Re: [PATCH] ntfs3: fix mount failure on volumes with fragmented MFT bitmap

On 3/28/26 11:53, Ruslan Elishev wrote:

> [You don't often get email from [email protected]. Learn why this is important at https://aka.ms/LearnAboutSenderIdentification ]
>
> When the $MFT's $BITMAP attribute is fragmented across multiple MFT
> records (base record + extent records), ntfs_fill_super() fails with
> -ENOENT during wnd_init() because the MFT bitmap's run list only
> contains runs from the base MFT record.
>
> The issue is that wnd_init() (which calls wnd_rescan()) is invoked
> before ni_load_all_mi(), so the extent MFT records containing
> additional $BITMAP runs have not been loaded yet. When wnd_rescan()
> tries to look up a VCN beyond the base record's runs, run_lookup_entry()
> fails and returns -ENOENT.
>
> This affects NTFS volumes with a large or heavily fragmented MFT, which
> is common on long-used Windows systems where the MFT bitmap's run list
> doesn't fit in the base MFT record and spills into extent records.
>
> Fix this by:
> 1. Moving ni_load_all_mi() before wnd_init() so all extent records
>     are available.
> 2. After ni_load_all_mi(), iterating through the attribute list to
>     find any $BITMAP extent attributes and unpacking their runs into
>     sbi->mft.bitmap.run before wnd_init() is called.
>
> Tested on a 664GB NTFS volume with 86 MFT bitmap runs spanning
> records 0 (VCN 0-105) and 17 (VCN 106-165). Before the fix, mount
> fails with -ENOENT. After the fix, mount succeeds and all read/write
> operations work correctly. Stress-tested with 8 test categories
> (large file integrity, 10K small files, copy, move, delete/recreate
> cycles, concurrent writes, deep directories, overwrite persistence).
>
> Signed-off-by: Ruslan Elishev <[email protected]>
> ---
>   fs/ntfs3/super.c | 39 +++++++++++++++++++++++++++++++++------
>   1 file changed, 33 insertions(+), 6 deletions(-)
>
> diff --git a/fs/ntfs3/super.c b/fs/ntfs3/super.c
> index abcdef..123456 100644
> --- a/fs/ntfs3/super.c
> +++ b/fs/ntfs3/super.c
> @@ -1364,16 +1364,43 @@
>          tt = inode->i_size >> sbi->record_bits;
>          sbi->mft.next_free = MFT_REC_USER;
>
> -       err = wnd_init(&sbi->mft.bitmap, sb, tt);
> -       if (err)
> -               goto put_inode_out;
> -
>          err = ni_load_all_mi(ni);
>          if (err) {
>                  ntfs_err(sb, "Failed to load $MFT's subrecords (%d).", err);
>                  goto put_inode_out;
>          }
>
> +       /* Merge MFT bitmap runs from extent records loaded by ni_load_all_mi. */
> +       {
> +               struct ATTRIB *a = NULL;
> +               struct ATTR_LIST_ENTRY *le = NULL;
> +
> +               while ((a = ni_enum_attr_ex(ni, a, &le, NULL))) {
> +                       CLST svcn, evcn;
> +                       u16 roff;
> +
> +                       if (a->type != ATTR_BITMAP || !a->non_res)
> +                               continue;
> +
> +                       svcn = le64_to_cpu(a->nres.svcn);
> +                       if (!svcn)
> +                               continue; /* Base record runs already loaded. */
> +
> +                       evcn = le64_to_cpu(a->nres.evcn);
> +                       roff = le16_to_cpu(a->nres.run_off);
> +
> +                       err = run_unpack_ex(&sbi->mft.bitmap.run, sbi,
> +                                           MFT_REC_MFT, svcn, evcn, svcn,
> +                                           Add2Ptr(a, roff),
> +                                           le32_to_cpu(a->size) - roff);
> +                       if (err < 0) {
> +                               ntfs_err(sb, "Failed to unpack $MFT bitmap extent (%d).", err);
> +                               goto put_inode_out;
> +                       }
> +                       err = 0;
> +               }
> +       }
> +
> +       err = wnd_init(&sbi->mft.bitmap, sb, tt);
> +       if (err)
> +               goto put_inode_out;
> +
>          sbi->mft.ni = ni;
>
>          /* Load $Bitmap. */
> --
> 2.43.0

Hello,

I've failed to apply your patch as it is. Git produced an error.
But I've taken your changes unmodified. Thanks for the patch.

Regards,
Konstantin