NAME

vnsubr, vn_close, vn_default_error, vn_isunder, vn_lock, vn_marktext, vn_rdwr, vn_open, vn_stat, vn_writechk —

high-level convenience functions for vnode operations

SYNOPSIS

#include <sys/param.h>
#include <sys/lock.h>
#include <sys/vnode.h>

int
vn_close(struct vnode *vp, int flags, struct ucred *cred, struct proc *p);

int
vn_default_error(void *v);

int
vn_isunder(struct vnode *dvp, struct vnode *rvp, struct proc *p);

int
vn_lock(struct vnode *vp, int flags);

void
vn_marktext(struct vnode *vp);

int
vn_open(struct nameidata *ndp, int fmode, int cmode);

int
vn_rdwr(enum uio_rw rw, struct vnode *vp, caddr_t base, int len, off_t offset, enum uio_seg segflg, int ioflg, struct ucred *cred, size_t *aresid, struct proc *p);

int
vn_stat(struct vnode *vp, struct stat *sb, struct proc *p);

int
vn_writechk(struct vnode *vp);

DESCRIPTION

The high-level functions described in this page are convenience functions for simplified access to the vnode operations described in VOP_LOOKUP(9).

vn_close(vp, flags, cred, p)

Common code for a vnode close. The argument vp is the unlocked vnode of the vnode to close. vn_close() simply locks the vnode, invokes the vnode operation VOP_CLOSE() and calls vput(9) to return the vnode to the freelist or holdlist. Note that vn_close() expects an unlocked, referenced vnode and will dereference the vnode prior to returning. If the operation is successful, zero is returned; otherwise an appropriate error is returned.

vn_default_error(v)

A generic "default" routine that just returns error. It is used by a file system to specify unsupported operations in the vnode operations vector.

vn_isunder(dvp, rvp, p)

Common code to check if one directory specified by the vnode rvp can be found inside the directory specified by the vnode dvp. The argument p is the calling process. vn_isunder() is intended to be used in chroot(2), chdir(2), fchdir(2), etc., to ensure that chroot(2) actually means something. If the operation is successful, zero is returned; otherwise 1 is returned.

vn_lock(vp, flags)

Acquire the vnode lock. Certain file system operations require that the vnode lock be held when they are called.

The vn_lock() function must not be called when the vnode's reference count is zero. Instead, the vget(9) function should be used.

In addition to the flags accepted by VOP_LOCK(9), the LK_RETRY flag may be used. LK_RETRY causes vn_lock() to return the vnode even if it has been reclaimed. It must not be used with LK_NOWAIT.

The vn_lock() function can sleep.

vn_marktext(vp)

Common code to mark the vnode vp as being the text of a running process.

vn_open(ndp, fmode, cmode)

Common code for vnode open operations. The pathname is described in the nameidata pointer (see namei(9)). The arguments fmode and cmode specify the open(2) file mode and the access permissions for creation. vn_open() checks permissions and invokes the VOP_OPEN(9) or VOP_CREATE(9) vnode operations. If the operation is successful, zero is returned; otherwise an appropriate error code is returned.

vn_rdwr(rw, vp, base, len, offset, segflg, ioflg, cred, aresid, p);

Common code to package up an I/O request on a vnode into a uio and then perform the I/O. The argument rw specifies whether the I/O is a read (UIO_READ) or write (UIO_WRITE) operation. The unlocked vnode is specified by vp. The arguments p and cred are the calling process and its credentials. The remaining arguments specify the uio parameters. For further information on these parameters, see uiomove(9).

vn_stat(vp, sb, p)

Common code for a vnode stat operation. The vnode is specified by the argument vp, and sb is the buffer in which to store the stat information. The argument p is the calling process. vn_stat() basically calls the vnode operation VOP_GETATTR(9) and transfers the contents of a vattr structure into a struct stat. If the operation is successful, zero is returned; otherwise an appropriate error code is returned.

vn_writechk(vp)

Common code to check for write permission on the vnode vp. A vnode is read-only if it is in use as a process's text image. If the vnode is read-only, ETXTBSY is returned; otherwise zero is returned to indicate that the vnode can be written to.

ERRORS

[ETXTBSY]: Cannot write to a vnode since it is a process's text image.
[ENOENT]: The vnode has been reclaimed and is dead. This error is only returned if the LK_RETRY flag is not passed to vn_lock().
[EBUSY]: The LK_NOWAIT flag was set and vn_lock() would have slept.

CODE REFERENCES

This section describes places within the OpenBSD source tree where actual code implementing or using the vnode framework can be found. All pathnames are relative to /usr/src.

The high-level convenience functions are implemented within the files sys/kern/vfs_vnops.c and sys/sys/vnode.h.

BUGS

The locking discipline is bizarre. Many vnode operations are passed locked vnodes on entry but release the lock before they exit. Discussions with Kirk McKusick indicate that locking discipline evolved out of the pre-VFS way of doing inode locking. In addition, the current locking discipline may actually save lines of code, especially if the number of file systems is fewer than the number of call sites. However, the VFS interface would require less wizardry if the locking discipline were simpler.

The locking discipline is used in some places to attempt to make a series of operations atomic (e.g., permissions check + operation). This does not work for non-local file systems that do not support locking (e.g., NFS).

Are vnode locks even necessary? The security checks can be moved into the individual file systems. Each file system can have the responsibility of ensuring that vnode operations are suitably atomic.

The LK_NOWAIT flag does prevent the caller from sleeping.

The locking discipline as it relates to shared locks has yet to be defined.

OpenBSD manual page server