UBIFS: add new flash file system

This is a new flash file system. See
http://www.linux-mtd.infradead.org/doc/ubifs.html

Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
Signed-off-by: Adrian Hunter <ext-adrian.hunter@nokia.com>

1 files changed, 2956 insertions, 0 deletions

diff --git a/fs/ubifs/tnc.c b/fs/ubifs/tnc.c
new file mode 100644
index 00000000000..e909f4a9644
--- /dev/null
+++ b/fs/ubifs/tnc.c
@@ -0,0 +1,2956 @@
+/*
+ * This file is part of UBIFS.
+ *
+ * Copyright (C) 2006-2008 Nokia Corporation.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 as published by
+ * the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, write to the Free Software Foundation, Inc., 51
+ * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
+ *
+ * Authors: Adrian Hunter
+ *          Artem Bityutskiy (Битюцкий Артём)
+ */
+
+/*
+ * This file implements TNC (Tree Node Cache) which caches indexing nodes of
+ * the UBIFS B-tree.
+ *
+ * At the moment the locking rules of the TNC tree are quite simple and
+ * straightforward. We just have a mutex and lock it when we traverse the
+ * tree. If a znode is not in memory, we read it from flash while still having
+ * the mutex locked.
+ */
+
+#include <linux/crc32.h>
+#include "ubifs.h"
+
+/*
+ * Returned codes of 'matches_name()' and 'fallible_matches_name()' functions.
+ * @NAME_LESS: name corresponding to the first argument is less than second
+ * @NAME_MATCHES: names match
+ * @NAME_GREATER: name corresponding to the second argument is greater than
+ *                first
+ * @NOT_ON_MEDIA: node referred by zbranch does not exist on the media
+ *
+ * These constants were introduce to improve readability.
+ */
+enum {
+	NAME_LESS    = 0,
+	NAME_MATCHES = 1,
+	NAME_GREATER = 2,
+	NOT_ON_MEDIA = 3,
+};
+
+/**
+ * insert_old_idx - record an index node obsoleted since the last commit start.
+ * @c: UBIFS file-system description object
+ * @lnum: LEB number of obsoleted index node
+ * @offs: offset of obsoleted index node
+ *
+ * Returns %0 on success, and a negative error code on failure.
+ *
+ * For recovery, there must always be a complete intact version of the index on
+ * flash at all times. That is called the "old index". It is the index as at the
+ * time of the last successful commit. Many of the index nodes in the old index
+ * may be dirty, but they must not be erased until the next successful commit
+ * (at which point that index becomes the old index).
+ *
+ * That means that the garbage collection and the in-the-gaps method of
+ * committing must be able to determine if an index node is in the old index.
+ * Most of the old index nodes can be found by looking up the TNC using the
+ * 'lookup_znode()' function. However, some of the old index nodes may have
+ * been deleted from the current index or may have been changed so much that
+ * they cannot be easily found. In those cases, an entry is added to an RB-tree.
+ * That is what this function does. The RB-tree is ordered by LEB number and
+ * offset because they uniquely identify the old index node.
+ */
+static int insert_old_idx(struct ubifs_info *c, int lnum, int offs)
+{
+	struct ubifs_old_idx *old_idx, *o;
+	struct rb_node **p, *parent = NULL;
+
+	old_idx = kmalloc(sizeof(struct ubifs_old_idx), GFP_NOFS);
+	if (unlikely(!old_idx))
+		return -ENOMEM;
+	old_idx->lnum = lnum;
+	old_idx->offs = offs;
+
+	p = &c->old_idx.rb_node;
+	while (*p) {
+		parent = *p;
+		o = rb_entry(parent, struct ubifs_old_idx, rb);
+		if (lnum < o->lnum)
+			p = &(*p)->rb_left;
+		else if (lnum > o->lnum)
+			p = &(*p)->rb_right;
+		else if (offs < o->offs)
+			p = &(*p)->rb_left;
+		else if (offs > o->offs)
+			p = &(*p)->rb_right;
+		else {
+			ubifs_err("old idx added twice!");
+			kfree(old_idx);
+			return 0;
+		}
+	}
+	rb_link_node(&old_idx->rb, parent, p);
+	rb_insert_color(&old_idx->rb, &c->old_idx);
+	return 0;
+}
+
+/**
+ * insert_old_idx_znode - record a znode obsoleted since last commit start.
+ * @c: UBIFS file-system description object
+ * @znode: znode of obsoleted index node
+ *
+ * Returns %0 on success, and a negative error code on failure.
+ */
+int insert_old_idx_znode(struct ubifs_info *c, struct ubifs_znode *znode)
+{
+	if (znode->parent) {
+		struct ubifs_zbranch *zbr;
+
+		zbr = &znode->parent->zbranch[znode->iip];
+		if (zbr->len)
+			return insert_old_idx(c, zbr->lnum, zbr->offs);
+	} else
+		if (c->zroot.len)
+			return insert_old_idx(c, c->zroot.lnum,
+					      c->zroot.offs);
+	return 0;
+}
+
+/**
+ * ins_clr_old_idx_znode - record a znode obsoleted since last commit start.
+ * @c: UBIFS file-system description object
+ * @znode: znode of obsoleted index node
+ *
+ * Returns %0 on success, and a negative error code on failure.
+ */
+static int ins_clr_old_idx_znode(struct ubifs_info *c,
+				 struct ubifs_znode *znode)
+{
+	int err;
+
+	if (znode->parent) {
+		struct ubifs_zbranch *zbr;
+
+		zbr = &znode->parent->zbranch[znode->iip];
+		if (zbr->len) {
+			err = insert_old_idx(c, zbr->lnum, zbr->offs);
+			if (err)
+				return err;
+			zbr->lnum = 0;
+			zbr->offs = 0;
+			zbr->len = 0;
+		}
+	} else
+		if (c->zroot.len) {
+			err = insert_old_idx(c, c->zroot.lnum, c->zroot.offs);
+			if (err)
+				return err;
+			c->zroot.lnum = 0;
+			c->zroot.offs = 0;
+			c->zroot.len = 0;
+		}
+	return 0;
+}
+
+/**
+ * destroy_old_idx - destroy the old_idx RB-tree.
+ * @c: UBIFS file-system description object
+ *
+ * During start commit, the old_idx RB-tree is used to avoid overwriting index
+ * nodes that were in the index last commit but have since been deleted.  This
+ * is necessary for recovery i.e. the old index must be kept intact until the
+ * new index is successfully written.  The old-idx RB-tree is used for the
+ * in-the-gaps method of writing index nodes and is destroyed every commit.
+ */
+void destroy_old_idx(struct ubifs_info *c)
+{
+	struct rb_node *this = c->old_idx.rb_node;
+	struct ubifs_old_idx *old_idx;
+
+	while (this) {
+		if (this->rb_left) {
+			this = this->rb_left;
+			continue;
+		} else if (this->rb_right) {
+			this = this->rb_right;
+			continue;
+		}
+		old_idx = rb_entry(this, struct ubifs_old_idx, rb);
+		this = rb_parent(this);
+		if (this) {
+			if (this->rb_left == &old_idx->rb)
+				this->rb_left = NULL;
+			else
+				this->rb_right = NULL;
+		}
+		kfree(old_idx);
+	}
+	c->old_idx = RB_ROOT;
+}
+
+/**
+ * copy_znode - copy a dirty znode.
+ * @c: UBIFS file-system description object
+ * @znode: znode to copy
+ *
+ * A dirty znode being committed may not be changed, so it is copied.
+ */
+static struct ubifs_znode *copy_znode(struct ubifs_info *c,
+				      struct ubifs_znode *znode)
+{
+	struct ubifs_znode *zn;
+
+	zn = kmalloc(c->max_znode_sz, GFP_NOFS);
+	if (unlikely(!zn))
+		return ERR_PTR(-ENOMEM);
+
+	memcpy(zn, znode, c->max_znode_sz);
+	zn->cnext = NULL;
+	__set_bit(DIRTY_ZNODE, &zn->flags);
+	__clear_bit(COW_ZNODE, &zn->flags);
+
+	ubifs_assert(!test_bit(OBSOLETE_ZNODE, &znode->flags));
+	__set_bit(OBSOLETE_ZNODE, &znode->flags);
+
+	if (znode->level != 0) {
+		int i;
+		const int n = zn->child_cnt;
+
+		/* The children now have new parent */
+		for (i = 0; i < n; i++) {
+			struct ubifs_zbranch *zbr = &zn->zbranch[i];
+
+			if (zbr->znode)
+				zbr->znode->parent = zn;
+		}
+	}
+
+	atomic_long_inc(&c->dirty_zn_cnt);
+	return zn;
+}
+
+/**
+ * add_idx_dirt - add dirt due to a dirty znode.
+ * @c: UBIFS file-system description object
+ * @lnum: LEB number of index node
+ * @dirt: size of index node
+ *
+ * This function updates lprops dirty space and the new size of the index.
+ */
+static int add_idx_dirt(struct ubifs_info *c, int lnum, int dirt)
+{
+	c->calc_idx_sz -= ALIGN(dirt, 8);
+	return ubifs_add_dirt(c, lnum, dirt);
+}
+
+/**
+ * dirty_cow_znode - ensure a znode is not being committed.
+ * @c: UBIFS file-system description object
+ * @zbr: branch of znode to check
+ *
+ * Returns dirtied znode on success or negative error code on failure.
+ */
+static struct ubifs_znode *dirty_cow_znode(struct ubifs_info *c,
+					   struct ubifs_zbranch *zbr)
+{
+	struct ubifs_znode *znode = zbr->znode;
+	struct ubifs_znode *zn;
+	int err;
+
+	if (!test_bit(COW_ZNODE, &znode->flags)) {
+		/* znode is not being committed */
+		if (!test_and_set_bit(DIRTY_ZNODE, &znode->flags)) {
+			atomic_long_inc(&c->dirty_zn_cnt);
+			atomic_long_dec(&c->clean_zn_cnt);
+			atomic_long_dec(&ubifs_clean_zn_cnt);
+			err = add_idx_dirt(c, zbr->lnum, zbr->len);
+			if (unlikely(err))
+				return ERR_PTR(err);
+		}
+		return znode;
+	}
+
+	zn = copy_znode(c, znode);
+	if (unlikely(IS_ERR(zn)))
+		return zn;
+
+	if (zbr->len) {
+		err = insert_old_idx(c, zbr->lnum, zbr->offs);
+		if (unlikely(err))
+			return ERR_PTR(err);
+		err = add_idx_dirt(c, zbr->lnum, zbr->len);
+	} else
+		err = 0;
+
+	zbr->znode = zn;
+	zbr->lnum = 0;
+	zbr->offs = 0;
+	zbr->len = 0;
+
+	if (unlikely(err))
+		return ERR_PTR(err);
+	return zn;
+}
+
+/**
+ * lnc_add - add a leaf node to the leaf node cache.
+ * @c: UBIFS file-system description object
+ * @zbr: zbranch of leaf node
+ * @node: leaf node
+ *
+ * Leaf nodes are non-index nodes directory entry nodes or data nodes. The
+ * purpose of the leaf node cache is to save re-reading the same leaf node over
+ * and over again. Most things are cached by VFS, however the file system must
+ * cache directory entries for readdir and for resolving hash collisions. The
+ * present implementation of the leaf node cache is extremely simple, and
+ * allows for error returns that are not used but that may be needed if a more
+ * complex implementation is created.
+ *
+ * Note, this function does not add the @node object to LNC directly, but
+ * allocates a copy of the object and adds the copy to LNC. The reason for this
+ * is that @node has been allocated outside of the TNC subsystem and will be
+ * used with @c->tnc_mutex unlock upon return from the TNC subsystem. But LNC
+ * may be changed at any time, e.g. freed by the shrinker.
+ */
+static int lnc_add(struct ubifs_info *c, struct ubifs_zbranch *zbr,
+		   const void *node)
+{
+	int err;
+	void *lnc_node;
+	const struct ubifs_dent_node *dent = node;
+
+	ubifs_assert(!zbr->leaf);
+	ubifs_assert(zbr->len != 0);
+	ubifs_assert(is_hash_key(c, &zbr->key));
+
+	err = ubifs_validate_entry(c, dent);
+	if (err) {
+		dbg_dump_stack();
+		dbg_dump_node(c, dent);
+		return err;
+	}
+
+	lnc_node = kmalloc(zbr->len, GFP_NOFS);
+	if (!lnc_node)
+		/* We don't have to have the cache, so no error */
+		return 0;
+
+	memcpy(lnc_node, node, zbr->len);
+	zbr->leaf = lnc_node;
+	return 0;
+}
+
+ /**
+ * lnc_add_directly - add a leaf node to the leaf-node-cache.
+ * @c: UBIFS file-system description object
+ * @zbr: zbranch of leaf node
+ * @node: leaf node
+ *
+ * This function is similar to 'lnc_add()', but it does not create a copy of
+ * @node but inserts @node to TNC directly.
+ */
+static int lnc_add_directly(struct ubifs_info *c, struct ubifs_zbranch *zbr,
+			    void *node)
+{
+	int err;
+
+	ubifs_assert(!zbr->leaf);
+	ubifs_assert(zbr->len != 0);
+
+	err = ubifs_validate_entry(c, node);
+	if (err) {
+		dbg_dump_stack();
+		dbg_dump_node(c, node);
+		return err;
+	}
+
+	zbr->leaf = node;
+	return 0;
+}
+
+/**
+ * lnc_free - remove a leaf node from the leaf node cache.
+ * @zbr: zbranch of leaf node
+ * @node: leaf node
+ */
+static void lnc_free(struct ubifs_zbranch *zbr)
+{
+	if (!zbr->leaf)
+		return;
+	kfree(zbr->leaf);
+	zbr->leaf = NULL;
+}
+
+/**
+ * tnc_read_node_nm - read a "hashed" leaf node.
+ * @c: UBIFS file-system description object
+ * @zbr: key and position of the node
+ * @node: node is returned here
+ *
+ * This function reads a "hashed" node defined by @zbr from the leaf node cache
+ * (in it is there) or from the hash media, in which case the node is also
+ * added to LNC. Returns zero in case of success or a negative negative error
+ * code in case of failure.
+ */
+static int tnc_read_node_nm(struct ubifs_info *c, struct ubifs_zbranch *zbr,
+			    void *node)
+{
+	int err;
+
+	ubifs_assert(is_hash_key(c, &zbr->key));
+
+	if (zbr->leaf) {
+		/* Read from the leaf node cache */
+		ubifs_assert(zbr->len != 0);
+		memcpy(node, zbr->leaf, zbr->len);
+		return 0;
+	}
+
+	err = ubifs_tnc_read_node(c, zbr, node);
+	if (err)
+		return err;
+
+	/* Add the node to the leaf node cache */
+	err = lnc_add(c, zbr, node);
+	return err;
+}
+
+/**
+ * try_read_node - read a node if it is a node.
+ * @c: UBIFS file-system description object
+ * @buf: buffer to read to
+ * @type: node type
+ * @len: node length (not aligned)
+ * @lnum: LEB number of node to read
+ * @offs: offset of node to read
+ *
+ * This function tries to read a node of known type and length, checks it and
+ * stores it in @buf. This function returns %1 if a node is present and %0 if
+ * a node is not present. A negative error code is returned for I/O errors.
+ * This function performs that same function as ubifs_read_node except that
+ * it does not require that there is actually a node present and instead
+ * the return code indicates if a node was read.
+ */
+static int try_read_node(const struct ubifs_info *c, void *buf, int type,
+			 int len, int lnum, int offs)
+{
+	int err, node_len;
+	struct ubifs_ch *ch = buf;
+	uint32_t crc, node_crc;
+
+	dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
+
+	err = ubi_read(c->ubi, lnum, buf, offs, len);
+	if (err) {
+		ubifs_err("cannot read node type %d from LEB %d:%d, error %d",
+			  type, lnum, offs, err);
+		return err;
+	}
+
+	if (le32_to_cpu(ch->magic) != UBIFS_NODE_MAGIC)
+		return 0;
+
+	if (ch->node_type != type)
+		return 0;
+
+	node_len = le32_to_cpu(ch->len);
+	if (node_len != len)
+		return 0;
+
+	crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
+	node_crc = le32_to_cpu(ch->crc);
+	if (crc != node_crc)
+		return 0;
+
+	return 1;
+}
+
+/**
+ * fallible_read_node - try to read a leaf node.
+ * @c: UBIFS file-system description object
+ * @key:  key of node to read
+ * @zbr:  position of node
+ * @node: node returned
+ *
+ * This function tries to read a node and returns %1 if the node is read, %0
+ * if the node is not present, and a negative error code in the case of error.
+ */
+static int fallible_read_node(struct ubifs_info *c, const union ubifs_key *key,
+			      struct ubifs_zbranch *zbr, void *node)
+{
+	int ret;
+
+	dbg_tnc("LEB %d:%d, key %s", zbr->lnum, zbr->offs, DBGKEY(key));
+
+	ret = try_read_node(c, node, key_type(c, key), zbr->len, zbr->lnum,
+			    zbr->offs);
+	if (ret == 1) {
+		union ubifs_key node_key;
+		struct ubifs_dent_node *dent = node;
+
+		/* All nodes have key in the same place */
+		key_read(c, &dent->key, &node_key);
+		if (keys_cmp(c, key, &node_key) != 0)
+			ret = 0;
+	}
+	if (ret == 0)
+		dbg_mnt("dangling branch LEB %d:%d len %d, key %s",
+			zbr->lnum, zbr->offs, zbr->len, DBGKEY(key));
+	return ret;
+}
+
+/**
+ * matches_name - determine if a direntry or xattr entry matches a given name.
+ * @c: UBIFS file-system description object
+ * @zbr: zbranch of dent
+ * @nm: name to match
+ *
+ * This function checks if xentry/direntry referred by zbranch @zbr matches name
+ * @nm. Returns %NAME_MATCHES if it does, %NAME_LESS if the name referred by
+ * @zbr is less than @nm, and %NAME_GREATER if it is greater than @nm. In case
+ * of failure, a negative error code is returned.
+ */
+static int matches_name(struct ubifs_info *c, struct ubifs_zbranch *zbr,
+			const struct qstr *nm)
+{
+	struct ubifs_dent_node *dent;
+	int nlen, err;
+
+	/* If possible, match against the dent in the leaf node cache */
+	if (!zbr->leaf) {
+		dent = kmalloc(zbr->len, GFP_NOFS);
+		if (!dent)
+			return -ENOMEM;
+
+		err = ubifs_tnc_read_node(c, zbr, dent);
+		if (err)
+			goto out_free;
+
+		/* Add the node to the leaf node cache */
+		err = lnc_add_directly(c, zbr, dent);
+		if (err)
+			goto out_free;
+	} else
+		dent = zbr->leaf;
+
+	nlen = le16_to_cpu(dent->nlen);
+	err = memcmp(dent->name, nm->name, min_t(int, nlen, nm->len));
+	if (err == 0) {
+		if (nlen == nm->len)
+			return NAME_MATCHES;
+		else if (nlen < nm->len)
+			return NAME_LESS;
+		else
+			return NAME_GREATER;
+	} else if (err < 0)
+		return NAME_LESS;
+	else
+		return NAME_GREATER;
+
+out_free:
+	kfree(dent);
+	return err;
+}
+
+/**
+ * get_znode - get a TNC znode that may not be loaded yet.
+ * @c: UBIFS file-system description object
+ * @znode: parent znode
+ * @n: znode branch slot number
+ *
+ * This function returns the znode or a negative error code.
+ */
+static struct ubifs_znode *get_znode(struct ubifs_info *c,
+				     struct ubifs_znode *znode, int n)
+{
+	struct ubifs_zbranch *zbr;
+
+	zbr = &znode->zbranch[n];
+	if (zbr->znode)
+		znode = zbr->znode;
+	else
+		znode = ubifs_load_znode(c, zbr, znode, n);
+	return znode;
+}
+
+/**
+ * tnc_next - find next TNC entry.
+ * @c: UBIFS file-system description object
+ * @zn: znode is passed and returned here
+ * @n: znode branch slot number is passed and returned here
+ *
+ * This function returns %0 if the next TNC entry is found, %-ENOENT if there is
+ * no next entry, or a negative error code otherwise.
+ */
+static int tnc_next(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
+{
+	struct ubifs_znode *znode = *zn;
+	int nn = *n;
+
+	nn += 1;
+	if (nn < znode->child_cnt) {
+		*n = nn;
+		return 0;
+	}
+	while (1) {
+		struct ubifs_znode *zp;
+
+		zp = znode->parent;
+		if (!zp)
+			return -ENOENT;
+		nn = znode->iip + 1;
+		znode = zp;
+		if (nn < znode->child_cnt) {
+			znode = get_znode(c, znode, nn);
+			if (IS_ERR(znode))
+				return PTR_ERR(znode);
+			while (znode->level != 0) {
+				znode = get_znode(c, znode, 0);
+				if (IS_ERR(znode))
+					return PTR_ERR(znode);
+			}
+			nn = 0;
+			break;
+		}
+	}
+	*zn = znode;
+	*n = nn;
+	return 0;
+}
+
+/**
+ * tnc_prev - find previous TNC entry.
+ * @c: UBIFS file-system description object
+ * @zn: znode is returned here
+ * @n: znode branch slot number is passed and returned here
+ *
+ * This function returns %0 if the previous TNC entry is found, %-ENOENT if
+ * there is no next entry, or a negative error code otherwise.
+ */
+static int tnc_prev(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
+{
+	struct ubifs_znode *znode = *zn;
+	int nn = *n;
+
+	if (nn > 0) {
+		*n = nn - 1;
+		return 0;
+	}
+	while (1) {
+		struct ubifs_znode *zp;
+
+		zp = znode->parent;
+		if (!zp)
+			return -ENOENT;
+		nn = znode->iip - 1;
+		znode = zp;
+		if (nn >= 0) {
+			znode = get_znode(c, znode, nn);
+			if (IS_ERR(znode))
+				return PTR_ERR(znode);
+			while (znode->level != 0) {
+				nn = znode->child_cnt - 1;
+				znode = get_znode(c, znode, nn);
+				if (IS_ERR(znode))
+					return PTR_ERR(znode);
+			}
+			nn = znode->child_cnt - 1;
+			break;
+		}
+	}
+	*zn = znode;
+	*n = nn;
+	return 0;
+}
+
+/**
+ * resolve_collision - resolve a collision.
+ * @c: UBIFS file-system description object
+ * @key: key of a directory or extended attribute entry
+ * @zn: znode is returned here
+ * @n: zbranch number is passed and returned here
+ * @nm: name of the entry
+ *
+ * This function is called for "hashed" keys to make sure that the found key
+ * really corresponds to the looked up node (directory or extended attribute
+ * entry). It returns %1 and sets @zn and @n if the collision is resolved.
+ * %0 is returned if @nm is not found and @zn and @n are set to the previous
+ * entry, i.e. to the entry after which @nm could follow if it were in TNC.
+ * This means that @n may be set to %-1 if the leftmost key in @zn is the
+ * previous one. A negative error code is returned on failures.
+ */
+static int resolve_collision(struct ubifs_info *c, const union ubifs_key *key,
+			     struct ubifs_znode **zn, int *n,
+			     const struct qstr *nm)
+{
+	int err;
+
+	err = matches_name(c, &(*zn)->zbranch[*n], nm);
+	if (unlikely(err < 0))
+		return err;
+	if (err == NAME_MATCHES)
+		return 1;
+
+	if (err == NAME_GREATER) {
+		/* Look left */
+		while (1) {
+			err = tnc_prev(c, zn, n);
+			if (err == -ENOENT) {
+				ubifs_assert(*n == 0);
+				*n = -1;
+				return 0;
+			}
+			if (err < 0)
+				return err;
+			if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
+				/*
+				 * We have found the branch after which we would
+				 * like to insert, but inserting in this znode
+				 * may still be wrong. Consider the following 3
+				 * znodes, in the case where we are resolving a
+				 * collision with Key2.
+				 *
+				 *                  znode zp
+				 *            ----------------------
+				 * level 1     |  Key0  |  Key1  |
+				 *            -----------------------
+				 *                 |            |
+				 *       znode za  |            |  znode zb
+				 *          ------------      ------------
+				 * level 0  |  Key0  |        |  Key2  |
+				 *          ------------      ------------
+				 *
+				 * The lookup finds Key2 in znode zb. Lets say
+				 * there is no match and the name is greater so
+				 * we look left. When we find Key0, we end up
+				 * here. If we return now, we will insert into
+				 * znode za at slot n = 1.  But that is invalid
+				 * according to the parent's keys.  Key2 must
+				 * be inserted into znode zb.
+				 *
+				 * Note, this problem is not relevant for the
+				 * case when we go right, because
+				 * 'tnc_insert()' would correct the parent key.
+				 */
+				if (*n == (*zn)->child_cnt - 1) {
+					err = tnc_next(c, zn, n);
+					if (err) {
+						/* Should be impossible */
+						ubifs_assert(0);
+						if (err == -ENOENT)
+							err = -EINVAL;
+						return err;
+					}
+					ubifs_assert(*n == 0);
+					*n = -1;
+				}
+				return 0;
+			}
+			err = matches_name(c, &(*zn)->zbranch[*n], nm);
+			if (err < 0)
+				return err;
+			if (err == NAME_LESS)
+				return 0;
+			if (err == NAME_MATCHES)
+				return 1;
+			ubifs_assert(err == NAME_GREATER);
+		}
+	} else {
+		int nn = *n;
+		struct ubifs_znode *znode = *zn;
+
+		/* Look right */
+		while (1) {
+			err = tnc_next(c, &znode, &nn);
+			if (err == -ENOENT)
+				return 0;
+			if (err < 0)
+				return err;
+			if (keys_cmp(c, &znode->zbranch[nn].key, key))
+				return 0;
+			err = matches_name(c, &znode->zbranch[nn], nm);
+			if (err < 0)
+				return err;
+			if (err == NAME_GREATER)
+				return 0;
+			*zn = znode;
+			*n = nn;
+			if (err == NAME_MATCHES)
+				return 1;
+			ubifs_assert(err == NAME_LESS);
+		}
+	}
+}
+
+/**
+ * fallible_matches_name - determine if a dent matches a given name.
+ * @c: UBIFS file-system description object
+ * @zbr: zbranch of dent
+ * @nm: name to match
+ *
+ * This is a "fallible" version of 'matches_name()' function which does not
+ * panic if the direntry/xentry referred by @zbr does not exist on the media.
+ *
+ * This function checks if xentry/direntry referred by zbranch @zbr matches name
+ * @nm. Returns %NAME_MATCHES it does, %NAME_LESS if the name referred by @zbr
+ * is less than @nm, %NAME_GREATER if it is greater than @nm, and @NOT_ON_MEDIA
+ * if xentry/direntry referred by @zbr does not exist on the media. A negative
+ * error code is returned in case of failure.
+ */
+static int fallible_matches_name(struct ubifs_info *c,
+				 struct ubifs_zbranch *zbr,
+				 const struct qstr *nm)
+{
+	struct ubifs_dent_node *dent;
+	int nlen, err;
+
+	/* If possible, match against the dent in the leaf node cache */
+	if (!zbr->leaf) {
+		dent = kmalloc(zbr->len, GFP_NOFS);
+		if (!dent)
+			return -ENOMEM;
+
+		err = fallible_read_node(c, &zbr->key, zbr, dent);
+		if (err < 0)
+			goto out_free;
+		if (err == 0) {
+			/* The node was not present */
+			err = NOT_ON_MEDIA;
+			goto out_free;
+		}
+		ubifs_assert(err == 1);
+
+		err = lnc_add_directly(c, zbr, dent);
+		if (err)
+			goto out_free;
+	} else
+		dent = zbr->leaf;
+
+	nlen = le16_to_cpu(dent->nlen);
+	err = memcmp(dent->name, nm->name, min_t(int, nlen, nm->len));
+	if (err == 0) {
+		if (nlen == nm->len)
+			return NAME_MATCHES;
+		else if (nlen < nm->len)
+			return NAME_LESS;
+		else
+			return NAME_GREATER;
+	} else if (err < 0)
+		return NAME_LESS;
+	else
+		return NAME_GREATER;
+
+out_free:
+	kfree(dent);
+	return err;
+}
+
+/**
+ * fallible_resolve_collision - resolve a collision even if nodes are missing.
+ * @c: UBIFS file-system description object
+ * @key: key
+ * @zn: znode is returned here
+ * @n: branch number is passed and returned here
+ * @nm: name of directory entry
+ * @adding: indicates caller is adding a key to the TNC
+ *
+ * This is a "fallible" version of the 'resolve_collision()' function which
+ * does not panic if one of the nodes referred to by TNC does not exist on the
+ * media. This may happen when replaying the journal if a deleted node was
+ * Garbage-collected and the commit was not done. A branch that refers to a node
+ * that is not present is called a dangling branch. The following are the return
+ * codes for this function:
+ *  o if @nm was found, %1 is returned and @zn and @n are set to the found
+ *    branch;
+ *  o if we are @adding and @nm was not found, %0 is returned;
+ *  o if we are not @adding and @nm was not found, but a dangling branch was
+ *    found, then %1 is returned and @zn and @n are set to the dangling branch;
+ *  o a negative error code is returned in case of failure.
+ */
+static int fallible_resolve_collision(struct ubifs_info *c,
+				      const union ubifs_key *key,
+				      struct ubifs_znode **zn, int *n,
+				      const struct qstr *nm, int adding)
+{
+	struct ubifs_znode *o_znode = NULL, *znode = *zn;
+	int uninitialized_var(o_n), err, cmp, unsure = 0, nn = *n;
+
+	cmp = fallible_matches_name(c, &znode->zbranch[nn], nm);
+	if (unlikely(cmp < 0))
+		return cmp;
+	if (cmp == NAME_MATCHES)
+		return 1;
+	if (cmp == NOT_ON_MEDIA) {
+		o_znode = znode;
+		o_n = nn;
+		/*
+		 * We are unlucky and hit a dangling branch straight away.
+		 * Now we do not really know where to go to find the needed
+		 * branch - to the left or to the right. Well, let's try left.
+		 */
+		unsure = 1;
+	} else if (!adding)
+		unsure = 1; /* Remove a dangling branch wherever it is */
+
+	if (cmp == NAME_GREATER || unsure) {
+		/* Look left */
+		while (1) {
+			err = tnc_prev(c, zn, n);
+			if (err == -ENOENT) {
+				ubifs_assert(*n == 0);
+				*n = -1;
+				break;
+			}
+			if (err < 0)
+				return err;
+			if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
+				/* See comments in 'resolve_collision()' */
+				if (*n == (*zn)->child_cnt - 1) {
+					err = tnc_next(c, zn, n);
+					if (err) {
+						/* Should be impossible */
+						ubifs_assert(0);
+						if (err == -ENOENT)
+							err = -EINVAL;
+						return err;
+					}
+					ubifs_assert(*n == 0);
+					*n = -1;
+				}
+				break;
+			}
+			err = fallible_matches_name(c, &(*zn)->zbranch[*n], nm);
+			if (err < 0)
+				return err;
+			if (err == NAME_MATCHES)
+				return 1;
+			if (err == NOT_ON_MEDIA) {
+				o_znode = *zn;
+				o_n = *n;
+				continue;
+			}
+			if (!adding)
+				continue;
+			if (err == NAME_LESS)
+				break;
+			else
+				unsure = 0;
+		}
+	}
+
+	if (cmp == NAME_LESS || unsure) {
+		/* Look right */
+		*zn = znode;
+		*n = nn;
+		while (1) {
+			err = tnc_next(c, &znode, &nn);
+			if (err == -ENOENT)
+				break;
+			if (err < 0)
+				return err;
+			if (keys_cmp(c, &znode->zbranch[nn].key, key))
+				break;
+			err = fallible_matches_name(c, &znode->zbranch[nn], nm);
+			if (err < 0)
+				return err;
+			if (err == NAME_GREATER)
+				break;
+			*zn = znode;
+			*n = nn;
+			if (err == NAME_MATCHES)
+				return 1;
+			if (err == NOT_ON_MEDIA) {
+				o_znode = znode;
+				o_n = nn;
+			}
+		}
+	}
+
+	/* Never match a dangling branch when adding */
+	if (adding || !o_znode)
+		return 0;
+
+	dbg_mnt("dangling match LEB %d:%d len %d %s",
+		o_znode->zbranch[o_n].lnum, o_znode->zbranch[o_n].offs,
+		o_znode->zbranch[o_n].len, DBGKEY(key));
+	*zn = o_znode;
+	*n = o_n;
+	return 1;
+}
+
+/**
+ * matches_position - determine if a zbranch matches a given position.
+ * @zbr: zbranch of dent
+ * @lnum: LEB number of dent to match
+ * @offs: offset of dent to match
+ *
+ * This function returns %1 if @lnum:@offs matches, and %0 otherwise.
+ */
+static int matches_position(struct ubifs_zbranch *zbr, int lnum, int offs)
+{
+	if (zbr->lnum == lnum && zbr->offs == offs)
+		return 1;
+	else
+		return 0;
+}
+
+/**
+ * resolve_collision_directly - resolve a collision directly.
+ * @c: UBIFS file-system description object
+ * @key: key of directory entry
+ * @zn: znode is passed and returned here
+ * @n: zbranch number is passed and returned here
+ * @lnum: LEB number of dent node to match
+ * @offs: offset of dent node to match
+ *
+ * This function is used for "hashed" keys to make sure the found directory or
+ * extended attribute entry node is what was looked for. It is used when the
+ * flash address of the right node is known (@lnum:@offs) which makes it much
+ * easier to resolve collisions (no need to read entries and match full
+ * names). This function returns %1 and sets @zn and @n if the collision is
+ * resolved, %0 if @lnum:@offs is not found and @zn and @n are set to the
+ * previous directory entry. Otherwise a negative error code is returned.
+ */
+static int resolve_collision_directly(struct ubifs_info *c,
+				      const union ubifs_key *key,
+				      struct ubifs_znode **zn, int *n,
+				      int lnum, int offs)
+{
+	struct ubifs_znode *znode;
+	int nn, err;
+
+	znode = *zn;
+	nn = *n;
+	if (matches_position(&znode->zbranch[nn], lnum, offs))
+		return 1;
+
+	/* Look left */
+	while (1) {
+		err = tnc_prev(c, &znode, &nn);
+		if (err == -ENOENT)
+			break;
+		if (err < 0)
+			return err;
+		if (keys_cmp(c, &znode->zbranch[nn].key, key))
+			break;
+		if (matches_position(&znode->zbranch[nn], lnum, offs)) {
+			*zn = znode;
+			*n = nn;
+			return 1;
+		}
+	}
+
+	/* Look right */
+	znode = *zn;
+	nn = *n;
+	while (1) {
+		err = tnc_next(c, &znode, &nn);
+		if (err == -ENOENT)
+			return 0;
+		if (err < 0)
+			return err;
+		if (keys_cmp(c, &znode->zbranch[nn].key, key))
+			return 0;
+		*zn = znode;
+		*n = nn;
+		if (matches_position(&znode->zbranch[nn], lnum, offs))
+			return 1;
+	}
+}
+
+/**
+ * dirty_cow_bottom_up - dirty a znode and its ancestors.
+ * @c: UBIFS file-system description object
+ * @znode: znode to dirty
+ *
+ * If we do not have a unique key that resides in a znode, then we cannot
+ * dirty that znode from the top down (i.e. by using lookup_level0_dirty)
+ * This function records the path back to the last dirty ancestor, and then
+ * dirties the znodes on that path.
+ */
+static struct ubifs_znode *dirty_cow_bottom_up(struct ubifs_info *c,
+					       struct ubifs_znode *znode)
+{
+	struct ubifs_znode *zp;
+	int *path = c->bottom_up_buf, p = 0;
+
+	ubifs_assert(c->zroot.znode);
+	ubifs_assert(znode);
+	if (c->zroot.znode->level > BOTTOM_UP_HEIGHT) {
+		kfree(c->bottom_up_buf);
+		c->bottom_up_buf = kmalloc(c->zroot.znode->level * sizeof(int),
+					   GFP_NOFS);
+		if (!c->bottom_up_buf)
+			return ERR_PTR(-ENOMEM);
+		path = c->bottom_up_buf;
+	}
+	if (c->zroot.znode->level) {
+		/* Go up until parent is dirty */
+		while (1) {
+			int n;
+
+			zp = znode->parent;
+			if (!zp)
+				break;
+			n = znode->iip;
+			ubifs_assert(p < c->zroot.znode->level);
+			path[p++] = n;
+			if (!zp->cnext && ubifs_zn_dirty(znode))
+				break;
+			znode = zp;
+		}
+	}
+
+	/* Come back down, dirtying as we go */
+	while (1) {
+		struct ubifs_zbranch *zbr;
+
+		zp = znode->parent;
+		if (zp) {
+			ubifs_assert(path[p - 1] >= 0);
+			ubifs_assert(path[p - 1] < zp->child_cnt);
+			zbr = &zp->zbranch[path[--p]];
+			znode = dirty_cow_znode(c, zbr);
+		} else {
+			ubifs_assert(znode == c->zroot.znode);
+			znode = dirty_cow_znode(c, &c->zroot);
+		}
+		if (unlikely(IS_ERR(znode)) || !p)
+			break;
+		ubifs_assert(path[p - 1] >= 0);
+		ubifs_assert(path[p - 1] < znode->child_cnt);
+		znode = znode->zbranch[path[p - 1]].znode;
+	}
+
+	return znode;
+}
+
+/**
+ * ubifs_lookup_level0 - search for zero-level znode.
+ * @c: UBIFS file-system description object
+ * @key:  key to lookup
+ * @zn: znode is returned here
+ * @n: znode branch slot number is returned here
+ *
+ * This function looks up the TNC tree and search for zero-level znode which
+ * refers key @key. The found zero-level znode is returned in @zn. There are 3
+ * cases:
+ *   o exact match, i.e. the found zero-level znode contains key @key, then %1