Implement MAP_32BIT.

PiperOrigin-RevId: 196281052
Change-Id: Ie620a0f983a1bf2570d0003d4754611879335c1c

pkg/abi/linux/mm.go

@@ -31,6 +31,7 @@ const (
 	MAP_PRIVATE    = 1 << 1
 	MAP_FIXED      = 1 << 4
 	MAP_ANONYMOUS  = 1 << 5
+	MAP_32BIT      = 1 << 6 // arch/x86/include/uapi/asm/mman.h
 	MAP_GROWSDOWN  = 1 << 8
 	MAP_DENYWRITE  = 1 << 11
 	MAP_EXECUTABLE = 1 << 12

pkg/sentry/memmap/memmap.go

@@ -266,6 +266,12 @@ type MMapOpts struct {
 	// be replaced. If Unmap is true, Fixed must be true.
 	Unmap bool
 
+	// If Map32Bit is true, all addresses in the created mapping must fit in a
+	// 32-bit integer. (Note that the "end address" of the mapping, i.e. the
+	// address of the first byte *after* the mapping, need not fit in a 32-bit
+	// integer.) Map32Bit is ignored if Fixed is true.
+	Map32Bit bool
+
 	// Perms is the set of permissions to the applied to this mapping.
 	Perms usermem.AccessType
 

pkg/sentry/mm/vma.go

@@ -34,9 +34,10 @@ func (mm *MemoryManager) createVMALocked(ctx context.Context, opts memmap.MMapOp
 
 	// Find a useable range.
 	addr, err := mm.findAvailableLocked(opts.Length, findAvailableOpts{
-		Addr:  opts.Addr,
-		Fixed: opts.Fixed,
-		Unmap: opts.Unmap,
+		Addr:     opts.Addr,
+		Fixed:    opts.Fixed,
+		Unmap:    opts.Unmap,
+		Map32Bit: opts.Map32Bit,
 	})
 	if err != nil {
 		return vmaIterator{}, usermem.AddrRange{}, err
@@ -93,24 +94,40 @@ func (mm *MemoryManager) createVMALocked(ctx context.Context, opts memmap.MMapOp
 }
 
 type findAvailableOpts struct {
-	// Addr is a suggested address. Addr must be page-aligned.
-	Addr usermem.Addr
+	// These fields are equivalent to those in memmap.MMapOpts, except that:
+	//
+	// - Addr must be page-aligned.
+	//
+	// - Unmap allows existing guard pages in the returned range.
 
-	// Fixed is true if only the suggested address is acceptable.
-	Fixed bool
-
-	// Unmap is true if existing vmas and guard pages may exist in the returned
-	// range.
-	Unmap bool
+	Addr     usermem.Addr
+	Fixed    bool
+	Unmap    bool
+	Map32Bit bool
 }
 
+// map32Start/End are the bounds to which MAP_32BIT mappings are constrained,
+// and are equivalent to Linux's MAP32_BASE and MAP32_MAX respectively.
+const (
+	map32Start = 0x40000000
+	map32End   = 0x80000000
+)
+
 // findAvailableLocked finds an allocatable range.
 //
 // Preconditions: mm.mappingMu must be locked.
 func (mm *MemoryManager) findAvailableLocked(length uint64, opts findAvailableOpts) (usermem.Addr, error) {
+	if opts.Fixed {
+		opts.Map32Bit = false
+	}
+	allowedAR := mm.applicationAddrRange()
+	if opts.Map32Bit {
+		allowedAR = allowedAR.Intersect(usermem.AddrRange{map32Start, map32End})
+	}
+
 	// Does the provided suggestion work?
 	if ar, ok := opts.Addr.ToRange(length); ok {
-		if mm.applicationAddrRange().IsSupersetOf(ar) {
+		if allowedAR.IsSupersetOf(ar) {
 			if opts.Unmap {
 				return ar.Start, nil
 			}
@@ -132,6 +149,9 @@ func (mm *MemoryManager) findAvailableLocked(length uint64, opts findAvailableOp
 		alignment = usermem.HugePageSize
 	}
 
+	if opts.Map32Bit {
+		return mm.findLowestAvailableLocked(length, alignment, allowedAR)
+	}
 	if mm.layout.DefaultDirection == arch.MmapBottomUp {
 		return mm.findLowestAvailableLocked(length, alignment, usermem.AddrRange{mm.layout.BottomUpBase, mm.layout.MaxAddr})
 	}

pkg/sentry/syscalls/linux/sys_mmap.go

@@ -45,6 +45,7 @@ func Mmap(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC
 	private := flags&linux.MAP_PRIVATE != 0
 	shared := flags&linux.MAP_SHARED != 0
 	anon := flags&linux.MAP_ANONYMOUS != 0
+	map32bit := flags&linux.MAP_32BIT != 0
 
 	// Require exactly one of MAP_PRIVATE and MAP_SHARED.
 	if private == shared {
@@ -52,12 +53,13 @@ func Mmap(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC
 	}
 
 	opts := memmap.MMapOpts{
-		Length:  args[1].Uint64(),
-		Offset:  args[5].Uint64(),
-		Addr:    args[0].Pointer(),
-		Fixed:   fixed,
-		Unmap:   fixed,
-		Private: private,
+		Length:   args[1].Uint64(),
+		Offset:   args[5].Uint64(),
+		Addr:     args[0].Pointer(),
+		Fixed:    fixed,
+		Unmap:    fixed,
+		Map32Bit: map32bit,
+		Private:  private,
 		Perms: usermem.AccessType{
 			Read:    linux.PROT_READ&prot != 0,
 			Write:   linux.PROT_WRITE&prot != 0,