Skip to contentSkip to conversation

mm: add MAP_EXCLUSIVE to create exclusive user mappings

RFC PATCH In review Mike Rapoport 21 participants 61 messages Oct 27, 2019 – Dec 8, 2019

Subsystems (17)
Memory management x86 BPF MM core +13 more subsystems
lore
Share & tools
lore

Mike Rapoport

author maintainer
#0 lore raw

RFC PATCH 1/1 mm: add MAP_EXCLUSIVE to create exclusive user mappings

From: Mike Rapoport <[email protected]>

Hi,

The patch below aims to allow applications to create mappins that have pages visible only to the owning process. Such mappings could be used to store secrets so that these secrets are not visible neither to other processes nor to the kernel.

I've only tested the basic functionality, the changes should be verified against THP/migration/compaction. Yet, I'd appreciate early feedback.

Mike Rapoport (1):
mm: add MAP_EXCLUSIVE to create exclusive user mappings

 arch/x86/mm/fault.c                    | 14 ++++++++++
 fs/proc/task_mmu.c                     |  1 +
 include/linux/mm.h                     |  9 +++++++
 include/linux/page-flags.h             |  7 +++++
 include/linux/page_excl.h              | 49 ++++++++++++++++++++++++++++++++++
 include/trace/events/mmflags.h         |  9 ++++++-
 include/uapi/asm-generic/mman-common.h |  1 +
 kernel/fork.c                          |  3 ++-
 mm/Kconfig                             |  3 +++
 mm/gup.c                               |  8 ++++++
 mm/memory.c                            |  3 +++
 mm/mmap.c                              | 16 +++++++++++
 mm/page_alloc.c                        |  5 ++++
 13 files changed, 126 insertions(+), 2 deletions(-)
 create mode 100644 include/linux/page_excl.h
Signature · 1 line
2.7.4
author maintainer
#1 lore raw

RFC PATCH 1/1 mm: add MAP_EXCLUSIVE to create exclusive user mappings

From: Mike Rapoport <[email protected]>

The mappings created with MAP_EXCLUSIVE are visible only in the context of the owning process and can be used by applications to store secret information that will not be visible not only to other processes but to the kernel as well.

The pages in these mappings are removed from the kernel direct map and marked with PG_user_exclusive flag. When the exclusive area is unmapped, the pages are mapped back into the direct map.

The MAP_EXCLUSIVE flag implies MAP_POPULATE and MAP_LOCKED.

---
 arch/x86/mm/fault.c                    | 14 ++++++++++
 fs/proc/task_mmu.c                     |  1 +
 include/linux/mm.h                     |  9 +++++++
 include/linux/page-flags.h             |  7 +++++
 include/linux/page_excl.h              | 49 ++++++++++++++++++++++++++++++++++
 include/trace/events/mmflags.h         |  9 ++++++-
 include/uapi/asm-generic/mman-common.h |  1 +
 kernel/fork.c                          |  3 ++-
 mm/Kconfig                             |  3 +++
 mm/gup.c                               |  8 ++++++
 mm/memory.c                            |  3 +++
 mm/mmap.c                              | 16 +++++++++++
 mm/page_alloc.c                        |  5 ++++
 13 files changed, 126 insertions(+), 2 deletions(-)
 create mode 100644 include/linux/page_excl.h
arch/x86/mm/fault.c +14 -0
index 9ceacd1..8f73a75 100644
@@ -17,6 +17,7 @@
 #include <linux/context_tracking.h>	/* exception_enter(), ...	*/
 #include <linux/uaccess.h>		/* faulthandler_disabled()	*/
 #include <linux/efi.h>			/* efi_recover_from_page_fault()*/
+#include <linux/page_excl.h>		/* page_is_user_exclusive()	*/
 #include <linux/mm_types.h>
 
 #include <asm/cpufeature.h>		/* boot_cpu_has, ...		*/
@@ -1218,6 +1219,13 @@ static int fault_in_kernel_space(unsigned long address)
 	return address >= TASK_SIZE_MAX;
 }
 
+static bool fault_in_user_exclusive_page(unsigned long address)
+{
+	struct page *page = virt_to_page(address);
+
+	return page_is_user_exclusive(page);
+}
+
 /*
  * Called for all faults where 'address' is part of the kernel address
  * space.  Might get called for faults that originate from *code* that
@@ -1261,6 +1269,12 @@ do_kern_addr_fault(struct pt_regs *regs, unsigned long hw_error_code,
 	if (spurious_kernel_fault(hw_error_code, address))
 		return;
 
+	/* FIXME: warn and handle gracefully */
+	if (unlikely(fault_in_user_exclusive_page(address))) {
+		pr_err("page fault in user exclusive page at %lx", address);
+		force_sig_fault(SIGSEGV, SEGV_MAPERR, (void __user *)address);
+	}
+
 	/* kprobes don't want to hook the spurious faults: */
 	if (kprobe_page_fault(regs, X86_TRAP_PF))
 		return;
fs/proc/task_mmu.c +1 -0
index 9442631..99e14d1 100644
@@ -655,6 +655,7 @@ static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 #ifdef CONFIG_X86_INTEL_MPX
 		[ilog2(VM_MPX)]		= "mp",
 #endif
+		[ilog2(VM_EXCLUSIVE)]	= "xl",
 		[ilog2(VM_LOCKED)]	= "lo",
 		[ilog2(VM_IO)]		= "io",
 		[ilog2(VM_SEQ_READ)]	= "sr",
include/linux/mm.h +9 -0
index cc29227..9c43375 100644
@@ -298,11 +298,13 @@ extern unsigned int kobjsize(const void *objp);
 #define VM_HIGH_ARCH_BIT_2	34	/* bit only usable on 64-bit architectures */
 #define VM_HIGH_ARCH_BIT_3	35	/* bit only usable on 64-bit architectures */
 #define VM_HIGH_ARCH_BIT_4	36	/* bit only usable on 64-bit architectures */
+#define VM_HIGH_ARCH_BIT_5	37	/* bit only usable on 64-bit architectures */
 #define VM_HIGH_ARCH_0	BIT(VM_HIGH_ARCH_BIT_0)
 #define VM_HIGH_ARCH_1	BIT(VM_HIGH_ARCH_BIT_1)
 #define VM_HIGH_ARCH_2	BIT(VM_HIGH_ARCH_BIT_2)
 #define VM_HIGH_ARCH_3	BIT(VM_HIGH_ARCH_BIT_3)
 #define VM_HIGH_ARCH_4	BIT(VM_HIGH_ARCH_BIT_4)
+#define VM_HIGH_ARCH_5	BIT(VM_HIGH_ARCH_BIT_5)
 #endif /* CONFIG_ARCH_USES_HIGH_VMA_FLAGS */
 
 #ifdef CONFIG_ARCH_HAS_PKEYS
@@ -340,6 +342,12 @@ extern unsigned int kobjsize(const void *objp);
 # define VM_MPX		VM_NONE
 #endif
 
+#ifdef CONFIG_ARCH_USES_HIGH_VMA_FLAGS
+# define VM_EXCLUSIVE	VM_HIGH_ARCH_5
+#else
+# define VM_EXCLUSIVE	VM_NONE
+#endif
+
 #ifndef VM_GROWSUP
 # define VM_GROWSUP	VM_NONE
 #endif
@@ -2594,6 +2602,7 @@ struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
 #define FOLL_ANON	0x8000	/* don't do file mappings */
 #define FOLL_LONGTERM	0x10000	/* mapping lifetime is indefinite: see below */
 #define FOLL_SPLIT_PMD	0x20000	/* split huge pmd before returning */
+#define FOLL_EXCLUSIVE	0x40000	/* mapping is exclusive to owning mm */
 
 /*
  * NOTE on FOLL_LONGTERM:
include/linux/page-flags.h +7 -0
index f91cb88..32d0aee 100644
@@ -131,6 +131,9 @@ enum pageflags {
 	PG_young,
 	PG_idle,
 #endif
+#if defined(CONFIG_EXCLUSIVE_USER_PAGES)
+	PG_user_exclusive,
+#endif
 	__NR_PAGEFLAGS,
 
 	/* Filesystems */
@@ -431,6 +434,10 @@ TESTCLEARFLAG(Young, young, PF_ANY)
 PAGEFLAG(Idle, idle, PF_ANY)
 #endif
 
+#ifdef CONFIG_EXCLUSIVE_USER_PAGES
+__PAGEFLAG(UserExclusive, user_exclusive, PF_ANY)
+#endif
+
 /*
  * On an anonymous page mapped into a user virtual memory area,
  * page->mapping points to its anon_vma, not to a struct address_space;
include/linux/page_excl.h new file +49 -0

49 lines not on this page · raw message

include/trace/events/mmflags.h +8 -1
index a1675d4..2d3c14a 100644
@@ -79,6 +79,12 @@
 #define IF_HAVE_PG_IDLE(flag,string)
 #endif
 
+#ifdef CONFIG_EXCLUSIVE_USER_PAGES
+#define IF_HAVE_PG_USER_EXCLUSIVE(flag,string) ,{1UL << flag, string}
+#else
+#define IF_HAVE_PG_USER_EXCLUSIVE(flag,string)
+#endif
+
 #define __def_pageflag_names						\
 	{1UL << PG_locked,		"locked"	},		\
 	{1UL << PG_waiters,		"waiters"	},		\
@@ -105,7 +111,8 @@ IF_HAVE_PG_MLOCK(PG_mlocked,		"mlocked"	)		\
 IF_HAVE_PG_UNCACHED(PG_uncached,	"uncached"	)		\
 IF_HAVE_PG_HWPOISON(PG_hwpoison,	"hwpoison"	)		\
 IF_HAVE_PG_IDLE(PG_young,		"young"		)		\
-IF_HAVE_PG_IDLE(PG_idle,		"idle"		)
+IF_HAVE_PG_IDLE(PG_idle,		"idle"		)		\
+IF_HAVE_PG_USER_EXCLUSIVE(PG_user_exclusive,	"user_exclusive" )
 
 #define show_page_flags(flags)						\
 	(flags) ? __print_flags(flags, "|",				\
include/uapi/asm-generic/mman-common.h +1 -0
index c160a53..bf8f23e 100644
@@ -27,6 +27,7 @@
 #define MAP_HUGETLB		0x040000	/* create a huge page mapping */
 #define MAP_SYNC		0x080000 /* perform synchronous page faults for the mapping */
 #define MAP_FIXED_NOREPLACE	0x100000	/* MAP_FIXED which doesn't unmap underlying mapping */
+#define MAP_EXCLUSIVE		0x200000	/* mapping is exclusive to the owning task; the pages in it are dropped from the direct map */
 
 #define MAP_UNINITIALIZED 0x4000000	/* For anonymous mmap, memory could be
 					 * uninitialized */
kernel/fork.c +2 -1
index bcdf531..d63adec 100644
@@ -518,7 +518,8 @@ static __latent_entropy int dup_mmap(struct mm_struct *mm,
 	for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 		struct file *file;
 
-		if (mpnt->vm_flags & VM_DONTCOPY) {
+		if (mpnt->vm_flags & VM_DONTCOPY ||
+		    mpnt->vm_flags & VM_EXCLUSIVE) {
 			vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 			continue;
 		}
mm/Kconfig +3 -0
index a5dae9a..9d60141 100644
@@ -736,4 +736,7 @@ config ARCH_HAS_PTE_SPECIAL
 config ARCH_HAS_HUGEPD
 	bool
 
+config EXCLUSIVE_USER_PAGES
+        def_bool ARCH_USES_HIGH_VMA_FLAGS && ARCH_HAS_SET_DIRECT_MAP
+
 endmenu
mm/gup.c +8 -0
index 8f236a3..a98c0ca0 100644
@@ -17,6 +17,7 @@
 #include <linux/migrate.h>
 #include <linux/mm_inline.h>
 #include <linux/sched/mm.h>
+#include <linux/page_excl.h>
 
 #include <asm/mmu_context.h>
 #include <asm/pgtable.h>
@@ -868,6 +869,10 @@ static long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
 			ret = PTR_ERR(page);
 			goto out;
 		}
+
+		if (gup_flags & FOLL_EXCLUSIVE)
+			__set_page_user_exclusive(page);
+
 		if (pages) {
 			pages[i] = page;
 			flush_anon_page(vma, page, start);
@@ -1216,6 +1221,9 @@ long populate_vma_page_range(struct vm_area_struct *vma,
 	if (vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC))
 		gup_flags |= FOLL_FORCE;
 
+	if (vma->vm_flags & VM_EXCLUSIVE)
+		gup_flags |= FOLL_EXCLUSIVE;
+
 	/*
 	 * We made sure addr is within a VMA, so the following will
 	 * not result in a stack expansion that recurses back here.
mm/memory.c +3 -0
index b1ca51a..a4b4cff 100644
@@ -71,6 +71,7 @@
 #include <linux/dax.h>
 #include <linux/oom.h>
 #include <linux/numa.h>
+#include <linux/page_excl.h>
 
 #include <asm/io.h>
 #include <asm/mmu_context.h>
@@ -1062,6 +1063,8 @@ static unsigned long zap_pte_range(struct mmu_gather *tlb,
 			page_remove_rmap(page, false);
 			if (unlikely(page_mapcount(page) < 0))
 				print_bad_pte(vma, addr, ptent, page);
+			if (page_is_user_exclusive(page))
+				__clear_page_user_exclusive(page);
 			if (unlikely(__tlb_remove_page(tlb, page))) {
 				force_flush = 1;
 				addr += PAGE_SIZE;
mm/mmap.c +16 -0
index a7d8c84..d8cc82d 100644
@@ -1574,6 +1574,9 @@ unsigned long do_mmap(struct file *file, unsigned long addr,
 			vm_flags |= VM_NORESERVE;
 	}
 
+	if (flags & MAP_EXCLUSIVE)
+		vm_flags |= VM_EXCLUSIVE;
+
 	addr = mmap_region(file, addr, len, vm_flags, pgoff, uf);
 	if (!IS_ERR_VALUE(addr) &&
 	    ((vm_flags & VM_LOCKED) ||
@@ -1591,6 +1594,19 @@ unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len,
 
 	addr = untagged_addr(addr);
 
+	if (flags & MAP_EXCLUSIVE) {
+		/*
+		 * MAP_EXCLUSIVE is only supported for private
+		 * anonymous memory not backed by hugetlbfs
+		 */
+		if (!(flags & MAP_ANONYMOUS) || !(flags & MAP_PRIVATE) ||
+		    (flags & MAP_HUGETLB))
+			return -EINVAL;
+
+		/* and impies MAP_LOCKED and MAP_POPULATE */
+		flags |= (MAP_LOCKED | MAP_POPULATE);
+	}
+
 	if (!(flags & MAP_ANONYMOUS)) {
 		audit_mmap_fd(fd, flags);
 		file = fget(fd);
mm/page_alloc.c +5 -0
index ecc3dba..2f1de9d 100644
@@ -68,6 +68,7 @@
 #include <linux/lockdep.h>
 #include <linux/nmi.h>
 #include <linux/psi.h>
+#include <linux/page_excl.h>
 
 #include <asm/sections.h>
 #include <asm/tlbflush.h>
@@ -4779,6 +4780,10 @@ __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid,
 		page = NULL;
 	}
 
+	/* FIXME: should not happen! */
+	if (WARN_ON(page_is_user_exclusive(page)))
+		__clear_page_user_exclusive(page);
+
 	trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
 
 	return page;
Signature · 1 line
2.7.4
#2 lore raw
Quoted 10 lines · Mike Rapoport

From: Mike Rapoport <[email protected]>

The mappings created with MAP_EXCLUSIVE are visible only in the context of the owning process and can be used by applications to store secret information that will not be visible not only to other processes but to the kernel as well.

The pages in these mappings are removed from the kernel direct map and marked with PG_user_exclusive flag. When the exclusive area is unmapped, the pages are mapped back into the direct map.

I probably blind, but I don't see where you manipulate direct map...

Signature · 1 line
 Kirill A. Shutemov
author maintainer
#3 lore raw
Quoted 8 lines · Kirill A. Shutemov
Quoted 7 lines · Mike Rapoport

From: Mike Rapoport <[email protected]>

The mappings created with MAP_EXCLUSIVE are visible only in the context of the owning process and can be used by applications to store secret information that will not be visible not only to other processes but to the kernel as well.

Quoted 5 lines
Quoted 3 lines

The pages in these mappings are removed from the kernel direct map and marked with PG_user_exclusive flag. When the exclusive area is unmapped, the pages are mapped back into the direct map.

I probably blind, but I don't see where you manipulate direct map...

__get_user_pages() calls __set_page_user_exclusive() which in turn calls set_direct_map_invalid_noflush() that makes the page not present.

Quoted 2 lines
Signature · 1 line
 Kirill A. Shutemov
Signature · 2 lines
Sincerely yours,
Mike.
#4 lore raw
Quoted 9 lines · Mike Rapoport
Quoted 8 lines · Kirill A. Shutemov
Quoted 7 lines · Mike Rapoport

From: Mike Rapoport <[email protected]>

The mappings created with MAP_EXCLUSIVE are visible only in the context of the owning process and can be used by applications to store secret information that will not be visible not only to other processes but to the kernel as well.

Quoted 8 lines
Quoted 5 lines
Quoted 3 lines

The pages in these mappings are removed from the kernel direct map and marked with PG_user_exclusive flag. When the exclusive area is unmapped, the pages are mapped back into the direct map.

I probably blind, but I don't see where you manipulate direct map...

__get_user_pages() calls __set_page_user_exclusive() which in turn calls set_direct_map_invalid_noflush() that makes the page not present.

Ah. okay.

I think active use of this feature will lead to performance degradation of the system with time.

Setting a single 4k page non-present in the direct mapping will require splitting 2M or 1G page we usually map direct mapping with. And it's one way road. We don't have any mechanism to map the memory with huge page again after the application has freed the page.

It might be okay if all these pages cluster together, but I don't think we have a way to achieve it easily.

Signature · 1 line
 Kirill A. Shutemov
maintainer
#5 lore raw
Quoted 7 lines · Kirill A. Shutemov

I think active use of this feature will lead to performance degradation of the system with time.

Setting a single 4k page non-present in the direct mapping will require splitting 2M or 1G page we usually map direct mapping with. And it's one way road. We don't have any mechanism to map the memory with huge page again after the application has freed the page.

Right, we recently had a 'bug' where ftrace triggered something like this and facebook ran into it as a performance regression. So yes, this is a real concern.

#6 lore raw
Quoted 5 lines · Peter Zijlstra
Quoted 3 lines · Kirill A. Shutemov

I think active use of this feature will lead to performance degradation of the system with time.

Quoted 8 lines
Quoted 4 lines

Setting a single 4k page non-present in the direct mapping will require splitting 2M or 1G page we usually map direct mapping with. And it's one way road. We don't have any mechanism to map the memory with huge page again after the application has freed the page.

Right, we recently had a 'bug' where ftrace triggered something like this and facebook ran into it as a performance regression. So yes, this is a real concern.

Don't e/cBPF filters also break the direct map down to 4k pages when calling set_memory_ro() on the filter for 64 bit x86 and arm?

I've been wondering if the page allocator should make some effort to find a broken down page for anything that can be known will have direct map permissions changed (or if it already groups them somehow). But also, why any potential slowdown of 4k pages on the direct map hasn't been noticed for apps that do a lot of insertions and removals of BPF filters, if this is indeed the case.

maintainer
#7 lore raw
Quoted 15 lines · Rick P
Quoted 5 lines · Peter Zijlstra
Quoted 3 lines · Kirill A. Shutemov

I think active use of this feature will lead to performance degradation of the system with time.

Quoted 8 lines
Quoted 4 lines

Setting a single 4k page non-present in the direct mapping will require splitting 2M or 1G page we usually map direct mapping with. And it's one way road. We don't have any mechanism to map the memory with huge page again after the application has freed the page.

Right, we recently had a 'bug' where ftrace triggered something like this and facebook ran into it as a performance regression. So yes, this is a real concern.

Quoted 8 lines

Don't e/cBPF filters also break the direct map down to 4k pages when calling set_memory_ro() on the filter for 64 bit x86 and arm?

I've been wondering if the page allocator should make some effort to find a broken down page for anything that can be known will have direct map permissions changed (or if it already groups them somehow). But also, why any potential slowdown of 4k pages on the direct map hasn't been noticed for apps that do a lot of insertions and removals of BPF filters, if this is indeed the case.

That should be limited to the module range. Random data maps could shatter the world.

#8 lore raw
Quoted 28 lines · Peter Zijlstra
Quoted 19 lines · Rick P
Quoted 6 lines · Peter Zijlstra
Quoted 4 lines · Kirill A. Shutemov

I think active use of this feature will lead to performance degradation of
the system with time.

Quoted 8 lines
Quoted 4 lines

Setting a single 4k page non-present in the direct mapping will require splitting 2M or 1G page we usually map direct mapping with. And it's one way road. We don't have any mechanism to map the memory with huge page again after the application has freed the page.

Right, we recently had a 'bug' where ftrace triggered something like this and facebook ran into it as a performance regression. So yes, this is a real concern.

Don't e/cBPF filters also break the direct map down to 4k pages when calling set_memory_ro() on the filter for 64 bit x86 and arm?

Quoted 7 lines

I've been wondering if the page allocator should make some effort to find a broken down page for anything that can be known will have direct map permissions
changed (or if it already groups them somehow). But also, why any potential slowdown of 4k pages on the direct map hasn't been noticed for apps that do a
lot of insertions and removals of BPF filters, if this is indeed the case.

Quoted 2 lines

That should be limited to the module range. Random data maps could shatter the world.

BPF has one vmalloc space allocation for the byte code and one for the module space allocation for the JIT. Both get RO also set on the direct map alias of the pages, and reset RW when freed.

You mean shatter performance?

maintainer
#9 lore raw
Quoted 8 lines · Rick P
Quoted 2 lines · Peter Zijlstra

That should be limited to the module range. Random data maps could shatter the world.

BPF has one vmalloc space allocation for the byte code and one for the module space allocation for the JIT. Both get RO also set on the direct map alias of the pages, and reset RW when freed.

Argh, I didn't know they mapped the bytecode RO; why does it do that? It can throw out the bytecode once it's JIT'ed.

Quoted 1 line

You mean shatter performance?

Shatter (all) large pages.

#10 lore raw
Quoted 7 lines · Peter Zijlstra
Quoted 5 lines · Rick P
Quoted 2 lines · Peter Zijlstra

That should be limited to the module range. Random data maps could shatter the world.

Quoted 6 lines
Quoted 3 lines

BPF has one vmalloc space allocation for the byte code and one for the module space allocation for the JIT. Both get RO also set on the direct map alias of the pages, and reset RW when freed.

Argh, I didn't know they mapped the bytecode RO; why does it do that? It can throw out the bytecode once it's JIT'ed.

because of endless security "concerns" that some folks had.
Like what if something can exploit another bug in the kernel
and modify bytecode that was already verified
then interpreter will execute that modified bytecode.
Sort of similar reasoning why .text is read-only.
I think it's not a realistic attack, but I didn't bother to argue back then. The mere presence of interpreter itself is a real security concern. People that care about speculation attacks should
have CONFIG_BPF_JIT_ALWAYS_ON=y,
so modifying bytecode via another exploit will be pointless. Getting rid of RO for bytecode will save a ton of memory too, since we won't need to allocate full page for each small programs.

maintainer
#11 lore raw
Quoted 12 lines · Alexei Starovoitov
Quoted 11 lines · Peter Zijlstra
Quoted 8 lines · Rick P
Quoted 2 lines · Peter Zijlstra

That should be limited to the module range. Random data maps could shatter the world.

BPF has one vmalloc space allocation for the byte code and one for the module space allocation for the JIT. Both get RO also set on the direct map alias of the pages, and reset RW when freed.

Quoted 7 lines
Quoted 2 lines

Argh, I didn't know they mapped the bytecode RO; why does it do that? It can throw out the bytecode once it's JIT'ed.

because of endless security "concerns" that some folks had. Like what if something can exploit another bug in the kernel and modify bytecode that was already verified
then interpreter will execute that modified bytecode.

But when it's JIT'ed the bytecode is no longer of relevance, right? So any scenario with a JIT on can then toss the bytecode and certainly doesn't need to map it RO.

Quoted 5 lines

Sort of similar reasoning why .text is read-only.
I think it's not a realistic attack, but I didn't bother to argue back then. The mere presence of interpreter itself is a real security concern. People that care about speculation attacks should
have CONFIG_BPF_JIT_ALWAYS_ON=y,

This isn't about speculation attacks, it is about breaking buffer limits and being able to write to memory. And in that respect being able to change the current task state (write it's effective PID to 0) is much simpler than writing to text or bytecode, but if you cannot reach/find the task struct but can reach/find text..

Quoted 3 lines

so modifying bytecode via another exploit will be pointless. Getting rid of RO for bytecode will save a ton of memory too, since we won't need to allocate full page for each small programs.

So I'm thinking we can get rid of that for any scenario that has the JIT enabled -- not only JIT_ALWAYS_ON.

#12 lore raw
Quoted 17 lines · Peter Zijlstra
Quoted 15 lines · Alexei Starovoitov
Quoted 13 lines · Peter Zijlstra
Quoted 8 lines · Rick P
Quoted 2 lines · Peter Zijlstra

That should be limited to the module range. Random data maps could shatter the world.

BPF has one vmalloc space allocation for the byte code and one for the module space allocation for the JIT. Both get RO also set on the direct map alias of the pages, and reset RW when freed.

Argh, I didn't know they mapped the bytecode RO; why does it do that? It can throw out the bytecode once it's JIT'ed.

Quoted 8 lines
Quoted 4 lines

because of endless security "concerns" that some folks had. Like what if something can exploit another bug in the kernel and modify bytecode that was already verified
then interpreter will execute that modified bytecode.

But when it's JIT'ed the bytecode is no longer of relevance, right? So any scenario with a JIT on can then toss the bytecode and certainly doesn't need to map it RO.

We keep so called "xlated" bytecode around for debugging. It's the one that is actually running. It was modified through several stages of the verifier before being runnable by interpreter. When folks debug stuff in production they want to see
the whole thing. Both x86 asm and xlated bytecode.
xlated bytecode also sanitized before it's returned
back to user space.

Quoted 11 lines
Quoted 5 lines

Sort of similar reasoning why .text is read-only.
I think it's not a realistic attack, but I didn't bother to argue back then. The mere presence of interpreter itself is a real security concern. People that care about speculation attacks should
have CONFIG_BPF_JIT_ALWAYS_ON=y,

This isn't about speculation attacks, it is about breaking buffer limits and being able to write to memory. And in that respect being able to change the current task state (write it's effective PID to 0) is much simpler than writing to text or bytecode, but if you cannot reach/find the task struct but can reach/find text..

exactly. that's why RO bytecode was dubious to me from the beginning. For an attacker to write meaningful bytecode they need to know quite a few other kernel internal pointers.
If an exploit can write into memory there are plenty of easier targets.

Quoted 6 lines
Quoted 3 lines

so modifying bytecode via another exploit will be pointless. Getting rid of RO for bytecode will save a ton of memory too, since we won't need to allocate full page for each small programs.

So I'm thinking we can get rid of that for any scenario that has the JIT enabled -- not only JIT_ALWAYS_ON.

Sounds good to me. Happy to do that. Will add it to our todo list.

#13 lore raw
In reply to Peter Zijlstra
Quoted 3 lines · Peter Zijlstra
Quoted 1 line

You mean shatter performance?

Shatter (all) large pages.

So it looks like this is already happening then to some degree. It's not just BPF either, any module_alloc() user is going to do something similar with the direct map alias of the page they got for the text.

So there must be at least some usages where breaking the direct map down, for like a page to store a key or something, isn't totally horrible.

#14 lore raw
Quoted 11 lines · Rick P
Quoted 3 lines · Peter Zijlstra
Quoted 1 line

You mean shatter performance?

Shatter (all) large pages.

So it looks like this is already happening then to some degree. It's not just BPF either, any module_alloc() user is going to do something similar with the direct map alias of the page they got for the text.

So there must be at least some usages where breaking the direct map down, for like a page to store a key or something, isn't totally horrible.

The systems that really need large pages are the large ones. They have the same TLBs and data structures as really little systems, but orders of magnitude more address space. Modules and BPF are a (hopefully) drop in the bucket on small systems and they're really inconsequential on really big systems.

Modules also require privilege.

Allowing random user apps to fracture the direct map for every page of their memory or *lots* of pages of their memory is an entirely different kind of problem from modules. It takes a "drop in the bucket" fracturing and turns it into the common case.

#15 lore raw
Quoted 1 line · Dave Hansen

Modules also require privilege.

IMNHO, if BPF is fracturing large swaths the direct map with no privilege, it's only a matter of time until it starts to cause problems.
The fact that we do it today is only evidence that we have a ticking
time bomb, not that it's OK.

#16 lore raw
In reply to Kirill A. Shutemov
Quoted 25 lines · Kirill A. Shutemov
Quoted 9 lines · Mike Rapoport
Quoted 8 lines · Kirill A. Shutemov
Quoted 7 lines · Mike Rapoport

From: Mike Rapoport <[email protected]>

The mappings created with MAP_EXCLUSIVE are visible only in the context of the owning process and can be used by applications to store secret information that will not be visible not only to other processes but to the kernel as well.

Quoted 8 lines
Quoted 5 lines
Quoted 3 lines

The pages in these mappings are removed from the kernel direct map and marked with PG_user_exclusive flag. When the exclusive area is unmapped, the pages are mapped back into the direct map.

I probably blind, but I don't see where you manipulate direct map...

__get_user_pages() calls __set_page_user_exclusive() which in turn calls set_direct_map_invalid_noflush() that makes the page not present.

Ah. okay.

I think active use of this feature will lead to performance degradation of the system with time.

Quoted 7 lines

Setting a single 4k page non-present in the direct mapping will require splitting 2M or 1G page we usually map direct mapping with. And it's one way road. We don't have any mechanism to map the memory with huge page again after the application has freed the page.

It might be okay if all these pages cluster together, but I don't think we have a way to achieve it easily.

Still, it would be worth exploring what that would look like if not for MAP_EXCLUSIVE then set_mce_nospec() that wants to punch out poison pages from the direct map. In the case of pmem, where those pages are able to be repaired, it would be nice to also repair the mapping granularity of the direct map.

#17 lore raw
Quoted 31 lines · Dan Williams
Quoted 30 lines · Kirill A. Shutemov
Quoted 9 lines · Mike Rapoport
Quoted 8 lines · Kirill A. Shutemov
Quoted 7 lines · Mike Rapoport

From: Mike Rapoport <[email protected]>

The mappings created with MAP_EXCLUSIVE are visible only in the context of the owning process and can be used by applications to store secret information that will not be visible not only to other processes but to the kernel as well.

Quoted 8 lines
Quoted 5 lines
Quoted 3 lines

The pages in these mappings are removed from the kernel direct map and marked with PG_user_exclusive flag. When the exclusive area is unmapped, the pages are mapped back into the direct map.

I probably blind, but I don't see where you manipulate direct map...

__get_user_pages() calls __set_page_user_exclusive() which in turn calls set_direct_map_invalid_noflush() that makes the page not present.

Ah. okay.

I think active use of this feature will lead to performance degradation of the system with time.

Setting a single 4k page non-present in the direct mapping will require splitting 2M or 1G page we usually map direct mapping with. And it's one way road. We don't have any mechanism to map the memory with huge page again after the application has freed the page.

Quoted 8 lines
Quoted 2 lines

It might be okay if all these pages cluster together, but I don't think we have a way to achieve it easily.

Still, it would be worth exploring what that would look like if not for MAP_EXCLUSIVE then set_mce_nospec() that wants to punch out poison pages from the direct map. In the case of pmem, where those pages are able to be repaired, it would be nice to also repair the mapping granularity of the direct map.

The solution has to consist of two parts: finding a range to collapse and actually collapsing the range into a huge page.

Finding the collapsible range will likely require background scanning of the direct mapping as we do for THP with khugepaged. It should not too hard, but likely require long and tedious tuning to be effective, but not too disturbing for the system.

Alternatively, after any changes to the direct mapping, we can initiate checking if the range is collapsible. Up to 1G around the changed 4k. It might be more taxing than scanning if direct mapping changes often.

Collapsing itself appears to be simple: re-check if the range is collapsible under the lock, replace the page table with the huge page and flush the TLB.

But some CPUs don't like to have two TLB entries for the same memory with different sizes at the same time. See for instance AMD erratum 383.

Getting it right would require making the range not present, flush TLB and only then install huge page. That's what we do for userspace.

It will not fly for the direct mapping. There is no reasonable way to exclude other CPU from accessing the range while it's not present (call stop_machine()? :P). Moreover, the range may contain the code that doing the collapse or data required for it...

BTW, looks like current __split_large_page() in pageattr.c is susceptible to the errata. Maybe we can get away with the easy way...

Signature · 1 line
 Kirill A. Shutemov
maintainer
#18 lore raw
Quoted 6 lines · Kirill A. Shutemov

But some CPUs don't like to have two TLB entries for the same memory with different sizes at the same time. See for instance AMD erratum 383.

Getting it right would require making the range not present, flush TLB and only then install huge page. That's what we do for userspace.

Quoted 7 lines

It will not fly for the direct mapping. There is no reasonable way to exclude other CPU from accessing the range while it's not present (call stop_machine()? :P). Moreover, the range may contain the code that doing the collapse or data required for it...

BTW, looks like current __split_large_page() in pageattr.c is susceptible to the errata. Maybe we can get away with the easy way...

As you write above, there is just no way we can have a (temporary) hole in the direct map.

We are careful about that other errata, and make sure both translations are identical wrt everything else.

#19 lore raw
Quoted 12 lines · Peter Zijlstra
Quoted 11 lines · Kirill A. Shutemov

But some CPUs don't like to have two TLB entries for the same memory with different sizes at the same time. See for instance AMD erratum 383.

Getting it right would require making the range not present, flush TLB and only then install huge page. That's what we do for userspace.

It will not fly for the direct mapping. There is no reasonable way to exclude other CPU from accessing the range while it's not present (call stop_machine()? :P). Moreover, the range may contain the code that doing the collapse or data required for it...

Quoted 8 lines
Quoted 2 lines

BTW, looks like current __split_large_page() in pageattr.c is susceptible to the errata. Maybe we can get away with the easy way...

As you write above, there is just no way we can have a (temporary) hole in the direct map.

We are careful about that other errata, and make sure both translations are identical wrt everything else.

It's not clear if it is enough to avoid the issue. "under a highly specific and detailed set of conditions" is not very specific set of conditions :P

Signature · 1 line
 Kirill A. Shutemov
maintainer
#20 lore raw

AMD TLB errata, (Was: [PATCH RFC] mm: add MAP_EXCLUSIVE to create exclusive user mappings)

Quoted 16 lines · Kirill A. Shutemov
Quoted 15 lines · Peter Zijlstra
Quoted 13 lines · Kirill A. Shutemov

But some CPUs don't like to have two TLB entries for the same memory with different sizes at the same time. See for instance AMD erratum 383.

Getting it right would require making the range not present, flush TLB and only then install huge page. That's what we do for userspace.

It will not fly for the direct mapping. There is no reasonable way to exclude other CPU from accessing the range while it's not present (call stop_machine()? :P). Moreover, the range may contain the code that doing the collapse or data required for it...

BTW, looks like current __split_large_page() in pageattr.c is susceptible to the errata. Maybe we can get away with the easy way...

Quoted 8 lines
Quoted 5 lines

As you write above, there is just no way we can have a (temporary) hole in the direct map.

We are careful about that other errata, and make sure both translations are identical wrt everything else.

It's not clear if it is enough to avoid the issue. "under a highly specific and detailed set of conditions" is not very specific set of conditions :P

Yeah, I know ... :/ Tom is there any chance you could shed a little more light on that errata?

#21 lore raw

Re: AMD TLB errata, (Was: [PATCH RFC] mm: add MAP_EXCLUSIVE to create exclusive user mappings)

Quoted 20 lines · Peter Zijlstra
Quoted 19 lines · Kirill A. Shutemov
Quoted 18 lines · Peter Zijlstra
Quoted 6 lines · Kirill A. Shutemov

But some CPUs don't like to have two TLB entries for the same memory with different sizes at the same time. See for instance AMD erratum 383.

Getting it right would require making the range not present, flush TLB and only then install huge page. That's what we do for userspace.

Quoted 7 lines

It will not fly for the direct mapping. There is no reasonable way to exclude other CPU from accessing the range while it's not present (call stop_machine()? :P). Moreover, the range may contain the code that doing the collapse or data required for it...

BTW, looks like current __split_large_page() in pageattr.c is susceptible to the errata. Maybe we can get away with the easy way...

As you write above, there is just no way we can have a (temporary) hole in the direct map.

Quoted 8 lines
Quoted 5 lines
Quoted 2 lines

We are careful about that other errata, and make sure both translations are identical wrt everything else.

It's not clear if it is enough to avoid the issue. "under a highly specific and detailed set of conditions" is not very specific set of conditions :P

Yeah, I know ... :/ Tom is there any chance you could shed a little more light on that errata?

I talked with some of the hardware folks and if you maintain the same bits in the large and small pages (aside from the large page bit) until the flush, then the errata should not occur.

The errata really applies to mappings that end up with different attribute bits being set. Even then, it doesn't fail every time. There are other conditions required to make it fail.

Thanks,
Tom

Quoted 1 line
maintainer
#22 lore raw

Re: AMD TLB errata, (Was: [PATCH RFC] mm: add MAP_EXCLUSIVE to create exclusive user mappings)

Quoted 3 lines · Tom Lendacky

I talked with some of the hardware folks and if you maintain the same bits in the large and small pages (aside from the large page bit) until the flush, then the errata should not occur.

Excellent!

Thanks for digging that out Tom.

#23 lore raw
In reply to Kirill A. Shutemov
Quoted 54 lines · Kirill A. Shutemov
Quoted 31 lines · Dan Williams
Quoted 30 lines · Kirill A. Shutemov
Quoted 9 lines · Mike Rapoport
Quoted 8 lines · Kirill A. Shutemov
Quoted 7 lines · Mike Rapoport

From: Mike Rapoport <[email protected]>

The mappings created with MAP_EXCLUSIVE are visible only in the context of the owning process and can be used by applications to store secret information that will not be visible not only to other processes but to the kernel as well.

Quoted 8 lines
Quoted 5 lines
Quoted 3 lines

The pages in these mappings are removed from the kernel direct map and marked with PG_user_exclusive flag. When the exclusive area is unmapped, the pages are mapped back into the direct map.

I probably blind, but I don't see where you manipulate direct map...

__get_user_pages() calls __set_page_user_exclusive() which in turn calls set_direct_map_invalid_noflush() that makes the page not present.

Ah. okay.

I think active use of this feature will lead to performance degradation of the system with time.

Setting a single 4k page non-present in the direct mapping will require splitting 2M or 1G page we usually map direct mapping with. And it's one way road. We don't have any mechanism to map the memory with huge page again after the application has freed the page.

Quoted 8 lines
Quoted 2 lines

It might be okay if all these pages cluster together, but I don't think we have a way to achieve it easily.

Still, it would be worth exploring what that would look like if not for MAP_EXCLUSIVE then set_mce_nospec() that wants to punch out poison pages from the direct map. In the case of pmem, where those pages are able to be repaired, it would be nice to also repair the mapping granularity of the direct map.

The solution has to consist of two parts: finding a range to collapse and actually collapsing the range into a huge page.

Finding the collapsible range will likely require background scanning of the direct mapping as we do for THP with khugepaged. It should not too hard, but likely require long and tedious tuning to be effective, but not too disturbing for the system.

Alternatively, after any changes to the direct mapping, we can initiate checking if the range is collapsible. Up to 1G around the changed 4k. It might be more taxing than scanning if direct mapping changes often.

Quoted 6 lines

Collapsing itself appears to be simple: re-check if the range is collapsible under the lock, replace the page table with the huge page and flush the TLB.

But some CPUs don't like to have two TLB entries for the same memory with different sizes at the same time. See for instance AMD erratum 383.

That basic description would seem to defeat most (all?) interesting huge page use cases. For example dax makes no attempt to make sure aliased mappings of pmem are the same size between the direct map that the driver uses, and userspace dax mappings. So I assume there are more details than "all aliased mappings must be the same size".

Quoted 7 lines

Getting it right would require making the range not present, flush TLB and only then install huge page. That's what we do for userspace.

It will not fly for the direct mapping. There is no reasonable way to exclude other CPU from accessing the range while it's not present (call stop_machine()? :P). Moreover, the range may contain the code that doing the collapse or data required for it...

At least for pmem all the access points can be controlled. pmem is never used for kernel text at least in the dax mode where it is accessed via file-backed shared mappings, or the pmem driver. So when I say "direct-map repair" I mean the incidental direct-map that pmem uses since it maps pmem with arch_add_memory(), not the typical DRAM direct-map that may house kernel text. Poison consumed from the kernel DRAM direct-map is fatal, poison consumed from dax mappings and the pmem driver path is recoverable and repairable.

#24 lore raw
Quoted 7 lines · Dan Williams
Quoted 2 lines

But some CPUs don't like to have two TLB entries for the same memory with different sizes at the same time. See for instance AMD erratum 383.

That basic description would seem to defeat most (all?) interesting huge page use cases. For example dax makes no attempt to make sure aliased mappings of pmem are the same size between the direct map that the driver uses, and userspace dax mappings. So I assume there are more details than "all aliased mappings must be the same size".

These are about when large and small TLB entries could be held in the TLB at the same time for the same virtual address in the same process. It doesn't matter that two *different* mappings are using different page size.

Imagine you were *just* changing the page size. Without these errata, you could just skip flushing the TLB. You might use the old hardware page size for a while, but it will be functionally OK. With these errata, we need to ensure in software that the old TLB entries for the old page size are flushed before the new page size is established.

#25 lore raw
In reply to Kirill A. Shutemov
Quoted 7 lines · Kirill A. Shutemov

Setting a single 4k page non-present in the direct mapping will require splitting 2M or 1G page we usually map direct mapping with. And it's one way road. We don't have any mechanism to map the memory with huge page again after the application has freed the page.

It might be okay if all these pages cluster together, but I don't think we have a way to achieve it easily.

Set aside a special physical memory range for this and migrate the page to that physical memory range when MAP_EXCLUSIVE is specified?

Maybe some processors also have hardware ranges that offer additional protection for stuff like that?

author maintainer
#26 lore raw
Quoted 7 lines · Christopher Lameter
Quoted 5 lines · Kirill A. Shutemov

Setting a single 4k page non-present in the direct mapping will require splitting 2M or 1G page we usually map direct mapping with. And it's one way road. We don't have any mechanism to map the memory with huge page again after the application has freed the page.

Quoted 5 lines
Quoted 2 lines

It might be okay if all these pages cluster together, but I don't think we have a way to achieve it easily.

Set aside a special physical memory range for this and migrate the page to that physical memory range when MAP_EXCLUSIVE is specified?

I've talked with Thomas yesterday and he suggested something similar:

When the MAP_EXCLUSIVE request comes for the first time, we allocate a huge page for it and then use this page as a pool of 4K pages for subsequent requests. Once this huge page is full we allocate a new one and append it to the pool. When all the 4K pages that comprise the huge page are freed the huge page is collapsed.

And then on top of this we can look into compaction of the direct map.

Of course, this would work if the easy way of collapsing direct map pages Kirill mentioned on other mail will work.

Quoted 3 lines

Maybe some processors also have hardware ranges that offer additional protection for stuff like that?

Signature · 2 lines
Sincerely yours,
Mike.
#27 lore raw
Quoted 7 lines · Mike Rapoport

I've talked with Thomas yesterday and he suggested something similar:

When the MAP_EXCLUSIVE request comes for the first time, we allocate a huge page for it and then use this page as a pool of 4K pages for subsequent requests. Once this huge page is full we allocate a new one and append it to the pool. When all the 4K pages that comprise the huge page are freed the huge page is collapsed.

Or write a device driver that allows you to mmap a secure area and avoid all core kernel modifications?

/dev/securemem or so?

It may exist already.

author maintainer
#28 lore raw
Quoted 12 lines · Christopher Lameter
Quoted 7 lines · Mike Rapoport

I've talked with Thomas yesterday and he suggested something similar:

When the MAP_EXCLUSIVE request comes for the first time, we allocate a huge page for it and then use this page as a pool of 4K pages for subsequent requests. Once this huge page is full we allocate a new one and append it to the pool. When all the 4K pages that comprise the huge page are freed the huge page is collapsed.

Quoted 4 lines

Or write a device driver that allows you to mmap a secure area and avoid all core kernel modifications?

/dev/securemem or so?

A device driver will need to remove the secure area from the direct map and then we back to square one.

Quoted 2 lines

It may exist already.

Signature · 2 lines
Sincerely yours,
Mike.
#29 lore raw
Quoted 4 lines · Mike Rapoport
Quoted 1 line

/dev/securemem or so?

A device driver will need to remove the secure area from the direct map and then we back to square one.

We have avoided the need for modifications to kernel core code. And its a natural thing to treat this like special memory provided by a device driver.

#30 lore raw
In reply to Mike Rapoport
Quoted 126 lines · Mike Rapoport

From: Mike Rapoport <[email protected]>

The mappings created with MAP_EXCLUSIVE are visible only in the context of the owning process and can be used by applications to store secret information that will not be visible not only to other processes but to the kernel as well.

The pages in these mappings are removed from the kernel direct map and marked with PG_user_exclusive flag. When the exclusive area is unmapped, the pages are mapped back into the direct map.

The MAP_EXCLUSIVE flag implies MAP_POPULATE and MAP_LOCKED.

---
  arch/x86/mm/fault.c                    | 14 ++++++++++
  fs/proc/task_mmu.c                     |  1 +
  include/linux/mm.h                     |  9 +++++++
  include/linux/page-flags.h             |  7 +++++
  include/linux/page_excl.h              | 49 ++++++++++++++++++++++++++++++++++
  include/trace/events/mmflags.h         |  9 ++++++-
  include/uapi/asm-generic/mman-common.h |  1 +
  kernel/fork.c                          |  3 ++-
  mm/Kconfig                             |  3 +++
  mm/gup.c                               |  8 ++++++
  mm/memory.c                            |  3 +++
  mm/mmap.c                              | 16 +++++++++++
  mm/page_alloc.c                        |  5 ++++
  13 files changed, 126 insertions(+), 2 deletions(-)

create mode 100644 include/linux/page_excl.h

arch/x86/mm/fault.c +14 -0
index 9ceacd1..8f73a75 100644
@@ -17,6 +17,7 @@
  #include <linux/context_tracking.h>	/* exception_enter(), ...	*/
  #include <linux/uaccess.h>		/* faulthandler_disabled()	*/
  #include <linux/efi.h>			/* efi_recover_from_page_fault()*/
+#include <linux/page_excl.h>		/* page_is_user_exclusive()	*/
  #include <linux/mm_types.h>
 
  #include <asm/cpufeature.h>		/* boot_cpu_has, ...		*/
@@ -1218,6 +1219,13 @@ static int fault_in_kernel_space(unsigned long address)
  	return address >= TASK_SIZE_MAX;
  }
 
+static bool fault_in_user_exclusive_page(unsigned long address)
+{
+	struct page *page = virt_to_page(address);
+
+	return page_is_user_exclusive(page);
+}
+
  /*
   * Called for all faults where 'address' is part of the kernel address
   * space.  Might get called for faults that originate from *code* that
@@ -1261,6 +1269,12 @@ do_kern_addr_fault(struct pt_regs *regs, unsigned long hw_error_code,
  	if (spurious_kernel_fault(hw_error_code, address))
  		return;
 
+	/* FIXME: warn and handle gracefully */
+	if (unlikely(fault_in_user_exclusive_page(address))) {
+		pr_err("page fault in user exclusive page at %lx", address);
+		force_sig_fault(SIGSEGV, SEGV_MAPERR, (void __user *)address);
+	}
+
  	/* kprobes don't want to hook the spurious faults: */
  	if (kprobe_page_fault(regs, X86_TRAP_PF))
  		return;
fs/proc/task_mmu.c +1 -0
index 9442631..99e14d1 100644
@@ -655,6 +655,7 @@ static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
  #ifdef CONFIG_X86_INTEL_MPX
  		[ilog2(VM_MPX)]		= "mp",
  #endif
+		[ilog2(VM_EXCLUSIVE)]	= "xl",
  		[ilog2(VM_LOCKED)]	= "lo",
  		[ilog2(VM_IO)]		= "io",
  		[ilog2(VM_SEQ_READ)]	= "sr",
include/linux/mm.h +9 -0
index cc29227..9c43375 100644
@@ -298,11 +298,13 @@ extern unsigned int kobjsize(const void *objp);
  #define VM_HIGH_ARCH_BIT_2	34	/* bit only usable on 64-bit architectures */
  #define VM_HIGH_ARCH_BIT_3	35	/* bit only usable on 64-bit architectures */
  #define VM_HIGH_ARCH_BIT_4	36	/* bit only usable on 64-bit architectures */
+#define VM_HIGH_ARCH_BIT_5	37	/* bit only usable on 64-bit architectures */
  #define VM_HIGH_ARCH_0	BIT(VM_HIGH_ARCH_BIT_0)
  #define VM_HIGH_ARCH_1	BIT(VM_HIGH_ARCH_BIT_1)
  #define VM_HIGH_ARCH_2	BIT(VM_HIGH_ARCH_BIT_2)
  #define VM_HIGH_ARCH_3	BIT(VM_HIGH_ARCH_BIT_3)
  #define VM_HIGH_ARCH_4	BIT(VM_HIGH_ARCH_BIT_4)
+#define VM_HIGH_ARCH_5	BIT(VM_HIGH_ARCH_BIT_5)
  #endif /* CONFIG_ARCH_USES_HIGH_VMA_FLAGS */
 
  #ifdef CONFIG_ARCH_HAS_PKEYS
@@ -340,6 +342,12 @@ extern unsigned int kobjsize(const void *objp);
  # define VM_MPX		VM_NONE
  #endif
 
+#ifdef CONFIG_ARCH_USES_HIGH_VMA_FLAGS
+# define VM_EXCLUSIVE	VM_HIGH_ARCH_5
+#else
+# define VM_EXCLUSIVE	VM_NONE
+#endif
+
  #ifndef VM_GROWSUP
  # define VM_GROWSUP	VM_NONE
  #endif
@@ -2594,6 +2602,7 @@ struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
  #define FOLL_ANON	0x8000	/* don't do file mappings */
  #define FOLL_LONGTERM	0x10000	/* mapping lifetime is indefinite: see below */
  #define FOLL_SPLIT_PMD	0x20000	/* split huge pmd before returning */
+#define FOLL_EXCLUSIVE	0x40000	/* mapping is exclusive to owning mm */
 
  /*
   * NOTE on FOLL_LONGTERM:
include/linux/page-flags.h +0 -0
index f91cb88..32d0aee 100644

include/linux/page-flags.h enum pageflags { · line 136

Quoted 6 lines
include/linux/page-flags.h +3 -0
@@ -131,6 +131,9 @@ enum pageflags {
  	PG_young,
  	PG_idle,
  #endif
+#if defined(CONFIG_EXCLUSIVE_USER_PAGES)
+	PG_user_exclusive,
+#endif

Last time I tried to introduce a new page flag I learned that this is very much frowned upon. Best you can usually do is reuse another flag - if valid in that context.

Signature · 3 lines
Thanks,

David / dhildenb
#31 lore raw
In reply to Mike Rapoport
Quoted 3 lines · Mike Rapoport

The pages in these mappings are removed from the kernel direct map and marked with PG_user_exclusive flag. When the exclusive area is unmapped, the pages are mapped back into the direct map.

This looks fun. It's certainly simple.

But, the description is not really calling out the pros and cons very well. I'm also not sure that folks will use an interface like this that requires up-front, special code to do an allocation instead of something like madvise(). That's why protection keys ended up the way it did: if you do this as a mmap() replacement, you need to modify all *allocators* to be enabled for this. If you do it with mprotect()-style, you can apply it to existing allocations.

Some other random thoughts:

* The page flag is probably not a good idea. It would be probably better to set _PAGE_SPECIAL on the PTE and force get_user_pages() into the slow path.
* This really stops being "normal" memory. You can't do futexes on it, cant splice it. Probably need a more fleshed-out list of incompatible features.
* As Kirill noted, each 4k page ends up with a potential 1GB "blast radius" of demoted pages in the direct map. Not cool. This is probably a non-starter as it stands.
* The global TLB flushes are going to eat you alive. They probably border on a DoS on larger systems.
* Do we really want this user interface to dictate the kernel implementation? In other words, do we really want MAP_EXCLUSIVE, or do we want MAP_SECRET? One tells the kernel what do *do*, the other tells the kernel what the memory *IS*.
* There's a lot of other stuff going on in this area: XPFO, SEV, MKTME, Persistent Memory, where the kernel direct map is a liability in some way. We probably need some kind of overall, architected solution rather than five or ten things all poking at the direct map.

#32 lore raw
Quoted 17 lines · Dave Hansen
Quoted 3 lines · Mike Rapoport

The pages in these mappings are removed from the kernel direct map and marked with PG_user_exclusive flag. When the exclusive area is unmapped, the pages are mapped back into the direct map.

This looks fun. It's certainly simple.

But, the description is not really calling out the pros and cons very well. I'm also not sure that folks will use an interface like this that requires up-front, special code to do an allocation instead of something like madvise(). That's why protection keys ended up the way it did: if you do this as a mmap() replacement, you need to modify all *allocators* to be enabled for this. If you do it with mprotect()-style, you can apply it to existing allocations.

Some other random thoughts:

Quoted 15 lines

* The page flag is probably not a good idea. It would be probably better to set _PAGE_SPECIAL on the PTE and force get_user_pages() into the slow path.
* This really stops being "normal" memory. You can't do futexes on it, cant splice it. Probably need a more fleshed-out list of incompatible features.
* As Kirill noted, each 4k page ends up with a potential 1GB "blast radius" of demoted pages in the direct map. Not cool. This is probably a non-starter as it stands.
* The global TLB flushes are going to eat you alive. They probably border on a DoS on larger systems.
* Do we really want this user interface to dictate the kernel implementation? In other words, do we really want MAP_EXCLUSIVE, or do we want MAP_SECRET? One tells the kernel what do *do*, the other tells the kernel what the memory *IS*.

If we go that route, maybe MAP_USER_SECRET so that there's wiggle room in the event that there are different secret keepers that require different
implementations in the kernel? E.g. MAP_GUEST_SECRET for a KVM guest to
take the userspace VMM (Qemu) out of the TCB, i.e. the mapping would be accessible by the kernel (or just KVM?) and the KVM guest, but not userspace.

Quoted 5 lines

* There's a lot of other stuff going on in this area: XPFO, SEV, MKTME, Persistent Memory, where the kernel direct map is a liability in some way. We probably need some kind of overall, architected solution rather than five or ten things all poking at the direct map.

#33 lore raw
In reply to Dave Hansen
Quoted 13 lines · Dave Hansen

Some other random thoughts:

* The page flag is probably not a good idea. It would be probably better to set _PAGE_SPECIAL on the PTE and force get_user_pages() into the slow path.
* This really stops being "normal" memory. You can't do futexes on it, cant splice it. Probably need a more fleshed-out list of incompatible features.
* As Kirill noted, each 4k page ends up with a potential 1GB "blast radius" of demoted pages in the direct map. Not cool. This is probably a non-starter as it stands.
* The global TLB flushes are going to eat you alive. They probably border on a DoS on larger systems.

Quoted 8 lines

* Do we really want this user interface to dictate the kernel implementation? In other words, do we really want MAP_EXCLUSIVE, or do we want MAP_SECRET? One tells the kernel what do *do*, the other tells the kernel what the memory *IS*.
* There's a lot of other stuff going on in this area: XPFO, SEV, MKTME, Persistent Memory, where the kernel direct map is a liability in some way. We probably need some kind of overall, architected solution rather than five or ten things all poking at the direct map.

Another random set of thoughts:

- Should devices be permitted to DMA to/from MAP_SECRET pages?
- How about GUP? Can I ptrace my way into another process's secret pages?
- What if I splice() the page into a pipe?

author maintainer
#34 lore raw
Quoted 23 lines · Matthew Wilcox
Quoted 13 lines · Dave Hansen

Some other random thoughts:

* The page flag is probably not a good idea. It would be probably better to set _PAGE_SPECIAL on the PTE and force get_user_pages() into the slow path.
* This really stops being "normal" memory. You can't do futexes on it, cant splice it. Probably need a more fleshed-out list of incompatible features.
* As Kirill noted, each 4k page ends up with a potential 1GB "blast radius" of demoted pages in the direct map. Not cool. This is probably a non-starter as it stands.
* The global TLB flushes are going to eat you alive. They probably border on a DoS on larger systems.

Quoted 8 lines

* Do we really want this user interface to dictate the kernel implementation? In other words, do we really want MAP_EXCLUSIVE, or do we want MAP_SECRET? One tells the kernel what do *do*, the other tells the kernel what the memory *IS*.
* There's a lot of other stuff going on in this area: XPFO, SEV, MKTME, Persistent Memory, where the kernel direct map is a liability in some way. We probably need some kind of overall, architected solution rather than five or ten things all poking at the direct map.

Quoted 3 lines

Another random set of thoughts:

- Should devices be permitted to DMA to/from MAP_SECRET pages?

I can't say I have a clear cut yes or no here. One possible use case for such pages is to read a secrets from storage directly into them. On the other side, DMA to/from a device can be used to exploit those secrets...

Quoted 1 line

- How about GUP?

Do you mean GUP for "remote" memory? I'd say no.

Quoted 1 line

- Can I ptrace my way into another process's secret pages?

No.

Quoted 1 line

- What if I splice() the page into a pipe?

I think it should fail.

Signature · 2 lines
Sincerely yours,
Mike.
author maintainer
#35 lore raw
In reply to Dave Hansen
Quoted 7 lines · Dave Hansen
Quoted 3 lines · Mike Rapoport

The pages in these mappings are removed from the kernel direct map and marked with PG_user_exclusive flag. When the exclusive area is unmapped, the pages are mapped back into the direct map.

This looks fun. It's certainly simple.

Quoted 7 lines

But, the description is not really calling out the pros and cons very well. I'm also not sure that folks will use an interface like this that requires up-front, special code to do an allocation instead of something like madvise(). That's why protection keys ended up the way it did: if you do this as a mmap() replacement, you need to modify all *allocators* to be enabled for this. If you do it with mprotect()-style, you can apply it to existing allocations.

Actually, I've started with mprotect() and then realized that mmap() would be simpler, so I switched over to mmap().

Quoted 5 lines

Some other random thoughts:

* The page flag is probably not a good idea. It would be probably better to set _PAGE_SPECIAL on the PTE and force get_user_pages() into the slow path.

The page flag won't work on 32-bit, indeed. But do we really need such functionality on 32-bit?

Quoted 3 lines

* This really stops being "normal" memory. You can't do futexes on it, cant splice it. Probably need a more fleshed-out list of incompatible features.

True, my bad. I should have mentioned more than THP/compaction/migration.

Quoted 5 lines

* As Kirill noted, each 4k page ends up with a potential 1GB "blast radius" of demoted pages in the direct map. Not cool. This is probably a non-starter as it stands.
* The global TLB flushes are going to eat you alive. They probably border on a DoS on larger systems.

As I wrote in another email, we could use some kind of pooling to reduce the "blast radius" and that will reduce that amount of TLB flushes as well. The size of the MAP_EXCLUSIVE obeys the RLIMIT_MEMLOCK and we can add a system-wide limit for size of such allocations.

Quoted 4 lines

* Do we really want this user interface to dictate the kernel implementation? In other words, do we really want MAP_EXCLUSIVE, or do we want MAP_SECRET? One tells the kernel what do *do*, the other tells the kernel what the memory *IS*.

I hesitated quite some time between EXCLUSIVE and SECRET. I've settled down on EXCLUSIVE because in my view that better describes the fact that the region is only mapped in its owner address space. And as such it can be used to store secrets, but it can be used for other purposes as well.

Quoted 4 lines

* There's a lot of other stuff going on in this area: XPFO, SEV, MKTME, Persistent Memory, where the kernel direct map is a liability in some way. We probably need some kind of overall, architected solution rather than five or ten things all poking at the direct map.

Agree.

Signature · 2 lines
Sincerely yours,
Mike.
maintainer
#36 lore raw
In reply to Mike Rapoport
Quoted 59 lines · Mike Rapoport

From: Mike Rapoport <[email protected]>

The mappings created with MAP_EXCLUSIVE are visible only in the context of the owning process and can be used by applications to store secret information that will not be visible not only to other processes but to the kernel as well.

The pages in these mappings are removed from the kernel direct map and marked with PG_user_exclusive flag. When the exclusive area is unmapped, the pages are mapped back into the direct map.

The MAP_EXCLUSIVE flag implies MAP_POPULATE and MAP_LOCKED.

---
 arch/x86/mm/fault.c                    | 14 ++++++++++
 fs/proc/task_mmu.c                     |  1 +
 include/linux/mm.h                     |  9 +++++++
 include/linux/page-flags.h             |  7 +++++
 include/linux/page_excl.h              | 49 ++++++++++++++++++++++++++++++++++
 include/trace/events/mmflags.h         |  9 ++++++-
 include/uapi/asm-generic/mman-common.h |  1 +
 kernel/fork.c                          |  3 ++-
 mm/Kconfig                             |  3 +++
 mm/gup.c                               |  8 ++++++
 mm/memory.c                            |  3 +++
 mm/mmap.c                              | 16 +++++++++++
 mm/page_alloc.c                        |  5 ++++
 13 files changed, 126 insertions(+), 2 deletions(-)
 create mode 100644 include/linux/page_excl.h
arch/x86/mm/fault.c +8 -0
index 9ceacd1..8f73a75 100644
@@ -17,6 +17,7 @@
 #include <linux/context_tracking.h>    /* exception_enter(), ...       */
 #include <linux/uaccess.h>             /* faulthandler_disabled()      */
 #include <linux/efi.h>                 /* efi_recover_from_page_fault()*/
+#include <linux/page_excl.h>           /* page_is_user_exclusive()     */
 #include <linux/mm_types.h>
 
 #include <asm/cpufeature.h>            /* boot_cpu_has, ...            */
@@ -1218,6 +1219,13 @@ static int fault_in_kernel_space(unsigned long address)
        return address >= TASK_SIZE_MAX;
 }
 
+static bool fault_in_user_exclusive_page(unsigned long address)
+{
+       struct page *page = virt_to_page(address);
+
+       return page_is_user_exclusive(page);
+}
+
 /*
  * Called for all faults where 'address' is part of the kernel address
  * space.  Might get called for faults that originate from *code* that

arch/x86/mm/fault.c do_kern_addr_fault(struct pt_regs *regs, unsigned long hw_error_code, · line 1276

Quoted 8 lines
arch/x86/mm/fault.c +5 -0
@@ -1261,6 +1269,12 @@ do_kern_addr_fault(struct pt_regs *regs, unsigned long hw_error_code,
        if (spurious_kernel_fault(hw_error_code, address))
                return;
 
+       /* FIXME: warn and handle gracefully */
+       if (unlikely(fault_in_user_exclusive_page(address))) {
+               pr_err("page fault in user exclusive page at %lx", address);
+               force_sig_fault(SIGSEGV, SEGV_MAPERR, (void __user *)address);
+       }

Sending a signal here is not a reasonable thing to do in response to an unexpected kernel fault. You need to OOPS. Printing a nice message would be nice.

--Andy

#37 lore raw
In reply to Mike Rapoport
Quoted 11 lines · Mike Rapoport

From: Mike Rapoport <[email protected]>

The mappings created with MAP_EXCLUSIVE are visible only in the context of the owning process and can be used by applications to store secret information that will not be visible not only to other processes but to the kernel as well.

The pages in these mappings are removed from the kernel direct map and marked with PG_user_exclusive flag. When the exclusive area is unmapped, the pages are mapped back into the direct map.

Just a thought, the kernel is still able to indirectly read the contents of these pages by doing a kdump from kexec environment, right?. Also, I wonder what would happen if you map such pages via /dev/mem into another user space application and e.g., use them along with kvm [1].

[1] https://lwn.net/Articles/778240/

Signature · 3 lines
Thanks,

David / dhildenb
author maintainer
#38 lore raw
Quoted 8 lines · David Hildenbrand
Quoted 7 lines · Mike Rapoport

From: Mike Rapoport <[email protected]>

The mappings created with MAP_EXCLUSIVE are visible only in the context of the owning process and can be used by applications to store secret information that will not be visible not only to other processes but to the kernel as well.

Quoted 7 lines
Quoted 4 lines

The pages in these mappings are removed from the kernel direct map and marked with PG_user_exclusive flag. When the exclusive area is unmapped, the pages are mapped back into the direct map.

Just a thought, the kernel is still able to indirectly read the contents of these pages by doing a kdump from kexec environment, right?

Right.

Quoted 3 lines

Also, I wonder
what would happen if you map such pages via /dev/mem into another user space application and e.g., use them along with kvm [1].

Do you mean that one application creates MAP_EXCLUSIVE and another applications accesses the same physical pages via /dev/mem?

With /dev/mem all physical memory is visible...

Quoted 8 lines

[1] https://lwn.net/Articles/778240/

Signature · 3 lines
Thanks,

David / dhildenb
Signature · 2 lines
Sincerely yours,
Mike.
#39 lore raw
Quoted 17 lines · Mike Rapoport
Quoted 8 lines · David Hildenbrand
Quoted 7 lines · Mike Rapoport

From: Mike Rapoport <[email protected]>

The mappings created with MAP_EXCLUSIVE are visible only in the context of the owning process and can be used by applications to store secret information that will not be visible not only to other processes but to the kernel as well.

Quoted 7 lines
Quoted 4 lines

The pages in these mappings are removed from the kernel direct map and marked with PG_user_exclusive flag. When the exclusive area is unmapped, the pages are mapped back into the direct map.

Just a thought, the kernel is still able to indirectly read the contents of these pages by doing a kdump from kexec environment, right?

Quoted 8 lines

Right.

Quoted 3 lines

Also, I wonder
what would happen if you map such pages via /dev/mem into another user space application and e.g., use them along with kvm [1].

Do you mean that one application creates MAP_EXCLUSIVE and another applications accesses the same physical pages via /dev/mem?

Exactly.

Quoted 2 lines

With /dev/mem all physical memory is visible...

Okay, so the statement "information that will not be visible not only to other processes but to the kernel as well" is not correct. There are easy ways to access that information if you really want to (might require root permissions, though).

Signature · 3 lines
Thanks,

David / dhildenb
author maintainer
#40 lore raw
Quoted 30 lines · David Hildenbrand
Quoted 17 lines · Mike Rapoport
Quoted 8 lines · David Hildenbrand
Quoted 7 lines · Mike Rapoport

From: Mike Rapoport <[email protected]>

The mappings created with MAP_EXCLUSIVE are visible only in the context of the owning process and can be used by applications to store secret information that will not be visible not only to other processes but to the kernel as well.

Quoted 7 lines
Quoted 4 lines

The pages in these mappings are removed from the kernel direct map and marked with PG_user_exclusive flag. When the exclusive area is unmapped, the pages are mapped back into the direct map.

Just a thought, the kernel is still able to indirectly read the contents of these pages by doing a kdump from kexec environment, right?

Quoted 8 lines

Right.

Quoted 3 lines

Also, I wonder
what would happen if you map such pages via /dev/mem into another user space application and e.g., use them along with kvm [1].

Do you mean that one application creates MAP_EXCLUSIVE and another applications accesses the same physical pages via /dev/mem?

Exactly.

Quoted 1 line
Quoted 6 lines
Quoted 1 line

With /dev/mem all physical memory is visible...

Okay, so the statement "information that will not be visible not only to other processes but to the kernel as well" is not correct. There are easy ways to access that information if you really want to (might require root permissions, though).

Right, but /dev/mem is an easy way to extract any information in any environment if one has root permissions...

Quoted 6 lines
Signature · 3 lines
Thanks,

David / dhildenb
Signature · 2 lines
Sincerely yours,
Mike.
#41 lore raw
Quoted 34 lines · Mike Rapoport
Quoted 32 lines · David Hildenbrand
Quoted 17 lines · Mike Rapoport
Quoted 8 lines · David Hildenbrand
Quoted 7 lines · Mike Rapoport

From: Mike Rapoport <[email protected]>

The mappings created with MAP_EXCLUSIVE are visible only in the context of the owning process and can be used by applications to store secret information that will not be visible not only to other processes but to the kernel as well.

Quoted 7 lines
Quoted 4 lines

The pages in these mappings are removed from the kernel direct map and marked with PG_user_exclusive flag. When the exclusive area is unmapped, the pages are mapped back into the direct map.

Just a thought, the kernel is still able to indirectly read the contents of these pages by doing a kdump from kexec environment, right?

Quoted 8 lines

Right.

Quoted 3 lines

Also, I wonder
what would happen if you map such pages via /dev/mem into another user space application and e.g., use them along with kvm [1].

Do you mean that one application creates MAP_EXCLUSIVE and another applications accesses the same physical pages via /dev/mem?

Exactly.

Quoted 2 lines

With /dev/mem all physical memory is visible...

Quoted 8 lines
Quoted 4 lines

Okay, so the statement "information that will not be visible not only to other processes but to the kernel as well" is not correct. There are easy ways to access that information if you really want to (might require root permissions, though).

Right, but /dev/mem is an easy way to extract any information in any environment if one has root permissions...

I don't understand this concern with /dev/mem. Just add these pages to the growing list of the things /dev/mem is not allowed to touch.

#42 lore raw
In reply to Mike Rapoport
reproduce: wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross chmod +x ~/bin/make.cross # save the attached .config to linux build tree…

Hi Mike,

[FYI, it's a private test report for your RFC patch.] [auto build test ERROR on mmotm/master]

url:    https://github.com/0day-ci/linux/commits/Mike-Rapoport/mm-add-MAP_EXCLUSIVE-to-create-exclusive-user-mappings/20191029-232454
base:   git://git.cmpxchg.org/linux-mmotm.git master
config: parisc-c3000_defconfig (attached as .config)
compiler: hppa-linux-gcc (GCC) 7.4.0

reproduce:
wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross chmod +x ~/bin/make.cross
# save the attached .config to linux build tree
GCC_VERSION=7.4.0 make.cross ARCH=parisc

If you fix the issue, kindly add following tag

All errors (new ones prefixed by >>):

mm/mmap.c: In function 'do_mmap':

>> mm/mmap.c:1577:14: error: 'MAP_EXCLUSIVE' undeclared (first use in this function); did you mean 'RMAP_EXCLUSIVE'?
     if (flags & MAP_EXCLUSIVE)
                 ^~~~~~~~~~~~~
                 RMAP_EXCLUSIVE
   mm/mmap.c:1577:14: note: each undeclared identifier is reported only once for each function it appears in
   mm/mmap.c: In function 'ksys_mmap_pgoff':
   mm/mmap.c:1597:14: error: 'MAP_EXCLUSIVE' undeclared (first use in this function); did you mean 'RMAP_EXCLUSIVE'?
     if (flags & MAP_EXCLUSIVE) {
                 ^~~~~~~~~~~~~
                 RMAP_EXCLUSIVE

vim +1577 mm/mmap.c

1393
1394 /*
1395 * The caller must hold down_write(&current->mm->mmap_sem).
1396 */
1397 unsigned long do_mmap(struct file *file, unsigned long addr,
1398 unsigned long len, unsigned long prot,
1399 unsigned long flags, vm_flags_t vm_flags,
1400 unsigned long pgoff, unsigned long *populate,
1401 struct list_head *uf)
1402 {
1403 struct mm_struct *mm = current->mm;
1404 int pkey = 0;
1405
1406 *populate = 0;
1407
1408 if (!len)
1409 return -EINVAL;
1410
1411 /*
1412 * Does the application expect PROT_READ to imply PROT_EXEC?
1413 *
1414 * (the exception is when the underlying filesystem is noexec
1415 * mounted, in which case we dont add PROT_EXEC.)
1416 */
1417 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
1418 if (!(file && path_noexec(&file->f_path)))
1419 prot |= PROT_EXEC;
1420
1421 /* force arch specific MAP_FIXED handling in get_unmapped_area */
1422 if (flags & MAP_FIXED_NOREPLACE)
1423 flags |= MAP_FIXED;
1424
1425 if (!(flags & MAP_FIXED))
1426 addr = round_hint_to_min(addr);
1427
1428 /* Careful about overflows.. */
1429 len = PAGE_ALIGN(len);
1430 if (!len)
1431 return -ENOMEM;
1432
1433 /* offset overflow? */
1434 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
1435 return -EOVERFLOW;
1436
1437 /* Too many mappings? */
1438 if (mm->map_count > sysctl_max_map_count)
1439 return -ENOMEM;
1440
1441 /* Obtain the address to map to. we verify (or select) it and ensure
1442 * that it represents a valid section of the address space.
1443 */
1444 addr = get_unmapped_area(file, addr, len, pgoff, flags);
1445 if (offset_in_page(addr))
1446 return addr;
1447
1448 if (flags & MAP_FIXED_NOREPLACE) {
1449 struct vm_area_struct *vma = find_vma(mm, addr);
1450
1451 if (vma && vma->vm_start < addr + len)
1452 return -EEXIST;
1453 }
1454
1455 if (prot == PROT_EXEC) {
1456 pkey = execute_only_pkey(mm);
1457 if (pkey < 0)
1458 pkey = 0;
1459 }
1460
1461 /* Do simple checking here so the lower-level routines won't have
1462 * to. we assume access permissions have been handled by the open
1463 * of the memory object, so we don't do any here.
1464 */
1465 vm_flags |= calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) |
1466 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1467
1468 if (flags & MAP_LOCKED)
1469 if (!can_do_mlock())
1470 return -EPERM;
1471
1472 if (mlock_future_check(mm, vm_flags, len))
1473 return -EAGAIN;
1474
1475 if (file) {
1476 struct inode *inode = file_inode(file);
1477 unsigned long flags_mask;
1478
1479 if (!file_mmap_ok(file, inode, pgoff, len))
1480 return -EOVERFLOW;
1481
1482 flags_mask = LEGACY_MAP_MASK | file->f_op->mmap_supported_flags;
1483
1484 switch (flags & MAP_TYPE) {
1485 case MAP_SHARED:
1486 /*
1487 * Force use of MAP_SHARED_VALIDATE with non-legacy
1488 * flags. E.g. MAP_SYNC is dangerous to use with
1489 * MAP_SHARED as you don't know which consistency model
1490 * you will get. We silently ignore unsupported flags
1491 * with MAP_SHARED to preserve backward compatibility.
1492 */
1493 flags &= LEGACY_MAP_MASK;
1494 /* fall through */
1495 case MAP_SHARED_VALIDATE:
1496 if (flags & ~flags_mask)
1497 return -EOPNOTSUPP;
1498 if (prot & PROT_WRITE) {
1499 if (!(file->f_mode & FMODE_WRITE))
1500 return -EACCES;
1501 if (IS_SWAPFILE(file->f_mapping->host))
1502 return -ETXTBSY;
1503 }
1504
1505 /*
1506 * Make sure we don't allow writing to an append-only
1507 * file..
1508 */
1509 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1510 return -EACCES;
1511
1512 /*
1513 * Make sure there are no mandatory locks on the file.
1514 */
1515 if (locks_verify_locked(file))
1516 return -EAGAIN;
1517
1518 vm_flags |= VM_SHARED | VM_MAYSHARE;
1519 if (!(file->f_mode & FMODE_WRITE))
1520 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1521
1522 /* fall through */
1523 case MAP_PRIVATE:
1524 if (!(file->f_mode & FMODE_READ))
1525 return -EACCES;
1526 if (path_noexec(&file->f_path)) {
1527 if (vm_flags & VM_EXEC)
1528 return -EPERM;
1529 vm_flags &= ~VM_MAYEXEC;
1530 }
1531
1532 if (!file->f_op->mmap)
1533 return -ENODEV;
1534 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1535 return -EINVAL;
1536 break;
1537
1538 default:
1539 return -EINVAL;
1540 }
1541 } else {
1542 switch (flags & MAP_TYPE) {
1543 case MAP_SHARED:
1544 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1545 return -EINVAL;
1546 /*
1547 * Ignore pgoff.
1548 */
1549 pgoff = 0;
1550 vm_flags |= VM_SHARED | VM_MAYSHARE;
1551 break;
1552 case MAP_PRIVATE:
1553 /*
1554 * Set pgoff according to addr for anon_vma.
1555 */
1556 pgoff = addr >> PAGE_SHIFT;
1557 break;
1558 default:
1559 return -EINVAL;
1560 }
1561 }
1562
1563 /*
1564 * Set 'VM_NORESERVE' if we should not account for the
1565 * memory use of this mapping.
1566 */
1567 if (flags & MAP_NORESERVE) {
1568 /* We honor MAP_NORESERVE if allowed to overcommit */
1569 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1570 vm_flags |= VM_NORESERVE;
1571
1572 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1573 if (file && is_file_hugepages(file))
1574 vm_flags |= VM_NORESERVE;
1575 }
1576

> 1577		if (flags & MAP_EXCLUSIVE)

1578 vm_flags |= VM_EXCLUSIVE;
1579
1580 addr = mmap_region(file, addr, len, vm_flags, pgoff, uf);
1581 if (!IS_ERR_VALUE(addr) &&
1582 ((vm_flags & VM_LOCKED) ||
1583 (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
1584 *populate = len;
1585 return addr;
1586 }
1587

---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all Intel Corporation

#43 lore raw
In reply to Mike Rapoport

* Mike Rapoport:

Quoted 4 lines

The patch below aims to allow applications to create mappins that have pages visible only to the owning process. Such mappings could be used to store secrets so that these secrets are not visible neither to other processes nor to the kernel.

How is this expected to interact with CRIU?

Quoted 2 lines

I've only tested the basic functionality, the changes should be verified against THP/migration/compaction. Yet, I'd appreciate early feedback.

What are the expected semantics for VM migration? Should it fail?

author maintainer
#44 lore raw
Quoted 10 lines · Florian Weimer

* Mike Rapoport:

Quoted 1 line

The patch below aims to allow applications to create mappins that

have

Quoted 1 line

pages visible only to the owning process. Such mappings could be used

to

Quoted 2 lines

store secrets so that these secrets are not visible neither to other processes nor to the kernel.

How is this expected to interact with CRIU?

CRIU dumps the memory contents using a parasite code from inside the dumpee address space, so it would work the same way as for the other mappings. Of course, at the restore time the exclusive mapping should be recreated with the appropriate flags.

Quoted 5 lines
Quoted 1 line

I've only tested the basic functionality, the changes should be

verified

Quoted 1 line

against THP/migration/compaction. Yet, I'd appreciate early feedback.

What are the expected semantics for VM migration? Should it fail?

I don't quite follow. If qemu would use such mappings it would be able to transfer them during live migration.

Signature · 2 lines
Sincerely yours,
Mike
#45 lore raw

* Mike Rapoport:

Quoted 11 lines
Quoted 9 lines · Florian Weimer

* Mike Rapoport:

Quoted 1 line

The patch below aims to allow applications to create mappins that

have

Quoted 1 line

pages visible only to the owning process. Such mappings could be used

to

Quoted 2 lines

store secrets so that these secrets are not visible neither to other processes nor to the kernel.

Quoted 6 lines
Quoted 1 line

How is this expected to interact with CRIU?

CRIU dumps the memory contents using a parasite code from inside the dumpee address space, so it would work the same way as for the other mappings. Of course, at the restore time the exclusive mapping should be recreated with the appropriate flags.

Hmm, so it would use a bounce buffer to perform the extraction?

Quoted 8 lines
Quoted 5 lines
Quoted 1 line

I've only tested the basic functionality, the changes should be

verified

Quoted 1 line

against THP/migration/compaction. Yet, I'd appreciate early feedback.

What are the expected semantics for VM migration? Should it fail?

I don't quite follow. If qemu would use such mappings it would be able to transfer them during live migration.

I was wondering if the special state is supposed to bubble up to the host eventually.

author maintainer
#46 lore raw
Quoted 13 lines · Florian Weimer

* Mike Rapoport:

Quoted 11 lines
Quoted 9 lines · Florian Weimer

* Mike Rapoport:

Quoted 1 line

The patch below aims to allow applications to create mappins that

have

Quoted 1 line

pages visible only to the owning process. Such mappings could be used

to

Quoted 2 lines

store secrets so that these secrets are not visible neither to other processes nor to the kernel.

Quoted 8 lines
Quoted 6 lines
Quoted 1 line

How is this expected to interact with CRIU?

CRIU dumps the memory contents using a parasite code from inside the dumpee address space, so it would work the same way as for the other mappings. Of course, at the restore time the exclusive mapping should be recreated with the appropriate flags.

Hmm, so it would use a bounce buffer to perform the extraction?

At first I thought that CRIU would extract the memory contents from these mappings just as it does now using vmsplice(). But it seems that such mappings won't play well with pipes, so CRIU will need a bounce buffer indeed.

Quoted 11 lines
Quoted 8 lines
Quoted 5 lines
Quoted 1 line

I've only tested the basic functionality, the changes should be

verified

Quoted 1 line

against THP/migration/compaction. Yet, I'd appreciate early feedback.

What are the expected semantics for VM migration? Should it fail?

I don't quite follow. If qemu would use such mappings it would be able to transfer them during live migration.

I was wondering if the special state is supposed to bubble up to the host eventually.

Well, that was not intended.

Signature · 2 lines
Sincerely yours,
Mike.
#47 lore raw
In reply to Mike Rapoport
Quoted 7 lines · Mike Rapoport

From: Mike Rapoport <[email protected]>

Hi,

Quoted 7 lines

The patch below aims to allow applications to create mappins that have pages visible only to the owning process. Such mappings could be used to store secrets so that these secrets are not visible neither to other processes nor to the kernel.

I've only tested the basic functionality, the changes should be verified against THP/migration/compaction. Yet, I'd appreciate early feedback.

I’ve contemplated the concept a fair amount, and I think you should consider a change to the API. In particular, rather than having it be a MAP_ flag, make it a chardev. You can, at least at first, allow only MAP_SHARED, and admins can decide who gets to use it. It might also play better with the VM overall, and you won’t need a VM_ flag for it — you can just wire up .fault to do the right thing.

author maintainer
#48 lore raw
Quoted 15 lines · Andy Lutomirski
Quoted 7 lines · Mike Rapoport

From: Mike Rapoport <[email protected]>

Hi,

Quoted 7 lines

The patch below aims to allow applications to create mappins that have pages visible only to the owning process. Such mappings could be used to store secrets so that these secrets are not visible neither to other processes nor to the kernel.

I've only tested the basic functionality, the changes should be verified against THP/migration/compaction. Yet, I'd appreciate early feedback.

Quoted 6 lines

I’ve contemplated the concept a fair amount, and I think you should consider a change to the API. In particular, rather than having it be a MAP_ flag, make it a chardev. You can, at least at first, allow only MAP_SHARED, and admins can decide who gets to use it. It might also play better with the VM overall, and you won’t need a VM_ flag for it — you can just wire up .fault to do the right thing.

I think mmap()/mprotect()/madvise() are the natural APIs for such interface. Switching to a chardev doesn't solve the major problem of direct map fragmentation and defeats the ability to use exclusive memory mappings with the existing allocators, while mprotect() and madvise() do not.

Signature · 2 lines
Sincerely yours,
Mike.
maintainer
#49 lore raw
Quoted 24 lines · Mike Rapoport
Quoted 15 lines · Andy Lutomirski
Quoted 7 lines · Mike Rapoport

From: Mike Rapoport <[email protected]>

Hi,

Quoted 7 lines

The patch below aims to allow applications to create mappins that have pages visible only to the owning process. Such mappings could be used to store secrets so that these secrets are not visible neither to other processes nor to the kernel.

I've only tested the basic functionality, the changes should be verified against THP/migration/compaction. Yet, I'd appreciate early feedback.

Quoted 6 lines

I’ve contemplated the concept a fair amount, and I think you should consider a change to the API. In particular, rather than having it be a MAP_ flag, make it a chardev. You can, at least at first, allow only MAP_SHARED, and admins can decide who gets to use it. It might also play better with the VM overall, and you won’t need a VM_ flag for it — you can just wire up .fault to do the right thing.

Quoted 2 lines

I think mmap()/mprotect()/madvise() are the natural APIs for such interface.

Then you have a whole bunch of questions to answer. For example:

What happens if you mprotect() or similar when the mapping is already in use in a way that's incompatible with MAP_EXCLUSIVE?

Is it actually reasonable to malloc() some memory and then make it exclusive?

Are you permitted to map a file MAP_EXCLUSIVE? What does it mean?

What does MAP_PRIVATE | MAP_EXCLUSIVE do?

How does one pass exclusive memory via SCM_RIGHTS? (If it's a memfd-like or chardev interface, it's trivial. mmap(), not so much.)

And finally, there's my personal giant pet peeve: a major use of this will be for virtualization. I suspect that a lot of people would like the majority of KVM guest memory to be unmapped from the host pagetables. But people might also like for guest memory to be unmapped in *QEMU's* pagetables, and mmap() is a basically worthless interface for this. Getting fd-backed memory into a guest will take some possibly major work in the kernel, but getting vma-backed memory into a guest without mapping it in the host user address space seems much, much worse.

Quoted 4 lines

Switching to a chardev doesn't solve the major problem of direct map fragmentation and defeats the ability to use exclusive memory mappings with the existing allocators, while mprotect() and madvise() do not.

Will people really want to do malloc() and then remap it exclusive? This sounds dubiously useful at best.

author maintainer
#50 lore raw
Quoted 25 lines · Andy Lutomirski
Quoted 24 lines · Mike Rapoport
Quoted 21 lines · Andy Lutomirski
Quoted 7 lines · Mike Rapoport

From: Mike Rapoport <[email protected]>

Hi,

Quoted 7 lines

The patch below aims to allow applications to create mappins that have pages visible only to the owning process. Such mappings could be used to store secrets so that these secrets are not visible neither to other processes nor to the kernel.

I've only tested the basic functionality, the changes should be verified against THP/migration/compaction. Yet, I'd appreciate early feedback.

I’ve contemplated the concept a fair amount, and I think you should consider a change to the API. In particular, rather than having it be a MAP_ flag, make it a chardev. You can, at least at first, allow only MAP_SHARED, and admins can decide who gets to use it. It might also play better with the VM overall, and you won’t need a VM_ flag for it — you can just wire up .fault to do the right thing.

Quoted 7 lines
Quoted 2 lines

I think mmap()/mprotect()/madvise() are the natural APIs for such interface.

Then you have a whole bunch of questions to answer. For example:

What happens if you mprotect() or similar when the mapping is already in use in a way that's incompatible with MAP_EXCLUSIVE?

Then we refuse to mprotect()? Like in any other case when vm_flags are not compatible with required madvise()/mprotect() operation.

Quoted 3 lines

Is it actually reasonable to malloc() some memory and then make it exclusive?

Are you permitted to map a file MAP_EXCLUSIVE? What does it mean?

I'd limit MAP_EXCLUSIVE only to anonymous memory.

Quoted 1 line

What does MAP_PRIVATE | MAP_EXCLUSIVE do?

My preference is to have only mmap() and then the semantics is more clear:

MAP_PRIVATE | MAP_EXCLUSIVE creates a pre-populated region, marks it locked and drops the pages in this region from the direct map.
The pages are returned back on munmap().
Then there is no way to change an existing area to be exclusive or vice versa.

Quoted 2 lines

How does one pass exclusive memory via SCM_RIGHTS? (If it's a memfd-like or chardev interface, it's trivial. mmap(), not so much.)

Why passing such memory via SCM_RIGHTS would be useful?

Quoted 9 lines

And finally, there's my personal giant pet peeve: a major use of this will be for virtualization. I suspect that a lot of people would like the majority of KVM guest memory to be unmapped from the host pagetables. But people might also like for guest memory to be unmapped in *QEMU's* pagetables, and mmap() is a basically worthless interface for this. Getting fd-backed memory into a guest will take some possibly major work in the kernel, but getting vma-backed memory into a guest without mapping it in the host user address space seems much, much worse.

Well, in my view, the MAP_EXCLUSIVE is intended to keep small secrets rather than use it for the entire guest memory. I even considered adding a limit for the mapping size, but then I decided that since RLIMIT_MEMLOCK is anyway enforced there is no need for a new one.

I agree that getting fd-backed memory into a guest would be less pain that VMA, but KVM can already use memory outside the control of the kernel via /dev/map [1].

So unless I'm missing something here, there is no need to use MAP_EXCLUSIVE for the guest memory.

[1] https://lwn.net/Articles/778240/

Quoted 7 lines
Quoted 4 lines

Switching to a chardev doesn't solve the major problem of direct map fragmentation and defeats the ability to use exclusive memory mappings with the existing allocators, while mprotect() and madvise() do not.

Will people really want to do malloc() and then remap it exclusive? This sounds dubiously useful at best.

Again, my preference is to have mmap() only, but I see a value in this use case as well. Application developers allocate memory and then sometimes change its properties rather than go mmap() something. For such usage mprotect() may be usefull.

Signature · 2 lines
Sincerely yours,
Mike.
maintainer
#51 lore raw
Quoted 30 lines · Mike Rapoport
Quoted 28 lines · Andy Lutomirski
Quoted 26 lines · Mike Rapoport
Quoted 15 lines · Andy Lutomirski
Quoted 7 lines · Mike Rapoport

From: Mike Rapoport <[email protected]>

Hi,

Quoted 7 lines

The patch below aims to allow applications to create mappins that have pages visible only to the owning process. Such mappings could be used to store secrets so that these secrets are not visible neither to other processes nor to the kernel.

I've only tested the basic functionality, the changes should be verified against THP/migration/compaction. Yet, I'd appreciate early feedback.

Quoted 6 lines

I’ve contemplated the concept a fair amount, and I think you should consider a change to the API. In particular, rather than having it be a MAP_ flag, make it a chardev. You can, at least at first, allow only MAP_SHARED, and admins can decide who gets to use it. It might also play better with the VM overall, and you won’t need a VM_ flag for it — you can just wire up .fault to do the right thing.

I think mmap()/mprotect()/madvise() are the natural APIs for such interface.

Quoted 8 lines
Quoted 4 lines

Then you have a whole bunch of questions to answer. For example:

What happens if you mprotect() or similar when the mapping is already in use in a way that's incompatible with MAP_EXCLUSIVE?

Then we refuse to mprotect()? Like in any other case when vm_flags are not compatible with required madvise()/mprotect() operation.

I'm not talking about flags. I'm talking about the case where one thread (or RDMA or whatever) has get_user_pages()'d a mapping and another thread mprotect()s it MAP_EXCLUSIVE.

Quoted 8 lines
Quoted 3 lines

Is it actually reasonable to malloc() some memory and then make it exclusive?

Are you permitted to map a file MAP_EXCLUSIVE? What does it mean?

I'd limit MAP_EXCLUSIVE only to anonymous memory.

Quoted 1 line

What does MAP_PRIVATE | MAP_EXCLUSIVE do?

Quoted 7 lines

My preference is to have only mmap() and then the semantics is more clear:

MAP_PRIVATE | MAP_EXCLUSIVE creates a pre-populated region, marks it locked and drops the pages in this region from the direct map.
The pages are returned back on munmap().
Then there is no way to change an existing area to be exclusive or vice versa.

And what happens if you fork()? Limiting it to MAP_SHARED | MAP_EXCLUSIVE would about this particular nasty question.

Quoted 5 lines
Quoted 3 lines

How does one pass exclusive memory via SCM_RIGHTS? (If it's a memfd-like or chardev interface, it's trivial. mmap(), not so much.)

Why passing such memory via SCM_RIGHTS would be useful?

Suppose I want to put a secret into exclusive memory and then send that secret to some other process. The obvious approach would be to SCM_RIGHTS an fd over, but you can't do that with MAP_EXCLUSIVE as you've defined it. In general, there are lots of use cases for memfd and other fd-backed memory.

Quoted 11 lines
Quoted 10 lines

And finally, there's my personal giant pet peeve: a major use of this will be for virtualization. I suspect that a lot of people would like the majority of KVM guest memory to be unmapped from the host pagetables. But people might also like for guest memory to be unmapped in *QEMU's* pagetables, and mmap() is a basically worthless interface for this. Getting fd-backed memory into a guest will take some possibly major work in the kernel, but getting vma-backed memory into a guest without mapping it in the host user address space seems much, much worse.

Quoted 8 lines

Well, in my view, the MAP_EXCLUSIVE is intended to keep small secrets rather than use it for the entire guest memory. I even considered adding a limit for the mapping size, but then I decided that since RLIMIT_MEMLOCK is anyway enforced there is no need for a new one.

I agree that getting fd-backed memory into a guest would be less pain that VMA, but KVM can already use memory outside the control of the kernel via /dev/map [1].

That series doesn't address the problem I'm talking about at all. I'm saying that there is a legitimate use case where QEMU should *not* have a mapping of the memory. So QEMU would create some exclusive memory using /dev/exclusive_memory and would tell KVM to map it into the guest without mapping it into QEMU's address space at all.

(In fact, the way that SEV currently works is *functionally* like this, except that there's a bogus incoherent mapping in the QEMU process that is a giant can of worms.

IMO a major benefit of a chardev approach is that you don't need a new VM_ flag and you don't need to worry about wiring it up everywhere in the core mm code.

author maintainer
#52 lore raw
Quoted 101 lines · Andy Lutomirski
Quoted 30 lines · Mike Rapoport
Quoted 28 lines · Andy Lutomirski
Quoted 26 lines · Mike Rapoport
Quoted 15 lines · Andy Lutomirski
Quoted 7 lines · Mike Rapoport

From: Mike Rapoport <[email protected]>

Hi,

Quoted 7 lines

The patch below aims to allow applications to create mappins that have pages visible only to the owning process. Such mappings could be used to store secrets so that these secrets are not visible neither to other processes nor to the kernel.

I've only tested the basic functionality, the changes should be verified against THP/migration/compaction. Yet, I'd appreciate early feedback.

Quoted 6 lines

I’ve contemplated the concept a fair amount, and I think you should consider a change to the API. In particular, rather than having it be a MAP_ flag, make it a chardev. You can, at least at first, allow only MAP_SHARED, and admins can decide who gets to use it. It might also play better with the VM overall, and you won’t need a VM_ flag for it — you can just wire up .fault to do the right thing.

I think mmap()/mprotect()/madvise() are the natural APIs for such interface.

Quoted 8 lines
Quoted 4 lines

Then you have a whole bunch of questions to answer. For example:

What happens if you mprotect() or similar when the mapping is already in use in a way that's incompatible with MAP_EXCLUSIVE?

Then we refuse to mprotect()? Like in any other case when vm_flags are not compatible with required madvise()/mprotect() operation.

I'm not talking about flags. I'm talking about the case where one thread (or RDMA or whatever) has get_user_pages()'d a mapping and another thread mprotect()s it MAP_EXCLUSIVE.

Quoted 8 lines
Quoted 3 lines

Is it actually reasonable to malloc() some memory and then make it exclusive?

Are you permitted to map a file MAP_EXCLUSIVE? What does it mean?

I'd limit MAP_EXCLUSIVE only to anonymous memory.

Quoted 1 line

What does MAP_PRIVATE | MAP_EXCLUSIVE do?

Quoted 7 lines

My preference is to have only mmap() and then the semantics is more clear:

MAP_PRIVATE | MAP_EXCLUSIVE creates a pre-populated region, marks it locked and drops the pages in this region from the direct map.
The pages are returned back on munmap().
Then there is no way to change an existing area to be exclusive or vice versa.

And what happens if you fork()? Limiting it to MAP_SHARED | MAP_EXCLUSIVE would about this particular nasty question.

Quoted 5 lines
Quoted 3 lines

How does one pass exclusive memory via SCM_RIGHTS? (If it's a memfd-like or chardev interface, it's trivial. mmap(), not so much.)

Why passing such memory via SCM_RIGHTS would be useful?

Suppose I want to put a secret into exclusive memory and then send that secret to some other process. The obvious approach would be to SCM_RIGHTS an fd over, but you can't do that with MAP_EXCLUSIVE as you've defined it. In general, there are lots of use cases for memfd and other fd-backed memory.

Quoted 11 lines
Quoted 10 lines

And finally, there's my personal giant pet peeve: a major use of this will be for virtualization. I suspect that a lot of people would like the majority of KVM guest memory to be unmapped from the host pagetables. But people might also like for guest memory to be unmapped in *QEMU's* pagetables, and mmap() is a basically worthless interface for this. Getting fd-backed memory into a guest will take some possibly major work in the kernel, but getting vma-backed memory into a guest without mapping it in the host user address space seems much, much worse.

Quoted 8 lines

Well, in my view, the MAP_EXCLUSIVE is intended to keep small secrets rather than use it for the entire guest memory. I even considered adding a limit for the mapping size, but then I decided that since RLIMIT_MEMLOCK is anyway enforced there is no need for a new one.

I agree that getting fd-backed memory into a guest would be less pain that VMA, but KVM can already use memory outside the control of the kernel via /dev/map [1].

That series doesn't address the problem I'm talking about at all. I'm saying that there is a legitimate use case where QEMU should *not* have a mapping of the memory. So QEMU would create some exclusive memory using /dev/exclusive_memory and would tell KVM to map it into the guest without mapping it into QEMU's address space at all.

Quoted 8 lines

(In fact, the way that SEV currently works is *functionally* like this, except that there's a bogus incoherent mapping in the QEMU process that is a giant can of worms.

IMO a major benefit of a chardev approach is that you don't need a new VM_ flag and you don't need to worry about wiring it up everywhere in the core mm code.

Ok, at last I'm starting to see your and Christoph's point.

Just to reiterate, we can use fd-backed memory using /dev/exclusive_memory chardev (or some other name we'll pick after long bikeshedding) and then the .mmap method of this character device can do interesting things with the backing physical memory. Since the memory is not VMA-mapped, we do not have to find all the places in the core that might require a check of a VM_ flag to ensure there is no clashes with the exclusive memory.

Still, whatever we do with the mapping properties of this memory, we need a solution to the splitting of huge pages that map the direct map, but this is an orthogonal problem in a way.

Signature · 2 lines
Sincerely yours,
Mike.
author maintainer
#53 lore raw
In reply to Andy Lutomirski
Quoted 4 lines · Andy Lutomirski

IMO a major benefit of a chardev approach is that you don't need a new VM_ flag and you don't need to worry about wiring it up everywhere in the core mm code.

I've done a couple of experiments with secret/exclusive/whatever memory backed by a file-descriptor using a chardev and memfd_create syscall. There is indeed no need for VM_ flag, but there are still places that would require special care, e.g vm_normal_page(), madvise(DO_FORK), so it won't be completely free of core mm modifications.

Below is a POC that implements extension to memfd_create() that allows mapping of a "secret" memory. The "secrecy" mode should be explicitly set using ioctl(), for now I've implemented exclusive and uncached mappings.

The POC primarily indented to illustrate a possible userspace API for fd-based secret memory. The idea is that user will create a file descriptor using a system call. The user than has to use ioctl() to define the desired mode of operation and only when the mode is set it is possible to mmap() the memory. I.e something like

	fd = memfd_create("secret", MFD_SECRET);
	ioctl(fd, MFD_SECRET_UNCACHED);
	ptr = mmap(NULL, MAP_SIZE, PROT_READ | PROT_WRITE, MAP_SHARED,
		   fd, 0);

The ioctl() allows a lot of flexibility in how the secrecy should be defined. It could be either a request for a particular protection (e.g. exclusive, uncached) or something like "secrecy level" from "a bit more secret than normally" to "do your best even at the expense of performance". The POC implements the first option and the modes are mutually exclusive for now, but there is no fundamental reason they cannot be mixed.

I've chosen memfd over a chardev as it seem to play more neatly with anon_inodes and would allow simple (ab)use of the page cache for tracking pages allocated for the "secret" mappings as well as using address_space_operations for e.g. page migration callbacks.

The POC implementation uses set_memory/pageattr APIs to manipulate the direct map and does not address the direct map fragmentation issue.

Of course this is something that must be addresses, as well as modifications to core mm to required keep the secret memory secret, but I'd really like to focus on the userspace ABI first.

Quoted 1 line

From 0e1bd1b63f38685ffc5aab8c89b31086bde3da7b Mon Sep 17 00:00:00 2001

From: Mike Rapoport <[email protected]>
Date: Mon, 18 Nov 2019 09:32:22 +0200
Subject: [PATCH] mm: extend memfd with ability to create secret memory

---
 include/linux/memfd.h      |   9 ++
 include/uapi/linux/magic.h |   1 +
 include/uapi/linux/memfd.h |   6 +
 mm/Kconfig                 |   3 +
 mm/Makefile                |   1 +
 mm/memfd.c                 |  10 +-
 mm/secretmem.c             | 233 +++++++++++++++++++++++++++++++++++++
 7 files changed, 261 insertions(+), 2 deletions(-)
 create mode 100644 mm/secretmem.c
include/linux/memfd.h +9 -0
index 4f1600413f91..d3ca7285f51a 100644
@@ -13,4 +13,13 @@ static inline long memfd_fcntl(struct file *f, unsigned int c, unsigned long a)
 }
 #endif
 
+#ifdef CONFIG_MEMFD_SECRETMEM
+extern struct file *secretmem_file_create(const char *name, unsigned int flags);
+#else
+static inline struct file *secretmem_file_create(const char *name, unsigned int flags)
+{
+	return ERR_PTR(-EINVAL);
+}
+#endif
+
 #endif /* __LINUX_MEMFD_H */
include/uapi/linux/magic.h +1 -0
index 903cc2d2750b..3dad6208c8de 100644
@@ -94,5 +94,6 @@
 #define ZSMALLOC_MAGIC		0x58295829
 #define DMA_BUF_MAGIC		0x444d4142	/* "DMAB" */
 #define Z3FOLD_MAGIC		0x33
+#define SECRETMEM_MAGIC		0x5345434d	/* "SECM" */
 
 #endif /* __LINUX_MAGIC_H__ */
include/uapi/linux/memfd.h +6 -0
index 7a8a26751c23..3320a79b638d 100644
@@ -8,6 +8,12 @@
 #define MFD_CLOEXEC		0x0001U
 #define MFD_ALLOW_SEALING	0x0002U
 #define MFD_HUGETLB		0x0004U
+#define MFD_SECRET		0x0008U
+
+/* ioctls for secret memory */
+#define MFD_SECRET_IOCTL '-'
+#define MFD_SECRET_EXCLUSIVE	_IOW(MFD_SECRET_IOCTL, 0x13, unsigned long)
+#define MFD_SECRET_UNCACHED	_IOW(MFD_SECRET_IOCTL, 0x14, unsigned long)
 
 /*
  * Huge page size encoding when MFD_HUGETLB is specified, and a huge page
mm/Kconfig +3 -0
index a5dae9a7eb51..aa828f240287 100644
@@ -736,4 +736,7 @@ config ARCH_HAS_PTE_SPECIAL
 config ARCH_HAS_HUGEPD
 	bool
 
+config MEMFD_SECRETMEM
+        def_bool MEMFD_CREATE && ARCH_HAS_SET_DIRECT_MAP
+
 endmenu
mm/Makefile +1 -0
index d996846697ef..54cb8a60d698 100644
@@ -107,3 +107,4 @@ obj-$(CONFIG_PERCPU_STATS) += percpu-stats.o
 obj-$(CONFIG_ZONE_DEVICE) += memremap.o
 obj-$(CONFIG_HMM_MIRROR) += hmm.o
 obj-$(CONFIG_MEMFD_CREATE) += memfd.o
+obj-$(CONFIG_MEMFD_SECRETMEM) += secretmem.o
mm/memfd.c +8 -2
index 2647c898990c..3e1cc37e0389 100644
@@ -245,7 +245,8 @@ long memfd_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
 
-#define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING | MFD_HUGETLB)
+#define MFD_SECRET_MASK (MFD_CLOEXEC | MFD_SECRET)
+#define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING | MFD_HUGETLB | MFD_SECRET)
 
 SYSCALL_DEFINE2(memfd_create,
 		const char __user *, uname,
@@ -257,6 +258,9 @@ SYSCALL_DEFINE2(memfd_create,
 	char *name;
 	long len;
 
+	if (flags & ~(unsigned int)MFD_SECRET_MASK)
+		return -EINVAL;
+
 	if (!(flags & MFD_HUGETLB)) {
 		if (flags & ~(unsigned int)MFD_ALL_FLAGS)
 			return -EINVAL;
@@ -296,7 +300,9 @@ SYSCALL_DEFINE2(memfd_create,
 		goto err_name;
 	}
 
-	if (flags & MFD_HUGETLB) {
+	if (flags & MFD_SECRET) {
+		file = secretmem_file_create(name, flags);
+	} else if (flags & MFD_HUGETLB) {
 		struct user_struct *user = NULL;
 
 		file = hugetlb_file_setup(name, 0, VM_NORESERVE, &user,
mm/secretmem.c new file +233 -0

233 lines not on this page · raw message

Signature · 5 lines
2.24.0

-- 
Sincerely yours,
Mike.
#54 lore raw

Re: [PATCH] mm: extend memfd with ability to create secret memory

I love your patch! Yet something to improve:

Hi Mike,

I love your patch! Yet something to improve:

[auto build test ERROR on mmotm/master]
[also build test ERROR on linux/master v5.4]
[cannot apply to arm-soc/for-next linus/master next-20191208]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system. BTW, we also suggest to use '--base' option to specify the base tree in git format-patch, please see https://stackoverflow.com/a/37406982]

url:    https://github.com/0day-ci/linux/commits/Mike-Rapoport/mm-extend-memfd-with-ability-to-create-secret-memory/20191207-130906
base:   git://git.cmpxchg.org/linux-mmotm.git master
config: arm64-randconfig-a001-20191208 (attached as .config)
compiler: aarch64-linux-gcc (GCC) 7.5.0

reproduce:
wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross chmod +x ~/bin/make.cross
# save the attached .config to linux build tree
GCC_VERSION=7.5.0 make.cross ARCH=arm64

If you fix the issue, kindly add following tag

All errors (new ones prefixed by >>):

mm/secretmem.c: In function 'secretmem_fault':

>> mm/secretmem.c:45:9: error: implicit declaration of function 'set_pages_array_uc'; did you mean 'set_page_dirty_lock'? [-Werror=implicit-function-declaration]
      err = set_pages_array_uc(&page, 1);
            ^~~~~~~~~~~~~~~~~~
            set_page_dirty_lock
   mm/secretmem.c: In function 'secretmem_close':
>> mm/secretmem.c:75:4: error: implicit declaration of function 'set_pages_array_wb'; did you mean 'set_page_dirty'? [-Werror=implicit-function-declaration]
       set_pages_array_wb(&page, 1);
       ^~~~~~~~~~~~~~~~~~
       set_page_dirty

cc1: some warnings being treated as errors

vim +45 mm/secretmem.c

    21	
    22	static vm_fault_t secretmem_fault(struct vm_fault *vmf)
    23	{
    24		struct secretmem_state *state = vmf->vma->vm_file->private_data;
    25		struct address_space *mapping = vmf->vma->vm_file->f_mapping;
    26		pgoff_t offset = vmf->pgoff;
    27		unsigned long addr;
    28		struct page *page;
    29		int err;
    30	
    31		page = find_get_page(mapping, offset);
    32		if (!page) {
    33			page = pagecache_get_page(mapping, offset,
    34						  FGP_CREAT|FGP_FOR_MMAP,
    35						  vmf->gfp_mask);
    36			if (!page)
    37				return vmf_error(-ENOMEM);
    38	
    39			__SetPageUptodate(page);
    40		}
    41	
    42		if (state->mode == SECRETMEM_EXCLUSIVE)
    43			err = set_direct_map_invalid_noflush(page);
    44		else if (state->mode == SECRETMEM_UNCACHED)
  > 45			err = set_pages_array_uc(&page, 1);
    46		else
    47			BUG();
    48	
    49		if (err) {
    50			delete_from_page_cache(page);
    51			return vmf_error(err);
    52		}
    53	
    54		addr = (unsigned long)page_address(page);
    55		flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
    56	
    57		vmf->page = page;
    58		return  0;
    59	}
    60	
    61	static void secretmem_close(struct vm_area_struct *vma)
    62	{
    63		struct secretmem_state *state = vma->vm_file->private_data;
    64		struct address_space *mapping = vma->vm_file->f_mapping;
    65		struct page *page;
    66		pgoff_t index;
    67	
    68		xa_for_each(&mapping->i_pages, index, page) {
    69			get_page(page);
    70			lock_page(page);
    71	
    72			if (state->mode == SECRETMEM_EXCLUSIVE)
    73				set_direct_map_default_noflush(page);
    74			else if (state->mode == SECRETMEM_UNCACHED)
  > 75				set_pages_array_wb(&page, 1);
    76			else
    77				BUG();
    78	
    79			__ClearPageDirty(page);
    80			delete_from_page_cache(page);
    81	
    82			unlock_page(page);
    83			put_page(page);
    84		}
    85	}
    86	

---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/hyperkitty/list/[email protected] Intel Corporation

#55 lore raw
In reply to Mike Rapoport
Quoted 4 lines

The patch below aims to allow applications to create mappins that have pages visible only to the owning process. Such mappings could be used to store secrets so that these secrets are not visible neither to other processes nor to the kernel.

Hi Mike,

I have actually been looking into the closely related problem for the past couple of weeks (on and off). What is common here is the need for userspace to indicate to kernel that some pages contain secrets. And then there are actually a number of things that kernel can do to try to protect these secrets better. Unmap from direct map is one of them. Another thing is to map such pages as non-cached, which can help us to prevent or considerably restrict speculation on such pages. The initial proof of concept for marking pages as "UNCACHED" that I got from Dave Hansen was actually based on mlock2() and a new flag for it for this purpose. Since then I have been thinking on what interface suits the use case better and actually selected going with new madvise() flag instead because of all possible implications for fragmentation and performance. My logic was that we better allocate the secret data explicitly (using mmap()) to make sure that no other process data accidentally gets to suffer. Imagine I would allocate a buffer to hold a secret key, signal with mlock
to protect it and suddenly my other high throughput non-secret buffer
(which happened to live on the same page by chance) became very slow
and I don't even have an easy way (apart from mmap()ing it!) to guarantee that it won't be affected.

So, I ended up towards smth like:

secret_buffer = mmap(NULL, PAGE_SIZE, ...)
madvise(secret_buffer, size, MADV_SECRET)

I have work in progress code here:
https://github.com/ereshetova/linux/commits/madvise

I haven't sent it for review, because it is not ready yet and I am now working on trying to add the page wiping functionality. Otherwise it would be useless to protect the page during the time it is used in userspace, but then allow it to get reused by a different process later after it has been released back and userspace was stupid enough not to wipe the contents (or was crashed on purpose before it was able to wipe anything out).

We have also had some discussions with Tycho that XPFO can be also applied selectively for such "SECRET" marked pages and I know that he has also did some initial prototyping on this, so I think it would be great to decide on userspace interface first and then see how we can assemble together all these features.

The *very* far fetching goal for all of this would be something that Alan Cox suggested when I started looking into this - actually have a new libc function to allocate memory in a secure way, which can hide all the dancing with mmap()/madvise() (or/and potentially interaction with a chardev that Andy was suggesting also) and implement an efficient allocator for such secret pages. Openssl has its own version of "secure heap", which is essentially mmap area with additional MLOCK_ONFAULT and MADV_DONTDUMP flags for protection. Some other
apps or libs must use smth similar if they want additional protection, which makes them to reimplement the same concept again and again. Sadly or surprisingly other major libs like boringssl, mbedTLS or client like openssh do not user any mlock()/ madvise() flags for any additional protection of secrets that they hold in memory. Maybe if all of it would be behind a single secure API situation would start to change in userspace towards better.

Best Regards,
Elena.

.

#56 lore raw

Hi Elena, Mike,

Quoted 34 lines · Elena
Quoted 4 lines

The patch below aims to allow applications to create mappins that have pages visible only to the owning process. Such mappings could be used to store secrets so that these secrets are not visible neither to other processes nor to the kernel.

Hi Mike,

I have actually been looking into the closely related problem for the past couple of weeks (on and off). What is common here is the need for userspace to indicate to kernel that some pages contain secrets. And then there are actually a number of things that kernel can do to try to protect these secrets better. Unmap from direct map is one of them. Another thing is to map such pages as non-cached, which can help us to prevent or considerably restrict speculation on such pages. The initial proof of concept for marking pages as "UNCACHED" that I got from Dave Hansen was actually based on mlock2() and a new flag for it for this purpose. Since then I have been thinking on what interface suits the use case better and actually selected going with new madvise() flag instead because of all possible implications for fragmentation and performance. My logic was that we better allocate the secret data explicitly (using mmap()) to make sure that no other process data accidentally gets to suffer. Imagine I would allocate a buffer to hold a secret key, signal with mlock
to protect it and suddenly my other high throughput non-secret buffer
(which happened to live on the same page by chance) became very slow
and I don't even have an easy way (apart from mmap()ing it!) to guarantee that it won't be affected.

So, I ended up towards smth like:

secret_buffer = mmap(NULL, PAGE_SIZE, ...)
madvise(secret_buffer, size, MADV_SECRET)

I have work in progress code here:
https://github.com/ereshetova/linux/commits/madvise

Quoted 6 lines

I haven't sent it for review, because it is not ready yet and I am now working on trying to add the page wiping functionality. Otherwise it would be useless to protect the page during the time it is used in userspace, but then allow it to get reused by a different process later after it has been released back and userspace was stupid enough not to wipe the contents (or was crashed on purpose before it was able to wipe anything out).

I was looking at this and thinking that wiping during do_exit() might be a nice place, but I haven't tried anything yet.

Quoted 5 lines

We have also had some discussions with Tycho that XPFO can be also applied selectively for such "SECRET" marked pages and I know that he has also did some initial prototyping on this, so I think it would be great to decide on userspace interface first and then see how we can assemble together all these features.

Yep! Here's my tree with the direct un-mapping bits ported from XPFO:
https://github.com/tych0/linux/commits/madvise

As noted in one of the commit messages I think the bit math for page prot flags needs a bit of work, but the test passes, so :)

In any case, I'll try to look at Mike's patches later today.

Cheers,

Tycho

maintainer
#57 lore raw
In reply to Reshetova, Elena
Quoted 8 lines · Elena
Quoted 5 lines

The patch below aims to allow applications to create mappins that have pages visible only to the owning process. Such mappings could be used to store secrets so that these secrets are not visible neither to other processes nor to the kernel.

Hi Mike,

Quoted 11 lines

I have actually been looking into the closely related problem for the past couple of weeks (on and off). What is common here is the need for userspace to indicate to kernel that some pages contain secrets. And then there are actually a number of things that kernel can do to try to protect these secrets better. Unmap from direct map is one of them. Another thing is to map such pages as non-cached, which can help us to prevent or considerably restrict speculation on such pages. The initial proof of concept for marking pages as "UNCACHED" that I got from Dave Hansen was actually based on mlock2() and a new flag for it for this purpose. Since then I have been thinking on what interface suits the use case better and actually selected going with new madvise() flag instead because of all possible implications for fragmentation and performance.

Doing all of this with MAP_SECRET seems bad to me. If user code wants UC memory, it should ask for UC memory -- having the kernel involved in the decision to use UC memory is a bad idea, because the performance impact of using UC memory where user code wasn't expecting it wil be so bad that the system might as well not work at all. (For kicks, I once added a sysctl to turn off caching in CR0. I enabled it in gnome-shell. The system slowed down to such an extent that I was unable to enter the three or so keystrokes to turn it back off.)

EXCLUSIVE makes sense. Saying "don't ptrace this" makes sense. UC makes sense. But having one flag to rule them all does not make sense to me.

#58 lore raw
Quoted 6 lines

Doing all of this with MAP_SECRET seems bad to me. If user code wants
UC memory, it should ask for UC memory -- having the kernel involved in the decision to use UC memory is a bad idea, because the performance impact of using UC memory where user code wasn't expecting

The user has no idea that they want UC memory. It varies by platform what this means. There are some systems (eg in order uclinux devices, M68K, old atoms) for which it probably means 'no-op', there are those where UC helps, those it hinders, there are those where WC is probably sufficient. There are platforms where 'secret' memory might best be implemented by using on die memory pools or cache locking. It might even mean 'put me in a non HT cgroup'.

Secret might also mean 'not accessible by thunderbolt', or 'do not swap unless swap is encrypted' and other things.

IMHO the question is what is the actual semantic here. What are you asking for ? Does it mean "at any cost", what does it guarantee (100% or statistically), what level of guarantee is acceptable, what level is -EOPNOTSUPP or similar ?

I'm also wary of the focus always being on keys. If you decrypt a file I'm probably just as interested in the contents so can I mmap a file this way and if so what happens on the unmap. Yes key theft lets me do all sorts of theoretical long term bad stuff, but frequently data theft is sufficient to do lots of practical short term bad stuff. Also as an attacker I'm probably a script, and I don't want to be exposing my master long term because they want the footprints gone.

Quoted 2 lines

in gnome-shell. The system slowed down to such an extent that I was unable to enter the three or so keystrokes to turn it back off.)

Yes - and any uncached pages also need to be kept away from anything that the kernel touches under locks, or use in atomic user operations stuff. Copy on write of an uncached page for example is suddenly really slow and there are so many other cases we'd have to find and deal with.

Quoted 4 lines

EXCLUSIVE makes sense. Saying "don't ptrace this" makes sense. UC makes sense. But having one flag to rule them all does not make sense
to me.

We already support not ptracing, and if I can ptrace any of the code I can access all of its code/data so that one isn't hard and the LSM interfaces can do it. That one is easy - minus the fact that malware writers are big fans of anything that stops tracing...

Alan

#59 lore raw
In reply to Andy Lutomirski
Quoted 46 lines · Andy Lutomirski
Quoted 25 lines · Elena
Quoted 8 lines

The patch below aims to allow applications to create mappins that have
pages visible only to the owning process. Such mappings could be used to
store secrets so that these secrets are not visible neither to other
processes nor to the kernel.

Hi Mike,

I have actually been looking into the closely related problem for the past
couple of weeks (on and off). What is common here is the need for userspace
to indicate to kernel that some pages contain secrets. And then there are
actually a number of things that kernel can do to try to protect these secrets
better. Unmap from direct map is one of them. Another thing is to map such
pages as non-cached, which can help us to prevent or considerably restrict
speculation on such pages. The initial proof of concept for marking pages as

Quoted 8 lines

"UNCACHED" that I got from Dave Hansen was actually based on mlock2()
and a new flag for it for this purpose. Since then I have been thinking on what
interface suits the use case better and actually selected going with new madvise()
flag instead because of all possible implications for fragmentation and performance.

Doing all of this with MAP_SECRET seems bad to me. If user code wants UC memory, it should ask for UC memory -- having the kernel involved in the decision to use UC memory is a bad idea, because the performance impact of using UC memory where user code wasn't expecting it wil be so bad that the system might as well not work at all. (For kicks, I once added a sysctl to turn off caching in CR0. I enabled it in gnome-shell. The system slowed down to such an extent that I was unable to enter the three or so keystrokes to turn it back off.)

Quoted 3 lines

EXCLUSIVE makes sense. Saying "don't ptrace this" makes sense. UC makes sense. But having one flag to rule them all does not make sense to me.

So this is a usability problem. We have a memory flag that can be used for "secrecy" for some userspace value of the word and we have a load of internal properties depending on how the hardware works, including potentially some hardware additions like SEV or TME, that can be used to implement the property. If we expose our hardware vagaries, the user is really not going to know what to do ... and we have a limited number of flags to express this, so it stands to reason that we need to define "secrecy" for the user and then implement it using whatever flags we have. So I think no ptrace and no direct map make sense for pretty much any value of "secrecy". The UC bit seems to be an attempt to prevent exfiltration via L1TF or other cache side channels, so it looks like it should only be applied if the side channel mitigations aren't active ... which would tend to indicate it's a kernel decision as well.

In the use case in my head, I'd like MAP_EXCLUSIVE to mean the data in the user memory is difficult to exfiltrate for another tenant in a virtual system, even if they break containment, so effectively I want it protected against kernel exploitation and root in the host ... and I suppose I need to acknowledge that "protected" means best effort available on the platform, not no attacker can ever extract this.

James

maintainer
#60 lore raw
Quoted 52 lines · James Bottomley
Quoted 46 lines · Andy Lutomirski
Quoted 25 lines · Elena
Quoted 8 lines

The patch below aims to allow applications to create mappins that have
pages visible only to the owning process. Such mappings could be used to
store secrets so that these secrets are not visible neither to other
processes nor to the kernel.

Hi Mike,

I have actually been looking into the closely related problem for the past
couple of weeks (on and off). What is common here is the need for userspace
to indicate to kernel that some pages contain secrets. And then there are
actually a number of things that kernel can do to try to protect these secrets
better. Unmap from direct map is one of them. Another thing is to map such
pages as non-cached, which can help us to prevent or considerably restrict
speculation on such pages. The initial proof of concept for marking pages as

Quoted 8 lines

"UNCACHED" that I got from Dave Hansen was actually based on mlock2()
and a new flag for it for this purpose. Since then I have been thinking on what
interface suits the use case better and actually selected going with new madvise()
flag instead because of all possible implications for fragmentation and performance.

Doing all of this with MAP_SECRET seems bad to me. If user code wants UC memory, it should ask for UC memory -- having the kernel involved in the decision to use UC memory is a bad idea, because the performance impact of using UC memory where user code wasn't expecting it wil be so bad that the system might as well not work at all. (For kicks, I once added a sysctl to turn off caching in CR0. I enabled it in gnome-shell. The system slowed down to such an extent that I was unable to enter the three or so keystrokes to turn it back off.)

Quoted 3 lines

EXCLUSIVE makes sense. Saying "don't ptrace this" makes sense. UC makes sense. But having one flag to rule them all does not make sense to me.

Quoted 14 lines

So this is a usability problem. We have a memory flag that can be used for "secrecy" for some userspace value of the word and we have a load of internal properties depending on how the hardware works, including potentially some hardware additions like SEV or TME, that can be used to implement the property. If we expose our hardware vagaries, the user is really not going to know what to do ... and we have a limited number of flags to express this, so it stands to reason that we need to define "secrecy" for the user and then implement it using whatever flags we have. So I think no ptrace and no direct map make sense for pretty much any value of "secrecy". The UC bit seems to be an attempt to prevent exfiltration via L1TF or other cache side channels, so it looks like it should only be applied if the side channel mitigations aren't active ... which would tend to indicate it's a kernel decision as well.

I just don't think this will work in practice. Someone will say "hey, let's keep this giant buffer we do crypto from, or maybe even the entire data area of some critical service, secret". It will work *fine* at first. But then some kernel config changes and we can't do DMA, and now it breaks on some configs. Someone else will say "hey, I don't have L1TF or whatever mitigation, let's turn on UC", and everything goes to hell.

IMO the kernel should attempt to keep *all memory* secret. Specific applications that want greater levels of secrecy should opt in to more expensive things. Here's what's already on the table:

Exclusive / XPFO / XPO: allocation might be extremely expensive. Overuse might hurt performance due to huge page fragmentation DMA may not work. Otherwise it's peachy.

SEV: Works only in some contexts. The current kernel implementation is, IMO, unacceptable to the extent that I wish I could go back in time and NAK it.

TME: it's on or it's off. There's no room for a MAP_ flag here.

MKTME: of highly dubious value here. The only useful thing here I can thing it would be a MAP_NOTSECRET to opt *out* of encryption for a specific range. Other than that, it has all the same performance implications that EXCLUSIVE has.

UC: Performance hit is extreme. *Also* has the perf implications of exclusive. I can't imagine this making any sense except were the user application is written in the expectation that UC might be used so that the access patterns would be reasonable.

WC: Same issues as UC plus memory ordering issues such that unsuspecting applications will corrupt data.

Trying to bundle these together with kernel- or admin-only config seems like a lost cause.