*
* %sccs.include.redist.c%
*
- * @(#)union_vfsops.c 1.1 (Berkeley) %G%
+ * @(#)union_vfsops.c 8.14 (Berkeley) %G%
*/
/*
- * Null Layer
- * (See union_vnops.c for a description of what this does.)
+ * Union Layer
*/
#include <sys/param.h>
#include <sys/mount.h>
#include <sys/namei.h>
#include <sys/malloc.h>
-#include "union.h"
+#include <sys/filedesc.h>
+#include <sys/queue.h>
+#include <miscfs/union/union.h>
/*
* Mount union filesystem
{
int error = 0;
struct union_args args;
- struct vnode *lowerrootvp;
- struct vnode *upperrootvp;
- struct union_mount *um;
+ struct vnode *lowerrootvp = NULLVP;
+ struct vnode *upperrootvp = NULLVP;
+ struct union_mount *um = 0;
+ struct ucred *cred = 0;
+ struct ucred *scred;
+ struct vattr va;
+ char *cp;
+ int len;
u_int size;
#ifdef UNION_DIAGNOSTIC
/*
* Update is a no-op
*/
- if (mp->mnt_flag & MNT_UPDATE)
- return (EOPNOTSUPP);
+ if (mp->mnt_flag & MNT_UPDATE) {
+ /*
+ * Need to provide.
+ * 1. a way to convert between rdonly and rdwr mounts.
+ * 2. support for nfs exports.
+ */
+ error = EOPNOTSUPP;
+ goto bad;
+ }
/*
* Get argument
*/
if (error = copyin(data, (caddr_t)&args, sizeof(struct union_args)))
- return (error);
+ goto bad;
lowerrootvp = mp->mnt_vnodecovered;
VREF(lowerrootvp);
/*
- * Find upper node
+ * Find upper node.
*/
NDINIT(ndp, LOOKUP, FOLLOW|WANTPARENT,
UIO_USERSPACE, args.target, p);
- if (error = namei(ndp)) {
- vrele(lowerrootvp);
- return (error);
- }
+
+ if (error = namei(ndp))
+ goto bad;
+
upperrootvp = ndp->ni_vp;
vrele(ndp->ni_dvp);
ndp->ni_dvp = NULL;
if (upperrootvp->v_type != VDIR) {
- vrele(upperrootvp);
- return (EINVAL);
+ error = EINVAL;
+ goto bad;
}
um = (struct union_mount *) malloc(sizeof(struct union_mount),
/*
* Keep a held reference to the target vnodes.
* They are vrele'd in union_unmount.
+ *
+ * Depending on the _BELOW flag, the filesystems are
+ * viewed in a different order. In effect, this is the
+ * same as providing a mount under option to the mount syscall.
*/
- um->um_lowervp = lowerrootvp;
- um->um_uppervp = upperrootvp;
+
+ um->um_op = args.mntflags & UNMNT_OPMASK;
+ switch (um->um_op) {
+ case UNMNT_ABOVE:
+ um->um_lowervp = lowerrootvp;
+ um->um_uppervp = upperrootvp;
+ break;
+
+ case UNMNT_BELOW:
+ um->um_lowervp = upperrootvp;
+ um->um_uppervp = lowerrootvp;
+ break;
+
+ case UNMNT_REPLACE:
+ vrele(lowerrootvp);
+ lowerrootvp = NULLVP;
+ um->um_uppervp = upperrootvp;
+ um->um_lowervp = lowerrootvp;
+ break;
+
+ default:
+ error = EINVAL;
+ goto bad;
+ }
+
/*
- * Take a copy of the process's credentials. This isn't
- * quite right since the euid will always be zero and we
- * want to get the "real" users credentials. So fix up
- * the uid field after taking the copy.
+ * Unless the mount is readonly, ensure that the top layer
+ * supports whiteout operations
*/
- um->um_cred = crdup(p->p_ucred);
- um->um_cred->cr_uid = p->p_cred->p_ruid;
+ if ((mp->mnt_flag & MNT_RDONLY) == 0) {
+ error = VOP_WHITEOUT(um->um_uppervp, (struct componentname *) 0, LOOKUP);
+ if (error)
+ goto bad;
+ }
+
+ um->um_cred = p->p_ucred;
+ crhold(um->um_cred);
+ um->um_cmode = UN_DIRMODE &~ p->p_fd->fd_cmask;
+
+ /*
+ * Depending on what you think the MNT_LOCAL flag might mean,
+ * you may want the && to be || on the conditional below.
+ * At the moment it has been defined that the filesystem is
+ * only local if it is all local, ie the MNT_LOCAL flag implies
+ * that the entire namespace is local. If you think the MNT_LOCAL
+ * flag implies that some of the files might be stored locally
+ * then you will want to change the conditional.
+ */
+ if (um->um_op == UNMNT_ABOVE) {
+ if (((um->um_lowervp == NULLVP) ||
+ (um->um_lowervp->v_mount->mnt_flag & MNT_LOCAL)) &&
+ (um->um_uppervp->v_mount->mnt_flag & MNT_LOCAL))
+ mp->mnt_flag |= MNT_LOCAL;
+ }
+
+ /*
+ * Copy in the upper layer's RDONLY flag. This is for the benefit
+ * of lookup() which explicitly checks the flag, rather than asking
+ * the filesystem for it's own opinion. This means, that an update
+ * mount of the underlying filesystem to go from rdonly to rdwr
+ * will leave the unioned view as read-only.
+ */
+ mp->mnt_flag |= (um->um_uppervp->v_mount->mnt_flag & MNT_RDONLY);
- if ((lowerrootvp->v_mount->mnt_flag & MNT_LOCAL) ||
- (upperrootvp->v_mount->mnt_flag & MNT_LOCAL))
- mp->mnt_flag |= MNT_LOCAL;
mp->mnt_data = (qaddr_t) um;
getnewfsid(mp, MOUNT_UNION);
(void) copyinstr(path, mp->mnt_stat.f_mntonname, MNAMELEN - 1, &size);
bzero(mp->mnt_stat.f_mntonname + size, MNAMELEN - size);
- (void) copyinstr(args.target, mp->mnt_stat.f_mntfromname, MNAMELEN - 1,
- &size);
- bzero(mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
+
+ switch (um->um_op) {
+ case UNMNT_ABOVE:
+ cp = "<above>:";
+ break;
+ case UNMNT_BELOW:
+ cp = "<below>:";
+ break;
+ case UNMNT_REPLACE:
+ cp = "";
+ break;
+ }
+ len = strlen(cp);
+ bcopy(cp, mp->mnt_stat.f_mntfromname, len);
+
+ cp = mp->mnt_stat.f_mntfromname + len;
+ len = MNAMELEN - len;
+
+ (void) copyinstr(args.target, cp, len - 1, &size);
+ bzero(cp + size, len - size);
+
#ifdef UNION_DIAGNOSTIC
- printf("union_mount: upper %s, lower at %s\n",
+ printf("union_mount: from %s, on %s\n",
mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname);
#endif
return (0);
+
+bad:
+ if (um)
+ free(um, M_UFSMNT);
+ if (cred)
+ crfree(cred);
+ if (upperrootvp)
+ vrele(upperrootvp);
+ if (lowerrootvp)
+ vrele(lowerrootvp);
+ return (error);
}
/*
struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
struct vnode *um_rootvp;
int error;
+ int freeing;
int flags = 0;
extern int doforce;
if (error = union_root(mp, &um_rootvp))
return (error);
- if (um_rootvp->v_usecount > 1)
+
+ /*
+ * Keep flushing vnodes from the mount list.
+ * This is needed because of the un_pvp held
+ * reference to the parent vnode.
+ * If more vnodes have been freed on a given pass,
+ * the try again. The loop will iterate at most
+ * (d) times, where (d) is the maximum tree depth
+ * in the filesystem.
+ */
+ for (freeing = 0; vflush(mp, um_rootvp, flags) != 0;) {
+ struct vnode *vp;
+ int n;
+
+ /* count #vnodes held on mount list */
+ for (n = 0, vp = mp->mnt_vnodelist.lh_first;
+ vp != NULLVP;
+ vp = vp->v_mntvnodes.le_next)
+ n++;
+
+ /* if this is unchanged then stop */
+ if (n == freeing)
+ break;
+
+ /* otherwise try once more time */
+ freeing = n;
+ }
+
+ /* At this point the root vnode should have a single reference */
+ if (um_rootvp->v_usecount > 1) {
+ vput(um_rootvp);
return (EBUSY);
- if (error = vflush(mp, um_rootvp, flags))
- return (error);
+ }
#ifdef UNION_DIAGNOSTIC
- vprint("alias root of lower", um_rootvp);
+ vprint("union root", um_rootvp);
#endif
/*
* Discard references to upper and lower target vnodes.
*/
- vrele(um->um_lowervp);
+ if (um->um_lowervp)
+ vrele(um->um_lowervp);
vrele(um->um_uppervp);
crfree(um->um_cred);
/*
* Release reference on underlying root vnode
*/
- vrele(um_rootvp);
+ vput(um_rootvp);
/*
* And blow it away for future re-use
*/
*/
free(mp->mnt_data, M_UFSMNT); /* XXX */
mp->mnt_data = 0;
- return 0;
+ return (0);
}
int
{
struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
int error;
-
-#ifdef UNION_DIAGNOSTIC
- printf("union_root(mp = %x, lvp = %x, uvp = %x)\n", mp,
- um->um_lowervp,
- um->um_uppervp);
-#endif
+ int loselock;
/*
* Return locked reference to root.
*/
- error = union_allocvp(vpp, mp, (struct vnode *) 0,
+ VREF(um->um_uppervp);
+ if ((um->um_op == UNMNT_BELOW) &&
+ VOP_ISLOCKED(um->um_uppervp)) {
+ loselock = 1;
+ } else {
+ VOP_LOCK(um->um_uppervp);
+ loselock = 0;
+ }
+ if (um->um_lowervp)
+ VREF(um->um_lowervp);
+ error = union_allocvp(vpp, mp,
+ (struct vnode *) 0,
+ (struct vnode *) 0,
(struct componentname *) 0,
um->um_uppervp,
- um->um_lowervp);
- if (error == 0)
- (*vpp)->v_flag |= VROOT;
+ um->um_lowervp,
+ 1);
+
+ if (error) {
+ if (!loselock)
+ VOP_UNLOCK(um->um_uppervp);
+ vrele(um->um_uppervp);
+ if (um->um_lowervp)
+ vrele(um->um_lowervp);
+ } else {
+ if (loselock)
+ VTOUNION(*vpp)->un_flags &= ~UN_ULOCK;
+ }
return (error);
}
bzero(&mstat, sizeof(mstat));
- error = VFS_STATFS(um->um_lowervp->v_mount, &mstat, p);
- if (error)
- return (error);
+ if (um->um_lowervp) {
+ error = VFS_STATFS(um->um_lowervp->v_mount, &mstat, p);
+ if (error)
+ return (error);
+ }
/* now copy across the "interesting" information and fake the rest */
#if 0
if (error)
return (error);
- sbp->f_type = mstat.f_type;
+ sbp->f_type = MOUNT_UNION;
sbp->f_flags = mstat.f_flags;
sbp->f_bsize = mstat.f_bsize;
sbp->f_iosize = mstat.f_iosize;
* kind of sense. none of this makes sense though.
*/
- if (mstat.f_bsize != lbsize) {
+ if (mstat.f_bsize != lbsize)
sbp->f_blocks = sbp->f_blocks * lbsize / mstat.f_bsize;
- sbp->f_bfree = sbp->f_bfree * lbsize / mstat.f_bsize;
- sbp->f_bavail = sbp->f_bavail * lbsize / mstat.f_bsize;
- }
+
+ /*
+ * The "total" fields count total resources in all layers,
+ * the "free" fields count only those resources which are
+ * free in the upper layer (since only the upper layer
+ * is writeable).
+ */
sbp->f_blocks += mstat.f_blocks;
- sbp->f_bfree += mstat.f_bfree;
- sbp->f_bavail += mstat.f_bavail;
+ sbp->f_bfree = mstat.f_bfree;
+ sbp->f_bavail = mstat.f_bavail;
sbp->f_files += mstat.f_files;
- sbp->f_ffree += mstat.f_ffree;
+ sbp->f_ffree = mstat.f_ffree;
if (sbp != &mp->mnt_stat) {
bcopy(&mp->mnt_stat.f_fsid, &sbp->f_fsid, sizeof(sbp->f_fsid));