xref: /dragonfly/sys/vfs/hammer2/hammer2_xops.c (revision dcb5d66b)

/*
 * 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>
 * by Venkatesh Srinivas <vsrinivas@dragonflybsd.org>
 * by Daniel Flores (GSOC 2013 - mentored by Matthew Dillon, compression)
 *
 * 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.
 */
/*
 * Per-node backend for kernel filesystem interface.
 *
 * This executes a VOP concurrently on multiple nodes, each node via its own
 * thread, and competes to advance the original request.  The original
 * request is retired the moment all requirements are met, even if the
 * operation is still in-progress on some nodes.
 */
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/fcntl.h>
#include <sys/buf.h>
#include <sys/proc.h>
#include <sys/namei.h>
#include <sys/mount.h>
#include <sys/vnode.h>
#include <sys/mountctl.h>
#include <sys/dirent.h>
#include <sys/uio.h>
#include <sys/objcache.h>
#include <sys/event.h>
#include <sys/file.h>
#include <vfs/fifofs/fifo.h>

#include "hammer2.h"

/*
 * Determine if the specified directory is empty.
 *
 *	Returns 0 on success.
 *
 *	Returns HAMMER_ERROR_EAGAIN if caller must re-lookup the entry and
 *	retry. (occurs if we race a ripup on oparent or ochain).
 *
 *	Or returns a permanent HAMMER2_ERROR_* error mask.
 *
 * The caller must pass in an exclusively locked oparent and ochain.  This
 * function will handle the case where the chain is a directory entry or
 * the inode itself.  The original oparent,ochain will be locked upon return.
 *
 * This function will unlock the underlying oparent,ochain temporarily when
 * doing an inode lookup to avoid deadlocks.  The caller MUST handle the EAGAIN
 * result as this means that oparent is no longer the parent of ochain, or
 * that ochain was destroyed while it was unlocked.
 */
static
int
checkdirempty(hammer2_chain_t *oparent, hammer2_chain_t *ochain, int clindex)
{
	hammer2_chain_t *parent;
	hammer2_chain_t *chain;
	hammer2_key_t key_next;
	hammer2_key_t inum;
	int error;
	int didunlock;

	error = 0;
	didunlock = 0;

	/*
	 * Find the inode, set it up as a locked 'chain'.  ochain can be the
	 * inode itself, or it can be a directory entry.
	 */
	if (ochain->bref.type == HAMMER2_BREF_TYPE_DIRENT) {
		inum = ochain->bref.embed.dirent.inum;
		hammer2_chain_unlock(ochain);
		hammer2_chain_unlock(oparent);

		parent = NULL;
		chain = NULL;
		error = hammer2_chain_inode_find(ochain->pmp, inum,
						 clindex, 0,
						 &parent, &chain);
		if (parent) {
			hammer2_chain_unlock(parent);
			hammer2_chain_drop(parent);
		}
		didunlock = 1;
	} else {
		/*
		 * The directory entry *is* the directory inode
		 */
		chain = hammer2_chain_lookup_init(ochain, 0);
	}

	/*
	 * Determine if the directory is empty or not by checking its
	 * visible namespace (the area which contains directory entries).
	 */
	if (error == 0) {
		parent = chain;
		chain = NULL;
		if (parent) {
			chain = hammer2_chain_lookup(&parent, &key_next,
						     HAMMER2_DIRHASH_VISIBLE,
						     HAMMER2_KEY_MAX,
						     &error, 0);
		}
		if (chain) {
			error = HAMMER2_ERROR_ENOTEMPTY;
			hammer2_chain_unlock(chain);
			hammer2_chain_drop(chain);
		}
		hammer2_chain_lookup_done(parent);
	}

	if (didunlock) {
		hammer2_chain_lock(oparent, HAMMER2_RESOLVE_ALWAYS);
		hammer2_chain_lock(ochain, HAMMER2_RESOLVE_ALWAYS);
		if ((ochain->flags & HAMMER2_CHAIN_DELETED) ||
		    (oparent->flags & HAMMER2_CHAIN_DELETED) ||
		    ochain->parent != oparent) {
			kprintf("hammer2: debug: CHECKDIR inum %jd RETRY\n",
				inum);
			error = HAMMER2_ERROR_EAGAIN;
		}
	}
	return error;
}

/*
 * Backend for hammer2_vfs_root()
 *
 * This is called when a newly mounted PFS has not yet synchronized
 * to the inode_tid and modify_tid.
 */
void
hammer2_xop_ipcluster(hammer2_thread_t *thr, hammer2_xop_t *arg)
{
	hammer2_xop_ipcluster_t *xop = &arg->xop_ipcluster;
	hammer2_chain_t *chain;
	int error;

	chain = hammer2_inode_chain(xop->head.ip1, thr->clindex,
				    HAMMER2_RESOLVE_ALWAYS |
				    HAMMER2_RESOLVE_SHARED);
	if (chain)
		error = chain->error;
	else
		error = HAMMER2_ERROR_EIO;

	hammer2_xop_feed(&xop->head, chain, thr->clindex, error);
	if (chain) {
		hammer2_chain_unlock(chain);
		hammer2_chain_drop(chain);
	}
}

/*
 * Backend for hammer2_vop_readdir()
 */
void
hammer2_xop_readdir(hammer2_thread_t *thr, hammer2_xop_t *arg)
{
	hammer2_xop_readdir_t *xop = &arg->xop_readdir;
	hammer2_chain_t *parent;
	hammer2_chain_t *chain;
	hammer2_key_t key_next;
	hammer2_key_t lkey;
	int error = 0;

	lkey = xop->lkey;
	if (hammer2_debug & 0x0020)
		kprintf("xop_readdir %p lkey=%016jx\n", xop, lkey);

	/*
	 * The inode's chain is the iterator.  If we cannot acquire it our
	 * contribution ends here.
	 */
	parent = hammer2_inode_chain(xop->head.ip1, thr->clindex,
				     HAMMER2_RESOLVE_ALWAYS |
				     HAMMER2_RESOLVE_SHARED);
	if (parent == NULL) {
		kprintf("xop_readdir: NULL parent\n");
		goto done;
	}

	/*
	 * Directory scan [re]start and loop, the feed inherits the chain's
	 * lock so do not unlock it on the iteration.
	 */
	chain = hammer2_chain_lookup(&parent, &key_next, lkey, lkey,
				     &error, HAMMER2_LOOKUP_SHARED);
	if (chain == NULL) {
		chain = hammer2_chain_lookup(&parent, &key_next,
					     lkey, HAMMER2_KEY_MAX,
					     &error, HAMMER2_LOOKUP_SHARED);
	}
	while (chain) {
		error = hammer2_xop_feed(&xop->head, chain, thr->clindex, 0);
		if (error)
			goto break2;
		chain = hammer2_chain_next(&parent, chain, &key_next,
					   key_next, HAMMER2_KEY_MAX,
					   &error, HAMMER2_LOOKUP_SHARED);
	}
break2:
	if (chain) {
		hammer2_chain_unlock(chain);
		hammer2_chain_drop(chain);
	}
	hammer2_chain_unlock(parent);
	hammer2_chain_drop(parent);
done:
	hammer2_xop_feed(&xop->head, NULL, thr->clindex, error);
}

/*
 * Backend for hammer2_vop_nresolve()
 */
void
hammer2_xop_nresolve(hammer2_thread_t *thr, hammer2_xop_t *arg)
{
	hammer2_xop_nresolve_t *xop = &arg->xop_nresolve;
	hammer2_chain_t *parent;
	hammer2_chain_t *chain;
	const char *name;
	size_t name_len;
	hammer2_key_t key_next;
	hammer2_key_t lhc;
	int error;

	parent = hammer2_inode_chain(xop->head.ip1, thr->clindex,
				     HAMMER2_RESOLVE_ALWAYS |
				     HAMMER2_RESOLVE_SHARED);
	if (parent == NULL) {
		kprintf("xop_nresolve: NULL parent\n");
		chain = NULL;
		error = HAMMER2_ERROR_EIO;
		goto done;
	}
	name = xop->head.name1;
	name_len = xop->head.name1_len;

	/*
	 * Lookup the directory entry
	 */
	lhc = hammer2_dirhash(name, name_len);
	chain = hammer2_chain_lookup(&parent, &key_next,
				     lhc, lhc + HAMMER2_DIRHASH_LOMASK,
				     &error,
				     HAMMER2_LOOKUP_ALWAYS |
				     HAMMER2_LOOKUP_SHARED);
	while (chain) {
		if (hammer2_chain_dirent_test(chain, name, name_len))
			break;
		chain = hammer2_chain_next(&parent, chain, &key_next,
					   key_next,
					   lhc + HAMMER2_DIRHASH_LOMASK,
					   &error,
					   HAMMER2_LOOKUP_ALWAYS |
					   HAMMER2_LOOKUP_SHARED);
	}

	/*
	 * If the entry is a hardlink pointer, resolve it.
	 */
	if (chain && chain->error == 0) {
		if (chain->bref.type == HAMMER2_BREF_TYPE_DIRENT) {
			lhc = chain->bref.embed.dirent.inum;
			error = hammer2_chain_inode_find(chain->pmp,
							 lhc,
							 thr->clindex,
							 HAMMER2_LOOKUP_SHARED,
							 &parent,
							 &chain);
		}
	} else if (chain && error == 0) {
		error = chain->error;
	}
done:
	error = hammer2_xop_feed(&xop->head, chain, thr->clindex, error);
	if (chain) {
		hammer2_chain_unlock(chain);
		hammer2_chain_drop(chain);
	}
	if (parent) {
		hammer2_chain_unlock(parent);
		hammer2_chain_drop(parent);
	}
}

/*
 * Backend for hammer2_vop_nremove(), hammer2_vop_nrmdir(), and
 * backend for pfs_delete.
 *
 * This function locates and removes a directory entry, and will lookup
 * and return the underlying inode.  For directory entries the underlying
 * inode is not removed.  If the directory entry is the actual inode itself,
 * it may be conditonally removed and returned.
 *
 * WARNING!  Any target inode's nlinks may not be synchronized to the
 *	     in-memory inode.  The frontend's hammer2_inode_unlink_finisher()
 *	     is responsible for the final disposition of the actual inode.
 */
void
hammer2_xop_unlink(hammer2_thread_t *thr, hammer2_xop_t *arg)
{
	hammer2_xop_unlink_t *xop = &arg->xop_unlink;
	hammer2_chain_t *parent;
	hammer2_chain_t *chain;
	const char *name;
	size_t name_len;
	hammer2_key_t key_next;
	hammer2_key_t lhc;
	int error;

again:
	/*
	 * Requires exclusive lock
	 */
	parent = hammer2_inode_chain(xop->head.ip1, thr->clindex,
				     HAMMER2_RESOLVE_ALWAYS);
	chain = NULL;
	if (parent == NULL) {
		kprintf("xop_nresolve: NULL parent\n");
		error = HAMMER2_ERROR_EIO;
		goto done;
	}
	name = xop->head.name1;
	name_len = xop->head.name1_len;

	/*
	 * Lookup the directory entry
	 */
	lhc = hammer2_dirhash(name, name_len);
	chain = hammer2_chain_lookup(&parent, &key_next,
				     lhc, lhc + HAMMER2_DIRHASH_LOMASK,
				     &error, HAMMER2_LOOKUP_ALWAYS);
	while (chain) {
		if (hammer2_chain_dirent_test(chain, name, name_len))
			break;
		chain = hammer2_chain_next(&parent, chain, &key_next,
					   key_next,
					   lhc + HAMMER2_DIRHASH_LOMASK,
					   &error, HAMMER2_LOOKUP_ALWAYS);
	}

	/*
	 * The directory entry will either be a BREF_TYPE_DIRENT or a
	 * BREF_TYPE_INODE.  We always permanently delete DIRENTs, but
	 * must go by xop->dopermanent for BREF_TYPE_INODE.
	 *
	 * Note that the target chain's nlinks may not be synchronized with
	 * the in-memory hammer2_inode_t structure, so we don't try to do
	 * anything fancy here.  The frontend deals with nlinks
	 * synchronization.
	 */
	if (chain && chain->error == 0) {
		int dopermanent = xop->dopermanent & H2DOPERM_PERMANENT;
		int doforce = xop->dopermanent & H2DOPERM_FORCE;
		uint8_t type;

		/*
		 * If the directory entry is the actual inode then use its
		 * type for the directory typing tests, otherwise if it is
		 * a directory entry, pull the type field from the entry.
		 *
		 * Directory entries are always permanently deleted
		 * (because they aren't the actual inode).
		 */
		if (chain->bref.type == HAMMER2_BREF_TYPE_DIRENT) {
			type = chain->bref.embed.dirent.type;
			dopermanent |= HAMMER2_DELETE_PERMANENT;
		} else {
			type = chain->data->ipdata.meta.type;
		}

		/*
		 * Check directory typing and delete the entry.  Note that
		 * nlinks adjustments are made on the real inode by the
		 * frontend, not here.
		 *
		 * Unfortunately, checkdirempty() may have to unlock (parent).
		 * If it no longer matches chain->parent after re-locking,
		 * EAGAIN is returned.
		 */
		if (type == HAMMER2_OBJTYPE_DIRECTORY && doforce) {
			/*
			 * If doforce then execute the operation even if
			 * the directory is not empty or errored.  We
			 * ignore chain->error here, allowing an errored
			 * chain (aka directory entry) to still be deleted.
			 */
			error = hammer2_chain_delete(parent, chain,
					     xop->head.mtid, dopermanent);
		} else if (type == HAMMER2_OBJTYPE_DIRECTORY &&
			   (error = checkdirempty(parent, chain, thr->clindex)) != 0) {
			/*
			 * error may be EAGAIN or ENOTEMPTY
			 */
			if (error == HAMMER2_ERROR_EAGAIN) {
				hammer2_chain_unlock(chain);
				hammer2_chain_drop(chain);
				hammer2_chain_unlock(parent);
				hammer2_chain_drop(parent);
				goto again;
			}
		} else if (type == HAMMER2_OBJTYPE_DIRECTORY &&
		    xop->isdir == 0) {
			error = HAMMER2_ERROR_ENOTDIR;
		} else if (type != HAMMER2_OBJTYPE_DIRECTORY &&
			   xop->isdir >= 1) {
			error = HAMMER2_ERROR_EISDIR;
		} else {
			/*
			 * Delete the directory entry.  chain might also
			 * be a directly-embedded inode.
			 *
			 * Allow the deletion to proceed even if the chain
			 * is errored.  Give priority to error-on-delete over
			 * chain->error.
			 */
			error = hammer2_chain_delete(parent, chain,
						     xop->head.mtid,
						     dopermanent);
			if (error == 0)
				error = chain->error;
		}
	} else {
		if (chain && error == 0)
			error = chain->error;
	}

	/*
	 * If chain is a directory entry we must resolve it.  We do not try
	 * to manipulate the contents as it might not be synchronized with
	 * the frontend hammer2_inode_t, nor do we try to lookup the
	 * frontend hammer2_inode_t here (we are the backend!).
	 */
	if (chain && chain->bref.type == HAMMER2_BREF_TYPE_DIRENT &&
	    (xop->dopermanent & H2DOPERM_IGNINO) == 0) {
		int error2;

		lhc = chain->bref.embed.dirent.inum;

		error2 = hammer2_chain_inode_find(chain->pmp, lhc,
						  thr->clindex, 0,
						  &parent, &chain);
		if (error2) {
			kprintf("inode_find: %016jx %p failed\n",
				lhc, chain);
			error2 = 0;	/* silently ignore */
		}
		if (error == 0)
			error = error2;
	}

	/*
	 * Return the inode target for further action.  Typically used by
	 * hammer2_inode_unlink_finisher().
	 */
done:
	hammer2_xop_feed(&xop->head, chain, thr->clindex, error);
	if (chain) {
		hammer2_chain_unlock(chain);
		hammer2_chain_drop(chain);
		chain = NULL;
	}
	if (parent) {
		hammer2_chain_unlock(parent);
		hammer2_chain_drop(parent);
		parent = NULL;
	}
}

/*
 * Backend for hammer2_vop_nrename()
 *
 * This handles the backend rename operation.  Typically this renames
 * directory entries but can also be used to rename embedded inodes.
 *
 * NOTE! The frontend is responsible for updating the inode meta-data in
 *	 the file being renamed and for decrementing the target-replaced
 *	 inode's nlinks, if present.
 */
void
hammer2_xop_nrename(hammer2_thread_t *thr, hammer2_xop_t *arg)
{
	hammer2_xop_nrename_t *xop = &arg->xop_nrename;
	hammer2_pfs_t *pmp;
	hammer2_chain_t *parent;
	hammer2_chain_t *chain;
	hammer2_chain_t *tmp;
	hammer2_inode_t *ip;
	hammer2_key_t key_next;
	int error;

	/*
	 * We need the precise parent chain to issue the deletion.
	 *
	 * If this is a directory entry we must locate the underlying
	 * inode.  If it is an embedded inode we can act directly on it.
	 */
	ip = xop->head.ip2;
	pmp = ip->pmp;
	chain = NULL;
	error = 0;

	if (xop->ip_key & HAMMER2_DIRHASH_VISIBLE) {
		/*
		 * Find ip's direct parent chain.
		 */
		chain = hammer2_inode_chain(ip, thr->clindex,
					    HAMMER2_RESOLVE_ALWAYS);
		if (chain == NULL) {
			error = HAMMER2_ERROR_EIO;
			parent = NULL;
			goto done;
		}
		parent = hammer2_chain_getparent(chain, HAMMER2_RESOLVE_ALWAYS);
		if (parent == NULL) {
			error = HAMMER2_ERROR_EIO;
			goto done;
		}
	} else {
		/*
		 * The directory entry for the head.ip1 inode
		 * is in fdip, do a namespace search.
		 */
		hammer2_key_t lhc;
		const char *name;
		size_t name_len;

		parent = hammer2_inode_chain(xop->head.ip1, thr->clindex,
					     HAMMER2_RESOLVE_ALWAYS);
		if (parent == NULL) {
			kprintf("xop_nrename: NULL parent\n");
			error = HAMMER2_ERROR_EIO;
			goto done;
		}
		name = xop->head.name1;
		name_len = xop->head.name1_len;

		/*
		 * Lookup the directory entry
		 */
		lhc = hammer2_dirhash(name, name_len);
		chain = hammer2_chain_lookup(&parent, &key_next,
					     lhc, lhc + HAMMER2_DIRHASH_LOMASK,
					     &error, HAMMER2_LOOKUP_ALWAYS);
		while (chain) {
			if (hammer2_chain_dirent_test(chain, name, name_len))
				break;
			chain = hammer2_chain_next(&parent, chain, &key_next,
						   key_next,
						   lhc + HAMMER2_DIRHASH_LOMASK,
						   &error,
						   HAMMER2_LOOKUP_ALWAYS);
		}
	}

	if (chain == NULL) {
		/* XXX shouldn't happen, but does under fsstress */
		kprintf("hammer2_xop_rename: \"%s\" -> \"%s\"  ENOENT\n",
			xop->head.name1,
			xop->head.name2);
		if (error == 0)
			error = HAMMER2_ERROR_ENOENT;
		goto done;
	}

	if (chain->error) {
		error = chain->error;
		goto done;
	}

	/*
	 * Delete it, then create it in the new namespace.
	 *
	 * An error can occur if the chain being deleted requires
	 * modification and the media is full.
	 */
	error = hammer2_chain_delete(parent, chain, xop->head.mtid, 0);
	hammer2_chain_unlock(parent);
	hammer2_chain_drop(parent);
	parent = NULL;		/* safety */
	if (error)
		goto done;

	/*
	 * Adjust fields in the deleted chain appropriate for the rename
	 * operation.
	 *
	 * NOTE! For embedded inodes, the frontend will officially replicate
	 *	 the field adjustments, but we also do it here to maintain
	 *	 consistency in case of a crash.
	 */
	if (chain->bref.key != xop->lhc ||
	    xop->head.name1_len != xop->head.name2_len ||
	    bcmp(xop->head.name1, xop->head.name2, xop->head.name1_len) != 0) {
		if (chain->bref.type == HAMMER2_BREF_TYPE_INODE) {
			hammer2_inode_data_t *wipdata;

			error = hammer2_chain_modify(chain, xop->head.mtid,
						     0, 0);
			if (error == 0) {
				wipdata = &chain->data->ipdata;

				bzero(wipdata->filename,
				      sizeof(wipdata->filename));
				bcopy(xop->head.name2,
				      wipdata->filename,
				      xop->head.name2_len);
				wipdata->meta.name_key = xop->lhc;
				wipdata->meta.name_len = xop->head.name2_len;
			}
		}
		if (chain->bref.type == HAMMER2_BREF_TYPE_DIRENT) {
			if (xop->head.name2_len <=
			    sizeof(chain->bref.check.buf)) {
				/*
				 * Remove any related data buffer, we can
				 * embed the filename in the bref itself.
				 */
				error = hammer2_chain_resize(
						chain, xop->head.mtid, 0, 0, 0);
				if (error == 0) {
					error = hammer2_chain_modify(
							chain, xop->head.mtid,
							0, 0);
				}
				if (error == 0) {
					bzero(chain->bref.check.buf,
					      sizeof(chain->bref.check.buf));
					bcopy(xop->head.name2,
					      chain->bref.check.buf,
					      xop->head.name2_len);
				}
			} else {
				/*
				 * Associate a data buffer with the bref.
				 * Zero it for consistency.  Note that the
				 * data buffer is not 64KB so use chain->bytes
				 * instead of sizeof().
				 */
				error = hammer2_chain_resize(
					chain, xop->head.mtid, 0,
					hammer2_getradix(HAMMER2_ALLOC_MIN), 0);
				if (error == 0) {
					error = hammer2_chain_modify(
						    chain, xop->head.mtid,
						    0, 0);
				}
				if (error == 0) {
					bzero(chain->data->buf, chain->bytes);
					bcopy(xop->head.name2,
					      chain->data->buf,
					      xop->head.name2_len);
				}
			}
			if (error == 0) {
				chain->bref.embed.dirent.namlen =
					xop->head.name2_len;
			}
		}
	}

	/*
	 * The frontend will replicate this operation and is the real final
	 * authority, but adjust the inode's iparent field too if the inode
	 * is embedded in the directory.
	 */
	if (chain->bref.type == HAMMER2_BREF_TYPE_INODE &&
	    chain->data->ipdata.meta.iparent != xop->head.ip3->meta.inum) {
		hammer2_inode_data_t *wipdata;

		error = hammer2_chain_modify(chain, xop->head.mtid, 0, 0);
		if (error == 0) {
			wipdata = &chain->data->ipdata;
			wipdata->meta.iparent = xop->head.ip3->meta.inum;
		}
	}

	/*
	 * Destroy any matching target(s) before creating the new entry.
	 * This will result in some ping-ponging of the directory key
	 * iterator but that is ok.
	 */
	parent = hammer2_inode_chain(xop->head.ip3, thr->clindex,
				     HAMMER2_RESOLVE_ALWAYS);
	if (parent == NULL) {
		error = HAMMER2_ERROR_EIO;
		goto done;
	}

	/*
	 * Delete all matching directory entries.  That is, get rid of
	 * multiple duplicates if present, as a self-healing mechanism.
	 */
	if (error == 0) {
		tmp = hammer2_chain_lookup(&parent, &key_next,
					   xop->lhc & ~HAMMER2_DIRHASH_LOMASK,
					   xop->lhc | HAMMER2_DIRHASH_LOMASK,
					   &error,
					   HAMMER2_LOOKUP_ALWAYS);
		while (tmp) {
			int e2;
			if (hammer2_chain_dirent_test(tmp, xop->head.name2,
						      xop->head.name2_len)) {
				e2 = hammer2_chain_delete(parent, tmp,
							  xop->head.mtid, 0);
				if (error == 0 && e2)
					error = e2;
			}
			tmp = hammer2_chain_next(&parent, tmp, &key_next,
						 key_next,
						 xop->lhc |
						  HAMMER2_DIRHASH_LOMASK,
						 &error,
						 HAMMER2_LOOKUP_ALWAYS);
		}
	}
	if (error == 0) {
		/*
		 * A relookup is required before the create to properly
		 * position the parent chain.
		 */
		tmp = hammer2_chain_lookup(&parent, &key_next,
					   xop->lhc, xop->lhc,
					   &error, 0);
		KKASSERT(tmp == NULL);
		error = hammer2_chain_create(&parent, &chain,
					     pmp, HAMMER2_METH_DEFAULT,
					     xop->lhc, 0,
					     HAMMER2_BREF_TYPE_INODE,
					     HAMMER2_INODE_BYTES,
					     xop->head.mtid, 0, 0);
	}
done:
	hammer2_xop_feed(&xop->head, NULL, thr->clindex, error);
	if (parent) {
		hammer2_chain_unlock(parent);
		hammer2_chain_drop(parent);
	}
	if (chain) {
		hammer2_chain_unlock(chain);
		hammer2_chain_drop(chain);
	}
}

/*
 * Directory collision resolver scan helper (backend, threaded).
 *
 * Used by the inode create code to locate an unused lhc.
 */
void
hammer2_xop_scanlhc(hammer2_thread_t *thr, hammer2_xop_t *arg)
{
	hammer2_xop_scanlhc_t *xop = &arg->xop_scanlhc;
	hammer2_chain_t *parent;
	hammer2_chain_t *chain;
	hammer2_key_t key_next;
	int error = 0;

	parent = hammer2_inode_chain(xop->head.ip1, thr->clindex,
				     HAMMER2_RESOLVE_ALWAYS |
				     HAMMER2_RESOLVE_SHARED);
	if (parent == NULL) {
		kprintf("xop_nresolve: NULL parent\n");
		chain = NULL;
		error = HAMMER2_ERROR_EIO;
		goto done;
	}

	/*
	 * Lookup all possibly conflicting directory entries, the feed
	 * inherits the chain's lock so do not unlock it on the iteration.
	 */
	chain = hammer2_chain_lookup(&parent, &key_next,
				     xop->lhc,
				     xop->lhc + HAMMER2_DIRHASH_LOMASK,
				     &error,
				     HAMMER2_LOOKUP_ALWAYS |
				     HAMMER2_LOOKUP_SHARED);
	while (chain) {
		error = hammer2_xop_feed(&xop->head, chain, thr->clindex, 0);
		if (error) {
			hammer2_chain_unlock(chain);
			hammer2_chain_drop(chain);
			chain = NULL;	/* safety */
			goto done;
		}
		chain = hammer2_chain_next(&parent, chain, &key_next,
					   key_next,
					   xop->lhc + HAMMER2_DIRHASH_LOMASK,
					   &error,
					   HAMMER2_LOOKUP_ALWAYS |
					   HAMMER2_LOOKUP_SHARED);
	}
done:
	hammer2_xop_feed(&xop->head, NULL, thr->clindex, error);
	if (parent) {
		hammer2_chain_unlock(parent);
		hammer2_chain_drop(parent);
	}
}

/*
 * Generic lookup of a specific key.
 *
 * Used by the inode hidden directory code to find the hidden directory.
 */
void
hammer2_xop_lookup(hammer2_thread_t *thr, hammer2_xop_t *arg)
{
	hammer2_xop_scanlhc_t *xop = &arg->xop_scanlhc;
	hammer2_chain_t *parent;
	hammer2_chain_t *chain;
	hammer2_key_t key_next;
	int error = 0;

	parent = hammer2_inode_chain(xop->head.ip1, thr->clindex,
				     HAMMER2_RESOLVE_ALWAYS |
				     HAMMER2_RESOLVE_SHARED);
	chain = NULL;
	if (parent == NULL) {
		error = HAMMER2_ERROR_EIO;
		goto done;
	}

	/*
	 * Lookup all possibly conflicting directory entries, the feed
	 * inherits the chain's lock so do not unlock it on the iteration.
	 */
	chain = hammer2_chain_lookup(&parent, &key_next,
				     xop->lhc, xop->lhc,
				     &error,
				     HAMMER2_LOOKUP_ALWAYS |
				     HAMMER2_LOOKUP_SHARED);
	if (error == 0) {
		if (chain)
			error = chain->error;
		else
			error = HAMMER2_ERROR_ENOENT;
	}
	hammer2_xop_feed(&xop->head, chain, thr->clindex, error);

done:
	if (chain) {
		hammer2_chain_unlock(chain);
		hammer2_chain_drop(chain);
	}
	if (parent) {
		hammer2_chain_unlock(parent);
		hammer2_chain_drop(parent);
	}
}

/*
 * Generic scan
 *
 * WARNING! Fed chains must be locked shared so ownership can be transfered
 *	    and to prevent frontend/backend stalls that would occur with an
 *	    exclusive lock.  The shared lock also allows chain->data to be
 *	    retained.
 */
void
hammer2_xop_scanall(hammer2_thread_t *thr, hammer2_xop_t *arg)
{
	hammer2_xop_scanall_t *xop = &arg->xop_scanall;
	hammer2_chain_t *parent;
	hammer2_chain_t *chain;
	hammer2_key_t key_next;
	int error = 0;

	/*
	 * Assert required flags.
	 */
	KKASSERT(xop->resolve_flags & HAMMER2_RESOLVE_SHARED);
	KKASSERT(xop->lookup_flags & HAMMER2_LOOKUP_SHARED);

	/*
	 * The inode's chain is the iterator.  If we cannot acquire it our
	 * contribution ends here.
	 */
	parent = hammer2_inode_chain(xop->head.ip1, thr->clindex,
				     xop->resolve_flags);
	if (parent == NULL) {
		kprintf("xop_readdir: NULL parent\n");
		goto done;
	}

	/*
	 * Generic scan of exact records.  Note that indirect blocks are
	 * automatically recursed and will not be returned.
	 */
	chain = hammer2_chain_lookup(&parent, &key_next,
				     xop->key_beg, xop->key_end,
				     &error, xop->lookup_flags);
	while (chain) {
		error = hammer2_xop_feed(&xop->head, chain, thr->clindex, 0);
		if (error)
			goto break2;
		chain = hammer2_chain_next(&parent, chain, &key_next,
					   key_next, xop->key_end,
					   &error, xop->lookup_flags);
	}
break2:
	if (chain) {
		hammer2_chain_unlock(chain);
		hammer2_chain_drop(chain);
	}
	hammer2_chain_unlock(parent);
	hammer2_chain_drop(parent);
done:
	hammer2_xop_feed(&xop->head, NULL, thr->clindex, error);
}