vfs/hammer2/hammer2_chain.c

/*
 * Copyright (c) 2011-2018 The DragonFly Project.  All rights reserved.
 *
 * This code is derived from software contributed to The DragonFly Project
 * by Matthew Dillon <dillon@dragonflybsd.org>
 * and Venkatesh Srinivas <vsrinivas@dragonflybsd.org>
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 *
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in
 *    the documentation and/or other materials provided with the
 *    distribution.
 * 3. Neither the name of The DragonFly Project nor the names of its
 *    contributors may be used to endorse or promote products derived
 *    from this software without specific, prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 */
/*
 * This subsystem implements most of the core support functions for
 * the hammer2_chain structure.
 *
 * Chains are the in-memory version on media objects (volume header, inodes,
 * indirect blocks, data blocks, etc).  Chains represent a portion of the
 * HAMMER2 topology.
 *
 * Chains are no-longer delete-duplicated.  Instead, the original in-memory
 * chain will be moved along with its block reference (e.g. for things like
 * renames, hardlink operations, modifications, etc), and will be indexed
 * on a secondary list for flush handling instead of propagating a flag
 * upward to the root.
 *
 * Concurrent front-end operations can still run against backend flushes
 * as long as they do not cross the current flush boundary.  An operation
 * running above the current flush (in areas not yet flushed) can become
 * part of the current flush while ano peration running below the current
 * flush can become part of the next flush.
 */
#include <sys/cdefs.h>
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/types.h>
#include <sys/lock.h>
#include <sys/kern_syscall.h>
#include <sys/uuid.h>

#include <crypto/sha2/sha2.h>

#include "hammer2.h"

static hammer2_chain_t *hammer2_chain_create_indirect(
		hammer2_chain_t *parent,
		hammer2_key_t key, int keybits,
		hammer2_tid_t mtid, int for_type, int *errorp);
static hammer2_io_t *hammer2_chain_drop_data(hammer2_chain_t *chain);
static hammer2_chain_t *hammer2_combined_find(
		hammer2_chain_t *parent,
		hammer2_blockref_t *base, int count,
		hammer2_key_t *key_nextp,
		hammer2_key_t key_beg, hammer2_key_t key_end,
		hammer2_blockref_t **bresp);

static struct krate krate_h2me = { .freq = 1 };

/*
 * Basic RBTree for chains (core->rbtree and core->dbtree).  Chains cannot
 * overlap in the RB trees.  Deleted chains are moved from rbtree to either
 * dbtree or to dbq.
 *
 * Chains in delete-duplicate sequences can always iterate through core_entry
 * to locate the live version of the chain.
 */
RB_GENERATE(hammer2_chain_tree, hammer2_chain, rbnode, hammer2_chain_cmp);

int
hammer2_chain_cmp(hammer2_chain_t *chain1, hammer2_chain_t *chain2)
{
	hammer2_key_t c1_beg;
	hammer2_key_t c1_end;
	hammer2_key_t c2_beg;
	hammer2_key_t c2_end;

	/*
	 * Compare chains.  Overlaps are not supposed to happen and catch
	 * any software issues early we count overlaps as a match.
	 */
	c1_beg = chain1->bref.key;
	c1_end = c1_beg + ((hammer2_key_t)1 << chain1->bref.keybits) - 1;
	c2_beg = chain2->bref.key;
	c2_end = c2_beg + ((hammer2_key_t)1 << chain2->bref.keybits) - 1;

	if (c1_end < c2_beg)	/* fully to the left */
		return(-1);
	if (c1_beg > c2_end)	/* fully to the right */
		return(1);
	return(0);		/* overlap (must not cross edge boundary) */
}

/*
 * Assert that a chain has no media data associated with it.
 */
static __inline void
hammer2_chain_assert_no_data(hammer2_chain_t *chain)
{
	KKASSERT(chain->dio == NULL);
	if (chain->bref.type != HAMMER2_BREF_TYPE_VOLUME &&
	    chain->bref.type != HAMMER2_BREF_TYPE_FREEMAP &&
	    chain->data) {
		panic("hammer2_assert_no_data: chain %p still has data", chain);
	}
}

/*
 * Make a chain visible to the flusher.  The flusher needs to be able to
 * do flushes of subdirectory chains or single files so it does a top-down
 * recursion using the ONFLUSH flag for the recursion.  It locates MODIFIED
 * or UPDATE chains and flushes back up the chain to the volume root.
 *
 * This routine sets ONFLUSH upward until it hits the volume root.  For
 * simplicity we ignore PFSROOT boundaries whos rules can be complex.
 * Extra ONFLUSH flagging doesn't hurt the filesystem.
 */
void
hammer2_chain_setflush(hammer2_chain_t *chain)
{
	hammer2_chain_t *parent;

	if ((chain->flags & HAMMER2_CHAIN_ONFLUSH) == 0) {
		hammer2_spin_sh(&chain->core.spin);
		while ((chain->flags & HAMMER2_CHAIN_ONFLUSH) == 0) {
			atomic_set_int(&chain->flags, HAMMER2_CHAIN_ONFLUSH);
			if ((parent = chain->parent) == NULL)
				break;
			hammer2_spin_sh(&parent->core.spin);
			hammer2_spin_unsh(&chain->core.spin);
			chain = parent;
		}
		hammer2_spin_unsh(&chain->core.spin);
	}
}

/*
 * Allocate a new disconnected chain element representing the specified
 * bref.  chain->refs is set to 1 and the passed bref is copied to
 * chain->bref.  chain->bytes is derived from the bref.
 *
 * chain->pmp inherits pmp unless the chain is an inode (other than the
 * super-root inode).
 *
 * NOTE: Returns a referenced but unlocked (because there is no core) chain.
 */
hammer2_chain_t *
hammer2_chain_alloc(hammer2_dev_t *hmp, hammer2_pfs_t *pmp,
		    hammer2_blockref_t *bref)
{
	hammer2_chain_t *chain;
	u_int bytes;

	/*
	 * Special case - radix of 0 indicates a chain that does not
	 * need a data reference (context is completely embedded in the
	 * bref).
	 */
	if ((int)(bref->data_off & HAMMER2_OFF_MASK_RADIX))
		bytes = 1U << (int)(bref->data_off & HAMMER2_OFF_MASK_RADIX);
	else
		bytes = 0;

	atomic_add_long(&hammer2_chain_allocs, 1);

	/*
	 * Construct the appropriate system structure.
	 */
	switch(bref->type) {
	case HAMMER2_BREF_TYPE_DIRENT:
	case HAMMER2_BREF_TYPE_INODE:
	case HAMMER2_BREF_TYPE_INDIRECT:
	case HAMMER2_BREF_TYPE_FREEMAP_NODE:
	case HAMMER2_BREF_TYPE_DATA:
	case HAMMER2_BREF_TYPE_FREEMAP_LEAF:
		/*
		 * Chain's are really only associated with the hmp but we
		 * maintain a pmp association for per-mount memory tracking
		 * purposes.  The pmp can be NULL.
		 */
		chain = kmalloc(sizeof(*chain), hmp->mchain, M_WAITOK | M_ZERO);
		break;
	case HAMMER2_BREF_TYPE_VOLUME:
	case HAMMER2_BREF_TYPE_FREEMAP:
		/*
		 * Only hammer2_chain_bulksnap() calls this function with these
		 * types.
		 */
		chain = kmalloc(sizeof(*chain), hmp->mchain, M_WAITOK | M_ZERO);
		break;
	default:
		chain = NULL;
		panic("hammer2_chain_alloc: unrecognized blockref type: %d",
		      bref->type);
	}

	/*
	 * Initialize the new chain structure.  pmp must be set to NULL for
	 * chains belonging to the super-root topology of a device mount.
	 */
	if (pmp == hmp->spmp)
		chain->pmp = NULL;
	else
		chain->pmp = pmp;
	chain->hmp = hmp;
	chain->bref = *bref;
	chain->bytes = bytes;
	chain->refs = 1;
	chain->flags = HAMMER2_CHAIN_ALLOCATED;

	/*
	 * Set the PFS boundary flag if this chain represents a PFS root.
	 */
	if (bref->flags & HAMMER2_BREF_FLAG_PFSROOT)
		atomic_set_int(&chain->flags, HAMMER2_CHAIN_PFSBOUNDARY);
	hammer2_chain_core_init(chain);

	return (chain);
}

/*
 * Initialize a chain's core structure.  This structure used to be allocated
 * but is now embedded.
 *
 * The core is not locked.  No additional refs on the chain are made.
 * (trans) must not be NULL if (core) is not NULL.
 */
void
hammer2_chain_core_init(hammer2_chain_t *chain)
{
	/*
	 * Fresh core under nchain (no multi-homing of ochain's
	 * sub-tree).
	 */
	RB_INIT(&chain->core.rbtree);	/* live chains */
	hammer2_mtx_init(&chain->lock, "h2chain");
}

/*
 * Add a reference to a chain element, preventing its destruction.
 *
 * (can be called with spinlock held)
 */
void
hammer2_chain_ref(hammer2_chain_t *chain)
{
	if (atomic_fetchadd_int(&chain->refs, 1) == 0) {
		/*
		 * Just flag that the chain was used and should be recycled
		 * on the LRU if it encounters it later.
		 */
		if (chain->flags & HAMMER2_CHAIN_ONLRU)
			atomic_set_int(&chain->flags, HAMMER2_CHAIN_LRUHINT);

#if 0
		/*
		 * REMOVED - reduces contention, lru_list is more heuristical
		 * now.
		 *
		 * 0->non-zero transition must ensure that chain is removed
		 * from the LRU list.
		 *
		 * NOTE: Already holding lru_spin here so we cannot call
		 *	 hammer2_chain_ref() to get it off lru_list, do
		 *	 it manually.
		 */
		if (chain->flags & HAMMER2_CHAIN_ONLRU) {
			hammer2_pfs_t *pmp = chain->pmp;
			hammer2_spin_ex(&pmp->lru_spin);
			if (chain->flags & HAMMER2_CHAIN_ONLRU) {
				atomic_add_int(&pmp->lru_count, -1);
				atomic_clear_int(&chain->flags,
						 HAMMER2_CHAIN_ONLRU);
				TAILQ_REMOVE(&pmp->lru_list, chain, lru_node);
			}
			hammer2_spin_unex(&pmp->lru_spin);
		}
#endif
	}
}

/*
 * Ref a locked chain and force the data to be held across an unlock.
 * Chain must be currently locked.  The user of the chain who desires
 * to release the hold must call hammer2_chain_lock_unhold() to relock
 * and unhold the chain, then unlock normally, or may simply call
 * hammer2_chain_drop_unhold() (which is safer against deadlocks).
 */
void
hammer2_chain_ref_hold(hammer2_chain_t *chain)
{
	atomic_add_int(&chain->lockcnt, 1);
	hammer2_chain_ref(chain);
}

/*
 * Insert the chain in the core rbtree.
 *
 * Normal insertions are placed in the live rbtree.  Insertion of a deleted
 * chain is a special case used by the flush code that is placed on the
 * unstaged deleted list to avoid confusing the live view.
 */
#define HAMMER2_CHAIN_INSERT_SPIN	0x0001
#define HAMMER2_CHAIN_INSERT_LIVE	0x0002
#define HAMMER2_CHAIN_INSERT_RACE	0x0004

static
int
hammer2_chain_insert(hammer2_chain_t *parent, hammer2_chain_t *chain,
		     int flags, int generation)
{
	hammer2_chain_t *xchain;
	int error = 0;

	if (flags & HAMMER2_CHAIN_INSERT_SPIN)
		hammer2_spin_ex(&parent->core.spin);

	/*
	 * Interlocked by spinlock, check for race
	 */
	if ((flags & HAMMER2_CHAIN_INSERT_RACE) &&
	    parent->core.generation != generation) {
		error = HAMMER2_ERROR_EAGAIN;
		goto failed;
	}

	/*
	 * Insert chain
	 */
	xchain = RB_INSERT(hammer2_chain_tree, &parent->core.rbtree, chain);
	KASSERT(xchain == NULL,
		("hammer2_chain_insert: collision %p %p (key=%016jx)",
		chain, xchain, chain->bref.key));
	atomic_set_int(&chain->flags, HAMMER2_CHAIN_ONRBTREE);
	chain->parent = parent;
	++parent->core.chain_count;
	++parent->core.generation;	/* XXX incs for _get() too, XXX */

	/*
	 * We have to keep track of the effective live-view blockref count
	 * so the create code knows when to push an indirect block.
	 */
	if (flags & HAMMER2_CHAIN_INSERT_LIVE)
		atomic_add_int(&parent->core.live_count, 1);
failed:
	if (flags & HAMMER2_CHAIN_INSERT_SPIN)
		hammer2_spin_unex(&parent->core.spin);
	return error;
}

/*
 * Drop the caller's reference to the chain.  When the ref count drops to
 * zero this function will try to disassociate the chain from its parent and
 * deallocate it, then recursely drop the parent using the implied ref
 * from the chain's chain->parent.
 *
 * Nobody should own chain's mutex on the 1->0 transition, unless this drop
 * races an acquisition by another cpu.  Therefore we can loop if we are
 * unable to acquire the mutex, and refs is unlikely to be 1 unless we again
 * race against another drop.
 */
static hammer2_chain_t *hammer2_chain_lastdrop(hammer2_chain_t *chain,
				int depth);
static void hammer2_chain_lru_flush(hammer2_pfs_t *pmp);

void
hammer2_chain_drop(hammer2_chain_t *chain)
{
	u_int refs;

	if (hammer2_debug & 0x200000)
		Debugger("drop");

	KKASSERT(chain->refs > 0);

	while (chain) {
		refs = chain->refs;
		cpu_ccfence();
		KKASSERT(refs > 0);

		if (refs == 1) {
			if (hammer2_mtx_ex_try(&chain->lock) == 0)
				chain = hammer2_chain_lastdrop(chain, 0);
			/* retry the same chain, or chain from lastdrop */
		} else {
			if (atomic_cmpset_int(&chain->refs, refs, refs - 1))
				break;
			/* retry the same chain */
		}
		cpu_pause();
	}
}

/*
 * Unhold a held and probably not-locked chain, ensure that the data is
 * dropped on the 1->0 transition of lockcnt by obtaining an exclusive
 * lock and then simply unlocking the chain.
 */
void
hammer2_chain_drop_unhold(hammer2_chain_t *chain)
{
	u_int lockcnt;
	int iter = 0;

	for (;;) {
		lockcnt = chain->lockcnt;
		cpu_ccfence();
		if (lockcnt > 1) {
			if (atomic_cmpset_int(&chain->lockcnt,
					      lockcnt, lockcnt - 1)) {
				break;
			}
		} else if (hammer2_mtx_ex_try(&chain->lock) == 0) {
			hammer2_chain_unlock(chain);
			break;
		} else {
			/*
			 * This situation can easily occur on SMP due to
			 * the gap inbetween the 1->0 transition and the
			 * final unlock.  We cannot safely block on the
			 * mutex because lockcnt might go above 1.
			 *
			 * XXX Sleep for one tick if it takes too long.
			 */
			if (++iter > 1000) {
				if (iter > 1000 + hz) {
					kprintf("hammer2: h2race1 %p\n", chain);
					iter = 1000;
				}
				tsleep(&iter, 0, "h2race1", 1);
			}
			cpu_pause();
		}
	}
	hammer2_chain_drop(chain);
}

/*
 * Handles the (potential) last drop of chain->refs from 1->0.  Called with
 * the mutex exclusively locked, refs == 1, and lockcnt 0.  SMP races are
 * possible against refs and lockcnt.  We must dispose of the mutex on chain.
 *
 * This function returns an unlocked chain for recursive drop or NULL.  It
 * can return the same chain if it determines it has raced another ref.
 *
 * --
 *
 * When two chains need to be recursively dropped we use the chain we
 * would otherwise free to placehold the additional chain.  It's a bit
 * convoluted but we can't just recurse without potentially blowing out
 * the kernel stack.
 *
 * The chain cannot be freed if it has any children.
 * The chain cannot be freed if flagged MODIFIED unless we can dispose of it.
 * The chain cannot be freed if flagged UPDATE unless we can dispose of it.
 * Any dedup registration can remain intact.
 *
 * The core spinlock is allowed to nest child-to-parent (not parent-to-child).
 */
static
hammer2_chain_t *
hammer2_chain_lastdrop(hammer2_chain_t *chain, int depth)
{
	hammer2_pfs_t *pmp;
	hammer2_dev_t *hmp;
	hammer2_chain_t *parent;
	hammer2_chain_t *rdrop;
#if 0
	hammer2_io_t *dio;
#endif

#if 0
	/*
	 * On last drop if there is no parent and data_off is good (at
	 * least does not represent the volume root), the modified chain
	 * is probably going to be destroyed.  We have to make sure that
	 * the data area is not registered for dedup.
	 *
	 * XXX removed. In fact, we do not have to make sure that the
	 *     data area is not registered for dedup.  The data area
	 *     can, in fact, still be used for dedup because it is
	 *     still allocated in the freemap and the underlying I/O
	 *     will still be flushed.
	 */
	if (chain->parent == NULL &&
	    (chain->flags & HAMMER2_CHAIN_MODIFIED) &&
	    (chain->bref.data_off & ~HAMMER2_OFF_MASK_RADIX)) {
		hmp = chain->hmp;
		hammer2_io_dedup_delete(hmp, chain->bref.type,
					chain->bref.data_off, chain->bytes);
	}
#endif
	/*
	 * We need chain's spinlock to interlock the sub-tree test.
	 * We already have chain's mutex, protecting chain->parent.
	 *
	 * Remember that chain->refs can be in flux.
	 */
	hammer2_spin_ex(&chain->core.spin);

	if (chain->parent != NULL) {
		/*
		 * If the chain has a parent the UPDATE bit prevents scrapping
		 * as the chain is needed to properly flush the parent.  Try
		 * to complete the 1->0 transition and return NULL.  Retry
		 * (return chain) if we are unable to complete the 1->0
		 * transition, else return NULL (nothing more to do).
		 *
		 * If the chain has a parent the MODIFIED bit prevents
		 * scrapping.
		 *
		 * Chains with UPDATE/MODIFIED are *not* put on the LRU list!
		 */
		if (chain->flags & (HAMMER2_CHAIN_UPDATE |
				    HAMMER2_CHAIN_MODIFIED)) {
			if (atomic_cmpset_int(&chain->refs, 1, 0)) {
				hammer2_spin_unex(&chain->core.spin);
#if 0
				dio = hammer2_chain_drop_data(chain, 0);
				if (dio)
					hammer2_io_bqrelse(&dio);
#endif
				hammer2_chain_assert_no_data(chain);
				hammer2_mtx_unlock(&chain->lock);
				chain = NULL;
			} else {
				hammer2_spin_unex(&chain->core.spin);
				hammer2_mtx_unlock(&chain->lock);
			}
			return (chain);
		}
		/* spinlock still held */
	} else if (chain->bref.type == HAMMER2_BREF_TYPE_VOLUME ||
		   chain->bref.type == HAMMER2_BREF_TYPE_FREEMAP) {
		/*
		 * Retain the static vchain and fchain.  Clear bits that
		 * are not relevant.  Do not clear the MODIFIED bit,
		 * and certainly do not put it on the delayed-flush queue.
		 */
		atomic_clear_int(&chain->flags, HAMMER2_CHAIN_UPDATE);
	} else {
		/*
		 * The chain has no parent and can be flagged for destruction.
		 * Since it has no parent, UPDATE can also be cleared.
		 */
		atomic_set_int(&chain->flags, HAMMER2_CHAIN_DESTROY);
		if (chain->flags & HAMMER2_CHAIN_UPDATE)
			atomic_clear_int(&chain->flags, HAMMER2_CHAIN_UPDATE);

		/*
		 * If the chain has children we must still flush the chain.
		 * Any dedup is already handled by the underlying DIO, so
		 * we do not have to specifically flush it here.
		 *
		 * In the case where it has children, the DESTROY flag test
		 * in the flush code will prevent unnecessary flushes of
		 * MODIFIED chains that are not flagged DEDUP so don't worry
		 * about that here.
		 */
		if (chain->core.chain_count) {
			/*
			 * Put on flushq (should ensure refs > 1), retry
			 * the drop.
			 */
			hammer2_spin_unex(&chain->core.spin);
			hammer2_delayed_flush(chain);
			hammer2_mtx_unlock(&chain->lock);

			return(chain);	/* retry drop */
		}

		/*
		 * Otherwise we can scrap the MODIFIED bit if it is set,
		 * and continue along the freeing path.
		 *
		 * Be sure to clean-out any dedup bits.  Without a parent
		 * this chain will no longer be visible to the flush code.
		 * Easy check data_off to avoid the volume root.
		 */
		if (chain->flags & HAMMER2_CHAIN_MODIFIED) {
			atomic_clear_int(&chain->flags, HAMMER2_CHAIN_MODIFIED);
			atomic_add_long(&hammer2_count_modified_chains, -1);
			if (chain->pmp)
				hammer2_pfs_memory_wakeup(chain->pmp);
		}
		/* spinlock still held */
	}

	/* spinlock still held */
#if 0
	dio = NULL;
#endif

	/*
	 * If any children exist we must leave the chain intact with refs == 0.
	 * They exist because chains are retained below us which have refs or
	 * may require flushing.
	 *
	 * Retry (return chain) if we fail to transition the refs to 0, else
	 * return NULL indication nothing more to do.
	 *
	 * Chains with children are NOT put on the LRU list.
	 */
	if (chain->core.chain_count) {
		if (atomic_cmpset_int(&chain->refs, 1, 0)) {
			hammer2_spin_unex(&chain->core.spin);
			hammer2_chain_assert_no_data(chain);
			hammer2_mtx_unlock(&chain->lock);
			chain = NULL;
		} else {
			hammer2_spin_unex(&chain->core.spin);
			hammer2_mtx_unlock(&chain->lock);
		}
		return (chain);
	}
	/* spinlock still held */
	/* no chains left under us */

	/*
	 * chain->core has no children left so no accessors can get to our
	 * chain from there.  Now we have to lock the parent core to interlock
	 * remaining possible accessors that might bump chain's refs before
	 * we can safely drop chain's refs with intent to free the chain.
	 */
	hmp = chain->hmp;
	pmp = chain->pmp;	/* can be NULL */
	rdrop = NULL;

	parent = chain->parent;

	/*
	 * WARNING! chain's spin lock is still held here, and other spinlocks
	 *	    will be acquired and released in the code below.  We
	 *	    cannot be making fancy procedure calls!
	 */

	/*
	 * We can cache the chain if it is associated with a pmp
	 * and not flagged as being destroyed or requesting a full
	 * release.  In this situation the chain is not removed
	 * from its parent, i.e. it can still be looked up.
	 *
	 * We intentionally do not cache DATA chains because these
	 * were likely used to load data into the logical buffer cache
	 * and will not be accessed again for some time.
	 */
	if ((chain->flags &
	     (HAMMER2_CHAIN_DESTROY | HAMMER2_CHAIN_RELEASE)) == 0 &&
	    chain->pmp &&
	    chain->bref.type != HAMMER2_BREF_TYPE_DATA) {
		if (parent)
			hammer2_spin_ex(&parent->core.spin);
		if (atomic_cmpset_int(&chain->refs, 1, 0) == 0) {
			/*
			 * 1->0 transition failed, retry.  Do not drop
			 * the chain's data yet!
			 */
			if (parent)
				hammer2_spin_unex(&parent->core.spin);
			hammer2_spin_unex(&chain->core.spin);
			hammer2_mtx_unlock(&chain->lock);

			return(chain);
		}

		/*
		 * Success
		 */
#if 0
		dio = hammer2_chain_drop_data(chain, 1);
#endif
		hammer2_chain_assert_no_data(chain);

		/*
		 * Make sure we are on the LRU list, clean up excessive
		 * LRU entries.  We can only really drop one but there might
		 * be other entries that we can remove from the lru_list
		 * without dropping.
		 *
		 * NOTE: HAMMER2_CHAIN_ONLRU may only be safely set when
		 *	 chain->core.spin AND pmp->lru_spin are held, but
		 *	 can be safely cleared only holding pmp->lru_spin.
		 */
		if ((chain->flags & HAMMER2_CHAIN_ONLRU) == 0) {
			hammer2_spin_ex(&pmp->lru_spin);
			if ((chain->flags & HAMMER2_CHAIN_ONLRU) == 0) {
				atomic_set_int(&chain->flags,
					       HAMMER2_CHAIN_ONLRU);
				TAILQ_INSERT_TAIL(&pmp->lru_list,
						  chain, lru_node);
				atomic_add_int(&pmp->lru_count, 1);
			}
			if (pmp->lru_count < HAMMER2_LRU_LIMIT)
				depth = 1;	/* disable lru_list flush */
			hammer2_spin_unex(&pmp->lru_spin);
		} else {
			/* disable lru flush */
			depth = 1;
		}

		if (parent) {
			hammer2_spin_unex(&parent->core.spin);
			parent = NULL;	/* safety */
		}
		hammer2_spin_unex(&chain->core.spin);
		hammer2_mtx_unlock(&chain->lock);
#if 0
		if (dio)
			hammer2_io_bqrelse(&dio);
#endif

		/*
		 * lru_list hysteresis (see above for depth overrides).
		 * Note that depth also prevents excessive lastdrop recursion.
		 */
		if (depth == 0)
			hammer2_chain_lru_flush(pmp);

		return NULL;
		/* NOT REACHED */
	}

	/*
	 * Make sure we are not on the LRU list.
	 */
	if (chain->flags & HAMMER2_CHAIN_ONLRU) {
		hammer2_spin_ex(&pmp->lru_spin);
		if (chain->flags & HAMMER2_CHAIN_ONLRU) {
			atomic_add_int(&pmp->lru_count, -1);
			atomic_clear_int(&chain->flags, HAMMER2_CHAIN_ONLRU);
			TAILQ_REMOVE(&pmp->lru_list, chain, lru_node);
		}
		hammer2_spin_unex(&pmp->lru_spin);
	}

	/*
	 * Spinlock the parent and try to drop the last ref on chain.
	 * On success determine if we should dispose of the chain
	 * (remove the chain from its parent, etc).
	 *
	 * (normal core locks are top-down recursive but we define
	 * core spinlocks as bottom-up recursive, so this is safe).
	 */
	if (parent) {
		hammer2_spin_ex(&parent->core.spin);
		if (atomic_cmpset_int(&chain->refs, 1, 0) == 0) {
#if 0
			/* XXX remove, don't try to drop data on fail */
			hammer2_spin_unex(&parent->core.spin);
			dio = hammer2_chain_drop_data(chain, 0);
			hammer2_spin_unex(&chain->core.spin);
			if (dio)
				hammer2_io_bqrelse(&dio);
#endif
			/*
			 * 1->0 transition failed, retry.
			 */
			hammer2_spin_unex(&parent->core.spin);
			hammer2_spin_unex(&chain->core.spin);
			hammer2_mtx_unlock(&chain->lock);

			return(chain);
		}

		/*
		 * 1->0 transition successful, parent spin held to prevent
		 * new lookups, chain spinlock held to protect parent field.
		 * Remove chain from the parent.
		 *
		 * If the chain is being removed from the parent's btree but
		 * is not bmapped, we have to adjust live_count downward.  If
		 * it is bmapped then the blockref is retained in the parent
		 * as is its associated live_count.  This case can occur when
		 * a chain added to the topology is unable to flush and is
		 * then later deleted.
		 */
		if (chain->flags & HAMMER2_CHAIN_ONRBTREE) {
			if ((parent->flags & HAMMER2_CHAIN_COUNTEDBREFS) &&
			    (chain->flags & HAMMER2_CHAIN_BMAPPED) == 0) {
				atomic_add_int(&parent->core.live_count, -1);
			}
			RB_REMOVE(hammer2_chain_tree,
				  &parent->core.rbtree, chain);
			atomic_clear_int(&chain->flags, HAMMER2_CHAIN_ONRBTREE);
			--parent->core.chain_count;
			chain->parent = NULL;
		}

		/*
		 * If our chain was the last chain in the parent's core the
		 * core is now empty and its parent might have to be
		 * re-dropped if it has 0 refs.
		 */
		if (parent->core.chain_count == 0) {
			rdrop = parent;
			atomic_add_int(&rdrop->refs, 1);
			/*
			if (atomic_cmpset_int(&rdrop->refs, 0, 1) == 0)
				rdrop = NULL;
			*/
		}
		hammer2_spin_unex(&parent->core.spin);
		parent = NULL;	/* safety */
		/* FALL THROUGH */
	} else {
		/*
		 * No-parent case.
		 */
		if (atomic_cmpset_int(&chain->refs, 1, 0) == 0) {
			/*
			 * 1->0 transition failed, retry.
			 */
			hammer2_spin_unex(&parent->core.spin);
			hammer2_spin_unex(&chain->core.spin);
			hammer2_mtx_unlock(&chain->lock);

			return(chain);
		}
	}

	/*
	 * Successful 1->0 transition, no parent, no children... no way for
	 * anyone to ref this chain any more.  We can clean-up and free it.
	 *
	 * We still have the core spinlock, and core's chain_count is 0.
	 * Any parent spinlock is gone.
	 */
	hammer2_spin_unex(&chain->core.spin);
	hammer2_chain_assert_no_data(chain);
	hammer2_mtx_unlock(&chain->lock);
	KKASSERT(RB_EMPTY(&chain->core.rbtree) &&
		 chain->core.chain_count == 0);

	/*
	 * All locks are gone, no pointers remain to the chain, finish
	 * freeing it.
	 */
	KKASSERT((chain->flags & (HAMMER2_CHAIN_UPDATE |
				  HAMMER2_CHAIN_MODIFIED)) == 0);
#if 0
	dio = hammer2_chain_drop_data(chain, 1);
	if (dio)
		hammer2_io_bqrelse(&dio);
#endif

	/*
	 * Once chain resources are gone we can use the now dead chain
	 * structure to placehold what might otherwise require a recursive
	 * drop, because we have potentially two things to drop and can only
	 * return one directly.
	 */
	if (chain->flags & HAMMER2_CHAIN_ALLOCATED) {
		atomic_clear_int(&chain->flags, HAMMER2_CHAIN_ALLOCATED);
		chain->hmp = NULL;
		kfree(chain, hmp->mchain);
	}

	/*
	 * Possible chaining loop when parent re-drop needed.
	 */
	return(rdrop);
}

/*
 * Heuristical flush of the LRU, try to reduce the number of entries
 * on the LRU to (HAMMER2_LRU_LIMIT * 2 / 3).  This procedure is called
 * only when lru_count exceeds HAMMER2_LRU_LIMIT.
 */
static
void
hammer2_chain_lru_flush(hammer2_pfs_t *pmp)
{
	hammer2_chain_t *chain;

again:
	chain = NULL;
	hammer2_spin_ex(&pmp->lru_spin);
	while (pmp->lru_count > HAMMER2_LRU_LIMIT * 2 / 3) {
		/*
		 * Pick a chain off the lru_list, just recycle it quickly
		 * if LRUHINT is set (the chain was ref'd but left on
		 * the lru_list, so cycle to the end).
		 */
		chain = TAILQ_FIRST(&pmp->lru_list);
		TAILQ_REMOVE(&pmp->lru_list, chain, lru_node);

		if (chain->flags & HAMMER2_CHAIN_LRUHINT) {
			atomic_clear_int(&chain->flags, HAMMER2_CHAIN_LRUHINT);
			TAILQ_INSERT_TAIL(&pmp->lru_list, chain, lru_node);
			chain = NULL;
			continue;
		}

		/*
		 * Ok, we are off the LRU.  We must adjust refs before we
		 * can safely clear the ONLRU flag.
		 */
		atomic_add_int(&pmp->lru_count, -1);
		if (atomic_cmpset_int(&chain->refs, 0, 1)) {
			atomic_clear_int(&chain->flags, HAMMER2_CHAIN_ONLRU);
			atomic_set_int(&chain->flags, HAMMER2_CHAIN_RELEASE);
			break;
		}
		atomic_clear_int(&chain->flags, HAMMER2_CHAIN_ONLRU);
		chain = NULL;
	}
	hammer2_spin_unex(&pmp->lru_spin);
	if (chain == NULL)
		return;

	/*
	 * If we picked a chain off the lru list we may be able to lastdrop
	 * it.  Use a depth of 1 to prevent excessive lastdrop recursion.
	 */
	while (chain) {
		u_int refs;

		refs = chain->refs;
		cpu_ccfence();
		KKASSERT(refs > 0);

		if (refs == 1) {
			if (hammer2_mtx_ex_try(&chain->lock) == 0)
				chain = hammer2_chain_lastdrop(chain, 1);
			/* retry the same chain, or chain from lastdrop */
		} else {
			if (atomic_cmpset_int(&chain->refs, refs, refs - 1))
				break;
			/* retry the same chain */
		}
		cpu_pause();
	}
	goto again;
}

/*
 * On last lock release.
 */
static hammer2_io_t *
hammer2_chain_drop_data(hammer2_chain_t *chain)
{
	hammer2_io_t *dio;

	if ((dio = chain->dio) != NULL) {
		chain->dio = NULL;
		chain->data = NULL;
	} else {
		switch(chain->bref.type) {
		case HAMMER2_BREF_TYPE_VOLUME:
		case HAMMER2_BREF_TYPE_FREEMAP:
			break;
		default:
			if (chain->data != NULL) {
				hammer2_spin_unex(&chain->core.spin);
				panic("chain data not null: "
				      "chain %p bref %016jx.%02x "
				      "refs %d parent %p dio %p data %p",
				      chain, chain->bref.data_off,
				      chain->bref.type, chain->refs,
				      chain->parent,
				      chain->dio, chain->data);
			}
			KKASSERT(chain->data == NULL);
			break;
		}
	}
	return dio;
}

/*
 * Lock a referenced chain element, acquiring its data with I/O if necessary,
 * and specify how you would like the data to be resolved.
 *
 * If an I/O or other fatal error occurs, chain->error will be set to non-zero.
 *
 * The lock is allowed to recurse, multiple locking ops will aggregate
 * the requested resolve types.  Once data is assigned it will not be
 * removed until the last unlock.
 *
 * HAMMER2_RESOLVE_NEVER - Do not resolve the data element.
 *			   (typically used to avoid device/logical buffer
 *			    aliasing for data)
 *
 * HAMMER2_RESOLVE_MAYBE - Do not resolve data elements for chains in
 *			   the INITIAL-create state (indirect blocks only).
 *
 *			   Do not resolve data elements for DATA chains.
 *			   (typically used to avoid device/logical buffer
 *			    aliasing for data)
 *
 * HAMMER2_RESOLVE_ALWAYS- Always resolve the data element.
 *
 * HAMMER2_RESOLVE_SHARED- (flag) The chain is locked shared, otherwise
 *			   it will be locked exclusive.
 *
 * HAMMER2_RESOLVE_NONBLOCK- (flag) The chain is locked non-blocking.  If
 *			   the lock fails, EAGAIN is returned.
 *
 * NOTE: Embedded elements (volume header, inodes) are always resolved
 *	 regardless.
 *
 * NOTE: Specifying HAMMER2_RESOLVE_ALWAYS on a newly-created non-embedded
 *	 element will instantiate and zero its buffer, and flush it on
 *	 release.
 *
 * NOTE: (data) elements are normally locked RESOLVE_NEVER or RESOLVE_MAYBE
 *	 so as not to instantiate a device buffer, which could alias against
 *	 a logical file buffer.  However, if ALWAYS is specified the
 *	 device buffer will be instantiated anyway.
 *
 * NOTE: The return value is always 0 unless NONBLOCK is specified, in which
 *	 case it can be either 0 or EAGAIN.
 *
 * WARNING! This function blocks on I/O if data needs to be fetched.  This
 *	    blocking can run concurrent with other compatible lock holders
 *	    who do not need data returning.  The lock is not upgraded to
 *	    exclusive during a data fetch, a separate bit is used to
 *	    interlock I/O.  However, an exclusive lock holder can still count
 *	    on being interlocked against an I/O fetch managed by a shared
 *	    lock holder.
 */
int
hammer2_chain_lock(hammer2_chain_t *chain, int how)
{
	KKASSERT(chain->refs > 0);

	if (how & HAMMER2_RESOLVE_NONBLOCK) {
		/*
		 * For non-blocking operation attempt to get the lock
		 * before bumping lockcnt, just so we don't have to deal
		 * with dropping lockcnt (and dealing with the underlying
		 * data) if we fail.
		 *
		 * NOTE: LOCKAGAIN must always succeed without blocking.
		 */
		if (how & HAMMER2_RESOLVE_SHARED) {
			if (how & HAMMER2_RESOLVE_LOCKAGAIN) {
				hammer2_mtx_sh_again(&chain->lock);
			} else {
				if (hammer2_mtx_sh_try(&chain->lock) != 0)
					return EAGAIN;
			}
		} else {
			if (hammer2_mtx_ex_try(&chain->lock) != 0)
				return EAGAIN;
		}
		atomic_add_int(&chain->lockcnt, 1);
		++curthread->td_tracker;
	} else {
		/*
		 * Lock the element.  Recursive locks are allowed.  lockcnt
		 * ensures that data is left intact.
		 */
		atomic_add_int(&chain->lockcnt, 1);

		/*
		 * Get the appropriate lock.  If LOCKAGAIN is flagged with
		 * SHARED the caller expects a shared lock to already be
		 * present and we are giving it another ref.  This case must
		 * importantly not block if there is a pending exclusive lock
		 * request.
		 */
		if (how & HAMMER2_RESOLVE_SHARED) {
			if (how & HAMMER2_RESOLVE_LOCKAGAIN) {
				hammer2_mtx_sh_again(&chain->lock);
			} else {
				hammer2_mtx_sh(&chain->lock);
			}
		} else {
			hammer2_mtx_ex(&chain->lock);
		}
		++curthread->td_tracker;
	}

	/*
	 * If we already have a valid data pointer make sure the data is
	 * synchronized to the current cpu, and then no further action is
	 * necessary.
	 */
	if (chain->data) {
		if (chain->dio)
			hammer2_io_bkvasync(chain->dio);
		return 0;
	}

	/*
	 * Do we have to resolve the data?  This is generally only
	 * applicable to HAMMER2_BREF_TYPE_DATA which is special-cased.
	 * Other BREF types expects the data to be there.
	 */
	switch(how & HAMMER2_RESOLVE_MASK) {
	case HAMMER2_RESOLVE_NEVER:
		return 0;
	case HAMMER2_RESOLVE_MAYBE:
		if (chain->flags & HAMMER2_CHAIN_INITIAL)
			return 0;
		if (chain->bref.type == HAMMER2_BREF_TYPE_DATA)
			return 0;
#if 0
		if (chain->bref.type == HAMMER2_BREF_TYPE_FREEMAP_NODE)
			return 0;
		if (chain->bref.type == HAMMER2_BREF_TYPE_FREEMAP_LEAF)
			return 0;
#endif
		/* fall through */
	case HAMMER2_RESOLVE_ALWAYS:
	default:
		break;
	}

	/*
	 * Caller requires data
	 */
	hammer2_chain_load_data(chain);

	return 0;
}

/*
 * Lock the chain, retain the hold, and drop the data persistence count.
 * The data should remain valid because we never transitioned lockcnt
 * through 0.
 */
void
hammer2_chain_lock_unhold(hammer2_chain_t *chain, int how)
{
	hammer2_chain_lock(chain, how);
	atomic_add_int(&chain->lockcnt, -1);
}

#if 0
/*
 * Downgrade an exclusive chain lock to a shared chain lock.
 *
 * NOTE: There is no upgrade equivalent due to the ease of
 *	 deadlocks in that direction.
 */
void
hammer2_chain_lock_downgrade(hammer2_chain_t *chain)
{
	hammer2_mtx_downgrade(&chain->lock);
}
#endif

/*
 * Issue I/O and install chain->data.  Caller must hold a chain lock, lock
 * may be of any type.
 *
 * Once chain->data is set it cannot be disposed of until all locks are
 * released.
 *
 * Make sure the data is synchronized to the current cpu.
 */
void
hammer2_chain_load_data(hammer2_chain_t *chain)
{
	hammer2_blockref_t *bref;
	hammer2_dev_t *hmp;
	hammer2_io_t *dio;
	char *bdata;
	int error;

	/*
	 * Degenerate case, data already present, or chain has no media
	 * reference to load.
	 */
	if (chain->data) {
		if (chain->dio)
			hammer2_io_bkvasync(chain->dio);
		return;
	}
	if ((chain->bref.data_off & ~HAMMER2_OFF_MASK_RADIX) == 0)
		return;

	hmp = chain->hmp;
	KKASSERT(hmp != NULL);

	/*
	 * Gain the IOINPROG bit, interlocked block.
	 */
	for (;;) {
		u_int oflags;
		u_int nflags;

		oflags = chain->flags;
		cpu_ccfence();
		if (oflags & HAMMER2_CHAIN_IOINPROG) {
			nflags = oflags | HAMMER2_CHAIN_IOSIGNAL;
			tsleep_interlock(&chain->flags, 0);
			if (atomic_cmpset_int(&chain->flags, oflags, nflags)) {
				tsleep(&chain->flags, PINTERLOCKED,
					"h2iocw", 0);
			}
			/* retry */
		} else {
			nflags = oflags | HAMMER2_CHAIN_IOINPROG;
			if (atomic_cmpset_int(&chain->flags, oflags, nflags)) {
				break;
			}
			/* retry */
		}
	}

	/*
	 * We own CHAIN_IOINPROG
	 *
	 * Degenerate case if we raced another load.
	 */
	if (chain->data) {
		if (chain->dio)
			hammer2_io_bkvasync(chain->dio);
		goto done;
	}

	/*
	 * We must resolve to a device buffer, either by issuing I/O or
	 * by creating a zero-fill element.  We do not mark the buffer
	 * dirty when creating a zero-fill element (the hammer2_chain_modify()
	 * API must still be used to do that).
	 *
	 * The device buffer is variable-sized in powers of 2 down
	 * to HAMMER2_MIN_ALLOC (typically 1K).  A 64K physical storage
	 * chunk always contains buffers of the same size. (XXX)
	 *
	 * The minimum physical IO size may be larger than the variable
	 * block size.
	 */
	bref = &chain->bref;

	/*
	 * The getblk() optimization can only be used on newly created
	 * elements if the physical block size matches the request.
	 */
	if (chain->flags & HAMMER2_CHAIN_INITIAL) {
		error = hammer2_io_new(hmp, bref->type,
				       bref->data_off, chain->bytes,
				       &chain->dio);
	} else {
		error = hammer2_io_bread(hmp, bref->type,
					 bref->data_off, chain->bytes,
					 &chain->dio);
		hammer2_adjreadcounter(&chain->bref, chain->bytes);
	}
	if (error) {
		chain->error = HAMMER2_ERROR_EIO;
		kprintf("hammer2_chain_lock: I/O error %016jx: %d\n",
			(intmax_t)bref->data_off, error);
		hammer2_io_bqrelse(&chain->dio);
		goto done;
	}
	chain->error = 0;

	/*
	 * This isn't perfect and can be ignored on OSs which do not have
	 * an indication as to whether a buffer is coming from cache or
	 * if I/O was actually issued for the read.  TESTEDGOOD will work
	 * pretty well without the B_IOISSUED logic because chains are
	 * cached, but in that situation (without B_IOISSUED) it will not
	 * detect whether a re-read via I/O is corrupted verses the original
	 * read.
	 *
	 * We can't re-run the CRC on every fresh lock.  That would be
	 * insanely expensive.
	 *
	 * If the underlying kernel buffer covers the entire chain we can
	 * use the B_IOISSUED indication to determine if we have to re-run
	 * the CRC on chain data for chains that managed to stay cached
	 * across the kernel disposal of the original buffer.
	 */
	if ((dio = chain->dio) != NULL && dio->bp) {
		struct buf *bp = dio->bp;

		if (dio->psize == chain->bytes &&
		    (bp->b_flags & B_IOISSUED)) {
			atomic_clear_int(&chain->flags,
					 HAMMER2_CHAIN_TESTEDGOOD);
			bp->b_flags &= ~B_IOISSUED;
		}
	}

	/*
	 * NOTE: A locked chain's data cannot be modified without first
	 *	 calling hammer2_chain_modify().
	 */

	/*
	 * Clear INITIAL.  In this case we used io_new() and the buffer has
	 * been zero'd and marked dirty.
	 *
	 * NOTE: hammer2_io_data() call issues bkvasync()
	 */
	bdata = hammer2_io_data(chain->dio, chain->bref.data_off);

	if (chain->flags & HAMMER2_CHAIN_INITIAL) {
		atomic_clear_int(&chain->flags, HAMMER2_CHAIN_INITIAL);
		chain->bref.flags |= HAMMER2_BREF_FLAG_ZERO;
	} else if (chain->flags & HAMMER2_CHAIN_MODIFIED) {
		/*
		 * check data not currently synchronized due to
		 * modification.  XXX assumes data stays in the buffer
		 * cache, which might not be true (need biodep on flush
		 * to calculate crc?  or simple crc?).
		 */
	} else if ((chain->flags & HAMMER2_CHAIN_TESTEDGOOD) == 0) {
		if (hammer2_chain_testcheck(chain, bdata) == 0) {
			chain->error = HAMMER2_ERROR_CHECK;
		} else {
			atomic_set_int(&chain->flags, HAMMER2_CHAIN_TESTEDGOOD);
		}
	}

	/*
	 * Setup the data pointer, either pointing it to an embedded data
	 * structure and copying the data from the buffer, or pointing it
	 * into the buffer.
	 *
	 * The buffer is not retained when copying to an embedded data
	 * structure in order to avoid potential deadlocks or recursions
	 * on the same physical buffer.
	 *
	 * WARNING! Other threads can start using the data the instant we
	 *	    set chain->data non-NULL.
	 */
	switch (bref->type) {
	case HAMMER2_BREF_TYPE_VOLUME:
	case HAMMER2_BREF_TYPE_FREEMAP:
		/*
		 * Copy data from bp to embedded buffer
		 */
		panic("hammer2_chain_load_data: unresolved volume header");
		break;
	case HAMMER2_BREF_TYPE_DIRENT:
		KKASSERT(chain->bytes != 0);
		/* fall through */
	case HAMMER2_BREF_TYPE_INODE:
	case HAMMER2_BREF_TYPE_FREEMAP_LEAF:
	case HAMMER2_BREF_TYPE_INDIRECT:
	case HAMMER2_BREF_TYPE_DATA:
	case HAMMER2_BREF_TYPE_FREEMAP_NODE:
	default:
		/*
		 * Point data at the device buffer and leave dio intact.
		 */
		chain->data = (void *)bdata;
		break;
	}

	/*
	 * Release HAMMER2_CHAIN_IOINPROG and signal waiters if requested.
	 */
done:
	for (;;) {
		u_int oflags;
		u_int nflags;

		oflags = chain->flags;
		nflags = oflags & ~(HAMMER2_CHAIN_IOINPROG |
				    HAMMER2_CHAIN_IOSIGNAL);
		KKASSERT(oflags & HAMMER2_CHAIN_IOINPROG);
		if (atomic_cmpset_int(&chain->flags, oflags, nflags)) {
			if (oflags & HAMMER2_CHAIN_IOSIGNAL)
				wakeup(&chain->flags);
			break;
		}
	}
}

/*
 * Unlock and deref a chain element.
 *
 * Remember that the presence of children under chain prevent the chain's
 * destruction but do not add additional references, so the dio will still
 * be dropped.
 */
void
hammer2_chain_unlock(hammer2_chain_t *chain)
{
	hammer2_io_t *dio;
	u_int lockcnt;
	int iter = 0;

	--curthread->td_tracker;

	/*
	 * If multiple locks are present (or being attempted) on this
	 * particular chain we can just unlock, drop refs, and return.
	 *
	 * Otherwise fall-through on the 1->0 transition.
	 */
	for (;;) {
		lockcnt = chain->lockcnt;
		KKASSERT(lockcnt > 0);
		cpu_ccfence();
		if (lockcnt > 1) {
			if (atomic_cmpset_int(&chain->lockcnt,
					      lockcnt, lockcnt - 1)) {
				hammer2_mtx_unlock(&chain->lock);
				return;
			}
		} else if (hammer2_mtx_upgrade_try(&chain->lock) == 0) {
			/* while holding the mutex exclusively */
			if (atomic_cmpset_int(&chain->lockcnt, 1, 0))
				break;
		} else {
			/*
			 * This situation can easily occur on SMP due to
			 * the gap inbetween the 1->0 transition and the
			 * final unlock.  We cannot safely block on the
			 * mutex because lockcnt might go above 1.
			 *
			 * XXX Sleep for one tick if it takes too long.
			 */
			if (++iter > 1000) {
				if (iter > 1000 + hz) {
					kprintf("hammer2: h2race2 %p\n", chain);
					iter = 1000;
				}
				tsleep(&iter, 0, "h2race2", 1);
			}
			cpu_pause();
		}
		/* retry */
	}

	/*
	 * Last unlock / mutex upgraded to exclusive.  Drop the data
	 * reference.
	 */
	dio = hammer2_chain_drop_data(chain);
	if (dio)
		hammer2_io_bqrelse(&dio);
	hammer2_mtx_unlock(&chain->lock);
}

/*
 * Unlock and hold chain data intact
 */
void
hammer2_chain_unlock_hold(hammer2_chain_t *chain)
{
	atomic_add_int(&chain->lockcnt, 1);
	hammer2_chain_unlock(chain);
}

/*
 * Helper to obtain the blockref[] array base and count for a chain.
 *
 * XXX Not widely used yet, various use cases need to be validated and
 *     converted to use this function.
 */
static
hammer2_blockref_t *
hammer2_chain_base_and_count(hammer2_chain_t *parent, int *countp)
{
	hammer2_blockref_t *base;
	int count;

	if (parent->flags & HAMMER2_CHAIN_INITIAL) {
		base = NULL;

		switch(parent->bref.type) {
		case HAMMER2_BREF_TYPE_INODE:
			count = HAMMER2_SET_COUNT;
			break;
		case HAMMER2_BREF_TYPE_INDIRECT:
		case HAMMER2_BREF_TYPE_FREEMAP_NODE:
			count = parent->bytes / sizeof(hammer2_blockref_t);
			break;
		case HAMMER2_BREF_TYPE_VOLUME:
			count = HAMMER2_SET_COUNT;
			break;
		case HAMMER2_BREF_TYPE_FREEMAP:
			count = HAMMER2_SET_COUNT;
			break;
		default:
			panic("hammer2_chain_create_indirect: "
			      "unrecognized blockref type: %d",
			      parent->bref.type);
			count = 0;
			break;
		}
	} else {
		switch(parent->bref.type) {
		case HAMMER2_BREF_TYPE_INODE:
			base = &parent->data->ipdata.u.blockset.blockref[0];
			count = HAMMER2_SET_COUNT;
			break;
		case HAMMER2_BREF_TYPE_INDIRECT:
		case HAMMER2_BREF_TYPE_FREEMAP_NODE:
			base = &parent->data->npdata[0];
			count = parent->bytes / sizeof(hammer2_blockref_t);
			break;
		case HAMMER2_BREF_TYPE_VOLUME:
			base = &parent->data->voldata.
					sroot_blockset.blockref[0];
			count = HAMMER2_SET_COUNT;
			break;
		case HAMMER2_BREF_TYPE_FREEMAP:
			base = &parent->data->blkset.blockref[0];
			count = HAMMER2_SET_COUNT;
			break;
		default:
			panic("hammer2_chain_create_indirect: "
			      "unrecognized blockref type: %d",
			      parent->bref.type);
			count = 0;
			break;
		}
	}
	*countp = count;

	return base;
}

/*
 * This counts the number of live blockrefs in a block array and
 * also calculates the point at which all remaining blockrefs are empty.
 * This routine can only be called on a live chain.
 *
 * Caller holds the chain locked, but possibly with a shared lock.  We
 * must use an exclusive spinlock to prevent corruption.
 *
 * NOTE: Flag is not set until after the count is complete, allowing
 *	 callers to test the flag without holding the spinlock.
 *
 * NOTE: If base is NULL the related chain is still in the INITIAL
 *	 state and there are no blockrefs to count.
 *
 * NOTE: live_count may already have some counts accumulated due to
 *	 creation and deletion and could even be initially negative.
 */
void
hammer2_chain_countbrefs(hammer2_chain_t *chain,
			 hammer2_blockref_t *base, int count)
{
	hammer2_spin_ex(&chain->core.spin);
        if ((chain->flags & HAMMER2_CHAIN_COUNTEDBREFS) == 0) {
		if (base) {
			while (--count >= 0) {
				if (base[count].type)
					break;
			}
			chain->core.live_zero = count + 1;
			while (count >= 0) {
				if (base[count].type)
					atomic_add_int(&chain->core.live_count,
						       1);
				--count;
			}
		} else {
			chain->core.live_zero = 0;
		}
		/* else do not modify live_count */
		atomic_set_int(&chain->flags, HAMMER2_CHAIN_COUNTEDBREFS);
	}
	hammer2_spin_unex(&chain->core.spin);
}

/*
 * Resize the chain's physical storage allocation in-place.  This function does
 * not usually adjust the data pointer and must be followed by (typically) a
 * hammer2_chain_modify() call to copy any old data over and adjust the
 * data pointer.
 *
 * Chains can be resized smaller without reallocating the storage.  Resizing
 * larger will reallocate the storage.  Excess or prior storage is reclaimed
 * asynchronously at a later time.
 *
 * An nradix value of 0 is special-cased to mean that the storage should
 * be disassociated, that is the chain is being resized to 0 bytes (not 1
 * byte).
 *
 * Must be passed an exclusively locked parent and chain.
 *
 * This function is mostly used with DATA blocks locked RESOLVE_NEVER in order
 * to avoid instantiating a device buffer that conflicts with the vnode data
 * buffer.  However, because H2 can compress or encrypt data, the chain may
 * have a dio assigned to it in those situations, and they do not conflict.
 *
 * XXX return error if cannot resize.
 */
int
hammer2_chain_resize(hammer2_chain_t *chain,
		     hammer2_tid_t mtid, hammer2_off_t dedup_off,
		     int nradix, int flags)
{
	hammer2_dev_t *hmp;
	size_t obytes;
	size_t nbytes;
	int error;

	hmp = chain->hmp;

	/*
	 * Only data and indirect blocks can be resized for now.
	 * (The volu root, inodes, and freemap elements use a fixed size).
	 */
	KKASSERT(chain != &hmp->vchain);
	KKASSERT(chain->bref.type == HAMMER2_BREF_TYPE_DATA ||
		 chain->bref.type == HAMMER2_BREF_TYPE_INDIRECT ||
		 chain->bref.type == HAMMER2_BREF_TYPE_DIRENT);

	/*
	 * Nothing to do if the element is already the proper size
	 */
	obytes = chain->bytes;
	nbytes = (nradix) ? (1U << nradix) : 0;
	if (obytes == nbytes)
		return (chain->error);

	/*
	 * Make sure the old data is instantiated so we can copy it.  If this
	 * is a data block, the device data may be superfluous since the data
	 * might be in a logical block, but compressed or encrypted data is
	 * another matter.
	 *
	 * NOTE: The modify will set BMAPUPD for us if BMAPPED is set.
	 */
	error = hammer2_chain_modify(chain, mtid, dedup_off, 0);
	if (error)
		return error;

	/*
	 * Relocate the block, even if making it smaller (because different
	 * block sizes may be in different regions).
	 *
	 * NOTE: Operation does not copy the data and may only be used
	 *	  to resize data blocks in-place, or directory entry blocks
	 *	  which are about to be modified in some manner.
	 */
	error = hammer2_freemap_alloc(chain, nbytes);
	if (error)
		return error;

	chain->bytes = nbytes;

	/*
	 * We don't want the followup chain_modify() to try to copy data
	 * from the old (wrong-sized) buffer.  It won't know how much to
	 * copy.  This case should only occur during writes when the
	 * originator already has the data to write in-hand.
	 */
	if (chain->dio) {
		KKASSERT(chain->bref.type == HAMMER2_BREF_TYPE_DATA ||
			 chain->bref.type == HAMMER2_BREF_TYPE_DIRENT);
		hammer2_io_brelse(&chain->dio);
		chain->data = NULL;
	}
	return (chain->error);
}

/*
 * Set the chain modified so its data can be changed by the caller, or
 * install deduplicated data.  The caller must call this routine for each
 * set of modifications it makes, even if the chain is already flagged
 * MODIFIED.
 *
 * Sets bref.modify_tid to mtid only if mtid != 0.  Note that bref.modify_tid
 * is a CLC (cluster level change) field and is not updated by parent
 * propagation during a flush.
 *
 * Returns an appropriate HAMMER2_ERROR_* code, which will generally reflect
 * chain->error except for HAMMER2_ERROR_ENOSPC.  If the allocation fails
 * due to no space available, HAMMER2_ERROR_ENOSPC is returned and the chain
 * remains unmodified with its old data ref intact and chain->error
 * unchanged.
 *
 *				 Dedup Handling
 *
 * If the DEDUPABLE flag is set in the chain the storage must be reallocated
 * even if the chain is still flagged MODIFIED.  In this case the chain's
 * DEDUPABLE flag will be cleared once the new storage has been assigned.
 *
 * If the caller passes a non-zero dedup_off we will use it to assign the
 * new storage.  The MODIFIED flag will be *CLEARED* in this case, and
 * DEDUPABLE will be set (NOTE: the UPDATE flag is always set).  The caller
 * must not modify the data content upon return.
 */
int
hammer2_chain_modify(hammer2_chain_t *chain, hammer2_tid_t mtid,
		     hammer2_off_t dedup_off, int flags)
{
	hammer2_blockref_t obref;
	hammer2_dev_t *hmp;
	hammer2_io_t *dio;
	int error;
	int wasinitial;
	int setmodified;
	int setupdate;
	int newmod;
	char *bdata;

	hmp = chain->hmp;
	obref = chain->bref;
	KKASSERT((chain->flags & HAMMER2_CHAIN_FICTITIOUS) == 0);

	/*
	 * Data is not optional for freemap chains (we must always be sure
	 * to copy the data on COW storage allocations).
	 */
	if (chain->bref.type == HAMMER2_BREF_TYPE_FREEMAP_NODE ||
	    chain->bref.type == HAMMER2_BREF_TYPE_FREEMAP_LEAF) {
		KKASSERT((chain->flags & HAMMER2_CHAIN_INITIAL) ||
			 (flags & HAMMER2_MODIFY_OPTDATA) == 0);
	}

	/*
	 * Data must be resolved if already assigned, unless explicitly
	 * flagged otherwise.  If we cannot safety load the data the
	 * modification fails and we return early.
	 */
	if (chain->data == NULL && chain->bytes != 0 &&
	    (flags & HAMMER2_MODIFY_OPTDATA) == 0 &&
	    (chain->bref.data_off & ~HAMMER2_OFF_MASK_RADIX)) {
		hammer2_chain_load_data(chain);
		if (chain->error)
			return (chain->error);
	}
	error = 0;

	/*
	 * Set MODIFIED to indicate that the chain has been modified.  A new
	 * allocation is required when modifying a chain.
	 *
	 * Set UPDATE to ensure that the blockref is updated in the parent.
	 *
	 * If MODIFIED is already set determine if we can reuse the assigned
	 * data block or if we need a new data block.
	 */
	if ((chain->flags & HAMMER2_CHAIN_MODIFIED) == 0) {
		/*
		 * Must set modified bit.
		 */
		atomic_add_long(&hammer2_count_modified_chains, 1);
		atomic_set_int(&chain->flags, HAMMER2_CHAIN_MODIFIED);
		hammer2_pfs_memory_inc(chain->pmp);  /* can be NULL */
		setmodified = 1;

		/*
		 * We may be able to avoid a copy-on-write if the chain's
		 * check mode is set to NONE and the chain's current
		 * modify_tid is beyond the last explicit snapshot tid.
		 *
		 * This implements HAMMER2's overwrite-in-place feature.
		 *
		 * NOTE! This data-block cannot be used as a de-duplication
		 *	 source when the check mode is set to NONE.
		 */
		if ((chain->bref.type == HAMMER2_BREF_TYPE_DATA ||
		     chain->bref.type == HAMMER2_BREF_TYPE_DIRENT) &&
		    (chain->flags & HAMMER2_CHAIN_INITIAL) == 0 &&
		    (chain->flags & HAMMER2_CHAIN_DEDUPABLE) == 0 &&
		    HAMMER2_DEC_CHECK(chain->bref.methods) ==
		     HAMMER2_CHECK_NONE &&
		    chain->pmp &&
		    chain->bref.modify_tid >
		     chain->pmp->iroot->meta.pfs_lsnap_tid) {
			/*
			 * Sector overwrite allowed.
			 */
			newmod = 0;
		} else {
			/*
			 * Sector overwrite not allowed, must copy-on-write.
			 */
			newmod = 1;
		}
	} else if (chain->flags & HAMMER2_CHAIN_DEDUPABLE) {
		/*
		 * If the modified chain was registered for dedup we need
		 * a new allocation.  This only happens for delayed-flush
		 * chains (i.e. which run through the front-end buffer
		 * cache).
		 */
		newmod = 1;
		setmodified = 0;
	} else {
		/*
		 * Already flagged modified, no new allocation is needed.
		 */
		newmod = 0;
		setmodified = 0;
	}

	/*
	 * Flag parent update required.
	 */
	if ((chain->flags & HAMMER2_CHAIN_UPDATE) == 0) {
		atomic_set_int(&chain->flags, HAMMER2_CHAIN_UPDATE);
		setupdate = 1;
	} else {
		setupdate = 0;
	}

	/*
	 * The modification or re-modification requires an allocation and
	 * possible COW.  If an error occurs, the previous content and data
	 * reference is retained and the modification fails.
	 *
	 * If dedup_off is non-zero, the caller is requesting a deduplication
	 * rather than a modification.  The MODIFIED bit is not set and the
	 * data offset is set to the deduplication offset.  The data cannot
	 * be modified.
	 *
	 * NOTE: The dedup offset is allowed to be in a partially free state
	 *	 and we must be sure to reset it to a fully allocated state
	 *	 to force two bulkfree passes to free it again.
	 *
	 * NOTE: Only applicable when chain->bytes != 0.
	 *
	 * XXX can a chain already be marked MODIFIED without a data
	 * assignment?  If not, assert here instead of testing the case.
	 */
	if (chain != &hmp->vchain && chain != &hmp->fchain &&
	    chain->bytes) {
		if ((chain->bref.data_off & ~HAMMER2_OFF_MASK_RADIX) == 0 ||
		     newmod
		) {
			/*
			 * NOTE: We do not have to remove the dedup
			 *	 registration because the area is still
			 *	 allocated and the underlying DIO will
			 *	 still be flushed.
			 */
			if (dedup_off) {
				chain->bref.data_off = dedup_off;
				chain->bytes = 1 << (dedup_off &
						     HAMMER2_OFF_MASK_RADIX);
				chain->error = 0;
				atomic_clear_int(&chain->flags,
						 HAMMER2_CHAIN_MODIFIED);
				atomic_add_long(&hammer2_count_modified_chains,
						-1);
				if (chain->pmp)
					hammer2_pfs_memory_wakeup(chain->pmp);
				hammer2_freemap_adjust(hmp, &chain->bref,
						HAMMER2_FREEMAP_DORECOVER);
				atomic_set_int(&chain->flags,
						HAMMER2_CHAIN_DEDUPABLE);
			} else {
				error = hammer2_freemap_alloc(chain,
							      chain->bytes);
				atomic_clear_int(&chain->flags,
						HAMMER2_CHAIN_DEDUPABLE);
			}
		}
	}

	/*
	 * Stop here if error.  We have to undo any flag bits we might
	 * have set above.
	 */
	if (error) {
		if (setmodified) {
			atomic_clear_int(&chain->flags, HAMMER2_CHAIN_MODIFIED);
			atomic_add_long(&hammer2_count_modified_chains, -1);
			if (chain->pmp)
				hammer2_pfs_memory_wakeup(chain->pmp);
		}
		if (setupdate) {
			atomic_clear_int(&chain->flags, HAMMER2_CHAIN_UPDATE);
		}
		return error;
	}

	/*
	 * Update mirror_tid and modify_tid.  modify_tid is only updated
	 * if not passed as zero (during flushes, parent propagation passes
	 * the value 0).
	 *
	 * NOTE: chain->pmp could be the device spmp.
	 */
	chain->bref.mirror_tid = hmp->voldata.mirror_tid + 1;
	if (mtid)
		chain->bref.modify_tid = mtid;

	/*
	 * Set BMAPUPD to tell the flush code that an existing blockmap entry
	 * requires updating as well as to tell the delete code that the
	 * chain's blockref might not exactly match (in terms of physical size
	 * or block offset) the one in the parent's blocktable.  The base key
	 * of course will still match.
	 */
	if (chain->flags & HAMMER2_CHAIN_BMAPPED)
		atomic_set_int(&chain->flags, HAMMER2_CHAIN_BMAPUPD);

	/*
	 * Short-cut data blocks which the caller does not need an actual
	 * data reference to (aka OPTDATA), as long as the chain does not
	 * already have a data pointer to the data.  This generally means
	 * that the modifications are being done via the logical buffer cache.
	 * The INITIAL flag relates only to the device data buffer and thus
	 * remains unchange in this situation.
	 *
	 * This code also handles bytes == 0 (most dirents).
	 */
	if (chain->bref.type == HAMMER2_BREF_TYPE_DATA &&
	    (flags & HAMMER2_MODIFY_OPTDATA) &&
	    chain->data == NULL) {
		KKASSERT(chain->dio == NULL);
		goto skip2;
	}

	/*
	 * Clearing the INITIAL flag (for indirect blocks) indicates that
	 * we've processed the uninitialized storage allocation.
	 *
	 * If this flag is already clear we are likely in a copy-on-write
	 * situation but we have to be sure NOT to bzero the storage if
	 * no data is present.
	 */
	if (chain->flags & HAMMER2_CHAIN_INITIAL) {
		atomic_clear_int(&chain->flags, HAMMER2_CHAIN_INITIAL);
		wasinitial = 1;
	} else {
		wasinitial = 0;
	}

	/*
	 * Instantiate data buffer and possibly execute COW operation
	 */
	switch(chain->bref.type) {
	case HAMMER2_BREF_TYPE_VOLUME:
	case HAMMER2_BREF_TYPE_FREEMAP:
		/*
		 * The data is embedded, no copy-on-write operation is
		 * needed.
		 */
		KKASSERT(chain->dio == NULL);
		break;
	case HAMMER2_BREF_TYPE_DIRENT:
		/*
		 * The data might be fully embedded.
		 */
		if (chain->bytes == 0) {
			KKASSERT(chain->dio == NULL);
			break;
		}
		/* fall through */
	case HAMMER2_BREF_TYPE_INODE:
	case HAMMER2_BREF_TYPE_FREEMAP_LEAF:
	case HAMMER2_BREF_TYPE_DATA:
	case HAMMER2_BREF_TYPE_INDIRECT:
	case HAMMER2_BREF_TYPE_FREEMAP_NODE:
		/*
		 * Perform the copy-on-write operation
		 *
		 * zero-fill or copy-on-write depending on whether
		 * chain->data exists or not and set the dirty state for
		 * the new buffer.  hammer2_io_new() will handle the
		 * zero-fill.
		 *
		 * If a dedup_off was supplied this is an existing block
		 * and no COW, copy, or further modification is required.
		 */
		KKASSERT(chain != &hmp->vchain && chain != &hmp->fchain);

		if (wasinitial && dedup_off == 0) {
			error = hammer2_io_new(hmp, chain->bref.type,
					       chain->bref.data_off,
					       chain->bytes, &dio);
		} else {
			error = hammer2_io_bread(hmp, chain->bref.type,
						 chain->bref.data_off,
						 chain->bytes, &dio);
		}
		hammer2_adjreadcounter(&chain->bref, chain->bytes);

		/*
		 * If an I/O error occurs make sure callers cannot accidently
		 * modify the old buffer's contents and corrupt the filesystem.
		 *
		 * NOTE: hammer2_io_data() call issues bkvasync()
		 */
		if (error) {
			kprintf("hammer2_chain_modify: hmp=%p I/O error\n",
				hmp);
			chain->error = HAMMER2_ERROR_EIO;
			hammer2_io_brelse(&dio);
			hammer2_io_brelse(&chain->dio);
			chain->data = NULL;
			break;
		}
		chain->error = 0;
		bdata = hammer2_io_data(dio, chain->bref.data_off);

		if (chain->data) {
			/*
			 * COW (unless a dedup).
			 */
			KKASSERT(chain->dio != NULL);
			if (chain->data != (void *)bdata && dedup_off == 0) {
				bcopy(chain->data, bdata, chain->bytes);
			}
		} else if (wasinitial == 0) {
			/*
			 * We have a problem.  We were asked to COW but
			 * we don't have any data to COW with!
			 */
			panic("hammer2_chain_modify: having a COW %p\n",
			      chain);
		}

		/*
		 * Retire the old buffer, replace with the new.  Dirty or
		 * redirty the new buffer.
		 *
		 * WARNING! The system buffer cache may have already flushed
		 *	    the buffer, so we must be sure to [re]dirty it
		 *	    for further modification.
		 *
		 *	    If dedup_off was supplied, the caller is not
		 *	    expected to make any further modification to the
		 *	    buffer.
		 */
		if (chain->dio)
			hammer2_io_bqrelse(&chain->dio);
		chain->data = (void *)bdata;
		chain->dio = dio;
		if (dedup_off == 0)
			hammer2_io_setdirty(dio);
		break;
	default:
		panic("hammer2_chain_modify: illegal non-embedded type %d",
		      chain->bref.type);
		break;

	}
skip2:
	/*
	 * setflush on parent indicating that the parent must recurse down
	 * to us.  Do not call on chain itself which might already have it
	 * set.
	 */
	if (chain->parent)
		hammer2_chain_setflush(chain->parent);
	return (chain->error);
}

/*
 * Modify the chain associated with an inode.
 */
int
hammer2_chain_modify_ip(hammer2_inode_t *ip, hammer2_chain_t *chain,
			hammer2_tid_t mtid, int flags)
{
	int error;

	hammer2_inode_modify(ip);
	error = hammer2_chain_modify(chain, mtid, 0, flags);

	return error;
}

/*
 * Volume header data locks
 */
void
hammer2_voldata_lock(hammer2_dev_t *hmp)
{
	lockmgr(&hmp->vollk, LK_EXCLUSIVE);
}

void
hammer2_voldata_unlock(hammer2_dev_t *hmp)
{
	lockmgr(&hmp->vollk, LK_RELEASE);
}

void
hammer2_voldata_modify(hammer2_dev_t *hmp)
{
	if ((hmp->vchain.flags & HAMMER2_CHAIN_MODIFIED) == 0) {
		atomic_add_long(&hammer2_count_modified_chains, 1);
		atomic_set_int(&hmp->vchain.flags, HAMMER2_CHAIN_MODIFIED);
		hammer2_pfs_memory_inc(hmp->vchain.pmp);
	}
}

/*
 * This function returns the chain at the nearest key within the specified
 * range.  The returned chain will be referenced but not locked.
 *
 * This function will recurse through chain->rbtree as necessary and will
 * return a *key_nextp suitable for iteration.  *key_nextp is only set if
 * the iteration value is less than the current value of *key_nextp.
 *
 * The caller should use (*key_nextp) to calculate the actual range of
 * the returned element, which will be (key_beg to *key_nextp - 1), because
 * there might be another element which is superior to the returned element
 * and overlaps it.
 *
 * (*key_nextp) can be passed as key_beg in an iteration only while non-NULL
 * chains continue to be returned.  On EOF (*key_nextp) may overflow since
 * it will wind up being (key_end + 1).
 *
 * WARNING!  Must be called with child's spinlock held.  Spinlock remains
 *	     held through the operation.
 */
struct hammer2_chain_find_info {
	hammer2_chain_t		*best;
	hammer2_key_t		key_beg;
	hammer2_key_t		key_end;
	hammer2_key_t		key_next;
};

static int hammer2_chain_find_cmp(hammer2_chain_t *child, void *data);
static int hammer2_chain_find_callback(hammer2_chain_t *child, void *data);

static
hammer2_chain_t *
hammer2_chain_find(hammer2_chain_t *parent, hammer2_key_t *key_nextp,
			  hammer2_key_t key_beg, hammer2_key_t key_end)
{
	struct hammer2_chain_find_info info;

	info.best = NULL;
	info.key_beg = key_beg;
	info.key_end = key_end;
	info.key_next = *key_nextp;

	RB_SCAN(hammer2_chain_tree, &parent->core.rbtree,
		hammer2_chain_find_cmp, hammer2_chain_find_callback,
		&info);
	*key_nextp = info.key_next;
#if 0
	kprintf("chain_find %p %016jx:%016jx next=%016jx\n",
		parent, key_beg, key_end, *key_nextp);
#endif

	return (info.best);
}

static
int
hammer2_chain_find_cmp(hammer2_chain_t *child, void *data)
{
	struct hammer2_chain_find_info *info = data;
	hammer2_key_t child_beg;
	hammer2_key_t child_end;

	child_beg = child->bref.key;
	child_end = child_beg + ((hammer2_key_t)1 << child->bref.keybits) - 1;

	if (child_end < info->key_beg)
		return(-1);
	if (child_beg > info->key_end)
		return(1);
	return(0);
}

static
int
hammer2_chain_find_callback(hammer2_chain_t *child, void *data)
{
	struct hammer2_chain_find_info *info = data;
	hammer2_chain_t *best;
	hammer2_key_t child_end;

	/*
	 * WARNING! Layerq is scanned forwards, exact matches should keep
	 *	    the existing info->best.
	 */
	if ((best = info->best) == NULL) {
		/*
		 * No previous best.  Assign best
		 */
		info->best = child;
	} else if (best->bref.key <= info->key_beg &&
		   child->bref.key <= info->key_beg) {
		/*
		 * Illegal overlap.
		 */
		KKASSERT(0);
		/*info->best = child;*/
	} else if (child->bref.key < best->bref.key) {
		/*
		 * Child has a nearer key and best is not flush with key_beg.
		 * Set best to child.  Truncate key_next to the old best key.
		 */
		info->best = child;
		if (info->key_next > best->bref.key || info->key_next == 0)
			info->key_next = best->bref.key;
	} else if (child->bref.key == best->bref.key) {
		/*
		 * If our current best is flush with the child then this
		 * is an illegal overlap.
		 *
		 * key_next will automatically be limited to the smaller of
		 * the two end-points.
		 */
		KKASSERT(0);
		info->best = child;
	} else {
		/*
		 * Keep the current best but truncate key_next to the child's
		 * base.
		 *
		 * key_next will also automatically be limited to the smaller
		 * of the two end-points (probably not necessary for this case
		 * but we do it anyway).
		 */
		if (info->key_next > child->bref.key || info->key_next == 0)
			info->key_next = child->bref.key;
	}

	/*
	 * Always truncate key_next based on child's end-of-range.
	 */
	child_end = child->bref.key + ((hammer2_key_t)1 << child->bref.keybits);
	if (child_end && (info->key_next > child_end || info->key_next == 0))
		info->key_next = child_end;

	return(0);
}

/*
 * Retrieve the specified chain from a media blockref, creating the
 * in-memory chain structure which reflects it.  The returned chain is
 * held and locked according to (how) (HAMMER2_RESOLVE_*).  The caller must
 * handle crc-checks and so forth, and should check chain->error before
 * assuming that the data is good.
 *
 * To handle insertion races pass the INSERT_RACE flag along with the
 * generation number of the core.  NULL will be returned if the generation
 * number changes before we have a chance to insert the chain.  Insert
 * races can occur because the parent might be held shared.
 *
 * Caller must hold the parent locked shared or exclusive since we may
 * need the parent's bref array to find our block.
 *
 * WARNING! chain->pmp is always set to NULL for any chain representing
 *	    part of the super-root topology.
 */
hammer2_chain_t *
hammer2_chain_get(hammer2_chain_t *parent, int generation,
		  hammer2_blockref_t *bref, int how)
{
	hammer2_dev_t *hmp = parent->hmp;
	hammer2_chain_t *chain;
	int error;

	/*
	 * Allocate a chain structure representing the existing media
	 * entry.  Resulting chain has one ref and is not locked.
	 */
	if (bref->flags & HAMMER2_BREF_FLAG_PFSROOT)
		chain = hammer2_chain_alloc(hmp, NULL, bref);
	else
		chain = hammer2_chain_alloc(hmp, parent->pmp, bref);
	/* ref'd chain returned */

	/*
	 * Flag that the chain is in the parent's blockmap so delete/flush
	 * knows what to do with it.
	 */
	atomic_set_int(&chain->flags, HAMMER2_CHAIN_BMAPPED);

	/*
	 * chain must be locked to avoid unexpected ripouts
	 */
	hammer2_chain_lock(chain, how);

	/*
	 * Link the chain into its parent.  A spinlock is required to safely
	 * access the RBTREE, and it is possible to collide with another
	 * hammer2_chain_get() operation because the caller might only hold
	 * a shared lock on the parent.
	 *
	 * NOTE: Get races can occur quite often when we distribute
	 *	 asynchronous read-aheads across multiple threads.
	 */
	KKASSERT(parent->refs > 0);
	error = hammer2_chain_insert(parent, chain,
				     HAMMER2_CHAIN_INSERT_SPIN |
				     HAMMER2_CHAIN_INSERT_RACE,
				     generation);
	if (error) {
		KKASSERT((chain->flags & HAMMER2_CHAIN_ONRBTREE) == 0);
		/*kprintf("chain %p get race\n", chain);*/
		hammer2_chain_unlock(chain);
		hammer2_chain_drop(chain);
		chain = NULL;
	} else {
		KKASSERT(chain->flags & HAMMER2_CHAIN_ONRBTREE);
	}

	/*
	 * Return our new chain referenced but not locked, or NULL if
	 * a race occurred.
	 */
	return (chain);
}

/*
 * Lookup initialization/completion API
 */
hammer2_chain_t *
hammer2_chain_lookup_init(hammer2_chain_t *parent, int flags)
{
	hammer2_chain_ref(parent);
	if (flags & HAMMER2_LOOKUP_SHARED) {
		hammer2_chain_lock(parent, HAMMER2_RESOLVE_ALWAYS |
					   HAMMER2_RESOLVE_SHARED);
	} else {
		hammer2_chain_lock(parent, HAMMER2_RESOLVE_ALWAYS);
	}
	return (parent);
}

void
hammer2_chain_lookup_done(hammer2_chain_t *parent)
{
	if (parent) {
		hammer2_chain_unlock(parent);
		hammer2_chain_drop(parent);
	}
}

/*
 * Take the locked chain and return a locked parent.  The chain remains
 * locked on return, but may have to be temporarily unlocked to acquire
 * the parent.  Because of this, (chain) must be stable and cannot be
 * deleted while it was temporarily unlocked (typically means that (chain)
 * is an inode).
 *
 * Pass HAMMER2_RESOLVE_* flags in flags.
 *
 * This will work even if the chain is errored, and the caller can check
 * parent->error on return if desired since the parent will be locked.
 *
 * This function handles the lock order reversal.
 */
hammer2_chain_t *
hammer2_chain_getparent(hammer2_chain_t *chain, int flags)
{
	hammer2_chain_t *parent;

	/*
	 * Be careful of order, chain must be unlocked before parent
	 * is locked below to avoid a deadlock.  Try it trivially first.
	 */
	parent = chain->parent;
	if (parent == NULL)
		panic("hammer2_chain_getparent: no parent");
	hammer2_chain_ref(parent);
	if (hammer2_chain_lock(parent, flags|HAMMER2_RESOLVE_NONBLOCK) == 0)
		return parent;

	for (;;) {
		hammer2_chain_unlock(chain);
		hammer2_chain_lock(parent, flags);
		hammer2_chain_lock(chain, flags);

		/*
		 * Parent relinking races are quite common.  We have to get
		 * it right or we will blow up the block table.
		 */
		if (chain->parent == parent)
			break;
		hammer2_chain_unlock(parent);
		hammer2_chain_drop(parent);
		cpu_ccfence();
		parent = chain->parent;
		if (parent == NULL)
			panic("hammer2_chain_getparent: no parent");
		hammer2_chain_ref(parent);
	}
	return parent;
}

/*
 * Take the locked chain and return a locked parent.  The chain is unlocked
 * and dropped.  *chainp is set to the returned parent as a convenience.
 * Pass HAMMER2_RESOLVE_* flags in flags.
 *
 * This will work even if the chain is errored, and the caller can check
 * parent->error on return if desired since the parent will be locked.
 *
 * The chain does NOT need to be stable.  We use a tracking structure
 * to track the expected parent if the chain is deleted out from under us.
 *
 * This function handles the lock order reversal.
 */
hammer2_chain_t *
hammer2_chain_repparent(hammer2_chain_t **chainp, int flags)
{
	hammer2_chain_t *chain;
	hammer2_chain_t *parent;
	struct hammer2_reptrack reptrack;
	struct hammer2_reptrack **repp;

	/*
	 * Be careful of order, chain must be unlocked before parent
	 * is locked below to avoid a deadlock.  Try it trivially first.
	 */
	chain = *chainp;
	parent = chain->parent;
	if (parent == NULL) {
		hammer2_spin_unex(&chain->core.spin);
		panic("hammer2_chain_repparent: no parent");
	}
	hammer2_chain_ref(parent);
	if (hammer2_chain_lock(parent, flags|HAMMER2_RESOLVE_NONBLOCK) == 0) {
		hammer2_chain_unlock(chain);
		hammer2_chain_drop(chain);
		*chainp = parent;

		return parent;
	}

	/*
	 * Ok, now it gets a bit nasty.  There are multiple situations where
	 * the parent might be in the middle of a deletion, or where the child
	 * (chain) might be deleted the instant we let go of its lock.
	 * We can potentially end up in a no-win situation!
	 *
	 * In particular, the indirect_maintenance() case can cause these
	 * situations.
	 *
	 * To deal with this we install a reptrack structure in the parent
	 * This reptrack structure 'owns' the parent ref and will automatically
	 * migrate to the parent's parent if the parent is deleted permanently.
	 */
	hammer2_spin_init(&reptrack.spin, "h2reptrk");
	reptrack.chain = parent;
	hammer2_chain_ref(parent);		/* for the reptrack */

	hammer2_spin_ex(&parent->core.spin);
	reptrack.next = parent->core.reptrack;
	parent->core.reptrack = &reptrack;
	hammer2_spin_unex(&parent->core.spin);

	hammer2_chain_unlock(chain);
	hammer2_chain_drop(chain);
	chain = NULL;	/* gone */

	/*
	 * At the top of this loop, chain is gone and parent is refd both
	 * by us explicitly AND via our reptrack.  We are attempting to
	 * lock parent.
	 */
	for (;;) {
		hammer2_chain_lock(parent, flags);

		if (reptrack.chain == parent)
			break;
		hammer2_chain_unlock(parent);
		hammer2_chain_drop(parent);

		kprintf("hammer2: debug REPTRACK %p->%p\n",
			parent, reptrack.chain);
		hammer2_spin_ex(&reptrack.spin);
		parent = reptrack.chain;
		hammer2_chain_ref(parent);
		hammer2_spin_unex(&reptrack.spin);
	}

	/*
	 * Once parent is locked and matches our reptrack, our reptrack
	 * will be stable and we have our parent.  We can unlink our
	 * reptrack.
	 *
	 * WARNING!  Remember that the chain lock might be shared.  Chains
	 *	     locked shared have stable parent linkages.
	 */
	hammer2_spin_ex(&parent->core.spin);
	repp = &parent->core.reptrack;
	while (*repp != &reptrack)
		repp = &(*repp)->next;
	*repp = reptrack.next;
	hammer2_spin_unex(&parent->core.spin);

	hammer2_chain_drop(parent);	/* reptrack ref */
	*chainp = parent;		/* return parent lock+ref */

	return parent;
}

/*
 * Dispose of any linked reptrack structures in (chain) by shifting them to
 * (parent).  Both (chain) and (parent) must be exclusively locked.
 *
 * This is interlocked against any children of (chain) on the other side.
 * No children so remain as-of when this is called so we can test
 * core.reptrack without holding the spin-lock.
 *
 * Used whenever the caller intends to permanently delete chains related
 * to topological recursions (BREF_TYPE_INDIRECT, BREF_TYPE_FREEMAP_NODE),
 * where the chains underneath the node being deleted are given a new parent
 * above the node being deleted.
 */
static
void
hammer2_chain_repchange(hammer2_chain_t *parent, hammer2_chain_t *chain)
{
	struct hammer2_reptrack *reptrack;

	KKASSERT(chain->core.live_count == 0 && RB_EMPTY(&chain->core.rbtree));
	while (chain->core.reptrack) {
		hammer2_spin_ex(&parent->core.spin);
		hammer2_spin_ex(&chain->core.spin);
		reptrack = chain->core.reptrack;
		if (reptrack == NULL) {
			hammer2_spin_unex(&chain->core.spin);
			hammer2_spin_unex(&parent->core.spin);
			break;
		}
		hammer2_spin_ex(&reptrack->spin);
		chain->core.reptrack = reptrack->next;
		reptrack->chain = parent;
		reptrack->next = parent->core.reptrack;
		parent->core.reptrack = reptrack;
		hammer2_chain_ref(parent);		/* reptrack */

		hammer2_spin_unex(&chain->core.spin);
		hammer2_spin_unex(&parent->core.spin);
		kprintf("hammer2: debug repchange %p %p->%p\n",
			reptrack, chain, parent);
		hammer2_chain_drop(chain);		/* reptrack */
	}
}

/*
 * Locate the first chain whos key range overlaps (key_beg, key_end) inclusive.
 * (*parentp) typically points to an inode but can also point to a related
 * indirect block and this function will recurse upwards and find the inode
 * or the nearest undeleted indirect block covering the key range.
 *
 * This function unconditionally sets *errorp, replacing any previous value.
 *
 * (*parentp) must be exclusive or shared locked (depending on flags) and
 * referenced and can be an inode or an existing indirect block within the
 * inode.
 *
 * If (*parent) is errored out, this function will not attempt to recurse
 * the radix tree and will return NULL along with an appropriate *errorp.
 * If NULL is returned and *errorp is 0, the requested lookup could not be
 * located.
 *
 * On return (*parentp) will be modified to point at the deepest parent chain
 * element encountered during the search, as a helper for an insertion or
 * deletion.
 *
 * The new (*parentp) will be locked shared or exclusive (depending on flags),
 * and referenced, and the old will be unlocked and dereferenced (no change
 * if they are both the same).  This is particularly important if the caller
 * wishes to insert a new chain, (*parentp) will be set properly even if NULL
 * is returned, as long as no error occurred.
 *
 * The matching chain will be returned locked according to flags.
 *
 * --
 *
 * NULL is returned if no match was found, but (*parentp) will still
 * potentially be adjusted.
 *
 * On return (*key_nextp) will point to an iterative value for key_beg.
 * (If NULL is returned (*key_nextp) is set to (key_end + 1)).
 *
 * This function will also recurse up the chain if the key is not within the
 * current parent's range.  (*parentp) can never be set to NULL.  An iteration
 * can simply allow (*parentp) to float inside the loop.
 *
 * NOTE!  chain->data is not always resolved.  By default it will not be
 *	  resolved for BREF_TYPE_DATA, FREEMAP_NODE, or FREEMAP_LEAF.  Use
 *	  HAMMER2_LOOKUP_ALWAYS to force resolution (but be careful w/
 *	  BREF_TYPE_DATA as the device buffer can alias the logical file
 *	  buffer).
 */

hammer2_chain_t *
hammer2_chain_lookup(hammer2_chain_t **parentp, hammer2_key_t *key_nextp,
		     hammer2_key_t key_beg, hammer2_key_t key_end,
		     int *errorp, int flags)
{
	hammer2_dev_t *hmp;
	hammer2_chain_t *parent;
	hammer2_chain_t *chain;
	hammer2_blockref_t *base;
	hammer2_blockref_t *bref;
	hammer2_blockref_t bcopy;
	hammer2_key_t scan_beg;
	hammer2_key_t scan_end;
	int count = 0;
	int how_always = HAMMER2_RESOLVE_ALWAYS;
	int how_maybe = HAMMER2_RESOLVE_MAYBE;
	int how;
	int generation;
	int maxloops = 300000;
	volatile hammer2_mtx_t save_mtx;

	if (flags & HAMMER2_LOOKUP_ALWAYS) {
		how_maybe = how_always;
		how = HAMMER2_RESOLVE_ALWAYS;
	} else if (flags & HAMMER2_LOOKUP_NODATA) {
		how = HAMMER2_RESOLVE_NEVER;
	} else {
		how = HAMMER2_RESOLVE_MAYBE;
	}
	if (flags & HAMMER2_LOOKUP_SHARED) {
		how_maybe |= HAMMER2_RESOLVE_SHARED;
		how_always |= HAMMER2_RESOLVE_SHARED;
		how |= HAMMER2_RESOLVE_SHARED;
	}

	/*
	 * Recurse (*parentp) upward if necessary until the parent completely
	 * encloses the key range or we hit the inode.
	 *
	 * Handle races against the flusher deleting indirect nodes on its
	 * way back up by continuing to recurse upward past the deletion.
	 */
	parent = *parentp;
	hmp = parent->hmp;
	*errorp = 0;

	while (parent->bref.type == HAMMER2_BREF_TYPE_INDIRECT ||
	       parent->bref.type == HAMMER2_BREF_TYPE_FREEMAP_NODE) {
		scan_beg = parent->bref.key;
		scan_end = scan_beg +
			   ((hammer2_key_t)1 << parent->bref.keybits) - 1;
		if ((parent->flags & HAMMER2_CHAIN_DELETED) == 0) {
			if (key_beg >= scan_beg && key_end <= scan_end)
				break;
		}
		parent = hammer2_chain_repparent(parentp, how_maybe);
	}
again:
	if (--maxloops == 0)
		panic("hammer2_chain_lookup: maxloops");
	/*
	 * Locate the blockref array.  Currently we do a fully associative
	 * search through the array.
	 */
	switch(parent->bref.type) {
	case HAMMER2_BREF_TYPE_INODE:
		/*
		 * Special shortcut for embedded data returns the inode
		 * itself.  Callers must detect this condition and access
		 * the embedded data (the strategy code does this for us).
		 *
		 * This is only applicable to regular files and softlinks.
		 *
		 * We need a second lock on parent.  Since we already have
		 * a lock we must pass LOCKAGAIN to prevent unexpected
		 * blocking (we don't want to block on a second shared
		 * ref if an exclusive lock is pending)
		 */
		if (parent->data->ipdata.meta.op_flags &
		    HAMMER2_OPFLAG_DIRECTDATA) {
			if (flags & HAMMER2_LOOKUP_NODIRECT) {
				chain = NULL;
				*key_nextp = key_end + 1;
				goto done;
			}
			hammer2_chain_ref(parent);
			hammer2_chain_lock(parent, how_always |
						   HAMMER2_RESOLVE_LOCKAGAIN);
			*key_nextp = key_end + 1;
			return (parent);
		}
		base = &parent->data->ipdata.u.blockset.blockref[0];
		count = HAMMER2_SET_COUNT;
		break;
	case HAMMER2_BREF_TYPE_FREEMAP_NODE:
	case HAMMER2_BREF_TYPE_INDIRECT:
		/*
		 * Handle MATCHIND on the parent
		 */
		if (flags & HAMMER2_LOOKUP_MATCHIND) {
			scan_beg = parent->bref.key;
			scan_end = scan_beg +
			       ((hammer2_key_t)1 << parent->bref.keybits) - 1;
			if (key_beg == scan_beg && key_end == scan_end) {
				chain = parent;
				hammer2_chain_ref(chain);
				hammer2_chain_lock(chain, how_maybe);
				*key_nextp = scan_end + 1;
				goto done;
			}
		}

		/*
		 * Optimize indirect blocks in the INITIAL state to avoid
		 * I/O.
		 */
		if (parent->flags & HAMMER2_CHAIN_INITIAL) {
			base = NULL;
		} else {
			if (parent->data == NULL) {
				kprintf("parent->data is NULL %p\n", parent);
				while (1)
					tsleep(parent, 0, "xxx", 0);
			}
			base = &parent->data->npdata[0];
		}
		count = parent->bytes / sizeof(hammer2_blockref_t);
		break;
	case HAMMER2_BREF_TYPE_VOLUME:
		base = &parent->data->voldata.sroot_blockset.blockref[0];
		count = HAMMER2_SET_COUNT;
		break;
	case HAMMER2_BREF_TYPE_FREEMAP:
		base = &parent->data->blkset.blockref[0];
		count = HAMMER2_SET_COUNT;
		break;
	default:
		kprintf("hammer2_chain_lookup: unrecognized "
			"blockref(B) type: %d",
			parent->bref.type);
		while (1)
			tsleep(&base, 0, "dead", 0);
		panic("hammer2_chain_lookup: unrecognized "
		      "blockref(B) type: %d",
		      parent->bref.type);
		base = NULL;	/* safety */
		count = 0;	/* safety */
	}

	/*
	 * No lookup is possible if the parent is errored.  We delayed
	 * this check as long as we could to ensure that the parent backup,
	 * embedded data, and MATCHIND code could still execute.
	 */
	if (parent->error) {
		*errorp = parent->error;
		return NULL;
	}

	/*
	 * Merged scan to find next candidate.
	 *
	 * hammer2_base_*() functions require the parent->core.live_* fields
	 * to be synchronized.
	 *
	 * We need to hold the spinlock to access the block array and RB tree
	 * and to interlock chain creation.
	 */
	if ((parent->flags & HAMMER2_CHAIN_COUNTEDBREFS) == 0)
		hammer2_chain_countbrefs(parent, base, count);

	/*
	 * Combined search
	 */
	hammer2_spin_ex(&parent->core.spin);
	chain = hammer2_combined_find(parent, base, count,
				      key_nextp,
				      key_beg, key_end,
				      &bref);
	generation = parent->core.generation;

	/*
	 * Exhausted parent chain, iterate.
	 */
	if (bref == NULL) {
		KKASSERT(chain == NULL);
		hammer2_spin_unex(&parent->core.spin);
		if (key_beg == key_end)	/* short cut single-key case */
			return (NULL);

		/*
		 * Stop if we reached the end of the iteration.
		 */
		if (parent->bref.type != HAMMER2_BREF_TYPE_INDIRECT &&
		    parent->bref.type != HAMMER2_BREF_TYPE_FREEMAP_NODE) {
			return (NULL);
		}

		/*
		 * Calculate next key, stop if we reached the end of the
		 * iteration, otherwise go up one level and loop.
		 */
		key_beg = parent->bref.key +
			  ((hammer2_key_t)1 << parent->bref.keybits);
		if (key_beg == 0 || key_beg > key_end)
			return (NULL);
		parent = hammer2_chain_repparent(parentp, how_maybe);
		goto again;
	}

	/*
	 * Selected from blockref or in-memory chain.
	 */
	bcopy = *bref;
	if (chain == NULL) {
		hammer2_spin_unex(&parent->core.spin);
		if (bcopy.type == HAMMER2_BREF_TYPE_INDIRECT ||
		    bcopy.type == HAMMER2_BREF_TYPE_FREEMAP_NODE) {
			chain = hammer2_chain_get(parent, generation,
						  &bcopy, how_maybe);
		} else {
			chain = hammer2_chain_get(parent, generation,
						  &bcopy, how);
		}
		if (chain == NULL)
			goto again;
	} else {
		hammer2_chain_ref(chain);
		hammer2_spin_unex(&parent->core.spin);

		/*
		 * chain is referenced but not locked.  We must lock the
		 * chain to obtain definitive state.
		 */
		if (bcopy.type == HAMMER2_BREF_TYPE_INDIRECT ||
		    bcopy.type == HAMMER2_BREF_TYPE_FREEMAP_NODE) {
			hammer2_chain_lock(chain, how_maybe);
		} else {
			hammer2_chain_lock(chain, how);
		}
		KKASSERT(chain->parent == parent);
	}
	if (bcmp(&bcopy, &chain->bref, sizeof(bcopy)) ||
	    chain->parent != parent) {
		hammer2_chain_unlock(chain);
		hammer2_chain_drop(chain);
		chain = NULL;	/* SAFETY */
		goto again;
	}


	/*
	 * Skip deleted chains (XXX cache 'i' end-of-block-array? XXX)
	 *
	 * NOTE: Chain's key range is not relevant as there might be
	 *	 one-offs within the range that are not deleted.
	 *
	 * NOTE: Lookups can race delete-duplicate because
	 *	 delete-duplicate does not lock the parent's core
	 *	 (they just use the spinlock on the core).
	 */
	if (chain->flags & HAMMER2_CHAIN_DELETED) {
		kprintf("skip deleted chain %016jx.%02x key=%016jx\n",
			chain->bref.data_off, chain->bref.type,
			chain->bref.key);
		hammer2_chain_unlock(chain);
		hammer2_chain_drop(chain);
		chain = NULL;	/* SAFETY */
		key_beg = *key_nextp;
		if (key_beg == 0 || key_beg > key_end)
			return(NULL);
		goto again;
	}

	/*
	 * If the chain element is an indirect block it becomes the new
	 * parent and we loop on it.  We must maintain our top-down locks
	 * to prevent the flusher from interfering (i.e. doing a
	 * delete-duplicate and leaving us recursing down a deleted chain).
	 *
	 * The parent always has to be locked with at least RESOLVE_MAYBE
	 * so we can access its data.  It might need a fixup if the caller
	 * passed incompatible flags.  Be careful not to cause a deadlock
	 * as a data-load requires an exclusive lock.
	 *
	 * If HAMMER2_LOOKUP_MATCHIND is set and the indirect block's key
	 * range is within the requested key range we return the indirect
	 * block and do NOT loop.  This is usually only used to acquire
	 * freemap nodes.
	 */
	if (chain->bref.type == HAMMER2_BREF_TYPE_INDIRECT ||
	    chain->bref.type == HAMMER2_BREF_TYPE_FREEMAP_NODE) {
		save_mtx = parent->lock;
		hammer2_chain_unlock(parent);
		hammer2_chain_drop(parent);
		*parentp = parent = chain;
		chain = NULL;	/* SAFETY */
		goto again;
	}
done:
	/*
	 * All done, return the locked chain.
	 *
	 * If the caller does not want a locked chain, replace the lock with
	 * a ref.  Perhaps this can eventually be optimized to not obtain the
	 * lock in the first place for situations where the data does not
	 * need to be resolved.
	 *
	 * NOTE! A chain->error must be tested by the caller upon return.
	 *	 *errorp is only set based on issues which occur while
	 *	 trying to reach the chain.
	 */
	return (chain);
}

/*
 * After having issued a lookup we can iterate all matching keys.
 *
 * If chain is non-NULL we continue the iteration from just after it's index.
 *
 * If chain is NULL we assume the parent was exhausted and continue the
 * iteration at the next parent.
 *
 * If a fatal error occurs (typically an I/O error), a dummy chain is
 * returned with chain->error and error-identifying information set.  This
 * chain will assert if you try to do anything fancy with it.
 *
 * XXX Depending on where the error occurs we should allow continued iteration.
 *
 * parent must be locked on entry and remains locked throughout.  chain's
 * lock status must match flags.  Chain is always at least referenced.
 *
 * WARNING!  The MATCHIND flag does not apply to this function.
 */
hammer2_chain_t *
hammer2_chain_next(hammer2_chain_t **parentp, hammer2_chain_t *chain,
		   hammer2_key_t *key_nextp,
		   hammer2_key_t key_beg, hammer2_key_t key_end,
		   int *errorp, int flags)
{
	hammer2_chain_t *parent;
	int how_maybe;

	/*
	 * Calculate locking flags for upward recursion.
	 */
	how_maybe = HAMMER2_RESOLVE_MAYBE;
	if (flags & HAMMER2_LOOKUP_SHARED)
		how_maybe |= HAMMER2_RESOLVE_SHARED;

	parent = *parentp;
	*errorp = 0;

	/*
	 * Calculate the next index and recalculate the parent if necessary.
	 */
	if (chain) {
		key_beg = chain->bref.key +
			  ((hammer2_key_t)1 << chain->bref.keybits);
		hammer2_chain_unlock(chain);
		hammer2_chain_drop(chain);

		/*
		 * chain invalid past this point, but we can still do a
		 * pointer comparison w/parent.
		 *
		 * Any scan where the lookup returned degenerate data embedded
		 * in the inode has an invalid index and must terminate.
		 */
		if (chain == parent)
			return(NULL);
		if (key_beg == 0 || key_beg > key_end)
			return(NULL);
		chain = NULL;
	} else if (parent->bref.type != HAMMER2_BREF_TYPE_INDIRECT &&
		   parent->bref.type != HAMMER2_BREF_TYPE_FREEMAP_NODE) {
		/*
		 * We reached the end of the iteration.
		 */
		return (NULL);
	} else {
		/*
		 * Continue iteration with next parent unless the current
		 * parent covers the range.
		 *
		 * (This also handles the case of a deleted, empty indirect
		 * node).
		 */
		key_beg = parent->bref.key +
			  ((hammer2_key_t)1 << parent->bref.keybits);
		if (key_beg == 0 || key_beg > key_end)
			return (NULL);
		parent = hammer2_chain_repparent(parentp, how_maybe);
	}

	/*
	 * And execute
	 */
	return (hammer2_chain_lookup(parentp, key_nextp,
				     key_beg, key_end,
				     errorp, flags));
}

/*
 * Caller wishes to iterate chains under parent, loading new chains into
 * chainp.  Caller must initialize *chainp to NULL and *firstp to 1, and
 * then call hammer2_chain_scan() repeatedly until a non-zero return.
 * During the scan, *firstp will be set to 0 and (*chainp) will be replaced
 * with the returned chain for the scan.  The returned *chainp will be
 * locked and referenced.  Any prior contents will be unlocked and dropped.
 *
 * Caller should check the return value.  A normal scan EOF will return
 * exactly HAMMER2_ERROR_EOF.  Any other non-zero value indicates an
 * error trying to access parent data.  Any error in the returned chain
 * must be tested separately by the caller.
 *
 * (*chainp) is dropped on each scan, but will only be set if the returned
 * element itself can recurse.  Leaf elements are NOT resolved, loaded, or
 * returned via *chainp.  The caller will get their bref only.
 *
 * The raw scan function is similar to lookup/next but does not seek to a key.
 * Blockrefs are iterated via first_bref = (parent, NULL) and
 * next_chain = (parent, bref).
 *
 * The passed-in parent must be locked and its data resolved.  The function
 * nominally returns a locked and referenced *chainp != NULL for chains
 * the caller might need to recurse on (and will dipose of any *chainp passed
 * in).  The caller must check the chain->bref.type either way.
 */
int
hammer2_chain_scan(hammer2_chain_t *parent, hammer2_chain_t **chainp,
		   hammer2_blockref_t *bref, int *firstp,
		   int flags)
{
	hammer2_dev_t *hmp;
	hammer2_blockref_t *base;
	hammer2_blockref_t *bref_ptr;
	hammer2_key_t key;
	hammer2_key_t next_key;
	hammer2_chain_t *chain = NULL;
	int count = 0;
	int how_always = HAMMER2_RESOLVE_ALWAYS;
	int how_maybe = HAMMER2_RESOLVE_MAYBE;
	int how;
	int generation;
	int maxloops = 300000;
	int error;

	hmp = parent->hmp;
	error = 0;

	/*
	 * Scan flags borrowed from lookup.
	 */
	if (flags & HAMMER2_LOOKUP_ALWAYS) {
		how_maybe = how_always;
		how = HAMMER2_RESOLVE_ALWAYS;
	} else if (flags & HAMMER2_LOOKUP_NODATA) {
		how = HAMMER2_RESOLVE_NEVER;
	} else {
		how = HAMMER2_RESOLVE_MAYBE;
	}
	if (flags & HAMMER2_LOOKUP_SHARED) {
		how_maybe |= HAMMER2_RESOLVE_SHARED;
		how_always |= HAMMER2_RESOLVE_SHARED;
		how |= HAMMER2_RESOLVE_SHARED;
	}

	/*
	 * Calculate key to locate first/next element, unlocking the previous
	 * element as we go.  Be careful, the key calculation can overflow.
	 *
	 * (also reset bref to NULL)
	 */
	if (*firstp) {
		key = 0;
		*firstp = 0;
	} else {
		key = bref->key + ((hammer2_key_t)1 << bref->keybits);
		if ((chain = *chainp) != NULL) {
			*chainp = NULL;
			hammer2_chain_unlock(chain);
			hammer2_chain_drop(chain);
			chain = NULL;
		}
		if (key == 0) {
			error |= HAMMER2_ERROR_EOF;
			goto done;
		}
	}

again:
	if (parent->error) {
		error = parent->error;
		goto done;
	}
	if (--maxloops == 0)
		panic("hammer2_chain_scan: maxloops");

	/*
	 * Locate the blockref array.  Currently we do a fully associative
	 * search through the array.
	 */
	switch(parent->bref.type) {
	case HAMMER2_BREF_TYPE_INODE:
		/*
		 * An inode with embedded data has no sub-chains.
		 *
		 * WARNING! Bulk scan code may pass a static chain marked
		 *	    as BREF_TYPE_INODE with a copy of the volume
		 *	    root blockset to snapshot the volume.
		 */
		if (parent->data->ipdata.meta.op_flags &
		    HAMMER2_OPFLAG_DIRECTDATA) {
			error |= HAMMER2_ERROR_EOF;
			goto done;
		}
		base = &parent->data->ipdata.u.blockset.blockref[0];
		count = HAMMER2_SET_COUNT;
		break;
	case HAMMER2_BREF_TYPE_FREEMAP_NODE:
	case HAMMER2_BREF_TYPE_INDIRECT:
		/*
		 * Optimize indirect blocks in the INITIAL state to avoid
		 * I/O.
		 */
		if (parent->flags & HAMMER2_CHAIN_INITIAL) {
			base = NULL;
		} else {
			if (parent->data == NULL)
				panic("parent->data is NULL");
			base = &parent->data->npdata[0];
		}
		count = parent->bytes / sizeof(hammer2_blockref_t);
		break;
	case HAMMER2_BREF_TYPE_VOLUME:
		base = &parent->data->voldata.sroot_blockset.blockref[0];
		count = HAMMER2_SET_COUNT;
		break;
	case HAMMER2_BREF_TYPE_FREEMAP:
		base = &parent->data->blkset.blockref[0];
		count = HAMMER2_SET_COUNT;
		break;
	default:
		panic("hammer2_chain_lookup: unrecognized blockref type: %d",
		      parent->bref.type);
		base = NULL;	/* safety */
		count = 0;	/* safety */
	}

	/*
	 * Merged scan to find next candidate.
	 *
	 * hammer2_base_*() functions require the parent->core.live_* fields
	 * to be synchronized.
	 *
	 * We need to hold the spinlock to access the block array and RB tree
	 * and to interlock chain creation.
	 */
	if ((parent->flags & HAMMER2_CHAIN_COUNTEDBREFS) == 0)
		hammer2_chain_countbrefs(parent, base, count);

	next_key = 0;
	bref_ptr = NULL;
	hammer2_spin_ex(&parent->core.spin);
	chain = hammer2_combined_find(parent, base, count,
				      &next_key,
				      key, HAMMER2_KEY_MAX,
				      &bref_ptr);
	generation = parent->core.generation;

	/*
	 * Exhausted parent chain, we're done.
	 */
	if (bref_ptr == NULL) {
		hammer2_spin_unex(&parent->core.spin);
		KKASSERT(chain == NULL);
		error |= HAMMER2_ERROR_EOF;
		goto done;
	}

	/*
	 * Copy into the supplied stack-based blockref.
	 */
	*bref = *bref_ptr;

	/*
	 * Selected from blockref or in-memory chain.
	 */
	if (chain == NULL) {
		switch(bref->type) {
		case HAMMER2_BREF_TYPE_INODE:
		case HAMMER2_BREF_TYPE_FREEMAP_NODE:
		case HAMMER2_BREF_TYPE_INDIRECT:
		case HAMMER2_BREF_TYPE_VOLUME:
		case HAMMER2_BREF_TYPE_FREEMAP:
			/*
			 * Recursion, always get the chain
			 */
			hammer2_spin_unex(&parent->core.spin);
			chain = hammer2_chain_get(parent, generation,
						  bref, how);
			if (chain == NULL)
				goto again;
			break;
		default:
			/*
			 * No recursion, do not waste time instantiating
			 * a chain, just iterate using the bref.
			 */
			hammer2_spin_unex(&parent->core.spin);
			break;
		}
	} else {
		/*
		 * Recursion or not we need the chain in order to supply
		 * the bref.
		 */
		hammer2_chain_ref(chain);
		hammer2_spin_unex(&parent->core.spin);
		hammer2_chain_lock(chain, how);
	}
	if (chain &&
	    (bcmp(bref, &chain->bref, sizeof(*bref)) ||
	     chain->parent != parent)) {
		hammer2_chain_unlock(chain);
		hammer2_chain_drop(chain);
		chain = NULL;
		goto again;
	}

	/*
	 * Skip deleted chains (XXX cache 'i' end-of-block-array? XXX)
	 *
	 * NOTE: chain's key range is not relevant as there might be
	 *	 one-offs within the range that are not deleted.
	 *
	 * NOTE: XXX this could create problems with scans used in
	 *	 situations other than mount-time recovery.
	 *
	 * NOTE: Lookups can race delete-duplicate because
	 *	 delete-duplicate does not lock the parent's core
	 *	 (they just use the spinlock on the core).
	 */
	if (chain && (chain->flags & HAMMER2_CHAIN_DELETED)) {
		hammer2_chain_unlock(chain);
		hammer2_chain_drop(chain);
		chain = NULL;

		key = next_key;
		if (key == 0) {
			error |= HAMMER2_ERROR_EOF;
			goto done;
		}
		goto again;
	}

done:
	/*
	 * All done, return the bref or NULL, supply chain if necessary.
	 */
	if (chain)
		*chainp = chain;
	return (error);
}

/*
 * Create and return a new hammer2 system memory structure of the specified
 * key, type and size and insert it under (*parentp).  This is a full
 * insertion, based on the supplied key/keybits, and may involve creating
 * indirect blocks and moving other chains around via delete/duplicate.
 *
 * THE CALLER MUST HAVE ALREADY PROPERLY SEEKED (*parentp) TO THE INSERTION
 * POINT SANS ANY REQUIRED INDIRECT BLOCK CREATIONS DUE TO THE ARRAY BEING
 * FULL.  This typically means that the caller is creating the chain after
 * doing a hammer2_chain_lookup().
 *
 * (*parentp) must be exclusive locked and may be replaced on return
 * depending on how much work the function had to do.
 *
 * (*parentp) must not be errored or this function will assert.
 *
 * (*chainp) usually starts out NULL and returns the newly created chain,
 * but if the caller desires the caller may allocate a disconnected chain
 * and pass it in instead.
 *
 * This function should NOT be used to insert INDIRECT blocks.  It is
 * typically used to create/insert inodes and data blocks.
 *
 * Caller must pass-in an exclusively locked parent the new chain is to
 * be inserted under, and optionally pass-in a disconnected, exclusively
 * locked chain to insert (else we create a new chain).  The function will
 * adjust (*parentp) as necessary, create or connect the chain, and
 * return an exclusively locked chain in *chainp.
 *
 * When creating a PFSROOT inode under the super-root, pmp is typically NULL
 * and will be reassigned.
 *
 * NOTE: returns HAMMER_ERROR_* flags
 */
int
hammer2_chain_create(hammer2_chain_t **parentp, hammer2_chain_t **chainp,
		     hammer2_pfs_t *pmp, int methods,
		     hammer2_key_t key, int keybits, int type, size_t bytes,
		     hammer2_tid_t mtid, hammer2_off_t dedup_off, int flags)
{
	hammer2_dev_t *hmp;
	hammer2_chain_t *chain;
	hammer2_chain_t *parent;
	hammer2_blockref_t *base;
	hammer2_blockref_t dummy;
	int allocated = 0;
	int error = 0;
	int count;
	int maxloops = 300000;

	/*
	 * Topology may be crossing a PFS boundary.
	 */
	parent = *parentp;
	KKASSERT(hammer2_mtx_owned(&parent->lock));
	KKASSERT(parent->error == 0);
	hmp = parent->hmp;
	chain = *chainp;

	if (chain == NULL) {
		/*
		 * First allocate media space and construct the dummy bref,
		 * then allocate the in-memory chain structure.  Set the
		 * INITIAL flag for fresh chains which do not have embedded
		 * data.
		 *
		 * XXX for now set the check mode of the child based on
		 *     the parent or, if the parent is an inode, the
		 *     specification in the inode.
		 */
		bzero(&dummy, sizeof(dummy));
		dummy.type = type;
		dummy.key = key;
		dummy.keybits = keybits;
		dummy.data_off = hammer2_getradix(bytes);

		/*
		 * Inherit methods from parent by default.  Primarily used
		 * for BREF_TYPE_DATA.  Non-data types *must* be set to
		 * a non-NONE check algorithm.
		 */
		if (methods == -1)
			dummy.methods = parent->bref.methods;
		else
			dummy.methods = (uint8_t)methods;

		if (type != HAMMER2_BREF_TYPE_DATA &&
		    HAMMER2_DEC_CHECK(dummy.methods) == HAMMER2_CHECK_NONE) {
			dummy.methods |=
				HAMMER2_ENC_CHECK(HAMMER2_CHECK_DEFAULT);
		}

		chain = hammer2_chain_alloc(hmp, pmp, &dummy);

		/*
		 * Lock the chain manually, chain_lock will load the chain
		 * which we do NOT want to do.  (note: chain->refs is set
		 * to 1 by chain_alloc() for us, but lockcnt is not).
		 */
		chain->lockcnt = 1;
		hammer2_mtx_ex(&chain->lock);
		allocated = 1;
		++curthread->td_tracker;

		/*
		 * Set INITIAL to optimize I/O.  The flag will generally be
		 * processed when we call hammer2_chain_modify().
		 *
		 * Recalculate bytes to reflect the actual media block
		 * allocation.  Handle special case radix 0 == 0 bytes.
		 */
		bytes = (size_t)(chain->bref.data_off & HAMMER2_OFF_MASK_RADIX);
		if (bytes)
			bytes = (hammer2_off_t)1 << bytes;
		chain->bytes = bytes;

		switch(type) {
		case HAMMER2_BREF_TYPE_VOLUME:
		case HAMMER2_BREF_TYPE_FREEMAP:
			panic("hammer2_chain_create: called with volume type");
			break;
		case HAMMER2_BREF_TYPE_INDIRECT:
			panic("hammer2_chain_create: cannot be used to"
			      "create indirect block");
			break;
		case HAMMER2_BREF_TYPE_FREEMAP_NODE:
			panic("hammer2_chain_create: cannot be used to"
			      "create freemap root or node");
			break;
		case HAMMER2_BREF_TYPE_FREEMAP_LEAF:
			KKASSERT(bytes == sizeof(chain->data->bmdata));
			/* fall through */
		case HAMMER2_BREF_TYPE_DIRENT:
		case HAMMER2_BREF_TYPE_INODE:
		case HAMMER2_BREF_TYPE_DATA:
		default:
			/*
			 * leave chain->data NULL, set INITIAL
			 */
			KKASSERT(chain->data == NULL);
			atomic_set_int(&chain->flags, HAMMER2_CHAIN_INITIAL);
			break;
		}
	} else {
		/*
		 * We are reattaching a previously deleted chain, possibly
		 * under a new parent and possibly with a new key/keybits.
		 * The chain does not have to be in a modified state.  The
		 * UPDATE flag will be set later on in this routine.
		 *
		 * Do NOT mess with the current state of the INITIAL flag.
		 */
		chain->bref.key = key;
		chain->bref.keybits = keybits;
		if (chain->flags & HAMMER2_CHAIN_DELETED)
			atomic_clear_int(&chain->flags, HAMMER2_CHAIN_DELETED);
		KKASSERT(chain->parent == NULL);
	}

	/*
	 * Set the appropriate bref flag if requested.
	 *
	 * NOTE! Callers can call this function to move chains without
	 *	 knowing about special flags, so don't clear bref flags
	 *	 here!
	 */
	if (flags & HAMMER2_INSERT_PFSROOT)
		chain->bref.flags |= HAMMER2_BREF_FLAG_PFSROOT;

	/*
	 * Calculate how many entries we have in the blockref array and
	 * determine if an indirect block is required.
	 */
again:
	if (--maxloops == 0)
		panic("hammer2_chain_create: maxloops");

	switch(parent->bref.type) {
	case HAMMER2_BREF_TYPE_INODE:
		if ((parent->data->ipdata.meta.op_flags &
		     HAMMER2_OPFLAG_DIRECTDATA) != 0) {
			kprintf("hammer2: parent set for direct-data! "
				"pkey=%016jx ckey=%016jx\n",
				parent->bref.key,
				chain->bref.key);
	        }
		KKASSERT((parent->data->ipdata.meta.op_flags &
			  HAMMER2_OPFLAG_DIRECTDATA) == 0);
		KKASSERT(parent->data != NULL);
		base = &parent->data->ipdata.u.blockset.blockref[0];
		count = HAMMER2_SET_COUNT;
		break;
	case HAMMER2_BREF_TYPE_INDIRECT:
	case HAMMER2_BREF_TYPE_FREEMAP_NODE:
		if (parent->flags & HAMMER2_CHAIN_INITIAL)
			base = NULL;
		else
			base = &parent->data->npdata[0];
		count = parent->bytes / sizeof(hammer2_blockref_t);
		break;
	case HAMMER2_BREF_TYPE_VOLUME:
		KKASSERT(parent->data != NULL);
		base = &parent->data->voldata.sroot_blockset.blockref[0];
		count = HAMMER2_SET_COUNT;
		break;
	case HAMMER2_BREF_TYPE_FREEMAP:
		KKASSERT(parent->data != NULL);
		base = &parent->data->blkset.blockref[0];
		count = HAMMER2_SET_COUNT;
		break;
	default:
		panic("hammer2_chain_create: unrecognized blockref type: %d",
		      parent->bref.type);
		base = NULL;
		count = 0;
		break;
	}

	/*
	 * Make sure we've counted the brefs
	 */
	if ((parent->flags & HAMMER2_CHAIN_COUNTEDBREFS) == 0)
		hammer2_chain_countbrefs(parent, base, count);

	KASSERT(parent->core.live_count >= 0 &&
		parent->core.live_count <= count,
		("bad live_count %d/%d (%02x, %d)",
			parent->core.live_count, count,
			parent->bref.type, parent->bytes));

	/*
	 * If no free blockref could be found we must create an indirect
	 * block and move a number of blockrefs into it.  With the parent
	 * locked we can safely lock each child in order to delete+duplicate
	 * it without causing a deadlock.
	 *
	 * This may return the new indirect block or the old parent depending
	 * on where the key falls.  NULL is returned on error.
	 */
	if (parent->core.live_count == count) {
		hammer2_chain_t *nparent;

		KKASSERT((flags & HAMMER2_INSERT_SAMEPARENT) == 0);

		nparent = hammer2_chain_create_indirect(parent, key, keybits,
							mtid, type, &error);
		if (nparent == NULL) {
			if (allocated)
				hammer2_chain_drop(chain);
			chain = NULL;
			goto done;
		}
		if (parent != nparent) {
			hammer2_chain_unlock(parent);
			hammer2_chain_drop(parent);
			parent = *parentp = nparent;
		}
		goto again;
	}

	if (chain->flags & HAMMER2_CHAIN_DELETED)
		kprintf("Inserting deleted chain @%016jx\n",
			chain->bref.key);

	/*
	 * Link the chain into its parent.
	 */
	if (chain->parent != NULL)
		panic("hammer2: hammer2_chain_create: chain already connected");
	KKASSERT(chain->parent == NULL);
	KKASSERT(parent->core.live_count < count);
	hammer2_chain_insert(parent, chain,
			     HAMMER2_CHAIN_INSERT_SPIN |
			     HAMMER2_CHAIN_INSERT_LIVE,
			     0);

	if (allocated) {
		/*
		 * Mark the newly created chain modified.  This will cause
		 * UPDATE to be set and process the INITIAL flag.
		 *
		 * Device buffers are not instantiated for DATA elements
		 * as these are handled by logical buffers.
		 *
		 * Indirect and freemap node indirect blocks are handled
		 * by hammer2_chain_create_indirect() and not by this
		 * function.
		 *
		 * Data for all other bref types is expected to be
		 * instantiated (INODE, LEAF).
		 */
		switch(chain->bref.type) {
		case HAMMER2_BREF_TYPE_DATA:
		case HAMMER2_BREF_TYPE_FREEMAP_LEAF:
		case HAMMER2_BREF_TYPE_DIRENT:
		case HAMMER2_BREF_TYPE_INODE:
			error = hammer2_chain_modify(chain, mtid, dedup_off,
						     HAMMER2_MODIFY_OPTDATA);
			break;
		default:
			/*
			 * Remaining types are not supported by this function.
			 * In particular, INDIRECT and LEAF_NODE types are
			 * handled by create_indirect().
			 */
			panic("hammer2_chain_create: bad type: %d",
			      chain->bref.type);
			/* NOT REACHED */
			break;
		}
	} else {
		/*
		 * When reconnecting a chain we must set UPDATE and
		 * setflush so the flush recognizes that it must update
		 * the bref in the parent.
		 */
		if ((chain->flags & HAMMER2_CHAIN_UPDATE) == 0)
			atomic_set_int(&chain->flags, HAMMER2_CHAIN_UPDATE);
	}

	/*
	 * We must setflush(parent) to ensure that it recurses through to
	 * chain.  setflush(chain) might not work because ONFLUSH is possibly
	 * already set in the chain (so it won't recurse up to set it in the
	 * parent).
	 */
	hammer2_chain_setflush(parent);

done:
	*chainp = chain;

	return (error);
}

/*
 * Move the chain from its old parent to a new parent.  The chain must have
 * already been deleted or already disconnected (or never associated) with
 * a parent.  The chain is reassociated with the new parent and the deleted
 * flag will be cleared (no longer deleted).  The chain's modification state
 * is not altered.
 *
 * THE CALLER MUST HAVE ALREADY PROPERLY SEEKED (parent) TO THE INSERTION
 * POINT SANS ANY REQUIRED INDIRECT BLOCK CREATIONS DUE TO THE ARRAY BEING
 * FULL.  This typically means that the caller is creating the chain after
 * doing a hammer2_chain_lookup().
 *
 * Neither (parent) or (chain) can be errored.
 *
 * If (parent) is non-NULL then the chain is inserted under the parent.
 *
 * If (parent) is NULL then the newly duplicated chain is not inserted
 * anywhere, similar to if it had just been chain_alloc()'d (suitable for
 * passing into hammer2_chain_create() after this function returns).
 *
 * WARNING! This function calls create which means it can insert indirect
 *	    blocks.  This can cause other unrelated chains in the parent to
 *	    be moved to a newly inserted indirect block in addition to the
 *	    specific chain.
 */
void
hammer2_chain_rename(hammer2_chain_t **parentp, hammer2_chain_t *chain,
		     hammer2_tid_t mtid, int flags)
{
	hammer2_blockref_t *bref;
	hammer2_dev_t *hmp;
	hammer2_chain_t *parent;
	size_t bytes;

	/*
	 * WARNING!  We should never resolve DATA to device buffers
	 *	     (XXX allow it if the caller did?), and since
	 *	     we currently do not have the logical buffer cache
	 *	     buffer in-hand to fix its cached physical offset
	 *	     we also force the modify code to not COW it. XXX
	 */
	hmp = chain->hmp;
	KKASSERT(chain->parent == NULL);
	KKASSERT(chain->error == 0);

	/*
	 * Now create a duplicate of the chain structure, associating
	 * it with the same core, making it the same size, pointing it
	 * to the same bref (the same media block).
	 *
	 * NOTE: Handle special radix == 0 case (means 0 bytes).
	 */
	bref = &chain->bref;
	bytes = (size_t)(bref->data_off & HAMMER2_OFF_MASK_RADIX);
	if (bytes)
		bytes = (hammer2_off_t)1 << bytes;

	/*
	 * If parent is not NULL the duplicated chain will be entered under
	 * the parent and the UPDATE bit set to tell flush to update
	 * the blockref.
	 *
	 * We must setflush(parent) to ensure that it recurses through to
	 * chain.  setflush(chain) might not work because ONFLUSH is possibly
	 * already set in the chain (so it won't recurse up to set it in the
	 * parent).
	 *
	 * Having both chains locked is extremely important for atomicy.
	 */
	if (parentp && (parent = *parentp) != NULL) {
		KKASSERT(hammer2_mtx_owned(&parent->lock));
		KKASSERT(parent->refs > 0);
		KKASSERT(parent->error == 0);

		hammer2_chain_create(parentp, &chain,
				     chain->pmp, HAMMER2_METH_DEFAULT,
				     bref->key, bref->keybits, bref->type,
				     chain->bytes, mtid, 0, flags);
		KKASSERT(chain->flags & HAMMER2_CHAIN_UPDATE);
		hammer2_chain_setflush(*parentp);
	}
}

/*
 * Helper function for deleting chains.
 *
 * The chain is removed from the live view (the RBTREE) as well as the parent's
 * blockmap.  Both chain and its parent must be locked.
 *
 * parent may not be errored.  chain can be errored.
 */
static int
_hammer2_chain_delete_helper(hammer2_chain_t *parent, hammer2_chain_t *chain,
			     hammer2_tid_t mtid, int flags)
{
	hammer2_dev_t *hmp;
	int error = 0;

	KKASSERT((chain->flags & (HAMMER2_CHAIN_DELETED |
				  HAMMER2_CHAIN_FICTITIOUS)) == 0);
	KKASSERT(chain->parent == parent);
	hmp = chain->hmp;

	if (chain->flags & HAMMER2_CHAIN_BMAPPED) {
		/*
		 * Chain is blockmapped, so there must be a parent.
		 * Atomically remove the chain from the parent and remove
		 * the blockmap entry.  The parent must be set modified
		 * to remove the blockmap entry.
		 */
		hammer2_blockref_t *base;
		int count;

		KKASSERT(parent != NULL);
		KKASSERT(parent->error == 0);
		KKASSERT((parent->flags & HAMMER2_CHAIN_INITIAL) == 0);
		error = hammer2_chain_modify(parent, mtid, 0, 0);
		if (error)
			goto done;

		/*
		 * Calculate blockmap pointer
		 */
		KKASSERT(chain->flags & HAMMER2_CHAIN_ONRBTREE);
		hammer2_spin_ex(&chain->core.spin);
		hammer2_spin_ex(&parent->core.spin);

		atomic_set_int(&chain->flags, HAMMER2_CHAIN_DELETED);
		atomic_add_int(&parent->core.live_count, -1);
		++parent->core.generation;
		RB_REMOVE(hammer2_chain_tree, &parent->core.rbtree, chain);
		atomic_clear_int(&chain->flags, HAMMER2_CHAIN_ONRBTREE);
		--parent->core.chain_count;
		chain->parent = NULL;

		switch(parent->bref.type) {
		case HAMMER2_BREF_TYPE_INODE:
			/*
			 * Access the inode's block array.  However, there
			 * is no block array if the inode is flagged
			 * DIRECTDATA.
			 */
			if (parent->data &&
			    (parent->data->ipdata.meta.op_flags &
			     HAMMER2_OPFLAG_DIRECTDATA) == 0) {
				base =
				   &parent->data->ipdata.u.blockset.blockref[0];
			} else {
				base = NULL;
			}
			count = HAMMER2_SET_COUNT;
			break;
		case HAMMER2_BREF_TYPE_INDIRECT:
		case HAMMER2_BREF_TYPE_FREEMAP_NODE:
			if (parent->data)
				base = &parent->data->npdata[0];
			else
				base = NULL;
			count = parent->bytes / sizeof(hammer2_blockref_t);
			break;
		case HAMMER2_BREF_TYPE_VOLUME:
			base = &parent->data->voldata.
					sroot_blockset.blockref[0];
			count = HAMMER2_SET_COUNT;
			break;
		case HAMMER2_BREF_TYPE_FREEMAP:
			base = &parent->data->blkset.blockref[0];
			count = HAMMER2_SET_COUNT;
			break;
		default:
			base = NULL;
			count = 0;
			panic("hammer2_flush_pass2: "
			      "unrecognized blockref type: %d",
			      parent->bref.type);
		}

		/*
		 * delete blockmapped chain from its parent.
		 *
		 * The parent is not affected by any statistics in chain
		 * which are pending synchronization.  That is, there is
		 * nothing to undo in the parent since they have not yet
		 * been incorporated into the parent.
		 *
		 * The parent is affected by statistics stored in inodes.
		 * Those have already been synchronized, so they must be
		 * undone.  XXX split update possible w/delete in middle?
		 */
		if (base) {
			hammer2_base_delete(parent, base, count, chain);
		}
		hammer2_spin_unex(&parent->core.spin);
		hammer2_spin_unex(&chain->core.spin);
	} else if (chain->flags & HAMMER2_CHAIN_ONRBTREE) {
		/*
		 * Chain is not blockmapped but a parent is present.
		 * Atomically remove the chain from the parent.  There is
		 * no blockmap entry to remove.
		 *
		 * Because chain was associated with a parent but not
		 * synchronized, the chain's *_count_up fields contain
		 * inode adjustment statistics which must be undone.
		 */
		hammer2_spin_ex(&chain->core.spin);
		hammer2_spin_ex(&parent->core.spin);
		atomic_set_int(&chain->flags, HAMMER2_CHAIN_DELETED);
		atomic_add_int(&parent->core.live_count, -1);
		++parent->core.generation;
		RB_REMOVE(hammer2_chain_tree, &parent->core.rbtree, chain);
		atomic_clear_int(&chain->flags, HAMMER2_CHAIN_ONRBTREE);
		--parent->core.chain_count;
		chain->parent = NULL;
		hammer2_spin_unex(&parent->core.spin);
		hammer2_spin_unex(&chain->core.spin);
	} else {
		/*
		 * Chain is not blockmapped and has no parent.  This
		 * is a degenerate case.
		 */
		atomic_set_int(&chain->flags, HAMMER2_CHAIN_DELETED);
	}
done:
	return error;
}

/*
 * Create an indirect block that covers one or more of the elements in the
 * current parent.  Either returns the existing parent with no locking or
 * ref changes or returns the new indirect block locked and referenced
 * and leaving the original parent lock/ref intact as well.
 *
 * If an error occurs, NULL is returned and *errorp is set to the H2 error.
 *
 * The returned chain depends on where the specified key falls.
 *
 * The key/keybits for the indirect mode only needs to follow three rules:
 *
 * (1) That all elements underneath it fit within its key space and
 *
 * (2) That all elements outside it are outside its key space.
 *
 * (3) When creating the new indirect block any elements in the current
 *     parent that fit within the new indirect block's keyspace must be
 *     moved into the new indirect block.
 *
 * (4) The keyspace chosen for the inserted indirect block CAN cover a wider
 *     keyspace the the current parent, but lookup/iteration rules will
 *     ensure (and must ensure) that rule (2) for all parents leading up
 *     to the nearest inode or the root volume header is adhered to.  This
 *     is accomplished by always recursing through matching keyspaces in
 *     the hammer2_chain_lookup() and hammer2_chain_next() API.
 *
 * The current implementation calculates the current worst-case keyspace by
 * iterating the current parent and then divides it into two halves, choosing
 * whichever half has the most elements (not necessarily the half containing
 * the requested key).
 *
 * We can also opt to use the half with the least number of elements.  This
 * causes lower-numbered keys (aka logical file offsets) to recurse through
 * fewer indirect blocks and higher-numbered keys to recurse through more.
 * This also has the risk of not moving enough elements to the new indirect
 * block and being forced to create several indirect blocks before the element
 * can be inserted.
 *
 * Must be called with an exclusively locked parent.
 *
 * NOTE: *errorp set to HAMMER_ERROR_* flags
 */
static int hammer2_chain_indkey_freemap(hammer2_chain_t *parent,
				hammer2_key_t *keyp, int keybits,
				hammer2_blockref_t *base, int count);
static int hammer2_chain_indkey_file(hammer2_chain_t *parent,
				hammer2_key_t *keyp, int keybits,
				hammer2_blockref_t *base, int count,
				int ncount);
static int hammer2_chain_indkey_dir(hammer2_chain_t *parent,
				hammer2_key_t *keyp, int keybits,
				hammer2_blockref_t *base, int count,
				int ncount);
static
hammer2_chain_t *
hammer2_chain_create_indirect(hammer2_chain_t *parent,
			      hammer2_key_t create_key, int create_bits,
			      hammer2_tid_t mtid, int for_type, int *errorp)
{
	hammer2_dev_t *hmp;
	hammer2_blockref_t *base;
	hammer2_blockref_t *bref;
	hammer2_blockref_t bcopy;
	hammer2_chain_t *chain;
	hammer2_chain_t *ichain;
	hammer2_chain_t dummy;
	hammer2_key_t key = create_key;
	hammer2_key_t key_beg;
	hammer2_key_t key_end;
	hammer2_key_t key_next;
	int keybits = create_bits;
	int count;
	int ncount;
	int nbytes;
	int loops;
	int error;
	int reason;
	int generation;
	int maxloops = 300000;

	/*
	 * Calculate the base blockref pointer or NULL if the chain
	 * is known to be empty.  We need to calculate the array count
	 * for RB lookups either way.
	 */
	hmp = parent->hmp;
	KKASSERT(hammer2_mtx_owned(&parent->lock));

	/*
	 * Pre-modify the parent now to avoid having to deal with error
	 * processing if we tried to later (in the middle of our loop).
	 */
	*errorp = hammer2_chain_modify(parent, mtid, 0, 0);
	if (*errorp) {
		kprintf("hammer2_create_indirect: error %08x %s\n",
			*errorp, hammer2_error_str(*errorp));
		return NULL;
	}

	/*hammer2_chain_modify(&parent, HAMMER2_MODIFY_OPTDATA);*/
	base = hammer2_chain_base_and_count(parent, &count);

	/*
	 * dummy used in later chain allocation (no longer used for lookups).
	 */
	bzero(&dummy, sizeof(dummy));

	/*
	 * How big should our new indirect block be?  It has to be at least
	 * as large as its parent for splits to work properly.
	 *
	 * The freemap uses a specific indirect block size.  The number of
	 * levels are built dynamically and ultimately depend on the size
	 * volume.  Because freemap blocks are taken from the reserved areas
	 * of the volume our goal is efficiency (fewer levels) and not so
	 * much to save disk space.
	 *
	 * The first indirect block level for a directory usually uses
	 * HAMMER2_IND_BYTES_MIN (4KB = 32 directory entries).  Due to
	 * the hash mechanism, this typically gives us a nominal
	 * 32 * 4 entries with one level of indirection.
	 *
	 * We use HAMMER2_IND_BYTES_NOM (16KB = 128 blockrefs) for FILE
	 * indirect blocks.  The initial 4 entries in the inode gives us
	 * 256KB.  Up to 4 indirect blocks gives us 32MB.  Three levels
	 * of indirection gives us 137GB, and so forth.  H2 can support
	 * huge file sizes but they are not typical, so we try to stick
	 * with compactness and do not use a larger indirect block size.
	 *
	 * We could use 64KB (PBUFSIZE), giving us 512 blockrefs, but
	 * due to the way indirect blocks are created this usually winds
	 * up being extremely inefficient for small files.  Even though
	 * 16KB requires more levels of indirection for very large files,
	 * the 16KB records can be ganged together into 64KB DIOs.
	 */
	if (for_type == HAMMER2_BREF_TYPE_FREEMAP_NODE ||
	    for_type == HAMMER2_BREF_TYPE_FREEMAP_LEAF) {
		nbytes = HAMMER2_FREEMAP_LEVELN_PSIZE;
	} else if (parent->bref.type == HAMMER2_BREF_TYPE_INODE) {
		if (parent->data->ipdata.meta.type ==
		    HAMMER2_OBJTYPE_DIRECTORY)
			nbytes = HAMMER2_IND_BYTES_MIN;	/* 4KB = 32 entries */
		else
			nbytes = HAMMER2_IND_BYTES_NOM;	/* 16KB = ~8MB file */

	} else {
		nbytes = HAMMER2_IND_BYTES_NOM;
	}
	if (nbytes < count * sizeof(hammer2_blockref_t)) {
		KKASSERT(for_type != HAMMER2_BREF_TYPE_FREEMAP_NODE &&
			 for_type != HAMMER2_BREF_TYPE_FREEMAP_LEAF);
		nbytes = count * sizeof(hammer2_blockref_t);
	}
	ncount = nbytes / sizeof(hammer2_blockref_t);

	/*
	 * When creating an indirect block for a freemap node or leaf
	 * the key/keybits must be fitted to static radix levels because
	 * particular radix levels use particular reserved blocks in the
	 * related zone.
	 *
	 * This routine calculates the key/radix of the indirect block
	 * we need to create, and whether it is on the high-side or the
	 * low-side.
	 */
	switch(for_type) {
	case HAMMER2_BREF_TYPE_FREEMAP_NODE:
	case HAMMER2_BREF_TYPE_FREEMAP_LEAF:
		keybits = hammer2_chain_indkey_freemap(parent, &key, keybits,
						       base, count);
		break;
	case HAMMER2_BREF_TYPE_DATA:
		keybits = hammer2_chain_indkey_file(parent, &key, keybits,
						    base, count, ncount);
		break;
	case HAMMER2_BREF_TYPE_DIRENT:
	case HAMMER2_BREF_TYPE_INODE:
		keybits = hammer2_chain_indkey_dir(parent, &key, keybits,
						   base, count, ncount);
		break;
	default:
		panic("illegal indirect block for bref type %d", for_type);
		break;
	}

	/*
	 * Normalize the key for the radix being represented, keeping the
	 * high bits and throwing away the low bits.
	 */
	key &= ~(((hammer2_key_t)1 << keybits) - 1);

	/*
	 * Ok, create our new indirect block
	 */
	if (for_type == HAMMER2_BREF_TYPE_FREEMAP_NODE ||
	    for_type == HAMMER2_BREF_TYPE_FREEMAP_LEAF) {
		dummy.bref.type = HAMMER2_BREF_TYPE_FREEMAP_NODE;
	} else {
		dummy.bref.type = HAMMER2_BREF_TYPE_INDIRECT;
	}
	dummy.bref.key = key;
	dummy.bref.keybits = keybits;
	dummy.bref.data_off = hammer2_getradix(nbytes);
	dummy.bref.methods =
		HAMMER2_ENC_CHECK(HAMMER2_DEC_CHECK(parent->bref.methods)) |
		HAMMER2_ENC_COMP(HAMMER2_COMP_NONE);

	ichain = hammer2_chain_alloc(hmp, parent->pmp, &dummy.bref);
	atomic_set_int(&ichain->flags, HAMMER2_CHAIN_INITIAL);
	hammer2_chain_lock(ichain, HAMMER2_RESOLVE_MAYBE);
	/* ichain has one ref at this point */

	/*
	 * We have to mark it modified to allocate its block, but use
	 * OPTDATA to allow it to remain in the INITIAL state.  Otherwise
	 * it won't be acted upon by the flush code.
	 */
	*errorp = hammer2_chain_modify(ichain, mtid, 0, HAMMER2_MODIFY_OPTDATA);
	if (*errorp) {
		kprintf("hammer2_alloc_indirect: error %08x %s\n",
			*errorp, hammer2_error_str(*errorp));
		hammer2_chain_unlock(ichain);
		hammer2_chain_drop(ichain);
		return NULL;
	}

	/*
	 * Iterate the original parent and move the matching brefs into
	 * the new indirect block.
	 *
	 * XXX handle flushes.
	 */
	key_beg = 0;
	key_end = HAMMER2_KEY_MAX;
	key_next = 0;	/* avoid gcc warnings */
	hammer2_spin_ex(&parent->core.spin);
	loops = 0;
	reason = 0;

	for (;;) {
		/*
		 * Parent may have been modified, relocating its block array.
		 * Reload the base pointer.
		 */
		base = hammer2_chain_base_and_count(parent, &count);

		if (++loops > 100000) {
		    hammer2_spin_unex(&parent->core.spin);
		    panic("excessive loops r=%d p=%p base/count %p:%d %016jx\n",
			  reason, parent, base, count, key_next);
		}

		/*
		 * NOTE: spinlock stays intact, returned chain (if not NULL)
		 *	 is not referenced or locked which means that we
		 *	 cannot safely check its flagged / deletion status
		 *	 until we lock it.
		 */
		chain = hammer2_combined_find(parent, base, count,
					      &key_next,
					      key_beg, key_end,
					      &bref);
		generation = parent->core.generation;
		if (bref == NULL)
			break;
		key_next = bref->key + ((hammer2_key_t)1 << bref->keybits);

		/*
		 * Skip keys that are not within the key/radix of the new
		 * indirect block.  They stay in the parent.
		 */
		if ((~(((hammer2_key_t)1 << keybits) - 1) &
		    (key ^ bref->key)) != 0) {
			goto next_key_spinlocked;
		}

		/*
		 * Load the new indirect block by acquiring the related
		 * chains (potentially from media as it might not be
		 * in-memory).  Then move it to the new parent (ichain).
		 *
		 * chain is referenced but not locked.  We must lock the
		 * chain to obtain definitive state.
		 */
		bcopy = *bref;
		if (chain) {
			/*
			 * Use chain already present in the RBTREE
			 */
			hammer2_chain_ref(chain);
			hammer2_spin_unex(&parent->core.spin);
			hammer2_chain_lock(chain, HAMMER2_RESOLVE_NEVER);
		} else {
			/*
			 * Get chain for blockref element.  _get returns NULL
			 * on insertion race.
			 */
			hammer2_spin_unex(&parent->core.spin);
			chain = hammer2_chain_get(parent, generation, &bcopy,
						  HAMMER2_RESOLVE_NEVER);
			if (chain == NULL) {
				reason = 1;
				hammer2_spin_ex(&parent->core.spin);
				continue;
			}
		}

		/*
		 * This is always live so if the chain has been deleted
		 * we raced someone and we have to retry.
		 *
		 * NOTE: Lookups can race delete-duplicate because
		 *	 delete-duplicate does not lock the parent's core
		 *	 (they just use the spinlock on the core).
		 *
		 *	 (note reversed logic for this one)
		 */
		if (bcmp(&bcopy, &chain->bref, sizeof(bcopy)) ||
		    chain->parent != parent ||
		    (chain->flags & HAMMER2_CHAIN_DELETED)) {
			hammer2_chain_unlock(chain);
			hammer2_chain_drop(chain);
			if (hammer2_debug & 0x0040) {
				kprintf("LOST PARENT RETRY "
				"RETRY (%p,%p)->%p %08x\n",
				parent, chain->parent, chain, chain->flags);
			}
			hammer2_spin_ex(&parent->core.spin);
			continue;
		}

		/*
		 * Shift the chain to the indirect block.
		 *
		 * WARNING! No reason for us to load chain data, pass NOSTATS
		 *	    to prevent delete/insert from trying to access
		 *	    inode stats (and thus asserting if there is no
		 *	    chain->data loaded).
		 *
		 * WARNING! The (parent, chain) deletion may modify the parent
		 *	    and invalidate the base pointer.
		 *
		 * WARNING! Parent must already be marked modified, so we
		 *	    can assume that chain_delete always suceeds.
		 *
		 * WARNING! hammer2_chain_repchange() does not have to be
		 *	    called (and doesn't work anyway because we are
		 *	    only doing a partial shift).  A recursion that is
		 *	    in-progress can continue at the current parent
		 *	    and will be able to properly find its next key.
		 */
		error = hammer2_chain_delete(parent, chain, mtid, 0);
		KKASSERT(error == 0);
		hammer2_chain_rename(&ichain, chain, mtid, 0);
		hammer2_chain_unlock(chain);
		hammer2_chain_drop(chain);
		KKASSERT(parent->refs > 0);
		chain = NULL;
		base = NULL;	/* safety */
		hammer2_spin_ex(&parent->core.spin);
next_key_spinlocked:
		if (--maxloops == 0)
			panic("hammer2_chain_create_indirect: maxloops");
		reason = 4;
		if (key_next == 0 || key_next > key_end)
			break;
		key_beg = key_next;
		/* loop */
	}
	hammer2_spin_unex(&parent->core.spin);

	/*
	 * Insert the new indirect block into the parent now that we've
	 * cleared out some entries in the parent.  We calculated a good
	 * insertion index in the loop above (ichain->index).
	 *
	 * We don't have to set UPDATE here because we mark ichain
	 * modified down below (so the normal modified -> flush -> set-moved
	 * sequence applies).
	 *
	 * The insertion shouldn't race as this is a completely new block
	 * and the parent is locked.
	 */
	base = NULL;	/* safety, parent modify may change address */
	KKASSERT((ichain->flags & HAMMER2_CHAIN_ONRBTREE) == 0);
	KKASSERT(parent->core.live_count < count);
	hammer2_chain_insert(parent, ichain,
			     HAMMER2_CHAIN_INSERT_SPIN |
			     HAMMER2_CHAIN_INSERT_LIVE,
			     0);

	/*
	 * Make sure flushes propogate after our manual insertion.
	 */
	hammer2_chain_setflush(ichain);
	hammer2_chain_setflush(parent);

	/*
	 * Figure out what to return.
	 */
	if (~(((hammer2_key_t)1 << keybits) - 1) &
		   (create_key ^ key)) {
		/*
		 * Key being created is outside the key range,
		 * return the original parent.
		 */
		hammer2_chain_unlock(ichain);
		hammer2_chain_drop(ichain);
	} else {
		/*
		 * Otherwise its in the range, return the new parent.
		 * (leave both the new and old parent locked).
		 */
		parent = ichain;
	}

	return(parent);
}

/*
 * Do maintenance on an indirect chain.  Both parent and chain are locked.
 *
 * Returns non-zero if (chain) is deleted, either due to being empty or
 * because its children were safely moved into the parent.
 */
int
hammer2_chain_indirect_maintenance(hammer2_chain_t *parent,
				   hammer2_chain_t *chain)
{
	hammer2_blockref_t *chain_base;
	hammer2_blockref_t *base;
	hammer2_blockref_t *bref;
	hammer2_blockref_t bcopy;
	hammer2_key_t key_next;
	hammer2_key_t key_beg;
	hammer2_key_t key_end;
	hammer2_chain_t *sub;
	int chain_count;
	int count;
	int error;
	int generation;

	/*
	 * Make sure we have an accurate live_count
	 */
	if ((chain->flags & (HAMMER2_CHAIN_INITIAL |
			     HAMMER2_CHAIN_COUNTEDBREFS)) == 0) {
		base = &chain->data->npdata[0];
		count = chain->bytes / sizeof(hammer2_blockref_t);
		hammer2_chain_countbrefs(chain, base, count);
	}

	/*
	 * If the indirect block is empty we can delete it.
	 * (ignore deletion error)
	 */
	if (chain->core.live_count == 0 && RB_EMPTY(&chain->core.rbtree)) {
		hammer2_chain_delete(parent, chain,
				     chain->bref.modify_tid,
				     HAMMER2_DELETE_PERMANENT);
		hammer2_chain_repchange(parent, chain);
		return 1;
	}

	base = hammer2_chain_base_and_count(parent, &count);

	if ((parent->flags & (HAMMER2_CHAIN_INITIAL |
			     HAMMER2_CHAIN_COUNTEDBREFS)) == 0) {
		hammer2_chain_countbrefs(parent, base, count);
	}

	/*
	 * Determine if we can collapse chain into parent, calculate
	 * hysteresis for chain emptiness.
	 */
	if (parent->core.live_count + chain->core.live_count - 1 > count)
		return 0;
	chain_count = chain->bytes / sizeof(hammer2_blockref_t);
	if (chain->core.live_count > chain_count * 3 / 4)
		return 0;

	/*
	 * Ok, theoretically we can collapse chain's contents into
	 * parent.  chain is locked, but any in-memory children of chain
	 * are not.  For this to work, we must be able to dispose of any
	 * in-memory children of chain.
	 *
	 * For now require that there are no in-memory children of chain.
	 *
	 * WARNING! Both chain and parent must remain locked across this
	 *	    entire operation.
	 */

	/*
	 * Parent must be marked modified.  Don't try to collapse it if we
	 * can't mark it modified.  Once modified, destroy chain to make room
	 * and to get rid of what will be a conflicting key (this is included
	 * in the calculation above).  Finally, move the children of chain
	 * into chain's parent.
	 *
	 * This order creates an accounting problem for bref.embed.stats
	 * because we destroy chain before we remove its children.  Any
	 * elements whos blockref is already synchronized will be counted
	 * twice.  To deal with the problem we clean out chain's stats prior
	 * to deleting it.
	 */
	error = hammer2_chain_modify(parent, 0, 0, 0);
	if (error) {
		krateprintf(&krate_h2me, "hammer2: indirect_maint: %s\n",
			    hammer2_error_str(error));
		return 0;
	}
	error = hammer2_chain_modify(chain, chain->bref.modify_tid, 0, 0);
	if (error) {
		krateprintf(&krate_h2me, "hammer2: indirect_maint: %s\n",
			    hammer2_error_str(error));
		return 0;
	}

	chain->bref.embed.stats.inode_count = 0;
	chain->bref.embed.stats.data_count = 0;
	error = hammer2_chain_delete(parent, chain,
				     chain->bref.modify_tid,
				     HAMMER2_DELETE_PERMANENT);
	KKASSERT(error == 0);

	/*
	 * The combined_find call requires core.spin to be held.  One would
	 * think there wouldn't be any conflicts since we hold chain
	 * exclusively locked, but the caching mechanism for 0-ref children
	 * does not require a chain lock.
	 */
	hammer2_spin_ex(&chain->core.spin);

	key_next = 0;
	key_beg = 0;
	key_end = HAMMER2_KEY_MAX;
	for (;;) {
		chain_base = &chain->data->npdata[0];
		chain_count = chain->bytes / sizeof(hammer2_blockref_t);
		sub = hammer2_combined_find(chain, chain_base, chain_count,
					    &key_next,
					    key_beg, key_end,
					    &bref);
		generation = chain->core.generation;
		if (bref == NULL)
			break;
		key_next = bref->key + ((hammer2_key_t)1 << bref->keybits);

		bcopy = *bref;
		if (sub) {
			hammer2_chain_ref(sub);
			hammer2_spin_unex(&chain->core.spin);
			hammer2_chain_lock(sub, HAMMER2_RESOLVE_NEVER);
		} else {
			hammer2_spin_unex(&chain->core.spin);
			sub = hammer2_chain_get(chain, generation, &bcopy,
						HAMMER2_RESOLVE_NEVER);
			if (sub == NULL) {
				hammer2_spin_ex(&chain->core.spin);
				continue;
			}
		}
		if (bcmp(&bcopy, &sub->bref, sizeof(bcopy)) ||
		    sub->parent != chain ||
		    (sub->flags & HAMMER2_CHAIN_DELETED)) {
			hammer2_chain_unlock(sub);
			hammer2_chain_drop(sub);
			hammer2_spin_ex(&chain->core.spin);
			sub = NULL;	/* safety */
			continue;
		}
		error = hammer2_chain_delete(chain, sub,
					     sub->bref.modify_tid, 0);
		KKASSERT(error == 0);
		hammer2_chain_rename(&parent, sub,
				     sub->bref.modify_tid,
				     HAMMER2_INSERT_SAMEPARENT);
		hammer2_chain_unlock(sub);
		hammer2_chain_drop(sub);
		hammer2_spin_ex(&chain->core.spin);

		if (key_next == 0)
			break;
		key_beg = key_next;
	}
	hammer2_spin_unex(&chain->core.spin);

	hammer2_chain_repchange(parent, chain);

	return 1;
}

/*
 * Freemap indirect blocks
 *
 * Calculate the keybits and highside/lowside of the freemap node the
 * caller is creating.
 *
 * This routine will specify the next higher-level freemap key/radix
 * representing the lowest-ordered set.  By doing so, eventually all
 * low-ordered sets will be moved one level down.
 *
 * We have to be careful here because the freemap reserves a limited
 * number of blocks for a limited number of levels.  So we can't just
 * push indiscriminately.
 */
int
hammer2_chain_indkey_freemap(hammer2_chain_t *parent, hammer2_key_t *keyp,
			     int keybits, hammer2_blockref_t *base, int count)
{
	hammer2_chain_t *chain;
	hammer2_blockref_t *bref;
	hammer2_key_t key;
	hammer2_key_t key_beg;
	hammer2_key_t key_end;
	hammer2_key_t key_next;
	int locount;
	int hicount;
	int maxloops = 300000;

	key = *keyp;
	locount = 0;
	hicount = 0;
	keybits = 64;

	/*
	 * Calculate the range of keys in the array being careful to skip
	 * slots which are overridden with a deletion.
	 */
	key_beg = 0;
	key_end = HAMMER2_KEY_MAX;
	hammer2_spin_ex(&parent->core.spin);

	for (;;) {
		if (--maxloops == 0) {
			panic("indkey_freemap shit %p %p:%d\n",
			      parent, base, count);
		}
		chain = hammer2_combined_find(parent, base, count,
					      &key_next,
					      key_beg, key_end,
					      &bref);

		/*
		 * Exhausted search
		 */
		if (bref == NULL)
			break;

		/*
		 * Skip deleted chains.
		 */
		if (chain && (chain->flags & HAMMER2_CHAIN_DELETED)) {
			if (key_next == 0 || key_next > key_end)
				break;
			key_beg = key_next;
			continue;
		}

		/*
		 * Use the full live (not deleted) element for the scan
		 * iteration.  HAMMER2 does not allow partial replacements.
		 *
		 * XXX should be built into hammer2_combined_find().
		 */
		key_next = bref->key + ((hammer2_key_t)1 << bref->keybits);

		if (keybits > bref->keybits) {
			key = bref->key;
			keybits = bref->keybits;
		} else if (keybits == bref->keybits && bref->key < key) {
			key = bref->key;
		}
		if (key_next == 0)
			break;
		key_beg = key_next;
	}
	hammer2_spin_unex(&parent->core.spin);

	/*
	 * Return the keybits for a higher-level FREEMAP_NODE covering
	 * this node.
	 */
	switch(keybits) {
	case HAMMER2_FREEMAP_LEVEL0_RADIX:
		keybits = HAMMER2_FREEMAP_LEVEL1_RADIX;
		break;
	case HAMMER2_FREEMAP_LEVEL1_RADIX:
		keybits = HAMMER2_FREEMAP_LEVEL2_RADIX;
		break;
	case HAMMER2_FREEMAP_LEVEL2_RADIX:
		keybits = HAMMER2_FREEMAP_LEVEL3_RADIX;
		break;
	case HAMMER2_FREEMAP_LEVEL3_RADIX:
		keybits = HAMMER2_FREEMAP_LEVEL4_RADIX;
		break;
	case HAMMER2_FREEMAP_LEVEL4_RADIX:
		keybits = HAMMER2_FREEMAP_LEVEL5_RADIX;
		break;
	case HAMMER2_FREEMAP_LEVEL5_RADIX:
		panic("hammer2_chain_indkey_freemap: level too high");
		break;
	default:
		panic("hammer2_chain_indkey_freemap: bad radix");
		break;
	}
	*keyp = key;

	return (keybits);
}

/*
 * File indirect blocks
 *
 * Calculate the key/keybits for the indirect block to create by scanning
 * existing keys.  The key being created is also passed in *keyp and can be
 * inside or outside the indirect block.  Regardless, the indirect block
 * must hold at least two keys in order to guarantee sufficient space.
 *
 * We use a modified version of the freemap's fixed radix tree, but taylored
 * for file data.  Basically we configure an indirect block encompassing the
 * smallest key.
 */
static int
hammer2_chain_indkey_file(hammer2_chain_t *parent, hammer2_key_t *keyp,
			    int keybits, hammer2_blockref_t *base, int count,
			    int ncount)
{
	hammer2_chain_t *chain;
	hammer2_blockref_t *bref;
	hammer2_key_t key;
	hammer2_key_t key_beg;
	hammer2_key_t key_end;
	hammer2_key_t key_next;
	int nradix;
	int locount;
	int hicount;
	int maxloops = 300000;

	key = *keyp;
	locount = 0;
	hicount = 0;
	keybits = 64;

	/*
	 * Calculate the range of keys in the array being careful to skip
	 * slots which are overridden with a deletion.
	 *
	 * Locate the smallest key.
	 */
	key_beg = 0;
	key_end = HAMMER2_KEY_MAX;
	hammer2_spin_ex(&parent->core.spin);

	for (;;) {
		if (--maxloops == 0) {
			panic("indkey_freemap shit %p %p:%d\n",
			      parent, base, count);
		}
		chain = hammer2_combined_find(parent, base, count,
					      &key_next,
					      key_beg, key_end,
					      &bref);

		/*
		 * Exhausted search
		 */
		if (bref == NULL)
			break;

		/*
		 * Skip deleted chains.
		 */
		if (chain && (chain->flags & HAMMER2_CHAIN_DELETED)) {
			if (key_next == 0 || key_next > key_end)
				break;
			key_beg = key_next;
			continue;
		}

		/*
		 * Use the full live (not deleted) element for the scan
		 * iteration.  HAMMER2 does not allow partial replacements.
		 *
		 * XXX should be built into hammer2_combined_find().
		 */
		key_next = bref->key + ((hammer2_key_t)1 << bref->keybits);

		if (keybits > bref->keybits) {
			key = bref->key;
			keybits = bref->keybits;
		} else if (keybits == bref->keybits && bref->key < key) {
			key = bref->key;
		}
		if (key_next == 0)
			break;
		key_beg = key_next;
	}
	hammer2_spin_unex(&parent->core.spin);

	/*
	 * Calculate the static keybits for a higher-level indirect block
	 * that contains the key.
	 */
	*keyp = key;

	switch(ncount) {
	case HAMMER2_IND_BYTES_MIN / sizeof(hammer2_blockref_t):
		nradix = HAMMER2_IND_RADIX_MIN - HAMMER2_BLOCKREF_RADIX;
		break;
	case HAMMER2_IND_BYTES_NOM / sizeof(hammer2_blockref_t):
		nradix = HAMMER2_IND_RADIX_NOM - HAMMER2_BLOCKREF_RADIX;
		break;
	case HAMMER2_IND_BYTES_MAX / sizeof(hammer2_blockref_t):
		nradix = HAMMER2_IND_RADIX_MAX - HAMMER2_BLOCKREF_RADIX;
		break;
	default:
		panic("bad ncount %d\n", ncount);
		nradix = 0;
		break;
	}

	/*
	 * The largest radix that can be returned for an indirect block is
	 * 63 bits.  (The largest practical indirect block radix is actually
	 * 62 bits because the top-level inode or volume root contains four
	 * entries, but allow 63 to be returned).
	 */
	if (nradix >= 64)
		nradix = 63;

	return keybits + nradix;
}

#if 1

/*
 * Directory indirect blocks.
 *
 * Covers both the inode index (directory of inodes), and directory contents
 * (filenames hardlinked to inodes).
 *
 * Because directory keys are hashed we generally try to cut the space in
 * half.  We accomodate the inode index (which tends to have linearly
 * increasing inode numbers) by ensuring that the keyspace is at least large
 * enough to fill up the indirect block being created.
 */
static int
hammer2_chain_indkey_dir(hammer2_chain_t *parent, hammer2_key_t *keyp,
			 int keybits, hammer2_blockref_t *base, int count,
			 int ncount)
{
	hammer2_blockref_t *bref;
	hammer2_chain_t	*chain;
	hammer2_key_t key_beg;
	hammer2_key_t key_end;
	hammer2_key_t key_next;
	hammer2_key_t key;
	int nkeybits;
	int locount;
	int hicount;
	int maxloops = 300000;

	/*
	 * NOTE: We can't take a shortcut here anymore for inodes because
	 *	 the root directory can contain a mix of inodes and directory
	 *	 entries (we used to just return 63 if parent->bref.type was
	 *	 HAMMER2_BREF_TYPE_INODE.
	 */
	key = *keyp;
	locount = 0;
	hicount = 0;

	/*
	 * Calculate the range of keys in the array being careful to skip
	 * slots which are overridden with a deletion.
	 */
	key_beg = 0;
	key_end = HAMMER2_KEY_MAX;
	hammer2_spin_ex(&parent->core.spin);

	for (;;) {
		if (--maxloops == 0) {
			panic("indkey_freemap shit %p %p:%d\n",
			      parent, base, count);
		}
		chain = hammer2_combined_find(parent, base, count,
					      &key_next,
					      key_beg, key_end,
					      &bref);

		/*
		 * Exhausted search
		 */
		if (bref == NULL)
			break;

		/*
		 * Deleted object
		 */
		if (chain && (chain->flags & HAMMER2_CHAIN_DELETED)) {
			if (key_next == 0 || key_next > key_end)
				break;
			key_beg = key_next;
			continue;
		}

		/*
		 * Use the full live (not deleted) element for the scan
		 * iteration.  HAMMER2 does not allow partial replacements.
		 *
		 * XXX should be built into hammer2_combined_find().
		 */
		key_next = bref->key + ((hammer2_key_t)1 << bref->keybits);

		/*
		 * Expand our calculated key range (key, keybits) to fit
		 * the scanned key.  nkeybits represents the full range
		 * that we will later cut in half (two halves @ nkeybits - 1).
		 */
		nkeybits = keybits;
		if (nkeybits < bref->keybits) {
			if (bref->keybits > 64) {
				kprintf("bad bref chain %p bref %p\n",
					chain, bref);
				Debugger("fubar");
			}
			nkeybits = bref->keybits;
		}
		while (nkeybits < 64 &&
		       (~(((hammer2_key_t)1 << nkeybits) - 1) &
		        (key ^ bref->key)) != 0) {
			++nkeybits;
		}

		/*
		 * If the new key range is larger we have to determine
		 * which side of the new key range the existing keys fall
		 * under by checking the high bit, then collapsing the
		 * locount into the hicount or vise-versa.
		 */
		if (keybits != nkeybits) {
			if (((hammer2_key_t)1 << (nkeybits - 1)) & key) {
				hicount += locount;
				locount = 0;
			} else {
				locount += hicount;
				hicount = 0;
			}
			keybits = nkeybits;
		}

		/*
		 * The newly scanned key will be in the lower half or the
		 * upper half of the (new) key range.
		 */
		if (((hammer2_key_t)1 << (nkeybits - 1)) & bref->key)
			++hicount;
		else
			++locount;

		if (key_next == 0)
			break;
		key_beg = key_next;
	}
	hammer2_spin_unex(&parent->core.spin);
	bref = NULL;	/* now invalid (safety) */

	/*
	 * Adjust keybits to represent half of the full range calculated
	 * above (radix 63 max) for our new indirect block.
	 */
	--keybits;

	/*
	 * Expand keybits to hold at least ncount elements.  ncount will be
	 * a power of 2.  This is to try to completely fill leaf nodes (at
	 * least for keys which are not hashes).
	 *
	 * We aren't counting 'in' or 'out', we are counting 'high side'
	 * and 'low side' based on the bit at (1LL << keybits).  We want
	 * everything to be inside in these cases so shift it all to
	 * the low or high side depending on the new high bit.
	 */
	while (((hammer2_key_t)1 << keybits) < ncount) {
		++keybits;
		if (key & ((hammer2_key_t)1 << keybits)) {
			hicount += locount;
			locount = 0;
		} else {
			locount += hicount;
			hicount = 0;
		}
	}

	if (hicount > locount)
		key |= (hammer2_key_t)1 << keybits;
	else
		key &= ~(hammer2_key_t)1 << keybits;

	*keyp = key;

	return (keybits);
}

#else

/*
 * Directory indirect blocks.
 *
 * Covers both the inode index (directory of inodes), and directory contents
 * (filenames hardlinked to inodes).
 *
 * Because directory keys are hashed we generally try to cut the space in
 * half.  We accomodate the inode index (which tends to have linearly
 * increasing inode numbers) by ensuring that the keyspace is at least large
 * enough to fill up the indirect block being created.
 */
static int
hammer2_chain_indkey_dir(hammer2_chain_t *parent, hammer2_key_t *keyp,
			 int keybits, hammer2_blockref_t *base, int count,
			 int ncount)
{
	hammer2_blockref_t *bref;
	hammer2_chain_t	*chain;
	hammer2_key_t key_beg;
	hammer2_key_t key_end;
	hammer2_key_t key_next;
	hammer2_key_t key;
	int nkeybits;
	int locount;
	int hicount;
	int maxloops = 300000;

	/*
	 * Shortcut if the parent is the inode.  In this situation the
	 * parent has 4+1 directory entries and we are creating an indirect
	 * block capable of holding many more.
	 */
	if (parent->bref.type == HAMMER2_BREF_TYPE_INODE) {
		return 63;
	}

	key = *keyp;
	locount = 0;
	hicount = 0;

	/*
	 * Calculate the range of keys in the array being careful to skip
	 * slots which are overridden with a deletion.
	 */
	key_beg = 0;
	key_end = HAMMER2_KEY_MAX;
	hammer2_spin_ex(&parent->core.spin);

	for (;;) {
		if (--maxloops == 0) {
			panic("indkey_freemap shit %p %p:%d\n",
			      parent, base, count);
		}
		chain = hammer2_combined_find(parent, base, count,
					      &key_next,
					      key_beg, key_end,
					      &bref);

		/*
		 * Exhausted search
		 */
		if (bref == NULL)
			break;

		/*
		 * Deleted object
		 */
		if (chain && (chain->flags & HAMMER2_CHAIN_DELETED)) {
			if (key_next == 0 || key_next > key_end)
				break;
			key_beg = key_next;
			continue;
		}

		/*
		 * Use the full live (not deleted) element for the scan
		 * iteration.  HAMMER2 does not allow partial replacements.
		 *
		 * XXX should be built into hammer2_combined_find().
		 */
		key_next = bref->key + ((hammer2_key_t)1 << bref->keybits);

		/*
		 * Expand our calculated key range (key, keybits) to fit
		 * the scanned key.  nkeybits represents the full range
		 * that we will later cut in half (two halves @ nkeybits - 1).
		 */
		nkeybits = keybits;
		if (nkeybits < bref->keybits) {
			if (bref->keybits > 64) {
				kprintf("bad bref chain %p bref %p\n",
					chain, bref);
				Debugger("fubar");
			}
			nkeybits = bref->keybits;
		}
		while (nkeybits < 64 &&
		       (~(((hammer2_key_t)1 << nkeybits) - 1) &
		        (key ^ bref->key)) != 0) {
			++nkeybits;
		}

		/*
		 * If the new key range is larger we have to determine
		 * which side of the new key range the existing keys fall
		 * under by checking the high bit, then collapsing the
		 * locount into the hicount or vise-versa.
		 */
		if (keybits != nkeybits) {
			if (((hammer2_key_t)1 << (nkeybits - 1)) & key) {
				hicount += locount;
				locount = 0;
			} else {
				locount += hicount;
				hicount = 0;
			}
			keybits = nkeybits;
		}

		/*
		 * The newly scanned key will be in the lower half or the
		 * upper half of the (new) key range.
		 */
		if (((hammer2_key_t)1 << (nkeybits - 1)) & bref->key)
			++hicount;
		else
			++locount;

		if (key_next == 0)
			break;
		key_beg = key_next;
	}
	hammer2_spin_unex(&parent->core.spin);
	bref = NULL;	/* now invalid (safety) */

	/*
	 * Adjust keybits to represent half of the full range calculated
	 * above (radix 63 max) for our new indirect block.
	 */
	--keybits;

	/*
	 * Expand keybits to hold at least ncount elements.  ncount will be
	 * a power of 2.  This is to try to completely fill leaf nodes (at
	 * least for keys which are not hashes).
	 *
	 * We aren't counting 'in' or 'out', we are counting 'high side'
	 * and 'low side' based on the bit at (1LL << keybits).  We want
	 * everything to be inside in these cases so shift it all to
	 * the low or high side depending on the new high bit.
	 */
	while (((hammer2_key_t)1 << keybits) < ncount) {
		++keybits;
		if (key & ((hammer2_key_t)1 << keybits)) {
			hicount += locount;
			locount = 0;
		} else {
			locount += hicount;
			hicount = 0;
		}
	}

	if (hicount > locount)
		key |= (hammer2_key_t)1 << keybits;
	else
		key &= ~(hammer2_key_t)1 << keybits;

	*keyp = key;

	return (keybits);
}

#endif

/*
 * Sets CHAIN_DELETED and remove the chain's blockref from the parent if
 * it exists.
 *
 * Both parent and chain must be locked exclusively.
 *
 * This function will modify the parent if the blockref requires removal
 * from the parent's block table.
 *
 * This function is NOT recursive.  Any entity already pushed into the
 * chain (such as an inode) may still need visibility into its contents,
 * as well as the ability to read and modify the contents.  For example,
 * for an unlinked file which is still open.
 *
 * Also note that the flusher is responsible for cleaning up empty
 * indirect blocks.
 */
int
hammer2_chain_delete(hammer2_chain_t *parent, hammer2_chain_t *chain,
		     hammer2_tid_t mtid, int flags)
{
	int error = 0;

	KKASSERT(hammer2_mtx_owned(&chain->lock));

	/*
	 * Nothing to do if already marked.
	 *
	 * We need the spinlock on the core whos RBTREE contains chain
	 * to protect against races.
	 */
	if ((chain->flags & HAMMER2_CHAIN_DELETED) == 0) {
		KKASSERT((chain->flags & HAMMER2_CHAIN_DELETED) == 0 &&
			 chain->parent == parent);
		error = _hammer2_chain_delete_helper(parent, chain,
						     mtid, flags);
	}

	/*
	 * Permanent deletions mark the chain as destroyed.
	 *
	 * NOTE: We do not setflush the chain unless the deletion is
	 *	 permanent, since the deletion of a chain does not actually
	 *	 require it to be flushed.
	 */
	if (error == 0) {
		if (flags & HAMMER2_DELETE_PERMANENT) {
			atomic_set_int(&chain->flags, HAMMER2_CHAIN_DESTROY);
			hammer2_chain_setflush(chain);
		}
	}

	return error;
}

/*
 * Returns the index of the nearest element in the blockref array >= elm.
 * Returns (count) if no element could be found.
 *
 * Sets *key_nextp to the next key for loop purposes but does not modify
 * it if the next key would be higher than the current value of *key_nextp.
 * Note that *key_nexp can overflow to 0, which should be tested by the
 * caller.
 *
 * WARNING!  Must be called with parent's spinlock held.  Spinlock remains
 *	     held through the operation.
 */
static int
hammer2_base_find(hammer2_chain_t *parent,
		  hammer2_blockref_t *base, int count,
		  hammer2_key_t *key_nextp,
		  hammer2_key_t key_beg, hammer2_key_t key_end)
{
	hammer2_blockref_t *scan;
	hammer2_key_t scan_end;
	int i;
	int limit;

	/*
	 * Require the live chain's already have their core's counted
	 * so we can optimize operations.
	 */
        KKASSERT(parent->flags & HAMMER2_CHAIN_COUNTEDBREFS);

	/*
	 * Degenerate case
	 */
	if (count == 0 || base == NULL)
		return(count);

	/*
	 * Sequential optimization using parent->cache_index.  This is
	 * the most likely scenario.
	 *
	 * We can avoid trailing empty entries on live chains, otherwise
	 * we might have to check the whole block array.
	 */
	i = parent->cache_index;	/* SMP RACE OK */
	cpu_ccfence();
	limit = parent->core.live_zero;
	if (i >= limit)
		i = limit - 1;
	if (i < 0)
		i = 0;
	KKASSERT(i < count);

	/*
	 * Search backwards
	 */
	scan = &base[i];
	while (i > 0 && (scan->type == 0 || scan->key > key_beg)) {
		--scan;
		--i;
	}
	parent->cache_index = i;

	/*
	 * Search forwards, stop when we find a scan element which
	 * encloses the key or until we know that there are no further
	 * elements.
	 */
	while (i < count) {
		if (scan->type != 0) {
			scan_end = scan->key +
				   ((hammer2_key_t)1 << scan->keybits) - 1;
			if (scan->key > key_beg || scan_end >= key_beg)
				break;
		}
		if (i >= limit)
			return (count);
		++scan;
		++i;
	}
	if (i != count) {
		parent->cache_index = i;
		if (i >= limit) {
			i = count;
		} else {
			scan_end = scan->key +
				   ((hammer2_key_t)1 << scan->keybits);
			if (scan_end && (*key_nextp > scan_end ||
					 *key_nextp == 0)) {
				*key_nextp = scan_end;
			}
		}
	}
	return (i);
}

/*
 * Do a combined search and return the next match either from the blockref
 * array or from the in-memory chain.  Sets *bresp to the returned bref in
 * both cases, or sets it to NULL if the search exhausted.  Only returns
 * a non-NULL chain if the search matched from the in-memory chain.
 *
 * When no in-memory chain has been found and a non-NULL bref is returned
 * in *bresp.
 *
 *
 * The returned chain is not locked or referenced.  Use the returned bref
 * to determine if the search exhausted or not.  Iterate if the base find
 * is chosen but matches a deleted chain.
 *
 * WARNING!  Must be called with parent's spinlock held.  Spinlock remains
 *	     held through the operation.
 */
hammer2_chain_t *
hammer2_combined_find(hammer2_chain_t *parent,
		      hammer2_blockref_t *base, int count,
		      hammer2_key_t *key_nextp,
		      hammer2_key_t key_beg, hammer2_key_t key_end,
		      hammer2_blockref_t **bresp)
{
	hammer2_blockref_t *bref;
	hammer2_chain_t *chain;
	int i;

	/*
	 * Lookup in block array and in rbtree.
	 */
	*key_nextp = key_end + 1;
	i = hammer2_base_find(parent, base, count, key_nextp,
			      key_beg, key_end);
	chain = hammer2_chain_find(parent, key_nextp, key_beg, key_end);

	/*
	 * Neither matched
	 */
	if (i == count && chain == NULL) {
		*bresp = NULL;
		return(NULL);
	}

	/*
	 * Only chain matched.
	 */
	if (i == count) {
		bref = &chain->bref;
		goto found;
	}

	/*
	 * Only blockref matched.
	 */
	if (chain == NULL) {
		bref = &base[i];
		goto found;
	}

	/*
	 * Both in-memory and blockref matched, select the nearer element.
	 *
	 * If both are flush with the left-hand side or both are the
	 * same distance away, select the chain.  In this situation the
	 * chain must have been loaded from the matching blockmap.
	 */
	if ((chain->bref.key <= key_beg && base[i].key <= key_beg) ||
	    chain->bref.key == base[i].key) {
		KKASSERT(chain->bref.key == base[i].key);
		bref = &chain->bref;
		goto found;
	}

	/*
	 * Select the nearer key
	 */
	if (chain->bref.key < base[i].key) {
		bref = &chain->bref;
	} else {
		bref = &base[i];
		chain = NULL;
	}

	/*
	 * If the bref is out of bounds we've exhausted our search.
	 */
found:
	if (bref->key > key_end) {
		*bresp = NULL;
		chain = NULL;
	} else {
		*bresp = bref;
	}
	return(chain);
}

/*
 * Locate the specified block array element and delete it.  The element
 * must exist.
 *
 * The spin lock on the related chain must be held.
 *
 * NOTE: live_count was adjusted when the chain was deleted, so it does not
 *	 need to be adjusted when we commit the media change.
 */
void
hammer2_base_delete(hammer2_chain_t *parent,
		    hammer2_blockref_t *base, int count,
		    hammer2_chain_t *chain)
{
	hammer2_blockref_t *elm = &chain->bref;
	hammer2_blockref_t *scan;
	hammer2_key_t key_next;
	int i;

	/*
	 * Delete element.  Expect the element to exist.
	 *
	 * XXX see caller, flush code not yet sophisticated enough to prevent
	 *     re-flushed in some cases.
	 */
	key_next = 0; /* max range */
	i = hammer2_base_find(parent, base, count, &key_next,
			      elm->key, elm->key);
	scan = &base[i];
	if (i == count || scan->type == 0 ||
	    scan->key != elm->key ||
	    ((chain->flags & HAMMER2_CHAIN_BMAPUPD) == 0 &&
	     scan->keybits != elm->keybits)) {
		hammer2_spin_unex(&parent->core.spin);
		panic("delete base %p element not found at %d/%d elm %p\n",
		      base, i, count, elm);
		return;
	}

	/*
	 * Update stats and zero the entry.
	 *
	 * NOTE: Handle radix == 0 (0 bytes) case.
	 */
	if ((int)(scan->data_off & HAMMER2_OFF_MASK_RADIX)) {
		parent->bref.embed.stats.data_count -= (hammer2_off_t)1 <<
				(int)(scan->data_off & HAMMER2_OFF_MASK_RADIX);
	}
	switch(scan->type) {
	case HAMMER2_BREF_TYPE_INODE:
		--parent->bref.embed.stats.inode_count;
		/* fall through */
	case HAMMER2_BREF_TYPE_DATA:
		if (parent->bref.leaf_count == HAMMER2_BLOCKREF_LEAF_MAX) {
			atomic_set_int(&chain->flags,
				       HAMMER2_CHAIN_HINT_LEAF_COUNT);
		} else {
			if (parent->bref.leaf_count)
				--parent->bref.leaf_count;
		}
		/* fall through */
	case HAMMER2_BREF_TYPE_INDIRECT:
		if (scan->type != HAMMER2_BREF_TYPE_DATA) {
			parent->bref.embed.stats.data_count -=
				scan->embed.stats.data_count;
			parent->bref.embed.stats.inode_count -=
				scan->embed.stats.inode_count;
		}
		if (scan->type == HAMMER2_BREF_TYPE_INODE)
			break;
		if (parent->bref.leaf_count == HAMMER2_BLOCKREF_LEAF_MAX) {
			atomic_set_int(&chain->flags,
				       HAMMER2_CHAIN_HINT_LEAF_COUNT);
		} else {
			if (parent->bref.leaf_count <= scan->leaf_count)
				parent->bref.leaf_count = 0;
			else
				parent->bref.leaf_count -= scan->leaf_count;
		}
		break;
	case HAMMER2_BREF_TYPE_DIRENT:
		if (parent->bref.leaf_count == HAMMER2_BLOCKREF_LEAF_MAX) {
			atomic_set_int(&chain->flags,
				       HAMMER2_CHAIN_HINT_LEAF_COUNT);
		} else {
			if (parent->bref.leaf_count)
				--parent->bref.leaf_count;
		}
	default:
		break;
	}

	bzero(scan, sizeof(*scan));

	/*
	 * We can only optimize parent->core.live_zero for live chains.
	 */
	if (parent->core.live_zero == i + 1) {
		while (--i >= 0 && base[i].type == 0)
			;
		parent->core.live_zero = i + 1;
	}

	/*
	 * Clear appropriate blockmap flags in chain.
	 */
	atomic_clear_int(&chain->flags, HAMMER2_CHAIN_BMAPPED |
					HAMMER2_CHAIN_BMAPUPD);
}

/*
 * Insert the specified element.  The block array must not already have the
 * element and must have space available for the insertion.
 *
 * The spin lock on the related chain must be held.
 *
 * NOTE: live_count was adjusted when the chain was deleted, so it does not
 *	 need to be adjusted when we commit the media change.
 */
void
hammer2_base_insert(hammer2_chain_t *parent,
		    hammer2_blockref_t *base, int count,
		    hammer2_chain_t *chain, hammer2_blockref_t *elm)
{
	hammer2_key_t key_next;
	hammer2_key_t xkey;
	int i;
	int j;
	int k;
	int l;
	int u = 1;

	/*
	 * Insert new element.  Expect the element to not already exist
	 * unless we are replacing it.
	 *
	 * XXX see caller, flush code not yet sophisticated enough to prevent
	 *     re-flushed in some cases.
	 */
	key_next = 0; /* max range */
	i = hammer2_base_find(parent, base, count, &key_next,
			      elm->key, elm->key);

	/*
	 * Shortcut fill optimization, typical ordered insertion(s) may not
	 * require a search.
	 */
	KKASSERT(i >= 0 && i <= count);

	/*
	 * Set appropriate blockmap flags in chain (if not NULL)
	 */
	if (chain)
		atomic_set_int(&chain->flags, HAMMER2_CHAIN_BMAPPED);

	/*
	 * Update stats and zero the entry
	 */
	if ((int)(elm->data_off & HAMMER2_OFF_MASK_RADIX)) {
		parent->bref.embed.stats.data_count += (hammer2_off_t)1 <<
				(int)(elm->data_off & HAMMER2_OFF_MASK_RADIX);
	}
	switch(elm->type) {
	case HAMMER2_BREF_TYPE_INODE:
		++parent->bref.embed.stats.inode_count;
		/* fall through */
	case HAMMER2_BREF_TYPE_DATA:
		if (parent->bref.leaf_count != HAMMER2_BLOCKREF_LEAF_MAX)
			++parent->bref.leaf_count;
		/* fall through */
	case HAMMER2_BREF_TYPE_INDIRECT:
		if (elm->type != HAMMER2_BREF_TYPE_DATA) {
			parent->bref.embed.stats.data_count +=
				elm->embed.stats.data_count;
			parent->bref.embed.stats.inode_count +=
				elm->embed.stats.inode_count;
		}
		if (elm->type == HAMMER2_BREF_TYPE_INODE)
			break;
		if (parent->bref.leaf_count + elm->leaf_count <
		    HAMMER2_BLOCKREF_LEAF_MAX) {
			parent->bref.leaf_count += elm->leaf_count;
		} else {
			parent->bref.leaf_count = HAMMER2_BLOCKREF_LEAF_MAX;
		}
		break;
	case HAMMER2_BREF_TYPE_DIRENT:
		if (parent->bref.leaf_count != HAMMER2_BLOCKREF_LEAF_MAX)
			++parent->bref.leaf_count;
		break;
	default:
		break;
	}


	/*
	 * We can only optimize parent->core.live_zero for live chains.
	 */
	if (i == count && parent->core.live_zero < count) {
		i = parent->core.live_zero++;
		base[i] = *elm;
		return;
	}

	xkey = elm->key + ((hammer2_key_t)1 << elm->keybits) - 1;
	if (i != count && (base[i].key < elm->key || xkey >= base[i].key)) {
		hammer2_spin_unex(&parent->core.spin);
		panic("insert base %p overlapping elements at %d elm %p\n",
		      base, i, elm);
	}

	/*
	 * Try to find an empty slot before or after.
	 */
	j = i;
	k = i;
	while (j > 0 || k < count) {
		--j;
		if (j >= 0 && base[j].type == 0) {
			if (j == i - 1) {
				base[j] = *elm;
			} else {
				bcopy(&base[j+1], &base[j],
				      (i - j - 1) * sizeof(*base));
				base[i - 1] = *elm;
			}
			goto validate;
		}
		++k;
		if (k < count && base[k].type == 0) {
			bcopy(&base[i], &base[i+1],
			      (k - i) * sizeof(hammer2_blockref_t));
			base[i] = *elm;

			/*
			 * We can only update parent->core.live_zero for live
			 * chains.
			 */
			if (parent->core.live_zero <= k)
				parent->core.live_zero = k + 1;
			u = 2;
			goto validate;
		}
	}
	panic("hammer2_base_insert: no room!");

	/*
	 * Debugging
	 */
validate:
	key_next = 0;
	for (l = 0; l < count; ++l) {
		if (base[l].type) {
			key_next = base[l].key +
				   ((hammer2_key_t)1 << base[l].keybits) - 1;
			break;
		}
	}
	while (++l < count) {
		if (base[l].type) {
			if (base[l].key <= key_next)
				panic("base_insert %d %d,%d,%d fail %p:%d", u, i, j, k, base, l);
			key_next = base[l].key +
				   ((hammer2_key_t)1 << base[l].keybits) - 1;

		}
	}

}

#if 0

/*
 * Sort the blockref array for the chain.  Used by the flush code to
 * sort the blockref[] array.
 *
 * The chain must be exclusively locked AND spin-locked.
 */
typedef hammer2_blockref_t *hammer2_blockref_p;

static
int
hammer2_base_sort_callback(const void *v1, const void *v2)
{
	hammer2_blockref_p bref1 = *(const hammer2_blockref_p *)v1;
	hammer2_blockref_p bref2 = *(const hammer2_blockref_p *)v2;

	/*
	 * Make sure empty elements are placed at the end of the array
	 */
	if (bref1->type == 0) {
		if (bref2->type == 0)
			return(0);
		return(1);
	} else if (bref2->type == 0) {
		return(-1);
	}

	/*
	 * Sort by key
	 */
	if (bref1->key < bref2->key)
		return(-1);
	if (bref1->key > bref2->key)
		return(1);
	return(0);
}

void
hammer2_base_sort(hammer2_chain_t *chain)
{
	hammer2_blockref_t *base;
	int count;

	switch(chain->bref.type) {
	case HAMMER2_BREF_TYPE_INODE:
		/*
		 * Special shortcut for embedded data returns the inode
		 * itself.  Callers must detect this condition and access
		 * the embedded data (the strategy code does this for us).
		 *
		 * This is only applicable to regular files and softlinks.
		 */
		if (chain->data->ipdata.meta.op_flags &
		    HAMMER2_OPFLAG_DIRECTDATA) {
			return;
		}
		base = &chain->data->ipdata.u.blockset.blockref[0];
		count = HAMMER2_SET_COUNT;
		break;
	case HAMMER2_BREF_TYPE_FREEMAP_NODE:
	case HAMMER2_BREF_TYPE_INDIRECT:
		/*
		 * Optimize indirect blocks in the INITIAL state to avoid
		 * I/O.
		 */
		KKASSERT((chain->flags & HAMMER2_CHAIN_INITIAL) == 0);
		base = &chain->data->npdata[0];
		count = chain->bytes / sizeof(hammer2_blockref_t);
		break;
	case HAMMER2_BREF_TYPE_VOLUME:
		base = &chain->data->voldata.sroot_blockset.blockref[0];
		count = HAMMER2_SET_COUNT;
		break;
	case HAMMER2_BREF_TYPE_FREEMAP:
		base = &chain->data->blkset.blockref[0];
		count = HAMMER2_SET_COUNT;
		break;
	default:
		kprintf("hammer2_chain_lookup: unrecognized "
			"blockref(A) type: %d",
		        chain->bref.type);
		while (1)
			tsleep(&base, 0, "dead", 0);
		panic("hammer2_chain_lookup: unrecognized "
		      "blockref(A) type: %d",
		      chain->bref.type);
		base = NULL;	/* safety */
		count = 0;	/* safety */
	}
	kqsort(base, count, sizeof(*base), hammer2_base_sort_callback);
}

#endif

/*
 * Chain memory management
 */
void
hammer2_chain_wait(hammer2_chain_t *chain)
{
	tsleep(chain, 0, "chnflw", 1);
}

const hammer2_media_data_t *
hammer2_chain_rdata(hammer2_chain_t *chain)
{
	KKASSERT(chain->data != NULL);
	return (chain->data);
}

hammer2_media_data_t *
hammer2_chain_wdata(hammer2_chain_t *chain)
{
	KKASSERT(chain->data != NULL);
	return (chain->data);
}

/*
 * Set the check data for a chain.  This can be a heavy-weight operation
 * and typically only runs on-flush.  For file data check data is calculated
 * when the logical buffers are flushed.
 */
void
hammer2_chain_setcheck(hammer2_chain_t *chain, void *bdata)
{
	chain->bref.flags &= ~HAMMER2_BREF_FLAG_ZERO;

	switch(HAMMER2_DEC_CHECK(chain->bref.methods)) {
	case HAMMER2_CHECK_NONE:
		break;
	case HAMMER2_CHECK_DISABLED:
		break;
	case HAMMER2_CHECK_ISCSI32:
		chain->bref.check.iscsi32.value =
			hammer2_icrc32(bdata, chain->bytes);
		break;
	case HAMMER2_CHECK_XXHASH64:
		chain->bref.check.xxhash64.value =
			XXH64(bdata, chain->bytes, XXH_HAMMER2_SEED);
		break;
	case HAMMER2_CHECK_SHA192:
		{
			SHA256_CTX hash_ctx;
			union {
				uint8_t digest[SHA256_DIGEST_LENGTH];
				uint64_t digest64[SHA256_DIGEST_LENGTH/8];
			} u;

			SHA256_Init(&hash_ctx);
			SHA256_Update(&hash_ctx, bdata, chain->bytes);
			SHA256_Final(u.digest, &hash_ctx);
			u.digest64[2] ^= u.digest64[3];
			bcopy(u.digest,
			      chain->bref.check.sha192.data,
			      sizeof(chain->bref.check.sha192.data));
		}
		break;
	case HAMMER2_CHECK_FREEMAP:
		chain->bref.check.freemap.icrc32 =
			hammer2_icrc32(bdata, chain->bytes);
		break;
	default:
		kprintf("hammer2_chain_setcheck: unknown check type %02x\n",
			chain->bref.methods);
		break;
	}
}

int
hammer2_chain_testcheck(hammer2_chain_t *chain, void *bdata)
{
	uint32_t check32;
	uint64_t check64;
	int r;

	if (chain->bref.flags & HAMMER2_BREF_FLAG_ZERO)
		return 1;

	switch(HAMMER2_DEC_CHECK(chain->bref.methods)) {
	case HAMMER2_CHECK_NONE:
		r = 1;
		break;
	case HAMMER2_CHECK_DISABLED:
		r = 1;
		break;
	case HAMMER2_CHECK_ISCSI32:
		check32 = hammer2_icrc32(bdata, chain->bytes);
		r = (chain->bref.check.iscsi32.value == check32);
		if (r == 0) {
			kprintf("chain %016jx.%02x meth=%02x CHECK FAIL "
				"(flags=%08x, bref/data %08x/%08x)\n",
				chain->bref.data_off,
				chain->bref.type,
				chain->bref.methods,
				chain->flags,
				chain->bref.check.iscsi32.value,
				check32);
		}
		hammer2_process_icrc32 += chain->bytes;
		break;
	case HAMMER2_CHECK_XXHASH64:
		check64 = XXH64(bdata, chain->bytes, XXH_HAMMER2_SEED);
		r = (chain->bref.check.xxhash64.value == check64);
		if (r == 0) {
			kprintf("chain %016jx.%02x key=%016jx "
				"meth=%02x CHECK FAIL "
				"(flags=%08x, bref/data %016jx/%016jx)\n",
				chain->bref.data_off,
				chain->bref.type,
				chain->bref.key,
				chain->bref.methods,
				chain->flags,
				chain->bref.check.xxhash64.value,
				check64);
		}
		hammer2_process_xxhash64 += chain->bytes;
		break;
	case HAMMER2_CHECK_SHA192:
		{
			SHA256_CTX hash_ctx;
			union {
				uint8_t digest[SHA256_DIGEST_LENGTH];
				uint64_t digest64[SHA256_DIGEST_LENGTH/8];
			} u;

			SHA256_Init(&hash_ctx);
			SHA256_Update(&hash_ctx, bdata, chain->bytes);
			SHA256_Final(u.digest, &hash_ctx);
			u.digest64[2] ^= u.digest64[3];
			if (bcmp(u.digest,
				 chain->bref.check.sha192.data,
			         sizeof(chain->bref.check.sha192.data)) == 0) {
				r = 1;
			} else {
				r = 0;
				kprintf("chain %016jx.%02x meth=%02x "
					"CHECK FAIL\n",
					chain->bref.data_off,
					chain->bref.type,
					chain->bref.methods);
			}
		}
		break;
	case HAMMER2_CHECK_FREEMAP:
		r = (chain->bref.check.freemap.icrc32 ==
		     hammer2_icrc32(bdata, chain->bytes));
		if (r == 0) {
			kprintf("chain %016jx.%02x meth=%02x "
				"CHECK FAIL\n",
				chain->bref.data_off,
				chain->bref.type,
				chain->bref.methods);
			kprintf("freemap.icrc %08x icrc32 %08x (%d)\n",
				chain->bref.check.freemap.icrc32,
				hammer2_icrc32(bdata, chain->bytes),
					       chain->bytes);
			if (chain->dio)
				kprintf("dio %p buf %016jx,%d bdata %p/%p\n",
					chain->dio, chain->dio->bp->b_loffset,
					chain->dio->bp->b_bufsize, bdata,
					chain->dio->bp->b_data);
		}

		break;
	default:
		kprintf("hammer2_chain_setcheck: unknown check type %02x\n",
			chain->bref.methods);
		r = 1;
		break;
	}
	return r;
}

/*
 * Acquire the chain and parent representing the specified inode for the
 * device at the specified cluster index.
 *
 * The flags passed in are LOOKUP flags, not RESOLVE flags.
 *
 * If we are unable to locate the hardlink, INVAL is returned and *chainp
 * will be NULL.  *parentp may still be set error or not, or NULL if the
 * parent itself could not be resolved.
 *
 * Caller must pass-in a valid (and locked), or NULL *parentp or *chainp.
 * This function replaces *parentp and *chainp.  Generally speaking, if
 * the caller found a directory entry and wants the inode, the caller should
 * pass the parent,chain representing the directory entry so this function
 * can dispose of it properly to avoid any possible lock order reversals.
 */
int
hammer2_chain_inode_find(hammer2_pfs_t *pmp, hammer2_key_t inum,
			 int clindex, int flags,
			 hammer2_chain_t **parentp, hammer2_chain_t **chainp)
{
	hammer2_chain_t *parent;
	hammer2_chain_t *rchain;
	hammer2_key_t key_dummy;
	int resolve_flags;
	int error;

	resolve_flags = (flags & HAMMER2_LOOKUP_SHARED) ?
			HAMMER2_RESOLVE_SHARED : 0;

	/*
	 * Caller expects us to replace these.
	 */
	if (*chainp) {
		hammer2_chain_unlock(*chainp);
		hammer2_chain_drop(*chainp);
		*chainp = NULL;
	}
	if (*parentp) {
		hammer2_chain_unlock(*parentp);
		hammer2_chain_drop(*parentp);
		*parentp = NULL;
	}

	/*
	 * Inodes hang off of the iroot (bit 63 is clear, differentiating
	 * inodes from root directory entries in the key lookup).
	 */
	parent = hammer2_inode_chain(pmp->iroot, clindex, resolve_flags);
	rchain = NULL;
	if (parent) {
		rchain = hammer2_chain_lookup(&parent, &key_dummy,
					      inum, inum,
					      &error, flags);
	} else {
		error = HAMMER2_ERROR_EIO;
	}
	*parentp = parent;
	*chainp = rchain;

	return error;
}

/*
 * Used by the bulkscan code to snapshot the synchronized storage for
 * a volume, allowing it to be scanned concurrently against normal
 * operation.
 */
hammer2_chain_t *
hammer2_chain_bulksnap(hammer2_dev_t *hmp)
{
	hammer2_chain_t *copy;

	copy = hammer2_chain_alloc(hmp, hmp->spmp, &hmp->vchain.bref);
	copy->data = kmalloc(sizeof(copy->data->voldata),
			     hmp->mchain,
			     M_WAITOK | M_ZERO);
	hammer2_voldata_lock(hmp);
	copy->data->voldata = hmp->volsync;
	hammer2_voldata_unlock(hmp);

	return copy;
}

void
hammer2_chain_bulkdrop(hammer2_chain_t *copy)
{
	KKASSERT(copy->bref.type == HAMMER2_BREF_TYPE_VOLUME);
	KKASSERT(copy->data);
	kfree(copy->data, copy->hmp->mchain);
	copy->data = NULL;
	atomic_add_long(&hammer2_chain_allocs, -1);
	hammer2_chain_drop(copy);
}

/*
 * Returns non-zero if the chain (INODE or DIRENT) matches the
 * filename.
 */
int
hammer2_chain_dirent_test(hammer2_chain_t *chain, const char *name,
			  size_t name_len)
{
	const hammer2_inode_data_t *ripdata;
	const hammer2_dirent_head_t *den;

	if (chain->bref.type == HAMMER2_BREF_TYPE_INODE) {
		ripdata = &chain->data->ipdata;
		if (ripdata->meta.name_len == name_len &&
		    bcmp(ripdata->filename, name, name_len) == 0) {
			return 1;
		}
	}
	if (chain->bref.type == HAMMER2_BREF_TYPE_DIRENT &&
	   chain->bref.embed.dirent.namlen == name_len) {
		den = &chain->bref.embed.dirent;
		if (name_len > sizeof(chain->bref.check.buf) &&
		    bcmp(chain->data->buf, name, name_len) == 0) {
			return 1;
		}
		if (name_len <= sizeof(chain->bref.check.buf) &&
		    bcmp(chain->bref.check.buf, name, name_len) == 0) {
			return 1;
		}
	}
	return 0;
}