On 2020/1/13 19:04, Ming Lei wrote:
> On Fri, Jan 10, 2020 at 01:10:20PM +0800, Zhiqiang Liu wrote:
>>
>> In brd_init func, rd_nr num of brd_device are firstly allocated
>> and add in brd_devices, then brd_devices are traversed to add each
>> brd_device by calling add_disk func. When allocating brd_device,
>> the disk->first_minor is set to i * max_part, if rd_nr * max_part
>> is larger than MINORMASK, two different brd_device may have the same
>> devt, then only one of them can be successfully added.
>> when rmmod brd.ko, it will cause oops when calling brd_exit.
>>
>> Follow those steps:
>> # modprobe brd rd_nr=3 rd_size=102400 max_part=1048576
>> # rmmod brd
>> then, the oops will appear.
>>
>> Oops log:
>> [ 726.613722] Call trace:
>> [ 726.614175] kernfs_find_ns+0x24/0x130
>> [ 726.614852] kernfs_find_and_get_ns+0x44/0x68
>> [ 726.615749] sysfs_remove_group+0x38/0xb0
>> [ 726.616520] blk_trace_remove_sysfs+0x1c/0x28
>> [ 726.617320] blk_unregister_queue+0x98/0x100
>> [ 726.618105] del_gendisk+0x144/0x2b8
>> [ 726.618759] brd_exit+0x68/0x560 [brd]
>> [ 726.619501] __arm64_sys_delete_module+0x19c/0x2a0
>> [ 726.620384] el0_svc_common+0x78/0x130
>> [ 726.621057] el0_svc_handler+0x38/0x78
>> [ 726.621738] el0_svc+0x8/0xc
>> [ 726.622259] Code: aa0203f6 aa0103f7 aa1e03e0 d503201f (7940e260)
>>
>> Here, we add brd_check_par_valid func to check parameter
>> validation before register_blkdev func.
>>
>> Signed-off-by: Zhiqiang Liu <liuzhiqiang26@xxxxxxxxxx>
>> ---
>> drivers/block/brd.c | 33 ++++++++++++++++++++++++++-------
>> +static inline int brd_check_par_valid(void)
>> +{
>> + if (unlikely(!rd_nr))
>> + rd_nr = 1;
>> +
>> + if (unlikely(!max_part))
>> + max_part = 1;
>> +
>> + if (rd_nr * max_part > MINORMASK)
>> + return -EINVAL;
>> +
>> + return 0;
>> +
>> +}
>> +
>> static int __init brd_init(void)
>> {
>> struct brd_device *brd, *next;
>> - int i;
>> + int i, ret;
>>
>> /*
>> * brd module now has a feature to instantiate underlying device
>> @@ -488,11 +503,15 @@ static int __init brd_init(void)
>> * dynamically.
>> */
>>
>> + ret = brd_check_par_valid();
>> + if (ret) {
>> + pr_info("brd: invalid parameter setting!!!\n");
>> + return ret;
>> + }
>> +
>
> The max supported partition number is 256, see __alloc_disk_node().
> So even though one bigger number is passed to alloc_disk(), at most
> 256 partitions are allowed on that disk. Maybe you can apply the
> following way to avoid the issue:
>
> disk->first_minor = i * disk->minors;
>
> However, looks 'rd_nr' still needs to be validated(rd_nr < 2 ^ 23).
>
Thanks for your reply.
As you said, minors is limited to DISK_MAX_PARTS in __alloc_disk_node when
calling alloc_disk in brd_alloc func. We can set: disk->first_minor = i * disk->minors
as your suggestion.
As for 'rd_nr', I think we should make sure that 'disk->first_minor' should be not
larger than MINORMASK.
->add_disk
->device_add_disk
->__device_add_disk
->blk_alloc_devt
->MKDEV
If rd_nr > MINORMASK / min(max_part, DISK_MAX_PARTS), two different brd_device may have the same devt
by calling MKDEV in blk_alloc_devt func. For example, MKDEV(1, 0) is equal to MKDEV(1, MINORMASK + 1).
So we should check both rd_nr and max_part as follows,
static inline int brd_check_par_valid(void)
{
if (unlikely(!rd_nr))
rd_nr = 1;
if (unlikely(!max_part))
max_part = 1;
if (max_part > DISK_MAX_PARTS)
max_part = DISK_MAX_PARTS;
if (rd_nr > MINORMASK / max_part)
return -EINVAL;
return 0;
}
I will send the v2 patch.
