leaking them (unless we still have some bug in our code). Just to correct my answer here as well: Although we experimented with fixed mappings for per-cpu addresses, the current patch does not incorporate this yet, so it indeed still leaks. However, it is not a severe problem. The mapping of the required (per-cpu) variables would be at a fixed location in the user CR3, instead of the ones that are used in the kernel.
Re: [kernel-hardening] [RFC, PATCH] x86_64: KAISER - do not map kernel in user mode
On 08.05.2017 15:22, Mark Rutland wrote: Specifically, I think this does not align with the statement in 2.1 regarding the two TTBRs: This simplifies privilege checks and does not require any address translation for invalid memory accesses and thus no cache lookups. ... since the use of the TTBRs is orthogonal to privilege checks and/or the design of the TLBs. Ok, this is a good point, we will try to clarify this in the paper.
Re: [kernel-hardening] [RFC, PATCH] x86_64: KAISER - do not map kernel in user mode
On 08.05.2017 15:22, Mark Rutland wrote: Specifically, I think this does not align with the statement in 2.1 regarding the two TTBRs: This simplifies privilege checks and does not require any address translation for invalid memory accesses and thus no cache lookups. ... since the use of the TTBRs is orthogonal to privilege checks and/or the design of the TLBs. Ok, this is a good point, we will try to clarify this in the paper.
Re: [kernel-hardening] [RFC, PATCH] x86_64: KAISER - do not map kernel in user mode
On 05.05.2017 10:23, Daniel Gruss wrote: - How this approach prevent the hardware attacks you mentioned? You still have to keep a part of _text in the pagetable and an attacker could discover it no? (and deduce the kernel base address). These parts are moved to a different section (.user_mapped) which is at a possibly predictable location - the location of the randomized parts of the kernel is independent of the location of .user_mapped. The code/data footprint for .user_mapped is quite small, helping to reduce or eliminate the attack surface... We just discussed that in our group again: although we experimented with this part, it's not yet included in the patch. The solution we sketched is, as I wrote, we map the required (per-thread) variables in the user CR3 to a fixed location in memory. During the context switch, only this fixed part remains mapped but not the randomized pages. This is not a lot of work, because it's just mapping a few more pages and fixing a 1 or 2 lines in the context switch.
Re: [kernel-hardening] [RFC, PATCH] x86_64: KAISER - do not map kernel in user mode
On 05.05.2017 10:23, Daniel Gruss wrote: - How this approach prevent the hardware attacks you mentioned? You still have to keep a part of _text in the pagetable and an attacker could discover it no? (and deduce the kernel base address). These parts are moved to a different section (.user_mapped) which is at a possibly predictable location - the location of the randomized parts of the kernel is independent of the location of .user_mapped. The code/data footprint for .user_mapped is quite small, helping to reduce or eliminate the attack surface... We just discussed that in our group again: although we experimented with this part, it's not yet included in the patch. The solution we sketched is, as I wrote, we map the required (per-thread) variables in the user CR3 to a fixed location in memory. During the context switch, only this fixed part remains mapped but not the randomized pages. This is not a lot of work, because it's just mapping a few more pages and fixing a 1 or 2 lines in the context switch.
Re: [kernel-hardening] [RFC, PATCH] x86_64: KAISER - do not map kernel in user mode
While it may be the case that in practice ARM systems do not have such a side channel, I think that it is erroneous to believe that the architectural TTBR{0,1} split ensures this. The use of TTBR0 for user and TTBR1 for kernel is entirely a SW policy, and not an architectural requirement. It is possible to map data in TTBR1 which is accessible to userspace, and data in TTBR0 which is only accessible by the kernel. In either case, this is determined by the page tables themselves. Absolutely right, but TTBR0 and TTBR1 are usually used in this way. Given this, I think that the statements in the KAISER paper regarding the TTBRs (in section 2.1) are not quite right. Architecturally, permission checks and lookups cannot be elided based on the TTBR used. As we say in section 2.1, they are "typically" used in this way, and this prevents the attacks. Not just the presence of a second register, but the way how the two registers are used to split the translation tables for user and kernel.
Re: [kernel-hardening] [RFC, PATCH] x86_64: KAISER - do not map kernel in user mode
While it may be the case that in practice ARM systems do not have such a side channel, I think that it is erroneous to believe that the architectural TTBR{0,1} split ensures this. The use of TTBR0 for user and TTBR1 for kernel is entirely a SW policy, and not an architectural requirement. It is possible to map data in TTBR1 which is accessible to userspace, and data in TTBR0 which is only accessible by the kernel. In either case, this is determined by the page tables themselves. Absolutely right, but TTBR0 and TTBR1 are usually used in this way. Given this, I think that the statements in the KAISER paper regarding the TTBRs (in section 2.1) are not quite right. Architecturally, permission checks and lookups cannot be elided based on the TTBR used. As we say in section 2.1, they are "typically" used in this way, and this prevents the attacks. Not just the presence of a second register, but the way how the two registers are used to split the translation tables for user and kernel.
Re: [kernel-hardening] Re: [RFC, PATCH] x86_64: KAISER - do not map kernel in user mode
On 2017-05-08 00:02, Richard Weinberger wrote: Ahh, *very* recent is the keyword then. ;) I was a bit confused since in your paper the overhead is less than 1%. Yes, only for very recent platforms (Skylake). While working on the paper we were surprised that we found overheads that small. What platforms did you test? We tested it on multiple platforms for stability, but we only ran longer performance tests on different Skylake i7-6700K systems we mentioned in the paper. i.e. how does it perform on recent AMD systems? Unfortunately, we don't have any AMD systems at hand. I'm also not sure how AMD is affected by the issue in the first place. Although unlikely, there is the possibility that the problem of KASLR information leakage through microarchitectural side channels might be Intel-specific.
Re: [kernel-hardening] Re: [RFC, PATCH] x86_64: KAISER - do not map kernel in user mode
On 2017-05-08 00:02, Richard Weinberger wrote: Ahh, *very* recent is the keyword then. ;) I was a bit confused since in your paper the overhead is less than 1%. Yes, only for very recent platforms (Skylake). While working on the paper we were surprised that we found overheads that small. What platforms did you test? We tested it on multiple platforms for stability, but we only ran longer performance tests on different Skylake i7-6700K systems we mentioned in the paper. i.e. how does it perform on recent AMD systems? Unfortunately, we don't have any AMD systems at hand. I'm also not sure how AMD is affected by the issue in the first place. Although unlikely, there is the possibility that the problem of KASLR information leakage through microarchitectural side channels might be Intel-specific.
Re: [kernel-hardening] Re: [RFC, PATCH] x86_64: KAISER - do not map kernel in user mode
Just did a quick test on my main KVM host, a 8 core Intel(R) Xeon(R) CPU E3-1240 V2. KVM guests are 4.10 w/o CONFIG_KAISER and kvmconfig without CONFIG_PARAVIRT. Building a defconfig kernel within that guests is about 10% slower when CONFIG_KAISER is enabled. Thank you for testing it! :) Is this expected? It sounds plausible. First, I would expect any form of virtualization to increase the overhead. Second, for the processor (Ivy Bridge), I would have expected even higher performance overheads. KAISER utilizes very recent performance improvements in Intel processors... If it helps I can redo the same test also on bare metal. I'm not sure how we proceed here and if this would help, because I don't know what everyone expects. KAISER definitely introduces an overhead, no doubt about that. How much overhead it is depends on the specific hardware and may be very little on recent architectures and more on older machines. We are not proposing to enable KAISER by default, but to provide the config option to allow easy integration into hardened kernels where performance overheads may be acceptable (which depends on the specific use case and the specific hardware).
Re: [kernel-hardening] Re: [RFC, PATCH] x86_64: KAISER - do not map kernel in user mode
Just did a quick test on my main KVM host, a 8 core Intel(R) Xeon(R) CPU E3-1240 V2. KVM guests are 4.10 w/o CONFIG_KAISER and kvmconfig without CONFIG_PARAVIRT. Building a defconfig kernel within that guests is about 10% slower when CONFIG_KAISER is enabled. Thank you for testing it! :) Is this expected? It sounds plausible. First, I would expect any form of virtualization to increase the overhead. Second, for the processor (Ivy Bridge), I would have expected even higher performance overheads. KAISER utilizes very recent performance improvements in Intel processors... If it helps I can redo the same test also on bare metal. I'm not sure how we proceed here and if this would help, because I don't know what everyone expects. KAISER definitely introduces an overhead, no doubt about that. How much overhead it is depends on the specific hardware and may be very little on recent architectures and more on older machines. We are not proposing to enable KAISER by default, but to provide the config option to allow easy integration into hardened kernels where performance overheads may be acceptable (which depends on the specific use case and the specific hardware).
Re: [kernel-hardening] [RFC, PATCH] x86_64: KAISER - do not map kernel in user mode
On 2017-05-06 06:02, David Gens wrote: Assuming that their patch indeed leaks per-cpu addresses.. it might not necessarily be required to change it. I think we're not leaking them (unless we still have some bug in our code). The basic idea is that any part that is required for the context switch is at a fixed location (unrelated to the location of code / data / per-cpu data / ...) and thus does not reveal any randomized offsets. Then the attacker cannot gain any knowledge through the side channel anymore. For any attack the attacker could then only use the few KBs of memory that cannot be unmapped because of the way x86 works. Hardening these few KBs seems like an easier task than doing the same for the entire kernel. (The best solution would of course be Intel introducing CR3A and CR3B just like ARM has TTBR0 and TTBR1 - on ARM this entirely prevents any prefetch / double-fault side-channel attacks.)
Re: [kernel-hardening] [RFC, PATCH] x86_64: KAISER - do not map kernel in user mode
On 2017-05-06 06:02, David Gens wrote: Assuming that their patch indeed leaks per-cpu addresses.. it might not necessarily be required to change it. I think we're not leaking them (unless we still have some bug in our code). The basic idea is that any part that is required for the context switch is at a fixed location (unrelated to the location of code / data / per-cpu data / ...) and thus does not reveal any randomized offsets. Then the attacker cannot gain any knowledge through the side channel anymore. For any attack the attacker could then only use the few KBs of memory that cannot be unmapped because of the way x86 works. Hardening these few KBs seems like an easier task than doing the same for the entire kernel. (The best solution would of course be Intel introducing CR3A and CR3B just like ARM has TTBR0 and TTBR1 - on ARM this entirely prevents any prefetch / double-fault side-channel attacks.)
Re: [kernel-hardening] [RFC, PATCH] x86_64: KAISER - do not map kernel in user mode
On 2017-05-05 17:53, Jann Horn wrote: Ah, I think I understand. The kernel stacks are mapped, but cpu_current_top_of_stack isn't, so you can't find the stack until after the CR3 switch in the syscall handler? That's the idea. Only the absolute minimum that is required for a context switch remains mapped (+ it is mapped at an offset which does not depend on KASLR -> we do not leak the KASLR offsets).
Re: [kernel-hardening] [RFC, PATCH] x86_64: KAISER - do not map kernel in user mode
On 2017-05-05 17:53, Jann Horn wrote: Ah, I think I understand. The kernel stacks are mapped, but cpu_current_top_of_stack isn't, so you can't find the stack until after the CR3 switch in the syscall handler? That's the idea. Only the absolute minimum that is required for a context switch remains mapped (+ it is mapped at an offset which does not depend on KASLR -> we do not leak the KASLR offsets).
Re: [kernel-hardening] [RFC, PATCH] x86_64: KAISER - do not map kernel in user mode
On 04.05.2017 17:28, Thomas Garnier wrote: Please read the documentation on submitting patches [1] and coding style [2]. I will have a closer look at that. - How this approach prevent the hardware attacks you mentioned? You still have to keep a part of _text in the pagetable and an attacker could discover it no? (and deduce the kernel base address). These parts are moved to a different section (.user_mapped) which is at a possibly predictable location - the location of the randomized parts of the kernel is independent of the location of .user_mapped. The code/data footprint for .user_mapped is quite small, helping to reduce or eliminate the attack surface... You also need to make it clear that btb attacks are still possible. By just increasing the KASLR randomization range, btb attacks can be mitigated (for free). - What is the perf impact? It will vary for different machines. We have promising results (<1%) for an i7-6700K with representative benchmarks. However, for older systems or for workloads with a lot of pressure on some TLB levels, the performance may be much worse.
Re: [kernel-hardening] [RFC, PATCH] x86_64: KAISER - do not map kernel in user mode
On 04.05.2017 17:28, Thomas Garnier wrote: Please read the documentation on submitting patches [1] and coding style [2]. I will have a closer look at that. - How this approach prevent the hardware attacks you mentioned? You still have to keep a part of _text in the pagetable and an attacker could discover it no? (and deduce the kernel base address). These parts are moved to a different section (.user_mapped) which is at a possibly predictable location - the location of the randomized parts of the kernel is independent of the location of .user_mapped. The code/data footprint for .user_mapped is quite small, helping to reduce or eliminate the attack surface... You also need to make it clear that btb attacks are still possible. By just increasing the KASLR randomization range, btb attacks can be mitigated (for free). - What is the perf impact? It will vary for different machines. We have promising results (<1%) for an i7-6700K with representative benchmarks. However, for older systems or for workloads with a lot of pressure on some TLB levels, the performance may be much worse.
Re: [RFC, PATCH] x86_64: KAISER - do not map kernel in user mode
On 04.05.2017 17:47, Christoph Hellwig wrote: I'll try to read the paper. In the meantime: how different is your approach from then one here? https://lwn.net/Articles/39283/ and how different is the performance impact? The approach sounds very similar, but we have fewer changes because we don't want to change memory allocation but only split the virtual memory - everything can stay where it is. We found that the CR3 switch seems to be significantly improved in modern microarchitectures (we performed our performance tests on a Skylake i7-6700K). We think the TLB maybe uses the full CR3 base address as a tag, relaxing the necessity of flushing the entire TLB upon CR3 updates a bit. Direct runtime overhead is switching the CR3, but that's it. Indirectly, we're potentially increasing the number of TLB entries that are required on one or the other level of the TLB. For TLB-intense tasks this might lead to more significant performance penalties. I'm sure the overhead on older systems is larger than on recent systems.
Re: [RFC, PATCH] x86_64: KAISER - do not map kernel in user mode
On 04.05.2017 17:47, Christoph Hellwig wrote: I'll try to read the paper. In the meantime: how different is your approach from then one here? https://lwn.net/Articles/39283/ and how different is the performance impact? The approach sounds very similar, but we have fewer changes because we don't want to change memory allocation but only split the virtual memory - everything can stay where it is. We found that the CR3 switch seems to be significantly improved in modern microarchitectures (we performed our performance tests on a Skylake i7-6700K). We think the TLB maybe uses the full CR3 base address as a tag, relaxing the necessity of flushing the entire TLB upon CR3 updates a bit. Direct runtime overhead is switching the CR3, but that's it. Indirectly, we're potentially increasing the number of TLB entries that are required on one or the other level of the TLB. For TLB-intense tasks this might lead to more significant performance penalties. I'm sure the overhead on older systems is larger than on recent systems.
Re: [RFC, PATCH] x86_64: KAISER - do not map kernel in user mode
Sorry, missed a file in the first mail (from some code cleanup), the full patch is now attached. Cheers, Daniel >From c4b1831d44c6144d3762ccc72f0c4e71a0c713e5 Mon Sep 17 00:00:00 2001 From: Richard FellnerDate: Thu, 4 May 2017 14:16:44 +0200 Subject: [PATCH] KAISER: Kernel Address Isolation This patch introduces our implementation of KAISER (Kernel Address Isolation to have Side-channels Efficiently Removed), a kernel isolation technique to close hardware side channels on kernel address information. More information about the patch can be found on: https://github.com/IAIK/KAISER --- arch/x86/entry/entry_64.S| 17 arch/x86/entry/entry_64_compat.S | 7 +- arch/x86/include/asm/hw_irq.h| 2 +- arch/x86/include/asm/kaiser.h| 113 +++ arch/x86/include/asm/pgtable.h | 4 + arch/x86/include/asm/pgtable_64.h| 21 + arch/x86/include/asm/pgtable_types.h | 12 ++- arch/x86/include/asm/processor.h | 7 +- arch/x86/kernel/cpu/common.c | 4 +- arch/x86/kernel/espfix_64.c | 6 ++ arch/x86/kernel/head_64.S| 16 +++- arch/x86/kernel/irqinit.c| 2 +- arch/x86/kernel/process.c| 2 +- arch/x86/mm/Makefile | 2 +- arch/x86/mm/kaiser.c | 172 +++ arch/x86/mm/pageattr.c | 2 +- arch/x86/mm/pgtable.c| 28 +- include/asm-generic/vmlinux.lds.h| 11 ++- include/linux/percpu-defs.h | 30 ++ init/main.c | 5 + kernel/fork.c| 8 ++ security/Kconfig | 7 ++ 22 files changed, 461 insertions(+), 17 deletions(-) create mode 100644 arch/x86/include/asm/kaiser.h create mode 100644 arch/x86/mm/kaiser.c diff --git a/arch/x86/entry/entry_64.S b/arch/x86/entry/entry_64.S index 044d18e..631c7bf 100644 --- a/arch/x86/entry/entry_64.S +++ b/arch/x86/entry/entry_64.S @@ -37,6 +37,7 @@ #include #include #include +#include #include .code64 @@ -141,6 +142,7 @@ ENTRY(entry_SYSCALL_64) * it is too small to ever cause noticeable irq latency. */ SWAPGS_UNSAFE_STACK + SWITCH_KERNEL_CR3_NO_STACK /* * A hypervisor implementation might want to use a label * after the swapgs, so that it can do the swapgs @@ -223,6 +225,7 @@ entry_SYSCALL_64_fastpath: movq RIP(%rsp), %rcx movq EFLAGS(%rsp), %r11 RESTORE_C_REGS_EXCEPT_RCX_R11 + SWITCH_USER_CR3 movq RSP(%rsp), %rsp USERGS_SYSRET64 @@ -318,10 +321,12 @@ return_from_SYSCALL_64: syscall_return_via_sysret: /* rcx and r11 are already restored (see code above) */ RESTORE_C_REGS_EXCEPT_RCX_R11 + SWITCH_USER_CR3 movq RSP(%rsp), %rsp USERGS_SYSRET64 opportunistic_sysret_failed: + SWITCH_USER_CR3 SWAPGS jmp restore_c_regs_and_iret END(entry_SYSCALL_64) @@ -420,6 +425,7 @@ ENTRY(ret_from_fork) leaq FRAME_OFFSET(%rsp),%rdi /* pt_regs pointer */ call syscall_return_slowpath /* returns with IRQs disabled */ TRACE_IRQS_ON /* user mode is traced as IRQS on */ + SWITCH_USER_CR3 SWAPGS FRAME_END jmp restore_regs_and_iret @@ -476,6 +482,7 @@ END(irq_entries_start) * tracking that we're in kernel mode. */ SWAPGS + SWITCH_KERNEL_CR3 /* * We need to tell lockdep that IRQs are off. We can't do this until @@ -533,6 +540,7 @@ GLOBAL(retint_user) mov %rsp,%rdi call prepare_exit_to_usermode TRACE_IRQS_IRETQ + SWITCH_USER_CR3 SWAPGS jmp restore_regs_and_iret @@ -610,6 +618,7 @@ native_irq_return_ldt: pushq %rdi/* Stash user RDI */ SWAPGS + SWITCH_KERNEL_CR3 movq PER_CPU_VAR(espfix_waddr), %rdi movq %rax, (0*8)(%rdi) /* user RAX */ movq (1*8)(%rsp), %rax /* user RIP */ @@ -636,6 +645,7 @@ native_irq_return_ldt: * still points to an RO alias of the ESPFIX stack. */ orq PER_CPU_VAR(espfix_stack), %rax + SWITCH_USER_CR3 SWAPGS movq %rax, %rsp @@ -1034,6 +1044,7 @@ ENTRY(paranoid_entry) testl %edx, %edx js 1f/* negative -> in kernel */ SWAPGS + SWITCH_KERNEL_CR3 xorl %ebx, %ebx 1: ret END(paranoid_entry) @@ -1056,6 +1067,7 @@ ENTRY(paranoid_exit) testl %ebx, %ebx /* swapgs needed? */ jnz paranoid_exit_no_swapgs TRACE_IRQS_IRETQ + SWITCH_USER_CR3_NO_STACK SWAPGS_UNSAFE_STACK jmp paranoid_exit_restore paranoid_exit_no_swapgs: @@ -1085,6 +1097,7 @@ ENTRY(error_entry) * from user mode due to an IRET fault. */ SWAPGS + SWITCH_KERNEL_CR3 .Lerror_entry_from_usermode_after_swapgs: /* @@ -1136,6 +1149,7 @@ ENTRY(error_entry) * Switch to kernel gsbase: */ SWAPGS + SWITCH_KERNEL_CR3 /* * Pretend that the exception came from user mode: set up pt_regs @@ -1234,6 +1248,7 @@ ENTRY(nmi) */ SWAPGS_UNSAFE_STACK + SWITCH_KERNEL_CR3_NO_STACK cld movq %rsp, %rdx movq PER_CPU_VAR(cpu_current_top_of_stack), %rsp @@ -1274,6 +1289,7 @@ ENTRY(nmi) * Return back to user mode. We must *not*
Re: [RFC, PATCH] x86_64: KAISER - do not map kernel in user mode
Sorry, missed a file in the first mail (from some code cleanup), the full patch is now attached. Cheers, Daniel >From c4b1831d44c6144d3762ccc72f0c4e71a0c713e5 Mon Sep 17 00:00:00 2001 From: Richard Fellner Date: Thu, 4 May 2017 14:16:44 +0200 Subject: [PATCH] KAISER: Kernel Address Isolation This patch introduces our implementation of KAISER (Kernel Address Isolation to have Side-channels Efficiently Removed), a kernel isolation technique to close hardware side channels on kernel address information. More information about the patch can be found on: https://github.com/IAIK/KAISER --- arch/x86/entry/entry_64.S| 17 arch/x86/entry/entry_64_compat.S | 7 +- arch/x86/include/asm/hw_irq.h| 2 +- arch/x86/include/asm/kaiser.h| 113 +++ arch/x86/include/asm/pgtable.h | 4 + arch/x86/include/asm/pgtable_64.h| 21 + arch/x86/include/asm/pgtable_types.h | 12 ++- arch/x86/include/asm/processor.h | 7 +- arch/x86/kernel/cpu/common.c | 4 +- arch/x86/kernel/espfix_64.c | 6 ++ arch/x86/kernel/head_64.S| 16 +++- arch/x86/kernel/irqinit.c| 2 +- arch/x86/kernel/process.c| 2 +- arch/x86/mm/Makefile | 2 +- arch/x86/mm/kaiser.c | 172 +++ arch/x86/mm/pageattr.c | 2 +- arch/x86/mm/pgtable.c| 28 +- include/asm-generic/vmlinux.lds.h| 11 ++- include/linux/percpu-defs.h | 30 ++ init/main.c | 5 + kernel/fork.c| 8 ++ security/Kconfig | 7 ++ 22 files changed, 461 insertions(+), 17 deletions(-) create mode 100644 arch/x86/include/asm/kaiser.h create mode 100644 arch/x86/mm/kaiser.c diff --git a/arch/x86/entry/entry_64.S b/arch/x86/entry/entry_64.S index 044d18e..631c7bf 100644 --- a/arch/x86/entry/entry_64.S +++ b/arch/x86/entry/entry_64.S @@ -37,6 +37,7 @@ #include #include #include +#include #include .code64 @@ -141,6 +142,7 @@ ENTRY(entry_SYSCALL_64) * it is too small to ever cause noticeable irq latency. */ SWAPGS_UNSAFE_STACK + SWITCH_KERNEL_CR3_NO_STACK /* * A hypervisor implementation might want to use a label * after the swapgs, so that it can do the swapgs @@ -223,6 +225,7 @@ entry_SYSCALL_64_fastpath: movq RIP(%rsp), %rcx movq EFLAGS(%rsp), %r11 RESTORE_C_REGS_EXCEPT_RCX_R11 + SWITCH_USER_CR3 movq RSP(%rsp), %rsp USERGS_SYSRET64 @@ -318,10 +321,12 @@ return_from_SYSCALL_64: syscall_return_via_sysret: /* rcx and r11 are already restored (see code above) */ RESTORE_C_REGS_EXCEPT_RCX_R11 + SWITCH_USER_CR3 movq RSP(%rsp), %rsp USERGS_SYSRET64 opportunistic_sysret_failed: + SWITCH_USER_CR3 SWAPGS jmp restore_c_regs_and_iret END(entry_SYSCALL_64) @@ -420,6 +425,7 @@ ENTRY(ret_from_fork) leaq FRAME_OFFSET(%rsp),%rdi /* pt_regs pointer */ call syscall_return_slowpath /* returns with IRQs disabled */ TRACE_IRQS_ON /* user mode is traced as IRQS on */ + SWITCH_USER_CR3 SWAPGS FRAME_END jmp restore_regs_and_iret @@ -476,6 +482,7 @@ END(irq_entries_start) * tracking that we're in kernel mode. */ SWAPGS + SWITCH_KERNEL_CR3 /* * We need to tell lockdep that IRQs are off. We can't do this until @@ -533,6 +540,7 @@ GLOBAL(retint_user) mov %rsp,%rdi call prepare_exit_to_usermode TRACE_IRQS_IRETQ + SWITCH_USER_CR3 SWAPGS jmp restore_regs_and_iret @@ -610,6 +618,7 @@ native_irq_return_ldt: pushq %rdi/* Stash user RDI */ SWAPGS + SWITCH_KERNEL_CR3 movq PER_CPU_VAR(espfix_waddr), %rdi movq %rax, (0*8)(%rdi) /* user RAX */ movq (1*8)(%rsp), %rax /* user RIP */ @@ -636,6 +645,7 @@ native_irq_return_ldt: * still points to an RO alias of the ESPFIX stack. */ orq PER_CPU_VAR(espfix_stack), %rax + SWITCH_USER_CR3 SWAPGS movq %rax, %rsp @@ -1034,6 +1044,7 @@ ENTRY(paranoid_entry) testl %edx, %edx js 1f/* negative -> in kernel */ SWAPGS + SWITCH_KERNEL_CR3 xorl %ebx, %ebx 1: ret END(paranoid_entry) @@ -1056,6 +1067,7 @@ ENTRY(paranoid_exit) testl %ebx, %ebx /* swapgs needed? */ jnz paranoid_exit_no_swapgs TRACE_IRQS_IRETQ + SWITCH_USER_CR3_NO_STACK SWAPGS_UNSAFE_STACK jmp paranoid_exit_restore paranoid_exit_no_swapgs: @@ -1085,6 +1097,7 @@ ENTRY(error_entry) * from user mode due to an IRET fault. */ SWAPGS + SWITCH_KERNEL_CR3 .Lerror_entry_from_usermode_after_swapgs: /* @@ -1136,6 +1149,7 @@ ENTRY(error_entry) * Switch to kernel gsbase: */ SWAPGS + SWITCH_KERNEL_CR3 /* * Pretend that the exception came from user mode: set up pt_regs @@ -1234,6 +1248,7 @@ ENTRY(nmi) */ SWAPGS_UNSAFE_STACK + SWITCH_KERNEL_CR3_NO_STACK cld movq %rsp, %rdx movq PER_CPU_VAR(cpu_current_top_of_stack), %rsp @@ -1274,6 +1289,7 @@ ENTRY(nmi) * Return back to user mode. We must *not* do the normal exit * work,
[RFC, PATCH] x86_64: KAISER - do not map kernel in user mode
After several recent works [1,2,3] KASLR on x86_64 was basically considered dead by many researchers. We have been working on an efficient but effective fix for this problem and found that not mapping the kernel space when running in user mode is the solution to this problem [4] (the corresponding paper [5] will be presented at ESSoS17). With this RFC patch we allow anybody to configure their kernel with the flag CONFIG_KAISER to add our defense mechanism. If there are any questions we would love to answer them. We also appreciate any comments! Cheers, Daniel (+ the KAISER team from Graz University of Technology) [1] http://www.ieee-security.org/TC/SP2013/papers/4977a191.pdf [2] https://www.blackhat.com/docs/us-16/materials/us-16-Fogh-Using-Undocumented-CPU-Behaviour-To-See-Into-Kernel-Mode-And-Break-KASLR-In-The-Process.pdf [3] https://www.blackhat.com/docs/us-16/materials/us-16-Jang-Breaking-Kernel-Address-Space-Layout-Randomization-KASLR-With-Intel-TSX.pdf [4] https://github.com/IAIK/KAISER [5] https://gruss.cc/files/kaiser.pdf >From 03c413bc52f1ac253cf0f067605f367f3390d3f4 Mon Sep 17 00:00:00 2001 From: Richard FellnerDate: Thu, 4 May 2017 10:44:38 +0200 Subject: [PATCH] KAISER: Kernel Address Isolation This patch introduces our implementation of KAISER (Kernel Address Isolation to have Side-channels Efficiently Removed), a kernel isolation technique to close hardware side channels on kernel address information. More information about the patch can be found on: https://github.com/IAIK/KAISER --- arch/x86/entry/entry_64.S| 17 + arch/x86/entry/entry_64_compat.S | 7 ++- arch/x86/include/asm/hw_irq.h| 2 +- arch/x86/include/asm/pgtable.h | 4 arch/x86/include/asm/pgtable_64.h| 21 + arch/x86/include/asm/pgtable_types.h | 12 ++-- arch/x86/include/asm/processor.h | 7 ++- arch/x86/kernel/cpu/common.c | 4 ++-- arch/x86/kernel/espfix_64.c | 6 ++ arch/x86/kernel/head_64.S| 16 arch/x86/kernel/irqinit.c| 2 +- arch/x86/kernel/process.c| 2 +- arch/x86/mm/Makefile | 2 +- arch/x86/mm/pageattr.c | 2 +- arch/x86/mm/pgtable.c| 28 +++- include/asm-generic/vmlinux.lds.h| 11 ++- include/linux/percpu-defs.h | 30 ++ init/main.c | 5 + kernel/fork.c| 8 security/Kconfig | 7 +++ 20 files changed, 176 insertions(+), 17 deletions(-) diff --git a/arch/x86/entry/entry_64.S b/arch/x86/entry/entry_64.S index 044d18e..631c7bf 100644 --- a/arch/x86/entry/entry_64.S +++ b/arch/x86/entry/entry_64.S @@ -37,6 +37,7 @@ #include #include #include +#include #include .code64 @@ -141,6 +142,7 @@ ENTRY(entry_SYSCALL_64) * it is too small to ever cause noticeable irq latency. */ SWAPGS_UNSAFE_STACK + SWITCH_KERNEL_CR3_NO_STACK /* * A hypervisor implementation might want to use a label * after the swapgs, so that it can do the swapgs @@ -223,6 +225,7 @@ entry_SYSCALL_64_fastpath: movq RIP(%rsp), %rcx movq EFLAGS(%rsp), %r11 RESTORE_C_REGS_EXCEPT_RCX_R11 + SWITCH_USER_CR3 movq RSP(%rsp), %rsp USERGS_SYSRET64 @@ -318,10 +321,12 @@ return_from_SYSCALL_64: syscall_return_via_sysret: /* rcx and r11 are already restored (see code above) */ RESTORE_C_REGS_EXCEPT_RCX_R11 + SWITCH_USER_CR3 movq RSP(%rsp), %rsp USERGS_SYSRET64 opportunistic_sysret_failed: + SWITCH_USER_CR3 SWAPGS jmp restore_c_regs_and_iret END(entry_SYSCALL_64) @@ -420,6 +425,7 @@ ENTRY(ret_from_fork) leaq FRAME_OFFSET(%rsp),%rdi /* pt_regs pointer */ call syscall_return_slowpath /* returns with IRQs disabled */ TRACE_IRQS_ON /* user mode is traced as IRQS on */ + SWITCH_USER_CR3 SWAPGS FRAME_END jmp restore_regs_and_iret @@ -476,6 +482,7 @@ END(irq_entries_start) * tracking that we're in kernel mode. */ SWAPGS + SWITCH_KERNEL_CR3 /* * We need to tell lockdep that IRQs are off. We can't do this until @@ -533,6 +540,7 @@ GLOBAL(retint_user) mov %rsp,%rdi call prepare_exit_to_usermode TRACE_IRQS_IRETQ + SWITCH_USER_CR3 SWAPGS jmp restore_regs_and_iret @@ -610,6 +618,7 @@ native_irq_return_ldt: pushq %rdi/* Stash user RDI */ SWAPGS + SWITCH_KERNEL_CR3 movq PER_CPU_VAR(espfix_waddr), %rdi movq %rax, (0*8)(%rdi) /* user RAX */ movq (1*8)(%rsp), %rax /* user RIP */ @@ -636,6 +645,7 @@ native_irq_return_ldt: * still points to an RO alias of the ESPFIX stack. */ orq PER_CPU_VAR(espfix_stack), %rax + SWITCH_USER_CR3 SWAPGS movq %rax, %rsp @@ -1034,6 +1044,7 @@ ENTRY(paranoid_entry) testl %edx, %edx js 1f/* negative -> in kernel */ SWAPGS + SWITCH_KERNEL_CR3 xorl %ebx, %ebx 1: ret END(paranoid_entry) @@
[RFC, PATCH] x86_64: KAISER - do not map kernel in user mode
After several recent works [1,2,3] KASLR on x86_64 was basically considered dead by many researchers. We have been working on an efficient but effective fix for this problem and found that not mapping the kernel space when running in user mode is the solution to this problem [4] (the corresponding paper [5] will be presented at ESSoS17). With this RFC patch we allow anybody to configure their kernel with the flag CONFIG_KAISER to add our defense mechanism. If there are any questions we would love to answer them. We also appreciate any comments! Cheers, Daniel (+ the KAISER team from Graz University of Technology) [1] http://www.ieee-security.org/TC/SP2013/papers/4977a191.pdf [2] https://www.blackhat.com/docs/us-16/materials/us-16-Fogh-Using-Undocumented-CPU-Behaviour-To-See-Into-Kernel-Mode-And-Break-KASLR-In-The-Process.pdf [3] https://www.blackhat.com/docs/us-16/materials/us-16-Jang-Breaking-Kernel-Address-Space-Layout-Randomization-KASLR-With-Intel-TSX.pdf [4] https://github.com/IAIK/KAISER [5] https://gruss.cc/files/kaiser.pdf >From 03c413bc52f1ac253cf0f067605f367f3390d3f4 Mon Sep 17 00:00:00 2001 From: Richard Fellner Date: Thu, 4 May 2017 10:44:38 +0200 Subject: [PATCH] KAISER: Kernel Address Isolation This patch introduces our implementation of KAISER (Kernel Address Isolation to have Side-channels Efficiently Removed), a kernel isolation technique to close hardware side channels on kernel address information. More information about the patch can be found on: https://github.com/IAIK/KAISER --- arch/x86/entry/entry_64.S| 17 + arch/x86/entry/entry_64_compat.S | 7 ++- arch/x86/include/asm/hw_irq.h| 2 +- arch/x86/include/asm/pgtable.h | 4 arch/x86/include/asm/pgtable_64.h| 21 + arch/x86/include/asm/pgtable_types.h | 12 ++-- arch/x86/include/asm/processor.h | 7 ++- arch/x86/kernel/cpu/common.c | 4 ++-- arch/x86/kernel/espfix_64.c | 6 ++ arch/x86/kernel/head_64.S| 16 arch/x86/kernel/irqinit.c| 2 +- arch/x86/kernel/process.c| 2 +- arch/x86/mm/Makefile | 2 +- arch/x86/mm/pageattr.c | 2 +- arch/x86/mm/pgtable.c| 28 +++- include/asm-generic/vmlinux.lds.h| 11 ++- include/linux/percpu-defs.h | 30 ++ init/main.c | 5 + kernel/fork.c| 8 security/Kconfig | 7 +++ 20 files changed, 176 insertions(+), 17 deletions(-) diff --git a/arch/x86/entry/entry_64.S b/arch/x86/entry/entry_64.S index 044d18e..631c7bf 100644 --- a/arch/x86/entry/entry_64.S +++ b/arch/x86/entry/entry_64.S @@ -37,6 +37,7 @@ #include #include #include +#include #include .code64 @@ -141,6 +142,7 @@ ENTRY(entry_SYSCALL_64) * it is too small to ever cause noticeable irq latency. */ SWAPGS_UNSAFE_STACK + SWITCH_KERNEL_CR3_NO_STACK /* * A hypervisor implementation might want to use a label * after the swapgs, so that it can do the swapgs @@ -223,6 +225,7 @@ entry_SYSCALL_64_fastpath: movq RIP(%rsp), %rcx movq EFLAGS(%rsp), %r11 RESTORE_C_REGS_EXCEPT_RCX_R11 + SWITCH_USER_CR3 movq RSP(%rsp), %rsp USERGS_SYSRET64 @@ -318,10 +321,12 @@ return_from_SYSCALL_64: syscall_return_via_sysret: /* rcx and r11 are already restored (see code above) */ RESTORE_C_REGS_EXCEPT_RCX_R11 + SWITCH_USER_CR3 movq RSP(%rsp), %rsp USERGS_SYSRET64 opportunistic_sysret_failed: + SWITCH_USER_CR3 SWAPGS jmp restore_c_regs_and_iret END(entry_SYSCALL_64) @@ -420,6 +425,7 @@ ENTRY(ret_from_fork) leaq FRAME_OFFSET(%rsp),%rdi /* pt_regs pointer */ call syscall_return_slowpath /* returns with IRQs disabled */ TRACE_IRQS_ON /* user mode is traced as IRQS on */ + SWITCH_USER_CR3 SWAPGS FRAME_END jmp restore_regs_and_iret @@ -476,6 +482,7 @@ END(irq_entries_start) * tracking that we're in kernel mode. */ SWAPGS + SWITCH_KERNEL_CR3 /* * We need to tell lockdep that IRQs are off. We can't do this until @@ -533,6 +540,7 @@ GLOBAL(retint_user) mov %rsp,%rdi call prepare_exit_to_usermode TRACE_IRQS_IRETQ + SWITCH_USER_CR3 SWAPGS jmp restore_regs_and_iret @@ -610,6 +618,7 @@ native_irq_return_ldt: pushq %rdi/* Stash user RDI */ SWAPGS + SWITCH_KERNEL_CR3 movq PER_CPU_VAR(espfix_waddr), %rdi movq %rax, (0*8)(%rdi) /* user RAX */ movq (1*8)(%rsp), %rax /* user RIP */ @@ -636,6 +645,7 @@ native_irq_return_ldt: * still points to an RO alias of the ESPFIX stack. */ orq PER_CPU_VAR(espfix_stack), %rax + SWITCH_USER_CR3 SWAPGS movq %rax, %rsp @@ -1034,6 +1044,7 @@ ENTRY(paranoid_entry) testl %edx, %edx js 1f/* negative -> in kernel */ SWAPGS + SWITCH_KERNEL_CR3 xorl %ebx, %ebx 1: ret END(paranoid_entry) @@ -1056,6 +1067,7 @@
Re: [kernel-hardening] rowhammer protection [was Re: Getting interrupt every million cache misses]
On 01.11.2016 09:10, Pavel Machek wrote: cpu family : 6 model: 23 model name : Intel(R) Core(TM)2 Duo CPU E7400 @ 2.80GHz stepping : 10 microcode: 0xa07 so rowhammerjs/native is not available for this system. Bit mapping for memory hash functions would need to be reverse engineered for more effective attack. By coincidence, we wrote a tool to do that in software: https://github.com/IAIK/drama ;)
Re: [kernel-hardening] rowhammer protection [was Re: Getting interrupt every million cache misses]
On 01.11.2016 09:10, Pavel Machek wrote: cpu family : 6 model: 23 model name : Intel(R) Core(TM)2 Duo CPU E7400 @ 2.80GHz stepping : 10 microcode: 0xa07 so rowhammerjs/native is not available for this system. Bit mapping for memory hash functions would need to be reverse engineered for more effective attack. By coincidence, we wrote a tool to do that in software: https://github.com/IAIK/drama ;)
Re: [kernel-hardening] rowhammer protection [was Re: Getting interrupt every million cache misses]
On 01.11.2016 07:33, Ingo Molnar wrote: Can you suggest a method to find heavily rowhammer affected hardware? Only by testing it, or are there some chipset IDs ranges or dmidecode info that will pinpoint potentially affected machines? I have worked with many different systems both on running rowhammer attacks and testing defense mechanisms. So far, every Ivy Bridge i5 (DDR3) that I had access to was susceptible to bit flips - you will have highest chances with Ivy Bridge i5...
Re: [kernel-hardening] rowhammer protection [was Re: Getting interrupt every million cache misses]
On 01.11.2016 07:33, Ingo Molnar wrote: Can you suggest a method to find heavily rowhammer affected hardware? Only by testing it, or are there some chipset IDs ranges or dmidecode info that will pinpoint potentially affected machines? I have worked with many different systems both on running rowhammer attacks and testing defense mechanisms. So far, every Ivy Bridge i5 (DDR3) that I had access to was susceptible to bit flips - you will have highest chances with Ivy Bridge i5...
Re: [kernel-hardening] rowhammer protection [was Re: Getting interrupt every million cache misses]
On 30.10.2016 00:01, Pavel Machek wrote: Hmm, maybe I'm glad I don't have a new machine :-). I assume you still get _some_ bitflips with generic "rowhammer"? 1 or 2 every 20-30 minutes...
Re: [kernel-hardening] rowhammer protection [was Re: Getting interrupt every million cache misses]
On 30.10.2016 00:01, Pavel Machek wrote: Hmm, maybe I'm glad I don't have a new machine :-). I assume you still get _some_ bitflips with generic "rowhammer"? 1 or 2 every 20-30 minutes...
Re: [kernel-hardening] rowhammer protection [was Re: Getting interrupt every million cache misses]
On 29.10.2016 23:45, Pavel Machek wrote: indy/sandy/haswell/skylake, so I'll just use the generic version...?) yes, generic might work, but i never tested it on anything that old... on my system i have >30 bit flips per second (ivy bridge i5-3xxx) with the rowhammer-ivy test... sometimes even more than 100 per second...
Re: [kernel-hardening] rowhammer protection [was Re: Getting interrupt every million cache misses]
On 29.10.2016 23:45, Pavel Machek wrote: indy/sandy/haswell/skylake, so I'll just use the generic version...?) yes, generic might work, but i never tested it on anything that old... on my system i have >30 bit flips per second (ivy bridge i5-3xxx) with the rowhammer-ivy test... sometimes even more than 100 per second...
Re: [kernel-hardening] rowhammer protection [was Re: Getting interrupt every million cache misses]
On 29.10.2016 23:05, Pavel Machek wrote: So far I did bzip2 and kernel compilation. I believe I can prevent flips in rowhammer-test with bzip2 going from 4 seconds to 5 seconds... let me see. can you prevent bitflips in this one? https://github.com/IAIK/rowhammerjs/tree/master/native Ok, let me try that. Problem is that the machine I'm testing on takes 20 minutes to produce bit flip... will be lots faster with my code above ;)
Re: [kernel-hardening] rowhammer protection [was Re: Getting interrupt every million cache misses]
On 29.10.2016 23:05, Pavel Machek wrote: So far I did bzip2 and kernel compilation. I believe I can prevent flips in rowhammer-test with bzip2 going from 4 seconds to 5 seconds... let me see. can you prevent bitflips in this one? https://github.com/IAIK/rowhammerjs/tree/master/native Ok, let me try that. Problem is that the machine I'm testing on takes 20 minutes to produce bit flip... will be lots faster with my code above ;)
Re: [kernel-hardening] rowhammer protection [was Re: Getting interrupt every million cache misses]
On 29.10.2016 21:42, Pavel Machek wrote: Congratulations. Now I'd like to take away your toys :-). I'm would like you to do that, but I'm very confident you're not successful the way your starting ;) Not in my testing. Have you tried music/video reencoding? Games? Anything that works with a decent amount of memory but not too much hard disk i/o? Numbers are very clear there... First, I'm not at all sure lowest CPU speed would make any difference at all It would. I've seen many bitflips but none where the CPU operated in the lower frequency range. Second, going to lowest clock speed will reduce performance As does the countermeasure you propose... No, sorry, not going to play this particular whack-a-mole game. But you are already with the countermeasure you propose... Linux is designed for working hardware, and with bit flips, something is going to break. (Does Flip Feng Shui really depend on dedup?) Deduplication should be disabled not because of bit flips but because of information leakage (deduplication attacks, cache side-channel attacks, ...) Yes, Flip Feng Shui requires deduplication and does not work without. Disabling deduplication is what the authors recommend as a countermeasure. But it will be nowhere near complete fix, right? Will fix user attacking kernel, but not user1 attacking user2. You could put each "user" into separate 2MB region, but then you'd have to track who needs go go where. (Same uid is not enough, probably "can ptrace"?) Exactly. But preventing user2kernel is already a good start, and you would prevent that without any doubt and without any cost. user2user is something else to think about and more complicated because you have shared libraries + copy on write --> same problems as deduplication. I think it might make sense to discuss whether separating by uids or even pids would be viable. That'll still let remote server gain permissons of local user running web server... using javascript exploit right? And that's actually attack that I find most scary. Local user to root exploit is bad, but getting permissions of web browser from remote web server is very, very, very bad. Rowhammer.js skips the browser... it goes JS to full phys. memory access. Anyway, preventing Rowhammer from JS should be easy because even the slightest slow down should be enough to prevent any Rowhammer attack from JS. That is a simple fix that does not cost any runtime performance. Simple? Not really, I'm afraid. Feel free to try to implement it. I had a student who already implemented this in another OS, I'm confident it can be done in Linux as well... Cheers, Daniel
Re: [kernel-hardening] rowhammer protection [was Re: Getting interrupt every million cache misses]
On 29.10.2016 21:42, Pavel Machek wrote: Congratulations. Now I'd like to take away your toys :-). I'm would like you to do that, but I'm very confident you're not successful the way your starting ;) Not in my testing. Have you tried music/video reencoding? Games? Anything that works with a decent amount of memory but not too much hard disk i/o? Numbers are very clear there... First, I'm not at all sure lowest CPU speed would make any difference at all It would. I've seen many bitflips but none where the CPU operated in the lower frequency range. Second, going to lowest clock speed will reduce performance As does the countermeasure you propose... No, sorry, not going to play this particular whack-a-mole game. But you are already with the countermeasure you propose... Linux is designed for working hardware, and with bit flips, something is going to break. (Does Flip Feng Shui really depend on dedup?) Deduplication should be disabled not because of bit flips but because of information leakage (deduplication attacks, cache side-channel attacks, ...) Yes, Flip Feng Shui requires deduplication and does not work without. Disabling deduplication is what the authors recommend as a countermeasure. But it will be nowhere near complete fix, right? Will fix user attacking kernel, but not user1 attacking user2. You could put each "user" into separate 2MB region, but then you'd have to track who needs go go where. (Same uid is not enough, probably "can ptrace"?) Exactly. But preventing user2kernel is already a good start, and you would prevent that without any doubt and without any cost. user2user is something else to think about and more complicated because you have shared libraries + copy on write --> same problems as deduplication. I think it might make sense to discuss whether separating by uids or even pids would be viable. That'll still let remote server gain permissons of local user running web server... using javascript exploit right? And that's actually attack that I find most scary. Local user to root exploit is bad, but getting permissions of web browser from remote web server is very, very, very bad. Rowhammer.js skips the browser... it goes JS to full phys. memory access. Anyway, preventing Rowhammer from JS should be easy because even the slightest slow down should be enough to prevent any Rowhammer attack from JS. That is a simple fix that does not cost any runtime performance. Simple? Not really, I'm afraid. Feel free to try to implement it. I had a student who already implemented this in another OS, I'm confident it can be done in Linux as well... Cheers, Daniel
Re: Re: [kernel-hardening] rowhammer protection [was Re: Getting interrupt every million cache misses]
I think that this idea to mitigate Rowhammer is not a good approach. I wrote Rowhammer.js (we published a paper on that) and I had the first reproducible bit flips on DDR4 at both, increased and default refresh rates (published in our DRAMA paper). We have researched the number of cache misses induced from different applications in the past and there are many applications that cause more cache misses than Rowhammer (published in our Flush+Flush paper) they just cause them on different rows. Slowing down a system surely works, but you could also, as a mitigation just make this CPU core run at the lowest possible frequency. That would likely be more effective than the solution you suggest. Now, every Rowhammer attack exploits not only the DRAM effects but also the way the operating system organizes memory. Some papers exploit page deduplication and disabling page deduplication should be the default also for other reasons, such as information disclosure attacks. If page deduplication is disabled, attacks like Dedup est Machina and Flip Feng Shui are inherently not possible anymore. Most other attacks target page tables (the Google exploit, Rowhammer.js, Drammer). Now in Rowhammer.js we suggested a very simple fix, that is just an extension of what Linux already does. Unless out of memory page tables and user pages are not placed in the same 2MB region. We suggested that this behavior should be more strict even in memory pressure situations. If the OS can only find a page table that resides in the same 2MB region as a user page, the request should fail instead and the process requesting it should go out of memory. More generally, the attack surface is gone if the OS never places a page table in proximity of less than 2MB to a user page. That is a simple fix that does not cost any runtime performance. It mitigates all these scary attacks and won't even incur a memory cost in most situation.
Re: Re: [kernel-hardening] rowhammer protection [was Re: Getting interrupt every million cache misses]
I think that this idea to mitigate Rowhammer is not a good approach. I wrote Rowhammer.js (we published a paper on that) and I had the first reproducible bit flips on DDR4 at both, increased and default refresh rates (published in our DRAMA paper). We have researched the number of cache misses induced from different applications in the past and there are many applications that cause more cache misses than Rowhammer (published in our Flush+Flush paper) they just cause them on different rows. Slowing down a system surely works, but you could also, as a mitigation just make this CPU core run at the lowest possible frequency. That would likely be more effective than the solution you suggest. Now, every Rowhammer attack exploits not only the DRAM effects but also the way the operating system organizes memory. Some papers exploit page deduplication and disabling page deduplication should be the default also for other reasons, such as information disclosure attacks. If page deduplication is disabled, attacks like Dedup est Machina and Flip Feng Shui are inherently not possible anymore. Most other attacks target page tables (the Google exploit, Rowhammer.js, Drammer). Now in Rowhammer.js we suggested a very simple fix, that is just an extension of what Linux already does. Unless out of memory page tables and user pages are not placed in the same 2MB region. We suggested that this behavior should be more strict even in memory pressure situations. If the OS can only find a page table that resides in the same 2MB region as a user page, the request should fail instead and the process requesting it should go out of memory. More generally, the attack surface is gone if the OS never places a page table in proximity of less than 2MB to a user page. That is a simple fix that does not cost any runtime performance. It mitigates all these scary attacks and won't even incur a memory cost in most situation.