Inline vfs.VirtualFilesystem in Kernel struct

This saves one pointer dereference per VFS access.

Updates #1623

PiperOrigin-RevId: 295216176

8 files changed, 43 insertions, 24 deletions

diff --git a/pkg/sentry/vfs/dentry.go b/pkg/sentry/vfs/dentry.go
index 486a76475..35b208721 100644
--- a/pkg/sentry/vfs/dentry.go
+++ b/pkg/sentry/vfs/dentry.go
@@ -71,6 +71,8 @@ import (
 // lifetime. Dentry reference counts only indicate the extent to which VFS
 // requires Dentries to exist; Filesystems may elect to cache or discard
 // Dentries with zero references.
+//
+// +stateify savable
 type Dentry struct {
 	// parent is this Dentry's parent in this Filesystem. If this Dentry is
 	// independent, parent is nil.
@@ -89,7 +91,7 @@ type Dentry struct {
 	children map[string]*Dentry
 
 	// mu synchronizes disowning and mounting over this Dentry.
-	mu sync.Mutex
+	mu sync.Mutex `state:"nosave"`
 
 	// impl is the DentryImpl associated with this Dentry. impl is immutable.
 	// This should be the last field in Dentry.
diff --git a/pkg/sentry/vfs/device.go b/pkg/sentry/vfs/device.go
index 3af2aa58d..bda5576fa 100644
--- a/pkg/sentry/vfs/device.go
+++ b/pkg/sentry/vfs/device.go
@@ -56,6 +56,7 @@ type Device interface {
 	Open(ctx context.Context, mnt *Mount, d *Dentry, opts OpenOptions) (*FileDescription, error)
 }
 
+// +stateify savable
 type registeredDevice struct {
 	dev  Device
 	opts RegisterDeviceOptions
@@ -63,6 +64,8 @@ type registeredDevice struct {
 
 // RegisterDeviceOptions contains options to
 // VirtualFilesystem.RegisterDevice().
+//
+// +stateify savable
 type RegisterDeviceOptions struct {
 	// GroupName is the name shown for this device registration in
 	// /proc/devices. If GroupName is empty, this registration will not be
diff --git a/pkg/sentry/vfs/file_description_impl_util_test.go b/pkg/sentry/vfs/file_description_impl_util_test.go
index 8fa26418e..3a75d4d62 100644
--- a/pkg/sentry/vfs/file_description_impl_util_test.go
+++ b/pkg/sentry/vfs/file_description_impl_util_test.go
@@ -107,7 +107,10 @@ func (fd *testFD) SetStat(ctx context.Context, opts SetStatOptions) error {
 func TestGenCountFD(t *testing.T) {
 	ctx := contexttest.Context(t)
 
-	vfsObj := New() // vfs.New()
+	vfsObj := &VirtualFilesystem{}
+	if err := vfsObj.Init(); err != nil {
+		t.Fatalf("VFS init: %v", err)
+	}
 	fd := newTestFD(vfsObj, linux.O_RDWR, &genCount{})
 	defer fd.DecRef()
 
@@ -162,7 +165,10 @@ func TestGenCountFD(t *testing.T) {
 func TestWritable(t *testing.T) {
 	ctx := contexttest.Context(t)
 
-	vfsObj := New() // vfs.New()
+	vfsObj := &VirtualFilesystem{}
+	if err := vfsObj.Init(); err != nil {
+		t.Fatalf("VFS init: %v", err)
+	}
 	fd := newTestFD(vfsObj, linux.O_RDWR, &storeData{data: "init"})
 	defer fd.DecRef()
 
diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go
index a06a6caf3..556976d0b 100644
--- a/pkg/sentry/vfs/filesystem.go
+++ b/pkg/sentry/vfs/filesystem.go
@@ -29,6 +29,8 @@ import (
 // Filesystem methods require that a reference is held.
 //
 // Filesystem is analogous to Linux's struct super_block.
+//
+// +stateify savable
 type Filesystem struct {
 	// refs is the reference count. refs is accessed using atomic memory
 	// operations.
diff --git a/pkg/sentry/vfs/filesystem_type.go b/pkg/sentry/vfs/filesystem_type.go
index c58b70728..bb9cada81 100644
--- a/pkg/sentry/vfs/filesystem_type.go
+++ b/pkg/sentry/vfs/filesystem_type.go
@@ -44,6 +44,7 @@ type GetFilesystemOptions struct {
 	InternalData interface{}
 }
 
+// +stateify savable
 type registeredFilesystemType struct {
 	fsType FilesystemType
 	opts   RegisterFilesystemTypeOptions
diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go
index ad2c9fcf4..9912df799 100644
--- a/pkg/sentry/vfs/mount.go
+++ b/pkg/sentry/vfs/mount.go
@@ -38,6 +38,8 @@ import (
 //
 // Mount is analogous to Linux's struct mount. (gVisor does not distinguish
 // between struct mount and struct vfsmount.)
+//
+// +stateify savable
 type Mount struct {
 	// vfs, fs, and root are immutable. References are held on fs and root.
 	//
@@ -85,6 +87,8 @@ type Mount struct {
 // MountNamespace methods require that a reference is held.
 //
 // MountNamespace is analogous to Linux's struct mnt_namespace.
+//
+// +stateify savable
 type MountNamespace struct {
 	// root is the MountNamespace's root mount. root is immutable.
 	root *Mount
diff --git a/pkg/sentry/vfs/mount_unsafe.go b/pkg/sentry/vfs/mount_unsafe.go
index bd90d36c4..1fe766a44 100644
--- a/pkg/sentry/vfs/mount_unsafe.go
+++ b/pkg/sentry/vfs/mount_unsafe.go
@@ -64,6 +64,8 @@ func (mnt *Mount) storeKey(vd VirtualDentry) {
 // (provided mutation is sufficiently uncommon).
 //
 // mountTable.Init() must be called on new mountTables before use.
+//
+// +stateify savable
 type mountTable struct {
 	// mountTable is implemented as a seqcount-protected hash table that
 	// resolves collisions with linear probing, featuring Robin Hood insertion
@@ -75,8 +77,8 @@ type mountTable struct {
 	// intrinsics and inline assembly, limiting the performance of this
 	// approach.)
 
-	seq  sync.SeqCount
-	seed uint32 // for hashing keys
+	seq  sync.SeqCount `state:"nosave"`
+	seed uint32        // for hashing keys
 
 	// size holds both length (number of elements) and capacity (number of
 	// slots): capacity is stored as its base-2 log (referred to as order) in
@@ -89,7 +91,7 @@ type mountTable struct {
 	// length and cap in separate uint32s) for ~free.
 	size uint64
 
-	slots unsafe.Pointer // []mountSlot; never nil after Init
+	slots unsafe.Pointer `state:"nosave"` // []mountSlot; never nil after Init
 }
 
 type mountSlot struct {
diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go
index 51deae313..8f29031b2 100644
--- a/pkg/sentry/vfs/vfs.go
+++ b/pkg/sentry/vfs/vfs.go
@@ -46,11 +46,13 @@ import (
 //
 // There is no analogue to the VirtualFilesystem type in Linux, as the
 // equivalent state in Linux is global.
+//
+// +stateify savable
 type VirtualFilesystem struct {
 	// mountMu serializes mount mutations.
 	//
 	// mountMu is analogous to Linux's namespace_sem.
-	mountMu sync.Mutex
+	mountMu sync.Mutex `state:"nosave"`
 
 	// mounts maps (mount parent, mount point) pairs to mounts. (Since mounts
 	// are uniquely namespaced, including mount parent in the key correctly
@@ -89,44 +91,42 @@ type VirtualFilesystem struct {
 
 	// devices contains all registered Devices. devices is protected by
 	// devicesMu.
-	devicesMu sync.RWMutex
+	devicesMu sync.RWMutex `state:"nosave"`
 	devices   map[devTuple]*registeredDevice
 
 	// anonBlockDevMinor contains all allocated anonymous block device minor
 	// numbers. anonBlockDevMinorNext is a lower bound for the smallest
 	// unallocated anonymous block device number. anonBlockDevMinorNext and
 	// anonBlockDevMinor are protected by anonBlockDevMinorMu.
-	anonBlockDevMinorMu   sync.Mutex
+	anonBlockDevMinorMu   sync.Mutex `state:"nosave"`
 	anonBlockDevMinorNext uint32
 	anonBlockDevMinor     map[uint32]struct{}
 
 	// fsTypes contains all registered FilesystemTypes. fsTypes is protected by
 	// fsTypesMu.
-	fsTypesMu sync.RWMutex
+	fsTypesMu sync.RWMutex `state:"nosave"`
 	fsTypes   map[string]*registeredFilesystemType
 
 	// filesystems contains all Filesystems. filesystems is protected by
 	// filesystemsMu.
-	filesystemsMu sync.Mutex
+	filesystemsMu sync.Mutex `state:"nosave"`
 	filesystems   map[*Filesystem]struct{}
 }
 
-// New returns a new VirtualFilesystem with no mounts or FilesystemTypes.
-func New() *VirtualFilesystem {
-	vfs := &VirtualFilesystem{
-		mountpoints:           make(map[*Dentry]map[*Mount]struct{}),
-		devices:               make(map[devTuple]*registeredDevice),
-		anonBlockDevMinorNext: 1,
-		anonBlockDevMinor:     make(map[uint32]struct{}),
-		fsTypes:               make(map[string]*registeredFilesystemType),
-		filesystems:           make(map[*Filesystem]struct{}),
-	}
+// Init initializes a new VirtualFilesystem with no mounts or FilesystemTypes.
+func (vfs *VirtualFilesystem) Init() error {
+	vfs.mountpoints = make(map[*Dentry]map[*Mount]struct{})
+	vfs.devices = make(map[devTuple]*registeredDevice)
+	vfs.anonBlockDevMinorNext = 1
+	vfs.anonBlockDevMinor = make(map[uint32]struct{})
+	vfs.fsTypes = make(map[string]*registeredFilesystemType)
+	vfs.filesystems = make(map[*Filesystem]struct{})
 	vfs.mounts.Init()
 
 	// Construct vfs.anonMount.
 	anonfsDevMinor, err := vfs.GetAnonBlockDevMinor()
 	if err != nil {
-		panic(fmt.Sprintf("VirtualFilesystem.GetAnonBlockDevMinor() failed during VirtualFilesystem construction: %v", err))
+		return err
 	}
 	anonfs := anonFilesystem{
 		devMinor: anonfsDevMinor,
@@ -137,8 +137,7 @@ func New() *VirtualFilesystem {
 		fs:   &anonfs.vfsfs,
 		refs: 1,
 	}
-
-	return vfs
+	return nil
 }
 
 // PathOperation specifies the path operated on by a VFS method.