/*
* Create a vnode and in-memory inode. EAGAIN is returned if we race the
* creation of the in-memory inode.
*
* A locked, referenced vnode is returned in *vpp if no error occurs.
*/
int
user_getnewvnode(struct mount *mp, struct vnode **vpp, ino_t ino,
enum vtype vtype)
{
struct user_inode *ip;
struct user_mount *ump;
struct vnode *vp;
int error;
error = getnewvnode(VT_USERFS, mp, vpp, 0, LK_CANRECURSE);
if (error)
return (error);
vp = *vpp;
ip = kmalloc(sizeof(*ip), M_USERFSINODE, M_WAITOK|M_ZERO);
vinitvmio(vp, 0, PAGE_SIZE, 0);
ump = (void *)mp->mnt_data;
/*
* Snap together the new vnode/inode and determine if we have
* raced the tree.
*/
if (userfs_ino_rb_tree_RB_LOOKUP(&ump->rb_root, ino)) {
kfree(ip, M_USERFSINODE);
vp->v_type = VBAD;
vput(*vpp);
return (EAGAIN);
}
ip->inum = ino;
ip->vp = vp;
ip->ump = ump;
vp->v_data = ip;
vp->v_type = vtype;
userfs_ino_rb_tree_RB_INSERT(&ump->rb_root, ip);
return(0);
}
/*
* Get the vnode associated with the given inode, allocating the vnode if
* necessary. The vnode will be returned exclusively locked.
*
* The caller must lock the inode (shared or exclusive).
*
* Great care must be taken to avoid deadlocks and vnode acquisition/reclaim
* races.
*/
struct vnode *
hammer2_igetv(hammer2_inode_t *ip, int *errorp)
{
hammer2_inode_data_t *ipdata;
hammer2_pfsmount_t *pmp;
struct vnode *vp;
ccms_state_t ostate;
pmp = ip->pmp;
KKASSERT(pmp != NULL);
*errorp = 0;
ipdata = &ip->chain->data->ipdata;
for (;;) {
/*
* Attempt to reuse an existing vnode assignment. It is
* possible to race a reclaim so the vget() may fail. The
* inode must be unlocked during the vget() to avoid a
* deadlock against a reclaim.
*/
vp = ip->vp;
if (vp) {
/*
* Inode must be unlocked during the vget() to avoid
* possible deadlocks, but leave the ip ref intact.
*
* vnode is held to prevent destruction during the
* vget(). The vget() can still fail if we lost
* a reclaim race on the vnode.
*/
vhold(vp);
ostate = hammer2_inode_lock_temp_release(ip);
if (vget(vp, LK_EXCLUSIVE)) {
vdrop(vp);
hammer2_inode_lock_temp_restore(ip, ostate);
continue;
}
hammer2_inode_lock_temp_restore(ip, ostate);
vdrop(vp);
/* vp still locked and ref from vget */
if (ip->vp != vp) {
kprintf("hammer2: igetv race %p/%p\n",
ip->vp, vp);
vput(vp);
continue;
}
*errorp = 0;
break;
}
/*
* No vnode exists, allocate a new vnode. Beware of
* allocation races. This function will return an
* exclusively locked and referenced vnode.
*/
*errorp = getnewvnode(VT_HAMMER2, pmp->mp, &vp, 0, 0);
if (*errorp) {
kprintf("hammer2: igetv getnewvnode failed %d\n",
*errorp);
vp = NULL;
break;
}
/*
* Lock the inode and check for an allocation race.
*/
ostate = hammer2_inode_lock_upgrade(ip);
if (ip->vp != NULL) {
vp->v_type = VBAD;
vx_put(vp);
hammer2_inode_lock_downgrade(ip, ostate);
continue;
}
switch (ipdata->type) {
case HAMMER2_OBJTYPE_DIRECTORY:
vp->v_type = VDIR;
break;
case HAMMER2_OBJTYPE_REGFILE:
vp->v_type = VREG;
vinitvmio(vp, ipdata->size,
HAMMER2_LBUFSIZE,
(int)ipdata->size & HAMMER2_LBUFMASK);
break;
case HAMMER2_OBJTYPE_SOFTLINK:
/*
* XXX for now we are using the generic file_read
* and file_write code so we need a buffer cache
* association.
*/
vp->v_type = VLNK;
vinitvmio(vp, ipdata->size,
HAMMER2_LBUFSIZE,
(int)ipdata->size & HAMMER2_LBUFMASK);
break;
case HAMMER2_OBJTYPE_CDEV:
vp->v_type = VCHR;
/* fall through */
case HAMMER2_OBJTYPE_BDEV:
vp->v_ops = &pmp->mp->mnt_vn_spec_ops;
if (ipdata->type != HAMMER2_OBJTYPE_CDEV)
vp->v_type = VBLK;
addaliasu(vp, ipdata->rmajor, ipdata->rminor);
break;
case HAMMER2_OBJTYPE_FIFO:
vp->v_type = VFIFO;
vp->v_ops = &pmp->mp->mnt_vn_fifo_ops;
break;
default:
panic("hammer2: unhandled objtype %d", ipdata->type);
break;
}
if (ip == pmp->iroot)
vsetflags(vp, VROOT);
vp->v_data = ip;
ip->vp = vp;
hammer2_inode_ref(ip); /* vp association */
hammer2_inode_lock_downgrade(ip, ostate);
break;
}
/*
* Return non-NULL vp and *errorp == 0, or NULL vp and *errorp != 0.
*/
if (hammer2_debug & 0x0002) {
kprintf("igetv vp %p refs 0x%08x aux 0x%08x\n",
vp, vp->v_refcnt, vp->v_auxrefs);
}
return (vp);
}
int
ntfs_vgetex(struct mount *mp, ino_t ino, u_int32_t attrtype, char *attrname,
u_long lkflags, u_long flags, struct thread *td,
struct vnode **vpp)
{
int error;
struct ntfsmount *ntmp;
struct ntnode *ip;
struct fnode *fp;
struct vnode *vp;
enum vtype f_type;
dprintf(("ntfs_vgetex: ino: %ju, attr: 0x%x:%s, lkf: 0x%lx, f: 0x%lx\n",
(uintmax_t) ino, attrtype, attrname?attrname:"", lkflags, flags));
ntmp = VFSTONTFS(mp);
*vpp = NULL;
/* Get ntnode */
error = ntfs_ntlookup(ntmp, ino, &ip);
if (error) {
kprintf("ntfs_vget: ntfs_ntget failed\n");
return (error);
}
/* It may be not initialized fully, so force load it */
if (!(flags & VG_DONTLOADIN) && !(ip->i_flag & IN_LOADED)) {
error = ntfs_loadntnode(ntmp, ip);
if(error) {
kprintf("ntfs_vget: CAN'T LOAD ATTRIBUTES FOR INO: %"PRId64"\n",
ip->i_number);
ntfs_ntput(ip);
return (error);
}
}
error = ntfs_fget(ntmp, ip, attrtype, attrname, &fp);
if (error) {
kprintf("ntfs_vget: ntfs_fget failed\n");
ntfs_ntput(ip);
return (error);
}
f_type = VINT;
if (!(flags & VG_DONTVALIDFN) && !(fp->f_flag & FN_VALID)) {
if ((ip->i_frflag & NTFS_FRFLAG_DIR) &&
(fp->f_attrtype == NTFS_A_DATA && fp->f_attrname == NULL)) {
f_type = VDIR;
} else if (flags & VG_EXT) {
f_type = VINT;
fp->f_size = fp->f_allocated = 0;
} else {
f_type = VREG;
error = ntfs_filesize(ntmp, fp,
&fp->f_size, &fp->f_allocated);
if (error) {
ntfs_ntput(ip);
return (error);
}
}
fp->f_flag |= FN_VALID;
}
if (FTOV(fp)) {
VGET(FTOV(fp), lkflags);
*vpp = FTOV(fp);
ntfs_ntput(ip);
return (0);
}
error = getnewvnode(VT_NTFS, ntmp->ntm_mountp, &vp, VLKTIMEOUT, 0);
if(error) {
ntfs_frele(fp);
ntfs_ntput(ip);
return (error);
}
dprintf(("ntfs_vget: vnode: %p for ntnode: %ju\n", vp, (uintmax_t)ino));
fp->f_vp = vp;
vp->v_data = fp;
vp->v_type = f_type;
if (ino == NTFS_ROOTINO)
vsetflags(vp, VROOT);
/*
* Normal files use the buffer cache
*/
if (f_type == VREG)
vinitvmio(vp, fp->f_size, PAGE_SIZE, -1);
ntfs_ntput(ip);
KKASSERT(lkflags & LK_TYPE_MASK);
/* XXX leave vnode locked exclusively from getnewvnode */
*vpp = vp;
return (0);
}
static int
devfs_spec_open(struct vop_open_args *ap)
{
struct vnode *vp = ap->a_vp;
struct vnode *orig_vp = NULL;
struct devfs_node *node = DEVFS_NODE(vp);
struct devfs_node *newnode;
cdev_t dev, ndev = NULL;
int error = 0;
if (node) {
if (node->d_dev == NULL)
return ENXIO;
if (!devfs_node_is_accessible(node))
return ENOENT;
}
if ((dev = vp->v_rdev) == NULL)
return ENXIO;
vn_lock(vp, LK_UPGRADE | LK_RETRY);
if (node && ap->a_fp) {
devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_open: -1.1-\n");
lockmgr(&devfs_lock, LK_EXCLUSIVE);
ndev = devfs_clone(dev, node->d_dir.d_name,
node->d_dir.d_namlen,
ap->a_mode, ap->a_cred);
if (ndev != NULL) {
newnode = devfs_create_device_node(
DEVFS_MNTDATA(vp->v_mount)->root_node,
ndev, NULL, NULL);
/* XXX: possibly destroy device if this happens */
if (newnode != NULL) {
dev = ndev;
devfs_link_dev(dev);
devfs_debug(DEVFS_DEBUG_DEBUG,
"parent here is: %s, node is: |%s|\n",
((node->parent->node_type == Nroot) ?
"ROOT!" : node->parent->d_dir.d_name),
newnode->d_dir.d_name);
devfs_debug(DEVFS_DEBUG_DEBUG,
"test: %s\n",
((struct devfs_node *)(TAILQ_LAST(DEVFS_DENODE_HEAD(node->parent), devfs_node_head)))->d_dir.d_name);
/*
* orig_vp is set to the original vp if we cloned.
*/
/* node->flags |= DEVFS_CLONED; */
devfs_allocv(&vp, newnode);
orig_vp = ap->a_vp;
ap->a_vp = vp;
}
}
lockmgr(&devfs_lock, LK_RELEASE);
}
devfs_debug(DEVFS_DEBUG_DEBUG,
"devfs_spec_open() called on %s! \n",
dev->si_name);
/*
* Make this field valid before any I/O in ->d_open
*/
if (!dev->si_iosize_max)
/* XXX: old DFLTPHYS == 64KB dependency */
dev->si_iosize_max = min(MAXPHYS,64*1024);
if (dev_dflags(dev) & D_TTY)
vsetflags(vp, VISTTY);
/*
* Open underlying device
*/
vn_unlock(vp);
error = dev_dopen(dev, ap->a_mode, S_IFCHR, ap->a_cred, ap->a_fp);
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
/*
* Clean up any cloned vp if we error out.
*/
if (error) {
if (orig_vp) {
vput(vp);
ap->a_vp = orig_vp;
/* orig_vp = NULL; */
}
return error;
}
/*
* This checks if the disk device is going to be opened for writing.
* It will be only allowed in the cases where securelevel permits it
* and it's not mounted R/W.
*/
if ((dev_dflags(dev) & D_DISK) && (ap->a_mode & FWRITE) &&
(ap->a_cred != FSCRED)) {
/* Very secure mode. No open for writing allowed */
if (securelevel >= 2)
return EPERM;
/*
* If it is mounted R/W, do not allow to open for writing.
* In the case it's mounted read-only but securelevel
* is >= 1, then do not allow opening for writing either.
*/
if (vfs_mountedon(vp)) {
if (!(dev->si_mountpoint->mnt_flag & MNT_RDONLY))
return EBUSY;
else if (securelevel >= 1)
return EPERM;
}
}
if (dev_dflags(dev) & D_TTY) {
if (dev->si_tty) {
struct tty *tp;
tp = dev->si_tty;
if (!tp->t_stop) {
devfs_debug(DEVFS_DEBUG_DEBUG,
"devfs: no t_stop\n");
tp->t_stop = nottystop;
}
}
}
if (vn_isdisk(vp, NULL)) {
if (!dev->si_bsize_phys)
dev->si_bsize_phys = DEV_BSIZE;
vinitvmio(vp, IDX_TO_OFF(INT_MAX), PAGE_SIZE, -1);
}
vop_stdopen(ap);
#if 0
if (node)
nanotime(&node->atime);
#endif
/*
* If we replaced the vp the vop_stdopen() call will have loaded
* it into fp->f_data and vref()d the vp, giving us two refs. So
* instead of just unlocking it here we have to vput() it.
*/
if (orig_vp)
vput(vp);
/* Ugly pty magic, to make pty devices appear once they are opened */
if (node && (node->flags & DEVFS_PTY) == DEVFS_PTY)
node->flags &= ~DEVFS_INVISIBLE;
if (ap->a_fp) {
KKASSERT(ap->a_fp->f_type == DTYPE_VNODE);
KKASSERT((ap->a_fp->f_flag & FMASK) == (ap->a_mode & FMASK));
ap->a_fp->f_ops = &devfs_dev_fileops;
KKASSERT(ap->a_fp->f_data == (void *)vp);
}
return 0;
}
/*
* Allocates a new vnode for the node node or returns a new reference to
* an existing one if the node had already a vnode referencing it. The
* resulting locked vnode is returned in *vpp.
*
* Returns zero on success or an appropriate error code on failure.
*
* The caller must ensure that node cannot go away (usually by holding
* the related directory entry).
*
* If dnode is non-NULL this routine avoids deadlocking against it but
* can return EAGAIN. Caller must try again. The dnode lock will cycle
* in this case, it remains locked on return in all cases. dnode must
* be shared-locked.
*/
int
tmpfs_alloc_vp(struct mount *mp,
struct tmpfs_node *dnode, struct tmpfs_node *node, int lkflag,
struct vnode **vpp)
{
int error = 0;
struct vnode *vp;
loop:
/*
* Interlocked extraction from node. This can race many things.
* We have to get a soft reference on the vnode while we hold
* the node locked, then acquire it properly and check for races.
*/
TMPFS_NODE_LOCK(node);
if ((vp = node->tn_vnode) != NULL) {
KKASSERT((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
vhold(vp);
TMPFS_NODE_UNLOCK(node);
if (dnode) {
/*
* Special-case handling to avoid deadlocking against
* dnode. This case has been validated and occurs
* every so often during synth builds.
*/
if (vget(vp, (lkflag & ~LK_RETRY) |
LK_NOWAIT |
LK_EXCLUSIVE) != 0) {
TMPFS_NODE_UNLOCK(dnode);
if (vget(vp, (lkflag & ~LK_RETRY) |
LK_SLEEPFAIL |
LK_EXCLUSIVE) == 0) {
vn_unlock(vp);
}
vdrop(vp);
TMPFS_NODE_LOCK_SH(dnode);
return EAGAIN;
}
} else {
/*
* Normal path
*/
if (vget(vp, lkflag | LK_EXCLUSIVE) != 0) {
vdrop(vp);
goto loop;
}
}
if (node->tn_vnode != vp) {
vput(vp);
vdrop(vp);
goto loop;
}
vdrop(vp);
goto out;
}
/* vp is NULL */
/*
* This should never happen.
*/
if (node->tn_vpstate & TMPFS_VNODE_DOOMED) {
TMPFS_NODE_UNLOCK(node);
error = ENOENT;
goto out;
}
/*
* Interlock against other calls to tmpfs_alloc_vp() trying to
* allocate and assign a vp to node.
*/
if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
node->tn_vpstate |= TMPFS_VNODE_WANT;
error = tsleep(&node->tn_vpstate, PINTERLOCKED | PCATCH,
"tmpfs_alloc_vp", 0);
TMPFS_NODE_UNLOCK(node);
if (error)
return error;
goto loop;
}
node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
TMPFS_NODE_UNLOCK(node);
/*
* Allocate a new vnode (may block). The ALLOCATING flag should
* prevent a race against someone else assigning node->tn_vnode.
*/
error = getnewvnode(VT_TMPFS, mp, &vp, VLKTIMEOUT, LK_CANRECURSE);
if (error != 0)
goto unlock;
KKASSERT(node->tn_vnode == NULL);
KKASSERT(vp != NULL);
vp->v_data = node;
vp->v_type = node->tn_type;
/* Type-specific initialization. */
switch (node->tn_type) {
case VBLK:
/* FALLTHROUGH */
case VCHR:
/* FALLTHROUGH */
case VSOCK:
break;
case VREG:
/*
* VMIO is mandatory. Tmpfs also supports KVABIO
* for its tmpfs_strategy().
*/
vsetflags(vp, VKVABIO);
vinitvmio(vp, node->tn_size, TMPFS_BLKSIZE, -1);
break;
case VLNK:
break;
case VFIFO:
vp->v_ops = &mp->mnt_vn_fifo_ops;
break;
case VDIR:
break;
default:
panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
}
unlock:
TMPFS_NODE_LOCK(node);
KKASSERT(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
node->tn_vnode = vp;
if (node->tn_vpstate & TMPFS_VNODE_WANT) {
node->tn_vpstate &= ~TMPFS_VNODE_WANT;
TMPFS_NODE_UNLOCK(node);
wakeup(&node->tn_vpstate);
} else {
TMPFS_NODE_UNLOCK(node);
}
out:
*vpp = vp;
KKASSERT(IFF(error == 0, *vpp != NULL && vn_islocked(*vpp)));
return error;
}
