Implement mlock(), kind of.

Currently mlock() and friends do nothing whatsoever. However, mlocking
is directly application-visible in a number of ways; for example,
madvise(MADV_DONTNEED) and msync(MS_INVALIDATE) both fail on mlocked
regions. We handle this inconsistently: MADV_DONTNEED is too important
to not work, but MS_INVALIDATE is rejected.

Change MM to track mlocked regions in a manner consistent with Linux.
It still will not actually pin pages into host physical memory, but:

- mlock() will now cause sentry memory management to precommit mlocked
pages.

- MADV_DONTNEED and MS_INVALIDATE will interact with mlocked pages as
described above.

PiperOrigin-RevId: 225861605
Change-Id: Iee187204979ac9a4d15d0e037c152c0902c8d0ee

1 files changed, 37 insertions, 0 deletions

diff --git a/pkg/sentry/memmap/memmap.go b/pkg/sentry/memmap/memmap.go
index 28e2bed9b..cf20b11e3 100644
--- a/pkg/sentry/memmap/memmap.go
+++ b/pkg/sentry/memmap/memmap.go
@@ -243,6 +243,40 @@ type MappingIdentity interface {
 	Msync(ctx context.Context, mr MappableRange) error
 }
 
+// MLockMode specifies the memory locking behavior of a memory mapping.
+type MLockMode int
+
+// Note that the ordering of MLockModes is significant; see
+// mm.MemoryManager.defMLockMode.
+const (
+	// MLockNone specifies that a mapping has no memory locking behavior.
+	//
+	// This must be the zero value for MLockMode.
+	MLockNone MLockMode = iota
+
+	// MLockEager specifies that a mapping is memory-locked, as by mlock() or
+	// similar. Pages in the mapping should be made, and kept, resident in
+	// physical memory as soon as possible.
+	//
+	// As of this writing, MLockEager does not cause memory-locking to be
+	// requested from the host; it only affects the sentry's memory management
+	// behavior.
+	//
+	// MLockEager is analogous to Linux's VM_LOCKED.
+	MLockEager
+
+	// MLockLazy specifies that a mapping is memory-locked, as by mlock() or
+	// similar. Pages in the mapping should be kept resident in physical memory
+	// once they have been made resident due to e.g. a page fault.
+	//
+	// As of this writing, MLockLazy does not cause memory-locking to be
+	// requested from the host; in fact, it has virtually no effect, except for
+	// interactions between mlocked pages and other syscalls.
+	//
+	// MLockLazy is analogous to Linux's VM_LOCKED | VM_LOCKONFAULT.
+	MLockLazy
+)
+
 // MMapOpts specifies a request to create a memory mapping.
 type MMapOpts struct {
 	// Length is the length of the mapping.
@@ -303,6 +337,9 @@ type MMapOpts struct {
 	// mapping (see platform.AddressSpace.MapFile).
 	Precommit bool
 
+	// MLockMode specifies the memory locking behavior of the mapping.
+	MLockMode MLockMode
+
 	// Hint is the name used for the mapping in /proc/[pid]/maps. If Hint is
 	// empty, MappingIdentity.MappedName() will be used instead.
 	//