/*
* Destroy a node and all its descendants. If the node to be destroyed
* has a parent, the parent's mutex must be held.
*/
int
pfs_destroy(struct pfs_node *pn)
{
struct pfs_node *iter;
KASSERT(pn != NULL,
("%s(): node is NULL", __func__));
KASSERT(pn->pn_info != NULL,
("%s(): node has no pn_info", __func__));
if (pn->pn_parent)
pfs_detach_node(pn);
/* destroy children */
if (pn->pn_type == pfstype_dir ||
pn->pn_type == pfstype_procdir ||
pn->pn_type == pfstype_root) {
pfs_lock(pn);
while (pn->pn_nodes != NULL) {
iter = pn->pn_nodes;
pn->pn_nodes = iter->pn_next;
iter->pn_parent = NULL;
pfs_unlock(pn);
pfs_destroy(iter);
pfs_lock(pn);
}
pfs_unlock(pn);
}
/* revoke vnodes and fileno */
pfs_purge(pn);
/* callback to free any private resources */
if (pn->pn_destroy != NULL)
pn_destroy(pn);
/* destroy the node */
pfs_fileno_free(pn);
mtx_destroy(&pn->pn_mutex);
free(pn, M_PFSNODES);
return (0);
}
/*
* Locate a node by name
*/
struct pfs_node *
pfs_find_node(struct pfs_node *parent, const char *name)
{
struct pfs_node *pn;
pfs_lock(parent);
for (pn = parent->pn_nodes; pn != NULL; pn = pn->pn_next)
if (strcmp(pn->pn_name, name) == 0)
break;
pfs_unlock(parent);
return (pn);
}
/*
* Add a node to a directory
*/
static void
pfs_add_node(struct pfs_node *parent, struct pfs_node *pn)
{
#ifdef INVARIANTS
struct pfs_node *iter;
#endif
KASSERT(parent != NULL,
("%s(): parent is NULL", __func__));
KASSERT(pn->pn_parent == NULL,
("%s(): node already has a parent", __func__));
KASSERT(parent->pn_info != NULL,
("%s(): parent has no pn_info", __func__));
KASSERT(parent->pn_type == pfstype_dir ||
parent->pn_type == pfstype_procdir ||
parent->pn_type == pfstype_root,
("%s(): parent is not a directory", __func__));
#ifdef INVARIANTS
/* XXX no locking! */
if (pn->pn_type == pfstype_procdir)
for (iter = parent; iter != NULL; iter = iter->pn_parent)
KASSERT(iter->pn_type != pfstype_procdir,
("%s(): nested process directories", __func__));
for (iter = parent->pn_nodes; iter != NULL; iter = iter->pn_next) {
KASSERT(strcmp(pn->pn_name, iter->pn_name) != 0,
("%s(): homonymous siblings", __func__));
if (pn->pn_type == pfstype_procdir)
KASSERT(iter->pn_type != pfstype_procdir,
("%s(): sibling process directories", __func__));
}
#endif
pn->pn_parent = parent;
pfs_fileno_alloc(pn);
pfs_lock(parent);
pn->pn_next = parent->pn_nodes;
if ((parent->pn_flags & PFS_PROCDEP) != 0)
pn->pn_flags |= PFS_PROCDEP;
parent->pn_nodes = pn;
pfs_unlock(parent);
}
/*
* Detach a node from its aprent
*/
static void
pfs_detach_node(struct pfs_node *pn)
{
struct pfs_node *parent = pn->pn_parent;
struct pfs_node **iter;
KASSERT(parent != NULL, ("%s(): node has no parent", __func__));
KASSERT(parent->pn_info == pn->pn_info,
("%s(): parent has different pn_info", __func__));
pfs_lock(parent);
iter = &parent->pn_nodes;
while (*iter != NULL) {
if (*iter == pn) {
*iter = pn->pn_next;
break;
}
iter = &(*iter)->pn_next;
}
pn->pn_parent = NULL;
pfs_unlock(parent);
}
/*
* Return directory entries.
*/
static int
pfs_readdir(struct vop_readdir_args *va)
{
struct vnode *vn = va->a_vp;
struct pfs_vdata *pvd = vn->v_data;
struct pfs_node *pd = pvd->pvd_pn;
pid_t pid = pvd->pvd_pid;
struct proc *p, *proc;
struct pfs_node *pn;
struct dirent *entry;
struct uio *uio;
off_t offset;
int error, i, resid;
char *buf, *ent;
KASSERT(pd->pn_info == vn->v_mount->mnt_data,
("%s(): pn_info does not match mountpoint", __func__));
PFS_TRACE(("%s pid %lu", pd->pn_name, (unsigned long)pid));
pfs_assert_not_owned(pd);
if (vn->v_type != VDIR)
PFS_RETURN (ENOTDIR);
KASSERT_PN_IS_DIR(pd);
uio = va->a_uio;
/* only allow reading entire entries */
offset = uio->uio_offset;
resid = uio->uio_resid;
if (offset < 0 || offset % PFS_DELEN != 0 ||
(resid && resid < PFS_DELEN))
PFS_RETURN (EINVAL);
if (resid == 0)
PFS_RETURN (0);
/* can't do this while holding the proc lock... */
buf = malloc(resid, M_IOV, M_WAITOK | M_ZERO);
sx_slock(&allproc_lock);
pfs_lock(pd);
/* check if the directory is visible to the caller */
if (!pfs_visible(curthread, pd, pid, &proc)) {
sx_sunlock(&allproc_lock);
pfs_unlock(pd);
free(buf, M_IOV);
PFS_RETURN (ENOENT);
}
KASSERT(pid == NO_PID || proc != NULL,
("%s(): no process for pid %lu", __func__, (unsigned long)pid));
/* skip unwanted entries */
for (pn = NULL, p = NULL; offset > 0; offset -= PFS_DELEN) {
if (pfs_iterate(curthread, proc, pd, &pn, &p) == -1) {
/* nothing left... */
if (proc != NULL)
PROC_UNLOCK(proc);
pfs_unlock(pd);
sx_sunlock(&allproc_lock);
free(buf, M_IOV);
PFS_RETURN (0);
}
}
/* fill in entries */
ent = buf;
while (pfs_iterate(curthread, proc, pd, &pn, &p) != -1 &&
resid >= PFS_DELEN) {
entry = (struct dirent *)ent;
entry->d_reclen = PFS_DELEN;
entry->d_fileno = pn_fileno(pn, pid);
/* PFS_DELEN was picked to fit PFS_NAMLEN */
for (i = 0; i < PFS_NAMELEN - 1 && pn->pn_name[i] != '\0'; ++i)
entry->d_name[i] = pn->pn_name[i];
entry->d_name[i] = 0;
entry->d_namlen = i;
switch (pn->pn_type) {
case pfstype_procdir:
KASSERT(p != NULL,
("reached procdir node with p == NULL"));
entry->d_namlen = snprintf(entry->d_name,
PFS_NAMELEN, "%d", p->p_pid);
/* fall through */
case pfstype_root:
case pfstype_dir:
case pfstype_this:
case pfstype_parent:
entry->d_type = DT_DIR;
break;
case pfstype_file:
entry->d_type = DT_REG;
break;
case pfstype_symlink:
entry->d_type = DT_LNK;
break;
default:
panic("%s has unexpected node type: %d", pn->pn_name, pn->pn_type);
}
PFS_TRACE(("%s", entry->d_name));
offset += PFS_DELEN;
resid -= PFS_DELEN;
ent += PFS_DELEN;
}
if (proc != NULL)
PROC_UNLOCK(proc);
pfs_unlock(pd);
sx_sunlock(&allproc_lock);
PFS_TRACE(("%zd bytes", ent - buf));
error = uiomove(buf, ent - buf, uio);
free(buf, M_IOV);
PFS_RETURN (error);
}
/*
* 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);
}
/*
* Get file attributes
*/
static int
pfs_getattr(struct vop_getattr_args *va)
{
struct vnode *vn = va->a_vp;
struct pfs_vdata *pvd = vn->v_data;
struct pfs_node *pn = pvd->pvd_pn;
struct vattr *vap = va->a_vap;
struct proc *proc;
int error = 0;
PFS_TRACE(("%s", pn->pn_name));
pfs_assert_not_owned(pn);
if (!pfs_visible(curthread, pn, pvd->pvd_pid, &proc))
PFS_RETURN (ENOENT);
vap->va_type = vn->v_type;
vap->va_fileid = pn_fileno(pn, pvd->pvd_pid);
vap->va_flags = 0;
vap->va_blocksize = PAGE_SIZE;
vap->va_bytes = vap->va_size = 0;
vap->va_filerev = 0;
vap->va_fsid = vn->v_mount->mnt_stat.f_fsid.val[0];
vap->va_nlink = 1;
nanotime(&vap->va_ctime);
vap->va_atime = vap->va_mtime = vap->va_ctime;
switch (pn->pn_type) {
case pfstype_procdir:
case pfstype_root:
case pfstype_dir:
#if 0
pfs_lock(pn);
/* compute link count */
pfs_unlock(pn);
#endif
vap->va_mode = 0555;
break;
case pfstype_file:
case pfstype_symlink:
vap->va_mode = 0444;
break;
default:
printf("shouldn't be here!\n");
vap->va_mode = 0;
break;
}
if (proc != NULL) {
vap->va_uid = proc->p_ucred->cr_ruid;
vap->va_gid = proc->p_ucred->cr_rgid;
} else {
vap->va_uid = 0;
vap->va_gid = 0;
}
if (pn->pn_attr != NULL)
error = pn_attr(va->a_td, proc, pn, vap);
if(proc != NULL)
PROC_UNLOCK(proc);
PFS_RETURN (error);
}
/*
* Convert a vnode to its component name
*/
static int
pfs_vptocnp(struct vop_vptocnp_args *ap)
{
struct vnode *vp = ap->a_vp;
struct vnode **dvp = ap->a_vpp;
struct pfs_vdata *pvd = vp->v_data;
struct pfs_node *pd = pvd->pvd_pn;
struct pfs_node *pn;
struct mount *mp;
char *buf = ap->a_buf;
int *buflen = ap->a_buflen;
char pidbuf[PFS_NAMELEN];
pid_t pid = pvd->pvd_pid;
int len, i, error, locked;
i = *buflen;
error = 0;
pfs_lock(pd);
if (vp->v_type == VDIR && pd->pn_type == pfstype_root) {
*dvp = vp;
vhold(*dvp);
pfs_unlock(pd);
PFS_RETURN (0);
} else if (vp->v_type == VDIR && pd->pn_type == pfstype_procdir) {
len = snprintf(pidbuf, sizeof(pidbuf), "%d", pid);
i -= len;
if (i < 0) {
error = ENOMEM;
goto failed;
}
bcopy(pidbuf, buf + i, len);
} else {
len = strlen(pd->pn_name);
i -= len;
if (i < 0) {
error = ENOMEM;
goto failed;
}
bcopy(pd->pn_name, buf + i, len);
}
pn = pd->pn_parent;
pfs_unlock(pd);
mp = vp->v_mount;
error = vfs_busy(mp, 0);
if (error)
return (error);
/*
* vp is held by caller.
*/
locked = VOP_ISLOCKED(vp);
VOP_UNLOCK(vp, 0);
error = pfs_vncache_alloc(mp, dvp, pn, pid);
if (error) {
vn_lock(vp, locked | LK_RETRY);
vfs_unbusy(mp);
PFS_RETURN(error);
}
*buflen = i;
vhold(*dvp);
vput(*dvp);
vn_lock(vp, locked | LK_RETRY);
vfs_unbusy(mp);
PFS_RETURN (0);
failed:
pfs_unlock(pd);
PFS_RETURN(error);
}
