Reorder the managament of advisory locks on open files so that the advisory

lock is obtained before the write count is increased during open() and the
lock is released after the write count is decreased during close().

The first change closes a race where an open() that will block with O_SHLOCK
or O_EXLOCK can increase the write count while it waits.  If the process
holding the current lock on the file then tries to call exec() on the file
it has locked, it can fail with ETXTBUSY even though the advisory lock is
preventing other threads from succesfully completeing a writable open().

The second change closes a race where a read-only open() with O_SHLOCK or
O_EXLOCK may return successfully while the write count is non-zero due to
another descriptor that had the advisory lock and was blocking the open()
still being in the process of closing.  If the process that completed the
open() then attempts to call exec() on the file it locked, it can fail with
ETXTBUSY even though the other process that held a write lock has closed
the file and released the lock.

Reviewed by:	kib
MFC after:	1 month

2 files changed, 61 insertions, 43 deletions

diff --git a/sys/kern/vfs_syscalls.c b/sys/kern/vfs_syscalls.c
index 66ab5a3..cc6d93c 100644
--- a/sys/kern/vfs_syscalls.c
+++ b/sys/kern/vfs_syscalls.c
@@ -1093,8 +1093,7 @@ kern_openat(struct thread *td, int fd, char *path, enum uio_seg pathseg,
 	struct file *fp;
 	struct vnode *vp;
 	int cmode;
-	int type, indx = -1, error;
-	struct flock lf;
+	int indx = -1, error;
 	struct nameidata nd;
 	int vfslocked;
 	cap_rights_t rights_needed = CAP_LOOKUP;
@@ -1180,26 +1179,11 @@ kern_openat(struct thread *td, int fd, char *path, enum uio_seg pathseg,
 	if (fp->f_ops == &badfileops) {
 		KASSERT(vp->v_type != VFIFO, ("Unexpected fifo."));
 		fp->f_seqcount = 1;
-		finit(fp, flags & FMASK, DTYPE_VNODE, vp, &vnops);
+		finit(fp, (flags & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
+		    vp, &vnops);
 	}
 
 	VOP_UNLOCK(vp, 0);
-	if (fp->f_type == DTYPE_VNODE && (flags & (O_EXLOCK | O_SHLOCK)) != 0) {
-		lf.l_whence = SEEK_SET;
-		lf.l_start = 0;
-		lf.l_len = 0;
-		if (flags & O_EXLOCK)
-			lf.l_type = F_WRLCK;
-		else
-			lf.l_type = F_RDLCK;
-		type = F_FLOCK;
-		if ((flags & FNONBLOCK) == 0)
-			type |= F_WAIT;
-		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
-		    type)) != 0)
-			goto bad;
-		atomic_set_int(&fp->f_flag, FHASLOCK);
-	}
 	if (flags & O_TRUNC) {
 		error = fo_truncate(fp, 0, td->td_ucred, td);
 		if (error)
@@ -4483,9 +4467,8 @@ sys_fhopen(td, uap)
 	struct mount *mp;
 	struct vnode *vp;
 	struct fhandle fhp;
-	struct flock lf;
 	struct file *fp;
-	int fmode, error, type;
+	int fmode, error;
 	int vfslocked;
 	int indx;
 
@@ -4542,24 +4525,9 @@ sys_fhopen(td, uap)
 #endif
 	fp->f_vnode = vp;
 	fp->f_seqcount = 1;
-	finit(fp, fmode & FMASK, DTYPE_VNODE, vp, &vnops);
+	finit(fp, (fmode & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE, vp,
+	    &vnops);
 	VOP_UNLOCK(vp, 0);
-	if (fmode & (O_EXLOCK | O_SHLOCK)) {
-		lf.l_whence = SEEK_SET;
-		lf.l_start = 0;
-		lf.l_len = 0;
-		if (fmode & O_EXLOCK)
-			lf.l_type = F_WRLCK;
-		else
-			lf.l_type = F_RDLCK;
-		type = F_FLOCK;
-		if ((fmode & FNONBLOCK) == 0)
-			type |= F_WAIT;
-		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
-		    type)) != 0)
-			goto bad;
-		atomic_set_int(&fp->f_flag, FHASLOCK);
-	}
 	if (fmode & O_TRUNC) {
 		error = fo_truncate(fp, 0, td->td_ucred, td);
 		if (error)
diff --git a/sys/kern/vfs_vnops.c b/sys/kern/vfs_vnops.c
index b5dcadf..0ad2db8 100644
--- a/sys/kern/vfs_vnops.c
+++ b/sys/kern/vfs_vnops.c
@@ -229,8 +229,10 @@ int
 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred,
     struct thread *td, struct file *fp)
 {
+	struct mount *mp;
 	accmode_t accmode;
-	int error;
+	struct flock lf;
+	int error, have_flock, lock_flags, type;
 
 	VFS_ASSERT_GIANT(vp->v_mount);
 	if (vp->v_type == VLNK)
@@ -271,6 +273,51 @@ vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred,
 	if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
 		return (error);
 
+	if (fmode & (O_EXLOCK | O_SHLOCK)) {
+		KASSERT(fp != NULL, ("open with flock requires fp"));
+		lock_flags = VOP_ISLOCKED(vp);
+		VOP_UNLOCK(vp, 0);
+		lf.l_whence = SEEK_SET;
+		lf.l_start = 0;
+		lf.l_len = 0;
+		if (fmode & O_EXLOCK)
+			lf.l_type = F_WRLCK;
+		else
+			lf.l_type = F_RDLCK;
+		type = F_FLOCK;
+		if ((fmode & FNONBLOCK) == 0)
+			type |= F_WAIT;
+		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type);
+		have_flock = (error == 0);
+		vn_lock(vp, lock_flags | LK_RETRY);
+		if (error == 0 && vp->v_iflag & VI_DOOMED)
+			error = ENOENT;
+		/*
+		 * Another thread might have used this vnode as an
+		 * executable while the vnode lock was dropped.
+		 * Ensure the vnode is still able to be opened for
+		 * writing after the lock has been obtained.
+		 */
+		if (error == 0 && accmode & VWRITE)
+			error = vn_writechk(vp);
+		if (error) {
+			VOP_UNLOCK(vp, 0);
+			if (have_flock) {
+				lf.l_whence = SEEK_SET;
+				lf.l_start = 0;
+				lf.l_len = 0;
+				lf.l_type = F_UNLCK;
+				(void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf,
+				    F_FLOCK);
+			}
+			vn_start_write(vp, &mp, V_WAIT);
+			vn_lock(vp, lock_flags | LK_RETRY);
+			(void)VOP_CLOSE(vp, fmode, cred, td);
+			vn_finished_write(mp);
+			return (error);
+		}
+		fp->f_flag |= FHASLOCK;
+	}
 	if (fmode & FWRITE) {
 		vp->v_writecount++;
 		CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
@@ -1400,19 +1447,22 @@ vn_closefile(fp, td)
 	int error;
 
 	vp = fp->f_vnode;
+	fp->f_ops = &badfileops;
 
 	vfslocked = VFS_LOCK_GIANT(vp->v_mount);
+	if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK)
+		vref(vp);
+
+	error = vn_close(vp, fp->f_flag, fp->f_cred, td);
+
 	if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
 		lf.l_whence = SEEK_SET;
 		lf.l_start = 0;
 		lf.l_len = 0;
 		lf.l_type = F_UNLCK;
 		(void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
+		vrele(vp);
 	}
-
-	fp->f_ops = &badfileops;
-
-	error = vn_close(vp, fp->f_flag, fp->f_cred, td);
 	VFS_UNLOCK_GIANT(vfslocked);
 	return (error);
 }