在 2022/3/15 18:55, Gao Xiang 写道:
> On Tue, Mar 15, 2022 at 06:43:01PM +0800, Huang Jianan wrote:
>> 在 2022/3/15 15:51, Dongliang Mu 写道:
>>> From: Dongliang Mu <[email protected]>
>>>
>>> Syzkaller hit 'WARNING: kobject bug in erofs_unregister_sysfs'. This bug
>>> is triggered by injecting fault in kobject_init_and_add of
>>> erofs_unregister_sysfs.
>>>
>>> Fix this by remembering if kobject_init_and_add is successful.
>>>
>>> Note that I've tested the patch and the crash does not occur any more.
>>>
>>> Reported-by: syzkaller <[email protected]>
>>> Signed-off-by: Dongliang Mu <[email protected]>
>>> ---
>>> fs/erofs/internal.h | 1 +
>>> fs/erofs/sysfs.c | 9 ++++++---
>>> 2 files changed, 7 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/fs/erofs/internal.h b/fs/erofs/internal.h
>>> index 5aa2cf2c2f80..9e20665e3f68 100644
>>> --- a/fs/erofs/internal.h
>>> +++ b/fs/erofs/internal.h
>>> @@ -144,6 +144,7 @@ struct erofs_sb_info {
>>> u32 feature_incompat;
>>> /* sysfs support */
>>> + bool s_sysfs_inited;
>> Hi Dongliang,
>>
>> How about using sbi->s_kobj.state_in_sysfs to avoid adding a extra member in
>> sbi ?
> Ok, I have no tendency of these (I'm fine with either ways).
> I've seen some usage like:
>
> static inline int device_is_registered(struct device *dev)
> {
> return dev->kobj.state_in_sysfs;
> }
>
> But I'm still not sure if we need to rely on such internal
> interface.. More thoughts?
Yeah... It seems that it is better to use some of the interfaces
provided by kobject,
otherwise we should still maintain this state in sbi.
Thanks,
Jianan
> Thanks,
> Gao Xiang
>> Thanks,
>> Jianan
>>
>>> struct kobject s_kobj; /* /sys/fs/erofs/<devname> */
>>> struct completion s_kobj_unregister;
>>> };
>>> diff --git a/fs/erofs/sysfs.c b/fs/erofs/sysfs.c
>>> index dac252bc9228..2b48a4df19b4 100644
>>> --- a/fs/erofs/sysfs.c
>>> +++ b/fs/erofs/sysfs.c
>>> @@ -209,6 +209,7 @@ int erofs_register_sysfs(struct super_block *sb)
>>> "%s", sb->s_id);
>>> if (err)
>>> goto put_sb_kobj;
>>> + sbi->s_sysfs_inited = true;
>>> return 0;
>>> put_sb_kobj:
>>> @@ -221,9 +222,11 @@ void erofs_unregister_sysfs(struct super_block *sb)
>>> {
>>> struct erofs_sb_info *sbi = EROFS_SB(sb);
>>> - kobject_del(&sbi->s_kobj);
>>> - kobject_put(&sbi->s_kobj);
>>> - wait_for_completion(&sbi->s_kobj_unregister);
>>> + if (sbi->s_sysfs_inited) {
>>> + kobject_del(&sbi->s_kobj);
>>> + kobject_put(&sbi->s_kobj);
>>> + wait_for_completion(&sbi->s_kobj_unregister);
>>> + }
>>> }
>>> int __init erofs_init_sysfs(void)
On Tue, Mar 15, 2022 at 7:05 PM Huang Jianan <[email protected]> wrote:
>
> 在 2022/3/15 18:55, Gao Xiang 写道:
> > On Tue, Mar 15, 2022 at 06:43:01PM +0800, Huang Jianan wrote:
> >> 在 2022/3/15 15:51, Dongliang Mu 写道:
> >>> From: Dongliang Mu <[email protected]>
> >>>
> >>> Syzkaller hit 'WARNING: kobject bug in erofs_unregister_sysfs'. This bug
> >>> is triggered by injecting fault in kobject_init_and_add of
> >>> erofs_unregister_sysfs.
> >>>
> >>> Fix this by remembering if kobject_init_and_add is successful.
> >>>
> >>> Note that I've tested the patch and the crash does not occur any more.
> >>>
> >>> Reported-by: syzkaller <[email protected]>
> >>> Signed-off-by: Dongliang Mu <[email protected]>
> >>> ---
> >>> fs/erofs/internal.h | 1 +
> >>> fs/erofs/sysfs.c | 9 ++++++---
> >>> 2 files changed, 7 insertions(+), 3 deletions(-)
> >>>
> >>> diff --git a/fs/erofs/internal.h b/fs/erofs/internal.h
> >>> index 5aa2cf2c2f80..9e20665e3f68 100644
> >>> --- a/fs/erofs/internal.h
> >>> +++ b/fs/erofs/internal.h
> >>> @@ -144,6 +144,7 @@ struct erofs_sb_info {
> >>> u32 feature_incompat;
> >>> /* sysfs support */
> >>> + bool s_sysfs_inited;
> >> Hi Dongliang,
> >>
> >> How about using sbi->s_kobj.state_in_sysfs to avoid adding a extra member in
> >> sbi ?
> > Ok, I have no tendency of these (I'm fine with either ways).
> > I've seen some usage like:
> >
> > static inline int device_is_registered(struct device *dev)
> > {
> > return dev->kobj.state_in_sysfs;
> > }
> >
> > But I'm still not sure if we need to rely on such internal
> > interface.. More thoughts?
>
> Yeah... It seems that it is better to use some of the interfaces
> provided by kobject,
> otherwise we should still maintain this state in sbi.
>
I am fine with either way. Let me know if you reach to an agreement.
> Thanks,
> Jianan
>
> > Thanks,
> > Gao Xiang
> >> Thanks,
> >> Jianan
> >>
> >>> struct kobject s_kobj; /* /sys/fs/erofs/<devname> */
> >>> struct completion s_kobj_unregister;
> >>> };
> >>> diff --git a/fs/erofs/sysfs.c b/fs/erofs/sysfs.c
> >>> index dac252bc9228..2b48a4df19b4 100644
> >>> --- a/fs/erofs/sysfs.c
> >>> +++ b/fs/erofs/sysfs.c
> >>> @@ -209,6 +209,7 @@ int erofs_register_sysfs(struct super_block *sb)
> >>> "%s", sb->s_id);
> >>> if (err)
> >>> goto put_sb_kobj;
> >>> + sbi->s_sysfs_inited = true;
> >>> return 0;
> >>> put_sb_kobj:
> >>> @@ -221,9 +222,11 @@ void erofs_unregister_sysfs(struct super_block *sb)
> >>> {
> >>> struct erofs_sb_info *sbi = EROFS_SB(sb);
> >>> - kobject_del(&sbi->s_kobj);
> >>> - kobject_put(&sbi->s_kobj);
> >>> - wait_for_completion(&sbi->s_kobj_unregister);
> >>> + if (sbi->s_sysfs_inited) {
> >>> + kobject_del(&sbi->s_kobj);
> >>> + kobject_put(&sbi->s_kobj);
> >>> + wait_for_completion(&sbi->s_kobj_unregister);
> >>> + }
> >>> }
> >>> int __init erofs_init_sysfs(void)
>