Decouple filemem from platform and move it to pgalloc.MemoryFile.

This is in preparation for improved page cache reclaim, which requires
greater integration between the page cache and page allocator.

PiperOrigin-RevId: 238444706
Change-Id: Id24141b3678d96c7d7dc24baddd9be555bffafe4

12 files changed, 73 insertions, 49 deletions

diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD
index d9bbfb556..4d34bc733 100644
--- a/pkg/sentry/kernel/BUILD
+++ b/pkg/sentry/kernel/BUILD
@@ -173,6 +173,7 @@ go_library(
         "//pkg/sentry/loader",
         "//pkg/sentry/memmap",
         "//pkg/sentry/mm",
+        "//pkg/sentry/pgalloc",
         "//pkg/sentry/platform",
         "//pkg/sentry/safemem",
         "//pkg/sentry/socket/netlink/port",
@@ -212,7 +213,7 @@ go_test(
         "//pkg/sentry/kernel/kdefs",
         "//pkg/sentry/kernel/sched",
         "//pkg/sentry/limits",
-        "//pkg/sentry/platform",
+        "//pkg/sentry/pgalloc",
         "//pkg/sentry/time",
         "//pkg/sentry/usage",
         "//pkg/sentry/usermem",
diff --git a/pkg/sentry/kernel/contexttest/BUILD b/pkg/sentry/kernel/contexttest/BUILD
index 5769a3b28..bfb2a0b73 100644
--- a/pkg/sentry/kernel/contexttest/BUILD
+++ b/pkg/sentry/kernel/contexttest/BUILD
@@ -12,6 +12,7 @@ go_library(
         "//pkg/sentry/context",
         "//pkg/sentry/context/contexttest",
         "//pkg/sentry/kernel",
+        "//pkg/sentry/pgalloc",
         "//pkg/sentry/platform",
     ],
 )
diff --git a/pkg/sentry/kernel/contexttest/contexttest.go b/pkg/sentry/kernel/contexttest/contexttest.go
index 9eb18e7e8..eb56a6a07 100644
--- a/pkg/sentry/kernel/contexttest/contexttest.go
+++ b/pkg/sentry/kernel/contexttest/contexttest.go
@@ -22,6 +22,7 @@ import (
 	"gvisor.googlesource.com/gvisor/pkg/sentry/context"
 	"gvisor.googlesource.com/gvisor/pkg/sentry/context/contexttest"
 	"gvisor.googlesource.com/gvisor/pkg/sentry/kernel"
+	"gvisor.googlesource.com/gvisor/pkg/sentry/pgalloc"
 	"gvisor.googlesource.com/gvisor/pkg/sentry/platform"
 )
 
@@ -33,6 +34,7 @@ func Context(tb testing.TB) context.Context {
 	k := &kernel.Kernel{
 		Platform: platform.FromContext(ctx),
 	}
+	k.SetMemoryFile(pgalloc.MemoryFileFromContext(ctx))
 	ctx.(*contexttest.TestContext).RegisterValue(kernel.CtxKernel, k)
 	return ctx
 }
diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go
index c6afae2e6..3533fd8f7 100644
--- a/pkg/sentry/kernel/kernel.go
+++ b/pkg/sentry/kernel/kernel.go
@@ -58,6 +58,7 @@ import (
 	"gvisor.googlesource.com/gvisor/pkg/sentry/limits"
 	"gvisor.googlesource.com/gvisor/pkg/sentry/loader"
 	"gvisor.googlesource.com/gvisor/pkg/sentry/mm"
+	"gvisor.googlesource.com/gvisor/pkg/sentry/pgalloc"
 	"gvisor.googlesource.com/gvisor/pkg/sentry/platform"
 	"gvisor.googlesource.com/gvisor/pkg/sentry/socket/netlink/port"
 	sentrytime "gvisor.googlesource.com/gvisor/pkg/sentry/time"
@@ -89,12 +90,14 @@ type Kernel struct {
 
 	// All of the following fields are immutable unless otherwise specified.
 
-	// Platform is the platform that is used to execute tasks in the
-	// created Kernel. It is embedded so that Kernel can directly serve as
-	// Platform in mm logic and also serve as platform.MemoryProvider in
-	// filemem S/R logic.
+	// Platform is the platform that is used to execute tasks in the created
+	// Kernel. See comment on pgalloc.MemoryFileProvider for why Platform is
+	// embedded anonymously (the same issue applies).
 	platform.Platform `state:"nosave"`
 
+	// mf provides application memory.
+	mf *pgalloc.MemoryFile `state:"nosave"`
+
 	// See InitKernelArgs for the meaning of these fields.
 	featureSet                  *cpuid.FeatureSet
 	timekeeper                  *Timekeeper
@@ -229,7 +232,8 @@ type InitKernelArgs struct {
 
 // Init initialize the Kernel with no tasks.
 //
-// Callers must manually set Kernel.Platform before caling Init.
+// Callers must manually set Kernel.Platform and call Kernel.SetMemoryFile
+// before calling Init.
 func (k *Kernel) Init(args InitKernelArgs) error {
 	if args.FeatureSet == nil {
 		return fmt.Errorf("FeatureSet is nil")
@@ -332,15 +336,9 @@ func (k *Kernel) SaveTo(w io.Writer) error {
 	log.Infof("Kernel save stats: %s", &stats)
 	log.Infof("Kernel save took [%s].", time.Since(kernelStart))
 
-	// Save the memory state.
-	//
-	// FIXME: In the future, this should not be dispatched via
-	// an abstract memory type. This should be dispatched to a single
-	// memory implementation that belongs to the kernel. (There is
-	// currently a single implementation anyways, it just needs to be
-	// "unabstracted" and reparented appropriately.)
+	// Save the memory file's state.
 	memoryStart := time.Now()
-	if err := k.Platform.Memory().SaveTo(w); err != nil {
+	if err := k.mf.SaveTo(w); err != nil {
 		return err
 	}
 	log.Infof("Memory save took [%s].", time.Since(memoryStart))
@@ -418,13 +416,9 @@ func (ts *TaskSet) unregisterEpollWaiters() {
 }
 
 // LoadFrom returns a new Kernel loaded from args.
-func (k *Kernel) LoadFrom(r io.Reader, p platform.Platform, net inet.Stack) error {
+func (k *Kernel) LoadFrom(r io.Reader, net inet.Stack) error {
 	loadStart := time.Now()
-	if p == nil {
-		return fmt.Errorf("Platform is nil")
-	}
 
-	k.Platform = p
 	k.networkStack = net
 
 	initAppCores := k.applicationCores
@@ -438,11 +432,9 @@ func (k *Kernel) LoadFrom(r io.Reader, p platform.Platform, net inet.Stack) erro
 	log.Infof("Kernel load stats: %s", &stats)
 	log.Infof("Kernel load took [%s].", time.Since(kernelStart))
 
-	// Load the memory state.
-	//
-	// See the note in SaveTo.
+	// Load the memory file's state.
 	memoryStart := time.Now()
-	if err := k.Platform.Memory().LoadFrom(r); err != nil {
+	if err := k.mf.LoadFrom(r); err != nil {
 		return err
 	}
 	log.Infof("Memory load took [%s].", time.Since(memoryStart))
@@ -597,6 +589,10 @@ func (ctx *createProcessContext) Value(key interface{}) interface{} {
 		return ctx.k.RealtimeClock()
 	case limits.CtxLimits:
 		return ctx.args.Limits
+	case pgalloc.CtxMemoryFile:
+		return ctx.k.mf
+	case pgalloc.CtxMemoryFileProvider:
+		return ctx.k
 	case platform.CtxPlatform:
 		return ctx.k
 	case uniqueid.CtxGlobalUniqueID:
@@ -1018,6 +1014,17 @@ func (k *Kernel) NowMonotonic() int64 {
 	return now
 }
 
+// SetMemoryFile sets Kernel.mf. SetMemoryFile must be called before Init or
+// LoadFrom.
+func (k *Kernel) SetMemoryFile(mf *pgalloc.MemoryFile) {
+	k.mf = mf
+}
+
+// MemoryFile implements pgalloc.MemoryFileProvider.MemoryFile.
+func (k *Kernel) MemoryFile() *pgalloc.MemoryFile {
+	return k.mf
+}
+
 // SupervisorContext returns a Context with maximum privileges in k. It should
 // only be used by goroutines outside the control of the emulated kernel
 // defined by e.
@@ -1083,7 +1090,7 @@ func (k *Kernel) ListSockets(family int) []*refs.WeakRef {
 	socks := []*refs.WeakRef{}
 	if table, ok := k.socketTable[family]; ok {
 		socks = make([]*refs.WeakRef, 0, len(table))
-		for s, _ := range table {
+		for s := range table {
 			socks = append(socks, s)
 		}
 	}
@@ -1123,6 +1130,10 @@ func (ctx supervisorContext) Value(key interface{}) interface{} {
 	case limits.CtxLimits:
 		// No limits apply.
 		return limits.NewLimitSet()
+	case pgalloc.CtxMemoryFile:
+		return ctx.k.mf
+	case pgalloc.CtxMemoryFileProvider:
+		return ctx.k
 	case platform.CtxPlatform:
 		return ctx.k
 	case uniqueid.CtxGlobalUniqueID:
diff --git a/pkg/sentry/kernel/memevent/memory_events.go b/pkg/sentry/kernel/memevent/memory_events.go
index b6283c5d1..d09d6debf 100644
--- a/pkg/sentry/kernel/memevent/memory_events.go
+++ b/pkg/sentry/kernel/memevent/memory_events.go
@@ -95,7 +95,7 @@ func (m *MemoryEvents) run() {
 }
 
 func (m *MemoryEvents) emit() {
-	totalPlatform, err := m.k.Platform.Memory().TotalUsage()
+	totalPlatform, err := m.k.MemoryFile().TotalUsage()
 	if err != nil {
 		log.Warningf("Failed to fetch memory usage for memory events: %v", err)
 		return
diff --git a/pkg/sentry/kernel/shm/BUILD b/pkg/sentry/kernel/shm/BUILD
index f45770eef..bc2089872 100644
--- a/pkg/sentry/kernel/shm/BUILD
+++ b/pkg/sentry/kernel/shm/BUILD
@@ -20,6 +20,7 @@ go_library(
         "//pkg/sentry/kernel/auth",
         "//pkg/sentry/kernel/time",
         "//pkg/sentry/memmap",
+        "//pkg/sentry/pgalloc",
         "//pkg/sentry/platform",
         "//pkg/sentry/usage",
         "//pkg/sentry/usermem",
diff --git a/pkg/sentry/kernel/shm/shm.go b/pkg/sentry/kernel/shm/shm.go
index 96414d060..4525aabf4 100644
--- a/pkg/sentry/kernel/shm/shm.go
+++ b/pkg/sentry/kernel/shm/shm.go
@@ -45,6 +45,7 @@ import (
 	"gvisor.googlesource.com/gvisor/pkg/sentry/kernel/auth"
 	ktime "gvisor.googlesource.com/gvisor/pkg/sentry/kernel/time"
 	"gvisor.googlesource.com/gvisor/pkg/sentry/memmap"
+	"gvisor.googlesource.com/gvisor/pkg/sentry/pgalloc"
 	"gvisor.googlesource.com/gvisor/pkg/sentry/platform"
 	"gvisor.googlesource.com/gvisor/pkg/sentry/usage"
 	"gvisor.googlesource.com/gvisor/pkg/sentry/usermem"
@@ -199,19 +200,19 @@ func (r *Registry) FindOrCreate(ctx context.Context, pid int32, key Key, size ui
 //
 // Precondition: Caller must hold r.mu.
 func (r *Registry) newShm(ctx context.Context, pid int32, key Key, creator fs.FileOwner, perms fs.FilePermissions, size uint64) (*Shm, error) {
-	p := platform.FromContext(ctx)
-	if p == nil {
-		panic(fmt.Sprintf("context.Context %T lacks non-nil value for key %T", ctx, platform.CtxPlatform))
+	mfp := pgalloc.MemoryFileProviderFromContext(ctx)
+	if mfp == nil {
+		panic(fmt.Sprintf("context.Context %T lacks non-nil value for key %T", ctx, pgalloc.CtxMemoryFileProvider))
 	}
 
 	effectiveSize := uint64(usermem.Addr(size).MustRoundUp())
-	fr, err := p.Memory().Allocate(effectiveSize, usage.Anonymous)
+	fr, err := mfp.MemoryFile().Allocate(effectiveSize, usage.Anonymous)
 	if err != nil {
 		return nil, err
 	}
 
 	shm := &Shm{
-		p:             p,
+		mfp:           mfp,
 		registry:      r,
 		creator:       creator,
 		size:          size,
@@ -312,7 +313,7 @@ type Shm struct {
 	// destruction.
 	refs.AtomicRefCount
 
-	p platform.Platform
+	mfp pgalloc.MemoryFileProvider
 
 	// registry points to the shm registry containing this segment. Immutable.
 	registry *Registry
@@ -333,7 +334,7 @@ type Shm struct {
 	// Invariant: effectiveSize must be a multiple of usermem.PageSize.
 	effectiveSize uint64
 
-	// fr is the offset into platform.Memory() that backs this contents of this
+	// fr is the offset into mfp.MemoryFile() that backs this contents of this
 	// segment. Immutable.
 	fr platform.FileRange
 
@@ -452,7 +453,7 @@ func (s *Shm) Translate(ctx context.Context, required, optional memmap.MappableR
 		return []memmap.Translation{
 			{
 				Source: source,
-				File:   s.p.Memory(),
+				File:   s.mfp.MemoryFile(),
 				Offset: s.fr.Start + source.Start,
 			},
 		}, err
@@ -599,7 +600,7 @@ func (s *Shm) Set(ctx context.Context, ds *linux.ShmidDS) error {
 }
 
 func (s *Shm) destroy() {
-	s.p.Memory().DecRef(s.fr)
+	s.mfp.MemoryFile().DecRef(s.fr)
 	s.registry.remove(s)
 }
 
diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go
index 702e40cce..e9f133c0b 100644
--- a/pkg/sentry/kernel/task.go
+++ b/pkg/sentry/kernel/task.go
@@ -29,6 +29,7 @@ import (
 	"gvisor.googlesource.com/gvisor/pkg/sentry/kernel/sched"
 	ktime "gvisor.googlesource.com/gvisor/pkg/sentry/kernel/time"
 	"gvisor.googlesource.com/gvisor/pkg/sentry/limits"
+	"gvisor.googlesource.com/gvisor/pkg/sentry/pgalloc"
 	"gvisor.googlesource.com/gvisor/pkg/sentry/platform"
 	"gvisor.googlesource.com/gvisor/pkg/sentry/unimpl"
 	"gvisor.googlesource.com/gvisor/pkg/sentry/uniqueid"
@@ -587,6 +588,10 @@ func (t *Task) Value(key interface{}) interface{} {
 		return t.k.RealtimeClock()
 	case limits.CtxLimits:
 		return t.tg.limits
+	case pgalloc.CtxMemoryFile:
+		return t.k.mf
+	case pgalloc.CtxMemoryFileProvider:
+		return t.k
 	case platform.CtxPlatform:
 		return t.k
 	case uniqueid.CtxGlobalUniqueID:
diff --git a/pkg/sentry/kernel/task_context.go b/pkg/sentry/kernel/task_context.go
index ee3e49d17..d1c82f2aa 100644
--- a/pkg/sentry/kernel/task_context.go
+++ b/pkg/sentry/kernel/task_context.go
@@ -144,7 +144,7 @@ func (t *Task) Stack() *arch.Stack {
 //  * fs: Binary FeatureSet
 func (k *Kernel) LoadTaskImage(ctx context.Context, mounts *fs.MountNamespace, root, wd *fs.Dirent, maxTraversals *uint, filename string, argv, envv []string, fs *cpuid.FeatureSet) (*TaskContext, *syserr.Error) {
 	// Prepare a new user address space to load into.
-	m := mm.NewMemoryManager(k)
+	m := mm.NewMemoryManager(k, k)
 	defer m.DecUsers(ctx)
 
 	os, ac, name, err := loader.Load(ctx, m, mounts, root, wd, maxTraversals, fs, filename, argv, envv, k.extraAuxv, k.vdso)
diff --git a/pkg/sentry/kernel/timekeeper.go b/pkg/sentry/kernel/timekeeper.go
index 6bff80f13..d7bd85e78 100644
--- a/pkg/sentry/kernel/timekeeper.go
+++ b/pkg/sentry/kernel/timekeeper.go
@@ -21,6 +21,7 @@ import (
 
 	"gvisor.googlesource.com/gvisor/pkg/log"
 	ktime "gvisor.googlesource.com/gvisor/pkg/sentry/kernel/time"
+	"gvisor.googlesource.com/gvisor/pkg/sentry/pgalloc"
 	"gvisor.googlesource.com/gvisor/pkg/sentry/platform"
 	sentrytime "gvisor.googlesource.com/gvisor/pkg/sentry/time"
 )
@@ -85,9 +86,9 @@ type Timekeeper struct {
 // NewTimekeeper does not take ownership of paramPage.
 //
 // SetClocks must be called on the returned Timekeeper before it is usable.
-func NewTimekeeper(platform platform.Platform, paramPage platform.FileRange) (*Timekeeper, error) {
+func NewTimekeeper(mfp pgalloc.MemoryFileProvider, paramPage platform.FileRange) (*Timekeeper, error) {
 	return &Timekeeper{
-		params: NewVDSOParamPage(platform, paramPage),
+		params: NewVDSOParamPage(mfp, paramPage),
 	}, nil
 }
 
diff --git a/pkg/sentry/kernel/timekeeper_test.go b/pkg/sentry/kernel/timekeeper_test.go
index 71674c21c..6084bcb18 100644
--- a/pkg/sentry/kernel/timekeeper_test.go
+++ b/pkg/sentry/kernel/timekeeper_test.go
@@ -18,7 +18,7 @@ import (
 	"testing"
 
 	"gvisor.googlesource.com/gvisor/pkg/sentry/context/contexttest"
-	"gvisor.googlesource.com/gvisor/pkg/sentry/platform"
+	"gvisor.googlesource.com/gvisor/pkg/sentry/pgalloc"
 	sentrytime "gvisor.googlesource.com/gvisor/pkg/sentry/time"
 	"gvisor.googlesource.com/gvisor/pkg/sentry/usage"
 	"gvisor.googlesource.com/gvisor/pkg/sentry/usermem"
@@ -53,13 +53,13 @@ func (c *mockClocks) GetTime(id sentrytime.ClockID) (int64, error) {
 // SetClocks called.
 func stateTestClocklessTimekeeper(tb testing.TB) *Timekeeper {
 	ctx := contexttest.Context(tb)
-	p := platform.FromContext(ctx)
-	fr, err := p.Memory().Allocate(usermem.PageSize, usage.Anonymous)
+	mfp := pgalloc.MemoryFileProviderFromContext(ctx)
+	fr, err := mfp.MemoryFile().Allocate(usermem.PageSize, usage.Anonymous)
 	if err != nil {
 		tb.Fatalf("failed to allocate memory: %v", err)
 	}
 	return &Timekeeper{
-		params: NewVDSOParamPage(p, fr),
+		params: NewVDSOParamPage(mfp, fr),
 	}
 }
 
diff --git a/pkg/sentry/kernel/vdso.go b/pkg/sentry/kernel/vdso.go
index 0ec858a4a..3a35f1d00 100644
--- a/pkg/sentry/kernel/vdso.go
+++ b/pkg/sentry/kernel/vdso.go
@@ -18,6 +18,7 @@ import (
 	"fmt"
 
 	"gvisor.googlesource.com/gvisor/pkg/binary"
+	"gvisor.googlesource.com/gvisor/pkg/sentry/pgalloc"
 	"gvisor.googlesource.com/gvisor/pkg/sentry/platform"
 	"gvisor.googlesource.com/gvisor/pkg/sentry/safemem"
 	"gvisor.googlesource.com/gvisor/pkg/sentry/usermem"
@@ -55,9 +56,9 @@ type vdsoParams struct {
 //
 // +stateify savable
 type VDSOParamPage struct {
-	// The parameter page is fr, allocated from platform.Memory().
-	platform platform.Platform
-	fr       platform.FileRange
+	// The parameter page is fr, allocated from mfp.MemoryFile().
+	mfp pgalloc.MemoryFileProvider
+	fr  platform.FileRange
 
 	// seq is the current sequence count written to the page.
 	//
@@ -73,20 +74,20 @@ type VDSOParamPage struct {
 //
 // Preconditions:
 //
-// * fr is a single page allocated from platform.Memory(). VDSOParamPage does
+// * fr is a single page allocated from mfp.MemoryFile(). VDSOParamPage does
 //   not take ownership of fr; it must remain allocated for the lifetime of the
 //   VDSOParamPage.
 //
 // * VDSOParamPage must be the only writer to fr.
 //
-// * platform.Memory().MapInternal(fr) must return a single safemem.Block.
-func NewVDSOParamPage(platform platform.Platform, fr platform.FileRange) *VDSOParamPage {
-	return &VDSOParamPage{platform: platform, fr: fr}
+// * mfp.MemoryFile().MapInternal(fr) must return a single safemem.Block.
+func NewVDSOParamPage(mfp pgalloc.MemoryFileProvider, fr platform.FileRange) *VDSOParamPage {
+	return &VDSOParamPage{mfp: mfp, fr: fr}
 }
 
 // access returns a mapping of the param page.
 func (v *VDSOParamPage) access() (safemem.Block, error) {
-	bs, err := v.platform.Memory().MapInternal(v.fr, usermem.ReadWrite)
+	bs, err := v.mfp.MemoryFile().MapInternal(v.fr, usermem.ReadWrite)
 	if err != nil {
 		return safemem.Block{}, err
 	}