posix_fallocate: push vnop implementation into the fileop layer

This opens the door for other descriptor types to implement
posix_fallocate(2) as needed.

Reviewed by:	kib, bcr (manpages)
Differential Revision:	https://reviews.freebsd.org/D23042

3 files changed, 100 insertions, 93 deletions

diff --git a/sys/kern/sys_generic.c b/sys/kern/sys_generic.c
index cd78cfbfbba2..8e614a0accc6 100644
--- a/sys/kern/sys_generic.c
+++ b/sys/kern/sys_generic.c
@@ -819,6 +819,47 @@ out:
 }
 
 int
+sys_posix_fallocate(struct thread *td, struct posix_fallocate_args *uap)
+{
+	int error;
+
+	error = kern_posix_fallocate(td, uap->fd, uap->offset, uap->len);
+	return (kern_posix_error(td, error));
+}
+
+int
+kern_posix_fallocate(struct thread *td, int fd, off_t offset, off_t len)
+{
+	struct file *fp;
+	int error;
+
+	AUDIT_ARG_FD(fd);
+	if (offset < 0 || len <= 0)
+		return (EINVAL);
+	/* Check for wrap. */
+	if (offset > OFF_MAX - len)
+		return (EFBIG);
+	AUDIT_ARG_FD(fd);
+	error = fget(td, fd, &cap_pwrite_rights, &fp);
+	if (error != 0)
+		return (error);
+	AUDIT_ARG_FILE(td->td_proc, fp);
+	if ((fp->f_ops->fo_flags & DFLAG_SEEKABLE) == 0) {
+		error = ESPIPE;
+		goto out;
+	}
+	if ((fp->f_flag & FWRITE) == 0) {
+		error = EBADF;
+		goto out;
+	}
+
+	error = fo_fallocate(fp, offset, len, td);
+ out:
+	fdrop(fp, td);
+	return (error);
+}
+
+int
 poll_no_poll(int events)
 {
 	/*
diff --git a/sys/kern/vfs_syscalls.c b/sys/kern/vfs_syscalls.c
index b954e3a6dd45..a3b66f0bcf47 100644
--- a/sys/kern/vfs_syscalls.c
+++ b/sys/kern/vfs_syscalls.c
@@ -4565,99 +4565,6 @@ out:
 	return (error);
 }
 
-int
-kern_posix_fallocate(struct thread *td, int fd, off_t offset, off_t len)
-{
-	struct file *fp;
-	struct mount *mp;
-	struct vnode *vp;
-	off_t olen, ooffset;
-	int error;
-#ifdef AUDIT
-	int audited_vnode1 = 0;
-#endif
-
-	AUDIT_ARG_FD(fd);
-	if (offset < 0 || len <= 0)
-		return (EINVAL);
-	/* Check for wrap. */
-	if (offset > OFF_MAX - len)
-		return (EFBIG);
-	AUDIT_ARG_FD(fd);
-	error = fget(td, fd, &cap_pwrite_rights, &fp);
-	if (error != 0)
-		return (error);
-	AUDIT_ARG_FILE(td->td_proc, fp);
-	if ((fp->f_ops->fo_flags & DFLAG_SEEKABLE) == 0) {
-		error = ESPIPE;
-		goto out;
-	}
-	if ((fp->f_flag & FWRITE) == 0) {
-		error = EBADF;
-		goto out;
-	}
-	if (fp->f_type != DTYPE_VNODE) {
-		error = ENODEV;
-		goto out;
-	}
-	vp = fp->f_vnode;
-	if (vp->v_type != VREG) {
-		error = ENODEV;
-		goto out;
-	}
-
-	/* Allocating blocks may take a long time, so iterate. */
-	for (;;) {
-		olen = len;
-		ooffset = offset;
-
-		bwillwrite();
-		mp = NULL;
-		error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
-		if (error != 0)
-			break;
-		error = vn_lock(vp, LK_EXCLUSIVE);
-		if (error != 0) {
-			vn_finished_write(mp);
-			break;
-		}
-#ifdef AUDIT
-		if (!audited_vnode1) {
-			AUDIT_ARG_VNODE1(vp);
-			audited_vnode1 = 1;
-		}
-#endif
-#ifdef MAC
-		error = mac_vnode_check_write(td->td_ucred, fp->f_cred, vp);
-		if (error == 0)
-#endif
-			error = VOP_ALLOCATE(vp, &offset, &len);
-		VOP_UNLOCK(vp);
-		vn_finished_write(mp);
-
-		if (olen + ooffset != offset + len) {
-			panic("offset + len changed from %jx/%jx to %jx/%jx",
-			    ooffset, olen, offset, len);
-		}
-		if (error != 0 || len == 0)
-			break;
-		KASSERT(olen > len, ("Iteration did not make progress?"));
-		maybe_yield();
-	}
- out:
-	fdrop(fp, td);
-	return (error);
-}
-
-int
-sys_posix_fallocate(struct thread *td, struct posix_fallocate_args *uap)
-{
-	int error;
-
-	error = kern_posix_fallocate(td, uap->fd, uap->offset, uap->len);
-	return (kern_posix_error(td, error));
-}
-
 /*
  * Unlike madvise(2), we do not make a best effort to remember every
  * possible caching hint.  Instead, we remember the last setting with
diff --git a/sys/kern/vfs_vnops.c b/sys/kern/vfs_vnops.c
index f15ea0969b7e..61edb3165e01 100644
--- a/sys/kern/vfs_vnops.c
+++ b/sys/kern/vfs_vnops.c
@@ -103,6 +103,7 @@ static fo_kqfilter_t	vn_kqfilter;
 static fo_stat_t	vn_statfile;
 static fo_close_t	vn_closefile;
 static fo_mmap_t	vn_mmap;
+static fo_fallocate_t	vn_fallocate;
 
 struct 	fileops vnops = {
 	.fo_read = vn_io_fault,
@@ -119,6 +120,7 @@ struct 	fileops vnops = {
 	.fo_seek = vn_seek,
 	.fo_fill_kinfo = vn_fill_kinfo,
 	.fo_mmap = vn_mmap,
+	.fo_fallocate = vn_fallocate,
 	.fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
 };
 
@@ -3150,3 +3152,60 @@ out:
 	free(dat, M_TEMP);
 	return (error);
 }
+
+static int
+vn_fallocate(struct file *fp, off_t offset, off_t len, struct thread *td)
+{
+	struct mount *mp;
+	struct vnode *vp;
+	off_t olen, ooffset;
+	int error;
+#ifdef AUDIT
+	int audited_vnode1 = 0;
+#endif
+
+	vp = fp->f_vnode;
+	if (vp->v_type != VREG)
+		return (ENODEV);
+
+	/* Allocating blocks may take a long time, so iterate. */
+	for (;;) {
+		olen = len;
+		ooffset = offset;
+
+		bwillwrite();
+		mp = NULL;
+		error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
+		if (error != 0)
+			break;
+		error = vn_lock(vp, LK_EXCLUSIVE);
+		if (error != 0) {
+			vn_finished_write(mp);
+			break;
+		}
+#ifdef AUDIT
+		if (!audited_vnode1) {
+			AUDIT_ARG_VNODE1(vp);
+			audited_vnode1 = 1;
+		}
+#endif
+#ifdef MAC
+		error = mac_vnode_check_write(td->td_ucred, fp->f_cred, vp);
+		if (error == 0)
+#endif
+			error = VOP_ALLOCATE(vp, &offset, &len);
+		VOP_UNLOCK(vp);
+		vn_finished_write(mp);
+
+		if (olen + ooffset != offset + len) {
+			panic("offset + len changed from %jx/%jx to %jx/%jx",
+			    ooffset, olen, offset, len);
+		}
+		if (error != 0 || len == 0)
+			break;
+		KASSERT(olen > len, ("Iteration did not make progress?"));
+		maybe_yield();
+	}
+
+	return (error);
+}