[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <Z76KaW6iAsafDDbB@google.com>
Date: Wed, 26 Feb 2025 03:28:41 +0000
From: Jaegeuk Kim <jaegeuk@...nel.org>
To: Chao Yu <chao@...nel.org>
Cc: linux-f2fs-devel@...ts.sourceforge.net, linux-kernel@...r.kernel.org,
Leo Stone <leocstone@...il.com>,
syzbot+b01a36acd7007e273a83@...kaller.appspotmail.com
Subject: Re: [PATCH v4] f2fs: add check for deleted inode
On 02/24, Chao Yu wrote:
> On 2/14/25 09:44, Chao Yu wrote:
> > On 2/14/25 01:38, Jaegeuk Kim wrote:
> >> On 02/13, Chao Yu wrote:
> >>> On 2/13/25 00:47, Jaegeuk Kim wrote:
> >>>> On 02/12, Chao Yu wrote:
> >>>>> From: Leo Stone <leocstone@...il.com>
> >>>>>
> >>>>> The syzbot reproducer mounts a f2fs image, then tries to unlink an
> >>>>> existing file. However, the unlinked file already has a link count of 0
> >>>>> when it is read for the first time in do_read_inode().
> >>>>>
> >>>>> Add a check to sanity_check_inode() for i_nlink == 0.
> >>>>>
> >>>>> [Chao Yu: rebase the code and fix orphan inode recovery issue]
> >>>>> Reported-by: syzbot+b01a36acd7007e273a83@...kaller.appspotmail.com
> >>>>> Closes: https://syzkaller.appspot.com/bug?extid=b01a36acd7007e273a83
> >>>>> Fixes: 39a53e0ce0df ("f2fs: add superblock and major in-memory structure")
> >>>>> Signed-off-by: Leo Stone <leocstone@...il.com>
> >>>>> Signed-off-by: Chao Yu <chao@...nel.org>
> >>>>> ---
> >>>>> fs/f2fs/checkpoint.c | 4 ++++
> >>>>> fs/f2fs/f2fs.h | 1 +
> >>>>> fs/f2fs/inode.c | 6 ++++++
> >>>>> 3 files changed, 11 insertions(+)
> >>>>>
> >>>>> diff --git a/fs/f2fs/checkpoint.c b/fs/f2fs/checkpoint.c
> >>>>> index bd890738b94d..ada2c548645c 100644
> >>>>> --- a/fs/f2fs/checkpoint.c
> >>>>> +++ b/fs/f2fs/checkpoint.c
> >>>>> @@ -751,6 +751,8 @@ int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
> >>>>> if (is_sbi_flag_set(sbi, SBI_IS_WRITABLE))
> >>>>> f2fs_info(sbi, "orphan cleanup on readonly fs");
> >>>>>
> >>>>> + set_sbi_flag(sbi, SBI_ORPHAN_RECOVERY);
> >>>>
> >>>> What about using SBI_POR_DOING?
> >>>
> >>> SBI_POR_DOING will cover most flow of f2fs_fill_super(), I think we can add a
> >>> separated flag just covering f2fs_recover_orphan_inodes(), so that we can allow
> >>> iget() of root_inode and all inodes during roll-forward recovery to do sanity
> >>> check nlink w/ zero. What do you think?
> >>
> >> Can we do this sanity check after f2fs_iget in the f2fs_unlink() only?
> >
> > Sure, we need to cover f2fs_rename() as well, please check this:
> >
> > https://lore.kernel.org/all/67450f9a.050a0220.21d33d.0003.GAE@google.com
>
> Hi Jaegeuk,
>
> I'm testing this, seems there is a problem, once we opened an inode that
> has zeroed nlink, in f2fs_evict_inode(), the inode and all its data will be
> deleted, then leaving its stale dir entry in parent directory.
>
> What do you think using v4? so that we may has chance to repair it w/ fsck
> rather than just deleting it?
Do you mean v4 as the below change?
>
> ---
> fs/f2fs/namei.c | 19 +++++++++++++++++++
> 1 file changed, 19 insertions(+)
>
> diff --git a/fs/f2fs/namei.c b/fs/f2fs/namei.c
> index a278c7da8177..949621bc0d07 100644
> --- a/fs/f2fs/namei.c
> +++ b/fs/f2fs/namei.c
> @@ -547,6 +547,16 @@ static int f2fs_unlink(struct inode *dir, struct dentry *dentry)
> goto fail;
> }
>
> + if (unlikely(S_ISDIR(inode->i_mode) ?
> + inode->i_nlink <= 1 : inode->i_nlink == 0)) {
> + f2fs_err_ratelimited(sbi, "%s: inode (ino=%lx) has inconsistent nlink: %u, isdir: %d",
> + __func__, inode->i_ino, inode->i_nlink,
> + S_ISDIR(inode->i_mode));
> + err = -EFSCORRUPTED;
> + set_sbi_flag(sbi, SBI_NEED_FSCK);
> + goto fail;
> + }
> +
> err = f2fs_dquot_initialize(dir);
> if (err)
> goto fail;
> @@ -968,6 +978,15 @@ static int f2fs_rename(struct mnt_idmap *idmap, struct inode *old_dir,
> }
>
> if (new_inode) {
> + if (unlikely(old_is_dir ?
> + new_inode->i_nlink <= 1 : new_inode->i_nlink == 0)) {
> + f2fs_err_ratelimited(sbi, "%s: inode (ino=%lx) has inconsistent nlink: %u, isdir: %d",
> + __func__, new_inode->i_ino, new_inode->i_nlink,
> + S_ISDIR(new_inode->i_mode));
> + err = -EFSCORRUPTED;
> + set_sbi_flag(sbi, SBI_NEED_FSCK);
> + goto out_dir;
> + }
>
> err = -ENOTEMPTY;
> if (old_is_dir && !f2fs_empty_dir(new_inode))
> --
> 2.48.1.601.g30ceb7b040-goog
>
> >
> > Thanks,
> >
> >>
> >>>
> >>> Thanks,
> >>>
> >>>>
> >>>>> +
> >>>>> start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
> >>>>> orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
> >>>>>
> >>>>> @@ -778,9 +780,11 @@ int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
> >>>>> }
> >>>>> f2fs_put_page(page, 1);
> >>>>> }
> >>>>> +
> >>>>> /* clear Orphan Flag */
> >>>>> clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
> >>>>> out:
> >>>>> + clear_sbi_flag(sbi, SBI_ORPHAN_RECOVERY);
> >>>>> set_sbi_flag(sbi, SBI_IS_RECOVERED);
> >>>>>
> >>>>> return err;
> >>>>> diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> >>>>> index 05879c6dc4d6..1c75081c0c14 100644
> >>>>> --- a/fs/f2fs/f2fs.h
> >>>>> +++ b/fs/f2fs/f2fs.h
> >>>>> @@ -1322,6 +1322,7 @@ enum {
> >>>>> SBI_IS_CLOSE, /* specify unmounting */
> >>>>> SBI_NEED_FSCK, /* need fsck.f2fs to fix */
> >>>>> SBI_POR_DOING, /* recovery is doing or not */
> >>>>> + SBI_ORPHAN_RECOVERY, /* orphan inodes recovery is doing */
> >>>>> SBI_NEED_SB_WRITE, /* need to recover superblock */
> >>>>> SBI_NEED_CP, /* need to checkpoint */
> >>>>> SBI_IS_SHUTDOWN, /* shutdown by ioctl */
> >>>>> diff --git a/fs/f2fs/inode.c b/fs/f2fs/inode.c
> >>>>> index d6ad7810df69..02f1b69d03d8 100644
> >>>>> --- a/fs/f2fs/inode.c
> >>>>> +++ b/fs/f2fs/inode.c
> >>>>> @@ -386,6 +386,12 @@ static bool sanity_check_inode(struct inode *inode, struct page *node_page)
> >>>>> }
> >>>>> }
> >>>>>
> >>>>> + if (inode->i_nlink == 0 && !is_sbi_flag_set(sbi, SBI_ORPHAN_RECOVERY)) {
> >>>>> + f2fs_warn(sbi, "%s: inode (ino=%lx) has a link count of 0",
> >>>>> + __func__, inode->i_ino);
> >>>>> + return false;
> >>>>> + }
> >>>>> +
> >>>>> return true;
> >>>>> }
> >>>>>
> >>>>> --
> >>>>> 2.48.1.502.g6dc24dfdaf-goog
> >
Powered by blists - more mailing lists