ref: 406d9c3c653b53f4202f0de3f9b3e8a0a1eec70c
dir: /src/ext4_dir_idx.c/
/*
* Copyright (c) 2013 Grzegorz Kostka (kostka.grzegorz@gmail.com)
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
*
* - Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* - 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.
* - The name of the author may not be used to endorse or promote products
* derived from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``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 AUTHOR 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.
*/
/** @addtogroup lwext4
* @{
*/
/**
* @file ext4_dir_idx.c
* @brief Directory indexing procedures.
*/
#include "ext4_config.h"
#include "ext4_types.h"
#include "ext4_misc.h"
#include "ext4_errno.h"
#include "ext4_debug.h"
#include "ext4_trans.h"
#include "ext4_dir_idx.h"
#include "ext4_dir.h"
#include "ext4_blockdev.h"
#include "ext4_fs.h"
#include "ext4_super.h"
#include "ext4_inode.h"
#include "ext4_crc32.h"
#include "ext4_hash.h"
#include <string.h>
#include <stdlib.h>
/**@brief Get hash version used in directory index.
* @param root_info Pointer to root info structure of index
* @return Hash algorithm version
*/
static inline uint8_t
ext4_dir_dx_rinfo_get_hash_version(struct ext4_dir_idx_rinfo *ri)
{
return ri->hash_version;
}
/**@brief Set hash version, that will be used in directory index.
* @param root_info Pointer to root info structure of index
* @param v Hash algorithm version
*/
static inline void
ext4_dir_dx_rinfo_set_hash_version(struct ext4_dir_idx_rinfo *ri, uint8_t v)
{
ri->hash_version = v;
}
/**@brief Get length of root_info structure in bytes.
* @param root_info Pointer to root info structure of index
* @return Length of the structure
*/
static inline uint8_t
ext4_dir_dx_rinfo_get_info_length(struct ext4_dir_idx_rinfo *ri)
{
return ri->info_length;
}
/**@brief Set length of root_info structure in bytes.
* @param root_info Pointer to root info structure of index
* @param info_length Length of the structure
*/
static inline void
ext4_dir_dx_root_info_set_info_length(struct ext4_dir_idx_rinfo *ri,
uint8_t len)
{
ri->info_length = len;
}
/**@brief Get number of indirect levels of HTree.
* @param root_info Pointer to root info structure of index
* @return Height of HTree (actually only 0 or 1)
*/
static inline uint8_t
ext4_dir_dx_rinfo_get_indirect_levels(struct ext4_dir_idx_rinfo *ri)
{
return ri->indirect_levels;
}
/**@brief Set number of indirect levels of HTree.
* @param root_info Pointer to root info structure of index
* @param lvl Height of HTree (actually only 0 or 1)
*/
static inline void
ext4_dir_dx_rinfo_set_indirect_levels(struct ext4_dir_idx_rinfo *ri, uint8_t l)
{
ri->indirect_levels = l;
}
/**@brief Get maximum number of index node entries.
* @param climit Pointer to counlimit structure
* @return Maximum of entries in node
*/
static inline uint16_t
ext4_dir_dx_climit_get_limit(struct ext4_dir_idx_climit *climit)
{
return to_le16(climit->limit);
}
/**@brief Set maximum number of index node entries.
* @param climit Pointer to counlimit structure
* @param limit Maximum of entries in node
*/
static inline void
ext4_dir_dx_climit_set_limit(struct ext4_dir_idx_climit *climit, uint16_t limit)
{
climit->limit = to_le16(limit);
}
/**@brief Get current number of index node entries.
* @param climit Pointer to counlimit structure
* @return Number of entries in node
*/
static inline uint16_t
ext4_dir_dx_climit_get_count(struct ext4_dir_idx_climit *climit)
{
return to_le16(climit->count);
}
/**@brief Set current number of index node entries.
* @param climit Pointer to counlimit structure
* @param count Number of entries in node
*/
static inline void
ext4_dir_dx_climit_set_count(struct ext4_dir_idx_climit *climit, uint16_t count)
{
climit->count = to_le16(count);
}
/**@brief Get hash value of index entry.
* @param entry Pointer to index entry
* @return Hash value
*/
static inline uint32_t
ext4_dir_dx_entry_get_hash(struct ext4_dir_idx_entry *entry)
{
return to_le32(entry->hash);
}
/**@brief Set hash value of index entry.
* @param entry Pointer to index entry
* @param hash Hash value
*/
static inline void
ext4_dir_dx_entry_set_hash(struct ext4_dir_idx_entry *entry, uint32_t hash)
{
entry->hash = to_le32(hash);
}
/**@brief Get block address where child node is located.
* @param entry Pointer to index entry
* @return Block address of child node
*/
static inline uint32_t
ext4_dir_dx_entry_get_block(struct ext4_dir_idx_entry *entry)
{
return to_le32(entry->block);
}
/**@brief Set block address where child node is located.
* @param entry Pointer to index entry
* @param block Block address of child node
*/
static inline void
ext4_dir_dx_entry_set_block(struct ext4_dir_idx_entry *entry, uint32_t block)
{
entry->block = to_le32(block);
}
/**@brief Sort entry item.*/
struct ext4_dx_sort_entry {
uint32_t hash;
uint32_t rec_len;
void *dentry;
};
static int ext4_dir_dx_hash_string(struct ext4_hash_info *hinfo, int len,
const char *name)
{
return ext2_htree_hash(name, len, hinfo->seed, hinfo->hash_version,
&hinfo->hash, &hinfo->minor_hash);
}
#if CONFIG_META_CSUM_ENABLE
static uint32_t ext4_dir_dx_checksum(struct ext4_inode_ref *inode_ref, void *de,
int count_offset, int count,
struct ext4_dir_idx_tail *t)
{
uint32_t orig_cum, csum = 0;
struct ext4_sblock *sb = &inode_ref->fs->sb;
int sz;
/* Compute the checksum only if the filesystem supports it */
if (ext4_sb_feature_ro_com(sb, EXT4_FRO_COM_METADATA_CSUM)) {
uint32_t ino_index = to_le32(inode_ref->index);
uint32_t ino_gen;
ino_gen = to_le32(ext4_inode_get_generation(inode_ref->inode));
sz = count_offset + (count * sizeof(struct ext4_dir_idx_tail));
orig_cum = t->checksum;
t->checksum = 0;
/* First calculate crc32 checksum against fs uuid */
csum = ext4_crc32c(EXT4_CRC32_INIT, sb->uuid, sizeof(sb->uuid));
/* Then calculate crc32 checksum against inode number
* and inode generation */
csum = ext4_crc32c(csum, &ino_index, sizeof(ino_index));
csum = ext4_crc32c(csum, &ino_gen, sizeof(ino_gen));
/* After that calculate crc32 checksum against all the dx_entry */
csum = ext4_crc32c(csum, de, sz);
/* Finally calculate crc32 checksum for dx_tail */
csum = ext4_crc32c(csum, t, sizeof(struct ext4_dir_idx_tail));
t->checksum = orig_cum;
}
return csum;
}
static struct ext4_dir_idx_climit *
ext4_dir_dx_get_climit(struct ext4_inode_ref *inode_ref,
struct ext4_dir_en *dirent, int *offset)
{
struct ext4_dir_en *dp;
struct ext4_dir_idx_root *root;
struct ext4_sblock *sb = &inode_ref->fs->sb;
uint32_t block_size = ext4_sb_get_block_size(sb);
uint16_t entry_len = ext4_dir_en_get_entry_len(dirent);
int count_offset;
if (entry_len == 12) {
root = (struct ext4_dir_idx_root *)dirent;
dp = (struct ext4_dir_en *)&root->dots[1];
if (ext4_dir_en_get_entry_len(dp) != (block_size - 12))
return NULL;
if (root->info.reserved_zero)
return NULL;
if (root->info.info_length != sizeof(struct ext4_dir_idx_rinfo))
return NULL;
count_offset = 32;
} else if (entry_len == block_size) {
count_offset = 8;
} else {
return NULL;
}
if (offset)
*offset = count_offset;
return (struct ext4_dir_idx_climit *)(((char *)dirent) + count_offset);
}
/*
* BIG FAT NOTES:
* Currently we do not verify the checksum of HTree node.
*/
static bool ext4_dir_dx_csum_verify(struct ext4_inode_ref *inode_ref,
struct ext4_dir_en *de)
{
struct ext4_sblock *sb = &inode_ref->fs->sb;
uint32_t block_size = ext4_sb_get_block_size(sb);
int coff, limit, cnt;
if (ext4_sb_feature_ro_com(sb, EXT4_FRO_COM_METADATA_CSUM)) {
struct ext4_dir_idx_climit *climit;
climit = ext4_dir_dx_get_climit(inode_ref, de, &coff);
if (!climit) {
/* Directory seems corrupted. */
return true;
}
struct ext4_dir_idx_tail *t;
limit = ext4_dir_dx_climit_get_limit(climit);
cnt = ext4_dir_dx_climit_get_count(climit);
if (coff + (limit * sizeof(struct ext4_dir_idx_entry)) >
(block_size - sizeof(struct ext4_dir_idx_tail))) {
/* There is no space to hold the checksum */
return true;
}
t = (void *)(((struct ext4_dir_idx_entry *)climit) + limit);
uint32_t c;
c = to_le32(ext4_dir_dx_checksum(inode_ref, de, coff, cnt, t));
if (t->checksum != c)
return false;
}
return true;
}
static void ext4_dir_set_dx_csum(struct ext4_inode_ref *inode_ref,
struct ext4_dir_en *dirent)
{
int coff, limit, count;
struct ext4_sblock *sb = &inode_ref->fs->sb;
uint32_t block_size = ext4_sb_get_block_size(sb);
if (ext4_sb_feature_ro_com(sb, EXT4_FRO_COM_METADATA_CSUM)) {
struct ext4_dir_idx_climit *climit;
climit = ext4_dir_dx_get_climit(inode_ref, dirent, &coff);
if (!climit) {
/* Directory seems corrupted. */
return;
}
struct ext4_dir_idx_tail *t;
limit = ext4_dir_dx_climit_get_limit(climit);
count = ext4_dir_dx_climit_get_count(climit);
if (coff + (limit * sizeof(struct ext4_dir_idx_entry)) >
(block_size - sizeof(struct ext4_dir_idx_tail))) {
/* There is no space to hold the checksum */
return;
}
t = (void *)(((struct ext4_dir_idx_entry *)climit) + limit);
t->checksum = to_le32(ext4_dir_dx_checksum(inode_ref, dirent,
coff, count, t));
}
}
#else
#define ext4_dir_dx_csum_verify(...) true
#define ext4_dir_set_dx_csum(...)
#endif
/****************************************************************************/
int ext4_dir_dx_init(struct ext4_inode_ref *dir, struct ext4_inode_ref *parent)
{
/* Load block 0, where will be index root located */
ext4_fsblk_t fblock;
uint32_t iblock = 0;
bool need_append =
(ext4_inode_get_size(&dir->fs->sb, dir->inode)
< EXT4_DIR_DX_INIT_BCNT)
? true : false;
struct ext4_sblock *sb = &dir->fs->sb;
uint32_t block_size = ext4_sb_get_block_size(&dir->fs->sb);
struct ext4_block block;
int rc;
if (!need_append)
rc = ext4_fs_init_inode_dblk_idx(dir, iblock, &fblock);
else
rc = ext4_fs_append_inode_dblk(dir, &fblock, &iblock);
if (rc != EOK)
return rc;
rc = ext4_trans_block_get_noread(dir->fs->bdev, &block, fblock);
if (rc != EOK)
return rc;
/* Initialize pointers to data structures */
struct ext4_dir_idx_root *root = (void *)block.data;
struct ext4_dir_idx_rinfo *info = &(root->info);
memset(root, 0, sizeof(struct ext4_dir_idx_root));
struct ext4_dir_en *de;
/* Initialize dot entries */
de = (struct ext4_dir_en *)root->dots;
ext4_dir_write_entry(sb, de, 12, dir, ".", strlen("."));
de = (struct ext4_dir_en *)(root->dots + 1);
uint16_t elen = block_size - 12;
ext4_dir_write_entry(sb, de, elen, parent, "..", strlen(".."));
/* Initialize root info structure */
uint8_t hash_version = ext4_get8(&dir->fs->sb, default_hash_version);
ext4_dir_dx_rinfo_set_hash_version(info, hash_version);
ext4_dir_dx_rinfo_set_indirect_levels(info, 0);
ext4_dir_dx_root_info_set_info_length(info, 8);
/* Set limit and current number of entries */
struct ext4_dir_idx_climit *climit;
climit = (struct ext4_dir_idx_climit *)&root->en;
ext4_dir_dx_climit_set_count(climit, 1);
uint32_t entry_space;
entry_space = block_size - 2 * sizeof(struct ext4_dir_idx_dot_en) -
sizeof(struct ext4_dir_idx_rinfo);
if (ext4_sb_feature_ro_com(sb, EXT4_FRO_COM_METADATA_CSUM))
entry_space -= sizeof(struct ext4_dir_idx_tail);
uint16_t root_limit = entry_space / sizeof(struct ext4_dir_idx_entry);
ext4_dir_dx_climit_set_limit(climit, root_limit);
/* Append new block, where will be new entries inserted in the future */
iblock++;
if (!need_append)
rc = ext4_fs_init_inode_dblk_idx(dir, iblock, &fblock);
else
rc = ext4_fs_append_inode_dblk(dir, &fblock, &iblock);
if (rc != EOK) {
ext4_block_set(dir->fs->bdev, &block);
return rc;
}
struct ext4_block new_block;
rc = ext4_trans_block_get_noread(dir->fs->bdev, &new_block, fblock);
if (rc != EOK) {
ext4_block_set(dir->fs->bdev, &block);
return rc;
}
/* Fill the whole block with empty entry */
struct ext4_dir_en *be = (void *)new_block.data;
if (ext4_sb_feature_ro_com(sb, EXT4_FRO_COM_METADATA_CSUM)) {
uint16_t len = block_size - sizeof(struct ext4_dir_entry_tail);
ext4_dir_en_set_entry_len(be, len);
ext4_dir_en_set_name_len(sb, be, 0);
ext4_dir_en_set_inode_type(sb, be, EXT4_DE_UNKNOWN);
ext4_dir_init_entry_tail(EXT4_DIRENT_TAIL(be, block_size));
ext4_dir_set_csum(dir, be);
} else {
ext4_dir_en_set_entry_len(be, block_size);
}
ext4_dir_en_set_inode(be, 0);
ext4_trans_set_block_dirty(new_block.buf);
rc = ext4_block_set(dir->fs->bdev, &new_block);
if (rc != EOK) {
ext4_block_set(dir->fs->bdev, &block);
return rc;
}
/* Connect new block to the only entry in index */
struct ext4_dir_idx_entry *entry = root->en;
ext4_dir_dx_entry_set_block(entry, iblock);
ext4_dir_set_dx_csum(dir, (struct ext4_dir_en *)block.data);
ext4_trans_set_block_dirty(block.buf);
return ext4_block_set(dir->fs->bdev, &block);
}
/**@brief Initialize hash info structure necessary for index operations.
* @param hinfo Pointer to hinfo to be initialized
* @param root_block Root block (number 0) of index
* @param sb Pointer to superblock
* @param name_len Length of name to be computed hash value from
* @param name Name to be computed hash value from
* @return Standard error code
*/
static int ext4_dir_hinfo_init(struct ext4_hash_info *hinfo,
struct ext4_block *root_block,
struct ext4_sblock *sb, size_t name_len,
const char *name)
{
struct ext4_dir_idx_root *root;
root = (struct ext4_dir_idx_root *)root_block->data;
if ((root->info.hash_version != EXT2_HTREE_LEGACY) &&
(root->info.hash_version != EXT2_HTREE_HALF_MD4) &&
(root->info.hash_version != EXT2_HTREE_TEA))
return EXT4_ERR_BAD_DX_DIR;
/* Check unused flags */
if (root->info.unused_flags != 0)
return EXT4_ERR_BAD_DX_DIR;
/* Check indirect levels */
if (root->info.indirect_levels > 1)
return EXT4_ERR_BAD_DX_DIR;
/* Check if node limit is correct */
uint32_t block_size = ext4_sb_get_block_size(sb);
uint32_t entry_space = block_size;
entry_space -= 2 * sizeof(struct ext4_dir_idx_dot_en);
entry_space -= sizeof(struct ext4_dir_idx_rinfo);
if (ext4_sb_feature_ro_com(sb, EXT4_FRO_COM_METADATA_CSUM))
entry_space -= sizeof(struct ext4_dir_idx_tail);
entry_space = entry_space / sizeof(struct ext4_dir_idx_entry);
struct ext4_dir_idx_climit *climit = (void *)&root->en;
uint16_t limit = ext4_dir_dx_climit_get_limit(climit);
if (limit != entry_space)
return EXT4_ERR_BAD_DX_DIR;
/* Check hash version and modify if necessary */
hinfo->hash_version = ext4_dir_dx_rinfo_get_hash_version(&root->info);
if ((hinfo->hash_version <= EXT2_HTREE_TEA) &&
(ext4_sb_check_flag(sb, EXT4_SUPERBLOCK_FLAGS_UNSIGNED_HASH))) {
/* Use unsigned hash */
hinfo->hash_version += 3;
}
/* Load hash seed from superblock */
hinfo->seed = ext4_get8(sb, hash_seed);
/* Compute hash value of name */
if (name)
return ext4_dir_dx_hash_string(hinfo, name_len, name);
return EOK;
}
/**@brief Walk through index tree and load leaf with corresponding hash value.
* @param hinfo Initialized hash info structure
* @param inode_ref Current i-node
* @param root_block Root block (iblock 0), where is root node located
* @param dx_block Pointer to leaf node in dx_blocks array
* @param dx_blocks Array with the whole path from root to leaf
* @return Standard error code
*/
static int ext4_dir_dx_get_leaf(struct ext4_hash_info *hinfo,
struct ext4_inode_ref *inode_ref,
struct ext4_block *root_block,
struct ext4_dir_idx_block **dx_block,
struct ext4_dir_idx_block *dx_blocks)
{
struct ext4_dir_idx_root *root;
struct ext4_dir_idx_entry *entries;
struct ext4_dir_idx_entry *p;
struct ext4_dir_idx_entry *q;
struct ext4_dir_idx_entry *m;
struct ext4_dir_idx_entry *at;
ext4_fsblk_t fblk;
uint32_t block_size;
uint16_t limit;
uint16_t entry_space;
uint8_t ind_level;
int r;
struct ext4_dir_idx_block *tmp_dx_blk = dx_blocks;
struct ext4_block *tmp_blk = root_block;
struct ext4_sblock *sb = &inode_ref->fs->sb;
block_size = ext4_sb_get_block_size(sb);
root = (struct ext4_dir_idx_root *)root_block->data;
entries = (struct ext4_dir_idx_entry *)&root->en;
limit = ext4_dir_dx_climit_get_limit((void *)entries);
ind_level = ext4_dir_dx_rinfo_get_indirect_levels(&root->info);
/* Walk through the index tree */
while (true) {
uint16_t cnt = ext4_dir_dx_climit_get_count((void *)entries);
if ((cnt == 0) || (cnt > limit))
return EXT4_ERR_BAD_DX_DIR;
/* Do binary search in every node */
p = entries + 1;
q = entries + cnt - 1;
while (p <= q) {
m = p + (q - p) / 2;
if (ext4_dir_dx_entry_get_hash(m) > hinfo->hash)
q = m - 1;
else
p = m + 1;
}
at = p - 1;
/* Write results */
memcpy(&tmp_dx_blk->b, tmp_blk, sizeof(struct ext4_block));
tmp_dx_blk->entries = entries;
tmp_dx_blk->position = at;
/* Is algorithm in the leaf? */
if (ind_level == 0) {
*dx_block = tmp_dx_blk;
return EOK;
}
/* Goto child node */
uint32_t n_blk = ext4_dir_dx_entry_get_block(at);
ind_level--;
r = ext4_fs_get_inode_dblk_idx(inode_ref, n_blk, &fblk, false);
if (r != EOK)
return r;
r = ext4_trans_block_get(inode_ref->fs->bdev, tmp_blk, fblk);
if (r != EOK)
return r;
entries = ((struct ext4_dir_idx_node *)tmp_blk->data)->entries;
limit = ext4_dir_dx_climit_get_limit((void *)entries);
entry_space = block_size - sizeof(struct ext4_fake_dir_entry);
if (ext4_sb_feature_ro_com(sb, EXT4_FRO_COM_METADATA_CSUM))
entry_space -= sizeof(struct ext4_dir_idx_tail);
entry_space = entry_space / sizeof(struct ext4_dir_idx_entry);
if (limit != entry_space) {
ext4_block_set(inode_ref->fs->bdev, tmp_blk);
return EXT4_ERR_BAD_DX_DIR;
}
if (!ext4_dir_dx_csum_verify(inode_ref, (void *)tmp_blk->data)) {
ext4_dbg(DEBUG_DIR_IDX,
DBG_WARN "HTree checksum failed."
"Inode: %" PRIu32", "
"Block: %" PRIu32"\n",
inode_ref->index,
n_blk);
}
++tmp_dx_blk;
}
/* Unreachable */
return EOK;
}
/**@brief Check if the the next block would be checked during entry search.
* @param inode_ref Directory i-node
* @param hash Hash value to check
* @param dx_block Current block
* @param dx_blocks Array with path from root to leaf node
* @return Standard Error code
*/
static int ext4_dir_dx_next_block(struct ext4_inode_ref *inode_ref,
uint32_t hash,
struct ext4_dir_idx_block *dx_block,
struct ext4_dir_idx_block *dx_blocks)
{
int r;
uint32_t num_handles = 0;
ext4_fsblk_t blk_adr;
struct ext4_dir_idx_block *p = dx_block;
/* Try to find data block with next bunch of entries */
while (true) {
uint16_t cnt = ext4_dir_dx_climit_get_count((void *)p->entries);
p->position++;
if (p->position < p->entries + cnt)
break;
if (p == dx_blocks)
return EOK;
num_handles++;
p--;
}
/* Check hash collision (if not occurred - no next block cannot be
* used)*/
uint32_t current_hash = ext4_dir_dx_entry_get_hash(p->position);
if ((hash & 1) == 0) {
if ((current_hash & ~1) != hash)
return 0;
}
/* Fill new path */
while (num_handles--) {
uint32_t blk = ext4_dir_dx_entry_get_block(p->position);
r = ext4_fs_get_inode_dblk_idx(inode_ref, blk, &blk_adr, false);
if (r != EOK)
return r;
struct ext4_block b;
r = ext4_trans_block_get(inode_ref->fs->bdev, &b, blk_adr);
if (r != EOK)
return r;
if (!ext4_dir_dx_csum_verify(inode_ref, (void *)b.data)) {
ext4_dbg(DEBUG_DIR_IDX,
DBG_WARN "HTree checksum failed."
"Inode: %" PRIu32", "
"Block: %" PRIu32"\n",
inode_ref->index,
blk);
}
p++;
/* Don't forget to put old block (prevent memory leak) */
r = ext4_block_set(inode_ref->fs->bdev, &p->b);
if (r != EOK)
return r;
memcpy(&p->b, &b, sizeof(b));
p->entries = ((struct ext4_dir_idx_node *)b.data)->entries;
p->position = p->entries;
}
return ENOENT;
}
int ext4_dir_dx_find_entry(struct ext4_dir_search_result *result,
struct ext4_inode_ref *inode_ref, size_t name_len,
const char *name)
{
/* Load direct block 0 (index root) */
ext4_fsblk_t root_block_addr;
int rc2;
int rc;
rc = ext4_fs_get_inode_dblk_idx(inode_ref, 0, &root_block_addr, false);
if (rc != EOK)
return rc;
struct ext4_fs *fs = inode_ref->fs;
struct ext4_block root_block;
rc = ext4_trans_block_get(fs->bdev, &root_block, root_block_addr);
if (rc != EOK)
return rc;
if (!ext4_dir_dx_csum_verify(inode_ref, (void *)root_block.data)) {
ext4_dbg(DEBUG_DIR_IDX,
DBG_WARN "HTree root checksum failed."
"Inode: %" PRIu32", "
"Block: %" PRIu32"\n",
inode_ref->index,
(uint32_t)0);
}
/* Initialize hash info (compute hash value) */
struct ext4_hash_info hinfo;
rc = ext4_dir_hinfo_init(&hinfo, &root_block, &fs->sb, name_len, name);
if (rc != EOK) {
ext4_block_set(fs->bdev, &root_block);
return EXT4_ERR_BAD_DX_DIR;
}
/*
* Hardcoded number 2 means maximum height of index tree,
* specified in the Linux driver.
*/
struct ext4_dir_idx_block dx_blocks[2];
struct ext4_dir_idx_block *dx_block;
struct ext4_dir_idx_block *tmp;
rc = ext4_dir_dx_get_leaf(&hinfo, inode_ref, &root_block, &dx_block,
dx_blocks);
if (rc != EOK) {
ext4_block_set(fs->bdev, &root_block);
return EXT4_ERR_BAD_DX_DIR;
}
do {
/* Load leaf block */
uint32_t leaf_blk_idx;
ext4_fsblk_t leaf_block_addr;
struct ext4_block b;
leaf_blk_idx = ext4_dir_dx_entry_get_block(dx_block->position);
rc = ext4_fs_get_inode_dblk_idx(inode_ref, leaf_blk_idx,
&leaf_block_addr, false);
if (rc != EOK)
goto cleanup;
rc = ext4_trans_block_get(fs->bdev, &b, leaf_block_addr);
if (rc != EOK)
goto cleanup;
if (!ext4_dir_csum_verify(inode_ref, (void *)b.data)) {
ext4_dbg(DEBUG_DIR_IDX,
DBG_WARN "HTree leaf block checksum failed."
"Inode: %" PRIu32", "
"Block: %" PRIu32"\n",
inode_ref->index,
leaf_blk_idx);
}
/* Linear search inside block */
struct ext4_dir_en *de;
rc = ext4_dir_find_in_block(&b, &fs->sb, name_len, name, &de);
/* Found => return it */
if (rc == EOK) {
result->block = b;
result->dentry = de;
goto cleanup;
}
/* Not found, leave untouched */
rc2 = ext4_block_set(fs->bdev, &b);
if (rc2 != EOK)
goto cleanup;
if (rc != ENOENT)
goto cleanup;
/* check if the next block could be checked */
rc = ext4_dir_dx_next_block(inode_ref, hinfo.hash, dx_block,
&dx_blocks[0]);
if (rc < 0)
goto cleanup;
} while (rc == ENOENT);
/* Entry not found */
rc = ENOENT;
cleanup:
/* The whole path must be released (preventing memory leak) */
tmp = dx_blocks;
while (tmp <= dx_block) {
rc2 = ext4_block_set(fs->bdev, &tmp->b);
if (rc == EOK && rc2 != EOK)
rc = rc2;
++tmp;
}
return rc;
}
#if CONFIG_DIR_INDEX_COMB_SORT
#define SWAP_ENTRY(se1, se2) \
do { \
struct ext4_dx_sort_entry tmp = se1; \
se1 = se2; \
se2 = tmp; \
\
} while (0)
static void comb_sort(struct ext4_dx_sort_entry *se, uint32_t count)
{
struct ext4_dx_sort_entry *p, *q, *top = se + count - 1;
bool more;
/* Combsort */
while (count > 2) {
count = (count * 10) / 13;
if (count - 9 < 2)
count = 11;
for (p = top, q = p - count; q >= se; p--, q--)
if (p->hash < q->hash)
SWAP_ENTRY(*p, *q);
}
/* Bubblesort */
do {
more = 0;
q = top;
while (q-- > se) {
if (q[1].hash >= q[0].hash)
continue;
SWAP_ENTRY(*(q + 1), *q);
more = 1;
}
} while (more);
}
#else
/**@brief Compare function used to pass in quicksort implementation.
* It can compare two entries by hash value.
* @param arg1 First entry
* @param arg2 Second entry
* @param dummy Unused parameter, can be NULL
*
* @return Classic compare result
* (0: equal, -1: arg1 < arg2, 1: arg1 > arg2)
*/
static int ext4_dir_dx_entry_comparator(const void *arg1, const void *arg2)
{
struct ext4_dx_sort_entry *entry1 = (void *)arg1;
struct ext4_dx_sort_entry *entry2 = (void *)arg2;
if (entry1->hash == entry2->hash)
return 0;
if (entry1->hash < entry2->hash)
return -1;
else
return 1;
}
#endif
/**@brief Insert new index entry to block.
* Note that space for new entry must be checked by caller.
* @param inode_ref Directory i-node
* @param index_block Block where to insert new entry
* @param hash Hash value covered by child node
* @param iblock Logical number of child block
*
*/
static void
ext4_dir_dx_insert_entry(struct ext4_inode_ref *inode_ref __unused,
struct ext4_dir_idx_block *index_block,
uint32_t hash, uint32_t iblock)
{
struct ext4_dir_idx_entry *old_index_entry = index_block->position;
struct ext4_dir_idx_entry *new_index_entry = old_index_entry + 1;
struct ext4_dir_idx_climit *climit = (void *)index_block->entries;
struct ext4_dir_idx_entry *start_index = index_block->entries;
uint32_t count = ext4_dir_dx_climit_get_count(climit);
size_t bytes;
bytes = (uint8_t *)(start_index + count) - (uint8_t *)(new_index_entry);
memmove(new_index_entry + 1, new_index_entry, bytes);
ext4_dir_dx_entry_set_block(new_index_entry, iblock);
ext4_dir_dx_entry_set_hash(new_index_entry, hash);
ext4_dir_dx_climit_set_count(climit, count + 1);
ext4_dir_set_dx_csum(inode_ref, (void *)index_block->b.data);
ext4_trans_set_block_dirty(index_block->b.buf);
}
/**@brief Split directory entries to two parts preventing node overflow.
* @param inode_ref Directory i-node
* @param hinfo Hash info
* @param old_data_block Block with data to be split
* @param index_block Block where index entries are located
* @param new_data_block Output value for newly allocated data block
*/
static int ext4_dir_dx_split_data(struct ext4_inode_ref *inode_ref,
struct ext4_hash_info *hinfo,
struct ext4_block *old_data_block,
struct ext4_dir_idx_block *index_block,
struct ext4_block *new_data_block)
{
int rc = EOK;
struct ext4_sblock *sb = &inode_ref->fs->sb;
uint32_t block_size = ext4_sb_get_block_size(&inode_ref->fs->sb);
/* Allocate buffer for directory entries */
uint8_t *entry_buffer = ext4_malloc(block_size);
if (entry_buffer == NULL)
return ENOMEM;
/* dot entry has the smallest size available */
uint32_t max_ecnt = block_size / sizeof(struct ext4_dir_idx_dot_en);
/* Allocate sort entry */
struct ext4_dx_sort_entry *sort;
sort = ext4_malloc(max_ecnt * sizeof(struct ext4_dx_sort_entry));
if (sort == NULL) {
ext4_free(entry_buffer);
return ENOMEM;
}
uint32_t idx = 0;
uint32_t real_size = 0;
/* Initialize hinfo */
struct ext4_hash_info hinfo_tmp;
memcpy(&hinfo_tmp, hinfo, sizeof(struct ext4_hash_info));
/* Load all valid entries to the buffer */
struct ext4_dir_en *de = (void *)old_data_block->data;
uint8_t *entry_buffer_ptr = entry_buffer;
while ((void *)de < (void *)(old_data_block->data + block_size)) {
/* Read only valid entries */
if (ext4_dir_en_get_inode(de) && de->name_len) {
uint16_t len = ext4_dir_en_get_name_len(sb, de);
rc = ext4_dir_dx_hash_string(&hinfo_tmp, len,
(char *)de->name);
if (rc != EOK) {
ext4_free(sort);
ext4_free(entry_buffer);
return rc;
}
uint32_t rec_len = 8 + len;
if ((rec_len % 4) != 0)
rec_len += 4 - (rec_len % 4);
memcpy(entry_buffer_ptr, de, rec_len);
sort[idx].dentry = entry_buffer_ptr;
sort[idx].rec_len = rec_len;
sort[idx].hash = hinfo_tmp.hash;
entry_buffer_ptr += rec_len;
real_size += rec_len;
idx++;
}
size_t elen = ext4_dir_en_get_entry_len(de);
de = (void *)((uint8_t *)de + elen);
}
/* Sort all entries */
#if CONFIG_DIR_INDEX_COMB_SORT
comb_sort(sort, idx);
#else
qsort(sort, idx, sizeof(struct ext4_dx_sort_entry),
ext4_dir_dx_entry_comparator);
#endif
/* Allocate new block for store the second part of entries */
ext4_fsblk_t new_fblock;
uint32_t new_iblock;
rc = ext4_fs_append_inode_dblk(inode_ref, &new_fblock, &new_iblock);
if (rc != EOK) {
ext4_free(sort);
ext4_free(entry_buffer);
return rc;
}
/* Load new block */
struct ext4_block new_data_block_tmp;
rc = ext4_trans_block_get_noread(inode_ref->fs->bdev, &new_data_block_tmp,
new_fblock);
if (rc != EOK) {
ext4_free(sort);
ext4_free(entry_buffer);
return rc;
}
/*
* Distribute entries to two blocks (by size)
* - compute the half
*/
uint32_t new_hash = 0;
uint32_t current_size = 0;
uint32_t mid = 0;
uint32_t i;
for (i = 0; i < idx; ++i) {
if ((current_size + sort[i].rec_len) > (block_size / 2)) {
new_hash = sort[i].hash;
mid = i;
break;
}
current_size += sort[i].rec_len;
}
/* Check hash collision */
uint32_t continued = 0;
if (new_hash == sort[mid - 1].hash)
continued = 1;
uint32_t off = 0;
void *ptr;
if (ext4_sb_feature_ro_com(sb, EXT4_FRO_COM_METADATA_CSUM))
block_size -= sizeof(struct ext4_dir_entry_tail);
/* First part - to the old block */
for (i = 0; i < mid; ++i) {
ptr = old_data_block->data + off;
memcpy(ptr, sort[i].dentry, sort[i].rec_len);
struct ext4_dir_en *t = ptr;
if (i < (mid - 1))
ext4_dir_en_set_entry_len(t, sort[i].rec_len);
else
ext4_dir_en_set_entry_len(t, block_size - off);
off += sort[i].rec_len;
}
/* Second part - to the new block */
off = 0;
for (i = mid; i < idx; ++i) {
ptr = new_data_block_tmp.data + off;
memcpy(ptr, sort[i].dentry, sort[i].rec_len);
struct ext4_dir_en *t = ptr;
if (i < (idx - 1))
ext4_dir_en_set_entry_len(t, sort[i].rec_len);
else
ext4_dir_en_set_entry_len(t, block_size - off);
off += sort[i].rec_len;
}
block_size = ext4_sb_get_block_size(&inode_ref->fs->sb);
/* Do some steps to finish operation */
sb = &inode_ref->fs->sb;
if (ext4_sb_feature_ro_com(sb, EXT4_FRO_COM_METADATA_CSUM)) {
struct ext4_dir_entry_tail *t;
t = EXT4_DIRENT_TAIL(old_data_block->data, block_size);
ext4_dir_init_entry_tail(t);
t = EXT4_DIRENT_TAIL(new_data_block_tmp.data, block_size);
ext4_dir_init_entry_tail(t);
}
ext4_dir_set_csum(inode_ref, (void *)old_data_block->data);
ext4_dir_set_csum(inode_ref, (void *)new_data_block_tmp.data);
ext4_trans_set_block_dirty(old_data_block->buf);
ext4_trans_set_block_dirty(new_data_block_tmp.buf);
ext4_free(sort);
ext4_free(entry_buffer);
ext4_dir_dx_insert_entry(inode_ref, index_block, new_hash + continued,
new_iblock);
*new_data_block = new_data_block_tmp;
return EOK;
}
/**@brief Split index node and maybe some parent nodes in the tree hierarchy.
* @param inode_ref Directory i-node
* @param dx_blocks Array with path from root to leaf node
* @param dx_block Leaf block to be split if needed
* @return Error code
*/
static int
ext4_dir_dx_split_index(struct ext4_inode_ref *ino_ref,
struct ext4_dir_idx_block *dx_blks,
struct ext4_dir_idx_block *dxb,
struct ext4_dir_idx_block **new_dx_block)
{
struct ext4_sblock *sb = &ino_ref->fs->sb;
struct ext4_dir_idx_entry *e;
int r;
uint32_t block_size = ext4_sb_get_block_size(&ino_ref->fs->sb);
uint32_t entry_space = block_size - sizeof(struct ext4_fake_dir_entry);
uint32_t node_limit = entry_space / sizeof(struct ext4_dir_idx_entry);
bool meta_csum = ext4_sb_feature_ro_com(sb, EXT4_FRO_COM_METADATA_CSUM);
if (dxb == dx_blks)
e = ((struct ext4_dir_idx_root *)dxb->b.data)->en;
else
e = ((struct ext4_dir_idx_node *)dxb->b.data)->entries;
struct ext4_dir_idx_climit *climit = (struct ext4_dir_idx_climit *)e;
uint16_t leaf_limit = ext4_dir_dx_climit_get_limit(climit);
uint16_t leaf_count = ext4_dir_dx_climit_get_count(climit);
/* Check if is necessary to split index block */
if (leaf_limit == leaf_count) {
struct ext4_dir_idx_entry *ren;
ptrdiff_t levels = dxb - dx_blks;
ren = ((struct ext4_dir_idx_root *)dx_blks[0].b.data)->en;
struct ext4_dir_idx_climit *rclimit = (void *)ren;
uint16_t root_limit = ext4_dir_dx_climit_get_limit(rclimit);
uint16_t root_count = ext4_dir_dx_climit_get_count(rclimit);
/* Linux limitation */
if ((levels > 0) && (root_limit == root_count))
return ENOSPC;
/* Add new block to directory */
ext4_fsblk_t new_fblk;
uint32_t new_iblk;
r = ext4_fs_append_inode_dblk(ino_ref, &new_fblk, &new_iblk);
if (r != EOK)
return r;
/* load new block */
struct ext4_block b;
r = ext4_trans_block_get_noread(ino_ref->fs->bdev, &b, new_fblk);
if (r != EOK)
return r;
struct ext4_dir_idx_node *new_node = (void *)b.data;
struct ext4_dir_idx_entry *new_en = new_node->entries;
memset(&new_node->fake, 0, sizeof(struct ext4_fake_dir_entry));
new_node->fake.entry_length = block_size;
/* Split leaf node */
if (levels > 0) {
uint32_t count_left = leaf_count / 2;
uint32_t count_right = leaf_count - count_left;
uint32_t hash_right;
size_t sz;
struct ext4_dir_idx_climit *left_climit;
struct ext4_dir_idx_climit *right_climit;
hash_right = ext4_dir_dx_entry_get_hash(e + count_left);
/* Copy data to new node */
sz = count_right * sizeof(struct ext4_dir_idx_entry);
memcpy(new_en, e + count_left, sz);
/* Initialize new node */
left_climit = (struct ext4_dir_idx_climit *)e;
right_climit = (struct ext4_dir_idx_climit *)new_en;
ext4_dir_dx_climit_set_count(left_climit, count_left);
ext4_dir_dx_climit_set_count(right_climit, count_right);
if (meta_csum)
entry_space -= sizeof(struct ext4_dir_idx_tail);
ext4_dir_dx_climit_set_limit(right_climit, node_limit);
/* Which index block is target for new entry */
uint32_t position_index =
(dxb->position - dxb->entries);
if (position_index >= count_left) {
ext4_dir_set_dx_csum(
ino_ref,
(struct ext4_dir_en *)
dxb->b.data);
ext4_trans_set_block_dirty(dxb->b.buf);
struct ext4_block block_tmp = dxb->b;
dxb->b = b;
dxb->position =
new_en + position_index - count_left;
dxb->entries = new_en;
b = block_tmp;
}
/* Finally insert new entry */
ext4_dir_dx_insert_entry(ino_ref, dx_blks, hash_right,
new_iblk);
ext4_dir_set_dx_csum(ino_ref, (void*)dx_blks[0].b.data);
ext4_dir_set_dx_csum(ino_ref, (void*)dx_blks[1].b.data);
ext4_trans_set_block_dirty(dx_blks[0].b.buf);
ext4_trans_set_block_dirty(dx_blks[1].b.buf);
ext4_dir_set_dx_csum(ino_ref, (void *)b.data);
ext4_trans_set_block_dirty(b.buf);
return ext4_block_set(ino_ref->fs->bdev, &b);
} else {
size_t sz;
/* Copy data from root to child block */
sz = leaf_count * sizeof(struct ext4_dir_idx_entry);
memcpy(new_en, e, sz);
struct ext4_dir_idx_climit *new_climit = (void*)new_en;
if (meta_csum)
entry_space -= sizeof(struct ext4_dir_idx_tail);
ext4_dir_dx_climit_set_limit(new_climit, node_limit);
/* Set values in root node */
struct ext4_dir_idx_climit *new_root_climit = (void *)e;
ext4_dir_dx_climit_set_count(new_root_climit, 1);
ext4_dir_dx_entry_set_block(e, new_iblk);
struct ext4_dir_idx_root *r = (void *)dx_blks[0].b.data;
r->info.indirect_levels = 1;
/* Add new entry to the path */
dxb = dx_blks + 1;
dxb->position = dx_blks->position - e + new_en;
dxb->entries = new_en;
dxb->b = b;
*new_dx_block = dxb;
ext4_dir_set_dx_csum(ino_ref, (void*)dx_blks[0].b.data);
ext4_dir_set_dx_csum(ino_ref, (void*)dx_blks[1].b.data);
ext4_trans_set_block_dirty(dx_blks[0].b.buf);
ext4_trans_set_block_dirty(dx_blks[1].b.buf);
}
}
return EOK;
}
int ext4_dir_dx_add_entry(struct ext4_inode_ref *parent,
struct ext4_inode_ref *child, const char *name)
{
int rc2 = EOK;
int r;
/* Get direct block 0 (index root) */
ext4_fsblk_t rblock_addr;
r = ext4_fs_get_inode_dblk_idx(parent, 0, &rblock_addr, false);
if (r != EOK)
return r;
struct ext4_fs *fs = parent->fs;
struct ext4_block root_blk;
r = ext4_trans_block_get(fs->bdev, &root_blk, rblock_addr);
if (r != EOK)
return r;
if (!ext4_dir_dx_csum_verify(parent, (void*)root_blk.data)) {
ext4_dbg(DEBUG_DIR_IDX,
DBG_WARN "HTree root checksum failed."
"Inode: %" PRIu32", "
"Block: %" PRIu32"\n",
parent->index,
(uint32_t)0);
}
/* Initialize hinfo structure (mainly compute hash) */
uint32_t name_len = strlen(name);
struct ext4_hash_info hinfo;
r = ext4_dir_hinfo_init(&hinfo, &root_blk, &fs->sb, name_len, name);
if (r != EOK) {
ext4_block_set(fs->bdev, &root_blk);
return EXT4_ERR_BAD_DX_DIR;
}
/*
* Hardcoded number 2 means maximum height of index
* tree defined in Linux.
*/
struct ext4_dir_idx_block dx_blks[2];
struct ext4_dir_idx_block *dx_blk;
struct ext4_dir_idx_block *dx_it;
r = ext4_dir_dx_get_leaf(&hinfo, parent, &root_blk, &dx_blk, dx_blks);
if (r != EOK) {
r = EXT4_ERR_BAD_DX_DIR;
goto release_index;
}
/* Try to insert to existing data block */
uint32_t leaf_block_idx = ext4_dir_dx_entry_get_block(dx_blk->position);
ext4_fsblk_t leaf_block_addr;
r = ext4_fs_get_inode_dblk_idx(parent, leaf_block_idx,
&leaf_block_addr, false);
if (r != EOK)
goto release_index;
/*
* Check if there is needed to split index node
* (and recursively also parent nodes)
*/
r = ext4_dir_dx_split_index(parent, dx_blks, dx_blk, &dx_blk);
if (r != EOK)
goto release_target_index;
struct ext4_block target_block;
r = ext4_trans_block_get(fs->bdev, &target_block, leaf_block_addr);
if (r != EOK)
goto release_index;
if (!ext4_dir_csum_verify(parent,(void *)target_block.data)) {
ext4_dbg(DEBUG_DIR_IDX,
DBG_WARN "HTree leaf block checksum failed."
"Inode: %" PRIu32", "
"Block: %" PRIu32"\n",
parent->index,
leaf_block_idx);
}
/* Check if insert operation passed */
r = ext4_dir_try_insert_entry(&fs->sb, parent, &target_block, child,
name, name_len);
if (r == EOK)
goto release_target_index;
/* Split entries to two blocks (includes sorting by hash value) */
struct ext4_block new_block;
r = ext4_dir_dx_split_data(parent, &hinfo, &target_block, dx_blk,
&new_block);
if (r != EOK) {
rc2 = r;
goto release_target_index;
}
/* Where to save new entry */
uint32_t blk_hash = ext4_dir_dx_entry_get_hash(dx_blk->position + 1);
if (hinfo.hash >= blk_hash)
r = ext4_dir_try_insert_entry(&fs->sb, parent, &new_block,
child, name, name_len);
else
r = ext4_dir_try_insert_entry(&fs->sb, parent, &target_block,
child, name, name_len);
/* Cleanup */
r = ext4_block_set(fs->bdev, &new_block);
if (r != EOK)
return r;
/* Cleanup operations */
release_target_index:
rc2 = r;
r = ext4_block_set(fs->bdev, &target_block);
if (r != EOK)
return r;
release_index:
if (r != EOK)
rc2 = r;
dx_it = dx_blks;
while (dx_it <= dx_blk) {
r = ext4_block_set(fs->bdev, &dx_it->b);
if (r != EOK)
return r;
dx_it++;
}
return rc2;
}
int ext4_dir_dx_reset_parent_inode(struct ext4_inode_ref *dir,
uint32_t parent_inode)
{
/* Load block 0, where will be index root located */
ext4_fsblk_t fblock;
int rc = ext4_fs_get_inode_dblk_idx(dir, 0, &fblock, false);
if (rc != EOK)
return rc;
struct ext4_block block;
rc = ext4_trans_block_get(dir->fs->bdev, &block, fblock);
if (rc != EOK)
return rc;
if (!ext4_dir_dx_csum_verify(dir, (void *)block.data)) {
ext4_dbg(DEBUG_DIR_IDX,
DBG_WARN "HTree root checksum failed."
"Inode: %" PRIu32", "
"Block: %" PRIu32"\n",
dir->index,
(uint32_t)0);
}
/* Initialize pointers to data structures */
struct ext4_dir_idx_root *root = (void *)block.data;
/* Fill the inode field with a new parent ino. */
ext4_dx_dot_en_set_inode(&root->dots[1], parent_inode);
ext4_dir_set_dx_csum(dir, (void *)block.data);
ext4_trans_set_block_dirty(block.buf);
return ext4_block_set(dir->fs->bdev, &block);
}
/**
* @}
*/