#include <sys/param.h>
#include <sys/systm.h>
#include <sys/proc_internal.h>
#include <sys/kauth.h>
#include <sys/time.h>
#include <sys/kernel.h>
#include <sys/vnode_internal.h>
#include <sys/namei.h>
#include <sys/malloc.h>
#include <sys/file.h>
#include <sys/filedesc.h>
#include <sys/queue.h>
#include <sys/mount_internal.h>
#include <sys/stat.h>
#include <sys/ubc.h>
#include <sys/uio_internal.h>
#include <miscfs/union/union.h>
#if DIAGNOSTIC
#include <sys/proc.h>
#endif
#define NHASH 32
#define UNION_HASH(u, l) \
(((((unsigned long) (u)) + ((unsigned long) l)) >> 8) & (NHASH-1))
static LIST_HEAD(unhead, union_node) unhead[NHASH];
static int unvplock[NHASH];
int
union_init()
{
int i;
for (i = 0; i < NHASH; i++)
LIST_INIT(&unhead[i]);
bzero((caddr_t) unvplock, sizeof(unvplock));
}
static int
union_list_lock(ix)
int ix;
{
if (unvplock[ix] & UN_LOCKED) {
unvplock[ix] |= UN_WANT;
sleep((caddr_t) &unvplock[ix], PINOD);
return (1);
}
unvplock[ix] |= UN_LOCKED;
return (0);
}
static void
union_list_unlock(ix)
int ix;
{
unvplock[ix] &= ~UN_LOCKED;
if (unvplock[ix] & UN_WANT) {
unvplock[ix] &= ~UN_WANT;
wakeup((caddr_t) &unvplock[ix]);
}
}
void
union_updatevp(un, uppervp, lowervp)
struct union_node *un;
struct vnode *uppervp;
struct vnode *lowervp;
{
int ohash = UNION_HASH(un->un_uppervp, un->un_lowervp);
int nhash = UNION_HASH(uppervp, lowervp);
int docache = (lowervp != NULLVP || uppervp != NULLVP);
int lhash, uhash;
if (nhash < ohash) {
lhash = nhash;
uhash = ohash;
} else {
lhash = ohash;
uhash = nhash;
}
if (lhash != uhash)
while (union_list_lock(lhash))
continue;
while (union_list_lock(uhash))
continue;
if (ohash != nhash || !docache) {
if (un->un_flags & UN_CACHED) {
un->un_flags &= ~UN_CACHED;
LIST_REMOVE(un, un_cache);
}
}
if (ohash != nhash)
union_list_unlock(ohash);
if (un->un_lowervp != lowervp) {
if (un->un_lowervp) {
vnode_put(un->un_lowervp);
if (un->un_path) {
_FREE(un->un_path, M_TEMP);
un->un_path = 0;
}
if (un->un_dirvp) {
vnode_put(un->un_dirvp);
un->un_dirvp = NULLVP;
}
}
un->un_lowervp = lowervp;
un->un_lowersz = VNOVAL;
}
if (un->un_uppervp != uppervp) {
if (un->un_uppervp)
vnode_put(un->un_uppervp);
un->un_uppervp = uppervp;
un->un_uppersz = VNOVAL;
}
if (docache && (ohash != nhash)) {
LIST_INSERT_HEAD(&unhead[nhash], un, un_cache);
un->un_flags |= UN_CACHED;
}
union_list_unlock(nhash);
}
void
union_newlower(un, lowervp)
struct union_node *un;
struct vnode *lowervp;
{
union_updatevp(un, un->un_uppervp, lowervp);
}
void
union_newupper(un, uppervp)
struct union_node *un;
struct vnode *uppervp;
{
union_updatevp(un, uppervp, un->un_lowervp);
}
void
union_newsize(vp, uppersz, lowersz)
struct vnode *vp;
off_t uppersz, lowersz;
{
struct union_node *un;
off_t sz;
if (vp->v_type != VREG)
return;
un = VTOUNION(vp);
sz = VNOVAL;
if ((uppersz != VNOVAL) && (un->un_uppersz != uppersz)) {
un->un_uppersz = uppersz;
if (sz == VNOVAL)
sz = un->un_uppersz;
}
if ((lowersz != VNOVAL) && (un->un_lowersz != lowersz)) {
un->un_lowersz = lowersz;
if (sz == VNOVAL)
sz = un->un_lowersz;
}
if (sz != VNOVAL) {
#ifdef UNION_DIAGNOSTIC
printf("union: %s size now %ld\n",
uppersz != VNOVAL ? "upper" : "lower", (long) sz);
#endif
ubc_setsize(vp, sz);
}
}
int
union_allocvp(vpp, mp, undvp, dvp, cnp, uppervp, lowervp, docache)
struct vnode **vpp;
struct mount *mp;
struct vnode *undvp;
struct vnode *dvp;
struct componentname *cnp;
struct vnode *uppervp;
struct vnode *lowervp;
int docache;
{
int error;
struct union_node *un;
struct union_node **pp;
struct vnode *xlowervp = NULLVP;
struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
int hash;
int markroot;
int try;
struct union_node *unp;
struct vnode_fsparam vfsp;
enum vtype vtype;
if (uppervp == NULLVP && lowervp == NULLVP)
panic("union: unidentifiable allocation");
if (uppervp && lowervp && (uppervp->v_type != lowervp->v_type)) {
xlowervp = lowervp;
lowervp = NULLVP;
}
markroot = 0;
if ((uppervp == um->um_uppervp) &&
((lowervp == NULLVP) || lowervp == um->um_lowervp)) {
if (lowervp == NULLVP) {
lowervp = um->um_lowervp;
if (lowervp != NULLVP)
vnode_get(lowervp);
}
markroot = VROOT;
}
loop:
if (!docache) {
un = 0;
} else for (try = 0; try < 3; try++) {
switch (try) {
case 0:
if (lowervp == NULLVP)
continue;
hash = UNION_HASH(uppervp, lowervp);
break;
case 1:
if (uppervp == NULLVP)
continue;
hash = UNION_HASH(uppervp, NULLVP);
break;
case 2:
if (lowervp == NULLVP)
continue;
hash = UNION_HASH(NULLVP, lowervp);
break;
}
while (union_list_lock(hash))
continue;
for (un = unhead[hash].lh_first; un != 0;
un = un->un_cache.le_next) {
if ((un->un_lowervp == lowervp ||
un->un_lowervp == NULLVP) &&
(un->un_uppervp == uppervp ||
un->un_uppervp == NULLVP) &&
(UNIONTOV(un)->v_mount == mp)) {
if (vnode_get(UNIONTOV(un))) {
union_list_unlock(hash);
goto loop;
}
break;
}
}
union_list_unlock(hash);
if (un)
break;
}
if (un) {
if ((dvp != NULLVP) && (uppervp == dvp)) {
#if DIAGNOSTIC
if ((un->un_flags & UN_LOCKED) == 0)
panic("union: . not locked");
else if (current_proc() && un->un_pid != current_proc()->p_pid &&
un->un_pid > -1 && current_proc()->p_pid > -1)
panic("union: allocvp not lock owner");
#endif
} else {
if (un->un_flags & UN_LOCKED) {
vnode_put(UNIONTOV(un));
un->un_flags |= UN_WANT;
sleep((caddr_t) &un->un_flags, PINOD);
goto loop;
}
un->un_flags |= UN_LOCKED;
#if DIAGNOSTIC
if (current_proc())
un->un_pid = current_proc()->p_pid;
else
un->un_pid = -1;
#endif
}
if (uppervp != un->un_uppervp) {
union_newupper(un, uppervp);
} else if (uppervp) {
vnode_put(uppervp);
}
if (un->un_uppervp) {
un->un_flags |= UN_ULOCK;
un->un_flags &= ~UN_KLOCK;
}
if (lowervp != un->un_lowervp) {
union_newlower(un, lowervp);
if (cnp && (lowervp != NULLVP)) {
un->un_hash = cnp->cn_hash;
MALLOC(un->un_path, caddr_t, cnp->cn_namelen+1,
M_TEMP, M_WAITOK);
bcopy(cnp->cn_nameptr, un->un_path,
cnp->cn_namelen);
un->un_path[cnp->cn_namelen] = '\0';
vnode_get(dvp);
un->un_dirvp = dvp;
}
} else if (lowervp) {
vnode_put(lowervp);
}
*vpp = UNIONTOV(un);
return (0);
}
if (docache) {
hash = UNION_HASH(uppervp, lowervp);
if (union_list_lock(hash))
goto loop;
}
MALLOC(unp, void *, sizeof(struct union_node), M_TEMP, M_WAITOK);
if (uppervp)
vtype = uppervp->v_type;
else
vtype = lowervp->v_type;
vfsp.vnfs_mp = mp;
vfsp.vnfs_vtype = vtype;
vfsp.vnfs_str = "unionfs";
vfsp.vnfs_dvp = dvp;
vfsp.vnfs_fsnode = unp;
vfsp.vnfs_cnp = cnp;
vfsp.vnfs_vops = union_vnodeop_p;
vfsp.vnfs_rdev = 0;
vfsp.vnfs_filesize = 0;
vfsp.vnfs_flags = VNFS_NOCACHE | VNFS_CANTCACHE;
vfsp.vnfs_marksystem = 0;
vfsp.vnfs_markroot = markroot;
error = vnode_create(VNCREATE_FLAVOR, VCREATESIZE, &vfsp, vpp);
if (error) {
FREE(unp, M_TEMP);
if (uppervp) {
vnode_put(uppervp);
}
if (lowervp)
vnode_put(lowervp);
goto out;
}
(*vpp)->v_tag = VT_UNION;
un = VTOUNION(*vpp);
un->un_vnode = *vpp;
un->un_uppervp = uppervp;
un->un_uppersz = VNOVAL;
un->un_lowervp = lowervp;
un->un_lowersz = VNOVAL;
un->un_pvp = undvp;
if (undvp != NULLVP)
vnode_get(undvp);
un->un_dircache = 0;
un->un_openl = 0;
un->un_flags = UN_LOCKED;
if (un->un_uppervp)
un->un_flags |= UN_ULOCK;
#if DIAGNOSTIC
if (current_proc())
un->un_pid = current_proc()->p_pid;
else
un->un_pid = -1;
#endif
if (cnp && (lowervp != NULLVP)) {
un->un_hash = cnp->cn_hash;
un->un_path = _MALLOC(cnp->cn_namelen+1, M_TEMP, M_WAITOK);
bcopy(cnp->cn_nameptr, un->un_path, cnp->cn_namelen);
un->un_path[cnp->cn_namelen] = '\0';
vnode_get(dvp);
un->un_dirvp = dvp;
} else {
un->un_hash = 0;
un->un_path = 0;
un->un_dirvp = 0;
}
if (docache) {
LIST_INSERT_HEAD(&unhead[hash], un, un_cache);
un->un_flags |= UN_CACHED;
}
if (xlowervp)
vnode_put(xlowervp);
out:
if (docache)
union_list_unlock(hash);
return (error);
}
int
union_freevp(vp)
struct vnode *vp;
{
struct union_node *un = VTOUNION(vp);
if (un->un_flags & UN_CACHED) {
un->un_flags &= ~UN_CACHED;
LIST_REMOVE(un, un_cache);
}
if (un->un_pvp != NULLVP)
vnode_put(un->un_pvp);
if (un->un_uppervp != NULLVP)
vnode_put(un->un_uppervp);
if (un->un_lowervp != NULLVP)
vnode_put(un->un_lowervp);
if (un->un_dirvp != NULLVP)
vnode_put(un->un_dirvp);
if (un->un_path)
_FREE(un->un_path, M_TEMP);
FREE(vp->v_data, M_TEMP);
vp->v_data = 0;
return (0);
}
int
union_copyfile(struct vnode *fvp, struct vnode *tvp, kauth_cred_t cred,
struct proc *p)
{
char *bufp;
struct uio uio;
struct iovec_32 iov;
struct vfs_context context;
int error = 0;
context.vc_proc = p;
context.vc_ucred = cred;
#if 1
uio.uio_segflg = UIO_SYSSPACE;
#else
uio.uio_segflg = UIO_SYSSPACE32;
#endif
uio.uio_offset = 0;
bufp = _MALLOC(MAXPHYSIO, M_TEMP, M_WAITOK);
do {
off_t offset = uio.uio_offset;
uio.uio_iovs.iov32p = &iov;
uio.uio_iovcnt = 1;
iov.iov_base = (uintptr_t)bufp;
iov.iov_len = MAXPHYSIO;
uio_setresid(&uio, iov.iov_len);
uio.uio_rw = UIO_READ;
error = VNOP_READ(fvp, &uio, 0, &context);
if (error == 0) {
uio.uio_iovs.iov32p = &iov;
uio.uio_iovcnt = 1;
iov.iov_base = (uintptr_t)bufp;
iov.iov_len = MAXPHYSIO - uio_resid(&uio);
uio.uio_offset = offset;
uio.uio_rw = UIO_WRITE;
uio_setresid(&uio, iov.iov_len);
if (uio_resid(&uio) == 0)
break;
do {
error = VNOP_WRITE(tvp, &uio, 0, &context);
} while ((uio_resid(&uio) > 0) && (error == 0));
}
} while (error == 0);
_FREE(bufp, M_TEMP);
return (error);
}
int
union_copyup(struct union_node *un, int docopy, kauth_cred_t cred,
struct proc *p)
{
int error;
struct vnode *lvp, *uvp;
struct vfs_context context;
error = union_vn_create(&uvp, un, p);
if (error)
return (error);
context.vc_proc = p;
context.vc_ucred = cred;
union_newupper(un, uvp);
un->un_flags |= UN_ULOCK;
lvp = un->un_lowervp;
if (docopy) {
error = VNOP_OPEN(lvp, FREAD, &context);
if (error == 0) {
error = union_copyfile(lvp, uvp, cred, p);
(void) VNOP_CLOSE(lvp, FREAD, &context);
}
#ifdef UNION_DIAGNOSTIC
if (error == 0)
uprintf("union: copied up %s\n", un->un_path);
#endif
}
un->un_flags &= ~UN_ULOCK;
union_vn_close(uvp, FWRITE, cred, p);
un->un_flags |= UN_ULOCK;
if (error == 0) {
int i;
for (i = 0; i < un->un_openl; i++) {
(void) VNOP_CLOSE(lvp, FREAD, &context);
(void) VNOP_OPEN(uvp, FREAD, &context);
}
un->un_openl = 0;
}
return (error);
}
static int
union_relookup(um, dvp, vpp, cnp, cn, path, pathlen)
struct union_mount *um;
struct vnode *dvp;
struct vnode **vpp;
struct componentname *cnp;
struct componentname *cn;
char *path;
int pathlen;
{
int error;
cn->cn_namelen = pathlen;
cn->cn_pnbuf = _MALLOC_ZONE(cn->cn_namelen+1, M_NAMEI, M_WAITOK);
cn->cn_pnlen = cn->cn_namelen+1;
bcopy(path, cn->cn_pnbuf, cn->cn_namelen);
cn->cn_pnbuf[cn->cn_namelen] = '\0';
cn->cn_nameiop = CREATE;
cn->cn_flags = (LOCKPARENT|HASBUF|SAVENAME|SAVESTART|ISLASTCN);
#ifdef XXX_HELP_ME
cn->cn_proc = cnp->cn_proc;
if (um->um_op == UNMNT_ABOVE)
cn->cn_cred = cnp->cn_cred;
else
cn->cn_cred = um->um_cred;
#endif
cn->cn_context = cnp->cn_context;
cn->cn_nameptr = cn->cn_pnbuf;
cn->cn_hash = cnp->cn_hash;
cn->cn_consume = cnp->cn_consume;
vnode_get(dvp);
error = relookup(dvp, vpp, cn);
if (!error)
vnode_put(dvp);
return (error);
}
int
union_mkshadow(um, dvp, cnp, vpp)
struct union_mount *um;
struct vnode *dvp;
struct componentname *cnp;
struct vnode **vpp;
{
int error;
struct vnode_attr va;
struct componentname cn;
error = union_relookup(um, dvp, vpp, cnp, &cn,
cnp->cn_nameptr, cnp->cn_namelen);
if (error)
return (error);
if (*vpp) {
vnode_put(*vpp);
*vpp = NULLVP;
return (EEXIST);
}
VATTR_INIT(&va);
VATTR_SET(&va, va_type, VDIR);
VATTR_SET(&va, va_mode, um->um_cmode);
error = vn_create(dvp, vpp, &cn, &va, 0, cnp->cn_context);
return (error);
}
int
union_mkwhiteout(um, dvp, cnp, path)
struct union_mount *um;
struct vnode *dvp;
struct componentname *cnp;
char *path;
{
int error;
struct vnode *wvp;
struct componentname cn;
error = union_relookup(um, dvp, &wvp, cnp, &cn, path, strlen(path));
if (error) {
return (error);
}
if (wvp) {
vnode_put(dvp);
vnode_put(wvp);
return (EEXIST);
}
error = VNOP_WHITEOUT(dvp, &cn, CREATE, cnp->cn_context);
vnode_put(dvp);
return (error);
}
int
union_vn_create(vpp, un, p)
struct vnode **vpp;
struct union_node *un;
struct proc *p;
{
struct vnode *vp;
kauth_cred_t cred = p->p_ucred;
struct vnode_attr vat;
struct vnode_attr *vap = &vat;
struct vfs_context context;
int fmode = FFLAGS(O_WRONLY|O_CREAT|O_TRUNC|O_EXCL);
int error;
int cmode = UN_FILEMODE & ~p->p_fd->fd_cmask;
char *cp;
struct componentname cn;
*vpp = NULLVP;
context.vc_proc = p;
context.vc_ucred = p->p_ucred;
cn.cn_namelen = strlen(un->un_path);
cn.cn_pnbuf = (caddr_t) _MALLOC_ZONE(cn.cn_namelen+1,
M_NAMEI, M_WAITOK);
cn.cn_pnlen = cn.cn_namelen+1;
bcopy(un->un_path, cn.cn_pnbuf, cn.cn_namelen+1);
cn.cn_nameiop = CREATE;
cn.cn_flags = (LOCKPARENT|HASBUF|SAVENAME|SAVESTART|ISLASTCN);
cn.cn_context = &context;
cn.cn_nameptr = cn.cn_pnbuf;
cn.cn_hash = un->un_hash;
cn.cn_consume = 0;
vnode_get(un->un_dirvp);
if (error = relookup(un->un_dirvp, &vp, &cn))
return (error);
vnode_put(un->un_dirvp);
if (vp) {
vnode_put(un->un_dirvp);
vnode_put(vp);
return (EEXIST);
}
VATTR_INIT(vap);
VATTR_SET(vap, va_type, VREG);
VATTR_SET(vap, va_mode, cmode);
if (error = vn_create(un->un_dirvp, &vp, &cn, vap, 0, &context))
return (error);
if (error = VNOP_OPEN(vp, fmode, &context)) {
vnode_put(vp);
return (error);
}
vnode_lock(vp);
if (++vp->v_writecount <= 0)
panic("union: v_writecount");
vnode_unlock(vp);
*vpp = vp;
return (0);
}
int
union_vn_close(struct vnode *vp, int fmode, kauth_cred_t cred,
struct proc *p)
{
struct vfs_context context;
context.vc_proc = p;
context.vc_ucred = cred;
if (fmode & FWRITE) {
vnode_lock(vp);
--vp->v_writecount;
vnode_unlock(vp);
}
return (VNOP_CLOSE(vp, fmode, &context));
}
void
union_removed_upper(un)
struct union_node *un;
{
struct proc *p = current_proc();
union_newupper(un, NULLVP);
if (un->un_flags & UN_CACHED) {
un->un_flags &= ~UN_CACHED;
LIST_REMOVE(un, un_cache);
}
if (un->un_flags & UN_ULOCK) {
un->un_flags &= ~UN_ULOCK;
}
}
#if 0
struct vnode *
union_lowervp(vp)
struct vnode *vp;
{
struct union_node *un = VTOUNION(vp);
if ((un->un_lowervp != NULLVP) &&
(vp->v_type == un->un_lowervp->v_type)) {
if (vnode_get(un->un_lowervp) == 0)
return (un->un_lowervp);
}
return (NULLVP);
}
#endif
int
union_dowhiteout(struct union_node *un, vfs_context_t ctx)
{
struct vnode_attr va;
if (un->un_lowervp != NULLVP)
return (1);
VATTR_INIT(&va);
VATTR_WANTED(&va, va_flags);
if (vnode_getattr(un->un_uppervp, &va, ctx) == 0 &&
(va.va_flags & OPAQUE))
return (1);
return (0);
}
static void
union_dircache_r(vp, vppp, cntp)
struct vnode *vp;
struct vnode ***vppp;
int *cntp;
{
struct union_node *un;
if (vp->v_op != union_vnodeop_p) {
if (vppp) {
vnode_get(vp);
*(*vppp)++ = vp;
if (--(*cntp) == 0)
panic("union: dircache table too small");
} else {
(*cntp)++;
}
return;
}
un = VTOUNION(vp);
if (un->un_uppervp != NULLVP)
union_dircache_r(un->un_uppervp, vppp, cntp);
if (un->un_lowervp != NULLVP)
union_dircache_r(un->un_lowervp, vppp, cntp);
}
struct vnode *
union_dircache(vp, p)
struct vnode *vp;
struct proc *p;
{
int count;
struct vnode *nvp;
struct vnode **vpp;
struct vnode **dircache;
struct union_node *un;
int error;
dircache = VTOUNION(vp)->un_dircache;
nvp = NULLVP;
if (dircache == 0) {
count = 0;
union_dircache_r(vp, 0, &count);
count++;
dircache = (struct vnode **)
_MALLOC(count * sizeof(struct vnode *),
M_TEMP, M_WAITOK);
vpp = dircache;
union_dircache_r(vp, &vpp, &count);
*vpp = NULLVP;
vpp = dircache + 1;
} else {
vpp = dircache;
do {
if (*vpp++ == VTOUNION(vp)->un_uppervp)
break;
} while (*vpp != NULLVP);
}
if (*vpp == NULLVP)
goto out;
vnode_get(*vpp);
error = union_allocvp(&nvp, vp->v_mount, NULLVP, NULLVP, 0, *vpp, NULLVP, 0);
if (error)
goto out;
VTOUNION(vp)->un_dircache = 0;
un = VTOUNION(nvp);
un->un_dircache = dircache;
out:
return (nvp);
}