Add vfs.FileDescription to FD table

FD table now holds both VFS1 and VFS2 types and uses the correct
one based on what's set.

Parts of this CL are just initial changes (e.g. sys_read.go,
runsc/main.go) to serve as a template for the remaining changes.

Updates #1487
Updates #1623

PiperOrigin-RevId: 292023223

1 files changed, 82 insertions, 16 deletions

diff --git a/pkg/sentry/kernel/fd_table_unsafe.go b/pkg/sentry/kernel/fd_table_unsafe.go
index e9fdb0917..7fd97dc53 100644
--- a/pkg/sentry/kernel/fd_table_unsafe.go
+++ b/pkg/sentry/kernel/fd_table_unsafe.go
@@ -19,6 +19,7 @@ import (
 	"unsafe"
 
 	"gvisor.dev/gvisor/pkg/sentry/fs"
+	"gvisor.dev/gvisor/pkg/sentry/vfs"
 )
 
 type descriptorTable struct {
@@ -41,15 +42,38 @@ func (f *FDTable) init() {
 //
 //go:nosplit
 func (f *FDTable) get(fd int32) (*fs.File, FDFlags, bool) {
+	file, _, flags, ok := f.getAll(fd)
+	return file, flags, ok
+}
+
+// getVFS2 gets a file entry.
+//
+// The boolean indicates whether this was in range.
+//
+//go:nosplit
+func (f *FDTable) getVFS2(fd int32) (*vfs.FileDescription, FDFlags, bool) {
+	_, file, flags, ok := f.getAll(fd)
+	return file, flags, ok
+}
+
+// getAll gets a file entry.
+//
+// The boolean indicates whether this was in range.
+//
+//go:nosplit
+func (f *FDTable) getAll(fd int32) (*fs.File, *vfs.FileDescription, FDFlags, bool) {
 	slice := *(*[]unsafe.Pointer)(atomic.LoadPointer(&f.slice))
 	if fd >= int32(len(slice)) {
-		return nil, FDFlags{}, false
+		return nil, nil, FDFlags{}, false
 	}
 	d := (*descriptor)(atomic.LoadPointer(&slice[fd]))
 	if d == nil {
-		return nil, FDFlags{}, true
+		return nil, nil, FDFlags{}, true
 	}
-	return d.file, d.flags, true
+	if d.file != nil && d.fileVFS2 != nil {
+		panic("VFS1 and VFS2 files set")
+	}
+	return d.file, d.fileVFS2, d.flags, true
 }
 
 // set sets an entry.
@@ -59,6 +83,30 @@ func (f *FDTable) get(fd int32) (*fs.File, FDFlags, bool) {
 //
 // Precondition: mu must be held.
 func (f *FDTable) set(fd int32, file *fs.File, flags FDFlags) {
+	f.setAll(fd, file, nil, flags)
+}
+
+// setVFS2 sets an entry.
+//
+// This handles accounting changes, as well as acquiring and releasing the
+// reference needed by the table iff the file is different.
+//
+// Precondition: mu must be held.
+func (f *FDTable) setVFS2(fd int32, file *vfs.FileDescription, flags FDFlags) {
+	f.setAll(fd, nil, file, flags)
+}
+
+// setAll sets an entry.
+//
+// This handles accounting changes, as well as acquiring and releasing the
+// reference needed by the table iff the file is different.
+//
+// Precondition: mu must be held.
+func (f *FDTable) setAll(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) {
+	if file != nil && fileVFS2 != nil {
+		panic("VFS1 and VFS2 files set")
+	}
+
 	slice := *(*[]unsafe.Pointer)(atomic.LoadPointer(&f.slice))
 
 	// Grow the table as required.
@@ -71,33 +119,51 @@ func (f *FDTable) set(fd int32, file *fs.File, flags FDFlags) {
 		atomic.StorePointer(&f.slice, unsafe.Pointer(&slice))
 	}
 
-	// Create the new element.
-	var d *descriptor
-	if file != nil {
-		d = &descriptor{
-			file:  file,
-			flags: flags,
+	var desc *descriptor
+	if file != nil || fileVFS2 != nil {
+		desc = &descriptor{
+			file:     file,
+			fileVFS2: fileVFS2,
+			flags:    flags,
 		}
 	}
 
 	// Update the single element.
-	orig := (*descriptor)(atomic.SwapPointer(&slice[fd], unsafe.Pointer(d)))
+	orig := (*descriptor)(atomic.SwapPointer(&slice[fd], unsafe.Pointer(desc)))
 
 	// Acquire a table reference.
-	if file != nil && (orig == nil || file != orig.file) {
-		file.IncRef()
+	if desc != nil {
+		switch {
+		case desc.file != nil:
+			if orig == nil || desc.file != orig.file {
+				desc.file.IncRef()
+			}
+		case desc.fileVFS2 != nil:
+			if orig == nil || desc.fileVFS2 != orig.fileVFS2 {
+				desc.fileVFS2.IncRef()
+			}
+		}
 	}
 
 	// Drop the table reference.
-	if orig != nil && file != orig.file {
-		f.drop(orig.file)
+	if orig != nil {
+		switch {
+		case orig.file != nil:
+			if desc == nil || desc.file != orig.file {
+				f.drop(orig.file)
+			}
+		case orig.fileVFS2 != nil:
+			if desc == nil || desc.fileVFS2 != orig.fileVFS2 {
+				f.dropVFS2(orig.fileVFS2)
+			}
+		}
 	}
 
 	// Adjust used.
 	switch {
-	case orig == nil && file != nil:
+	case orig == nil && desc != nil:
 		atomic.AddInt32(&f.used, 1)
-	case orig != nil && file == nil:
+	case orig != nil && desc == nil:
 		atomic.AddInt32(&f.used, -1)
 	}
 }