summaryrefslogtreecommitdiffstats
path: root/libblkid/src/superblocks/vfat.c
diff options
context:
space:
mode:
authorDaniel Baumann <daniel.baumann@progress-linux.org>2024-05-06 02:42:50 +0000
committerDaniel Baumann <daniel.baumann@progress-linux.org>2024-05-06 02:42:50 +0000
commit8cb83eee5a58b1fad74c34094ce3afb9e430b5a4 (patch)
treea9b2e7baeca1be40eb734371e3c8b11b02294497 /libblkid/src/superblocks/vfat.c
parentInitial commit. (diff)
downloadutil-linux-upstream/2.33.1.tar.xz
util-linux-upstream/2.33.1.zip
Adding upstream version 2.33.1.upstream/2.33.1upstream
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to 'libblkid/src/superblocks/vfat.c')
-rw-r--r--libblkid/src/superblocks/vfat.c464
1 files changed, 464 insertions, 0 deletions
diff --git a/libblkid/src/superblocks/vfat.c b/libblkid/src/superblocks/vfat.c
new file mode 100644
index 0000000..5061587
--- /dev/null
+++ b/libblkid/src/superblocks/vfat.c
@@ -0,0 +1,464 @@
+/*
+ * Copyright (C) 1999 by Andries Brouwer
+ * Copyright (C) 1999, 2000, 2003 by Theodore Ts'o
+ * Copyright (C) 2001 by Andreas Dilger
+ * Copyright (C) 2004 Kay Sievers <kay.sievers@vrfy.org>
+ * Copyright (C) 2008 Karel Zak <kzak@redhat.com>
+ *
+ * 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 <errno.h>
+#include <ctype.h>
+#include <stdint.h>
+#include <inttypes.h>
+
+#include "pt-mbr.h"
+#include "superblocks.h"
+
+/* Yucky misaligned values */
+struct vfat_super_block {
+/* 00*/ unsigned char vs_ignored[3];
+/* 03*/ unsigned char vs_sysid[8];
+/* 0b*/ unsigned char vs_sector_size[2];
+/* 0d*/ uint8_t vs_cluster_size;
+/* 0e*/ uint16_t vs_reserved;
+/* 10*/ uint8_t vs_fats;
+/* 11*/ unsigned char vs_dir_entries[2];
+/* 13*/ unsigned char vs_sectors[2];
+/* 15*/ unsigned char vs_media;
+/* 16*/ uint16_t vs_fat_length;
+/* 18*/ uint16_t vs_secs_track;
+/* 1a*/ uint16_t vs_heads;
+/* 1c*/ uint32_t vs_hidden;
+/* 20*/ uint32_t vs_total_sect;
+/* 24*/ uint32_t vs_fat32_length;
+/* 28*/ uint16_t vs_flags;
+/* 2a*/ uint8_t vs_version[2];
+/* 2c*/ uint32_t vs_root_cluster;
+/* 30*/ uint16_t vs_fsinfo_sector;
+/* 32*/ uint16_t vs_backup_boot;
+/* 34*/ uint16_t vs_reserved2[6];
+/* 40*/ unsigned char vs_unknown[3];
+/* 43*/ unsigned char vs_serno[4];
+/* 47*/ unsigned char vs_label[11];
+/* 52*/ unsigned char vs_magic[8];
+/* 5a*/ unsigned char vs_dummy2[0x1fe - 0x5a];
+/*1fe*/ unsigned char vs_pmagic[2];
+} __attribute__((packed));
+
+/* Yucky misaligned values */
+struct msdos_super_block {
+/* 00*/ unsigned char ms_ignored[3];
+/* 03*/ unsigned char ms_sysid[8];
+/* 0b*/ unsigned char ms_sector_size[2];
+/* 0d*/ uint8_t ms_cluster_size;
+/* 0e*/ uint16_t ms_reserved;
+/* 10*/ uint8_t ms_fats;
+/* 11*/ unsigned char ms_dir_entries[2];
+/* 13*/ unsigned char ms_sectors[2]; /* =0 iff V3 or later */
+/* 15*/ unsigned char ms_media;
+/* 16*/ uint16_t ms_fat_length; /* Sectors per FAT */
+/* 18*/ uint16_t ms_secs_track;
+/* 1a*/ uint16_t ms_heads;
+/* 1c*/ uint32_t ms_hidden;
+/* V3 BPB */
+/* 20*/ uint32_t ms_total_sect; /* iff ms_sectors == 0 */
+/* V4 BPB */
+/* 24*/ unsigned char ms_unknown[3]; /* Phys drive no., resvd, V4 sig (0x29) */
+/* 27*/ unsigned char ms_serno[4];
+/* 2b*/ unsigned char ms_label[11];
+/* 36*/ unsigned char ms_magic[8];
+/* 3e*/ unsigned char ms_dummy2[0x1fe - 0x3e];
+/*1fe*/ unsigned char ms_pmagic[2];
+} __attribute__((packed));
+
+struct vfat_dir_entry {
+ uint8_t name[11];
+ uint8_t attr;
+ uint16_t time_creat;
+ uint16_t date_creat;
+ uint16_t time_acc;
+ uint16_t date_acc;
+ uint16_t cluster_high;
+ uint16_t time_write;
+ uint16_t date_write;
+ uint16_t cluster_low;
+ uint32_t size;
+} __attribute__((packed));
+
+struct fat32_fsinfo {
+ uint8_t signature1[4];
+ uint32_t reserved1[120];
+ uint8_t signature2[4];
+ uint32_t free_clusters;
+ uint32_t next_cluster;
+ uint32_t reserved2[4];
+} __attribute__((packed));
+
+/* maximum number of clusters */
+#define FAT12_MAX 0xFF4
+#define FAT16_MAX 0xFFF4
+#define FAT32_MAX 0x0FFFFFF6
+
+#define FAT_ATTR_VOLUME_ID 0x08
+#define FAT_ATTR_DIR 0x10
+#define FAT_ATTR_LONG_NAME 0x0f
+#define FAT_ATTR_MASK 0x3f
+#define FAT_ENTRY_FREE 0xe5
+
+static const char *no_name = "NO NAME ";
+
+#define unaligned_le16(x) \
+ (((unsigned char *) x)[0] + (((unsigned char *) x)[1] << 8))
+
+/*
+ * Look for LABEL (name) in the FAT root directory.
+ */
+static unsigned char *search_fat_label(blkid_probe pr,
+ uint64_t offset, uint32_t entries)
+{
+ struct vfat_dir_entry *ent, *dir = NULL;
+ uint32_t i;
+
+ DBG(LOWPROBE, ul_debug("\tlook for label in root-dir "
+ "(entries: %"PRIu32", offset: %"PRIu64")", entries, offset));
+
+ if (!blkid_probe_is_tiny(pr)) {
+ /* large disk, read whole root directory */
+ dir = (struct vfat_dir_entry *)
+ blkid_probe_get_buffer(pr,
+ offset,
+ (uint64_t) entries *
+ sizeof(struct vfat_dir_entry));
+ if (!dir)
+ return NULL;
+ }
+
+ for (i = 0; i < entries; i++) {
+ /*
+ * The root directory could be relatively large (4-16kB).
+ * Fortunately, the LABEL is usually the first entry in the
+ * directory. On tiny disks we call read() per entry.
+ */
+ if (!dir)
+ ent = (struct vfat_dir_entry *)
+ blkid_probe_get_buffer(pr,
+ (uint64_t) offset + (i *
+ sizeof(struct vfat_dir_entry)),
+ sizeof(struct vfat_dir_entry));
+ else
+ ent = &dir[i];
+
+ if (!ent || ent->name[0] == 0x00)
+ break;
+
+ if ((ent->name[0] == FAT_ENTRY_FREE) ||
+ (ent->cluster_high != 0 || ent->cluster_low != 0) ||
+ ((ent->attr & FAT_ATTR_MASK) == FAT_ATTR_LONG_NAME))
+ continue;
+
+ if ((ent->attr & (FAT_ATTR_VOLUME_ID | FAT_ATTR_DIR)) ==
+ FAT_ATTR_VOLUME_ID) {
+ DBG(LOWPROBE, ul_debug("\tfound fs LABEL at entry %d", i));
+ if (ent->name[0] == 0x05)
+ ent->name[0] = 0xE5;
+ return ent->name;
+ }
+ }
+ return NULL;
+}
+
+static int fat_valid_superblock(blkid_probe pr,
+ const struct blkid_idmag *mag,
+ struct msdos_super_block *ms,
+ struct vfat_super_block *vs,
+ uint32_t *cluster_count, uint32_t *fat_size)
+{
+ uint16_t sector_size, dir_entries, reserved;
+ uint32_t sect_count, __fat_size, dir_size, __cluster_count, fat_length;
+ uint32_t max_count;
+
+ /* extra check for FATs without magic strings */
+ if (mag->len <= 2) {
+ /* Old floppies have a valid MBR signature */
+ if (ms->ms_pmagic[0] != 0x55 || ms->ms_pmagic[1] != 0xAA)
+ return 0;
+
+ /*
+ * OS/2 and apparently DFSee will place a FAT12/16-like
+ * pseudo-superblock in the first 512 bytes of non-FAT
+ * filesystems --- at least JFS and HPFS, and possibly others.
+ * So we explicitly check for those filesystems at the
+ * FAT12/16 filesystem magic field identifier, and if they are
+ * present, we rule this out as a FAT filesystem, despite the
+ * FAT-like pseudo-header.
+ */
+ if ((memcmp(ms->ms_magic, "JFS ", 8) == 0) ||
+ (memcmp(ms->ms_magic, "HPFS ", 8) == 0)) {
+ DBG(LOWPROBE, ul_debug("\tJFS/HPFS detected"));
+ return 0;
+ }
+ }
+
+ /* fat counts(Linux kernel expects at least 1 FAT table) */
+ if (!ms->ms_fats)
+ return 0;
+ if (!ms->ms_reserved)
+ return 0;
+ if (!(0xf8 <= ms->ms_media || ms->ms_media == 0xf0))
+ return 0;
+ if (!is_power_of_2(ms->ms_cluster_size))
+ return 0;
+
+ sector_size = unaligned_le16(&ms->ms_sector_size);
+ if (!is_power_of_2(sector_size) ||
+ sector_size < 512 || sector_size > 4096)
+ return 0;
+
+ dir_entries = unaligned_le16(&ms->ms_dir_entries);
+ reserved = le16_to_cpu(ms->ms_reserved);
+ sect_count = unaligned_le16(&ms->ms_sectors);
+
+ if (sect_count == 0)
+ sect_count = le32_to_cpu(ms->ms_total_sect);
+
+ fat_length = le16_to_cpu(ms->ms_fat_length);
+ if (fat_length == 0)
+ fat_length = le32_to_cpu(vs->vs_fat32_length);
+
+ __fat_size = fat_length * ms->ms_fats;
+ dir_size = ((dir_entries * sizeof(struct vfat_dir_entry)) +
+ (sector_size-1)) / sector_size;
+
+ __cluster_count = (sect_count - (reserved + __fat_size + dir_size)) /
+ ms->ms_cluster_size;
+ if (!ms->ms_fat_length && vs->vs_fat32_length)
+ max_count = FAT32_MAX;
+ else
+ max_count = __cluster_count > FAT12_MAX ? FAT16_MAX : FAT12_MAX;
+
+ if (__cluster_count > max_count)
+ return 0;
+
+ if (fat_size)
+ *fat_size = __fat_size;
+ if (cluster_count)
+ *cluster_count = __cluster_count;
+
+ if (blkid_probe_is_wholedisk(pr)) {
+ /* OK, seems like FAT, but it's possible that we found boot
+ * sector with crazy FAT-like stuff (magic strings, media,
+ * etc..) before MBR. Let's make sure that there is no MBR with
+ * usable partition. */
+ unsigned char *buf = (unsigned char *) ms;
+
+ if (mbr_is_valid_magic(buf)) {
+ struct dos_partition *p0 = mbr_get_partition(buf, 0);
+
+ if (dos_partition_get_size(p0) != 0 &&
+ (p0->boot_ind == 0 || p0->boot_ind == 0x80)) {
+ DBG(LOWPROBE, ul_debug("\tMBR detected"));
+ return 0;
+ }
+ }
+ }
+
+ if (blkid_probe_is_bitlocker(pr))
+ return 0;
+
+ return 1; /* valid */
+}
+
+/* function prototype to avoid warnings (duplicate in partitions/dos.c) */
+extern int blkid_probe_is_vfat(blkid_probe pr);
+
+/*
+ * This function is used by MBR partition table parser to avoid
+ * misinterpretation of FAT filesystem.
+ */
+int blkid_probe_is_vfat(blkid_probe pr)
+{
+ struct vfat_super_block *vs;
+ struct msdos_super_block *ms;
+ const struct blkid_idmag *mag = NULL;
+ int rc;
+
+ rc = blkid_probe_get_idmag(pr, &vfat_idinfo, NULL, &mag);
+ if (rc < 0)
+ return rc; /* error */
+ if (rc != BLKID_PROBE_OK || !mag)
+ return 0;
+
+ ms = blkid_probe_get_sb(pr, mag, struct msdos_super_block);
+ if (!ms)
+ return errno ? -errno : 0;
+ vs = blkid_probe_get_sb(pr, mag, struct vfat_super_block);
+ if (!vs)
+ return errno ? -errno : 0;
+
+ return fat_valid_superblock(pr, mag, ms, vs, NULL, NULL);
+}
+
+/* FAT label extraction from the root directory taken from Kay
+ * Sievers's volume_id library */
+static int probe_vfat(blkid_probe pr, const struct blkid_idmag *mag)
+{
+ struct vfat_super_block *vs;
+ struct msdos_super_block *ms;
+ const unsigned char *vol_label = NULL;
+ const unsigned char *boot_label = NULL;
+ unsigned char *vol_serno = NULL, vol_label_buf[11];
+ uint16_t sector_size = 0, reserved;
+ uint32_t cluster_count, fat_size;
+ const char *version = NULL;
+
+ ms = blkid_probe_get_sb(pr, mag, struct msdos_super_block);
+ if (!ms)
+ return errno ? -errno : 1;
+
+ vs = blkid_probe_get_sb(pr, mag, struct vfat_super_block);
+ if (!vs)
+ return errno ? -errno : 1;
+
+ if (!fat_valid_superblock(pr, mag, ms, vs, &cluster_count, &fat_size))
+ return 1;
+
+ sector_size = unaligned_le16(&ms->ms_sector_size);
+ reserved = le16_to_cpu(ms->ms_reserved);
+
+ if (ms->ms_fat_length) {
+ /* the label may be an attribute in the root directory */
+ uint32_t root_start = (reserved + fat_size) * sector_size;
+ uint32_t root_dir_entries = unaligned_le16(&vs->vs_dir_entries);
+
+ vol_label = search_fat_label(pr, root_start, root_dir_entries);
+ if (vol_label) {
+ memcpy(vol_label_buf, vol_label, 11);
+ vol_label = vol_label_buf;
+ }
+
+ boot_label = ms->ms_label;
+ vol_serno = ms->ms_serno;
+
+ blkid_probe_set_value(pr, "SEC_TYPE", (unsigned char *) "msdos",
+ sizeof("msdos"));
+
+ if (cluster_count < FAT12_MAX)
+ version = "FAT12";
+ else if (cluster_count < FAT16_MAX)
+ version = "FAT16";
+
+ } else if (vs->vs_fat32_length) {
+ unsigned char *buf;
+ uint16_t fsinfo_sect;
+ int maxloop = 100;
+
+ /* Search the FAT32 root dir for the label attribute */
+ uint32_t buf_size = vs->vs_cluster_size * sector_size;
+ uint32_t start_data_sect = reserved + fat_size;
+ uint32_t entries = le32_to_cpu(vs->vs_fat32_length) *
+ sector_size / sizeof(uint32_t);
+ uint32_t next = le32_to_cpu(vs->vs_root_cluster);
+
+ while (next && next < entries && --maxloop) {
+ uint32_t next_sect_off;
+ uint64_t next_off, fat_entry_off;
+ int count;
+
+ next_sect_off = (next - 2) * vs->vs_cluster_size;
+ next_off = (uint64_t)(start_data_sect + next_sect_off) *
+ sector_size;
+
+ count = buf_size / sizeof(struct vfat_dir_entry);
+
+ vol_label = search_fat_label(pr, next_off, count);
+ if (vol_label) {
+ memcpy(vol_label_buf, vol_label, 11);
+ vol_label = vol_label_buf;
+ break;
+ }
+
+ /* get FAT entry */
+ fat_entry_off = ((uint64_t) reserved * sector_size) +
+ (next * sizeof(uint32_t));
+ buf = blkid_probe_get_buffer(pr, fat_entry_off, buf_size);
+ if (buf == NULL)
+ break;
+
+ /* set next cluster */
+ next = le32_to_cpu(*((uint32_t *) buf)) & 0x0fffffff;
+ }
+
+ version = "FAT32";
+
+ boot_label = vs->vs_label;
+ vol_serno = vs->vs_serno;
+
+ /*
+ * FAT32 should have a valid signature in the fsinfo block,
+ * but also allow all bytes set to '\0', because some volumes
+ * do not set the signature at all.
+ */
+ fsinfo_sect = le16_to_cpu(vs->vs_fsinfo_sector);
+ if (fsinfo_sect) {
+ struct fat32_fsinfo *fsinfo;
+
+ buf = blkid_probe_get_buffer(pr,
+ (uint64_t) fsinfo_sect * sector_size,
+ sizeof(struct fat32_fsinfo));
+ if (buf == NULL)
+ return errno ? -errno : 1;
+
+ fsinfo = (struct fat32_fsinfo *) buf;
+ if (memcmp(fsinfo->signature1, "\x52\x52\x61\x41", 4) != 0 &&
+ memcmp(fsinfo->signature1, "\x52\x52\x64\x41", 4) != 0 &&
+ memcmp(fsinfo->signature1, "\x00\x00\x00\x00", 4) != 0)
+ return 1;
+ if (memcmp(fsinfo->signature2, "\x72\x72\x41\x61", 4) != 0 &&
+ memcmp(fsinfo->signature2, "\x00\x00\x00\x00", 4) != 0)
+ return 1;
+ }
+ }
+
+ if (boot_label && memcmp(boot_label, no_name, 11))
+ blkid_probe_set_id_label(pr, "LABEL_FATBOOT", boot_label, 11);
+
+ if (vol_label)
+ blkid_probe_set_label(pr, vol_label, 11);
+
+ /* We can't just print them as %04X, because they are unaligned */
+ if (vol_serno)
+ blkid_probe_sprintf_uuid(pr, vol_serno, 4, "%02X%02X-%02X%02X",
+ vol_serno[3], vol_serno[2], vol_serno[1], vol_serno[0]);
+ if (version)
+ blkid_probe_set_version(pr, version);
+
+ return 0;
+}
+
+
+const struct blkid_idinfo vfat_idinfo =
+{
+ .name = "vfat",
+ .usage = BLKID_USAGE_FILESYSTEM,
+ .probefunc = probe_vfat,
+ .magics =
+ {
+ { .magic = "MSWIN", .len = 5, .sboff = 0x52 },
+ { .magic = "FAT32 ", .len = 8, .sboff = 0x52 },
+ { .magic = "MSDOS", .len = 5, .sboff = 0x36 },
+ { .magic = "FAT16 ", .len = 8, .sboff = 0x36 },
+ { .magic = "FAT12 ", .len = 8, .sboff = 0x36 },
+ { .magic = "FAT ", .len = 8, .sboff = 0x36 },
+ { .magic = "\353", .len = 1, },
+ { .magic = "\351", .len = 1, },
+ { .magic = "\125\252", .len = 2, .sboff = 0x1fe },
+ { NULL }
+ }
+};
+