vfs_subr.c

来自「基于组件方式开发操作系统的OSKIT源代码」· C语言 代码 · 共 2,733 行 · 第 1/5 页

	 * initialize uvm_object within vnode.
	 */

	uobj = &vp->v_uvm.u_obj;
	uobj->pgops = &uvm_vnodeops;
	TAILQ_INIT(&uobj->memq);
	vp->v_uvm.u_size = VSIZENOTSET;

	if (mp && error != EDEADLK)
		vfs_unbusy(mp);
	return (0);
}

/*
 * This is really just the reverse of getnewvnode(). Needed for
 * VFS_VGET functions who may need to push back a vnode in case
 * of a locking race.
 */
void
ungetnewvnode(vp)
	struct vnode *vp;
{
#ifdef DIAGNOSTIC
	if (vp->v_usecount != 1)
		panic("ungetnewvnode: busy vnode");
#endif
	vp->v_usecount--;
	insmntque(vp, NULL);
	vp->v_type = VBAD;

	simple_lock(&vp->v_interlock);
	/* 
	 * Insert at head of LRU list
	 */
	simple_lock(&vnode_free_list_slock);
	if (vp->v_holdcnt > 0)
		TAILQ_INSERT_HEAD(&vnode_hold_list, vp, v_freelist);
	else
		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
	simple_unlock(&vnode_free_list_slock); 
	simple_unlock(&vp->v_interlock);
}

/*
 * Move a vnode from one mount queue to another.
 */
void
insmntque(vp, mp)
	struct vnode *vp;
	struct mount *mp;
{

#ifdef DIAGNOSTIC
	if ((mp != NULL) &&
	    (mp->mnt_flag & MNT_UNMOUNT) &&
	    !(mp->mnt_flag & MNT_SOFTDEP) &&
	    vp->v_tag != VT_VFS) {
		panic("insmntque into dying filesystem");
	}
#endif
	
	simple_lock(&mntvnode_slock);
	/*
	 * Delete from old mount point vnode list, if on one.
	 */
	if (vp->v_mount != NULL)
		LIST_REMOVE(vp, v_mntvnodes);
	/*
	 * Insert into list of vnodes for the new mount point, if available.
	 */
	if ((vp->v_mount = mp) != NULL)
		LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes);
	simple_unlock(&mntvnode_slock);
}
#endif /*OSKIT*/

/*
 * Update outstanding I/O count and do wakeup if requested.
 */
void
vwakeup(bp)
	struct buf *bp;
{
	struct vnode *vp;

	if ((vp = bp->b_vp) != NULL) {
		if (--vp->v_numoutput < 0)
			panic("vwakeup: neg numoutput, vp %p", vp);
		if ((vp->v_flag & VBWAIT) && vp->v_numoutput <= 0) {
			vp->v_flag &= ~VBWAIT;
			wakeup((caddr_t)&vp->v_numoutput);
		}
	}
}

#ifndef OSKIT
/*
 * Flush out and invalidate all buffers associated with a vnode.
 * Called with the underlying vnode locked, which should prevent new dirty
 * buffers from being queued.
 */
int
vinvalbuf(vp, flags, cred, p, slpflag, slptimeo)
	struct vnode *vp;
	int flags;
	struct ucred *cred;
	struct proc *p;
	int slpflag, slptimeo;
{
	struct uvm_object *uobj = &vp->v_uvm.u_obj;
	struct buf *bp, *nbp;
	int s, error, rv;
	int flushflags = PGO_ALLPAGES|PGO_FREE|PGO_SYNCIO|
		(flags & V_SAVE ? PGO_CLEANIT : 0);

	/* XXXUBC this doesn't look at flags or slp* */
	if (vp->v_type == VREG) {
		simple_lock(&uobj->vmobjlock);
		rv = (uobj->pgops->pgo_flush)(uobj, 0, 0, flushflags);
		simple_unlock(&uobj->vmobjlock);
		if (!rv) {
			return EIO;
		}
	}
	if (flags & V_SAVE) {
		error = VOP_FSYNC(vp, cred, FSYNC_WAIT|FSYNC_RECLAIM, 0, 0, p);
		if (error)
		        return (error);
#ifdef DIAGNOSTIC
		s = splbio();
		if (vp->v_numoutput > 0 || !LIST_EMPTY(&vp->v_dirtyblkhd))
		        panic("vinvalbuf: dirty bufs, vp %p", vp);
		splx(s);
#endif
	}

	s = splbio();

restart:
	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
		nbp = LIST_NEXT(bp, b_vnbufs);
		if (bp->b_flags & B_BUSY) {
			bp->b_flags |= B_WANTED;
			error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1),
			    "vinvalbuf", slptimeo);
			if (error) {
				splx(s);
				return (error);
			}
			goto restart;
		}
		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
		brelse(bp);
	}

	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
		nbp = LIST_NEXT(bp, b_vnbufs);
		if (bp->b_flags & B_BUSY) {
			bp->b_flags |= B_WANTED;
			error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1),
			    "vinvalbuf", slptimeo);
			if (error) {
				splx(s);
				return (error);
			}
			goto restart;
		}
		/*
		 * XXX Since there are no node locks for NFS, I believe
		 * there is a slight chance that a delayed write will
		 * occur while sleeping just above, so check for it.
		 */
		if ((bp->b_flags & B_DELWRI) && (flags & V_SAVE)) {
#ifdef DEBUG
			printf("buffer still DELWRI\n");
#endif
			bp->b_flags |= B_BUSY | B_VFLUSH;
			VOP_BWRITE(bp);
			goto restart;
		}
		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
		brelse(bp);
	}

#ifdef DIAGNOSTIC
	if (!LIST_EMPTY(&vp->v_cleanblkhd) || !LIST_EMPTY(&vp->v_dirtyblkhd))
		panic("vinvalbuf: flush failed, vp %p", vp);
#endif

	splx(s);

	return (0);
}

/*
 * Destroy any in core blocks past the truncation length.
 * Called with the underlying vnode locked, which should prevent new dirty
 * buffers from being queued.
 */
int
vtruncbuf(vp, lbn, slpflag, slptimeo)
	struct vnode *vp;
	daddr_t lbn;
	int slpflag, slptimeo;
{
	struct uvm_object *uobj = &vp->v_uvm.u_obj;
	struct buf *bp, *nbp;
	int s, error, rv;

	s = splbio();
	if (vp->v_type == VREG) {
		simple_lock(&uobj->vmobjlock);
		rv = (uobj->pgops->pgo_flush)(uobj,
		    round_page(lbn << vp->v_mount->mnt_fs_bshift),
		    vp->v_uvm.u_size, PGO_FREE);
		simple_unlock(&uobj->vmobjlock);
		if (!rv) {
			splx(s);
			return EIO;
		}
	}

restart:
	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
		nbp = LIST_NEXT(bp, b_vnbufs);
		if (bp->b_lblkno < lbn)
			continue;
		if (bp->b_flags & B_BUSY) {
			bp->b_flags |= B_WANTED;
			error = tsleep(bp, slpflag | (PRIBIO + 1),
			    "vtruncbuf", slptimeo);
			if (error) {
				splx(s);
				return (error);
			}
			goto restart;
		}
		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
		brelse(bp);
	}

	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
		nbp = LIST_NEXT(bp, b_vnbufs);
		if (bp->b_lblkno < lbn)
			continue;
		if (bp->b_flags & B_BUSY) {
			bp->b_flags |= B_WANTED;
			error = tsleep(bp, slpflag | (PRIBIO + 1),
			    "vtruncbuf", slptimeo);
			if (error) {
				splx(s);
				return (error);
			}
			goto restart;
		}
		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
		brelse(bp);
	}

	splx(s);

	return (0);
}

void
vflushbuf(vp, sync)
	struct vnode *vp;
	int sync;
{
	struct uvm_object *uobj = &vp->v_uvm.u_obj;
	struct buf *bp, *nbp;
	int s;

	if (vp->v_type == VREG) {
		int flags = PGO_CLEANIT|PGO_ALLPAGES| (sync ? PGO_SYNCIO : 0);

		simple_lock(&uobj->vmobjlock);
		(uobj->pgops->pgo_flush)(uobj, 0, 0, flags);
		simple_unlock(&uobj->vmobjlock);
	}

loop:
	s = splbio();
	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
		nbp = LIST_NEXT(bp, b_vnbufs);
		if ((bp->b_flags & B_BUSY))
			continue;
		if ((bp->b_flags & B_DELWRI) == 0)
			panic("vflushbuf: not dirty, bp %p", bp);
		bp->b_flags |= B_BUSY | B_VFLUSH;
		splx(s);
		/*
		 * Wait for I/O associated with indirect blocks to complete,
		 * since there is no way to quickly wait for them below.
		 */
		if (bp->b_vp == vp || sync == 0)
			(void) bawrite(bp);
		else
			(void) bwrite(bp);
		goto loop;
	}
	if (sync == 0) {
		splx(s);
		return;
	}
	while (vp->v_numoutput) {
		vp->v_flag |= VBWAIT;
		tsleep((caddr_t)&vp->v_numoutput, PRIBIO + 1, "vflushbuf", 0);
	}
	splx(s);
	if (!LIST_EMPTY(&vp->v_dirtyblkhd)) {
		vprint("vflushbuf: dirty", vp);
		goto loop;
	}
}
#endif /*OSKIT*/

/*
 * Associate a buffer with a vnode.
 */
void
bgetvp(vp, bp)
	struct vnode *vp;
	struct buf *bp;
{
	int s;

	if (bp->b_vp)
		panic("bgetvp: not free, bp %p", bp);
	VHOLD(vp);
	s = splbio();
	bp->b_vp = vp;
	if (vp->v_type == VBLK || vp->v_type == VCHR)
		bp->b_dev = vp->v_rdev;
	else
		bp->b_dev = NODEV;
	/*
	 * Insert onto list for new vnode.
	 */
	bufinsvn(bp, &vp->v_cleanblkhd);
	splx(s);
}

/*
 * Disassociate a buffer from a vnode.
 */
void
brelvp(bp)
	struct buf *bp;
{
	struct vnode *vp;
	int s;

	if (bp->b_vp == NULL)
		panic("brelvp: vp NULL, bp %p", bp);

	s = splbio();
	vp = bp->b_vp;
	/*
	 * Delete from old vnode list, if on one.
	 */
	if (bp->b_vnbufs.le_next != NOLIST)
		bufremvn(bp);

	if (vp->v_type != VREG && (vp->v_flag & VONWORKLST) &&
	    LIST_FIRST(&vp->v_dirtyblkhd) == NULL) {
		vp->v_flag &= ~VONWORKLST;
		LIST_REMOVE(vp, v_synclist);
	}

	bp->b_vp = NULL;
	HOLDRELE(vp);
	splx(s);
}

#ifndef OSKIT
/*
 * Reassign a buffer from one vnode to another.
 * Used to assign file specific control information
 * (indirect blocks) to the vnode to which they belong.
 *
 * This function must be called at splbio().
 */
void
reassignbuf(bp, newvp)
	struct buf *bp;
	struct vnode *newvp;
{
	struct buflists *listheadp;
	int delay;

	/*
	 * Delete from old vnode list, if on one.
	 */
	if (bp->b_vnbufs.le_next != NOLIST)
		bufremvn(bp);
	/*
	 * If dirty, put on list of dirty buffers;
	 * otherwise insert onto list of clean buffers.
	 */
	if ((bp->b_flags & B_DELWRI) == 0) {
		listheadp = &newvp->v_cleanblkhd;
		if (newvp->v_type != VREG &&
		    (newvp->v_flag & VONWORKLST) &&
		    LIST_FIRST(&newvp->v_dirtyblkhd) == NULL) {
			newvp->v_flag &= ~VONWORKLST;
			LIST_REMOVE(newvp, v_synclist);
		}
	} else {
		listheadp = &newvp->v_dirtyblkhd;
		if ((newvp->v_flag & VONWORKLST) == 0) {
			switch (newvp->v_type) {
			case VDIR:
				delay = dirdelay;
				break;
			case VBLK:
				if (newvp->v_specmountpoint != NULL) {
					delay = metadelay;
					break;
				}
				/* fall through */
			default:
				delay = filedelay;
				break;
			}
			if (!newvp->v_mount ||
			    (newvp->v_mount->mnt_flag & MNT_ASYNC) == 0)
				vn_syncer_add_to_worklist(newvp, delay);
		}
	}
	bufinsvn(bp, listheadp);
}

/*
 * Create a vnode for a block device.
 * Used for root filesystem and swap areas.
 * Also used for memory file system special devices.
 */
int
bdevvp(dev, vpp)
	dev_t dev;
	struct vnode **vpp;
{

	return (getdevvp(dev, vpp, VBLK));
}

/*
 * Create a vnode for a character device.
 * Used for kernfs and some console handling.
 */
int
cdevvp(dev, vpp)
	dev_t dev;
	struct vnode **vpp;
{

	return (getdevvp(dev, vpp, VCHR));
}

/*
 * Create a vnode for a device.
 * Used by bdevvp (block device) for root file system etc.,
 * and by cdevvp (character device) for console and kernfs.
 */
int
getdevvp(dev, vpp, type)
	dev_t dev;
	struct vnode **vpp;
	enum vtype type;
{
	struct vnode *vp;
	struct vnode *nvp;
	int error;

	if (dev == NODEV) {
		*vpp = NULLVP;
		return (0);
	}
	error = getnewvnode(VT_NON, NULL, spec_vnodeop_p, &nvp);
	if (error) {
		*vpp = NULLVP;
		return (error);
	}
	vp = nvp;
	vp->v_type = type;
	if ((nvp = checkalias(vp, dev, NULL)) != 0) {
		vput(vp);
		vp = nvp;
	}
	*vpp = vp;
	return (0);
}

/*
 * Check to see if the new vnode represents a special device
 * for which we already have a vnode (either because of
 * bdevvp() or because of a different vnode representing
 * the same block device). If such an alias exists, deallocate
 * the existing contents and return the aliased vnode. The
 * caller is responsible for filling it with its new contents.
 */
struct vnode *
checkalias(nvp, nvp_rdev, mp)
	struct vnode *nvp;
	dev_t nvp_rdev;
	struct mount *mp;
{
	struct proc *p = curproc;       /* XXX */
	struct vnode *vp;
	struct vnode **vpp;

	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
		return (NULLVP);

	vpp = &speclisth[SPECHASH(nvp_rdev)];
loop:
	simple_lock(&spechash_slock);
	for (vp = *vpp; vp; vp = vp->v_specnext) {
		if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type)
			continue;
		/*
		 * Alias, but not in use, so flush it out.
		 */
		simple_lock(&vp->v_interlock);
		if (vp->v_usecount == 0) {
			simple_unlock(&spechash_slock);
			vgonel(vp, p);
			goto loop;
		}
		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK)) {
			simple_unlock(&spechash_slock);
			goto loop;
		}
		break;
	}
	if (vp == NULL || vp->v_tag != VT_NON || vp->v_type != VBLK) {
		MALLOC(nvp->v_specinfo, struct specinfo *,
			sizeof(struct specinfo), M_VNODE, M_WAITOK);
		nvp->v_rdev = nvp_rdev;
		nvp->v_hashchain = vpp;
		nvp->v_specnext = *vpp;
		nvp->v_specmountpoint = NULL;
		simple_unlock(&spechash_slock);
		nvp->v_speclockf = NULL;
		*vpp = nvp;
		if (vp != NULLVP) {