Hi, Jan.
We find a use after free problem when CONFIG_QUOTA is enabled, the detail of
this problem is below.
mount_bdev()
ext4_fill_super()
sb->s_root = d_make_root(root);
ext4_orphan_cleanup()
sb->s_flags |= SB_ACTIVE; <--- 1. mark sb active
ext4_orphan_get()
ext4_truncate()
ext4_block_truncate_page()
mark_buffer_dirty <--- 2. dirty inode
iput()
iput_final <--- 3. put into lru list
ext4_mark_recovery_complete <--- 4. failed and return error
sb->s_root = NULL;
deactivate_locked_super()
kill_block_super()
generic_shutdown_super()
<--- 5. did not evict_inodes
put_super()
__put_super()
<--- 6. put super block
Because of the truncated inodes was dirty and will write them back later, it
will trigger use after free problem. Now the question is why we need to set
SB_ACTIVE bit when enable CONFIG_QUOTA below?
#ifdef CONFIG_QUOTA
/* Needed for iput() to work correctly and not trash data */
sb->s_flags |= SB_ACTIVE;
This code was merged long long ago in v2.6.6, IIUC, it may not affect
the quota statistics it we evict inode directly in the last iput.
In order to slove this UAF problem, I'm not sure is there any side effect
if we just remove this code, or remove SB_ACTIVE and call evict_inodes()
in the error path of ext4_fill_super().
Could you give some suggestions?
Thanks,
Yi.
Hi!
On Mon 22-03-21 23:24:23, Zhang Yi wrote:
> We find a use after free problem when CONFIG_QUOTA is enabled, the detail of
> this problem is below.
>
> mount_bdev()
> ext4_fill_super()
> sb->s_root = d_make_root(root);
> ext4_orphan_cleanup()
> sb->s_flags |= SB_ACTIVE; <--- 1. mark sb active
> ext4_orphan_get()
> ext4_truncate()
> ext4_block_truncate_page()
> mark_buffer_dirty <--- 2. dirty inode
> iput()
> iput_final <--- 3. put into lru list
> ext4_mark_recovery_complete <--- 4. failed and return error
> sb->s_root = NULL;
> deactivate_locked_super()
> kill_block_super()
> generic_shutdown_super()
> <--- 5. did not evict_inodes
> put_super()
> __put_super()
> <--- 6. put super block
>
> Because of the truncated inodes was dirty and will write them back later, it
> will trigger use after free problem. Now the question is why we need to set
> SB_ACTIVE bit when enable CONFIG_QUOTA below?
>
> #ifdef CONFIG_QUOTA
> /* Needed for iput() to work correctly and not trash data */
> sb->s_flags |= SB_ACTIVE;
>
> This code was merged long long ago in v2.6.6, IIUC, it may not affect
> the quota statistics it we evict inode directly in the last iput.
> In order to slove this UAF problem, I'm not sure is there any side effect
> if we just remove this code, or remove SB_ACTIVE and call evict_inodes()
> in the error path of ext4_fill_super().
>
> Could you give some suggestions?
That's a very good question. I do remember that I've added this code back
then because otherwise orphan cleanup was loosing updates to quota files.
But you're right that now I don't see how that could be happening and it
would be nice if we could get rid of this hack (and even better if it also
fixes the problem you've found). I guess I'll just try and test this change
with various quota configurations to see whether something still breaks or
not. Thanks report!
Honza
--
Jan Kara <[email protected]>
SUSE Labs, CR
On 2021/3/23 1:25, Jan Kara wrote:
> Hi!
>
> On Mon 22-03-21 23:24:23, Zhang Yi wrote:
>> We find a use after free problem when CONFIG_QUOTA is enabled, the detail of
>> this problem is below.
>>
>> mount_bdev()
>> ext4_fill_super()
>> sb->s_root = d_make_root(root);
>> ext4_orphan_cleanup()
>> sb->s_flags |= SB_ACTIVE; <--- 1. mark sb active
>> ext4_orphan_get()
>> ext4_truncate()
>> ext4_block_truncate_page()
>> mark_buffer_dirty <--- 2. dirty inode
>> iput()
>> iput_final <--- 3. put into lru list
>> ext4_mark_recovery_complete <--- 4. failed and return error
>> sb->s_root = NULL;
>> deactivate_locked_super()
>> kill_block_super()
>> generic_shutdown_super()
>> <--- 5. did not evict_inodes
>> put_super()
>> __put_super()
>> <--- 6. put super block
>>
>> Because of the truncated inodes was dirty and will write them back later, it
>> will trigger use after free problem. Now the question is why we need to set
>> SB_ACTIVE bit when enable CONFIG_QUOTA below?
>>
>> #ifdef CONFIG_QUOTA
>> /* Needed for iput() to work correctly and not trash data */
>> sb->s_flags |= SB_ACTIVE;
>>
>> This code was merged long long ago in v2.6.6, IIUC, it may not affect
>> the quota statistics it we evict inode directly in the last iput.
>> In order to slove this UAF problem, I'm not sure is there any side effect
>> if we just remove this code, or remove SB_ACTIVE and call evict_inodes()
>> in the error path of ext4_fill_super().
>>
>> Could you give some suggestions?
>
> That's a very good question. I do remember that I've added this code back
> then because otherwise orphan cleanup was loosing updates to quota files.
> But you're right that now I don't see how that could be happening and it
> would be nice if we could get rid of this hack (and even better if it also
> fixes the problem you've found). I guess I'll just try and test this change
> with various quota configurations to see whether something still breaks or
> not. Thanks report!
>
Thanks for taking time to look at this, is this change OK under your various
quota test cases?
Thanks,
Yi.
On Mon 29-03-21 17:20:35, Zhang Yi wrote:
> On 2021/3/23 1:25, Jan Kara wrote:
> > Hi!
> >
> > On Mon 22-03-21 23:24:23, Zhang Yi wrote:
> >> We find a use after free problem when CONFIG_QUOTA is enabled, the detail of
> >> this problem is below.
> >>
> >> mount_bdev()
> >> ext4_fill_super()
> >> sb->s_root = d_make_root(root);
> >> ext4_orphan_cleanup()
> >> sb->s_flags |= SB_ACTIVE; <--- 1. mark sb active
> >> ext4_orphan_get()
> >> ext4_truncate()
> >> ext4_block_truncate_page()
> >> mark_buffer_dirty <--- 2. dirty inode
> >> iput()
> >> iput_final <--- 3. put into lru list
> >> ext4_mark_recovery_complete <--- 4. failed and return error
> >> sb->s_root = NULL;
> >> deactivate_locked_super()
> >> kill_block_super()
> >> generic_shutdown_super()
> >> <--- 5. did not evict_inodes
> >> put_super()
> >> __put_super()
> >> <--- 6. put super block
> >>
> >> Because of the truncated inodes was dirty and will write them back later, it
> >> will trigger use after free problem. Now the question is why we need to set
> >> SB_ACTIVE bit when enable CONFIG_QUOTA below?
> >>
> >> #ifdef CONFIG_QUOTA
> >> /* Needed for iput() to work correctly and not trash data */
> >> sb->s_flags |= SB_ACTIVE;
> >>
> >> This code was merged long long ago in v2.6.6, IIUC, it may not affect
> >> the quota statistics it we evict inode directly in the last iput.
> >> In order to slove this UAF problem, I'm not sure is there any side effect
> >> if we just remove this code, or remove SB_ACTIVE and call evict_inodes()
> >> in the error path of ext4_fill_super().
> >>
> >> Could you give some suggestions?
> >
> > That's a very good question. I do remember that I've added this code back
> > then because otherwise orphan cleanup was loosing updates to quota files.
> > But you're right that now I don't see how that could be happening and it
> > would be nice if we could get rid of this hack (and even better if it also
> > fixes the problem you've found). I guess I'll just try and test this change
> > with various quota configurations to see whether something still breaks or
> > not. Thanks report!
> >
>
> Thanks for taking time to look at this, is this change OK under your various
> quota test cases?
Yes, I did tests both with journalled quotas and with ext4 quota feature
and the quota accounting was correct after orphan recovery. So just
removing the SB_ACTIVE setting is fine AFAICT. Will you send a patch or
should I do it?
Honza
--
Jan Kara <[email protected]>
SUSE Labs, CR
On 2021/3/30 23:02, Jan Kara wrote:
> On Mon 29-03-21 17:20:35, Zhang Yi wrote:
>> On 2021/3/23 1:25, Jan Kara wrote:
>>> Hi!
>>>
>>> On Mon 22-03-21 23:24:23, Zhang Yi wrote:
>>>> We find a use after free problem when CONFIG_QUOTA is enabled, the detail of
>>>> this problem is below.
>>>>
>>>> mount_bdev()
>>>> ext4_fill_super()
>>>> sb->s_root = d_make_root(root);
>>>> ext4_orphan_cleanup()
>>>> sb->s_flags |= SB_ACTIVE; <--- 1. mark sb active
>>>> ext4_orphan_get()
>>>> ext4_truncate()
>>>> ext4_block_truncate_page()
>>>> mark_buffer_dirty <--- 2. dirty inode
>>>> iput()
>>>> iput_final <--- 3. put into lru list
>>>> ext4_mark_recovery_complete <--- 4. failed and return error
>>>> sb->s_root = NULL;
>>>> deactivate_locked_super()
>>>> kill_block_super()
>>>> generic_shutdown_super()
>>>> <--- 5. did not evict_inodes
>>>> put_super()
>>>> __put_super()
>>>> <--- 6. put super block
>>>>
>>>> Because of the truncated inodes was dirty and will write them back later, it
>>>> will trigger use after free problem. Now the question is why we need to set
>>>> SB_ACTIVE bit when enable CONFIG_QUOTA below?
>>>>
>>>> #ifdef CONFIG_QUOTA
>>>> /* Needed for iput() to work correctly and not trash data */
>>>> sb->s_flags |= SB_ACTIVE;
>>>>
>>>> This code was merged long long ago in v2.6.6, IIUC, it may not affect
>>>> the quota statistics it we evict inode directly in the last iput.
>>>> In order to slove this UAF problem, I'm not sure is there any side effect
>>>> if we just remove this code, or remove SB_ACTIVE and call evict_inodes()
>>>> in the error path of ext4_fill_super().
>>>>
>>>> Could you give some suggestions?
>>>
>>> That's a very good question. I do remember that I've added this code back
>>> then because otherwise orphan cleanup was loosing updates to quota files.
>>> But you're right that now I don't see how that could be happening and it
>>> would be nice if we could get rid of this hack (and even better if it also
>>> fixes the problem you've found). I guess I'll just try and test this change
>>> with various quota configurations to see whether something still breaks or
>>> not. Thanks report!
>>>
>>
>> Thanks for taking time to look at this, is this change OK under your various
>> quota test cases?
>
> Yes, I did tests both with journalled quotas and with ext4 quota feature
> and the quota accounting was correct after orphan recovery. So just
> removing the SB_ACTIVE setting is fine AFAICT. Will you send a patch or
> should I do it?
>
Thanks for testing this change, I will send a patch.
Yi.