Re: kernel BUG at fs/sysfs/group.c:65!

[Weng Meiling]

On 2014/10/9 20:43, Weng Meiling wrote:
> Hi guys,
> 
> I see the mails you discussed the BUG at fs/sysfs/group.c:65! triggered by duplicated sysfs link.
> 
> the detail mail:
> 
> https://lkml.org/lkml/2013/3/8/370
> 
> but it seems the problems has no conclusion. In our environment, we triggered the bug too, but for error ENOENT:
> 
> we use 3.4 kernel, and do virtual disk device create / remove for many times. Before remove we can see the devices:
> 
> #ll /sys/devices/virtual/block/
> drwxr-xr-x 7 root root 0 Oct  6 09:17 sd-1a
> drwxr-xr-x 7 root root 0 Oct  6 09:17 sd-2a
> 
> when the two virtual devices were removed, the directory block/ was delete too.
> 
> after many times create / remove, the kernel trigger the bug (just the main call trace):
> 
> [ 3965.441713] WARNING: at /usr/src/packages/BUILD/linux-3.4/lib/kobject.c:202 kobject_add_internal+0x11f/0x280()
> [ 3965.441716] Hardware name: Romley
> [ 3965.441718] kobject_add_internal failed for sd-1a (error: -2 parent: block)
> 
> [ 3965.441817] Call Trace:
> [ 3965.441820]  [<ffffffff8103717a>] warn_slowpath_common+0x7a/0xb0
> [ 3965.441823]  [<ffffffff81037251>] warn_slowpath_fmt+0x41/0x50
> [ 3965.441826]  [<ffffffff81215e0f>] kobject_add_internal+0x11f/0x280
> [ 3965.441830]  [<ffffffff81216267>] kobject_add+0x67/0xc0
> [ 3965.441833]  [<ffffffff812d2305>] device_add+0x105/0x6d0
> [ 3965.441836]  [<ffffffff812d0dbc>] ? dev_set_name+0x3c/0x40
> [ 3965.441839]  [<ffffffff812030ac>] add_disk+0x1bc/0x490
> 
> [ 3965.441912] kernel BUG at /usr/src/packages/BUILD/linux-3.4/fs/sysfs/group.c:65!
> [ 3965.441915] invalid opcode: 0000 [#1] SMP
> [ 3965.686738] Call Trace:
> [ 3965.686743]  [<ffffffff811a677e>] sysfs_create_group+0xe/0x10
> [ 3965.686748]  [<ffffffff810cfb04>] blk_trace_init_sysfs+0x14/0x20
> [ 3965.686753]  [<ffffffff811fcabb>] blk_register_queue+0x3b/0x120
> [ 3965.686756]  [<ffffffff812030bc>] add_disk+0x1cc/0x490
> 
> from the error "kobject_add_internal failed for sd-1a (error: -2 parent: block)", we found that the first
> warning was caused by the disk device's parent_sd was null when it was added into sysfs:
> 
> int sysfs_create_dir(struct kobject * kobj)
> {
>     ...
>     if (kobj->parent)
>         parent_sd = kobj->parent->sd;
>     else
>         parent_sd = &sysfs_root;
> 
>     if (!parent_sd)
>         return -ENOENT;
>     ...
> }
> 
> The virtual disk device was not added into sysfs because of the above failure, and the kobj->sd was
> not set, then trigger the bug when creating attribute group under the device's directory:
> 
> static int internal_create_group(struct kobject *kobj, int update,
>                  const struct attribute_group *grp)
> {
>     ...
>     BUG_ON(!kobj || (!update && !kobj->sd));
>     ...
> }
> 
> Walk the code, it seems there maybe a race between block/ remove and virtual disk devices' register:
> 
> when the two virtual devices were removed, the block/ directory's refcount became 0, will into:
> 
> path0(remove the block/)                                             path1(register virtual device sd-1a)
> 
> kobject_del(){                                                       get_device_parent(){
> ...                                                                  ...
>    sysfs_remove_dir(kobj);  //kobj->sd=0                                 spin_lock(&dev->class->p->glue_dirs.list_lock);

this should be kobj->sd = NULL
> ...                                                      <=========      list_for_each_entry(k, &dev->class->p->glue_dirs.list, entry)  //get parent kobject from kset list
>    kobj_kset_leave(kobj);   //remove kobj from kset list             ...
> }                                                                    }
> 
> If getting parent object between " kobj->sd=0 " and "remove_kset_leave(kobj)", the sysfs_create_dir() will return ENOENT and trigger the BUG later.
> 
this should be kobj->sd = NULL

> The lastest kernel seems to be the same. But I am not familiar with block device, I am not sure whether the analysis is right or I am missing something.
> what do you think about this situation? Any suggestion is appreciative. Thanks!
> 
> 



--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/