static void
set_rootvnode(void)
{
struct proc *p;
if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode))
panic("Cannot find root vnode");
VOP_UNLOCK(rootvnode, 0);
p = curthread->td_proc;
FILEDESC_XLOCK(p->p_fd);
if (p->p_fd->fd_cdir != NULL)
vrele(p->p_fd->fd_cdir);
p->p_fd->fd_cdir = rootvnode;
VREF(rootvnode);
if (p->p_fd->fd_rdir != NULL)
vrele(p->p_fd->fd_rdir);
p->p_fd->fd_rdir = rootvnode;
VREF(rootvnode);
FILEDESC_XUNLOCK(p->p_fd);
}
static __inline
struct vnode *
union_lock_upper(struct union_node *un, struct thread *td)
{
struct vnode *uppervp;
if ((uppervp = un->un_uppervp) != NULL) {
VREF(uppervp);
vn_lock(uppervp, LK_EXCLUSIVE | LK_CANRECURSE | LK_RETRY, td);
}
KASSERT((uppervp == NULL || vrefcnt(uppervp) > 0), ("uppervp usecount is 0"));
return(uppervp);
}
/*
* lookup. this is incredibly complicated in the
* general case, however for most pseudo-filesystems
* very little needs to be done.
*
* Locking isn't hard here, just poorly documented.
*
* If we're looking up ".", just vref the parent & return it.
*
* If we're looking up "..", unlock the parent, and lock "..". If everything
* went ok, try to re-lock the parent. We do this to prevent lock races.
*
* For anything else, get the needed node.
*
* We try to exit with the parent locked in error cases.
*/
int
ptyfs_lookup(void *v)
{
struct vop_lookup_args /* {
struct vnode * a_dvp;
struct vnode ** a_vpp;
struct componentname * a_cnp;
} */ *ap = v;
struct componentname *cnp = ap->a_cnp;
struct vnode **vpp = ap->a_vpp;
struct vnode *dvp = ap->a_dvp;
const char *pname = cnp->cn_nameptr;
struct ptyfsnode *ptyfs;
int pty, error;
*vpp = NULL;
if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)
return EROFS;
if (cnp->cn_namelen == 1 && *pname == '.') {
*vpp = dvp;
VREF(dvp);
return 0;
}
ptyfs = VTOPTYFS(dvp);
switch (ptyfs->ptyfs_type) {
case PTYFSroot:
/*
* Shouldn't get here with .. in the root node.
*/
if (cnp->cn_flags & ISDOTDOT)
return EIO;
pty = atoi(pname, cnp->cn_namelen);
if (pty < 0 || pty >= npty || pty_isfree(pty, 1))
break;
error = ptyfs_allocvp(dvp->v_mount, vpp, PTYFSpts, pty,
curlwp);
return error;
default:
return ENOTDIR;
}
return cnp->cn_nameiop == LOOKUP ? ENOENT : EROFS;
}
static __inline
struct vnode *
union_lock_other(struct union_node *un, struct thread *td)
{
struct vnode *vp;
if (un->un_uppervp != NULL) {
vp = union_lock_upper(un, td);
} else if ((vp = un->un_lowervp) != NULL) {
VREF(vp);
vn_lock(vp, LK_EXCLUSIVE | LK_CANRECURSE | LK_RETRY, td);
}
return(vp);
}
VMBlockVFSRoot(struct mount *mp, // IN: vmblock file system
int flags, // IN: lockmgr(9) flags
struct vnode **vpp, // OUT: root vnode
struct thread *td) // IN: caller's thread context
#endif
{
struct vnode *vp;
/*
* Return locked reference to root.
*/
vp = MNTTOVMBLOCKMNT(mp)->rootVnode;
VREF(vp);
compat_vn_lock(vp, flags | LK_RETRY, compat_td);
*vpp = vp;
return 0;
}
/*
* Change the trace vnode in a correct way (to avoid races).
*/
void
ktrsettracevnode(struct proc *p, struct vnode *newvp)
{
struct vnode *vp;
if (p->p_tracep == newvp) /* avoid work */
return;
if (newvp != NULL)
VREF(newvp);
vp = p->p_tracep;
p->p_tracep = newvp;
if (vp != NULL)
vrele(vp);
}
/*
* Increasing refcount of lower vnode is needed at least for the case
* when lower FS is NFS to do sillyrename if the file is in use.
* Unfortunately v_usecount is incremented in many places in
* the kernel and, as such, there may be races that result in
* the NFS client doing an extraneous silly rename, but that seems
* preferable to not doing a silly rename when it is needed.
*/
static int
null_remove(struct vop_remove_args *ap)
{
int retval, vreleit;
struct vnode *lvp;
if (vrefcnt(ap->a_vp) > 1) {
lvp = NULLVPTOLOWERVP(ap->a_vp);
VREF(lvp);
vreleit = 1;
} else
vreleit = 0;
retval = null_bypass(&ap->a_gen);
if (vreleit != 0)
vrele(lvp);
return (retval);
}
static int vboxvfs_root(struct mount *mp, int flags, struct vnode **vpp, struct thread *td)
{
int rc = 0;
struct sf_glob_info *pShFlGlobalInfo = VFSMP2SFGLOBINFO(mp);
struct vnode *vp;
printf("%s: Enter\n", __FUNCTION__);
vp = pShFlGlobalInfo->vnode_root;
VREF(vp);
vn_lock(vp, flags | LK_RETRY, td);
*vpp = vp;
printf("%s: Leave\n", __FUNCTION__);
return rc;
}
/*
* Increasing refcount of lower vnode is needed at least for the case
* when lower FS is NFS to do sillyrename if the file is in use.
* Unfortunately v_usecount is incremented in many places in
* the kernel and, as such, there may be races that result in
* the NFS client doing an extraneous silly rename, but that seems
* preferable to not doing a silly rename when it is needed.
*/
static int
crypto_remove(struct vop_remove_args *ap)
{
int retval, vreleit;
struct vnode *lvp, *vp;
vp = ap->a_vp;
if (vrefcnt(vp) > 1) {
lvp = NULLVPTOLOWERVP(vp);
VREF(lvp);
vreleit = 1;
} else
vreleit = 0;
VTONULL(vp)->crypto_flags |= NULLV_DROP;
retval = crypto_bypass(&ap->a_gen);
if (vreleit != 0)
vrele(lvp);
return (retval);
}
static int
xfs_dnlc_lock(struct vnode *dvp,
xfs_componentname *cnp,
struct vnode **res)
{
int error = 0;
/*
* Try to handle the (complex) BSD locking protocol.
*/
if (*res == dvp) { /* "." */
VREF(dvp);
} else if (cnp->cn_flags & ISDOTDOT) { /* ".." */
u_long vpid = dvp->v_id;
#ifdef HAVE_FREEBSD_THREAD
xfs_vfs_unlock(dvp, xfs_cnp_to_thread(cnp));
error = xfs_do_vget(*res, LK_EXCLUSIVE, xfs_cnp_to_thread(cnp));
xfs_vfs_writelock(dvp, xfs_cnp_to_thread(cnp));
#else
xfs_vfs_unlock(dvp, xfs_cnp_to_proc(cnp));
error = xfs_do_vget(*res, LK_EXCLUSIVE, xfs_cnp_to_proc(cnp));
xfs_vfs_writelock(dvp, xfs_cnp_to_proc(cnp));
#endif
if (error == 0 && dvp->v_id != vpid) {
vput(*res);
return 0;
}
} else {
#ifdef HAVE_FREEBSD_THREAD
error = xfs_do_vget(*res, LK_EXCLUSIVE, xfs_cnp_to_thread(cnp));
#else
error = xfs_do_vget(*res, LK_EXCLUSIVE, xfs_cnp_to_proc(cnp));
#endif
}
if (error == 0)
return -1;
else
return 0;
}
/*
* File table vnode ioctl routine.
*/
int
vn_ioctl(struct file *fp, u_long com, caddr_t data, struct proc *p)
{
struct vnode *vp = ((struct vnode *)fp->f_data);
struct vattr vattr;
int error;
switch (vp->v_type) {
case VREG:
case VDIR:
if (com == FIONREAD) {
error = VOP_GETATTR(vp, &vattr, p->p_ucred, p);
if (error)
return (error);
*(int *)data = vattr.va_size - fp->f_offset;
return (0);
}
if (com == FIBMAP)
return VOP_IOCTL(vp, com, data, fp->f_flag,
p->p_ucred, p);
if (com == FIONBIO || com == FIOASYNC) /* XXX */
return (0); /* XXX */
/* FALLTHROUGH */
default:
return (ENOTTY);
case VFIFO:
case VCHR:
case VBLK:
error = VOP_IOCTL(vp, com, data, fp->f_flag, p->p_ucred, p);
if (error == 0 && com == TIOCSCTTY) {
if (p->p_session->s_ttyvp)
vrele(p->p_session->s_ttyvp);
p->p_session->s_ttyvp = vp;
VREF(vp);
}
return (error);
}
}
/*
* We have to carry on the locking protocol on the null layer vnodes
* as we progress through the tree. We also have to enforce read-only
* if this layer is mounted read-only.
*/
static int
null_lookup(struct vop_lookup_args *ap)
{
struct componentname *cnp = ap->a_cnp;
struct vnode *dvp = ap->a_dvp;
int flags = cnp->cn_flags;
struct vnode *vp, *ldvp, *lvp;
int error;
if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
(cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
return (EROFS);
/*
* Although it is possible to call null_bypass(), we'll do
* a direct call to reduce overhead
*/
ldvp = NULLVPTOLOWERVP(dvp);
vp = lvp = NULL;
error = VOP_LOOKUP(ldvp, &lvp, cnp);
if (error == EJUSTRETURN && (flags & ISLASTCN) &&
(dvp->v_mount->mnt_flag & MNT_RDONLY) &&
(cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME))
error = EROFS;
if ((error == 0 || error == EJUSTRETURN) && lvp != NULL) {
if (ldvp == lvp) {
*ap->a_vpp = dvp;
VREF(dvp);
vrele(lvp);
} else {
error = null_nodeget(dvp->v_mount, lvp, &vp);
if (error)
vput(lvp);
else
*ap->a_vpp = vp;
}
}
return (error);
}
int
reiserfs_lookup(struct vop_cachedlookup_args *ap)
{
int error, retval;
struct vnode *vdp = ap->a_dvp;
struct vnode **vpp = ap->a_vpp;
struct componentname *cnp = ap->a_cnp;
int flags = cnp->cn_flags;
struct thread *td = cnp->cn_thread;
struct cpu_key *saved_ino;
struct vnode *vp;
struct vnode *pdp; /* Saved dp during symlink work */
struct reiserfs_node *dp;
struct reiserfs_dir_entry de;
INITIALIZE_PATH(path_to_entry);
char c = cnp->cn_nameptr[cnp->cn_namelen];
cnp->cn_nameptr[cnp->cn_namelen] = '\0';
reiserfs_log(LOG_DEBUG, "looking for `%s', %ld (%s)\n",
cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_pnbuf);
cnp->cn_nameptr[cnp->cn_namelen] = c;
vp = NULL;
dp = VTOI(vdp);
if (REISERFS_MAX_NAME(dp->i_reiserfs->s_blocksize) < cnp->cn_namelen)
return (ENAMETOOLONG);
reiserfs_log(LOG_DEBUG, "searching entry\n");
de.de_gen_number_bit_string = 0;
retval = reiserfs_find_entry(dp, cnp->cn_nameptr, cnp->cn_namelen,
&path_to_entry, &de);
pathrelse(&path_to_entry);
if (retval == NAME_FOUND) {
reiserfs_log(LOG_DEBUG, "found\n");
} else {
reiserfs_log(LOG_DEBUG, "not found\n");
}
if (retval == NAME_FOUND) {
#if 0
/* Hide the .reiserfs_priv directory */
if (reiserfs_xattrs(dp->i_reiserfs) &&
!old_format_only(dp->i_reiserfs) &&
REISERFS_SB(dp->i_reiserfs)->priv_root &&
REISERFS_SB(dp->i_reiserfs)->priv_root->d_inode &&
de.de_objectid == le32toh(INODE_PKEY(REISERFS_SB(
dp->i_reiserfs)->priv_root->d_inode)->k_objectid)) {
return (EACCES);
}
#endif
reiserfs_log(LOG_DEBUG, "reading vnode\n");
pdp = vdp;
if (flags & ISDOTDOT) {
saved_ino = (struct cpu_key *)&(de.de_dir_id);
VOP_UNLOCK(pdp, 0);
error = reiserfs_iget(vdp->v_mount,
saved_ino, &vp, td);
vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY);
if (error != 0)
return (error);
*vpp = vp;
} else if (de.de_objectid == dp->i_number &&
de.de_dir_id == dp->i_ino) {
VREF(vdp); /* We want ourself, ie "." */
*vpp = vdp;
} else {
if ((error = reiserfs_iget(vdp->v_mount,
(struct cpu_key *)&(de.de_dir_id), &vp, td)) != 0)
return (error);
*vpp = vp;
}
/*
* Propogate the priv_object flag so we know we're in the
* priv tree
*/
/*if (is_reiserfs_priv_object(dir))
REISERFS_I(inode)->i_flags |= i_priv_object;*/
} else {
if (retval == IO_ERROR) {
reiserfs_log(LOG_DEBUG, "IO error\n");
return (EIO);
}
return (ENOENT);
}
/* Insert name into cache if appropriate. */
if (cnp->cn_flags & MAKEENTRY)
cache_enter(vdp, *vpp, cnp);
reiserfs_log(LOG_DEBUG, "done\n");
return (0);
}
/* Note that we don't set CDirty here, this is OK because the unlink
* RPC is called synchronously */
int
afs_remove(OSI_VC_DECL(adp), char *aname, afs_ucred_t *acred)
{
struct vrequest treq;
register struct dcache *tdc;
struct VenusFid unlinkFid;
register afs_int32 code;
register struct vcache *tvc;
afs_size_t offset, len;
struct afs_fakestat_state fakestate;
OSI_VC_CONVERT(adp);
AFS_STATCNT(afs_remove);
afs_Trace2(afs_iclSetp, CM_TRACE_REMOVE, ICL_TYPE_POINTER, adp,
ICL_TYPE_STRING, aname);
if ((code = afs_InitReq(&treq, acred))) {
return code;
}
afs_InitFakeStat(&fakestate);
AFS_DISCON_LOCK();
code = afs_EvalFakeStat(&adp, &fakestate, &treq);
if (code)
goto done;
/* Check if this is dynroot */
if (afs_IsDynroot(adp)) {
code = afs_DynrootVOPRemove(adp, acred, aname);
goto done;
}
if (afs_IsDynrootMount(adp)) {
code = ENOENT;
goto done;
}
if (strlen(aname) > AFSNAMEMAX) {
code = ENAMETOOLONG;
goto done;
}
tagain:
code = afs_VerifyVCache(adp, &treq);
tvc = NULL;
if (code) {
code = afs_CheckCode(code, &treq, 23);
goto done;
}
/** If the volume is read-only, return error without making an RPC to the
* fileserver
*/
if (adp->f.states & CRO) {
code = EROFS;
goto done;
}
/* If we're running disconnected without logging, go no further... */
if (AFS_IS_DISCONNECTED && !AFS_IS_DISCON_RW) {
code = ENETDOWN;
goto done;
}
tdc = afs_GetDCache(adp, (afs_size_t) 0, &treq, &offset, &len, 1); /* test for error below */
ObtainWriteLock(&adp->lock, 142);
if (tdc)
ObtainSharedLock(&tdc->lock, 638);
/*
* Make sure that the data in the cache is current. We may have
* received a callback while we were waiting for the write lock.
*/
if (!(adp->f.states & CStatd)
|| (tdc && !hsame(adp->f.m.DataVersion, tdc->f.versionNo))) {
ReleaseWriteLock(&adp->lock);
if (tdc) {
ReleaseSharedLock(&tdc->lock);
afs_PutDCache(tdc);
}
goto tagain;
}
unlinkFid.Fid.Vnode = 0;
if (!tvc) {
tvc = osi_dnlc_lookup(adp, aname, WRITE_LOCK);
}
/* This should not be necessary since afs_lookup() has already
* done the work.
*/
if (!tvc)
if (tdc) {
code = afs_dir_Lookup(tdc, aname, &unlinkFid.Fid);
if (code == 0) {
afs_int32 cached = 0;
unlinkFid.Cell = adp->f.fid.Cell;
unlinkFid.Fid.Volume = adp->f.fid.Fid.Volume;
if (unlinkFid.Fid.Unique == 0) {
tvc =
afs_LookupVCache(&unlinkFid, &treq, &cached, adp,
aname);
} else {
ObtainReadLock(&afs_xvcache);
tvc = afs_FindVCache(&unlinkFid, 0, DO_STATS);
ReleaseReadLock(&afs_xvcache);
}
}
}
if (AFS_IS_DISCON_RW) {
if (!adp->f.shadow.vnode && !(adp->f.ddirty_flags & VDisconCreate)) {
/* Make shadow copy of parent dir. */
afs_MakeShadowDir(adp, tdc);
}
/* Can't hold a dcache lock whilst we're getting a vcache one */
if (tdc)
ReleaseSharedLock(&tdc->lock);
/* XXX - We're holding adp->lock still, and we've got no
* guarantee about whether the ordering matches the lock hierarchy */
ObtainWriteLock(&tvc->lock, 713);
/* If we were locally created, then we don't need to do very
* much beyond ensuring that we don't exist anymore */
if (tvc->f.ddirty_flags & VDisconCreate) {
afs_DisconRemoveDirty(tvc);
} else {
/* Add removed file vcache to dirty list. */
afs_DisconAddDirty(tvc, VDisconRemove, 1);
}
adp->f.m.LinkCount--;
ReleaseWriteLock(&tvc->lock);
if (tdc)
ObtainSharedLock(&tdc->lock, 714);
}
if (tvc && osi_Active(tvc)) {
/* about to delete whole file, prefetch it first */
ReleaseWriteLock(&adp->lock);
if (tdc)
ReleaseSharedLock(&tdc->lock);
ObtainWriteLock(&tvc->lock, 143);
FetchWholeEnchilada(tvc, &treq);
ReleaseWriteLock(&tvc->lock);
ObtainWriteLock(&adp->lock, 144);
/* Technically I don't think we need this back, but let's hold it
anyway; The "got" reference should actually be sufficient. */
if (tdc)
ObtainSharedLock(&tdc->lock, 640);
}
osi_dnlc_remove(adp, aname, tvc);
Tadp1 = adp;
#ifndef AFS_DARWIN80_ENV
Tadpr = VREFCOUNT(adp);
#endif
Ttvc = tvc;
Tnam = aname;
Tnam1 = 0;
#ifndef AFS_DARWIN80_ENV
if (tvc)
Ttvcr = VREFCOUNT(tvc);
#endif
#ifdef AFS_AIX_ENV
if (tvc && VREFCOUNT_GT(tvc, 2) && tvc->opens > 0
&& !(tvc->f.states & CUnlinked)) {
#else
if (tvc && VREFCOUNT_GT(tvc, 1) && tvc->opens > 0
&& !(tvc->f.states & CUnlinked)) {
#endif
char *unlname = afs_newname();
ReleaseWriteLock(&adp->lock);
if (tdc)
ReleaseSharedLock(&tdc->lock);
code = afsrename(adp, aname, adp, unlname, acred, &treq);
Tnam1 = unlname;
if (!code) {
struct VenusFid *oldmvid = NULL;
if (tvc->mvid)
oldmvid = tvc->mvid;
tvc->mvid = (struct VenusFid *)unlname;
if (oldmvid)
osi_FreeSmallSpace(oldmvid);
crhold(acred);
if (tvc->uncred) {
crfree(tvc->uncred);
}
tvc->uncred = acred;
tvc->f.states |= CUnlinked;
/* if rename succeeded, remove should not */
ObtainWriteLock(&tvc->lock, 715);
if (tvc->f.ddirty_flags & VDisconRemove) {
tvc->f.ddirty_flags &= ~VDisconRemove;
}
ReleaseWriteLock(&tvc->lock);
} else {
osi_FreeSmallSpace(unlname);
}
if (tdc)
afs_PutDCache(tdc);
afs_PutVCache(tvc);
} else {
code = afsremove(adp, tdc, tvc, aname, acred, &treq);
}
done:
afs_PutFakeStat(&fakestate);
#ifndef AFS_DARWIN80_ENV
/* we can't track by thread, it's not exported in the KPI; only do
this on !macos */
osi_Assert(!WriteLocked(&adp->lock) || (adp->lock.pid_writer != MyPidxx));
#endif
AFS_DISCON_UNLOCK();
return code;
}
/* afs_remunlink -- This tries to delete the file at the server after it has
* been renamed when unlinked locally but now has been finally released.
*
* CAUTION -- may be called with avc unheld. */
int
afs_remunlink(register struct vcache *avc, register int doit)
{
afs_ucred_t *cred;
char *unlname;
struct vcache *adp;
struct vrequest treq;
struct VenusFid dirFid;
register struct dcache *tdc;
afs_int32 code = 0;
if (NBObtainWriteLock(&avc->lock, 423))
return 0;
#if defined(AFS_DARWIN80_ENV)
if (vnode_get(AFSTOV(avc))) {
ReleaseWriteLock(&avc->lock);
return 0;
}
#endif
if (avc->mvid && (doit || (avc->f.states & CUnlinkedDel))) {
if ((code = afs_InitReq(&treq, avc->uncred))) {
ReleaseWriteLock(&avc->lock);
} else {
/* Must bump the refCount because GetVCache may block.
* Also clear mvid so no other thread comes here if we block.
*/
unlname = (char *)avc->mvid;
avc->mvid = NULL;
cred = avc->uncred;
avc->uncred = NULL;
#if defined(AFS_DARWIN_ENV) && !defined(AFS_DARWIN80_ENV)
VREF(AFSTOV(avc));
#else
AFS_FAST_HOLD(avc);
#endif
/* We'll only try this once. If it fails, just release the vnode.
* Clear after doing hold so that NewVCache doesn't find us yet.
*/
avc->f.states &= ~(CUnlinked | CUnlinkedDel);
ReleaseWriteLock(&avc->lock);
dirFid.Cell = avc->f.fid.Cell;
dirFid.Fid.Volume = avc->f.fid.Fid.Volume;
dirFid.Fid.Vnode = avc->f.parent.vnode;
dirFid.Fid.Unique = avc->f.parent.unique;
adp = afs_GetVCache(&dirFid, &treq, NULL, NULL);
if (adp) {
tdc = afs_FindDCache(adp, (afs_size_t) 0);
ObtainWriteLock(&adp->lock, 159);
if (tdc)
ObtainSharedLock(&tdc->lock, 639);
/* afsremove releases the adp & tdc locks, and does vn_rele(avc) */
code = afsremove(adp, tdc, avc, unlname, cred, &treq);
afs_PutVCache(adp);
} else {
/* we failed - and won't be back to try again. */
afs_PutVCache(avc);
}
osi_FreeSmallSpace(unlname);
crfree(cred);
}
} else {
#if defined(AFS_DARWIN80_ENV)
vnode_put(AFSTOV(avc));
#endif
ReleaseWriteLock(&avc->lock);
}
return code;
}
/*
* Look up a FFS dinode number to find its incore vnode, otherwise read it
* in from disk. If it is in core, wait for the lock bit to clear, then
* return the inode locked. Detection and handling of mount points must be
* done by the calling routine.
*/
int
ffs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
{
struct fs *fs;
struct inode *ip;
struct ufs1_dinode *dp1;
#ifdef FFS2
struct ufs2_dinode *dp2;
#endif
struct ufsmount *ump;
struct buf *bp;
struct vnode *vp;
dev_t dev;
int error;
ump = VFSTOUFS(mp);
dev = ump->um_dev;
retry:
if ((*vpp = ufs_ihashget(dev, ino)) != NULL)
return (0);
/* Allocate a new vnode/inode. */
if ((error = getnewvnode(VT_UFS, mp, ffs_vnodeop_p, &vp)) != 0) {
*vpp = NULL;
return (error);
}
#ifdef VFSDEBUG
vp->v_flag |= VLOCKSWORK;
#endif
/* XXX - we use the same pool for ffs and mfs */
ip = pool_get(&ffs_ino_pool, PR_WAITOK);
bzero((caddr_t)ip, sizeof(struct inode));
lockinit(&ip->i_lock, PINOD, "inode", 0, 0);
ip->i_ump = ump;
VREF(ip->i_devvp);
vp->v_data = ip;
ip->i_vnode = vp;
ip->i_fs = fs = ump->um_fs;
ip->i_dev = dev;
ip->i_number = ino;
ip->i_vtbl = &ffs_vtbl;
/*
* Put it onto its hash chain and lock it so that other requests for
* this inode will block if they arrive while we are sleeping waiting
* for old data structures to be purged or for the contents of the
* disk portion of this inode to be read.
*/
error = ufs_ihashins(ip);
if (error) {
/*
* VOP_INACTIVE will treat this as a stale file
* and recycle it quickly
*/
vrele(vp);
if (error == EEXIST)
goto retry;
return (error);
}
/* Read in the disk contents for the inode, copy into the inode. */
error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
(int)fs->fs_bsize, NOCRED, &bp);
if (error) {
/*
* The inode does not contain anything useful, so it would
* be misleading to leave it on its hash chain. With mode
* still zero, it will be unlinked and returned to the free
* list by vput().
*/
vput(vp);
brelse(bp);
*vpp = NULL;
return (error);
}
#ifdef FFS2
if (ip->i_ump->um_fstype == UM_UFS2) {
ip->i_din2 = pool_get(&ffs_dinode2_pool, PR_WAITOK);
dp2 = (struct ufs2_dinode *) bp->b_data + ino_to_fsbo(fs, ino);
*ip->i_din2 = *dp2;
} else
#endif
{
ip->i_din1 = pool_get(&ffs_dinode1_pool, PR_WAITOK);
dp1 = (struct ufs1_dinode *) bp->b_data + ino_to_fsbo(fs, ino);
*ip->i_din1 = *dp1;
}
brelse(bp);
if (DOINGSOFTDEP(vp))
softdep_load_inodeblock(ip);
else
ip->i_effnlink = DIP(ip, nlink);
/*
* Initialize the vnode from the inode, check for aliases.
* Note that the underlying vnode may have changed.
*/
error = ufs_vinit(mp, ffs_specop_p, FFS_FIFOOPS, &vp);
if (error) {
vput(vp);
*vpp = NULL;
return (error);
}
/*
* Set up a generation number for this inode if it does not
* already have one. This should only happen on old filesystems.
*/
if (DIP(ip, gen) == 0) {
DIP_ASSIGN(ip, gen, arc4random() & INT_MAX);
if (DIP(ip, gen) == 0 || DIP(ip, gen) == -1)
DIP_ASSIGN(ip, gen, 1); /* Shouldn't happen */
if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0)
ip->i_flag |= IN_MODIFIED;
}
/*
* Ensure that uid and gid are correct. This is a temporary
* fix until fsck has been changed to do the update.
*/
if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_inodefmt < FS_44INODEFMT) {
ip->i_ffs1_uid = ip->i_din1->di_ouid;
ip->i_ffs1_gid = ip->i_din1->di_ogid;
}
*vpp = vp;
return (0);
}
/*
* Look up a file or directory
*/
static int
pfs_lookup(struct vop_cachedlookup_args *va)
{
struct vnode *vn = va->a_dvp;
struct vnode **vpp = va->a_vpp;
struct componentname *cnp = va->a_cnp;
struct pfs_vdata *pvd = vn->v_data;
struct pfs_node *pd = pvd->pvd_pn;
struct pfs_node *pn, *pdn = NULL;
pid_t pid = pvd->pvd_pid;
char *pname;
int error, i, namelen, visible;
PFS_TRACE(("%.*s", (int)cnp->cn_namelen, cnp->cn_nameptr));
pfs_assert_not_owned(pd);
if (vn->v_type != VDIR)
PFS_RETURN (ENOTDIR);
KASSERT_PN_IS_DIR(pd);
error = VOP_ACCESS(vn, VEXEC, cnp->cn_cred, cnp->cn_thread);
if (error)
PFS_RETURN (error);
/*
* Don't support DELETE or RENAME. CREATE is supported so
* that O_CREAT will work, but the lookup will still fail if
* the file does not exist.
*/
if ((cnp->cn_flags & ISLASTCN) &&
(cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
PFS_RETURN (EOPNOTSUPP);
/* shortcut: check if the name is too long */
if (cnp->cn_namelen >= PFS_NAMELEN)
PFS_RETURN (ENOENT);
/* check that parent directory is visible... */
if (!pfs_visible(curthread, pd, pvd->pvd_pid, NULL))
PFS_RETURN (ENOENT);
/* self */
namelen = cnp->cn_namelen;
pname = cnp->cn_nameptr;
if (namelen == 1 && pname[0] == '.') {
pn = pd;
*vpp = vn;
VREF(vn);
PFS_RETURN (0);
}
/* parent */
if (cnp->cn_flags & ISDOTDOT) {
if (pd->pn_type == pfstype_root)
PFS_RETURN (EIO);
VOP_UNLOCK(vn, 0, cnp->cn_thread);
KASSERT(pd->pn_parent != NULL,
("%s(): non-root directory has no parent", __func__));
/*
* This one is tricky. Descendents of procdir nodes
* inherit their parent's process affinity, but
* there's no easy reverse mapping. For simplicity,
* we assume that if this node is a procdir, its
* parent isn't (which is correct as long as
* descendents of procdir nodes are never procdir
* nodes themselves)
*/
if (pd->pn_type == pfstype_procdir)
pid = NO_PID;
pfs_lock(pd);
pn = pd->pn_parent;
pfs_unlock(pd);
goto got_pnode;
}
pfs_lock(pd);
/* named node */
for (pn = pd->pn_nodes; pn != NULL; pn = pn->pn_next)
if (pn->pn_type == pfstype_procdir)
pdn = pn;
else if (pn->pn_name[namelen] == '\0' &&
bcmp(pname, pn->pn_name, namelen) == 0) {
pfs_unlock(pd);
goto got_pnode;
}
/* process dependent node */
if ((pn = pdn) != NULL) {
pid = 0;
for (pid = 0, i = 0; i < namelen && isdigit(pname[i]); ++i)
if ((pid = pid * 10 + pname[i] - '0') > PID_MAX)
break;
if (i == cnp->cn_namelen) {
pfs_unlock(pd);
goto got_pnode;
}
}
pfs_unlock(pd);
PFS_RETURN (ENOENT);
got_pnode:
pfs_assert_not_owned(pd);
pfs_assert_not_owned(pn);
visible = pfs_visible(curthread, pn, pid, NULL);
if (!visible) {
error = ENOENT;
goto failed;
}
error = pfs_vncache_alloc(vn->v_mount, vpp, pn, pid);
if (error)
goto failed;
if (cnp->cn_flags & ISDOTDOT)
vn_lock(vn, LK_EXCLUSIVE|LK_RETRY, cnp->cn_thread);
if (cnp->cn_flags & MAKEENTRY)
cache_enter(vn, *vpp, cnp);
PFS_RETURN (0);
failed:
if (cnp->cn_flags & ISDOTDOT)
vn_lock(vn, LK_EXCLUSIVE|LK_RETRY, cnp->cn_thread);
PFS_RETURN(error);
}
static int
common_fhtovp(struct mount * mp,
struct fid * fhp,
struct vnode ** vpp)
{
#ifdef ARLA_KNFS
struct netcred *np = NULL;
struct xfs_node *xn;
struct vnode *vp;
xfs_handle handle;
int error;
NNPFSDEB(XDEBVFOPS, ("xfs_fhtovp\n"));
if (fhp->fid_len != 16) {
printf("xfs_fhtovp: *PANIC* got a invalid length of a fid\n");
return EINVAL;
}
memcpy(&handle, fhp->fid_data, sizeof(handle));
NNPFSDEB(XDEBVFOPS, ("xfs_fhtovp: fid: %d.%d.%d.%d\n",
handle.a, handle.d, handle.c, handle.d));
NNPFSDEB(XDEBVFOPS, ("xfs_fhtovp: xfs_vnode_find\n"));
xn = xfs_node_find(&xfs[0].nodehead, &handle);
if (xn == NULL) {
struct xfs_message_getattr msg;
error = xfs_getnewvnode(xfs[0].mp, &vp, &handle);
if (error)
return error;
xfs_do_vget(vp, 0, curproc);
} else {
/* XXX access ? */
vp = XNODE_TO_VNODE(xn);
/* XXX wrong ? (we tell arla below) */
if (vp->v_usecount <= 0)
xfs_do_vget(vp, 0, curproc);
else
VREF(vp);
error = 0;
}
*vpp = vp;
if (error == 0) {
NNPFSDEB(XDEBVFOPS, ("xfs_fhtovp done\n"));
/*
* XXX tell arla about this node is hold by nfsd.
* There need to be code in xfs_write too.
*/
} else
NNPFSDEB(XDEBVFOPS, ("xfs_fhtovp failed (%d)\n", error));
return error;
#else /* !ARLA_KNFS */
return EOPNOTSUPP;
#endif /* !ARLA_KNFS */
}
static int
tmpfs_lookup(struct vop_cachedlookup_args *v)
{
struct vnode *dvp = v->a_dvp;
struct vnode **vpp = v->a_vpp;
struct componentname *cnp = v->a_cnp;
int error;
struct tmpfs_dirent *de;
struct tmpfs_node *dnode;
dnode = VP_TO_TMPFS_DIR(dvp);
*vpp = NULLVP;
/* Check accessibility of requested node as a first step. */
error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_thread);
if (error != 0)
goto out;
/* We cannot be requesting the parent directory of the root node. */
MPASS(IMPLIES(dnode->tn_type == VDIR &&
dnode->tn_dir.tn_parent == dnode,
!(cnp->cn_flags & ISDOTDOT)));
TMPFS_ASSERT_LOCKED(dnode);
if (dnode->tn_dir.tn_parent == NULL) {
error = ENOENT;
goto out;
}
if (cnp->cn_flags & ISDOTDOT) {
int ltype = 0;
ltype = VOP_ISLOCKED(dvp);
vhold(dvp);
VOP_UNLOCK(dvp, 0);
/* Allocate a new vnode on the matching entry. */
error = tmpfs_alloc_vp(dvp->v_mount, dnode->tn_dir.tn_parent,
cnp->cn_lkflags, vpp);
vn_lock(dvp, ltype | LK_RETRY);
vdrop(dvp);
} else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
VREF(dvp);
*vpp = dvp;
error = 0;
} else {
de = tmpfs_dir_lookup(dnode, NULL, cnp);
if (de != NULL && de->td_node == NULL)
cnp->cn_flags |= ISWHITEOUT;
if (de == NULL || de->td_node == NULL) {
/* The entry was not found in the directory.
* This is OK if we are creating or renaming an
* entry and are working on the last component of
* the path name. */
if ((cnp->cn_flags & ISLASTCN) &&
(cnp->cn_nameiop == CREATE || \
cnp->cn_nameiop == RENAME ||
(cnp->cn_nameiop == DELETE &&
cnp->cn_flags & DOWHITEOUT &&
cnp->cn_flags & ISWHITEOUT))) {
error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
cnp->cn_thread);
if (error != 0)
goto out;
/* Keep the component name in the buffer for
* future uses. */
cnp->cn_flags |= SAVENAME;
error = EJUSTRETURN;
} else
error = ENOENT;
} else {
struct tmpfs_node *tnode;
/* The entry was found, so get its associated
* tmpfs_node. */
tnode = de->td_node;
/* If we are not at the last path component and
* found a non-directory or non-link entry (which
* may itself be pointing to a directory), raise
* an error. */
if ((tnode->tn_type != VDIR &&
tnode->tn_type != VLNK) &&
!(cnp->cn_flags & ISLASTCN)) {
error = ENOTDIR;
goto out;
}
/* If we are deleting or renaming the entry, keep
* track of its tmpfs_dirent so that it can be
* easily deleted later. */
if ((cnp->cn_flags & ISLASTCN) &&
(cnp->cn_nameiop == DELETE ||
cnp->cn_nameiop == RENAME)) {
error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
cnp->cn_thread);
if (error != 0)
goto out;
/* Allocate a new vnode on the matching entry. */
error = tmpfs_alloc_vp(dvp->v_mount, tnode,
cnp->cn_lkflags, vpp);
if (error != 0)
goto out;
if ((dnode->tn_mode & S_ISTXT) &&
VOP_ACCESS(dvp, VADMIN, cnp->cn_cred, cnp->cn_thread) &&
VOP_ACCESS(*vpp, VADMIN, cnp->cn_cred, cnp->cn_thread)) {
error = EPERM;
vput(*vpp);
*vpp = NULL;
goto out;
}
cnp->cn_flags |= SAVENAME;
} else {
error = tmpfs_alloc_vp(dvp->v_mount, tnode,
cnp->cn_lkflags, vpp);
}
}
}
/* Store the result of this lookup in the cache. Avoid this if the
* request was for creation, as it does not improve timings on
* emprical tests. */
if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE)
cache_enter(dvp, *vpp, cnp);
out:
/* If there were no errors, *vpp cannot be null and it must be
* locked. */
MPASS(IFF(error == 0, *vpp != NULLVP && VOP_ISLOCKED(*vpp)));
return error;
}
/*
* Set up nameidata for a lookup() call and do it.
*
* If pubflag is set, this call is done for a lookup operation on the
* public filehandle. In that case we allow crossing mountpoints and
* absolute pathnames. However, the caller is expected to check that
* the lookup result is within the public fs, and deny access if
* it is not.
*
* nfs_namei() clears out garbage fields that namei() might leave garbage.
* This is mainly ni_vp and ni_dvp when an error occurs, and ni_dvp when no
* error occurs but the parent was not requested.
*
* dirp may be set whether an error is returned or not, and must be
* released by the caller.
*/
int
nfs_namei(struct nameidata *ndp, struct nfsrv_descript *nfsd,
fhandle_t *fhp, int len, struct nfssvc_sock *slp,
struct sockaddr *nam, struct mbuf **mdp,
caddr_t *dposp, struct vnode **retdirp, int v3, struct vattr *retdirattrp,
int *retdirattr_retp, int pubflag)
{
int i, rem;
struct mbuf *md;
char *fromcp, *tocp, *cp;
struct iovec aiov;
struct uio auio;
struct vnode *dp;
int error, rdonly, linklen;
struct componentname *cnp = &ndp->ni_cnd;
int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0;
*retdirp = NULL;
cnp->cn_flags |= NOMACCHECK;
cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
/*
* Copy the name from the mbuf list to ndp->ni_pnbuf
* and set the various ndp fields appropriately.
*/
fromcp = *dposp;
tocp = cnp->cn_pnbuf;
md = *mdp;
rem = mtod(md, caddr_t) + md->m_len - fromcp;
for (i = 0; i < len; i++) {
while (rem == 0) {
md = md->m_next;
if (md == NULL) {
error = EBADRPC;
goto out;
}
fromcp = mtod(md, caddr_t);
rem = md->m_len;
}
if (*fromcp == '\0' || (!pubflag && *fromcp == '/')) {
error = EACCES;
goto out;
}
*tocp++ = *fromcp++;
rem--;
}
*tocp = '\0';
*mdp = md;
*dposp = fromcp;
len = nfsm_rndup(len)-len;
if (len > 0) {
if (rem >= len)
*dposp += len;
else if ((error = nfs_adv(mdp, dposp, len, rem)) != 0)
goto out;
}
if (!pubflag && nfs_ispublicfh(fhp))
return (ESTALE);
/*
* Extract and set starting directory.
*/
error = nfsrv_fhtovp(fhp, 0, &dp, nfsd, slp, nam, &rdonly);
if (error)
goto out;
if (dp->v_type != VDIR) {
vput(dp);
error = ENOTDIR;
goto out;
}
if (rdonly)
cnp->cn_flags |= RDONLY;
/*
* Set return directory. Reference to dp is implicitly transfered
* to the returned pointer
*/
*retdirp = dp;
if (v3) {
*retdirattr_retp = VOP_GETATTR(dp, retdirattrp,
ndp->ni_cnd.cn_cred);
}
VOP_UNLOCK(dp, 0);
if (pubflag) {
/*
* Oh joy. For WebNFS, handle those pesky '%' escapes,
* and the 'native path' indicator.
*/
cp = uma_zalloc(namei_zone, M_WAITOK);
fromcp = cnp->cn_pnbuf;
tocp = cp;
if ((unsigned char)*fromcp >= WEBNFS_SPECCHAR_START) {
switch ((unsigned char)*fromcp) {
case WEBNFS_NATIVE_CHAR:
/*
* 'Native' path for us is the same
* as a path according to the NFS spec,
* just skip the escape char.
*/
fromcp++;
break;
/*
* More may be added in the future, range 0x80-0xff
*/
default:
error = EIO;
uma_zfree(namei_zone, cp);
goto out;
}
}
/*
* Translate the '%' escapes, URL-style.
*/
while (*fromcp != '\0') {
if (*fromcp == WEBNFS_ESC_CHAR) {
if (fromcp[1] != '\0' && fromcp[2] != '\0') {
fromcp++;
*tocp++ = HEXSTRTOI(fromcp);
fromcp += 2;
continue;
} else {
error = ENOENT;
uma_zfree(namei_zone, cp);
goto out;
}
} else
*tocp++ = *fromcp++;
}
*tocp = '\0';
uma_zfree(namei_zone, cnp->cn_pnbuf);
cnp->cn_pnbuf = cp;
}
ndp->ni_pathlen = (tocp - cnp->cn_pnbuf) + 1;
ndp->ni_segflg = UIO_SYSSPACE;
if (pubflag) {
ndp->ni_rootdir = rootvnode;
ndp->ni_loopcnt = 0;
if (cnp->cn_pnbuf[0] == '/')
dp = rootvnode;
} else {
cnp->cn_flags |= NOCROSSMOUNT;
}
/*
* Initialize for scan, set ni_startdir and bump ref on dp again
* because lookup() will dereference ni_startdir.
*/
cnp->cn_thread = curthread;
VREF(dp);
ndp->ni_startdir = dp;
if (!lockleaf)
cnp->cn_flags |= LOCKLEAF;
for (;;) {
cnp->cn_nameptr = cnp->cn_pnbuf;
/*
* Call lookup() to do the real work. If an error occurs,
* ndp->ni_vp and ni_dvp are left uninitialized or NULL and
* we do not have to dereference anything before returning.
* In either case ni_startdir will be dereferenced and NULLed
* out.
*/
error = lookup(ndp);
if (error)
break;
/*
* Check for encountering a symbolic link. Trivial
* termination occurs if no symlink encountered.
* Note: zfree is safe because error is 0, so we will
* not zfree it again when we break.
*/
if ((cnp->cn_flags & ISSYMLINK) == 0) {
if (cnp->cn_flags & (SAVENAME | SAVESTART))
cnp->cn_flags |= HASBUF;
else
uma_zfree(namei_zone, cnp->cn_pnbuf);
if (ndp->ni_vp && !lockleaf)
VOP_UNLOCK(ndp->ni_vp, 0);
break;
}
/*
* Validate symlink
*/
if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
VOP_UNLOCK(ndp->ni_dvp, 0);
if (!pubflag) {
error = EINVAL;
goto badlink2;
}
if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
error = ELOOP;
goto badlink2;
}
if (ndp->ni_pathlen > 1)
cp = uma_zalloc(namei_zone, M_WAITOK);
else
cp = cnp->cn_pnbuf;
aiov.iov_base = cp;
aiov.iov_len = MAXPATHLEN;
auio.uio_iov = &aiov;
auio.uio_iovcnt = 1;
auio.uio_offset = 0;
auio.uio_rw = UIO_READ;
auio.uio_segflg = UIO_SYSSPACE;
auio.uio_td = NULL;
auio.uio_resid = MAXPATHLEN;
error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
if (error) {
badlink1:
if (ndp->ni_pathlen > 1)
uma_zfree(namei_zone, cp);
badlink2:
vput(ndp->ni_vp);
vrele(ndp->ni_dvp);
break;
}
linklen = MAXPATHLEN - auio.uio_resid;
if (linklen == 0) {
error = ENOENT;
goto badlink1;
}
if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
error = ENAMETOOLONG;
goto badlink1;
}
/*
* Adjust or replace path
*/
if (ndp->ni_pathlen > 1) {
bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
uma_zfree(namei_zone, cnp->cn_pnbuf);
cnp->cn_pnbuf = cp;
} else
cnp->cn_pnbuf[linklen] = '\0';
ndp->ni_pathlen += linklen;
/*
* Cleanup refs for next loop and check if root directory
* should replace current directory. Normally ni_dvp
* becomes the new base directory and is cleaned up when
* we loop. Explicitly null pointers after invalidation
* to clarify operation.
*/
vput(ndp->ni_vp);
ndp->ni_vp = NULL;
if (cnp->cn_pnbuf[0] == '/') {
vrele(ndp->ni_dvp);
ndp->ni_dvp = ndp->ni_rootdir;
VREF(ndp->ni_dvp);
}
ndp->ni_startdir = ndp->ni_dvp;
ndp->ni_dvp = NULL;
}
if (!lockleaf)
cnp->cn_flags &= ~LOCKLEAF;
/*
* nfs_namei() guarentees that fields will not contain garbage
* whether an error occurs or not. This allows the caller to track
* cleanup state trivially.
*/
out:
if (error) {
uma_zfree(namei_zone, cnp->cn_pnbuf);
ndp->ni_vp = NULL;
ndp->ni_dvp = NULL;
ndp->ni_startdir = NULL;
cnp->cn_flags &= ~HASBUF;
} else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
ndp->ni_dvp = NULL;
}
return (error);
}
/*
* Rename system call.
* rename("foo", "bar");
* is essentially
* unlink("bar");
* link("foo", "bar");
* unlink("foo");
* but ``atomically''. Can't do full commit without saving state in the
* inode on disk which isn't feasible at this time. Best we can do is
* always guarantee the target exists.
*
* Basic algorithm is:
*
* 1) Bump link count on source while we're linking it to the
* target. This also ensure the inode won't be deleted out
* from underneath us while we work (it may be truncated by
* a concurrent `trunc' or `open' for creation).
* 2) Link source to destination. If destination already exists,
* delete it first.
* 3) Unlink source reference to inode if still around. If a
* directory was moved and the parent of the destination
* is different from the source, patch the ".." entry in the
* directory.
*/
static int
ext2_rename(struct vop_rename_args *ap)
{
struct vnode *tvp = ap->a_tvp;
struct vnode *tdvp = ap->a_tdvp;
struct vnode *fvp = ap->a_fvp;
struct vnode *fdvp = ap->a_fdvp;
struct componentname *tcnp = ap->a_tcnp;
struct componentname *fcnp = ap->a_fcnp;
struct inode *ip, *xp, *dp;
struct dirtemplate dirbuf;
int doingdirectory = 0, oldparent = 0, newparent = 0;
int error = 0;
u_char namlen;
#ifdef DIAGNOSTIC
if ((tcnp->cn_flags & HASBUF) == 0 ||
(fcnp->cn_flags & HASBUF) == 0)
panic("ext2_rename: no name");
#endif
/*
* Check for cross-device rename.
*/
if ((fvp->v_mount != tdvp->v_mount) ||
(tvp && (fvp->v_mount != tvp->v_mount))) {
error = EXDEV;
abortit:
if (tdvp == tvp)
vrele(tdvp);
else
vput(tdvp);
if (tvp)
vput(tvp);
vrele(fdvp);
vrele(fvp);
return (error);
}
if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
(VTOI(tdvp)->i_flags & APPEND))) {
error = EPERM;
goto abortit;
}
/*
* Renaming a file to itself has no effect. The upper layers should
* not call us in that case. Temporarily just warn if they do.
*/
if (fvp == tvp) {
printf("ext2_rename: fvp == tvp (can't happen)\n");
error = 0;
goto abortit;
}
if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
goto abortit;
dp = VTOI(fdvp);
ip = VTOI(fvp);
if (ip->i_nlink >= EXT2_LINK_MAX) {
VOP_UNLOCK(fvp, 0);
error = EMLINK;
goto abortit;
}
if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
|| (dp->i_flags & APPEND)) {
VOP_UNLOCK(fvp, 0);
error = EPERM;
goto abortit;
}
if ((ip->i_mode & IFMT) == IFDIR) {
/*
* Avoid ".", "..", and aliases of "." for obvious reasons.
*/
if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT ||
(ip->i_flag & IN_RENAME)) {
VOP_UNLOCK(fvp, 0);
error = EINVAL;
goto abortit;
}
ip->i_flag |= IN_RENAME;
oldparent = dp->i_number;
doingdirectory++;
}
vrele(fdvp);
/*
* When the target exists, both the directory
* and target vnodes are returned locked.
*/
dp = VTOI(tdvp);
xp = NULL;
if (tvp)
xp = VTOI(tvp);
/*
* 1) Bump link count while we're moving stuff
* around. If we crash somewhere before
* completing our work, the link count
* may be wrong, but correctable.
*/
ip->i_nlink++;
ip->i_flag |= IN_CHANGE;
if ((error = ext2_update(fvp, !DOINGASYNC(fvp))) != 0) {
VOP_UNLOCK(fvp, 0);
goto bad;
}
/*
* If ".." must be changed (ie the directory gets a new
* parent) then the source directory must not be in the
* directory hierarchy above the target, as this would
* orphan everything below the source directory. Also
* the user must have write permission in the source so
* as to be able to change "..". We must repeat the call
* to namei, as the parent directory is unlocked by the
* call to checkpath().
*/
error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
VOP_UNLOCK(fvp, 0);
if (oldparent != dp->i_number)
newparent = dp->i_number;
if (doingdirectory && newparent) {
if (error) /* write access check above */
goto bad;
if (xp != NULL)
vput(tvp);
error = ext2_checkpath(ip, dp, tcnp->cn_cred);
if (error)
goto out;
VREF(tdvp);
error = relookup(tdvp, &tvp, tcnp);
if (error)
goto out;
vrele(tdvp);
dp = VTOI(tdvp);
xp = NULL;
if (tvp)
xp = VTOI(tvp);
}
/*
* 2) If target doesn't exist, link the target
* to the source and unlink the source.
* Otherwise, rewrite the target directory
* entry to reference the source inode and
* expunge the original entry's existence.
*/
if (xp == NULL) {
if (dp->i_devvp != ip->i_devvp)
panic("ext2_rename: EXDEV");
/*
* Account for ".." in new directory.
* When source and destination have the same
* parent we don't fool with the link count.
*/
if (doingdirectory && newparent) {
if ((nlink_t)dp->i_nlink >= EXT2_LINK_MAX) {
error = EMLINK;
goto bad;
}
dp->i_nlink++;
dp->i_flag |= IN_CHANGE;
error = ext2_update(tdvp, !DOINGASYNC(tdvp));
if (error)
goto bad;
}
error = ext2_direnter(ip, tdvp, tcnp);
if (error) {
if (doingdirectory && newparent) {
dp->i_nlink--;
dp->i_flag |= IN_CHANGE;
(void)ext2_update(tdvp, 1);
}
goto bad;
}
vput(tdvp);
} else {
if (xp->i_devvp != dp->i_devvp || xp->i_devvp != ip->i_devvp)
panic("ext2_rename: EXDEV");
/*
* Short circuit rename(foo, foo).
*/
if (xp->i_number == ip->i_number)
panic("ext2_rename: same file");
/*
* If the parent directory is "sticky", then the user must
* own the parent directory, or the destination of the rename,
* otherwise the destination may not be changed (except by
* root). This implements append-only directories.
*/
if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 &&
tcnp->cn_cred->cr_uid != dp->i_uid &&
xp->i_uid != tcnp->cn_cred->cr_uid) {
error = EPERM;
goto bad;
}
/*
* Target must be empty if a directory and have no links
* to it. Also, ensure source and target are compatible
* (both directories, or both not directories).
*/
if ((xp->i_mode&IFMT) == IFDIR) {
if (! ext2_dirempty(xp, dp->i_number, tcnp->cn_cred) ||
xp->i_nlink > 2) {
error = ENOTEMPTY;
goto bad;
}
if (!doingdirectory) {
error = ENOTDIR;
goto bad;
}
cache_purge(tdvp);
} else if (doingdirectory) {
error = EISDIR;
goto bad;
}
error = ext2_dirrewrite(dp, ip, tcnp);
if (error)
goto bad;
/*
* If the target directory is in the same
* directory as the source directory,
* decrement the link count on the parent
* of the target directory.
*/
if (doingdirectory && !newparent) {
dp->i_nlink--;
dp->i_flag |= IN_CHANGE;
}
vput(tdvp);
/*
* Adjust the link count of the target to
* reflect the dirrewrite above. If this is
* a directory it is empty and there are
* no links to it, so we can squash the inode and
* any space associated with it. We disallowed
* renaming over top of a directory with links to
* it above, as the remaining link would point to
* a directory without "." or ".." entries.
*/
xp->i_nlink--;
if (doingdirectory) {
if (--xp->i_nlink != 0)
panic("ext2_rename: linked directory");
error = ext2_truncate(tvp, (off_t)0, IO_SYNC,
tcnp->cn_cred, tcnp->cn_thread);
}
xp->i_flag |= IN_CHANGE;
vput(tvp);
xp = NULL;
}
/*
* 3) Unlink the source.
*/
fcnp->cn_flags &= ~MODMASK;
fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
VREF(fdvp);
error = relookup(fdvp, &fvp, fcnp);
if (error == 0)
vrele(fdvp);
if (fvp != NULL) {
xp = VTOI(fvp);
dp = VTOI(fdvp);
} else {
/*
* From name has disappeared.
*/
if (doingdirectory)
panic("ext2_rename: lost dir entry");
vrele(ap->a_fvp);
return (0);
}
/*
* Ensure that the directory entry still exists and has not
* changed while the new name has been entered. If the source is
* a file then the entry may have been unlinked or renamed. In
* either case there is no further work to be done. If the source
* is a directory then it cannot have been rmdir'ed; its link
* count of three would cause a rmdir to fail with ENOTEMPTY.
* The IN_RENAME flag ensures that it cannot be moved by another
* rename.
*/
if (xp != ip) {
if (doingdirectory)
panic("ext2_rename: lost dir entry");
} else {
/*
* If the source is a directory with a
* new parent, the link count of the old
* parent directory must be decremented
* and ".." set to point to the new parent.
*/
if (doingdirectory && newparent) {
dp->i_nlink--;
dp->i_flag |= IN_CHANGE;
error = vn_rdwr(UIO_READ, fvp, (caddr_t)&dirbuf,
sizeof(struct dirtemplate), (off_t)0,
UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
tcnp->cn_cred, NOCRED, NULL, NULL);
if (error == 0) {
/* Like ufs little-endian: */
namlen = dirbuf.dotdot_type;
if (namlen != 2 ||
dirbuf.dotdot_name[0] != '.' ||
dirbuf.dotdot_name[1] != '.') {
ext2_dirbad(xp, (doff_t)12,
"rename: mangled dir");
} else {
dirbuf.dotdot_ino = newparent;
(void) vn_rdwr(UIO_WRITE, fvp,
(caddr_t)&dirbuf,
sizeof(struct dirtemplate),
(off_t)0, UIO_SYSSPACE,
IO_NODELOCKED | IO_SYNC |
IO_NOMACCHECK, tcnp->cn_cred,
NOCRED, NULL, NULL);
cache_purge(fdvp);
}
}
}
error = ext2_dirremove(fdvp, fcnp);
if (!error) {
xp->i_nlink--;
xp->i_flag |= IN_CHANGE;
}
xp->i_flag &= ~IN_RENAME;
}
if (dp)
vput(fdvp);
if (xp)
vput(fvp);
vrele(ap->a_fvp);
return (error);
bad:
if (xp)
vput(ITOV(xp));
vput(ITOV(dp));
out:
if (doingdirectory)
ip->i_flag &= ~IN_RENAME;
if (vn_lock(fvp, LK_EXCLUSIVE) == 0) {
ip->i_nlink--;
ip->i_flag |= IN_CHANGE;
ip->i_flag &= ~IN_RENAME;
vput(fvp);
} else
vrele(fvp);
return (error);
}
/*
* Convert a pathname into a pointer to a vnode.
*
* The FOLLOW flag is set when symbolic links are to be followed
* when they occur at the end of the name translation process.
* Symbolic links are always followed for all other pathname
* components other than the last.
*
* If the LOCKLEAF flag is set, a locked vnode is returned.
*
* The segflg defines whether the name is to be copied from user
* space or kernel space.
*
* Overall outline of namei:
*
* copy in name
* get starting directory
* while (!done && !error) {
* call lookup to search path.
* if symbolic link, massage name in buffer and continue
* }
*/
int
namei(struct nameidata *ndp)
{
struct filedesc *fdp; /* pointer to file descriptor state */
char *cp; /* pointer into pathname argument */
struct vnode *dp; /* the directory we are searching */
struct iovec aiov; /* uio for reading symbolic links */
struct uio auio;
int error, linklen;
struct componentname *cnp = &ndp->ni_cnd;
struct proc *p = cnp->cn_proc;
ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_proc->p_ucred;
#ifdef DIAGNOSTIC
if (!cnp->cn_cred || !cnp->cn_proc)
panic ("namei: bad cred/proc");
if (cnp->cn_nameiop & (~OPMASK))
panic ("namei: nameiop contaminated with flags");
if (cnp->cn_flags & OPMASK)
panic ("namei: flags contaminated with nameiops");
#endif
fdp = cnp->cn_proc->p_fd;
/*
* Get a buffer for the name to be translated, and copy the
* name into the buffer.
*/
if ((cnp->cn_flags & HASBUF) == 0)
cnp->cn_pnbuf = pool_get(&namei_pool, PR_WAITOK);
if (ndp->ni_segflg == UIO_SYSSPACE)
error = copystr(ndp->ni_dirp, cnp->cn_pnbuf,
MAXPATHLEN, &ndp->ni_pathlen);
else
error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf,
MAXPATHLEN, &ndp->ni_pathlen);
/*
* Fail on null pathnames
*/
if (error == 0 && ndp->ni_pathlen == 1)
error = ENOENT;
if (error) {
pool_put(&namei_pool, cnp->cn_pnbuf);
ndp->ni_vp = NULL;
return (error);
}
#ifdef KTRACE
if (KTRPOINT(cnp->cn_proc, KTR_NAMEI))
ktrnamei(cnp->cn_proc, cnp->cn_pnbuf);
#endif
#if NSYSTRACE > 0
if (ISSET(cnp->cn_proc->p_flag, P_SYSTRACE))
systrace_namei(ndp);
#endif
/*
* Strip trailing slashes, as requested
*/
if (cnp->cn_flags & STRIPSLASHES) {
char *end = cnp->cn_pnbuf + ndp->ni_pathlen - 2;
cp = end;
while (cp >= cnp->cn_pnbuf && (*cp == '/'))
cp--;
/* Still some remaining characters in the buffer */
if (cp >= cnp->cn_pnbuf) {
ndp->ni_pathlen -= (end - cp);
*(cp + 1) = '\0';
}
}
ndp->ni_loopcnt = 0;
/*
* Get starting point for the translation.
*/
if ((ndp->ni_rootdir = fdp->fd_rdir) == NULL)
ndp->ni_rootdir = rootvnode;
/*
* Check if starting from root directory or current directory.
*/
if (cnp->cn_pnbuf[0] == '/') {
dp = ndp->ni_rootdir;
VREF(dp);
} else {
dp = fdp->fd_cdir;
VREF(dp);
}
for (;;) {
if (!dp->v_mount) {
/* Give up if the directory is no longer mounted */
pool_put(&namei_pool, cnp->cn_pnbuf);
return (ENOENT);
}
cnp->cn_nameptr = cnp->cn_pnbuf;
ndp->ni_startdir = dp;
if ((error = lookup(ndp)) != 0) {
pool_put(&namei_pool, cnp->cn_pnbuf);
return (error);
}
/*
* Check for symbolic link
*/
if ((cnp->cn_flags & ISSYMLINK) == 0) {
if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
pool_put(&namei_pool, cnp->cn_pnbuf);
else
cnp->cn_flags |= HASBUF;
return (0);
}
if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN))
VOP_UNLOCK(ndp->ni_dvp, 0, p);
if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
error = ELOOP;
break;
}
if (ndp->ni_pathlen > 1)
cp = pool_get(&namei_pool, PR_WAITOK);
else
cp = cnp->cn_pnbuf;
aiov.iov_base = cp;
aiov.iov_len = MAXPATHLEN;
auio.uio_iov = &aiov;
auio.uio_iovcnt = 1;
auio.uio_offset = 0;
auio.uio_rw = UIO_READ;
auio.uio_segflg = UIO_SYSSPACE;
auio.uio_procp = cnp->cn_proc;
auio.uio_resid = MAXPATHLEN;
error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
if (error) {
badlink:
if (ndp->ni_pathlen > 1)
pool_put(&namei_pool, cp);
break;
}
linklen = MAXPATHLEN - auio.uio_resid;
if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
error = ENAMETOOLONG;
goto badlink;
}
if (ndp->ni_pathlen > 1) {
bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
pool_put(&namei_pool, cnp->cn_pnbuf);
cnp->cn_pnbuf = cp;
} else
cnp->cn_pnbuf[linklen] = '\0';
ndp->ni_pathlen += linklen;
vput(ndp->ni_vp);
dp = ndp->ni_dvp;
/*
* Check if root directory should replace current directory.
*/
if (cnp->cn_pnbuf[0] == '/') {
vrele(dp);
dp = ndp->ni_rootdir;
VREF(dp);
}
}
pool_put(&namei_pool, cnp->cn_pnbuf);
vrele(ndp->ni_dvp);
vput(ndp->ni_vp);
ndp->ni_vp = NULL;
return (error);
}
/*
* vp is the current namei directory
* ndp is the name to locate in that directory...
*/
static int
fdesc_lookup(struct vop_lookup_args *ap)
{
struct vnode **vpp = ap->a_vpp;
struct vnode *dvp = ap->a_dvp;
struct componentname *cnp = ap->a_cnp;
char *pname = cnp->cn_nameptr;
struct thread *td = cnp->cn_thread;
struct file *fp;
struct fdesc_get_ino_args arg;
cap_rights_t rights;
int nlen = cnp->cn_namelen;
u_int fd, fd1;
int error;
struct vnode *fvp;
if ((cnp->cn_flags & ISLASTCN) &&
(cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
error = EROFS;
goto bad;
}
if (cnp->cn_namelen == 1 && *pname == '.') {
*vpp = dvp;
VREF(dvp);
return (0);
}
if (VTOFDESC(dvp)->fd_type != Froot) {
error = ENOTDIR;
goto bad;
}
fd = 0;
/* the only time a leading 0 is acceptable is if it's "0" */
if (*pname == '0' && nlen != 1) {
error = ENOENT;
goto bad;
}
while (nlen--) {
if (*pname < '0' || *pname > '9') {
error = ENOENT;
goto bad;
}
fd1 = 10 * fd + *pname++ - '0';
if (fd1 < fd) {
error = ENOENT;
goto bad;
}
fd = fd1;
}
/*
* No rights to check since 'fp' isn't actually used.
*/
if ((error = fget(td, fd, cap_rights_init(&rights), &fp)) != 0)
goto bad;
/* Check if we're looking up ourselves. */
if (VTOFDESC(dvp)->fd_ix == FD_DESC + fd) {
/*
* In case we're holding the last reference to the file, the dvp
* will be re-acquired.
*/
vhold(dvp);
VOP_UNLOCK(dvp, 0);
fdrop(fp, td);
/* Re-aquire the lock afterwards. */
vn_lock(dvp, LK_RETRY | LK_EXCLUSIVE);
vdrop(dvp);
fvp = dvp;
if ((dvp->v_iflag & VI_DOOMED) != 0)
error = ENOENT;
} else {
/*
* Unlock our root node (dvp) when doing this, since we might
* deadlock since the vnode might be locked by another thread
* and the root vnode lock will be obtained afterwards (in case
* we're looking up the fd of the root vnode), which will be the
* opposite lock order. Vhold the root vnode first so we don't
* lose it.
*/
arg.ftype = Fdesc;
arg.fd_fd = fd;
arg.ix = FD_DESC + fd;
arg.fp = fp;
arg.td = td;
error = vn_vget_ino_gen(dvp, fdesc_get_ino_alloc, &arg,
LK_EXCLUSIVE, &fvp);
}
if (error)
goto bad;
*vpp = fvp;
return (0);
bad:
*vpp = NULL;
return (error);
}
/*
* This is the 10-Apr-92 bypass routine.
* This version has been optimized for speed, throwing away some
* safety checks. It should still always work, but it's not as
* robust to programmer errors.
*
* In general, we map all vnodes going down and unmap them on the way back.
* As an exception to this, vnodes can be marked "unmapped" by setting
* the Nth bit in operation's vdesc_flags.
*
* Also, some BSD vnode operations have the side effect of vrele'ing
* their arguments. With stacking, the reference counts are held
* by the upper node, not the lower one, so we must handle these
* side-effects here. This is not of concern in Sun-derived systems
* since there are no such side-effects.
*
* This makes the following assumptions:
* - only one returned vpp
* - no INOUT vpp's (Sun's vop_open has one of these)
* - the vnode operation vector of the first vnode should be used
* to determine what implementation of the op should be invoked
* - all mapped vnodes are of our vnode-type (NEEDSWORK:
* problems on rmdir'ing mount points and renaming?)
*/
int
null_bypass(struct vop_generic_args *ap)
{
struct vnode **this_vp_p;
int error;
struct vnode *old_vps[VDESC_MAX_VPS];
struct vnode **vps_p[VDESC_MAX_VPS];
struct vnode ***vppp;
struct vnodeop_desc *descp = ap->a_desc;
int reles, i;
if (null_bug_bypass)
printf ("null_bypass: %s\n", descp->vdesc_name);
#ifdef DIAGNOSTIC
/*
* We require at least one vp.
*/
if (descp->vdesc_vp_offsets == NULL ||
descp->vdesc_vp_offsets[0] == VDESC_NO_OFFSET)
panic ("null_bypass: no vp's in map");
#endif
/*
* Map the vnodes going in.
* Later, we'll invoke the operation based on
* the first mapped vnode's operation vector.
*/
reles = descp->vdesc_flags;
for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) {
if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
break; /* bail out at end of list */
vps_p[i] = this_vp_p =
VOPARG_OFFSETTO(struct vnode**,descp->vdesc_vp_offsets[i],ap);
/*
* We're not guaranteed that any but the first vnode
* are of our type. Check for and don't map any
* that aren't. (We must always map first vp or vclean fails.)
*/
if (i && (*this_vp_p == NULLVP ||
(*this_vp_p)->v_op != &null_vnodeops)) {
old_vps[i] = NULLVP;
} else {
old_vps[i] = *this_vp_p;
*(vps_p[i]) = NULLVPTOLOWERVP(*this_vp_p);
/*
* XXX - Several operations have the side effect
* of vrele'ing their vp's. We must account for
* that. (This should go away in the future.)
*/
if (reles & VDESC_VP0_WILLRELE)
VREF(*this_vp_p);
}
}
/*
* Call the operation on the lower layer
* with the modified argument structure.
*/
if (vps_p[0] && *vps_p[0])
error = VCALL(ap);
else {
printf("null_bypass: no map for %s\n", descp->vdesc_name);
error = EINVAL;
}
/*
* Maintain the illusion of call-by-value
* by restoring vnodes in the argument structure
* to their original value.
*/
reles = descp->vdesc_flags;
for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) {
if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
break; /* bail out at end of list */
if (old_vps[i]) {
*(vps_p[i]) = old_vps[i];
#if 0
if (reles & VDESC_VP0_WILLUNLOCK)
VOP_UNLOCK(*(vps_p[i]), 0);
#endif
if (reles & VDESC_VP0_WILLRELE)
vrele(*(vps_p[i]));
}
}
/*
* Map the possible out-going vpp
* (Assumes that the lower layer always returns
* a VREF'ed vpp unless it gets an error.)
*/
if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET &&
!(descp->vdesc_flags & VDESC_NOMAP_VPP) &&
!error) {
/*
* XXX - even though some ops have vpp returned vp's,
* several ops actually vrele this before returning.
* We must avoid these ops.
* (This should go away when these ops are regularized.)
*/
if (descp->vdesc_flags & VDESC_VPP_WILLRELE)
goto out;
vppp = VOPARG_OFFSETTO(struct vnode***,
descp->vdesc_vpp_offset,ap);
if (*vppp)
error = null_nodeget(old_vps[0]->v_mount, **vppp, *vppp);
}
out:
return (error);
}
/*
* Reacquire a path name component.
*/
int
relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
{
struct proc *p = cnp->cn_proc;
struct vnode *dp = 0; /* the directory we are searching */
int wantparent; /* 1 => wantparent or lockparent flag */
int rdonly; /* lookup read-only flag bit */
int error = 0;
#ifdef NAMEI_DIAGNOSTIC
int newhash; /* DEBUG: check name hash */
char *cp; /* DEBUG: check name ptr/len */
#endif
/*
* Setup: break out flag bits into variables.
*/
wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
rdonly = cnp->cn_flags & RDONLY;
cnp->cn_flags &= ~ISSYMLINK;
dp = dvp;
vn_lock(dp, LK_EXCLUSIVE | LK_RETRY, p);
/* dirloop: */
/*
* Search a new directory.
*
* The cn_hash value is for use by vfs_cache.
* The last component of the filename is left accessible via
* cnp->cn_nameptr for callers that need the name. Callers needing
* the name set the SAVENAME flag. When done, they assume
* responsibility for freeing the pathname buffer.
*/
#ifdef NAMEI_DIAGNOSTIC
cp = NULL;
newhash = hash32_stre(cnp->cn_nameptr, '/', &cp, HASHINIT);
if (newhash != cnp->cn_hash)
panic("relookup: bad hash");
if (cnp->cn_namelen != cp - cnp->cn_nameptr)
panic ("relookup: bad len");
if (*cp != 0)
panic("relookup: not last component");
printf("{%s}: ", cnp->cn_nameptr);
#endif
/*
* Check for degenerate name (e.g. / or "")
* which is a way of talking about a directory,
* e.g. like "/." or ".".
*/
if (cnp->cn_nameptr[0] == '\0')
panic("relookup: null name");
if (cnp->cn_flags & ISDOTDOT)
panic ("relookup: lookup on dot-dot");
/*
* We now have a segment name to search for, and a directory to search.
*/
if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
#ifdef DIAGNOSTIC
if (*vpp != NULL)
panic("leaf should be empty");
#endif
if (error != EJUSTRETURN)
goto bad;
/*
* If creating and at end of pathname, then can consider
* allowing file to be created.
*/
if (rdonly || (dvp->v_mount->mnt_flag & MNT_RDONLY)) {
error = EROFS;
goto bad;
}
/* ASSERT(dvp == ndp->ni_startdir) */
if (cnp->cn_flags & SAVESTART)
VREF(dvp);
/*
* We return with ni_vp NULL to indicate that the entry
* doesn't currently exist, leaving a pointer to the
* (possibly locked) directory inode in ndp->ni_dvp.
*/
return (0);
}
dp = *vpp;
#ifdef DIAGNOSTIC
/*
* Check for symbolic link
*/
if (dp->v_type == VLNK && (cnp->cn_flags & FOLLOW))
panic ("relookup: symlink found.");
#endif
/*
* Check for read-only file systems.
*/
if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME) {
/*
* Disallow directory write attempts on read-only
* file systems.
*/
if (rdonly || (dp->v_mount->mnt_flag & MNT_RDONLY) ||
(wantparent &&
(dvp->v_mount->mnt_flag & MNT_RDONLY))) {
error = EROFS;
goto bad2;
}
}
/* ASSERT(dvp == ndp->ni_startdir) */
if (cnp->cn_flags & SAVESTART)
VREF(dvp);
if (!wantparent)
vrele(dvp);
if ((cnp->cn_flags & LOCKLEAF) == 0)
VOP_UNLOCK(dp, 0, p);
return (0);
bad2:
if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN))
VOP_UNLOCK(dvp, 0, p);
vrele(dvp);
bad:
vput(dp);
*vpp = NULL;
return (error);
}
/* Common routine shared by sys___getcwd() and vn_isunder() */
int
vfs_getcwd_common(struct vnode *lvp, struct vnode *rvp, char **bpp, char *bufp,
int limit, int flags, struct proc *p)
{
struct filedesc *fdp = p->p_fd;
struct vnode *uvp = NULL;
char *bp = NULL;
int error, perms = VEXEC;
if (rvp == NULL) {
rvp = fdp->fd_rdir;
if (rvp == NULL)
rvp = rootvnode;
}
VREF(rvp);
VREF(lvp);
error = vn_lock(lvp, LK_EXCLUSIVE | LK_RETRY, p);
if (error) {
vrele(lvp);
lvp = NULL;
goto out;
}
if (bufp)
bp = *bpp;
if (lvp == rvp) {
if (bp)
*(--bp) = '/';
goto out;
}
/*
* This loop will terminate when we hit the root, VOP_READDIR() or
* VOP_LOOKUP() fails, or we run out of space in the user buffer.
*/
do {
if (lvp->v_type != VDIR) {
error = ENOTDIR;
goto out;
}
/* Check for access if caller cares */
if (flags & GETCWD_CHECK_ACCESS) {
error = VOP_ACCESS(lvp, perms, p->p_ucred, p);
if (error)
goto out;
perms = VEXEC|VREAD;
}
/* Step up if we're a covered vnode */
while (lvp->v_flag & VROOT) {
struct vnode *tvp;
if (lvp == rvp)
goto out;
tvp = lvp;
lvp = lvp->v_mount->mnt_vnodecovered;
vput(tvp);
if (lvp == NULL) {
error = ENOENT;
goto out;
}
VREF(lvp);
error = vn_lock(lvp, LK_EXCLUSIVE | LK_RETRY, p);
if (error) {
vrele(lvp);
lvp = NULL;
goto out;
}
}
/* Look in the name cache */
error = vfs_getcwd_getcache(&lvp, &uvp, &bp, bufp);
if (error == -1) {
/* If that fails, look in the directory */
error = vfs_getcwd_scandir(&lvp, &uvp, &bp, bufp, p);
}
if (error)
goto out;
#ifdef DIAGNOSTIC
if (lvp != NULL)
panic("getcwd: oops, forgot to null lvp");
if (bufp && (bp <= bufp)) {
panic("getcwd: oops, went back too far");
}
#endif
if (bp)
*(--bp) = '/';
lvp = uvp;
uvp = NULL;
limit--;
} while ((lvp != rvp) && (limit > 0));
out:
if (bpp)
*bpp = bp;
if (uvp)
vput(uvp);
if (lvp)
vput(lvp);
vrele(rvp);
return (error);
}
/*
* When we search a directory the blocks containing directory entries are
* read and examined. The directory entries contain information that would
* normally be in the inode of a unix filesystem. This means that some of
* a directory's contents may also be in memory resident denodes (sort of
* an inode). This can cause problems if we are searching while some other
* process is modifying a directory. To prevent one process from accessing
* incompletely modified directory information we depend upon being the
* sole owner of a directory block. bread/brelse provide this service.
* This being the case, when a process modifies a directory it must first
* acquire the disk block that contains the directory entry to be modified.
* Then update the disk block and the denode, and then write the disk block
* out to disk. This way disk blocks containing directory entries and in
* memory denode's will be in synch.
*/
static int
msdosfs_lookup_(struct vnode *vdp, struct vnode **vpp,
struct componentname *cnp, u_int64_t *dd_inum)
{
struct mbnambuf nb;
daddr_t bn;
int error;
int slotcount;
int slotoffset = 0;
int frcn;
u_long cluster;
int blkoff;
int diroff;
int blsize;
int isadir; /* ~0 if found direntry is a directory */
u_long scn; /* starting cluster number */
struct vnode *pdp;
struct denode *dp;
struct denode *tdp;
struct msdosfsmount *pmp;
struct buf *bp = 0;
struct direntry *dep = NULL;
u_char dosfilename[12];
int flags = cnp->cn_flags;
int nameiop = cnp->cn_nameiop;
int unlen;
u_int64_t inode1;
int wincnt = 1;
int chksum = -1, chksum_ok;
int olddos = 1;
#ifdef MSDOSFS_DEBUG
printf("msdosfs_lookup(): looking for %s\n", cnp->cn_nameptr);
#endif
dp = VTODE(vdp);
pmp = dp->de_pmp;
#ifdef MSDOSFS_DEBUG
printf("msdosfs_lookup(): vdp %p, dp %p, Attr %02x\n",
vdp, dp, dp->de_Attributes);
#endif
restart:
if (vpp != NULL)
*vpp = NULL;
/*
* If they are going after the . or .. entry in the root directory,
* they won't find it. DOS filesystems don't have them in the root
* directory. So, we fake it. deget() is in on this scam too.
*/
if ((vdp->v_vflag & VV_ROOT) && cnp->cn_nameptr[0] == '.' &&
(cnp->cn_namelen == 1 ||
(cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.'))) {
isadir = ATTR_DIRECTORY;
scn = MSDOSFSROOT;
#ifdef MSDOSFS_DEBUG
printf("msdosfs_lookup(): looking for . or .. in root directory\n");
#endif
cluster = MSDOSFSROOT;
blkoff = MSDOSFSROOT_OFS;
goto foundroot;
}
switch (unix2dosfn((const u_char *)cnp->cn_nameptr, dosfilename,
cnp->cn_namelen, 0, pmp)) {
case 0:
return (EINVAL);
case 1:
break;
case 2:
wincnt = winSlotCnt((const u_char *)cnp->cn_nameptr,
cnp->cn_namelen, pmp) + 1;
break;
case 3:
olddos = 0;
wincnt = winSlotCnt((const u_char *)cnp->cn_nameptr,
cnp->cn_namelen, pmp) + 1;
break;
}
if (pmp->pm_flags & MSDOSFSMNT_SHORTNAME) {
wincnt = 1;
olddos = 1;
}
unlen = winLenFixup(cnp->cn_nameptr, cnp->cn_namelen);
/*
* Suppress search for slots unless creating
* file and at end of pathname, in which case
* we watch for a place to put the new file in
* case it doesn't already exist.
*/
slotcount = wincnt;
if ((nameiop == CREATE || nameiop == RENAME) &&
(flags & ISLASTCN))
slotcount = 0;
#ifdef MSDOSFS_DEBUG
printf("msdosfs_lookup(): dos version of filename %s, length %ld\n",
dosfilename, cnp->cn_namelen);
#endif
/*
* Search the directory pointed at by vdp for the name pointed at
* by cnp->cn_nameptr.
*/
tdp = NULL;
mbnambuf_init(&nb);
/*
* The outer loop ranges over the clusters that make up the
* directory. Note that the root directory is different from all
* other directories. It has a fixed number of blocks that are not
* part of the pool of allocatable clusters. So, we treat it a
* little differently. The root directory starts at "cluster" 0.
*/
diroff = 0;
for (frcn = 0;; frcn++) {
error = pcbmap(dp, frcn, &bn, &cluster, &blsize);
if (error) {
if (error == E2BIG)
break;
return (error);
}
error = bread(pmp->pm_devvp, bn, blsize, NOCRED, &bp);
if (error) {
brelse(bp);
return (error);
}
for (blkoff = 0; blkoff < blsize;
blkoff += sizeof(struct direntry),
diroff += sizeof(struct direntry)) {
dep = (struct direntry *)(bp->b_data + blkoff);
/*
* If the slot is empty and we are still looking
* for an empty then remember this one. If the
* slot is not empty then check to see if it
* matches what we are looking for. If the slot
* has never been filled with anything, then the
* remainder of the directory has never been used,
* so there is no point in searching it.
*/
if (dep->deName[0] == SLOT_EMPTY ||
dep->deName[0] == SLOT_DELETED) {
/*
* Drop memory of previous long matches
*/
chksum = -1;
mbnambuf_init(&nb);
if (slotcount < wincnt) {
slotcount++;
slotoffset = diroff;
}
if (dep->deName[0] == SLOT_EMPTY) {
brelse(bp);
goto notfound;
}
} else {
/*
* If there wasn't enough space for our winentries,
* forget about the empty space
*/
if (slotcount < wincnt)
slotcount = 0;
/*
* Check for Win95 long filename entry
*/
if (dep->deAttributes == ATTR_WIN95) {
if (pmp->pm_flags & MSDOSFSMNT_SHORTNAME)
continue;
chksum = win2unixfn(&nb,
(struct winentry *)dep, chksum,
pmp);
continue;
}
chksum = winChkName(&nb,
(const u_char *)cnp->cn_nameptr, unlen,
chksum, pmp);
if (chksum == -2) {
chksum = -1;
continue;
}
/*
* Ignore volume labels (anywhere, not just
* the root directory).
*/
if (dep->deAttributes & ATTR_VOLUME) {
chksum = -1;
continue;
}
/*
* Check for a checksum or name match
*/
chksum_ok = (chksum == winChksum(dep->deName));
if (!chksum_ok
&& (!olddos || bcmp(dosfilename, dep->deName, 11))) {
chksum = -1;
continue;
}
#ifdef MSDOSFS_DEBUG
printf("msdosfs_lookup(): match blkoff %d, diroff %d\n",
blkoff, diroff);
#endif
/*
* Remember where this directory
* entry came from for whoever did
* this lookup.
*/
dp->de_fndoffset = diroff;
if (chksum_ok && nameiop == RENAME) {
/*
* Target had correct long name
* directory entries, reuse them
* as needed.
*/
dp->de_fndcnt = wincnt - 1;
} else {
/*
* Long name directory entries
* not present or corrupt, can only
* reuse dos directory entry.
*/
dp->de_fndcnt = 0;
}
goto found;
}
} /* for (blkoff = 0; .... */
/*
* Release the buffer holding the directory cluster just
* searched.
*/
brelse(bp);
} /* for (frcn = 0; ; frcn++) */
notfound:
/*
* We hold no disk buffers at this point.
*/
/*
* Fixup the slot description to point to the place where
* we might put the new DOS direntry (putting the Win95
* long name entries before that)
*/
if (!slotcount) {
slotcount = 1;
slotoffset = diroff;
}
if (wincnt > slotcount)
slotoffset += sizeof(struct direntry) * (wincnt - slotcount);
/*
* If we get here we didn't find the entry we were looking for. But
* that's ok if we are creating or renaming and are at the end of
* the pathname and the directory hasn't been removed.
*/
#ifdef MSDOSFS_DEBUG
printf("msdosfs_lookup(): op %d, refcnt %ld\n",
nameiop, dp->de_refcnt);
printf(" slotcount %d, slotoffset %d\n",
slotcount, slotoffset);
#endif
if ((nameiop == CREATE || nameiop == RENAME) &&
(flags & ISLASTCN) && dp->de_refcnt != 0) {
/*
* Access for write is interpreted as allowing
* creation of files in the directory.
*/
error = VOP_ACCESS(vdp, VWRITE, cnp->cn_cred, cnp->cn_thread);
if (error)
return (error);
/*
* Return an indication of where the new directory
* entry should be put.
*/
dp->de_fndoffset = slotoffset;
dp->de_fndcnt = wincnt - 1;
/*
* We return with the directory locked, so that
* the parameters we set up above will still be
* valid if we actually decide to do a direnter().
* We return ni_vp == NULL to indicate that the entry
* does not currently exist; we leave a pointer to
* the (locked) directory inode in ndp->ni_dvp.
* The pathname buffer is saved so that the name
* can be obtained later.
*
* NB - if the directory is unlocked, then this
* information cannot be used.
*/
cnp->cn_flags |= SAVENAME;
return (EJUSTRETURN);
}
#if 0
/*
* Insert name into cache (as non-existent) if appropriate.
*
* XXX Negative caching is broken for msdosfs because the name
* cache doesn't understand peculiarities such as case insensitivity
* and 8.3 filenames. Hence, it may not invalidate all negative
* entries if a file with this name is later created.
*/
if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
cache_enter(vdp, *vpp, cnp);
#endif
return (ENOENT);
found:
/*
* NOTE: We still have the buffer with matched directory entry at
* this point.
*/
isadir = dep->deAttributes & ATTR_DIRECTORY;
scn = getushort(dep->deStartCluster);
if (FAT32(pmp)) {
scn |= getushort(dep->deHighClust) << 16;
if (scn == pmp->pm_rootdirblk) {
/*
* There should actually be 0 here.
* Just ignore the error.
*/
scn = MSDOSFSROOT;
}
}
if (isadir) {
cluster = scn;
if (cluster == MSDOSFSROOT)
blkoff = MSDOSFSROOT_OFS;
else
blkoff = 0;
} else if (cluster == MSDOSFSROOT)
blkoff = diroff;
/*
* Now release buf to allow deget to read the entry again.
* Reserving it here and giving it to deget could result
* in a deadlock.
*/
brelse(bp);
bp = 0;
foundroot:
/*
* If we entered at foundroot, then we are looking for the . or ..
* entry of the filesystems root directory. isadir and scn were
* setup before jumping here. And, bp is already null.
*/
if (FAT32(pmp) && scn == MSDOSFSROOT)
scn = pmp->pm_rootdirblk;
if (dd_inum != NULL) {
*dd_inum = (uint64_t)pmp->pm_bpcluster * scn + blkoff;
return (0);
}
/*
* If deleting, and at end of pathname, return
* parameters which can be used to remove file.
*/
if (nameiop == DELETE && (flags & ISLASTCN)) {
/*
* Don't allow deleting the root.
*/
if (blkoff == MSDOSFSROOT_OFS)
return (EBUSY);
/*
* Write access to directory required to delete files.
*/
error = VOP_ACCESS(vdp, VWRITE, cnp->cn_cred, cnp->cn_thread);
if (error)
return (error);
/*
* Return pointer to current entry in dp->i_offset.
* Save directory inode pointer in ndp->ni_dvp for dirremove().
*/
if (dp->de_StartCluster == scn && isadir) { /* "." */
VREF(vdp);
*vpp = vdp;
return (0);
}
error = deget(pmp, cluster, blkoff, &tdp);
if (error)
return (error);
*vpp = DETOV(tdp);
return (0);
}
/*
* If rewriting (RENAME), return the inode and the
* information required to rewrite the present directory
* Must get inode of directory entry to verify it's a
* regular file, or empty directory.
*/
if (nameiop == RENAME && (flags & ISLASTCN)) {
if (blkoff == MSDOSFSROOT_OFS)
return (EBUSY);
error = VOP_ACCESS(vdp, VWRITE, cnp->cn_cred, cnp->cn_thread);
if (error)
return (error);
/*
* Careful about locking second inode.
* This can only occur if the target is ".".
*/
if (dp->de_StartCluster == scn && isadir)
return (EISDIR);
if ((error = deget(pmp, cluster, blkoff, &tdp)) != 0)
return (error);
*vpp = DETOV(tdp);
cnp->cn_flags |= SAVENAME;
return (0);
}
/*
* Step through the translation in the name. We do not `vput' the
* directory because we may need it again if a symbolic link
* is relative to the current directory. Instead we save it
* unlocked as "pdp". We must get the target inode before unlocking
* the directory to insure that the inode will not be removed
* before we get it. We prevent deadlock by always fetching
* inodes from the root, moving down the directory tree. Thus
* when following backward pointers ".." we must unlock the
* parent directory before getting the requested directory.
*/
pdp = vdp;
if (flags & ISDOTDOT) {
error = msdosfs_deget_dotdot(pdp, cluster, blkoff, vpp);
if (error) {
*vpp = NULL;
return (error);
}
/*
* Recheck that ".." still points to the inode we
* looked up before pdp lock was dropped.
*/
error = msdosfs_lookup_(pdp, NULL, cnp, &inode1);
if (error) {
vput(*vpp);
*vpp = NULL;
return (error);
}
if (VTODE(*vpp)->de_inode != inode1) {
vput(*vpp);
goto restart;
}
} else if (dp->de_StartCluster == scn && isadir) {
VREF(vdp); /* we want ourself, ie "." */
*vpp = vdp;
} else {
if ((error = deget(pmp, cluster, blkoff, &tdp)) != 0)
return (error);
*vpp = DETOV(tdp);
}
/*
* Insert name into cache if appropriate.
*/
if (cnp->cn_flags & MAKEENTRY)
cache_enter(vdp, *vpp, cnp);
return (0);
}
/*
* internal version with extra arguments to allow accessing resource fork
*/
int
hfs_vget_internal(struct mount *mp, ino_t ino, uint8_t fork,
struct vnode **vpp)
{
struct hfsmount *hmp;
struct hfsnode *hnode;
struct vnode *vp;
hfs_callback_args cbargs;
hfs_cnid_t cnid;
hfs_catalog_keyed_record_t rec;
hfs_catalog_key_t key; /* the search key used to find this file on disk */
dev_t dev;
int error;
#ifdef HFS_DEBUG
printf("vfsop = hfs_vget()\n");
#endif /* HFS_DEBUG */
hnode = NULL;
vp = NULL;
hmp = VFSTOHFS(mp);
dev = hmp->hm_dev;
cnid = (hfs_cnid_t)ino;
if (fork != HFS_RSRCFORK)
fork = HFS_DATAFORK;
retry:
/* Check if this vnode has already been allocated. If so, just return it. */
if ((*vpp = hfs_nhashget(dev, cnid, fork, LK_EXCLUSIVE)) != NULL)
return 0;
/* Allocate a new vnode/inode. */
if ((error = getnewvnode(VT_HFS, mp, hfs_vnodeop_p, &vp)) != 0)
goto error;
MALLOC(hnode, struct hfsnode *, sizeof(struct hfsnode), M_TEMP,
M_WAITOK + M_ZERO);
/*
* If someone beat us to it while sleeping in getnewvnode(),
* push back the freshly allocated vnode we don't need, and return.
*/
mutex_enter(&hfs_hashlock);
if (hfs_nhashget(dev, cnid, fork, 0) != NULL) {
mutex_exit(&hfs_hashlock);
ungetnewvnode(vp);
FREE(hnode, M_TEMP);
goto retry;
}
vp->v_vflag |= VV_LOCKSWORK;
vp->v_data = hnode;
genfs_node_init(vp, &hfs_genfsops);
hnode->h_vnode = vp;
hnode->h_hmp = hmp;
hnode->dummy = 0x1337BABE;
/*
* We need to put this vnode into the hash chain and lock it so that other
* requests for this inode will block if they arrive while we are sleeping
* waiting for old data structures to be purged or for the contents of the
* disk portion of this inode to be read. The hash chain requires the node's
* device and cnid to be known. Since this information was passed in the
* arguments, fill in the appropriate hfsnode fields without reading having
* to read the disk.
*/
hnode->h_dev = dev;
hnode->h_rec.u.cnid = cnid;
hnode->h_fork = fork;
hfs_nhashinsert(hnode);
mutex_exit(&hfs_hashlock);
/*
* Read catalog record from disk.
*/
hfslib_init_cbargs(&cbargs);
if (hfslib_find_catalog_record_with_cnid(&hmp->hm_vol, cnid,
&rec, &key, &cbargs) != 0) {
vput(vp);
error = EBADF;
goto error;
}
memcpy(&hnode->h_rec, &rec, sizeof(hnode->h_rec));
hnode->h_parent = key.parent_cnid;
/* XXX Eventually need to add an "ignore permissions" mount option */
/*
* Now convert some of the catalog record's fields into values that make
* sense on this system.
*/
/* DATE AND TIME */
/*
* Initialize the vnode from the hfsnode, check for aliases.
* Note that the underlying vnode may change.
*/
hfs_vinit(mp, hfs_specop_p, hfs_fifoop_p, &vp);
hnode->h_devvp = hmp->hm_devvp;
VREF(hnode->h_devvp); /* Increment the ref count to the volume's device. */
/* Make sure UVM has allocated enough memory. (?) */
if (hnode->h_rec.u.rec_type == HFS_REC_FILE) {
if (hnode->h_fork == HFS_DATAFORK)
uvm_vnp_setsize(vp,
hnode->h_rec.file.data_fork.logical_size);
else
uvm_vnp_setsize(vp,
hnode->h_rec.file.rsrc_fork.logical_size);
}
else
uvm_vnp_setsize(vp, 0); /* no directly reading directories */
*vpp = vp;
return 0;
error:
*vpp = NULL;
return error;
}
/*
* Routine: macx_swapon
* Function:
* Syscall interface to add a file to backing store
*/
int
macx_swapon(
char *filename,
int flags,
long size,
long priority)
{
struct vnode *vp = 0;
struct nameidata nd, *ndp;
struct proc *p = current_proc();
pager_file_t pf;
register int error;
kern_return_t kr;
mach_port_t backing_store;
memory_object_default_t default_pager;
int i;
boolean_t funnel_state;
struct vattr vattr;
AUDIT_MACH_SYSCALL_ENTER(AUE_SWAPON);
AUDIT_ARG(value, priority);
funnel_state = thread_funnel_set(kernel_flock, TRUE);
ndp = &nd;
if ((error = suser(p->p_ucred, &p->p_acflag)))
goto swapon_bailout;
if(default_pager_init_flag == 0) {
start_def_pager(NULL);
default_pager_init_flag = 1;
}
/*
* Get a vnode for the paging area.
*/
NDINIT(ndp, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNPATH1, UIO_USERSPACE,
filename, p);
if ((error = namei(ndp)))
goto swapon_bailout;
vp = ndp->ni_vp;
if (vp->v_type != VREG) {
error = EINVAL;
VOP_UNLOCK(vp, 0, p);
goto swapon_bailout;
}
UBCINFOCHECK("macx_swapon", vp);
if (error = VOP_GETATTR(vp, &vattr, p->p_ucred, p)) {
VOP_UNLOCK(vp, 0, p);
goto swapon_bailout;
}
if (vattr.va_size < (u_quad_t)size) {
vattr_null(&vattr);
vattr.va_size = (u_quad_t)size;
error = VOP_SETATTR(vp, &vattr, p->p_ucred, p);
if (error) {
VOP_UNLOCK(vp, 0, p);
goto swapon_bailout;
}
}
/* add new backing store to list */
i = 0;
while(bs_port_table[i].vp != 0) {
if(i == MAX_BACKING_STORE)
break;
i++;
}
if(i == MAX_BACKING_STORE) {
error = ENOMEM;
VOP_UNLOCK(vp, 0, p);
goto swapon_bailout;
}
/* remember the vnode. This vnode has namei() reference */
bs_port_table[i].vp = vp;
/*
* Look to see if we are already paging to this file.
*/
/* make certain the copy send of kernel call will work */
default_pager = MEMORY_OBJECT_DEFAULT_NULL;
kr = host_default_memory_manager(host_priv_self(), &default_pager, 0);
if(kr != KERN_SUCCESS) {
error = EAGAIN;
VOP_UNLOCK(vp, 0, p);
bs_port_table[i].vp = 0;
goto swapon_bailout;
}
kr = default_pager_backing_store_create(default_pager,
-1, /* default priority */
0, /* default cluster size */
&backing_store);
memory_object_default_deallocate(default_pager);
if(kr != KERN_SUCCESS) {
error = ENOMEM;
VOP_UNLOCK(vp, 0, p);
bs_port_table[i].vp = 0;
goto swapon_bailout;
}
/*
* NOTE: we are able to supply PAGE_SIZE here instead of
* an actual record size or block number because:
* a: we do not support offsets from the beginning of the
* file (allowing for non page size/record modulo offsets.
* b: because allow paging will be done modulo page size
*/
VOP_UNLOCK(vp, 0, p);
kr = default_pager_add_file(backing_store, vp, PAGE_SIZE,
((int)vattr.va_size)/PAGE_SIZE);
if(kr != KERN_SUCCESS) {
bs_port_table[i].vp = 0;
if(kr == KERN_INVALID_ARGUMENT)
error = EINVAL;
else
error = ENOMEM;
goto swapon_bailout;
}
bs_port_table[i].bs = (void *)backing_store;
error = 0;
if (!ubc_hold(vp))
panic("macx_swapon: hold");
/* Mark this vnode as being used for swapfile */
SET(vp->v_flag, VSWAP);
ubc_setcred(vp, p);
/*
* take an extra reference on the vnode to keep
* vnreclaim() away from this vnode.
*/
VREF(vp);
/* Hold on to the namei reference to the paging file vnode */
vp = 0;
swapon_bailout:
if (vp) {
vrele(vp);
}
(void) thread_funnel_set(kernel_flock, FALSE);
AUDIT_MACH_SYSCALL_EXIT(error);
return(error);
}
void
doit_in_col_major (const char * description,
const int N, const int NRHS,
double A[N][N], double X[NRHS][N], double B[NRHS][N],
double expected_X[NRHS][N])
{
lapack_int Anrows = N;
lapack_int Ancols = N;
lapack_int ldA = Anrows; /* leading dimension of A */
lapack_int ldB = N; /* leading dimension of B */
/* Result of computation: permuted matrix A decomposed in LU. */
double packedLU[Ancols][Anrows];
/* Result of computation: tuple of partial pivot indexes representing
the permutation matrix. */
lapack_int ipiv_dim = MIN(Anrows, Ancols);
lapack_int ipiv[ipiv_dim];
/* Result of computation: error code, zero if success. */
lapack_int info;
/* Data needed to reconstruct A from the results: permutation
vector. */
int perms[Anrows];
/* Data needed to reconstruct A from the results: permutation matrix,
such that A = PLU. */
int Pnrows = Anrows;
int Pncols = Anrows;
int P[Pncols][Pnrows];
/* Lower-triangular factor L. */
lapack_int Lnrows = Anrows;
lapack_int Lncols = MIN(Anrows,Ancols);
lapack_int ldL = Lncols;
double L[Lncols][Lnrows];
/* Upper-triangular factor U. */
lapack_int Unrows = MIN(Anrows,Ancols);
lapack_int Uncols = Ancols;
lapack_int ldU = Uncols;
double U[Uncols][Unrows];
/* Data needed to reconstruct A from the results: product A1 = LU,
such that A = P A1. */
double A1[Ancols][Anrows];
/* Data needed to reconstruct A from the results:
*
* reconstructed_A_ipiv = P A1 = PLU
*
* reconstructed by applying IPIV to A1 backwards.
*/
double reconstructed_A_ipiv[Ancols][Anrows];
/* Data needed to reconstruct A from the results:
*
* reconstructed_A_P = P A1 = PLU
*
* reconstructed by left-multiplying A1 by the permutations matrix P.
*/
double reconstructed_A_P[Ancols][Anrows];
/* Load the original coefficients matrix from A to packedLU. The LU
factorisation result of dgesv() will be stored in packedLU,
overwriting it. */
memcpy(packedLU, A, sizeof(double) * Anrows * Ancols);
/* Load the right-hand side from B to X. The unknowns result of
dgesv() will be stored in X, overwriting it. */
memcpy(X, B, sizeof(double) * N * NRHS);
/* Do it. */
info = LAPACKE_dgesv(LAPACK_COL_MAJOR, N, NRHS,
MREF(packedLU), ldA, VREF(ipiv), MREF(X), ldB);
/* If something went wrong in the function call INFO is non-zero: exit
with failure. */
if (0 != info) {
printf("Error computing solution with row-major operands: INFO=%d.\n", info);
exit(EXIT_FAILURE);
}
/* Reconstructing A from the results. */
{
col_major_PLU_permutation_matrix_from_ipiv (Anrows, Ancols, ipiv, perms, P);
real_col_major_split_LU(Anrows, Ancols, MIN(Anrows, Ancols), packedLU, L, U);
/* Multiply L and U to verify that the result is indeed PA; we need
* CBLAS for this. In general DGEMM does:
*
* \alpha A B + \beta C
*
* where A, B and C are matrices. We need to inspect both the
* header file "cblas.h" and the source file "dgemm.f" for the
* documentation of the parameters; the prototype of "cblas_dgemm()"
* is:
*
* void cblas_dgemm(const enum CBLAS_ORDER Order,
* const enum CBLAS_TRANSPOSE TransA,
* const enum CBLAS_TRANSPOSE TransB,
* const int M, const int N, const int K,
* const double alpha,
* const double *A, const int lda,
* const double *B, const int ldb,
* const double beta,
* double *C, const int ldc);
*
* In our case all the matrices are in col-major order and we the
* representations in the arrays A and B are not transposed, so: M
* is the number of rows of A and C; N is the number of columns of B
* and of columns of C; K is the number of columns of A and rows of
* B. In other words:
*
* A has dimensions M x K
* B has dimensions K x N
* C has dimensions M x N
*
* obviously the product AB has dimensions M x N.
*
* Here we want to do:
*
* A1 = 1.0 L U + 0 A1
*
* where A1 is a matrix whose contents at input are not important,
* and whose contents at output are the result of the operation.
*/
{
double alpha = 1.0;
double beta = 0.0;
cblas_dgemm(CblasColMajor, CblasNoTrans, CblasNoTrans,
Anrows, Ancols, Lncols,
alpha, MREF(L), ldL, MREF(U), ldU, beta, MREF(A1), ldA);
real_col_major_apply_ipiv (Anrows, Ancols, ipiv, BACKWARD_IPIV_APPLICATION,
reconstructed_A_ipiv, A1);
real_col_major_apply_permutation_matrix (Anrows, Ancols, reconstructed_A_P, P, A1);
}
}
printf("Column-major dgesv results, %s:\n", description);
/* Result verification. */
{
compare_real_col_major_result_and_expected_result("computed unknowns",
N, NRHS, X, expected_X);
compare_real_col_major_result_and_expected_result("reconstructed A with IPIV application",
Anrows, Ancols, reconstructed_A_ipiv, A);
compare_real_col_major_result_and_expected_result("reconstructed A with P application",
Anrows, Ancols, reconstructed_A_P, A);
}
/* Results logging. */
{
print_real_col_major_matrix("X, resulting unknowns", N, NRHS, X);
print_real_col_major_matrix("A, original coefficient matrix", Anrows, Ancols, A);
print_col_major_PLU_partial_pivoting_vectors_and_matrix (Anrows, Ancols, ipiv, perms, P);
print_real_col_major_matrix("packedLU representing L and U packed in single matrix",
Anrows, Ancols, packedLU);
print_real_col_major_matrix("L, elements of packedLU", Lnrows, Lncols, L);
print_real_col_major_matrix("U, elements of packedLU", Unrows, Uncols, U);
print_real_col_major_matrix("A1 = LU, it must be such that A = PR", Anrows, Ancols, A1);
print_real_col_major_matrix("reconstructed_A_ipiv = PA1 = PLU, it must be such that A = reconstructed_A",
Anrows, Ancols, reconstructed_A_ipiv);
print_real_col_major_matrix("reconstructed_A_P = PA1 = PLU, it must be such that A = reconstructed_A",
Anrows, Ancols, reconstructed_A_P);
}
}
/*
* lookup. this is incredibly complicated in the
* general case, however for most pseudo-filesystems
* very little needs to be done.
*
* unless you want to get a migraine, just make sure your
* filesystem doesn't do any locking of its own. otherwise
* read and inwardly digest ufs_lookup().
*/
int
procfs_lookup(void *v)
{
struct vop_lookup_args *ap = v;
struct componentname *cnp = ap->a_cnp;
struct vnode **vpp = ap->a_vpp;
struct vnode *dvp = ap->a_dvp;
char *pname = cnp->cn_nameptr;
struct proc *curp = curproc;
const struct proc_target *pt;
struct vnode *fvp;
pid_t pid;
struct pfsnode *pfs;
struct proc *p = NULL;
int i, error, wantpunlock, iscurproc = 0, isself = 0;
*vpp = NULL;
cnp->cn_flags &= ~PDIRUNLOCK;
if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)
return (EROFS);
if (cnp->cn_namelen == 1 && *pname == '.') {
*vpp = dvp;
VREF(dvp);
return (0);
}
wantpunlock = (~cnp->cn_flags & (LOCKPARENT | ISLASTCN));
pfs = VTOPFS(dvp);
switch (pfs->pfs_type) {
case Proot:
if (cnp->cn_flags & ISDOTDOT)
return (EIO);
iscurproc = CNEQ(cnp, "curproc", 7);
isself = CNEQ(cnp, "self", 4);
if (iscurproc || isself) {
error = procfs_allocvp(dvp->v_mount, vpp, 0,
iscurproc ? Pcurproc : Pself);
if ((error == 0) && (wantpunlock)) {
VOP_UNLOCK(dvp, 0, curp);
cnp->cn_flags |= PDIRUNLOCK;
}
return (error);
}
for (i = 0; i < nproc_root_targets; i++) {
pt = &proc_root_targets[i];
if (cnp->cn_namelen == pt->pt_namlen &&
memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
(pt->pt_valid == NULL ||
(*pt->pt_valid)(p, dvp->v_mount)))
break;
}
if (i != nproc_root_targets) {
error = procfs_allocvp(dvp->v_mount, vpp, 0,
pt->pt_pfstype);
if ((error == 0) && (wantpunlock)) {
VOP_UNLOCK(dvp, 0, curp);
cnp->cn_flags |= PDIRUNLOCK;
}
return (error);
}
pid = atopid(pname, cnp->cn_namelen);
if (pid == NO_PID)
break;
p = pfind(pid);
if (p == 0)
break;
error = procfs_allocvp(dvp->v_mount, vpp, pid, Pproc);
if ((error == 0) && wantpunlock) {
VOP_UNLOCK(dvp, 0, curp);
cnp->cn_flags |= PDIRUNLOCK;
}
return (error);
case Pproc:
/*
* do the .. dance. We unlock the directory, and then
* get the root dir. That will automatically return ..
* locked. Then if the caller wanted dvp locked, we
* re-lock.
*/
if (cnp->cn_flags & ISDOTDOT) {
VOP_UNLOCK(dvp, 0, p);
cnp->cn_flags |= PDIRUNLOCK;
error = procfs_root(dvp->v_mount, vpp);
if ((error == 0) && (wantpunlock == 0) &&
((error = vn_lock(dvp, LK_EXCLUSIVE, curp)) == 0))
cnp->cn_flags &= ~PDIRUNLOCK;
return (error);
}
p = pfind(pfs->pfs_pid);
if (p == 0)
break;
for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
if (cnp->cn_namelen == pt->pt_namlen &&
bcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
(pt->pt_valid == NULL ||
(*pt->pt_valid)(p, dvp->v_mount)))
goto found;
}
break;
found:
if (pt->pt_pfstype == Pfile) {
fvp = p->p_textvp;
/* We already checked that it exists. */
VREF(fvp);
vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY, curp);
if (wantpunlock) {
VOP_UNLOCK(dvp, 0, curp);
cnp->cn_flags |= PDIRUNLOCK;
}
*vpp = fvp;
return (0);
}
error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
pt->pt_pfstype);
if ((error == 0) && (wantpunlock)) {
VOP_UNLOCK(dvp, 0, curp);
cnp->cn_flags |= PDIRUNLOCK;
}
return (error);
default:
return (ENOTDIR);
}
return (cnp->cn_nameiop == LOOKUP ? ENOENT : EROFS);
}
