Re: [PATCH v2] f2fs: fix sbi->extent_list corruption issue

From: Jaegeuk Kim
Date: Wed Dec 19 2018 - 18:42:47 EST


On 12/19, Chao Yu wrote:
> On 2018/12/19 6:47, Jaegeuk Kim wrote:
> > On 12/18, Chao Yu wrote:
> >> On 2018/12/14 22:25, Jaegeuk Kim wrote:
> >>> On 12/14, Sahitya Tummala wrote:
> >>>> On Wed, Dec 12, 2018 at 11:36:08AM +0800, Chao Yu wrote:
> >>>>> On 2018/12/12 11:17, Sahitya Tummala wrote:
> >>>>>> On Fri, Dec 07, 2018 at 05:47:31PM +0800, Chao Yu wrote:
> >>>>>>> On 2018/12/1 4:33, Jaegeuk Kim wrote:
> >>>>>>>> On 11/29, Sahitya Tummala wrote:
> >>>>>>>>>
> >>>>>>>>> On Tue, Nov 27, 2018 at 09:42:39AM +0800, Chao Yu wrote:
> >>>>>>>>>> On 2018/11/27 8:30, Jaegeuk Kim wrote:
> >>>>>>>>>>> On 11/26, Sahitya Tummala wrote:
> >>>>>>>>>>>> When there is a failure in f2fs_fill_super() after/during
> >>>>>>>>>>>> the recovery of fsync'd nodes, it frees the current sbi and
> >>>>>>>>>>>> retries again. This time the mount is successful, but the files
> >>>>>>>>>>>> that got recovered before retry, still holds the extent tree,
> >>>>>>>>>>>> whose extent nodes list is corrupted since sbi and sbi->extent_list
> >>>>>>>>>>>> is freed up. The list_del corruption issue is observed when the
> >>>>>>>>>>>> file system is getting unmounted and when those recoverd files extent
> >>>>>>>>>>>> node is being freed up in the below context.
> >>>>>>>>>>>>
> >>>>>>>>>>>> list_del corruption. prev->next should be fffffff1e1ef5480, but was (null)
> >>>>>>>>>>>> <...>
> >>>>>>>>>>>> kernel BUG at kernel/msm-4.14/lib/list_debug.c:53!
> >>>>>>>>>>>> task: fffffff1f46f2280 task.stack: ffffff8008068000
> >>>>>>>>>>>> lr : __list_del_entry_valid+0x94/0xb4
> >>>>>>>>>>>> pc : __list_del_entry_valid+0x94/0xb4
> >>>>>>>>>>>> <...>
> >>>>>>>>>>>> Call trace:
> >>>>>>>>>>>> __list_del_entry_valid+0x94/0xb4
> >>>>>>>>>>>> __release_extent_node+0xb0/0x114
> >>>>>>>>>>>> __free_extent_tree+0x58/0x7c
> >>>>>>>>>>>> f2fs_shrink_extent_tree+0xdc/0x3b0
> >>>>>>>>>>>> f2fs_leave_shrinker+0x28/0x7c
> >>>>>>>>>>>> f2fs_put_super+0xfc/0x1e0
> >>>>>>>>>>>> generic_shutdown_super+0x70/0xf4
> >>>>>>>>>>>> kill_block_super+0x2c/0x5c
> >>>>>>>>>>>> kill_f2fs_super+0x44/0x50
> >>>>>>>>>>>> deactivate_locked_super+0x60/0x8c
> >>>>>>>>>>>> deactivate_super+0x68/0x74
> >>>>>>>>>>>> cleanup_mnt+0x40/0x78
> >>>>>>>>>>>> __cleanup_mnt+0x1c/0x28
> >>>>>>>>>>>> task_work_run+0x48/0xd0
> >>>>>>>>>>>> do_notify_resume+0x678/0xe98
> >>>>>>>>>>>> work_pending+0x8/0x14
> >>>>>>>>>>>>
> >>>>>>>>>>>> Fix this by cleaning up inodes, extent tree and nodes of those
> >>>>>>>>>>>> recovered files before freeing up sbi and before next retry.
> >>>>>>>>>>>>
> >>>>>>>>>>>> Signed-off-by: Sahitya Tummala <stummala@xxxxxxxxxxxxxx>
> >>>>>>>>>>>> ---
> >>>>>>>>>>>> v2:
> >>>>>>>>>>>> -call evict_inodes() and f2fs_shrink_extent_tree() to cleanup inodes
> >>>>>>>>>>>>
> >>>>>>>>>>>> fs/f2fs/f2fs.h | 1 +
> >>>>>>>>>>>> fs/f2fs/shrinker.c | 2 +-
> >>>>>>>>>>>> fs/f2fs/super.c | 13 ++++++++++++-
> >>>>>>>>>>>> 3 files changed, 14 insertions(+), 2 deletions(-)
> >>>>>>>>>>>>
> >>>>>>>>>>>> diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> >>>>>>>>>>>> index 1e03197..aaee63b 100644
> >>>>>>>>>>>> --- a/fs/f2fs/f2fs.h
> >>>>>>>>>>>> +++ b/fs/f2fs/f2fs.h
> >>>>>>>>>>>> @@ -3407,6 +3407,7 @@ struct rb_entry *f2fs_lookup_rb_tree_ret(struct rb_root_cached *root,
> >>>>>>>>>>>> bool f2fs_check_rb_tree_consistence(struct f2fs_sb_info *sbi,
> >>>>>>>>>>>> struct rb_root_cached *root);
> >>>>>>>>>>>> unsigned int f2fs_shrink_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink);
> >>>>>>>>>>>> +unsigned long __count_extent_cache(struct f2fs_sb_info *sbi);
> >>>>>>>>>>>> bool f2fs_init_extent_tree(struct inode *inode, struct f2fs_extent *i_ext);
> >>>>>>>>>>>> void f2fs_drop_extent_tree(struct inode *inode);
> >>>>>>>>>>>> unsigned int f2fs_destroy_extent_node(struct inode *inode);
> >>>>>>>>>>>> diff --git a/fs/f2fs/shrinker.c b/fs/f2fs/shrinker.c
> >>>>>>>>>>>> index 9e13db9..7e3c13b 100644
> >>>>>>>>>>>> --- a/fs/f2fs/shrinker.c
> >>>>>>>>>>>> +++ b/fs/f2fs/shrinker.c
> >>>>>>>>>>>> @@ -30,7 +30,7 @@ static unsigned long __count_free_nids(struct f2fs_sb_info *sbi)
> >>>>>>>>>>>> return count > 0 ? count : 0;
> >>>>>>>>>>>> }
> >>>>>>>>>>>>
> >>>>>>>>>>>> -static unsigned long __count_extent_cache(struct f2fs_sb_info *sbi)
> >>>>>>>>>>>> +unsigned long __count_extent_cache(struct f2fs_sb_info *sbi)
> >>>>>>>>>>>> {
> >>>>>>>>>>>> return atomic_read(&sbi->total_zombie_tree) +
> >>>>>>>>>>>> atomic_read(&sbi->total_ext_node);
> >>>>>>>>>>>> diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
> >>>>>>>>>>>> index af58b2c..769e7b1 100644
> >>>>>>>>>>>> --- a/fs/f2fs/super.c
> >>>>>>>>>>>> +++ b/fs/f2fs/super.c
> >>>>>>>>>>>> @@ -3016,6 +3016,16 @@ static void f2fs_tuning_parameters(struct f2fs_sb_info *sbi)
> >>>>>>>>>>>> sbi->readdir_ra = 1;
> >>>>>>>>>>>> }
> >>>>>>>>>>>>
> >>>>>>>>>>>> +static void f2fs_cleanup_inodes(struct f2fs_sb_info *sbi)
> >>>>>>>>>>>> +{
> >>>>>>>>>>>> + struct super_block *sb = sbi->sb;
> >>>>>>>>>>>> +
> >>>>>>>>>>>> + sync_filesystem(sb);
> >>>>>>>>>>>
> >>>>>>>>>>> This writes another checkpoint, which would not be what this retrial intended.
> >>>>>>>>>>
> >>>>>>>>>> Actually, checkpoint will not be triggered due to SBI_POR_DOING flag check
> >>>>>>>>>> as below:
> >>>>>>>>>>
> >>>>>>>>>> int f2fs_sync_fs(struct super_block *sb, int sync)
> >>>>>>>>>> {
> >>>>>>>>>> ...
> >>>>>>>>>> if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
> >>>>>>>>>> return -EAGAIN;
> >>>>>>>>>> ...
> >>>>>>>>>> }
> >>>>>>>>>>
> >>>>>>>>>> And also all dirty data/node won't be persisted due to SBI_POR_DOING flag,
> >>>>>>>>>> IIUC.
> >>>>>>>>>>
> >>>>>>>>>
> >>>>>>>>> Thanks Chao for the clarification.
> >>>>>>>>>
> >>>>>>>>> Hi Jaegeuk,
> >>>>>>>>>
> >>>>>>>>> Do you still have any further concerns or comments on this patch?
> >>>>>>>>
> >>>>>>>> Could you try the below first?
> >>>>>>>>
> >>>>>>>> -- How about adding a condition in f2fs_may_extent_tree() when adding extents?
> >>>>>>>> -- Likewise, if (shrinker is not registered) return false;
> >>>>>>>>
> >>>>>>>> If we can fix what you described directly, I don't want to rely on such the
> >>>>>>>> assumptions saying we won't do checkpoint. This flow literally says syncing
> >>>>>>>> and evicting cached objects, which opposed to what we'd like to drop all caches
> >>>>>>>> and restart fill_super again.
> >>>>>>>>
> >>>>>>>> Let me consider this as a final resolution.
> >>>>>>>
> >>>>>>> Jaegeuk,
> >>>>>>>
> >>>>>>> Still I want to ask, what kind of scenario we have to add retry logic in
> >>>>>>> fill_super for? As in android scenario, it must be extreme rare case that
> >>>>>>> system runs out-of-memory in boot time...at least, I didn't get any kind of
> >>>>>>> report like that.
> >>>>>>>
> >>>>>> Hi Chao,
> >>>>>
> >>>>> Hi Sahitya,
> >>>>>
> >>>>> Thanks for letting me know that, I git-blamed the code, and found the
> >>>>> original intention is like what you described:
> >>>>>
> >>>>> commit ed2e621a95d704e6a4e904cc00524e8cbddda0c2
> >>>>> Author: Jaegeuk Kim <jaegeuk@xxxxxxxxxx>
> >>>>> Date: Fri Aug 8 15:37:41 2014 -0700
> >>>>>
> >>>>> f2fs: give a chance to mount again when encountering errors
> >>>>>
> >>>>> This patch gives another chance to try mount process when we encounter
> >>>>> an error.
> >>>>> This makes an effect on the roll-forward recovery failures as well.
> >>>>>
> >>>>> But I doubt that if we failed in recovery, maybe there is corruption in
> >>>>> this image, would it be better to fail the mount, and let user fsck it and
> >>>>> retry the mount? otherwise, the corruption may be expanded...
> >>>
> >>> The problem was there was no way to recover roll-forward area by fsck. IOWs,
> >>> mount was failing all the time. I don't think roll-forward itself can corrupt
> >>
> >> I got your concern, IMO, if mount fails, it will be better to let user
> >> decide how to handle it.
> >
> > Roll-forward is not based on user decision, but f2fs does internally. So, I'm
>
> Yup, IMO without roll-forward, data may lose, and posix compliance can be
> corrupted, f2fs should do roll-forward internally as possible as it can.
>
> > in doubt we have to ask users on any failed case here.
> > I don't catch why we need to revert this which has been landed for a long time.
>
> Actually, what I mean is mount can fail due to different reason, but we
> handle it with fixed retry method by dropping recovery, it may be not flexible.
>
> For example, first fill_super fails due to no memory, then second
> fill_super runs w/o recovery, if we succeed, we may lose fsynced data. I
> don't think it make sense.

Then, that's another issue, and yeah, we can prepare a patch for that.

>
> Thanks,
>
> >
> >>
> >> If mount fails due to:
> >>
> >> 1) recovery, user can run fsck and/or try disable_roll_forward or
> >> norecovery option in another mount;
> >> 2) -EINVAL caused by sanity, user can run fsck and retry mount.
> >> 3) -ENOMEM caused low memory in system, user can add more memory and retry
> >> mount.
> >> ...
> >>
> >> Thanks,
> >>
> >>> the image more. Please report, if you have any issue on this.>
> >>>>>
> >>>>
> >>>> Hi Jaegeuk,
> >>>>
> >>>> How do you think about this? If you think it is okay, then I will fix the
> >>>> sbi->extent_list corruption issue, by removing the retry logic. Otherwise,
> >>>> I will fix it in the extent handling as you have suggested earlier.
> >>>
> >>> I'd like to keep retry logic, so could you please test what I suggested above?
> >>>
> >>> Thanks,
> >>>
> >>>>
> >>>> Thanks,
> >>>>
> >>>>> Thanks,
> >>>>>
> >>>>>>
> >>>>>> In my case, the first boot up has a failure in recovery as below -
> >>>>>>
> >>>>>> F2FS-fs (mmcblk0p75): find_fsync_dnodes: detect looped node chain, blkaddr:1979471, next:1979472
> >>>>>> F2FS-fs (mmcblk0p75): Cannot recover all fsync data errno=-22
> >>>>>>
> >>>>>> But in the second attempt, retry will be set to false and because of that
> >>>>>> recover_fsync_data() is skipped. This helped mount to be successful in
> >>>>>> the second attempt.
> >>>>>>
> >>>>>> Thanks,
> >>>>>>
> >>>>>>> Thanks,
> >>>>>>>
> >>>>>>>>
> >>>>>>>> Thanks,
> >>>>>>>>
> >>>>>>>>>
> >>>>>>>>> Thanks,
> >>>>>>>>> Sahitya.
> >>>>>>>>>
> >>>>>>>>>> Thanks,
> >>>>>>>>>>
> >>>>>>>>>>> How about adding a condition in f2fs_may_extent_tree() when adding extents?
> >>>>>>>>>>> Likewise, if (shrinker is not registered) return false;
> >>>>>>>>>>>
> >>>>>>>>>>>
> >>>>>>>>>>>> + shrink_dcache_sb(sb);
> >>>>>>>>>>>> + evict_inodes(sb);
> >>>>>>>>>>>> + f2fs_shrink_extent_tree(sbi, __count_extent_cache(sbi));
> >>>>>>>>>>>> +}
> >>>>>>>>>>>> +
> >>>>>>>>>>>> static int f2fs_fill_super(struct super_block *sb, void *data, int silent)
> >>>>>>>>>>>> {
> >>>>>>>>>>>> struct f2fs_sb_info *sbi;
> >>>>>>>>>>>> @@ -3402,6 +3412,8 @@ static int f2fs_fill_super(struct super_block *sb, void *data, int silent)
> >>>>>>>>>>>> * falls into an infinite loop in f2fs_sync_meta_pages().
> >>>>>>>>>>>> */
> >>>>>>>>>>>> truncate_inode_pages_final(META_MAPPING(sbi));
> >>>>>>>>>>>> + /* cleanup recovery and quota inodes */
> >>>>>>>>>>>> + f2fs_cleanup_inodes(sbi);
> >>>>>>>>>>>> f2fs_unregister_sysfs(sbi);
> >>>>>>>>>>>> free_root_inode:
> >>>>>>>>>>>> dput(sb->s_root);
> >>>>>>>>>>>> @@ -3445,7 +3457,6 @@ static int f2fs_fill_super(struct super_block *sb, void *data, int silent)
> >>>>>>>>>>>> /* give only one another chance */
> >>>>>>>>>>>> if (retry) {
> >>>>>>>>>>>> retry = false;
> >>>>>>>>>>>> - shrink_dcache_sb(sb);
> >>>>>>>>>>>> goto try_onemore;
> >>>>>>>>>>>> }
> >>>>>>>>>>>> return err;
> >>>>>>>>>>>> --
> >>>>>>>>>>>> Qualcomm India Private Limited, on behalf of Qualcomm Innovation Center, Inc.
> >>>>>>>>>>>> Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project.
> >>>>>>>>>>>
> >>>>>>>>>>> .
> >>>>>>>>>>>
> >>>>>>>>>>
> >>>>>>>>>
> >>>>>>>>> --
> >>>>>>>>> --
> >>>>>>>>> Sent by a consultant of the Qualcomm Innovation Center, Inc.
> >>>>>>>>> The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum.
> >>>>>>>>
> >>>>>>>> .
> >>>>>>>>
> >>>>>>>
> >>>>>>
> >>>>>
> >>>>
> >>>> --
> >>>> --
> >>>> Sent by a consultant of the Qualcomm Innovation Center, Inc.
> >>>> The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum.
> >>>
> >>> .
> >>>
> >
> > .
> >