kern/src/ns/chan.c

/* Copyright © 1994-1999 Lucent Technologies Inc.  All rights reserved.
 * Portions Copyright © 1997-1999 Vita Nuova Limited
 * Portions Copyright © 2000-2007 Vita Nuova Holdings Limited
 *                                (www.vitanuova.com)
 * Revisions Copyright © 2000-2007 Lucent Technologies Inc. and others
 *
 * Modified for the Akaros operating system:
 * Copyright (c) 2013-2014 The Regents of the University of California
 * Copyright (c) 2013-2015 Google Inc.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a copy
 * of this software and associated documentation files (the "Software"), to deal
 * in the Software without restriction, including without limitation the rights
 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
 * copies of the Software, and to permit persons to whom the Software is
 * furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice shall be included in
 * all copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE. */

#include <slab.h>
#include <kmalloc.h>
#include <kref.h>
#include <string.h>
#include <stdio.h>
#include <assert.h>
#include <error.h>
#include <cpio.h>
#include <pmap.h>
#include <smp.h>
#include <syscall.h>

struct chan *kern_slash;

char *channame(struct chan *c)
{	/* DEBUGGING */
	if (c == NULL)
		return "<NULL chan>";
	if (c->name == NULL)
		return "<NULL name>";
	if (c->name->s == NULL)
		return "<NULL name.s>";
	return c->name->s;
}

enum {
	CNAMESLOP = 20
};

struct {
	spinlock_t lock;
	int fid;
	struct chan *free;
	struct chan *list;
} chanalloc;

typedef struct Elemlist Elemlist;

struct Elemlist {
	char *name;		/* copy of name, so '/' can be overwritten */
	int ARRAY_SIZEs;
	char **elems;
	int *off;
	int mustbedir;
};

struct walk_helper {
	bool can_mount;
	bool no_follow;
	unsigned int nr_loops;
};
#define WALK_MAX_NR_LOOPS 8

static struct chan *walk_symlink(struct chan *symlink, struct walk_helper *wh,
                                 unsigned int nr_names_left);

#define SEP(c) ((c) == 0 || (c) == '/')
void cleancname(struct cname *);

int isdotdot(char *p)
{
	return p[0] == '.' && p[1] == '.' && p[2] == '\0';
}

int emptystr(char *s)
{
	if (s == NULL)
		return 1;
	if (s[0] == '\0')
		return 1;
	return 0;
}

/*
 * Atomically replace *p with copy of s
 */
void kstrdup(char **p, char *s)
{
	int n;
	char *t, *prev;

	n = strlen(s) + 1;
	/* if it's a user, we can wait for memory; if not, something's very
	 * wrong */
	if (current) {
		t = kzmalloc(n, 0);
	} else {
		t = kzmalloc(n, 0);
		if (t == NULL)
			panic("kstrdup: no memory");
	}
	memmove(t, s, n);

	prev = atomic_swap_ptr((void**)p, t);
	kfree(prev);
}

static void chan_release(struct kref *kref)
{
	struct chan *c = container_of(kref, struct chan, ref);
	ERRSTACK(1);

	/* We can be called from RCU callbacks, but close methods can block.  In
	 * those cases, and any other context that cannot block, we need to
	 * defer our work to a kernel message. */
	if (!can_block(this_pcpui_ptr())) {
		run_as_rkm(chan_release, kref);
		return;
	}
	/* this style discards the error from close().  picture it as
	 * if (waserror()) { } else { close(); } chanfree_no_matter_what();  */
	if (!waserror()) {
		printd("releasing chan %p, type %d\n", c, c->type);
		/* -1 means there is no dev yet.  wants a noop for close() */
		if (c->type != -1)
			devtab[c->type].close(c);
	}
	/* need to poperror regardless of whether we error'd or not */
	poperror();
	/* and chan free no matter what */
	chanfree(c);
}

struct chan *newchan(void)
{
	struct chan *c;

	spin_lock(&(&chanalloc)->lock);
	c = chanalloc.free;
	if (c != 0)
		chanalloc.free = c->next;
	spin_unlock(&(&chanalloc)->lock);

	if (c == NULL) {
		c = kzmalloc(sizeof(struct chan), 0);
		spin_lock(&(&chanalloc)->lock);
		c->fid = ++chanalloc.fid;
		c->link = chanalloc.list;
		chanalloc.list = c;
		spin_unlock(&(&chanalloc)->lock);
		spinlock_init(&c->lock);
		qlock_init(&c->umqlock);
	}

	/* if you get an error before associating with a dev, cclose skips
	 * calling the dev's close */
	c->type = -1;
	c->flag = 0;
	kref_init(&c->ref, chan_release, 1);
	c->dev = 0;
	c->offset = 0;
	c->iounit = 0;
	c->umh = 0;
	c->uri = 0;
	c->dri = 0;
	c->aux = 0;
	c->mchan = 0;
	c->mcp = 0;
	c->mux = 0;
	c->mqid.path = 0;
	c->mqid.vers = 0;
	c->mqid.type = 0;
	c->name = 0;
	c->buf = NULL;
	c->mountpoint = NULL;
	return c;
}

static void __cname_release(struct kref *kref)
{
	struct cname *n = container_of(kref, struct cname, ref);
	kfree(n->s);
	kfree(n);
}

struct cname *newcname(char *s)
{
	struct cname *n;
	int i;

	n = kzmalloc(sizeof(*n), 0);
	i = strlen(s);
	n->len = i;
	n->alen = i + CNAMESLOP;
	n->s = kzmalloc(n->alen, 0);
	memmove(n->s, s, i + 1);
	kref_init(&n->ref, __cname_release, 1);
	return n;
}

void cnameclose(struct cname *n)
{
	if (n == NULL)
		return;
	kref_put(&n->ref);
}

struct cname *addelem(struct cname *n, char *s)
{
	int i, a;
	char *t;
	struct cname *new;

	if (s[0] == '.' && s[1] == '\0')
		return n;

	if (kref_refcnt(&n->ref) > 1) {
		/* copy on write */
		new = newcname(n->s);
		cnameclose(n);
		n = new;
	}

	i = strlen(s);
	if (n->len + 1 + i + 1 > n->alen) {
		a = n->len + 1 + i + 1 + CNAMESLOP;
		t = kzmalloc(a, 0);
		memmove(t, n->s, n->len + 1);
		kfree(n->s);
		n->s = t;
		n->alen = a;
	}
	/* don't insert extra slash if one is present */
	if (n->len > 0 && n->s[n->len - 1] != '/' && s[0] != '/')
		n->s[n->len++] = '/';
	memmove(n->s + n->len, s, i + 1);
	n->len += i;
	if (isdotdot(s))
		cleancname(n);
	return n;
}

void chanfree(struct chan *c)
{
	c->flag = CFREE;

	if (c->umh != NULL) {
		putmhead(c->umh);
		c->umh = NULL;
	}
	if (c->umc != NULL) {
		cclose(c->umc);
		c->umc = NULL;
	}
	if (c->mux != NULL) {
		//
		muxclose(c->mux);
		c->mux = NULL;
	}
	if (c->mchan != NULL) {
		cclose(c->mchan);
		c->mchan = NULL;
	}

	cnameclose(c->name);
	if (c->buf)
		kfree(c->buf);
	c->buf = NULL;
	c->bufused = 0;
	c->ateof = 0;

	spin_lock(&(&chanalloc)->lock);
	c->next = chanalloc.free;
	chanalloc.free = c;
	spin_unlock(&(&chanalloc)->lock);
}

void cclose(struct chan *c)
{
	if (c == 0)
		return;

	if (c->flag & CFREE)
		panic("double cclose %p.  maybe kfunc channame and hexdump", c);

	kref_put(&c->ref);
}

/* convenience wrapper for interposition.  if you do use this, don't forget
 * about the kref_get_not_zero in plan9setup() */
void chan_incref(struct chan *c)
{
	kref_get(&c->ref, 1);
}

/*
 * Make sure we have the only copy of c.  (Copy on write.)
 */
struct chan *cunique(struct chan *c)
{
	struct chan *nc;

	if (kref_refcnt(&c->ref) != 1) {
		nc = cclone(c);
		cclose(c);
		c = nc;
	}

	return c;
}

int eqqid(struct qid a, struct qid b)
{
	return a.path == b.path && a.vers == b.vers;
}

int eqchan(struct chan *a, struct chan *b, int pathonly)
{
	if (a->qid.path != b->qid.path)
		return 0;
	if (!pathonly && a->qid.vers != b->qid.vers)
		return 0;
	if (a->type != b->type)
		return 0;
	if (a->dev != b->dev)
		return 0;
	return 1;
}

int eqchantdqid(struct chan *a, int type, int dev, struct qid qid, int pathonly)
{
	if (a->qid.path != qid.path)
		return 0;
	if (!pathonly && a->qid.vers != qid.vers)
		return 0;
	if (a->type != type)
		return 0;
	if (a->dev != dev)
		return 0;
	return 1;
}

static void mh_release(struct kref *kref)
{
	struct mhead *mh = container_of(kref, struct mhead, ref);

	mh->mount = (struct mount *)0xCafeBeef;
	kfree(mh);
}

struct mhead *newmhead(struct chan *from)
{
	struct mhead *mh;

	mh = kzmalloc(sizeof(struct mhead), 0);
	kref_init(&mh->ref, mh_release, 1);
	rwinit(&mh->lock);
	mh->from = from;
	chan_incref(from);

/*
	n = from->name->len;
	if(n >= sizeof(mh->fromname))
		n = sizeof(mh->fromname)-1;
	memmove(mh->fromname, from->name->s, n);
	mh->fromname[n] = 0;
*/
	return mh;
}

int cmount(struct chan *new, struct chan *old, int flag, char *spec)
{
	ERRSTACK(1);
	struct pgrp *pg;
	int order, flg;
	struct mhead *m, **l, *mh;
	struct mount *nm, *f, *um, **h;

	/* Can't bind pointing to a symlink, since it vastly complicates namec
	 * and walk.  In particular, walk() only follows mounts on the
	 * intermediate path elements.  Grep 'ntry - 1'.  Because of that,
	 * walk() can end on a symlink.  Having domount() follow symlinks is a
	 * pain: undomount. */
	if (new->qid.type & QTSYMLINK)
		error(EINVAL, "cannot bind a symlink");

	/* Can bind anything onto a symlink's name.  Otherwise, both the old and
	 * the new must agree on whether or not it is a directory. */
	if (!(old->qid.type & QTSYMLINK) &&
	    (QTDIR & (old->qid.type ^ new->qid.type)))
		error(EINVAL, ERROR_FIXME);

	if (old->umh)
		printd("cmount old extra umh\n");

	order = flag & MORDER;

	if ((old->qid.type & QTDIR) == 0 && order != MREPL)
		error(EINVAL, ERROR_FIXME);

	mh = new->umh;

	/*
	 * Not allowed to bind when the old directory
	 * is itself a union.  (Maybe it should be allowed, but I don't see
	 * what the semantics would be.)
	 *
	 * We need to check mh->mount->next to tell unions apart from
	 * simple mount points, so that things like
	 *  mount -c fd /root
	 *  bind -c /root /
	 * work.  The check of mount->mflag catches things like
	 *  mount fd /root
	 *  bind -c /root /
	 *
	 * This is far more complicated than it should be, but I don't
	 * see an easier way at the moment.     -rsc
	 */
	if ((flag & MCREATE) && mh && mh->mount
		&& (mh->mount->next || !(mh->mount->mflag & MCREATE)))
		error(EEXIST, ERROR_FIXME);

	pg = current->pgrp;
	wlock(&pg->ns);

	l = &MOUNTH(pg, old->qid);
	for (m = *l; m; m = m->hash) {
		if (eqchan(m->from, old, 1))
			break;
		l = &m->hash;
	}

	if (m == NULL) {
		/*
		 *  nothing mounted here yet.  create a mount
		 *  head and add to the hash table.
		 */
		m = newmhead(old);
		*l = m;

		/*
		 *  if this is a union mount, add the old
		 *  node to the mount chain.
		 */
		if (order != MREPL)
			m->mount = newmount(m, old, 0, 0);
	}
	wlock(&m->lock);
	if (waserror()) {
		wunlock(&m->lock);
		nexterror();
	}
	wunlock(&pg->ns);

	nm = newmount(m, new, flag, spec);
	if (mh != NULL && mh->mount != NULL) {
		/*
		 *  copy a union when binding it onto a directory
		 */
		flg = order;
		if (order == MREPL)
			flg = MAFTER;
		h = &nm->next;
		um = mh->mount;
		for (um = um->next; um; um = um->next) {
			f = newmount(m, um->to, flg, um->spec);
			*h = f;
			h = &f->next;
		}
	}

	if (m->mount && order == MREPL) {
		mountfree(m->mount);
		m->mount = 0;
	}

	if (flag & MCREATE)
		nm->mflag |= MCREATE;

	if (m->mount && order == MAFTER) {
		for (f = m->mount; f->next; f = f->next) ;
		f->next = nm;
	} else {
		for (f = nm; f->next; f = f->next) ;
		f->next = m->mount;
		m->mount = nm;
	}

	wunlock(&m->lock);
	poperror();
	return nm->mountid;
}

void cunmount(struct chan *mnt, struct chan *mounted)
{
	struct pgrp *pg;
	struct mhead *m, **l;
	struct mount *f, **p;

	if (mnt->umh)	/* should not happen */
		printd("cunmount newp extra umh %p has %p\n", mnt, mnt->umh);

	/*
	 * It _can_ happen that mounted->umh is non-NULL,
	 * because mounted is the result of namec(Aopen)
	 * (see sysfile.c:/^sysunmount).
	 * If we open a union directory, it will have a umh.
	 * Although surprising, this is okay, since the
	 * cclose will take care of freeing the umh.
	 */

	pg = current->pgrp;
	wlock(&pg->ns);

	l = &MOUNTH(pg, mnt->qid);
	for (m = *l; m; m = m->hash) {
		if (eqchan(m->from, mnt, 1))
			break;
		l = &m->hash;
	}

	if (m == 0) {
		wunlock(&pg->ns);
		error(ENOENT, ERROR_FIXME);
	}

	wlock(&m->lock);
	if (mounted == 0) {
		*l = m->hash;
		wunlock(&pg->ns);
		mountfree(m->mount);
		m->mount = NULL;
		cclose(m->from);
		wunlock(&m->lock);
		putmhead(m);
		return;
	}

	p = &m->mount;
	for (f = *p; f; f = f->next) {
		/* BUG: Needs to be 2 pass */
		if (eqchan(f->to, mounted, 1) ||
			(f->to->mchan && eqchan(f->to->mchan, mounted, 1))) {
			*p = f->next;
			f->next = 0;
			mountfree(f);
			if (m->mount == NULL) {
				*l = m->hash;
				cclose(m->from);
				wunlock(&m->lock);
				wunlock(&pg->ns);
				putmhead(m);
				return;
			}
			wunlock(&m->lock);
			wunlock(&pg->ns);
			return;
		}
		p = &f->next;
	}
	wunlock(&m->lock);
	wunlock(&pg->ns);
	error(ENOENT, ERROR_FIXME);
}

struct chan *cclone(struct chan *c)
{
	struct chan *nc;
	struct walkqid *wq;

	wq = devtab[c->type].walk(c, NULL, NULL, 0);
	if (wq == NULL)
		error(EFAIL, "clone failed");
	nc = wq->clone;
	kfree(wq);
	nc->name = c->name;
	if (c->name)
		kref_get(&c->name->ref, 1);
	return nc;
}

/* Helper: is something mounted on the chan? */
static bool is_mount_point(struct chan *c)
{
	struct pgrp *pg;
	struct mhead *m;
	int type = c->type;
	int dev = c->dev;
	struct qid qid = c->qid;

	if (!current)
		return false;
	pg = current->pgrp;
	rlock(&pg->ns);
	for (m = MOUNTH(pg, qid); m; m = m->hash) {
		rlock(&m->lock);
		if (!m->from) {
			runlock(&m->lock);
			continue;
		}
		if (eqchantdqid(m->from, type, dev, qid, 1)) {
			runlock(&m->lock);
			runlock(&pg->ns);
			return true;
		}
		runlock(&m->lock);
	}
	runlock(&pg->ns);
	return false;
}

int findmount(struct chan **cp, struct mhead **mp, int type, int dev,
	      struct qid qid)
{
	struct pgrp *pg;
	struct mhead *m;

	if (!current)
		return 0;
	pg = current->pgrp;
	rlock(&pg->ns);
	for (m = MOUNTH(pg, qid); m; m = m->hash) {
		rlock(&m->lock);
		if (m->from == NULL) {
			printd("m %p m->from 0\n", m);
			runlock(&m->lock);
			continue;
		}
		if (eqchantdqid(m->from, type, dev, qid, 1)) {
			runlock(&pg->ns);
			if (mp != NULL) {
				kref_get(&m->ref, 1);
				if (*mp != NULL)
					putmhead(*mp);
				*mp = m;
			}
			if (*cp != NULL)
				cclose(*cp);
			chan_incref(m->mount->to);
			*cp = m->mount->to;
			runlock(&m->lock);
			return 1;
		}
		runlock(&m->lock);
	}

	runlock(&pg->ns);
	return 0;
}

int domount(struct chan **cp, struct mhead **mp)
{
	return findmount(cp, mp, (*cp)->type, (*cp)->dev, (*cp)->qid);
}

struct chan *undomount(struct chan *c, struct cname *name)
{
	ERRSTACK(1);
	struct chan *nc;
	struct pgrp *pg;
	struct mount *t;
	struct mhead **h, **he, *f;

	if (!current)
		return c;
	pg = current->pgrp;
	rlock(&pg->ns);
	if (waserror()) {
		runlock(&pg->ns);
		nexterror();
	}

	he = &pg->mnthash[MNTHASH];
	for (h = pg->mnthash; h < he; h++) {
		for (f = *h; f; f = f->hash) {
			if (strcmp(f->from->name->s, name->s) != 0)
				continue;
			for (t = f->mount; t; t = t->next) {
				if (eqchan(c, t->to, 1)) {
					/*
					 * We want to come out on the left hand
					 * side of the mount point using the
					 * element of the union that we entered
					 * on.  To do this, find the element
					 * that has a from name of c->name->s.
					 */
					if (strcmp(t->head->from->name->s,
						   name->s) != 0)
						continue;
					nc = t->head->from;
					chan_incref(nc);
					cclose(c);
					c = nc;
					break;
				}
			}
		}
	}
	poperror();
	runlock(&pg->ns);
	return c;
}

/*
 * Either walks all the way or not at all.  No partial results in *cp.
 * *nerror is the number of names to display in an error message.
 */
int walk(struct chan **cp, char **names, int nnames, struct walk_helper *wh,
         int *nerror)
{
	int dev, dotdot, i, n, nhave, ntry, type;
	struct chan *c, *nc, *lastmountpoint = NULL;
	struct cname *cname;
	struct mount *f;
	struct mhead *mh, *nmh;
	struct walkqid *wq;

	c = *cp;
	chan_incref(c);
	cname = c->name;
	kref_get(&cname->ref, 1);
	mh = NULL;

	/*
	 * While we haven't gotten all the way down the path:
	 *    1. step through a mount point, if any
	 *    2. send a walk request for initial dotdot or initial prefix
	 *    without dotdot
	 *    3. move to the first mountpoint along the way.
	 *    4. repeat.
	 *
	 * An invariant is that each time through the loop, c is on the
	 * undomount side of the mount point, and c's name is cname.
	 */
	for (nhave = 0; nhave < nnames; nhave += n) {
		/* We only allow symlink when they are first and it's .. (see
		 * below) */
		if ((c->qid.type & (QTDIR | QTSYMLINK)) == 0) {
			if (nerror)
				*nerror = nhave;
			cnameclose(cname);
			cclose(c);
			set_error(ENOTDIR, ERROR_FIXME);
			if (mh != NULL)
				putmhead(mh);
			return -1;
		}
		ntry = nnames - nhave;
		if (ntry > MAXWELEM)
			ntry = MAXWELEM;
		dotdot = 0;
		for (i = 0; i < ntry; i++) {
			if (isdotdot(names[nhave + i])) {
				if (i == 0) {
					dotdot = 1;
					ntry = 1;
				} else
					ntry = i;
				break;
			}
		}

		if (!dotdot && wh->can_mount)
			domount(&c, &mh);
		/* Bug - the only time we walk from a symlink should be during
		 * walk_symlink, which should have given us a dotdot. */
		if ((c->qid.type & QTSYMLINK) && !dotdot)
			panic("Got a walk from a symlink that wasn't ..!");

		type = c->type;
		dev = c->dev;

		if ((wq = devtab[type].walk(c, NULL, names + nhave, ntry)) ==
		    NULL) {
			/* try a union mount, if any */
			if (mh && wh->can_mount) {
				/*
				 * mh->mount == c, so start at mh->mount->next
				 */
				rlock(&mh->lock);
				for (f = mh->mount->next; f; f = f->next)
					if ((wq =
					     devtab[f->to->type].walk(f->to,
								      NULL,
								      names +
								      nhave,
								      ntry)) !=
					    NULL)
						break;
				runlock(&mh->lock);
				if (f != NULL) {
					type = f->to->type;
					dev = f->to->dev;
				}
			}
			if (wq == NULL) {
				cclose(c);
				cnameclose(cname);
				if (nerror)
					*nerror = nhave + 1;
				if (mh != NULL)
					putmhead(mh);
				return -1;
			}
		}

		nmh = NULL;
		if (dotdot) {
			assert(wq->nqid == 1);
			assert(wq->clone != NULL);

			cname = addelem(cname, "..");
			nc = undomount(wq->clone, cname);
			n = 1;
		} else {
			nc = NULL;
			if (wh->can_mount)
				for (i = 0; i < wq->nqid && i < ntry - 1; i++)
					if (findmount(&nc, &nmh, type, dev,
						      wq->qid[i]))
						break;
			if (nc == NULL) {	/* no mount points along path */
				if (wq->clone == NULL) {
					cclose(c);
					cnameclose(cname);
					if (wq->nqid == 0 ||
					    (wq->qid[wq->nqid - 1].type &
					     QTDIR)) {
						if (nerror)
							*nerror = nhave +
								wq->nqid + 1;
						set_error(ENOENT,
							  "walk failed");
					} else {
						if (nerror)
							*nerror = nhave +
								wq->nqid;
						set_error(ENOTDIR,
							  "walk failed");
					}
					kfree(wq);
					if (mh != NULL)
						putmhead(mh);
					return -1;
				}
				n = wq->nqid;
				nc = wq->clone;
			} else {	/* stopped early, at a mount point */
				if (wq->clone != NULL) {
					cclose(wq->clone);
					wq->clone = NULL;
				}
				lastmountpoint = nc;
				n = i + 1;
			}
			if (nc->qid.type & QTSYMLINK) {
				struct chan *old_nc = nc;

				nc = walk_symlink(old_nc, wh,
						  nnames - nhave - n);
				if (!nc) {
					/* walk_symlink() set error.
					 * This seems to be the standard
					 * walk() error-cleanup. */
					if (nerror)
						*nerror = nhave + n;
					cclose(c);
					cclose(old_nc);
					cnameclose(cname);
					kfree(wq);
					if (mh != NULL)
						putmhead(mh);
					return -1;
				}
			}
			for (i = 0; i < n; i++)
				cname = addelem(cname, names[nhave + i]);
		}
		cclose(c);
		c = nc;
		putmhead(mh);
		mh = nmh;
		kfree(wq);
	}

	putmhead(mh);

	c = cunique(c);

	if (c->umh != NULL) {	//BUG
		printd("walk umh\n");
		putmhead(c->umh);
		c->umh = NULL;
	}

	cnameclose(c->name);
	c->name = cname;
	c->mountpoint = lastmountpoint;

	cclose(*cp);
	*cp = c;
	if (nerror)
		*nerror = 0;
	return 0;
}

/*
 * c is a mounted non-creatable directory.  find a creatable one.
 */
struct chan *createdir(struct chan *c, struct mhead *m)
{
	ERRSTACK(1);
	struct chan *nc;
	struct mount *f;

	rlock(&m->lock);
	if (waserror()) {
		runlock(&m->lock);
		nexterror();
	}
	for (f = m->mount; f; f = f->next) {
		if (f->mflag & MCREATE) {
			nc = cclone(f->to);
			runlock(&m->lock);
			poperror();
			cclose(c);
			return nc;
		}
	}
	error(EPERM, ERROR_FIXME);
	poperror();
	return 0;
}

/*
 * In place, rewrite name to compress multiple /, eliminate ., and process ..
 */
void cleancname(struct cname *n)
{
	char *p;

	if (n->s[0] == '#') {
		p = strchr(n->s, '/');
		if (p == NULL)
			return;
		cleanname(p);

		/*
		 * The correct name is #i rather than #i/,
		 * but the correct name of #/ is #/.
		 */
		if (strcmp(p, "/") == 0 && n->s[1] != '/')
			*p = '\0';
	} else
		cleanname(n->s);
	n->len = strlen(n->s);
}

static void growparse(Elemlist * e)
{
	char **new;
	int *inew;
	enum { Delta = 8 };

	if (e->ARRAY_SIZEs % Delta == 0) {
		new = kzmalloc((e->ARRAY_SIZEs + Delta) * sizeof(char *), 0);
		memmove(new, e->elems, e->ARRAY_SIZEs * sizeof(char *));
		kfree(e->elems);
		e->elems = new;
		inew = kzmalloc((e->ARRAY_SIZEs + Delta + 1) * sizeof(int), 0);
		memmove(inew, e->off, e->ARRAY_SIZEs * sizeof(int));
		kfree(e->off);
		e->off = inew;
	}
}

/*
 * The name is known to be valid.
 * Copy the name so slashes can be overwritten.
 * An empty string will set ARRAY_SIZE=0.
 * A path ending in / or /. or /.//./ etc. will have
 * e.mustbedir = 1, so that we correctly
 * reject, e.g., "/adm/users/." when /adm/users is a file
 * rather than a directory.
 */
static void parsename(char *name, Elemlist * e)
{
	char *slash;

	kstrdup(&e->name, name);
	name = e->name;
	e->ARRAY_SIZEs = 0;
	e->elems = NULL;
	e->off = kzmalloc(sizeof(int), 0);
	e->off[0] = skipslash(name) - name;
	for (;;) {
		name = skipslash(name);
		if (*name == '\0') {
			e->mustbedir = 1;
			break;
		}
		growparse(e);

		e->elems[e->ARRAY_SIZEs++] = name;
		/* we may want to do this again some day
		   slash = utfrune(name, '/');
		 */
		slash = strchr(name, '/');
		if (slash == NULL) {
			e->off[e->ARRAY_SIZEs] = name + strlen(name) - e->name;
			e->mustbedir = 0;
			break;
		}
		e->off[e->ARRAY_SIZEs] = slash - e->name;
		*slash++ = '\0';
		name = slash;
	}
}

void *memrchr(void *va, int c, long n)
{
	uint8_t *a, *e;

	a = va;
	for (e = a + n - 1; e > a; e--)
		if (*e == c)
			return e;
	return NULL;
}

/*
 * Turn a name into a channel.
 * &name[0] is known to be a valid address.  It may be a kernel address.
 *
 * Opening with amode Aopen, Acreate, or Aremove guarantees
 * that the result will be the only reference to that particular fid.
 * This is necessary since we might pass the result to
 * devtab[].remove().
 *
 * Opening Atodir, Amount, or Aaccess does not guarantee this.
 *
 * Opening Aaccess can, under certain conditions, return a
 * correct Chan* but with an incorrect struct cname attached.
 * Since the functions that open Aaccess (sysstat, syswstat, sys_stat)
 * do not use the struct cname*, this avoids an unnecessary clone.
 *
 * The classic namec() is broken into a front end to get the starting point and
 * a __namec_from, which does the guts of the lookup.  */
static struct chan *__namec_from(struct chan *c, char *aname, int amode,
                                 int omode, uint32_t perm,
                                 struct walk_helper *wh, void *ext)
{
	ERRSTACK(2);
	int len, npath;
	struct chan *cnew, *renamee;
	struct cname *cname;
	Elemlist e;
	struct mhead *m;
	char tmperrbuf[ERRMAX];
	int saved_errno;
	// Rune r;

	static_assert(!(CINTERNAL_FLAGS & CEXTERNAL_FLAGS));

	e.name = NULL;
	e.elems = NULL;
	e.off = NULL;
	e.ARRAY_SIZEs = 0;
	if (waserror()) {
		cclose(c);
		kfree(e.name);
		kfree(e.elems);
		kfree(e.off);
		//dumpmount();
		nexterror();
	}

	/*
	 * Build a list of elements in the path.
	 */
	parsename(aname, &e);

	if (e.mustbedir)
		omode &= ~O_NOFOLLOW;

	switch (amode) {
	case Acreate:
		/* perm must have DMDIR if last element is / or /. */
		if (e.mustbedir && !(perm & DMDIR)) {
			npath = e.ARRAY_SIZEs;
			error(EINVAL, "create without DMDIR");
		}
		/* don't try to walk the last path element just yet. */
		if (e.ARRAY_SIZEs == 0)
			error(EEXIST, ERROR_FIXME);
		e.ARRAY_SIZEs--;
		/* We're dropping the last element, which O_NOFOLLOW applied to.
		 * Not sure if there are any legit reasons to have O_NOFOLLOW
		 * with create.*/
		omode &= ~O_NOFOLLOW;
		break;
	case Arename:
		if (e.ARRAY_SIZEs == 0)
			error(EINVAL, "rename needs at least one name");
		e.ARRAY_SIZEs--;
		omode &= ~O_NOFOLLOW;
		break;
	/* the difference for stat and lstat (Aaccess) are handled in sysfile.c
	 */
	case Abind:
	case Amount:
	case Aremove:
		omode |= O_NOFOLLOW;
		break;
	}

	if (omode & O_NOFOLLOW)
		wh->no_follow = true;

	if (walk(&c, e.elems, e.ARRAY_SIZEs, wh, &npath) < 0) {
		if (npath < 0 || npath > e.ARRAY_SIZEs) {
			printd("namec %s walk error npath=%d\n", aname, npath);
			error(EFAIL, "walk failed");
		}
		/* Old plan 9 errors would jump here for the magic error
		 * parsing. */
NameError:
		if (current_errstr()[0]) {
			/* errstr is set, just stick with it and error out */
			error_jmp();
		} else {
			error(EFAIL, "Name to chan lookup failed");
		}
		/* brho: skipping the namec custom error string business, since
		 * it hides the underlying failure.  implement this if you want
		 * the old stuff. */
#if 0
		strlcpy(tmperrbuf, current->errstr, sizeof(tmperrbuf));
		// prefix was name - aname, the start pt
		len = prefix + e.off[npath];
		if (len < ERRMAX / 3 || (name = memrchr(aname, '/', len)) ==
		    NULL || name == aname)
			snprintf(get_cur_genbuf(), sizeof current->genbuf,
				 "%.*s", len, aname);
		else
			snprintf(get_cur_genbuf(), sizeof current->genbuf,
				 "...%.*s", (int)(len - (name - aname)), name);
		snprintf(current->errstr, ERRMAX, "%#q %s", get_cur_genbuf(),
				 tmperrbuf);
#endif
	}

	if (e.mustbedir && !(c->qid.type & QTDIR)) {
		npath = e.ARRAY_SIZEs;
		error(ENOTDIR, "not a dir, but mustbedir.  trailing slash?");
	}

	if ((amode == Aopen) && (omode & O_EXEC) && (c->qid.type & QTDIR)) {
		npath = e.ARRAY_SIZEs;
		error(EFAIL, "cannot exec directory");
	}

	switch (amode) {
	case Aaccess:
		if (wh->can_mount)
			domount(&c, NULL);
		break;

	case Abind:
		m = NULL;
		if (wh->can_mount)
			domount(&c, &m);
		if (c->umh != NULL)
			putmhead(c->umh);
		c->umh = m;
		break;

	case Aremove:
	case Aopen:
Open:
		/* save the name; domount might change c */
		cname = c->name;
		kref_get(&cname->ref, 1);
		m = NULL;
		if (wh->can_mount)
			domount(&c, &m);

		/* our own copy to open or remove */
		c = cunique(c);

		/* now it's our copy anyway, we can put the name back */
		cnameclose(c->name);
		c->name = cname;

		switch (amode) {
		case Aremove:
			putmhead(m);
			break;

		case Aopen:
		case Acreate:
			if (c->umh != NULL) {
				printd("cunique umh\n");
				putmhead(c->umh);
				c->umh = NULL;
			}

			/* only save the mount head if it's a multiple element
			 * union */
			if (m && m->mount && m->mount->next)
				c->umh = m;
			else
				putmhead(m);
			/* here is where convert omode/vfs flags to c->flags.
			 * careful, O_CLOEXEC and O_REMCLO are in there.  might
			 * need to change that. */
			c->flag |= omode & CEXTERNAL_FLAGS;
			c = devtab[c->type].open(c,
						 omode & ~O_CLOEXEC);
			/* if you get this from a dev, in the dev's open, you
			 * are probably saving mode directly, without passing it
			 * through openmode. */
			if (c->mode & O_TRUNC)
				error(EFAIL,
				      "Device %s open failed to clear O_TRUNC",
				      devtab[c->type].name);
			break;
		}
		break;

	case Atodir:
		/*
		 * Directories (e.g. for cd) are left before the mount point,
		 * so one may mount on / or . and see the effect.
		 */
		if (!(c->qid.type & QTDIR))
			error(ENOTDIR, ERROR_FIXME);
		break;

	case Amount:
		/*
		 * When mounting on an already mounted upon directory,
		 * one wants subsequent mounts to be attached to the
		 * original directory, not the replacement.  Don't domount.
		 */
		break;

	case Arename:
		/* We already walked to the parent of new_path, which is in c.
		 * We're a lot like create here - need to find mounts, etc.  On
		 * the way out, we putmhead if we have an m, and clean up our
		 * chans.  On success, c becomes cnew (thus close the old c).
		 * On failure, we just close cnew. */
		if (!(c->qid.type & QTDIR))
			error(ENOTDIR, "rename target parent is not a dir");
		e.ARRAY_SIZEs++;
		m = NULL;
		cnew = NULL;
		if (waserror()) {
			/* rename or createdir failed */
			cclose(cnew);
			if (m)
				putmhead(m);
			nexterror();	/* safe since we're in a waserror() */
		}
		if (wh->can_mount && findmount(&cnew, &m, c->type, c->dev,
		                               c->qid)) {
			cnew = createdir(cnew, m);
		} else {
			cnew = c;
			chan_incref(cnew);
		}
		cnew = cunique(cnew);
		cnameclose(cnew->name);
		cnew->name = c->name;
		kref_get(&cnew->name->ref, 1);
		/* At this point, we have our new_path parent chan (cnew) and
		 * the renamee chan */
		renamee = ext;
		if (cnew->type != renamee->type)
			error(EXDEV, "can't rename across device types");

		devtab[cnew->type].rename(renamee, cnew,
		                          e.elems[e.ARRAY_SIZEs - 1], 0);
		poperror();

		if (m)
			putmhead(m);
		cclose(c);
		c = cnew;
		c->name = addelem(c->name, e.elems[e.ARRAY_SIZEs - 1]);
		break;

	case Acreate:
		/*
		 * We've already walked all but the last element.
		 * If the last exists, try to open it OTRUNC.
		 * If omode&OEXCL is set, just give up.
		 */
		e.ARRAY_SIZEs++;
		if (walk(&c, e.elems + e.ARRAY_SIZEs - 1, 1, wh, NULL) == 0) {
			if (omode & O_EXCL)
				error(EEXIST, ERROR_FIXME);
			omode |= O_TRUNC;
			goto Open;
		}

		/*
		 * The semantics of the create(2) system call are that if the
		 * file exists and can be written, it is to be opened with
		 * truncation.  On the other hand, the create(5) message fails
		 * if the file exists.
		 *
		 * If we get two create(2) calls happening simultaneously, they
		 * might both get here and send create(5) messages, but only one
		 * of the messages will succeed.  To provide the expected
		 * create(2) semantics, the call with the failed message needs
		 * to try the above walk again, opening for truncation.  This
		 * correctly solves the create/create race, in the sense that
		 * any observable outcome can be explained as one happening
		 * before the other.  The create/create race is quite common.
		 * For example, it happens when two rc subshells simultaneously
		 * update the same environment variable.
		 *
		 * The implementation still admits a create/create/remove race:
		 * (A) walk to file, fails
		 * (B) walk to file, fails
		 * (A) create file, succeeds, returns
		 * (B) create file, fails
		 * (A) remove file, succeeds, returns
		 * (B) walk to file, return failure.
		 *
		 * This is hardly as common as the create/create race, and is
		 * really not too much worse than what might happen if (B) got a
		 * hold of a file descriptor and then the file was removed --
		 * either way (B) can't do anything with the result of the
		 * create call.  So we don't care about this race.
		 *
		 * Applications that care about more fine-grained decision of
		 * the races can use the OEXCL flag to get at the underlying
		 * create(5) semantics; by default we provide the common case.
		 *
		 * We need to stay behind the mount point in case we
		 * need to do the first walk again (should the create fail).
		 *
		 * We also need to cross the mount point and find the directory
		 * in the union in which we should be creating.
		 *
		 * The channel staying behind is c, the one moving forward is
		 * cnew.
		 */
		m = NULL;
		cnew = NULL;	/* is this assignment necessary? */
		/* discard error */
		if (!waserror()) {	/* try create */
			if (wh->can_mount &&
			    findmount(&cnew, &m, c->type, c->dev, c->qid))
				cnew = createdir(cnew, m);
			else {
				cnew = c;
				chan_incref(cnew);
			}

			/*
			 * We need our own copy of the Chan because we're about
			 * to send a create, which will move it.  Once we have
			 * our own copy, we can fix the name, which might be
			 * wrong if findmount gave us a new Chan.
			 */
			cnew = cunique(cnew);
			cnameclose(cnew->name);
			cnew->name = c->name;
			kref_get(&cnew->name->ref, 1);

			cnew->flag |= omode & CEXTERNAL_FLAGS;
			devtab[cnew->type].create(cnew,
						  e.elems[e.ARRAY_SIZEs - 1],
						  omode & ~(O_EXCL | O_CLOEXEC),
						  perm, ext);
			poperror();

			if (m)
				putmhead(m);
			cclose(c);
			c = cnew;
			c->name = addelem(c->name, e.elems[e.ARRAY_SIZEs - 1]);
			break;
		}

		/* create failed */
		cclose(cnew);
		if (m)
			putmhead(m);
		if (omode & O_EXCL)
			nexterror();	/* safe since we're in a waserror() */
		poperror();	/* matching the if(!waserror) */

		/* save error, so walk doesn't clobber our existing errstr */
		strlcpy(tmperrbuf, current_errstr(), sizeof(tmperrbuf));
		saved_errno = get_errno();
		/* note: we depend that walk does not error */
		if (walk(&c, e.elems + e.ARRAY_SIZEs - 1, 1, wh, NULL) < 0) {
			set_errno(saved_errno);
			/* Report the error we had originally */
			error(EFAIL, tmperrbuf);
		}
		strlcpy(current_errstr(), tmperrbuf, MAX_ERRSTR_LEN);
		omode |= O_TRUNC;
		goto Open;

	default:
		panic("unknown namec access %d\n", amode);
	}

	poperror();

	if (e.ARRAY_SIZEs > 0)
		strlcpy(get_cur_genbuf(), e.elems[e.ARRAY_SIZEs - 1],
			GENBUF_SZ);
	else
		strlcpy(get_cur_genbuf(), ".", GENBUF_SZ);

	kfree(e.name);
	kfree(e.elems);
	kfree(e.off);

	return c;
}

struct chan *namec(char *name, int amode, int omode, uint32_t perm, void *ext)
{
	struct walk_helper wh = {.can_mount = true};
	struct chan *c;
	char *devname, *devspec;
	int n, devtype;

	if (name[0] == '\0')
		error(EFAIL, "empty file name");
	validname(name, 1);
	/*
	 * Find the starting off point (the current slash, the root of
	 * a device tree, or the current dot) as well as the name to
	 * evaluate starting there.
	 */
	switch (name[0]) {
	case '/':
		if (current)
			c = current->slash;
		else
			c = kern_slash;
		chan_incref(c);
		break;

	case '#':
		wh.can_mount = false;
		devname = get_cur_genbuf();
		devname[0] = '\0';
		n = 0;
		name++; /* drop the # */
		while ((*name != '\0') && (*name != '/')) {
			if (n >= GENBUF_SZ - 1)
				error(ENAMETOOLONG, ERROR_FIXME);
			devname[n++] = *name++;
		}
		devname[n] = '\0';
		/* for a name #foo.spec, devname = foo\0, devspec = spec\0.
		 * genbuf contains foo\0spec\0.  for no spec, devspec = \0 */
		devspec = strchr(devname, '.');
		if (devspec) {
			*devspec = '\0';
			devspec++;
		} else {
			devspec = &devname[n];
		}
		/* These devices have special attach functions that treat the
		 * char * as a blob pointer */
		if (!strcmp(devname, "mnt"))
			error(EINVAL, "can't namec-attach #mnt");
		if (!strcmp(devname, "gtfs"))
			error(EINVAL, "can't namec-attach #gtfs");
		/* TODO: deal with this "nodevs" business. */
		#if 0
		/*
		 *  the nodevs exceptions are
		 *  |  it only gives access to pipes you create
		 *  e  this process's environment
		 *  s  private file2chan creation space
		 *  D private secure sockets name space
		 *  a private TLS name space
		 */
		if (current->pgrp->nodevs &&
			//          (utfrune("|esDa", r) == NULL
			((strchr("|esDa", get_cur_genbuf()[1]) == NULL)
			 || (get_cur_genbuf()[1] == 's'	// || r == 's'
				 && get_cur_genbuf()[n] != '\0')))
			error(EINVAL, ERROR_FIXME);
		#endif
		devtype = devno(devname, 1);
		if (devtype == -1)
			error(EFAIL, "Unknown #device %s (spec %s)", devname,
			      devspec);
		c = devtab[devtype].attach(devspec);
		break;
	default:
		/* this case also covers \0 */
		c = current->dot;
		if (!c)
			panic("no dot!");
		chan_incref(c);
		break;
	}
	return __namec_from(c, name, amode, omode, perm, &wh, ext);
}

struct chan *namec_from(struct chan *c, char *name, int amode, int omode,
                        uint32_t perm, void *ext)
{
	struct walk_helper wh = {.can_mount = true};

	if (name[0] == '\0') {
		/* Our responsibility to cclose 'c' on our error */
		cclose(c);
		error(EFAIL, "empty file name");
	}
	validname(name, 1);
	return __namec_from(c, name, amode, omode, perm, &wh, ext);
}

/*
 * name is valid. skip leading / and ./ as much as possible
 */
char *skipslash(char *name)
{
	while (name[0] == '/'
		   || (name[0] == '.' && (name[1] == 0 || name[1] == '/')))
		name++;
	return name;
}

char isfrog[256] = {
	 /*NUL*/ 1, 1, 1, 1, 1, 1, 1, 1,
	 /*BKS*/ 1, 1, 1, 1, 1, 1, 1, 1,
	 /*DLE*/ 1, 1, 1, 1, 1, 1, 1, 1,
	 /*CAN*/ 1, 1, 1, 1, 1, 1, 1, 1,
	['/'] 1,
	[0x7f] 1,
};

/*
 * Check that the name
 *  a) is in valid memory.
 *  b) is shorter than 2^16 bytes, so it can fit in a 9P string field.
 *  c) contains no frogs.
 * The first byte is known to be addressible by the requester, so the
 * routine works for kernel and user memory both.
 * The parameter slashok flags whether a slash character is an error
 * or a valid character.
 */
void validname(char *aname, int slashok)
{
	char *ename, *name;
	int c;

	name = aname;
	ename = memchr(name, 0, (1 << 16));

	if (ename == NULL || ename - name >= (1 << 16))
		error(EINVAL, "Name too long");

	while (*name) {
		/* all characters above '~' are ok */
		c = *(uint8_t *) name;
#if 0
		if (c >= Runeself)
			name += chartorune(&r, name);
#endif
		if (c >= 0x7f) {
			error(EFAIL, "Akaros doesn't do UTF-8");
		} else {
			if (isfrog[c])
				if (!slashok || c != '/') {
					error(EINVAL, "%s (%p), at char %c",
					      aname, aname, c);
				}
			name++;
		}
	}
}

void isdir(struct chan *c)
{
	if (c->qid.type & QTDIR)
		return;
	error(ENOTDIR, ERROR_FIXME);
}

/*
 * This is necessary because there are many
 * pointers to the top of a given mount list:
 *
 *	- the mhead in the namespace hash table
 *	- the mhead in chans returned from findmount:
 *	  used in namec and then by unionread.
 *	- the mhead in chans returned from createdir:
 *	  used in the open/create race protect, which is gone.
 *
 * The RWlock in the Mhead protects the mount list it contains.
 * The mount list is deleted when we cunmount.
 * The RWlock ensures that nothing is using the mount list at that time.
 *
 * It is okay to replace c->mh with whatever you want as
 * long as you are sure you have a unique reference to it.
 *
 * This comment might belong somewhere else.
 */
void putmhead(struct mhead *m)
{
	if (m)
		kref_put(&m->ref);
}

/* Given s, make a copy of a string with padding bytes in front.  Returns a
 * pointer to the start of the string and the memory to free in str_store.
 *
 * Free str_store with kfree. */
static char *pad_and_strdup(char *s, int padding, char **str_store)
{
	char *store = kzmalloc(strlen(s) + 1 + padding, MEM_WAIT);

	strlcpy(store + padding, s, strlen(s) + 1);
	*str_store = store;
	return store + padding;
}

/* Walks a symlink c.  Returns the target chan, which could be the symlink
 * itself, if we're NO_FOLLOW.  On success, we'll decref the symlink and give
 * you a ref counted result.
 *
 * Returns NULL on error, and does not close the symlink.  Like regular walk, it
 * is all or nothing. */
static struct chan *walk_symlink(struct chan *symlink, struct walk_helper *wh,
                                 unsigned int nr_names_left)
{
	struct dir *dir;
	char *link_name, *link_store;
	struct chan *from;
	bool old_nofollow;
	Elemlist e = {0};

	/* mildly expensive: need to rlock the namespace */
	if (is_mount_point(symlink))
		return symlink;
	if (!nr_names_left && wh->no_follow)
		return symlink;
	if (wh->nr_loops >= WALK_MAX_NR_LOOPS) {
		set_error(ELOOP, "too many nested symlinks in walk");
		return NULL;
	}
	dir = chandirstat(symlink);
	if (!dir) {
		/* Should propagate the error from dev.stat() */
		return NULL;
	}
	if (!(dir->mode & DMSYMLINK)) {
		set_error(ELOOP, "symlink isn't a symlink!");
		kfree(dir);
		return NULL;
	}
	link_name = pad_and_strdup(dir->ext, 3, &link_store);
	kfree(dir);

	if (link_name[0] == '/') {
		if (current)
			from = current->slash;
		else
			from = kern_slash;
	} else {
		from = symlink;
		link_name -= 3;
		strncpy(link_name, "../", 3);
		if (!from->name)
			from->name = newcname("");
	}
	/* we close this ref on failure or it gets walked to the result. */
	chan_incref(from);

	parsename(link_name, &e);
	kfree(link_store);

	wh->nr_loops++;
	/* no_follow applies to the outermost walk, i.e. the one that the
	 * original namec performs.  At this point, we've decided that we're
	 * going to try and follow a symlink: even if its no_follow, that only
	 * applies to the last link in the original path.  Our sub-walks are not
	 * no_follow.
	 *
	 * Note the other wh vars need to stay with the walk: nr_loops,
	 * since its our method of detecting symlink loops, and can_mount, which
	 * is a property of the overall namec() call. */
	old_nofollow = wh->no_follow;
	wh->no_follow = false;
	if (walk(&from, e.elems, e.ARRAY_SIZEs, wh, NULL) < 0) {
		cclose(from);
		from = NULL;
	} else {
		/* We can still have a successful walk and have the new 'from'
		 * be a symlink.  We'd need walk_symlink to return a symlink
		 * chan, which happens if the symlink is a mount point. */
		cclose(symlink);
	}
	wh->no_follow = old_nofollow;
	wh->nr_loops--;

	kfree(e.name);
	kfree(e.elems);
	kfree(e.off);
	return from;
}