diff options
author | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-07 14:30:35 +0000 |
---|---|---|
committer | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-07 14:30:35 +0000 |
commit | 378c18e5f024ac5a8aef4cb40d7c9aa9633d144c (patch) | |
tree | 44dfb6ca500d32cabd450649b322a42e70a30683 /libblkid/src/superblocks/befs.c | |
parent | Initial commit. (diff) | |
download | util-linux-upstream.tar.xz util-linux-upstream.zip |
Adding upstream version 2.38.1.upstream/2.38.1upstream
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to '')
-rw-r--r-- | libblkid/src/superblocks/befs.c | 544 |
1 files changed, 544 insertions, 0 deletions
diff --git a/libblkid/src/superblocks/befs.c b/libblkid/src/superblocks/befs.c new file mode 100644 index 0000000..211feae --- /dev/null +++ b/libblkid/src/superblocks/befs.c @@ -0,0 +1,544 @@ +/* + * Copyright (C) 2010 Jeroen Oortwijn <oortwijn@gmail.com> + * + * Partly based on the Haiku BFS driver by + * Axel Dörfler <axeld@pinc-software.de> + * + * Also inspired by the Linux BeFS driver by + * Will Dyson <will_dyson@pobox.com>, et al. + * + * This file may be redistributed under the terms of the + * GNU Lesser General Public License. + */ +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> +#include <string.h> +#include <inttypes.h> + +#include "superblocks.h" + +#define B_OS_NAME_LENGTH 0x20 +#define SUPER_BLOCK_MAGIC1 0x42465331 /* BFS1 */ +#define SUPER_BLOCK_MAGIC2 0xdd121031 +#define SUPER_BLOCK_MAGIC3 0x15b6830e +#define SUPER_BLOCK_FS_ENDIAN 0x42494745 /* BIGE */ +#define INODE_MAGIC1 0x3bbe0ad9 +#define BPLUSTREE_MAGIC 0x69f6c2e8 +#define BPLUSTREE_NULL -1LL +#define NUM_DIRECT_BLOCKS 12 +#define B_UINT64_TYPE 0x554c4c47 /* ULLG */ +#define KEY_NAME "be:volume_id" +#define KEY_SIZE 8 + +#define FS16_TO_CPU(value, fs_is_le) (fs_is_le ? le16_to_cpu(value) \ + : be16_to_cpu(value)) +#define FS32_TO_CPU(value, fs_is_le) (fs_is_le ? le32_to_cpu(value) \ + : be32_to_cpu(value)) +#define FS64_TO_CPU(value, fs_is_le) (fs_is_le ? le64_to_cpu(value) \ + : be64_to_cpu(value)) + +typedef struct block_run { + int32_t allocation_group; + uint16_t start; + uint16_t len; +} __attribute__((packed)) block_run, inode_addr; + +struct befs_super_block { + char name[B_OS_NAME_LENGTH]; + int32_t magic1; + int32_t fs_byte_order; + uint32_t block_size; + uint32_t block_shift; + int64_t num_blocks; + int64_t used_blocks; + int32_t inode_size; + int32_t magic2; + int32_t blocks_per_ag; + int32_t ag_shift; + int32_t num_ags; + int32_t flags; + block_run log_blocks; + int64_t log_start; + int64_t log_end; + int32_t magic3; + inode_addr root_dir; + inode_addr indices; + int32_t pad[8]; +} __attribute__((packed)); + +typedef struct data_stream { + block_run direct[NUM_DIRECT_BLOCKS]; + int64_t max_direct_range; + block_run indirect; + int64_t max_indirect_range; + block_run double_indirect; + int64_t max_double_indirect_range; + int64_t size; +} __attribute__((packed)) data_stream; + +struct befs_inode { + int32_t magic1; + inode_addr inode_num; + int32_t uid; + int32_t gid; + int32_t mode; + int32_t flags; + int64_t create_time; + int64_t last_modified_time; + inode_addr parent; + inode_addr attributes; + uint32_t type; + int32_t inode_size; + uint32_t etc; + data_stream data; + int32_t pad[4]; + int32_t small_data[0]; +} __attribute__((packed)); + +struct small_data { + uint32_t type; + uint16_t name_size; + uint16_t data_size; + char name[0]; +} __attribute__((packed)); + +struct bplustree_header { + uint32_t magic; + uint32_t node_size; + uint32_t max_number_of_levels; + uint32_t data_type; + int64_t root_node_pointer; + int64_t free_node_pointer; + int64_t maximum_size; +} __attribute__((packed)); + +struct bplustree_node { + int64_t left_link; + int64_t right_link; + int64_t overflow_link; + uint16_t all_key_count; + uint16_t all_key_length; + char name[0]; +} __attribute__((packed)); + +static unsigned char *get_block_run(blkid_probe pr, const struct befs_super_block *bs, + const struct block_run *br, int fs_le) +{ + return blkid_probe_get_buffer(pr, + ((uint64_t) FS32_TO_CPU(br->allocation_group, fs_le) + << FS32_TO_CPU(bs->ag_shift, fs_le) + << FS32_TO_CPU(bs->block_shift, fs_le)) + + ((uint64_t) FS16_TO_CPU(br->start, fs_le) + << FS32_TO_CPU(bs->block_shift, fs_le)), + (uint64_t) FS16_TO_CPU(br->len, fs_le) + << FS32_TO_CPU(bs->block_shift, fs_le)); +} + +static unsigned char *get_custom_block_run(blkid_probe pr, + const struct befs_super_block *bs, + const struct block_run *br, + int64_t offset, uint32_t length, int fs_le) +{ + if (offset + length > (int64_t) FS16_TO_CPU(br->len, fs_le) + << FS32_TO_CPU(bs->block_shift, fs_le)) + return NULL; + + return blkid_probe_get_buffer(pr, + ((uint64_t) FS32_TO_CPU(br->allocation_group, fs_le) + << FS32_TO_CPU(bs->ag_shift, fs_le) + << FS32_TO_CPU(bs->block_shift, fs_le)) + + ((uint64_t) FS16_TO_CPU(br->start, fs_le) + << FS32_TO_CPU(bs->block_shift, fs_le)) + + offset, + length); +} + +static unsigned char *get_tree_node(blkid_probe pr, const struct befs_super_block *bs, + const struct data_stream *ds, + int64_t start, uint32_t length, int fs_le) +{ + if (start < (int64_t) FS64_TO_CPU(ds->max_direct_range, fs_le)) { + int64_t br_len; + size_t i; + + for (i = 0; i < NUM_DIRECT_BLOCKS; i++) { + br_len = (int64_t) FS16_TO_CPU(ds->direct[i].len, fs_le) + << FS32_TO_CPU(bs->block_shift, fs_le); + if (start < br_len) + return get_custom_block_run(pr, bs, + &ds->direct[i], + start, length, fs_le); + start -= br_len; + } + } else if (start < (int64_t) FS64_TO_CPU(ds->max_indirect_range, fs_le)) { + struct block_run *br; + int64_t max_br, br_len, i; + + start -= FS64_TO_CPU(ds->max_direct_range, fs_le); + max_br = ((int64_t) FS16_TO_CPU(ds->indirect.len, fs_le) + << FS32_TO_CPU(bs->block_shift, fs_le)) + / sizeof(struct block_run); + + br = (struct block_run *) get_block_run(pr, bs, &ds->indirect, + fs_le); + if (!br) + return NULL; + + for (i = 0; i < max_br; i++) { + br_len = (int64_t) FS16_TO_CPU(br[i].len, fs_le) + << FS32_TO_CPU(bs->block_shift, fs_le); + if (start < br_len) + return get_custom_block_run(pr, bs, &br[i], + start, length, fs_le); + start -= br_len; + } + } else if (start < (int64_t) FS64_TO_CPU(ds->max_double_indirect_range, fs_le)) { + struct block_run *br; + int64_t max_br, di_br_size, br_per_di_br, di_index, i_index; + + start -= (int64_t) FS64_TO_CPU(ds->max_indirect_range, fs_le); + + di_br_size = (int64_t) FS16_TO_CPU(ds->double_indirect.len, + fs_le) << FS32_TO_CPU(bs->block_shift, fs_le); + if (di_br_size == 0) + return NULL; + + br_per_di_br = di_br_size / sizeof(struct block_run); + if (br_per_di_br == 0) + return NULL; + + di_index = start / (br_per_di_br * di_br_size); + i_index = (start % (br_per_di_br * di_br_size)) / di_br_size; + start = (start % (br_per_di_br * di_br_size)) % di_br_size; + + if (di_index >= br_per_di_br) + return NULL; /* Corrupt? */ + + br = (struct block_run *) get_block_run(pr, bs, + &ds->double_indirect, fs_le); + if (!br) + return NULL; + + max_br = ((int64_t)FS16_TO_CPU(br[di_index].len, fs_le) + << FS32_TO_CPU(bs->block_shift, fs_le)) + / sizeof(struct block_run); + + if (i_index >= max_br) + return NULL; /* Corrupt? */ + + br = (struct block_run *) get_block_run(pr, bs, &br[di_index], + fs_le); + if (!br) + return NULL; + + return get_custom_block_run(pr, bs, &br[i_index], start, length, + fs_le); + } + return NULL; +} + +#define BAD_KEYS -2 + +static int32_t compare_keys(const char keys1[], uint16_t keylengths1[], + int32_t index, const char *key2, + uint16_t keylength2, uint16_t all_key_length, + int fs_le) +{ + const char *key1; + uint16_t keylength1, keystart1; + int32_t result; + + keystart1 = index == 0 ? 0 : FS16_TO_CPU(keylengths1[index - 1], fs_le); + keylength1 = FS16_TO_CPU(keylengths1[index], fs_le) - keystart1; + + if (keystart1 + keylength1 > all_key_length) + return BAD_KEYS; /* Corrupt? */ + + key1 = &keys1[keystart1]; + + result = strncmp(key1, key2, min(keylength1, keylength2)); + + if (result == 0) + return keylength1 - keylength2; + + if (result < 0) /* Don't clash with BAD_KEYS */ + result = -1; + + return result; +} + +static int64_t get_key_value(blkid_probe pr, const struct befs_super_block *bs, + const struct befs_inode *bi, const char *key, int fs_le) +{ + struct bplustree_header *bh; + struct bplustree_node *bn; + uint16_t *keylengths; + int64_t *values; + int64_t node_pointer; + uint32_t bn_size, all_key_count, all_key_length; + uint32_t keylengths_offset, values_offset; + int32_t first, last, mid, cmp; + int loop_detect = 0; + + errno = 0; + bh = (struct bplustree_header *) get_tree_node(pr, bs, &bi->data, 0, + sizeof(struct bplustree_header), fs_le); + if (!bh) + return errno ? -errno : -ENOENT; + + if ((int32_t) FS32_TO_CPU(bh->magic, fs_le) != BPLUSTREE_MAGIC) + return -ENOENT; + + node_pointer = FS64_TO_CPU(bh->root_node_pointer, fs_le); + bn_size = FS32_TO_CPU(bh->node_size, fs_le); + + if (bn_size < sizeof(struct bplustree_node)) + return -ENOENT; /* Corrupt? */ + + do { + errno = 0; + + bn = (struct bplustree_node *) get_tree_node(pr, bs, &bi->data, + node_pointer, bn_size, fs_le); + if (!bn) + return errno ? -errno : -ENOENT; + + all_key_count = FS16_TO_CPU(bn->all_key_count, fs_le); + all_key_length = FS16_TO_CPU(bn->all_key_length, fs_le); + keylengths_offset = + (sizeof(struct bplustree_node) + all_key_length + + sizeof(int64_t) - 1) & ~(sizeof(int64_t) - 1); + values_offset = keylengths_offset + + all_key_count * sizeof(uint16_t); + + if (values_offset + all_key_count * sizeof(uint64_t) > bn_size) + return -ENOENT; /* Corrupt? */ + + keylengths = (uint16_t *) ((uint8_t *) bn + keylengths_offset); + values = (int64_t *) ((uint8_t *) bn + values_offset); + + first = 0; + mid = 0; + last = all_key_count - 1; + + cmp = compare_keys(bn->name, keylengths, last, key, + strlen(key), all_key_length, fs_le); + if (cmp == BAD_KEYS) + return -ENOENT; + + if (cmp == 0) { + if ((int64_t) FS64_TO_CPU(bn->overflow_link, fs_le) + == BPLUSTREE_NULL) + return FS64_TO_CPU(values[last], fs_le); + + node_pointer = FS64_TO_CPU(values[last], fs_le); + } else if (cmp < 0) + node_pointer = FS64_TO_CPU(bn->overflow_link, fs_le); + else { + while (first <= last) { + mid = (first + last) / 2; + + cmp = compare_keys(bn->name, keylengths, mid, + key, strlen(key), + all_key_length, fs_le); + if (cmp == BAD_KEYS) + return -ENOENT; + + if (cmp == 0) { + if ((int64_t) FS64_TO_CPU(bn->overflow_link, + fs_le) == BPLUSTREE_NULL) + return FS64_TO_CPU(values[mid], + fs_le); + break; + } + + if (cmp < 0) + first = mid + 1; + else + last = mid - 1; + } + if (cmp < 0) + node_pointer = FS64_TO_CPU(values[mid + 1], + fs_le); + else + node_pointer = FS64_TO_CPU(values[mid], fs_le); + } + } while (++loop_detect < 100 && + (int64_t) FS64_TO_CPU(bn->overflow_link, fs_le) + != BPLUSTREE_NULL); + return 0; +} + +static int get_uuid(blkid_probe pr, const struct befs_super_block *bs, + uint64_t * const uuid, int fs_le) +{ + struct befs_inode *bi; + struct small_data *sd; + uint64_t bi_size, offset, sd_size, sd_total_size; + + bi = (struct befs_inode *) get_block_run(pr, bs, &bs->root_dir, fs_le); + if (!bi) + return errno ? -errno : BLKID_PROBE_NONE; + + if (FS32_TO_CPU(bi->magic1, fs_le) != INODE_MAGIC1) + return BLKID_PROBE_NONE; + + bi_size = (uint64_t)FS16_TO_CPU(bs->root_dir.len, fs_le) << + FS32_TO_CPU(bs->block_shift, fs_le); + sd_total_size = min(bi_size - sizeof(struct befs_inode), + (uint64_t)FS32_TO_CPU(bi->inode_size, fs_le)); + + offset = 0; + + while (offset + sizeof(struct small_data) <= sd_total_size) { + sd = (struct small_data *) ((uint8_t *)bi->small_data + offset); + sd_size = sizeof(struct small_data) + + FS16_TO_CPU(sd->name_size, fs_le) + 3 + + FS16_TO_CPU(sd->data_size, fs_le) + 1; + + if (offset + sd_size > sd_total_size) + break; + + if (FS32_TO_CPU(sd->type, fs_le) == B_UINT64_TYPE + && FS16_TO_CPU(sd->name_size, fs_le) == strlen(KEY_NAME) + && FS16_TO_CPU(sd->data_size, fs_le) == KEY_SIZE + && strcmp(sd->name, KEY_NAME) == 0) { + + memcpy(uuid, + sd->name + FS16_TO_CPU(sd->name_size, fs_le) + 3, + sizeof(uint64_t)); + + break; + } + + if (FS32_TO_CPU(sd->type, fs_le) == 0 + && FS16_TO_CPU(sd->name_size, fs_le) == 0 + && FS16_TO_CPU(sd->data_size, fs_le) == 0) + break; + + offset += sd_size; + } + + if (*uuid == 0 + && (FS32_TO_CPU(bi->attributes.allocation_group, fs_le) != 0 + || FS16_TO_CPU(bi->attributes.start, fs_le) != 0 + || FS16_TO_CPU(bi->attributes.len, fs_le) != 0)) { + int64_t value; + + bi = (struct befs_inode *) get_block_run(pr, bs, + &bi->attributes, fs_le); + if (!bi) + return errno ? -errno : BLKID_PROBE_NONE; + + if (FS32_TO_CPU(bi->magic1, fs_le) != INODE_MAGIC1) + return BLKID_PROBE_NONE; + + value = get_key_value(pr, bs, bi, KEY_NAME, fs_le); + if (value < 0) + return value == -ENOENT ? BLKID_PROBE_NONE : value; + + if (value > 0) { + bi = (struct befs_inode *) blkid_probe_get_buffer(pr, + value << FS32_TO_CPU(bs->block_shift, fs_le), + FS32_TO_CPU(bs->block_size, fs_le)); + if (!bi) + return errno ? -errno : BLKID_PROBE_NONE; + + if (FS32_TO_CPU(bi->magic1, fs_le) != INODE_MAGIC1) + return 1; + + if (FS32_TO_CPU(bi->type, fs_le) == B_UINT64_TYPE + && FS64_TO_CPU(bi->data.size, fs_le) == KEY_SIZE + && FS16_TO_CPU(bi->data.direct[0].len, fs_le) + == 1) { + uint64_t *attr_data; + + attr_data = (uint64_t *) get_block_run(pr, bs, + &bi->data.direct[0], fs_le); + if (!attr_data) + return errno ? -errno : BLKID_PROBE_NONE; + + *uuid = *attr_data; + } + } + } + return 0; +} + +static int probe_befs(blkid_probe pr, const struct blkid_idmag *mag) +{ + struct befs_super_block *bs; + const char *version = NULL; + uint64_t volume_id = 0; + uint32_t block_size, block_shift; + int fs_le, ret; + + bs = (struct befs_super_block *) blkid_probe_get_buffer(pr, + mag->sboff - B_OS_NAME_LENGTH, + sizeof(struct befs_super_block)); + if (!bs) + return errno ? -errno : BLKID_PROBE_NONE; + + if (le32_to_cpu(bs->magic1) == SUPER_BLOCK_MAGIC1 + && le32_to_cpu(bs->magic2) == SUPER_BLOCK_MAGIC2 + && le32_to_cpu(bs->magic3) == SUPER_BLOCK_MAGIC3 + && le32_to_cpu(bs->fs_byte_order) == SUPER_BLOCK_FS_ENDIAN) { + fs_le = 1; + version = "little-endian"; + } else if (be32_to_cpu(bs->magic1) == SUPER_BLOCK_MAGIC1 + && be32_to_cpu(bs->magic2) == SUPER_BLOCK_MAGIC2 + && be32_to_cpu(bs->magic3) == SUPER_BLOCK_MAGIC3 + && be32_to_cpu(bs->fs_byte_order) == SUPER_BLOCK_FS_ENDIAN) { + fs_le = 0; + version = "big-endian"; + } else + return BLKID_PROBE_NONE; + + block_size = FS32_TO_CPU(bs->block_size, fs_le); + block_shift = FS32_TO_CPU(bs->block_shift, fs_le); + + if (block_shift < 10 || block_shift > 13 || + block_size != 1U << block_shift) + return BLKID_PROBE_NONE; + + ret = get_uuid(pr, bs, &volume_id, fs_le); + + if (ret != 0) + return ret; + + /* + * all checks pass, set LABEL, VERSION and UUID + */ + if (*bs->name != '\0') + blkid_probe_set_label(pr, (unsigned char *) bs->name, + sizeof(bs->name)); + if (version) + blkid_probe_set_version(pr, version); + + if (volume_id) + blkid_probe_sprintf_uuid(pr, (unsigned char *) &volume_id, + sizeof(volume_id), "%016" PRIx64, + FS64_TO_CPU(volume_id, fs_le)); + + blkid_probe_set_block_size(pr, block_size); + + return BLKID_PROBE_OK; +} + +const struct blkid_idinfo befs_idinfo = +{ + .name = "befs", + .usage = BLKID_USAGE_FILESYSTEM, + .probefunc = probe_befs, + .minsz = 1024 * 1440, + .magics = { + { .magic = "BFS1", .len = 4, .sboff = B_OS_NAME_LENGTH }, + { .magic = "1SFB", .len = 4, .sboff = B_OS_NAME_LENGTH }, + { .magic = "BFS1", .len = 4, .sboff = 0x200 + + B_OS_NAME_LENGTH }, + { .magic = "1SFB", .len = 4, .sboff = 0x200 + + B_OS_NAME_LENGTH }, + { NULL } + } +}; |