lio_listio: validate aio_lio_opcode

Previously, we would accept any kind of LIO_* opcode, including ones
that were intended for in-kernel use only like LIO_SYNC (which is not
defined in userland).  The situation became more serious with
022ca2fc7fe08d51f33a1d23a9be49e6d132914e.  After that revision, setting
aio_lio_opcode to LIO_WRITEV or LIO_READV would trigger an assertion.

Note that POSIX does not specify what should happen if aio_lio_opcode is
invalid.

MFC-with:	022ca2fc7fe08d51f33a1d23a9be49e6d132914e
Reviewed by:	jhb, tmunro, 0mp
Differential Revision:	<https://reviews.freebsd.org/D28078

4 files changed, 74 insertions, 7 deletions

diff --git a/lib/libc/sys/aio_return.2 b/lib/libc/sys/aio_return.2
index d94fcc7eba62..aebe978fe140 100644
--- a/lib/libc/sys/aio_return.2
+++ b/lib/libc/sys/aio_return.2
@@ -24,7 +24,7 @@
 .\"
 .\" $FreeBSD$
 .\"
-.Dd January 2, 2021
+.Dd January 11, 2021
 .Dt AIO_RETURN 2
 .Os
 .Sh NAME
@@ -75,6 +75,12 @@ The
 .Fa iocb
 argument
 does not reference a completed asynchronous I/O request.
+.It Bq Er EINVAL
+The I/O operation was submitted with
+.Fn lio_listio ,
+and the value of the
+.Fa aio_lio_opcode
+is invalid.
 .El
 .Sh SEE ALSO
 .Xr aio_cancel 2 ,
diff --git a/sys/kern/vfs_aio.c b/sys/kern/vfs_aio.c
index 8666d6ea4217..bc0d7e04c9d5 100644
--- a/sys/kern/vfs_aio.c
+++ b/sys/kern/vfs_aio.c
@@ -1556,16 +1556,26 @@ aio_aqueue(struct thread *td, struct aiocb *ujob, struct aioliojob *lj,
 		goto err2;
 	}
 
+	/* Get the opcode. */
+	if (type == LIO_NOP) {
+		switch (job->uaiocb.aio_lio_opcode) {
+		case LIO_WRITE:
+		case LIO_NOP:
+		case LIO_READ:
+			opcode = job->uaiocb.aio_lio_opcode;
+			break;
+		default:
+			error = EINVAL;
+			goto err2;
+		}
+	} else
+		opcode = job->uaiocb.aio_lio_opcode = type;
+
 	ksiginfo_init(&job->ksi);
 
 	/* Save userspace address of the job info. */
 	job->ujob = ujob;
 
-	/* Get the opcode. */
-	if (type != LIO_NOP)
-		job->uaiocb.aio_lio_opcode = type;
-	opcode = job->uaiocb.aio_lio_opcode;
-
 	/*
 	 * Validate the opcode and fetch the file object for the specified
 	 * file descriptor.
diff --git a/sys/sys/aio.h b/sys/sys/aio.h
index dbfbadcd1254..ee928b8bf846 100644
--- a/sys/sys/aio.h
+++ b/sys/sys/aio.h
@@ -43,7 +43,7 @@
 #define	LIO_NOP			0x0
 #define LIO_WRITE		0x1
 #define	LIO_READ		0x2
-#ifdef _KERNEL
+#if defined(_KERNEL) || defined(_WANT_ALL_LIO_OPCODES)
 #define	LIO_SYNC		0x3
 #define	LIO_MLOCK		0x4
 #define	LIO_WRITEV		0x5
diff --git a/tests/sys/aio/lio_test.c b/tests/sys/aio/lio_test.c
index b6965cc9a4ee..fb519aac978d 100644
--- a/tests/sys/aio/lio_test.c
+++ b/tests/sys/aio/lio_test.c
@@ -26,8 +26,11 @@
  * $FreeBSD$
  */
 
+#define _WANT_ALL_LIO_OPCODES
+
 #include <sys/param.h>
 #include <sys/event.h>
+#include <sys/uio.h>
 
 #include <aio.h>
 #include <fcntl.h>
@@ -198,6 +201,53 @@ ATF_TC_BODY(lio_listio_empty_nowait_thread, tc)
 	ATF_REQUIRE_EQ(0, sem_destroy(&completions));
 }
 
+/*
+ * Only select opcodes are allowed with lio_listio
+ */
+ATF_TC_WITHOUT_HEAD(lio_listio_invalid_opcode);
+ATF_TC_BODY(lio_listio_invalid_opcode, tc)
+{
+	struct aiocb sync_cb, mlock_cb, writev_cb, readv_cb;
+	struct aiocb *list[] = {&sync_cb, &mlock_cb, &writev_cb, &readv_cb};
+	struct iovec iov;
+	int fd;
+
+	fd = open("testfile", O_CREAT | O_RDWR);
+	ATF_REQUIRE_MSG(fd >= 0, "open: %s", strerror(errno));
+
+	bzero(&sync_cb, sizeof(sync_cb));
+	sync_cb.aio_fildes = fd;
+	sync_cb.aio_lio_opcode = LIO_SYNC;
+
+	bzero(&mlock_cb, sizeof(mlock_cb));
+	mlock_cb.aio_lio_opcode = LIO_MLOCK;
+
+	iov.iov_base = NULL;
+	iov.iov_len = 0;
+
+	bzero(&readv_cb, sizeof(readv_cb));
+	readv_cb.aio_fildes = fd;
+	readv_cb.aio_lio_opcode = LIO_READV;
+	readv_cb.aio_iov = &iov;
+	readv_cb.aio_iovcnt = 1;
+
+	bzero(&writev_cb, sizeof(writev_cb));
+	writev_cb.aio_fildes = fd;
+	writev_cb.aio_lio_opcode = LIO_WRITEV;
+	writev_cb.aio_iov = &iov;
+	writev_cb.aio_iovcnt = 1;
+
+	ATF_CHECK_ERRNO(EIO, lio_listio(LIO_WAIT, list, nitems(list), NULL));
+	ATF_CHECK_EQ(EINVAL, aio_error(&sync_cb));
+	ATF_CHECK_ERRNO(EINVAL, aio_return(&sync_cb) < 0);
+	ATF_CHECK_EQ(EINVAL, aio_error(&mlock_cb));
+	ATF_CHECK_ERRNO(EINVAL, aio_return(&mlock_cb) < 0);
+	ATF_CHECK_EQ(EINVAL, aio_error(&readv_cb));
+	ATF_CHECK_ERRNO(EINVAL, aio_return(&readv_cb) < 0);
+	ATF_CHECK_EQ(EINVAL, aio_error(&writev_cb));
+	ATF_CHECK_ERRNO(EINVAL, aio_return(&writev_cb) < 0);
+}
+
 
 ATF_TP_ADD_TCS(tp)
 {
@@ -207,6 +257,7 @@ ATF_TP_ADD_TCS(tp)
 	ATF_TP_ADD_TC(tp, lio_listio_empty_nowait_signal);
 	ATF_TP_ADD_TC(tp, lio_listio_empty_nowait_thread);
 	ATF_TP_ADD_TC(tp, lio_listio_empty_wait);
+	ATF_TP_ADD_TC(tp, lio_listio_invalid_opcode);
 
 	return (atf_no_error());
 }