struct fsw_dnode_stat *sb);
static fsw_status_t fsw_ext4_get_extent(struct fsw_ext4_volume *vol, struct fsw_ext4_dnode *dno,
struct fsw_extent *extent);
+static fsw_status_t fsw_ext4_get_by_blkaddr(struct fsw_ext4_volume *vol, struct fsw_ext4_dnode *dno,
+ struct fsw_extent *extent);
+static fsw_status_t fsw_ext4_get_by_extent(struct fsw_ext4_volume *vol, struct fsw_ext4_dnode *dno,
+ struct fsw_extent *extent);
static fsw_status_t fsw_ext4_dir_lookup(struct fsw_ext4_volume *vol, struct fsw_ext4_dnode *dno,
struct fsw_string *lookup_name, struct fsw_ext4_dnode **child_dno);
fsw_ext4_readlink,
};
+
+static __inline int test_root(fsw_u32 a, int b)
+{
+ fsw_u32 num = b;
+
+ while (a > num)
+ num *= b;
+ return num == a;
+}
+
+static int fsw_ext4_group_sparse(fsw_u32 group)
+{
+ if (group <= 1)
+ return 1;
+ if (!(group & 1))
+ return 0;
+ return (test_root(group, 7) || test_root(group, 5) ||
+ test_root(group, 3));
+}
+
+/* calculate the first block number of the group */
+static __inline fsw_u32
+fsw_ext4_group_first_block_no(struct ext4_super_block *sb, fsw_u32 group_no)
+{
+ return group_no * (fsw_u32)EXT4_BLOCKS_PER_GROUP(sb) +
+ sb->s_first_data_block;
+}
+
/**
* Mount an ext4 volume. Reads the superblock and constructs the
* root directory dnode.
fsw_status_t status;
void *buffer;
fsw_u32 blocksize;
- fsw_u32 groupcnt, groupno, gdesc_per_block, gdesc_bno, gdesc_index;
+ fsw_u32 groupcnt, groupno, gdesc_per_block, gdesc_bno, gdesc_index, metabg_of_gdesc;
struct ext4_group_desc *gdesc;
int i;
struct fsw_string s;
FSW_MSG_DEBUG((FSW_MSGSTR("fsw_ext4_volume_mount: Incompat flag %x\n"), vol->sb->s_feature_incompat));
if (vol->sb->s_rev_level == EXT4_DYNAMIC_REV &&
- (vol->sb->s_feature_incompat & ~(EXT4_FEATURE_INCOMPAT_FILETYPE | EXT4_FEATURE_INCOMPAT_RECOVER)))
+ (vol->sb->s_feature_incompat & ~(EXT4_FEATURE_INCOMPAT_FILETYPE | EXT4_FEATURE_INCOMPAT_RECOVER |
+ EXT4_FEATURE_INCOMPAT_EXTENTS | EXT4_FEATURE_INCOMPAT_FLEX_BG |
+ EXT4_FEATURE_INCOMPAT_META_BG)))
return FSW_UNSUPPORTED;
- if (vol->sb->s_rev_level == EXT4_DYNAMIC_REV &&
- (vol->sb->s_feature_incompat & EXT4_FEATURE_INCOMPAT_RECOVER))
- {
- FSW_MSG_DEBUG((FSW_MSGSTR("fsw_ext4_volume_mount: This ext3 file system needs recovery\n")));
- // Print(L"Ext4 WARNING: This file system needs recovery, trying to use it anyway.\n");
- }
+ if (vol->sb->s_rev_level == EXT4_DYNAMIC_REV &&
+ (vol->sb->s_feature_incompat & EXT4_FEATURE_INCOMPAT_RECOVER))
+ {
+ FSW_MSG_DEBUG((FSW_MSGSTR("fsw_ext4_volume_mount: This ext3 file system needs recovery\n")));
+ // Print(L"Ext4 WARNING: This file system needs recovery, trying to use it anyway.\n");
+ }
- // set real blocksize
blocksize = EXT4_BLOCK_SIZE(vol->sb);
+ if (blocksize < EXT4_MIN_BLOCK_SIZE || blocksize > EXT4_MAX_BLOCK_SIZE)
+ return FSW_UNSUPPORTED;
+
+ // set real blocksize
fsw_set_blocksize(vol, blocksize, blocksize);
// get other info from superblock
vol->ind_bcnt = EXT4_ADDR_PER_BLOCK(vol->sb);
vol->dind_bcnt = vol->ind_bcnt * vol->ind_bcnt;
- vol->inode_size = EXT4_INODE_SIZE(vol->sb);
+ vol->inode_size = vol->sb->s_inode_size;//EXT4_INODE_SIZE(vol->sb);
for (i = 0; i < 16; i++)
if (vol->sb->s_volume_name[i] == 0)
if (status)
return status;
- // read the group descriptors to get inode table offsets
- groupcnt = ((vol->sb->s_inodes_count - 2) / vol->sb->s_inodes_per_group) + 1;
- gdesc_per_block = (vol->g.phys_blocksize / sizeof(struct ext4_group_desc));
+ // size of group descriptor depends on feature....
+ if (!(vol->sb->s_feature_incompat & EXT4_FEATURE_INCOMPAT_64BIT)) {
+ // Default minimal group descriptor size... (this might not be set in old ext2 filesystems, therefor set it!)
+ vol->sb->s_desc_size = EXT4_MIN_DESC_SIZE;
+ }
+
+ // Calculate group descriptor count the way the kernel does it...
+ groupcnt = (vol->sb->s_blocks_count_lo - vol->sb->s_first_data_block +
+ vol->sb->s_blocks_per_group - 1) / vol->sb->s_blocks_per_group;
+ // Descriptors in one block... s_desc_size needs to be set! (Usually 128 since normal block
+ // descriptors are 32 byte and block size is 4096)
+ gdesc_per_block = EXT4_DESC_PER_BLOCK(vol->sb);
+
+ // Read the group descriptors to get inode table offsets
status = fsw_alloc(sizeof(fsw_u32) * groupcnt, &vol->inotab_bno);
if (status)
return status;
+
+ // Loop through all block group descriptors in order to get inode table locations
for (groupno = 0; groupno < groupcnt; groupno++) {
- // get the block group descriptor
- gdesc_bno = (vol->sb->s_first_data_block + 1) + groupno / gdesc_per_block;
+
+ // Calculate the block number which contains the block group descriptor we look for
+ if(vol->sb->s_feature_incompat & EXT4_FEATURE_INCOMPAT_META_BG && groupno >= vol->sb->s_first_meta_bg)
+ {
+ // If option meta_bg is set, the block group descriptor is in meta block group...
+ metabg_of_gdesc = (fsw_u32)(groupno / gdesc_per_block) * gdesc_per_block;
+ gdesc_bno = fsw_ext4_group_first_block_no(vol->sb, metabg_of_gdesc);
+ // We need to know if the block group in questition has a super block, if yes, the
+ // block group descriptors are in the next block number
+ if(!(vol->sb->s_feature_ro_compat & EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) || fsw_ext4_group_sparse(metabg_of_gdesc))
+ gdesc_bno += 1;
+ }
+ else
+ {
+ // All group descriptors follow the super block (+1)
+ gdesc_bno = (vol->sb->s_first_data_block + 1) + groupno / gdesc_per_block;
+ }
gdesc_index = groupno % gdesc_per_block;
+
+ // Get block if necessary...
status = fsw_block_get(vol, gdesc_bno, 1, (void **)&buffer);
if (status)
return status;
- gdesc = ((struct ext4_group_desc *)(buffer)) + gdesc_index;
+
+ // Get group descriptor table and block number of inode table...
+ gdesc = (struct ext4_group_desc *)((char *)buffer + gdesc_index * vol->sb->s_desc_size);
vol->inotab_bno[groupno] = gdesc->bg_inode_table_lo;
+
fsw_block_release(vol, gdesc_bno, buffer);
}
if (dno->raw)
return FSW_SUCCESS;
- FSW_MSG_DEBUG((FSW_MSGSTR("fsw_ext4_dnode_fill: inode %d\n"), dno->g.dnode_id));
// read the inode block
groupno = (dno->g.dnode_id - 1) / vol->sb->s_inodes_per_group;
ino_in_group / (vol->g.phys_blocksize / vol->inode_size);
ino_index = ino_in_group % (vol->g.phys_blocksize / vol->inode_size);
status = fsw_block_get(vol, ino_bno, 2, (void **)&buffer);
+
if (status)
return status;
else
dno->g.type = FSW_DNODE_TYPE_SPECIAL;
+ FSW_MSG_DEBUG((FSW_MSGSTR("fsw_ext4_dnode_fill: inode flags %x\n"), dno->raw->i_flags));
+ FSW_MSG_DEBUG((FSW_MSGSTR("fsw_ext4_dnode_fill: i_mode %x\n"), dno->raw->i_mode));
return FSW_SUCCESS;
}
* on the dnode before. Our task here is to get the physical disk block number for
* the requested logical block number.
*
- * TODO...
- * The ext2 file system does not use extents, but stores a list of block numbers
- * using the usual direct, indirect, double-indirect, triple-indirect scheme. To
- * optimize access, this function checks if the following file blocks are mapped
- * to consecutive disk blocks and returns a combined extent if possible.
+ * The ext4 file system usually uses extents do to store those disk block numbers.
+ * However, since ext4 is backward compatible, depending on inode flags the old direct
+ * and indirect addressing scheme can still be in place...
*/
static fsw_status_t fsw_ext4_get_extent(struct fsw_ext4_volume *vol, struct fsw_ext4_dnode *dno,
struct fsw_extent *extent)
{
- fsw_status_t status;
- fsw_u32 bno, release_bno, buf_bcnt, file_bcnt;
- fsw_u32 *buffer;
- int path[5], i;
-
// Preconditions: The caller has checked that the requested logical block
// is within the file's size. The dnode has complete information, i.e.
// fsw_ext4_dnode_read_info was called successfully on it.
-
+ FSW_MSG_DEBUG((FSW_MSGSTR("fsw_ext4_get_extent: inode %d, block %d\n"), dno->g.dnode_id, extent->log_start));
extent->type = FSW_EXTENT_TYPE_PHYSBLOCK;
extent->log_count = 1;
+
+ if(dno->raw->i_flags & 1 << EXT4_INODE_EXTENTS)
+ {
+ FSW_MSG_DEBUG((FSW_MSGSTR("fsw_ext4_get_extent: inode %d uses extents\n"), dno->g.dnode_id));
+ return fsw_ext4_get_by_extent(vol, dno, extent);
+ }
+ else
+ {
+ FSW_MSG_DEBUG((FSW_MSGSTR("fsw_ext4_get_extent: inode %d uses direct/indirect block addressing\n"),
+ dno->g.dnode_id));
+ return fsw_ext4_get_by_blkaddr(vol, dno, extent);
+ }
+}
+
+/**
+ * New ext4 extents...
+ */
+static fsw_status_t fsw_ext4_get_by_extent(struct fsw_ext4_volume *vol, struct fsw_ext4_dnode *dno,
+ struct fsw_extent *extent)
+{
+ fsw_status_t status;
+ fsw_u32 bno, release_bno, buf_offset, file_bcnt;
+ int ext_cnt;
+ void *buffer;
+
+ struct ext4_extent_header *ext4_extent_header;
+ struct ext4_extent_idx *ext4_extent_idx;
+ struct ext4_extent *ext4_extent;
+
+ // Logical block requested by core...
+ bno = extent->log_start;
+
+ // First buffer is the i_block field from inode...
+ buffer = (void *)dno->raw->i_block;
+ buf_offset = 0;
+ while(1) {
+ ext4_extent_header = (struct ext4_extent_header *)((char *)buffer + buf_offset);
+ buf_offset += sizeof(struct ext4_extent_header);
+ FSW_MSG_DEBUG((FSW_MSGSTR("fsw_ext4_get_by_extent: extent header with %d entries\n"),
+ ext4_extent_header->eh_entries));
+ if(ext4_extent_header->eh_magic != EXT4_EXT_MAGIC)
+ return FSW_VOLUME_CORRUPTED;
+
+ for(ext_cnt = 0;ext_cnt < ext4_extent_header->eh_entries;ext_cnt++)
+ {
+ if(ext4_extent_header->eh_depth == 0)
+ {
+ // Leaf node, the header follows actual extents
+ ext4_extent = (struct ext4_extent *)((char *)buffer + buf_offset);
+ buf_offset += sizeof(struct ext4_extent);
+ FSW_MSG_DEBUG((FSW_MSGSTR("fsw_ext4_get_by_extent: extent node cover %d...\n"), ext4_extent->ee_block));
+
+ // Is the requested block in this extent?
+ if(bno >= ext4_extent->ee_block && bno < ext4_extent->ee_block + ext4_extent->ee_len)
+ {
+ extent->phys_start = ext4_extent->ee_start_lo + (bno - ext4_extent->ee_block);
+ extent->log_count = ext4_extent->ee_len - (bno - ext4_extent->ee_block);
+ return FSW_SUCCESS;
+ }
+ }
+ else
+ {
+ FSW_MSG_DEBUG((FSW_MSGSTR("fsw_ext4_get_by_extent: index extents, depth %d\n"),
+ ext4_extent_header->eh_depth));
+ ext4_extent_idx = (struct ext4_extent_idx *)((char *)buffer + buf_offset);
+ buf_offset += sizeof(struct ext4_extent_idx);
+
+ FSW_MSG_DEBUG((FSW_MSGSTR("fsw_ext4_get_by_extent: index node covers block %d...\n"),
+ ext4_extent_idx->ei_block));
+ if(bno >= ext4_extent_idx->ei_block)
+ {
+ // Follow extent tree...
+ status = fsw_block_get(vol, ext4_extent_idx->ei_leaf_lo, 1, (void **)&buffer);
+ if (status)
+ return status;
+ buf_offset = 0;
+ break;
+ }
+ }
+ }
+ }
+
+ return FSW_NOT_FOUND;
+}
+
+/**
+ * The ext2/ext3 file system does not use extents, but stores a list of block numbers
+ * using the usual direct, indirect, double-indirect, triple-indirect scheme. To
+ * optimize access, this function checks if the following file blocks are mapped
+ * to consecutive disk blocks and returns a combined extent if possible.
+ */
+static fsw_status_t fsw_ext4_get_by_blkaddr(struct fsw_ext4_volume *vol, struct fsw_ext4_dnode *dno,
+ struct fsw_extent *extent)
+{
+ fsw_status_t status;
+ fsw_u32 bno, release_bno, buf_bcnt, file_bcnt;
+ int path[5], i;
+ fsw_u32 *buffer;
bno = extent->log_start;
// try direct block pointers in the inode
}
}
}
-
+
// follow the indirection path
buffer = dno->raw->i_block;
buf_bcnt = EXT4_NDIR_BLOCKS;
// Preconditions: The caller has checked that dno is a directory node. The caller
// has opened a storage handle to the directory's storage and keeps it around between
// calls.
+ FSW_MSG_DEBUG((FSW_MSGSTR("fsw_ext4_dir_read: started reading dir\n")));
while (1) {
// read next entry