fs/fat/: add support for DOS 1.x formatted volumes
authorConrad Meyer <cse.cem@gmail.com>
Fri, 6 Jun 2014 21:36:37 +0000 (14:36 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 6 Jun 2014 23:08:10 +0000 (16:08 -0700)
Add structure for parsed BPB information, struct fat_bios_param_block,
and move all of the deserialization and validation logic from
fat_fill_super() into fat_read_bpb().

Add a 'dos1xfloppy' mount option to infer DOS 2.x BIOS Parameter Block
defaults from block device geometry for ancient floppies and floppy
images, as a fall-back from the default BPB parsing logic.

When fat_read_bpb() finds an invalid FAT filesystem and dos1xfloppy is
set, fall back to fat_read_static_bpb().  fat_read_static_bpb()
validates that the entire BPB is zero, and that the floppy has a
DOS-style 8086 code bootstrapping header.  Then it fills in default BPB
values from media size and a table.[0]

Media size is assumed to be static for archaic FAT volumes.  See also:
[1].

Fixes kernel.org bug #42617.

[0]: https://en.wikipedia.org/wiki/File_Allocation_Table#Exceptions
[1]: http://www.win.tue.nl/~aeb/linux/fs/fat/fat-1.html

[hirofumi@mail.parknet.co.jp: fix missed error code]
Signed-off-by: Conrad Meyer <cse.cem@gmail.com>
Acked-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Tested-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Documentation/filesystems/vfat.txt
fs/fat/fat.h
fs/fat/inode.c

index 4a93e98b290a085793c495c9635b97335c58da56..ce1126aceed8fc739aba64b1ff61ae3ce7cbc90b 100644 (file)
@@ -172,6 +172,11 @@ nfs=stale_rw|nostale_ro
                To maintain backward compatibility, '-o nfs' is also accepted,
                defaulting to stale_rw
 
+dos1xfloppy  -- If set, use a fallback default BIOS Parameter Block
+               configuration, determined by backing device size. These static
+               parameters match defaults assumed by DOS 1.x for 160 kiB,
+               180 kiB, 320 kiB, and 360 kiB floppies and floppy images.
+
 
 <bool>: 0,1,yes,no,true,false
 
index 7c31f4bc74a9d3e2e358baa27688e329b77d9fc5..e0c4ba39a377b407bb19c50f830443ab4f7f3188 100644 (file)
@@ -52,7 +52,8 @@ struct fat_mount_options {
                 usefree:1,        /* Use free_clusters for FAT32 */
                 tz_set:1,         /* Filesystem timestamps' offset set */
                 rodir:1,          /* allow ATTR_RO for directory */
-                discard:1;        /* Issue discard requests on deletions */
+                discard:1,        /* Issue discard requests on deletions */
+                dos1xfloppy:1;    /* Assume default BPB for DOS 1.x floppies */
 };
 
 #define FAT_HASH_BITS  8
index b3361fe2bcb590183150fac820d6ffbc1f46e9f9..babff0f40696aea5654505544c95e07653eca4ce 100644 (file)
 #define CONFIG_FAT_DEFAULT_IOCHARSET   ""
 #endif
 
+#define KB_IN_SECTORS 2
+
+/*
+ * A deserialized copy of the on-disk structure laid out in struct
+ * fat_boot_sector.
+ */
+struct fat_bios_param_block {
+       u16     fat_sector_size;
+       u8      fat_sec_per_clus;
+       u16     fat_reserved;
+       u8      fat_fats;
+       u16     fat_dir_entries;
+       u16     fat_sectors;
+       u16     fat_fat_length;
+       u32     fat_total_sect;
+
+       u8      fat16_state;
+       u32     fat16_vol_id;
+
+       u32     fat32_length;
+       u32     fat32_root_cluster;
+       u16     fat32_info_sector;
+       u8      fat32_state;
+       u32     fat32_vol_id;
+};
+
 static int fat_default_codepage = CONFIG_FAT_DEFAULT_CODEPAGE;
 static char fat_default_iocharset[] = CONFIG_FAT_DEFAULT_IOCHARSET;
 
+static struct fat_floppy_defaults {
+       unsigned nr_sectors;
+       unsigned sec_per_clus;
+       unsigned dir_entries;
+       unsigned media;
+       unsigned fat_length;
+} floppy_defaults[] = {
+{
+       .nr_sectors = 160 * KB_IN_SECTORS,
+       .sec_per_clus = 1,
+       .dir_entries = 64,
+       .media = 0xFE,
+       .fat_length = 1,
+},
+{
+       .nr_sectors = 180 * KB_IN_SECTORS,
+       .sec_per_clus = 1,
+       .dir_entries = 64,
+       .media = 0xFC,
+       .fat_length = 2,
+},
+{
+       .nr_sectors = 320 * KB_IN_SECTORS,
+       .sec_per_clus = 2,
+       .dir_entries = 112,
+       .media = 0xFF,
+       .fat_length = 1,
+},
+{
+       .nr_sectors = 360 * KB_IN_SECTORS,
+       .sec_per_clus = 2,
+       .dir_entries = 112,
+       .media = 0xFD,
+       .fat_length = 2,
+},
+};
 
 static int fat_add_cluster(struct inode *inode)
 {
@@ -853,6 +915,8 @@ static int fat_show_options(struct seq_file *m, struct dentry *root)
                seq_puts(m, ",nfs=stale_rw");
        if (opts->discard)
                seq_puts(m, ",discard");
+       if (opts->dos1xfloppy)
+               seq_puts(m, ",dos1xfloppy");
 
        return 0;
 }
@@ -867,7 +931,7 @@ enum {
        Opt_uni_xl_no, Opt_uni_xl_yes, Opt_nonumtail_no, Opt_nonumtail_yes,
        Opt_obsolete, Opt_flush, Opt_tz_utc, Opt_rodir, Opt_err_cont,
        Opt_err_panic, Opt_err_ro, Opt_discard, Opt_nfs, Opt_time_offset,
-       Opt_nfs_stale_rw, Opt_nfs_nostale_ro, Opt_err,
+       Opt_nfs_stale_rw, Opt_nfs_nostale_ro, Opt_err, Opt_dos1xfloppy,
 };
 
 static const match_table_t fat_tokens = {
@@ -900,6 +964,7 @@ static const match_table_t fat_tokens = {
        {Opt_nfs_stale_rw, "nfs"},
        {Opt_nfs_stale_rw, "nfs=stale_rw"},
        {Opt_nfs_nostale_ro, "nfs=nostale_ro"},
+       {Opt_dos1xfloppy, "dos1xfloppy"},
        {Opt_obsolete, "conv=binary"},
        {Opt_obsolete, "conv=text"},
        {Opt_obsolete, "conv=auto"},
@@ -1102,6 +1167,9 @@ static int parse_options(struct super_block *sb, char *options, int is_vfat,
                case Opt_nfs_nostale_ro:
                        opts->nfs = FAT_NFS_NOSTALE_RO;
                        break;
+               case Opt_dos1xfloppy:
+                       opts->dos1xfloppy = 1;
+                       break;
 
                /* msdos specific */
                case Opt_dots:
@@ -1247,6 +1315,169 @@ static unsigned long calc_fat_clusters(struct super_block *sb)
        return sbi->fat_length * sb->s_blocksize * 8 / sbi->fat_bits;
 }
 
+static bool fat_bpb_is_zero(struct fat_boot_sector *b)
+{
+       if (get_unaligned_le16(&b->sector_size))
+               return false;
+       if (b->sec_per_clus)
+               return false;
+       if (b->reserved)
+               return false;
+       if (b->fats)
+               return false;
+       if (get_unaligned_le16(&b->dir_entries))
+               return false;
+       if (get_unaligned_le16(&b->sectors))
+               return false;
+       if (b->media)
+               return false;
+       if (b->fat_length)
+               return false;
+       if (b->secs_track)
+               return false;
+       if (b->heads)
+               return false;
+       return true;
+}
+
+static int fat_read_bpb(struct super_block *sb, struct fat_boot_sector *b,
+       int silent, struct fat_bios_param_block *bpb)
+{
+       int error = -EINVAL;
+
+       /* Read in BPB ... */
+       memset(bpb, 0, sizeof(*bpb));
+       bpb->fat_sector_size = get_unaligned_le16(&b->sector_size);
+       bpb->fat_sec_per_clus = b->sec_per_clus;
+       bpb->fat_reserved = le16_to_cpu(b->reserved);
+       bpb->fat_fats = b->fats;
+       bpb->fat_dir_entries = get_unaligned_le16(&b->dir_entries);
+       bpb->fat_sectors = get_unaligned_le16(&b->sectors);
+       bpb->fat_fat_length = le16_to_cpu(b->fat_length);
+       bpb->fat_total_sect = le32_to_cpu(b->total_sect);
+
+       bpb->fat16_state = b->fat16.state;
+       bpb->fat16_vol_id = get_unaligned_le32(b->fat16.vol_id);
+
+       bpb->fat32_length = le32_to_cpu(b->fat32.length);
+       bpb->fat32_root_cluster = le32_to_cpu(b->fat32.root_cluster);
+       bpb->fat32_info_sector = le16_to_cpu(b->fat32.info_sector);
+       bpb->fat32_state = b->fat32.state;
+       bpb->fat32_vol_id = get_unaligned_le32(b->fat32.vol_id);
+
+       /* Validate this looks like a FAT filesystem BPB */
+       if (!bpb->fat_reserved) {
+               if (!silent)
+                       fat_msg(sb, KERN_ERR,
+                               "bogus number of reserved sectors");
+               goto out;
+       }
+       if (!bpb->fat_fats) {
+               if (!silent)
+                       fat_msg(sb, KERN_ERR, "bogus number of FAT structure");
+               goto out;
+       }
+
+       /*
+        * Earlier we checked here that b->secs_track and b->head are nonzero,
+        * but it turns out valid FAT filesystems can have zero there.
+        */
+
+       if (!fat_valid_media(b->media)) {
+               if (!silent)
+                       fat_msg(sb, KERN_ERR, "invalid media value (0x%02x)",
+                               (unsigned)b->media);
+               goto out;
+       }
+
+       if (!is_power_of_2(bpb->fat_sector_size)
+           || (bpb->fat_sector_size < 512)
+           || (bpb->fat_sector_size > 4096)) {
+               if (!silent)
+                       fat_msg(sb, KERN_ERR, "bogus logical sector size %u",
+                              (unsigned)bpb->fat_sector_size);
+               goto out;
+       }
+
+       if (!is_power_of_2(bpb->fat_sec_per_clus)) {
+               if (!silent)
+                       fat_msg(sb, KERN_ERR, "bogus sectors per cluster %u",
+                               (unsigned)bpb->fat_sec_per_clus);
+               goto out;
+       }
+
+       error = 0;
+
+out:
+       return error;
+}
+
+static int fat_read_static_bpb(struct super_block *sb,
+       struct fat_boot_sector *b, int silent,
+       struct fat_bios_param_block *bpb)
+{
+       static const char *notdos1x = "This doesn't look like a DOS 1.x volume";
+
+       struct fat_floppy_defaults *fdefaults = NULL;
+       int error = -EINVAL;
+       sector_t bd_sects;
+       unsigned i;
+
+       bd_sects = i_size_read(sb->s_bdev->bd_inode) / SECTOR_SIZE;
+
+       /* 16-bit DOS 1.x reliably wrote bootstrap short-jmp code */
+       if (b->ignored[0] != 0xeb || b->ignored[2] != 0x90) {
+               if (!silent)
+                       fat_msg(sb, KERN_ERR,
+                               "%s; no bootstrapping code", notdos1x);
+               goto out;
+       }
+
+       /*
+        * If any value in this region is non-zero, it isn't archaic
+        * DOS.
+        */
+       if (!fat_bpb_is_zero(b)) {
+               if (!silent)
+                       fat_msg(sb, KERN_ERR,
+                               "%s; DOS 2.x BPB is non-zero", notdos1x);
+               goto out;
+       }
+
+       for (i = 0; i < ARRAY_SIZE(floppy_defaults); i++) {
+               if (floppy_defaults[i].nr_sectors == bd_sects) {
+                       fdefaults = &floppy_defaults[i];
+                       break;
+               }
+       }
+
+       if (fdefaults == NULL) {
+               if (!silent)
+                       fat_msg(sb, KERN_WARNING,
+                               "This looks like a DOS 1.x volume, but isn't a recognized floppy size (%llu sectors)",
+                               (u64)bd_sects);
+               goto out;
+       }
+
+       if (!silent)
+               fat_msg(sb, KERN_INFO,
+                       "This looks like a DOS 1.x volume; assuming default BPB values");
+
+       memset(bpb, 0, sizeof(*bpb));
+       bpb->fat_sector_size = SECTOR_SIZE;
+       bpb->fat_sec_per_clus = fdefaults->sec_per_clus;
+       bpb->fat_reserved = 1;
+       bpb->fat_fats = 2;
+       bpb->fat_dir_entries = fdefaults->dir_entries;
+       bpb->fat_sectors = fdefaults->nr_sectors;
+       bpb->fat_fat_length = fdefaults->fat_length;
+
+       error = 0;
+
+out:
+       return error;
+}
+
 /*
  * Read the super block of an MS-DOS FS.
  */
@@ -1256,12 +1487,11 @@ int fat_fill_super(struct super_block *sb, void *data, int silent, int isvfat,
        struct inode *root_inode = NULL, *fat_inode = NULL;
        struct inode *fsinfo_inode = NULL;
        struct buffer_head *bh;
-       struct fat_boot_sector *b;
+       struct fat_bios_param_block bpb;
        struct msdos_sb_info *sbi;
        u16 logical_sector_size;
        u32 total_sectors, total_clusters, fat_clusters, rootdir_sectors;
        int debug;
-       unsigned int media;
        long error;
        char buf[50];
 
@@ -1298,100 +1528,72 @@ int fat_fill_super(struct super_block *sb, void *data, int silent, int isvfat,
                goto out_fail;
        }
 
-       b = (struct fat_boot_sector *) bh->b_data;
-       if (!b->reserved) {
-               if (!silent)
-                       fat_msg(sb, KERN_ERR, "bogus number of reserved sectors");
-               brelse(bh);
-               goto out_invalid;
-       }
-       if (!b->fats) {
-               if (!silent)
-                       fat_msg(sb, KERN_ERR, "bogus number of FAT structure");
-               brelse(bh);
-               goto out_invalid;
-       }
-
-       /*
-        * Earlier we checked here that b->secs_track and b->head are nonzero,
-        * but it turns out valid FAT filesystems can have zero there.
-        */
+       error = fat_read_bpb(sb, (struct fat_boot_sector *)bh->b_data, silent,
+               &bpb);
+       if (error == -EINVAL && sbi->options.dos1xfloppy)
+               error = fat_read_static_bpb(sb,
+                       (struct fat_boot_sector *)bh->b_data, silent, &bpb);
+       brelse(bh);
 
-       media = b->media;
-       if (!fat_valid_media(media)) {
-               if (!silent)
-                       fat_msg(sb, KERN_ERR, "invalid media value (0x%02x)",
-                              media);
-               brelse(bh);
-               goto out_invalid;
-       }
-       logical_sector_size = get_unaligned_le16(&b->sector_size);
-       if (!is_power_of_2(logical_sector_size)
-           || (logical_sector_size < 512)
-           || (logical_sector_size > 4096)) {
-               if (!silent)
-                       fat_msg(sb, KERN_ERR, "bogus logical sector size %u",
-                              logical_sector_size);
-               brelse(bh);
-               goto out_invalid;
-       }
-       sbi->sec_per_clus = b->sec_per_clus;
-       if (!is_power_of_2(sbi->sec_per_clus)) {
-               if (!silent)
-                       fat_msg(sb, KERN_ERR, "bogus sectors per cluster %u",
-                              sbi->sec_per_clus);
-               brelse(bh);
+       if (error == -EINVAL)
                goto out_invalid;
-       }
+       else if (error)
+               goto out_fail;
+
+       logical_sector_size = bpb.fat_sector_size;
+       sbi->sec_per_clus = bpb.fat_sec_per_clus;
 
+       error = -EIO;
        if (logical_sector_size < sb->s_blocksize) {
                fat_msg(sb, KERN_ERR, "logical sector size too small for device"
                       " (logical sector size = %u)", logical_sector_size);
-               brelse(bh);
                goto out_fail;
        }
+
        if (logical_sector_size > sb->s_blocksize) {
-               brelse(bh);
+               struct buffer_head *bh_resize;
 
                if (!sb_set_blocksize(sb, logical_sector_size)) {
                        fat_msg(sb, KERN_ERR, "unable to set blocksize %u",
                               logical_sector_size);
                        goto out_fail;
                }
-               bh = sb_bread(sb, 0);
-               if (bh == NULL) {
+
+               /* Verify that the larger boot sector is fully readable */
+               bh_resize = sb_bread(sb, 0);
+               if (bh_resize == NULL) {
                        fat_msg(sb, KERN_ERR, "unable to read boot sector"
                               " (logical sector size = %lu)",
                               sb->s_blocksize);
                        goto out_fail;
                }
-               b = (struct fat_boot_sector *) bh->b_data;
+               brelse(bh_resize);
        }
 
        mutex_init(&sbi->s_lock);
        sbi->cluster_size = sb->s_blocksize * sbi->sec_per_clus;
        sbi->cluster_bits = ffs(sbi->cluster_size) - 1;
-       sbi->fats = b->fats;
+       sbi->fats = bpb.fat_fats;
        sbi->fat_bits = 0;              /* Don't know yet */
-       sbi->fat_start = le16_to_cpu(b->reserved);
-       sbi->fat_length = le16_to_cpu(b->fat_length);
+       sbi->fat_start = bpb.fat_reserved;
+       sbi->fat_length = bpb.fat_fat_length;
        sbi->root_cluster = 0;
        sbi->free_clusters = -1;        /* Don't know yet */
        sbi->free_clus_valid = 0;
        sbi->prev_free = FAT_START_ENT;
        sb->s_maxbytes = 0xffffffff;
 
-       if (!sbi->fat_length && b->fat32.length) {
+       if (!sbi->fat_length && bpb.fat32_length) {
                struct fat_boot_fsinfo *fsinfo;
                struct buffer_head *fsinfo_bh;
 
                /* Must be FAT32 */
                sbi->fat_bits = 32;
-               sbi->fat_length = le32_to_cpu(b->fat32.length);
-               sbi->root_cluster = le32_to_cpu(b->fat32.root_cluster);
+               sbi->fat_length = bpb.fat32_length;
+               sbi->root_cluster = bpb.fat32_root_cluster;
 
                /* MC - if info_sector is 0, don't multiply by 0 */
-               sbi->fsinfo_sector = le16_to_cpu(b->fat32.info_sector);
+               sbi->fsinfo_sector = bpb.fat32_info_sector;
                if (sbi->fsinfo_sector == 0)
                        sbi->fsinfo_sector = 1;
 
@@ -1399,7 +1601,6 @@ int fat_fill_super(struct super_block *sb, void *data, int silent, int isvfat,
                if (fsinfo_bh == NULL) {
                        fat_msg(sb, KERN_ERR, "bread failed, FSINFO block"
                               " (sector = %lu)", sbi->fsinfo_sector);
-                       brelse(bh);
                        goto out_fail;
                }
 
@@ -1422,35 +1623,28 @@ int fat_fill_super(struct super_block *sb, void *data, int silent, int isvfat,
 
        /* interpret volume ID as a little endian 32 bit integer */
        if (sbi->fat_bits == 32)
-               sbi->vol_id = (((u32)b->fat32.vol_id[0]) |
-                                       ((u32)b->fat32.vol_id[1] << 8) |
-                                       ((u32)b->fat32.vol_id[2] << 16) |
-                                       ((u32)b->fat32.vol_id[3] << 24));
+               sbi->vol_id = bpb.fat32_vol_id;
        else /* fat 16 or 12 */
-               sbi->vol_id = (((u32)b->fat16.vol_id[0]) |
-                                       ((u32)b->fat16.vol_id[1] << 8) |
-                                       ((u32)b->fat16.vol_id[2] << 16) |
-                                       ((u32)b->fat16.vol_id[3] << 24));
+               sbi->vol_id = bpb.fat16_vol_id;
 
        sbi->dir_per_block = sb->s_blocksize / sizeof(struct msdos_dir_entry);
        sbi->dir_per_block_bits = ffs(sbi->dir_per_block) - 1;
 
        sbi->dir_start = sbi->fat_start + sbi->fats * sbi->fat_length;
-       sbi->dir_entries = get_unaligned_le16(&b->dir_entries);
+       sbi->dir_entries = bpb.fat_dir_entries;
        if (sbi->dir_entries & (sbi->dir_per_block - 1)) {
                if (!silent)
                        fat_msg(sb, KERN_ERR, "bogus directory-entries per block"
                               " (%u)", sbi->dir_entries);
-               brelse(bh);
                goto out_invalid;
        }
 
        rootdir_sectors = sbi->dir_entries
                * sizeof(struct msdos_dir_entry) / sb->s_blocksize;
        sbi->data_start = sbi->dir_start + rootdir_sectors;
-       total_sectors = get_unaligned_le16(&b->sectors);
+       total_sectors = bpb.fat_sectors;
        if (total_sectors == 0)
-               total_sectors = le32_to_cpu(b->total_sect);
+               total_sectors = bpb.fat_total_sect;
 
        total_clusters = (total_sectors - sbi->data_start) / sbi->sec_per_clus;
 
@@ -1459,9 +1653,9 @@ int fat_fill_super(struct super_block *sb, void *data, int silent, int isvfat,
 
        /* some OSes set FAT_STATE_DIRTY and clean it on unmount. */
        if (sbi->fat_bits == 32)
-               sbi->dirty = b->fat32.state & FAT_STATE_DIRTY;
+               sbi->dirty = bpb.fat32_state & FAT_STATE_DIRTY;
        else /* fat 16 or 12 */
-               sbi->dirty = b->fat16.state & FAT_STATE_DIRTY;
+               sbi->dirty = bpb.fat16_state & FAT_STATE_DIRTY;
 
        /* check that FAT table does not overflow */
        fat_clusters = calc_fat_clusters(sb);
@@ -1470,7 +1664,6 @@ int fat_fill_super(struct super_block *sb, void *data, int silent, int isvfat,
                if (!silent)
                        fat_msg(sb, KERN_ERR, "count of clusters too big (%u)",
                               total_clusters);
-               brelse(bh);
                goto out_invalid;
        }
 
@@ -1483,8 +1676,6 @@ int fat_fill_super(struct super_block *sb, void *data, int silent, int isvfat,
        if (sbi->prev_free < FAT_START_ENT)
                sbi->prev_free = FAT_START_ENT;
 
-       brelse(bh);
-
        /* set up enough so that it can read an inode */
        fat_hash_init(sb);
        dir_hash_init(sb);