Implement pwritev2.

Implement pwritev2 and associated unit tests.
Clean up preadv2 unit tests.
Tag RWF_ flags in both preadv2 and pwritev2 with associated bug tickets.

PiperOrigin-RevId: 226222119
Change-Id: Ieb22672418812894ba114bbc88e67f1dd50de620

4 files changed, 82 insertions, 15 deletions

diff --git a/pkg/abi/linux/file.go b/pkg/abi/linux/file.go
index ac49ae9a6..ae33f4a4d 100644
--- a/pkg/abi/linux/file.go
+++ b/pkg/abi/linux/file.go
@@ -152,9 +152,10 @@ const (
 
 // Values for preadv2/pwritev2.
 const (
-	RWF_HIPRI = 0x0001
-	RWF_DSYNC = 0X0002
-	RWF_SYNC  = 0x0004
+	RWF_HIPRI = 0x00000001
+	RWF_DSYNC = 0x00000002
+	RWF_SYNC  = 0x00000004
+	RWF_VALID = RWF_HIPRI | RWF_DSYNC | RWF_SYNC
 )
 
 // Stat represents struct stat.
diff --git a/pkg/sentry/syscalls/linux/linux64.go b/pkg/sentry/syscalls/linux/linux64.go
index cc5ebb955..e855590e6 100644
--- a/pkg/sentry/syscalls/linux/linux64.go
+++ b/pkg/sentry/syscalls/linux/linux64.go
@@ -377,7 +377,7 @@ var AMD64 = &kernel.SyscallTable{
 		// Syscalls after 325 are "backports" from versions of Linux after 4.4.
 		//	326: CopyFileRange,
 		327: Preadv2,
-		//	328: Pwritev2,  // Pwritev2, TODO
+		328: Pwritev2,
 	},
 
 	Emulate: map[usermem.Addr]uintptr{
diff --git a/pkg/sentry/syscalls/linux/sys_read.go b/pkg/sentry/syscalls/linux/sys_read.go
index cbb9eb9f8..b6df4d9d4 100644
--- a/pkg/sentry/syscalls/linux/sys_read.go
+++ b/pkg/sentry/syscalls/linux/sys_read.go
@@ -188,14 +188,20 @@ func Preadv(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal
 }
 
 // Preadv2 implements linux syscall preadv2(2).
+// TODO: Implement RWF_HIPRI functionality.
 func Preadv2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+	// While the syscall is
+	// preadv2(int fd, struct iovec* iov, int iov_cnt, off_t offset, int flags)
+	// the linux internal call
+	// (https://elixir.bootlin.com/linux/v4.18/source/fs/read_write.c#L1248)
+	// splits the offset argument into a high/low value for compatibility with
+	// 32-bit architectures. The flags argument is the 5th argument.
+
 	fd := kdefs.FD(args[0].Int())
 	addr := args[1].Pointer()
 	iovcnt := int(args[2].Int())
 	offset := args[3].Int64()
-	flags := int(args[4].Int())
-
-	validFlags := linux.RWF_HIPRI
+	flags := int(args[5].Int())
 
 	file := t.FDMap().GetFile(fd)
 	if file == nil {
@@ -219,14 +225,8 @@ func Preadv2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca
 	}
 
 	// Check flags field.
-	if flags != 0 {
-		if flags&^validFlags != 0 {
-			return 0, nil, syserror.EINVAL
-		}
-		// RWF_HIPRI must be called on a file with O_DIRECT flag set.
-		if flags&linux.RWF_HIPRI != 0 && !file.Flags().Direct {
-			return 0, nil, syserror.EINVAL
-		}
+	if flags&^linux.RWF_VALID != 0 {
+		return 0, nil, syserror.EOPNOTSUPP
 	}
 
 	// Read the iovecs that specify the destination of the read.
diff --git a/pkg/sentry/syscalls/linux/sys_write.go b/pkg/sentry/syscalls/linux/sys_write.go
index 08e263112..750a098cd 100644
--- a/pkg/sentry/syscalls/linux/sys_write.go
+++ b/pkg/sentry/syscalls/linux/sys_write.go
@@ -187,6 +187,72 @@ func Pwritev(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca
 	return uintptr(n), nil, handleIOError(t, n != 0, err, kernel.ERESTARTSYS, "pwritev", file)
 }
 
+// Pwritev2 implements linux syscall pwritev2(2).
+// TODO: Implement RWF_HIPRI functionality.
+// TODO: Implement O_SYNC and D_SYNC functionality.
+func Pwritev2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+	// While the syscall is
+	// pwritev2(int fd, struct iovec* iov, int iov_cnt, off_t offset, int flags)
+	// the linux internal call
+	// (https://elixir.bootlin.com/linux/v4.18/source/fs/read_write.c#L1354)
+	// splits the offset argument into a high/low value for compatibility with
+	// 32-bit architectures. The flags argument is the 5th argument.
+
+	fd := kdefs.FD(args[0].Int())
+	addr := args[1].Pointer()
+	iovcnt := int(args[2].Int())
+	offset := args[3].Int64()
+	flags := int(args[5].Int())
+
+	if int(args[4].Int())&0x4 == 1 {
+		return 0, nil, syserror.EACCES
+	}
+
+	file := t.FDMap().GetFile(fd)
+	if file == nil {
+		return 0, nil, syserror.EBADF
+	}
+	defer file.DecRef()
+
+	// Check that the offset is legitimate.
+	if offset < -1 {
+		return 0, nil, syserror.EINVAL
+	}
+
+	// Is writing at an offset supported?
+	if offset > -1 && !file.Flags().Pwrite {
+		return 0, nil, syserror.ESPIPE
+	}
+
+	if flags&^linux.RWF_VALID != 0 {
+		return uintptr(flags), nil, syserror.EOPNOTSUPP
+	}
+
+	// Check that the file is writeable.
+	if !file.Flags().Write {
+		return 0, nil, syserror.EBADF
+	}
+
+	// Read the iovecs that specify the source of the write.
+	src, err := t.IovecsIOSequence(addr, iovcnt, usermem.IOOpts{
+		AddressSpaceActive: true,
+	})
+	if err != nil {
+		return 0, nil, err
+	}
+
+	// If pwritev2 is called with an offset of -1, writev is called.
+	if offset == -1 {
+		n, err := writev(t, file, src)
+		t.IOUsage().AccountWriteSyscall(n)
+		return uintptr(n), nil, handleIOError(t, n != 0, err, kernel.ERESTARTSYS, "pwritev2", file)
+	}
+
+	n, err := pwritev(t, file, src, offset)
+	t.IOUsage().AccountWriteSyscall(n)
+	return uintptr(n), nil, handleIOError(t, n != 0, err, kernel.ERESTARTSYS, "pwritev2", file)
+}
+
 func writev(t *kernel.Task, f *fs.File, src usermem.IOSequence) (int64, error) {
 	n, err := f.Writev(t, src)
 	if err != syserror.ErrWouldBlock || f.Flags().NonBlocking {