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, 54 insertions, 18 deletions

diff --git a/pkg/sentry/syscalls/linux/error.go b/pkg/sentry/syscalls/linux/error.go
index 60469549d..64de56ac5 100644
--- a/pkg/sentry/syscalls/linux/error.go
+++ b/pkg/sentry/syscalls/linux/error.go
@@ -22,6 +22,7 @@ import (
 	"gvisor.dev/gvisor/pkg/metric"
 	"gvisor.dev/gvisor/pkg/sentry/fs"
 	"gvisor.dev/gvisor/pkg/sentry/kernel"
+	"gvisor.dev/gvisor/pkg/sentry/vfs"
 	"gvisor.dev/gvisor/pkg/sync"
 	"gvisor.dev/gvisor/pkg/syserror"
 )
@@ -31,20 +32,58 @@ var (
 	partialResultOnce   sync.Once
 )
 
+// HandleIOErrorVFS2 handles special error cases for partial results. For some
+// errors, we may consume the error and return only the partial read/write.
+//
+// op and f are used only for panics.
+func HandleIOErrorVFS2(t *kernel.Task, partialResult bool, err, intr error, op string, f *vfs.FileDescription) error {
+	known, err := handleIOErrorImpl(t, partialResult, err, intr, op)
+	if err != nil {
+		return err
+	}
+	if !known {
+		// An unknown error is encountered with a partial read/write.
+		fs := f.Mount().Filesystem().VirtualFilesystem()
+		root := vfs.RootFromContext(t)
+		name, _ := fs.PathnameWithDeleted(t, root, f.VirtualDentry())
+		log.Traceback("Invalid request partialResult %v and err (type %T) %v for %s operation on %q", partialResult, err, err, op, name)
+		partialResultOnce.Do(partialResultMetric.Increment)
+	}
+	return nil
+}
+
 // handleIOError handles special error cases for partial results. For some
 // errors, we may consume the error and return only the partial read/write.
 //
 // op and f are used only for panics.
 func handleIOError(t *kernel.Task, partialResult bool, err, intr error, op string, f *fs.File) error {
+	known, err := handleIOErrorImpl(t, partialResult, err, intr, op)
+	if err != nil {
+		return err
+	}
+	if !known {
+		// An unknown error is encountered with a partial read/write.
+		name, _ := f.Dirent.FullName(nil /* ignore chroot */)
+		log.Traceback("Invalid request partialResult %v and err (type %T) %v for %s operation on %q, %T", partialResult, err, err, op, name, f.FileOperations)
+		partialResultOnce.Do(partialResultMetric.Increment)
+	}
+	return nil
+}
+
+// handleIOError handles special error cases for partial results. For some
+// errors, we may consume the error and return only the partial read/write.
+//
+// Returns false if error is unknown.
+func handleIOErrorImpl(t *kernel.Task, partialResult bool, err, intr error, op string) (bool, error) {
 	switch err {
 	case nil:
 		// Typical successful syscall.
-		return nil
+		return true, nil
 	case io.EOF:
 		// EOF is always consumed. If this is a partial read/write
 		// (result != 0), the application will see that, otherwise
 		// they will see 0.
-		return nil
+		return true, nil
 	case syserror.ErrExceedsFileSizeLimit:
 		// Ignore partialResult because this error only applies to
 		// normal files, and for those files we cannot accumulate
@@ -53,20 +92,20 @@ func handleIOError(t *kernel.Task, partialResult bool, err, intr error, op strin
 		// Do not consume the error and return it as EFBIG.
 		// Simultaneously send a SIGXFSZ per setrlimit(2).
 		t.SendSignal(kernel.SignalInfoNoInfo(linux.SIGXFSZ, t, t))
-		return syserror.EFBIG
+		return true, syserror.EFBIG
 	case syserror.ErrInterrupted:
 		// The syscall was interrupted. Return nil if it completed
 		// partially, otherwise return the error code that the syscall
 		// needs (to indicate to the kernel what it should do).
 		if partialResult {
-			return nil
+			return true, nil
 		}
-		return intr
+		return true, intr
 	}
 
 	if !partialResult {
 		// Typical syscall error.
-		return err
+		return true, err
 	}
 
 	switch err {
@@ -75,14 +114,14 @@ func handleIOError(t *kernel.Task, partialResult bool, err, intr error, op strin
 		// read/write.  Like ErrWouldBlock, since we have a
 		// partial read/write, we consume the error and return
 		// the partial result.
-		return nil
+		return true, nil
 	case syserror.EFAULT:
 		// EFAULT is only shown the user if nothing was
 		// read/written. If we read something (this case), they see
 		// a partial read/write. They will then presumably try again
 		// with an incremented buffer, which will EFAULT with
 		// result == 0.
-		return nil
+		return true, nil
 	case syserror.EPIPE:
 		// Writes to a pipe or socket will return EPIPE if the other
 		// side is gone. The partial write is returned. EPIPE will be
@@ -90,32 +129,29 @@ func handleIOError(t *kernel.Task, partialResult bool, err, intr error, op strin
 		//
 		// TODO(gvisor.dev/issue/161): In some cases SIGPIPE should
 		// also be sent to the application.
-		return nil
+		return true, nil
 	case syserror.ENOSPC:
 		// Similar to EPIPE. Return what we wrote this time, and let
 		// ENOSPC be returned on the next call.
-		return nil
+		return true, nil
 	case syserror.ECONNRESET:
 		// For TCP sendfile connections, we may have a reset. But we
 		// should just return n as the result.
-		return nil
+		return true, nil
 	case syserror.ErrWouldBlock:
 		// Syscall would block, but completed a partial read/write.
 		// This case should only be returned by IssueIO for nonblocking
 		// files. Since we have a partial read/write, we consume
 		// ErrWouldBlock, returning the partial result.
-		return nil
+		return true, nil
 	}
 
 	switch err.(type) {
 	case kernel.SyscallRestartErrno:
 		// Identical to the EINTR case.
-		return nil
+		return true, nil
 	}
 
-	// An unknown error is encountered with a partial read/write.
-	name, _ := f.Dirent.FullName(nil /* ignore chroot */)
-	log.Traceback("Invalid request partialResult %v and err (type %T) %v for %s operation on %q, %T", partialResult, err, err, op, name, f.FileOperations)
-	partialResultOnce.Do(partialResultMetric.Increment)
-	return nil
+	// Error is unknown and cannot be properly handled.
+	return false, nil
 }