Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: quoted-printable
From:   Andreas Dilger <adilger@dilger.ca>
Mime-Version: 1.0 (1.0)
Subject: Re: [PATCH] ext4: Fix rec_len verify error
Date:   Sat, 29 Jul 2023 13:09:38 -0600
Message-Id: <8B200C7E-F6D1-4395-9776-4B521319CD8F@dilger.ca>
References: <20230729061357.1702891-1-zhangshida@kylinos.cn>
Cc:     tytso@mit.edu, adilger.kernel@dilger.ca,
        linux-kernel@vger.kernel.org, zhangshida@kylinos.cn
In-Reply-To: <20230729061357.1702891-1-zhangshida@kylinos.cn>
To:     zhangshida <starzhangzsd@gmail.com>
Precedence: bulk

On Jul 29, 2023, at 00:14, zhangshida <starzhangzsd@gmail.com> wrote:
>=20
> =EF=BB=BFFrom: Shida Zhang <zhangshida@kylinos.cn>
>=20
> with the configuration PAGE_SIZE 64k and filesystem blocksize 64k,
> a problem occurred when more than 13 millon files were directly created
> under a directory:
>=20
> EXT4-fs error (device xx): ext4_dx_csum_set:492: inode #xxxx: comm xxxxx: d=
ir seems corrupt?  Run e2fsck -D.
> EXT4-fs error (device xx): ext4_dx_csum_verify:463: inode #xxxx: comm xxxx=
x: dir seems corrupt?  Run e2fsck -D.
> EXT4-fs error (device xx): dx_probe:856: inode #xxxx: block 8188: comm xxx=
xx: Directory index failed checksum
>=20
> when enough files are created, the fake_dirent->reclen will be 0xffff.
> it doesn't equal to blocksize 65536, i.e. 0x10000.
>=20
> But it is not the same condition when blocksize equals to 4k.
> when enough files are created, the fake_dirent->reclen will be 0x1000.
> it equals to blocksize 4k, i.e. 0x1000.
>=20
> The problem seems to be related to the limitation of the 16-bit field
> when the blocksize is set to 64k. To address this, a special condition
> was introduced to handle it properly.
>=20
> Signed-off-by: Shida Zhang <zhangshida@kylinos.cn>
> ---
> fs/ext4/namei.c | 3 +++
> 1 file changed, 3 insertions(+)
>=20
> diff --git a/fs/ext4/namei.c b/fs/ext4/namei.c
> index 0caf6c730ce3..a422cff25216 100644
> --- a/fs/ext4/namei.c
> +++ b/fs/ext4/namei.c
> @@ -458,6 +458,9 @@ static struct dx_countlimit *get_dx_countlimit(struct i=
node *inode,
>            root->info_length !=3D sizeof(struct dx_root_info))
>            return NULL;
>        count_offset =3D 32;
> +    } else if ((EXT4_BLOCK_SIZE(inode->i_sb) =3D=3D 65536)
> +        && (le16_to_cpu(dirent->rec_len) =3D=3D 65535)) {
> +        count_offset =3D 8;

This should be moved up to the first if-block that is checking the block siz=
e:=20

=00=00=00=00=00=00=00=00        if (le16_to_cpu(dirent->rec_len) =3D=3D EXT4=
_BLOCK_SIZE(inode->i_sb) ||
             (le16_to_cpu(dirent->rec_len) =3D=3D 65535 &&
              EXT4_BLOCK_SIZE(inode->i_sb) >=3D 65536))
=00=00=00=00=00=00=00=00=00=00=00=00=00=00=00=00                count_offset=
 =3D 8;

since this is really the same case.=20

Ecen better would be to use ext4_rec_len_from_disk() to check the
length so that it keeps this large PAGE_SIZE logic in one place, and
does not add overhead on systems with smaller PAGE_SIZE:

        int blocksize =3D EXT4_BLOCK_SIZE(inode->i_sb);

        if (ext4_rec_len_from_disk(dirent->rec_len, blocksize) =3D=3D blocks=
ize)
                count_offset =3D 8;

Cheers, Andreas

>    } else
>        return NULL;
>=20
> --=20
> 2.27.0
>=20