Re: [PATCH 6/6] rq-qos: fix uaf in rq_qos_done_io()

["yukuai (C)" <yukuai3@huawei.com>]

On 2021/09/24 9:49, Ming Lei wrote:
> On Fri, Sep 24, 2021 at 09:23:42AM +0800, yukuai (C) wrote:
>> On 2021/09/24 8:41, Ming Lei wrote:
>>> On Thu, Sep 23, 2021 at 09:46:31PM +0800, Yu Kuai wrote:
>>>> our test report a uaf:
>>>>
>>>> [  142.925504] ==================================================================
>>>> [  142.929084] BUG: KASAN: use-after-free in __rq_qos_done_bio+0x57/0x90
>>>> [  142.931131] Read of size 8 at addr ffff88810306d858 by task blkdiscard/858
>>>> [  142.933289]
>>>> [  142.933798] CPU: 1 PID: 858 Comm: blkdiscard Not tainted 5.15.0-rc1-00004-g18bc2dec41ab-d4
>>>> [  142.936580] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS ?-20190727_0738364
>>>> [  142.939318] Call Trace:
>>>> [  142.939662]  ? dump_stack_lvl+0x73/0x9f
>>>> [  142.940197]  ? print_address_description.constprop.0+0x2f/0x250
>>>> [  142.941004]  ? __rq_qos_done_bio+0x57/0x90
>>>> [  142.941564]  ? __rq_qos_done_bio+0x57/0x90
>>>> [  142.942132]  ? kasan_report.cold+0x81/0x165
>>>> [  142.942710]  ? __rq_qos_done_bio+0x57/0x90
>>>> [  142.943282]  ? __asan_load8+0x74/0x110
>>>> [  142.943798]  ? __rq_qos_done_bio+0x57/0x90
>>>> [  142.944365]  ? bio_endio+0x142/0x430
>>>> [  142.944864]  ? submit_bio_checks+0x178/0xef0
>>>> [  142.945456]  ? trace_event_raw_event_block_rq_requeue+0x300/0x300
>>>> [  142.946283]  ? mempool_alloc+0xe9/0x2f0
>>>> [  142.946812]  ? remove_element+0x130/0x130
>>>> [  142.947371]  ? init_timer_key+0x83/0x1b0
>>>> [  142.947917]  ? submit_bio_noacct+0x86/0x9c0
>>>> [  142.948496]  ? blk_queue_enter+0x6d0/0x6d0
>>>> [  142.949066]  ? bio_alloc_bioset+0x1b2/0x3a0
>>>> [  142.949649]  ? __rcu_read_unlock+0x45/0x370
>>>> [  142.950227]  ? bvec_alloc+0x120/0x120
>>>> [  142.950732]  ? submit_bio+0x60/0x230
>>>> [  142.951230]  ? blk_next_bio+0x4f/0x70
>>>> [  142.951740]  ? __blkdev_issue_discard+0x257/0x520
>>>> [  142.952387]  ? __blkdev_issue_write_zeroes+0x270/0x270
>>>> [  142.953089]  ? bd_abort_claiming+0x70/0x70
>>>> [  142.953652]  ? __kasan_check_write+0x20/0x30
>>>> [  142.954236]  ? _raw_spin_lock+0xaf/0x130
>>>> [  142.954769]  ? _raw_read_lock_bh+0xa0/0xa0
>>>> [  142.955328]  ? __get_locked_pte+0x1b3/0x310
>>>> [  142.955897]  ? _raw_spin_unlock+0x3b/0x80
>>>> [  142.956444]  ? blkdev_issue_discard+0xd3/0x1a0
>>>> [  142.957051]  ? blkdev_issue_write_same+0x540/0x540
>>>> [  142.957708]  ? _raw_spin_lock+0xaf/0x130
>>>> [  142.958244]  ? bd_abort_claiming+0x70/0x70
>>>> [  142.958805]  ? wake_up_bit+0x46/0x50
>>>> [  142.959302]  ? preempt_count_sub+0x14/0x160
>>>> [  142.959877]  ? _raw_spin_unlock+0x3b/0x80
>>>> [  142.960428]  ? bd_abort_claiming+0x65/0x70
>>>> [  142.960993]  ? blk_ioctl_discard+0x1bd/0x240
>>>> [  142.961582]  ? blkdev_bszset+0x1c0/0x1c0
>>>> [  142.962118]  ? special_mapping_fault+0x6f/0x200
>>>> [  142.962743]  ? __do_fault+0x80/0x410
>>>> [  142.963241]  ? blkdev_common_ioctl+0x6c9/0x1190
>>>> [  142.963877]  ? ioctl_file_clone+0x110/0x110
>>>> [  142.964457]  ? blk_ioctl_discard+0x240/0x240
>>>> [  142.965038]  ? copy_page_range+0x2b60/0x2b60
>>>> [  142.965623]  ? vfs_getattr_nosec+0x177/0x190
>>>> [  142.966214]  ? __ia32_compat_sys_newfstat+0x40/0x40
>>>> [  142.966885]  ? blkdev_ioctl+0x180/0x4b0
>>>> [  142.967409]  ? blkdev_common_ioctl+0x1190/0x1190
>>>> [  142.968033]  ? handle_mm_fault+0x3c2/0x660
>>>> [  142.968590]  ? __kasan_check_write+0x20/0x30
>>>> [  142.969172]  ? block_ioctl+0x7d/0xa0
>>>> [  142.969666]  ? __x64_sys_ioctl+0xd5/0x150
>>>> [  142.970224]  ? do_syscall_64+0x35/0x80
>>>> [  142.970733]  ? entry_SYSCALL_64_after_hwframe+0x44/0xae
>>>> [  142.971441]
>>>> [  142.971653] Allocated by task 283:
>>>> [  142.972117]  kasan_save_stack+0x23/0x60
>>>> [  142.972637]  set_alloc_info+0x46/0x70
>>>> [  142.973136]  __kasan_kmalloc+0x8d/0xd0
>>>> [  142.973639]  kmem_cache_alloc_trace+0x3e7/0x820
>>>> [  142.974254]  wbt_init+0x40/0x430
>>>> [  142.974694]  wbt_enable_default+0xbb/0x100
>>>> [  142.975248]  blk_register_queue+0x216/0x3e0
>>>> [  142.975812]  device_add_disk+0x4ac/0x880
>>>> [  142.976358]  sd_probe+0x690/0x910
>>>> [  142.976809]  really_probe+0x5c3/0x800
>>>> [  142.977306]  __driver_probe_device+0x233/0x330
>>>> [  142.977907]  driver_probe_device+0x69/0x140
>>>> [  142.978466]  __device_attach_driver+0x125/0x210
>>>> [  142.979081]  bus_for_each_drv+0x10e/0x1b0
>>>> [  142.979615]  __device_attach_async_helper+0x175/0x230
>>>> [  142.980302]  async_run_entry_fn+0x7b/0x310
>>>> [  142.980859]  process_one_work+0x46a/0xa80
>>>> [  142.981400]  worker_thread+0x33d/0x8d0
>>>> [  142.981917]  kthread+0x282/0x300
>>>> [  142.982363]  ret_from_fork+0x1f/0x30
>>>> [  142.982862]
>>>> [  142.983077] Freed by task 863:
>>>> [  142.983501]  kasan_save_stack+0x23/0x60
>>>> [  142.984029]  kasan_set_track+0x24/0x40
>>>> [  142.984547]  kasan_set_free_info+0x30/0x60
>>>> [  142.985115]  __kasan_slab_free+0x137/0x210
>>>> [  142.985678]  kfree+0x10b/0x570
>>>> [  142.986106]  wbt_exit+0x68/0x80
>>>> [  142.986535]  rq_qos_exit+0x5f/0x80
>>>> [  142.987002]  blk_cleanup_queue+0xdb/0x250
>>>> [  142.987546]  __scsi_remove_device+0xb1/0x2e0
>>>> [  142.988131]  scsi_remove_device+0x38/0x60
>>>> [  142.988676]  sdev_store_delete+0x73/0x100
>>>> [  142.989230]  dev_attr_store+0x40/0x70
>>>> [  142.989730]  sysfs_kf_write+0x89/0xc0
>>>> [  142.990233]  kernfs_fop_write_iter+0x21d/0x340
>>>> [  142.990839]  new_sync_write+0x27e/0x3a0
>>>> [  142.991362]  vfs_write+0x46e/0x630
>>>> [  142.991834]  ksys_write+0xcd/0x1e0
>>>> [  142.992300]  __x64_sys_write+0x46/0x60
>>>> [  142.992814]  do_syscall_64+0x35/0x80
>>>> [  142.993311]  entry_SYSCALL_64_after_hwframe+0x44/0xae
>>>> [  142.994213] The buggy address belongs to the object at ffff88810306d800
>>>> [  142.994213]  which belongs to the cache kmalloc-256 of size 256
>>>> [  142.995889] The buggy address is located 88 bytes inside of
>>>> [  142.995889]  256-byte region [ffff88810306d800, ffff88810306d900)
>>>> [  142.997448] The buggy address belongs to the page:
>>>> [  142.998102] page:0000000069471149 refcount:1 mapcount:0 mapping:0000000000000000 index:0xc
>>>> [  142.999372] head:0000000069471149 order:2 compound_mapcount:0 compound_pincount:0
>>>> [  143.000375] flags: 0x2fffff80010200(slab|head|node=0|zone=2|lastcpupid=0x1fffff)
>>>> [  143.001403] raw: 002fffff80010200 0000000000000000 0000000100000001 ffff88810004cb40
>>>> [  143.002455] raw: 0000000000000000 0000000000200020 00000001ffffffff 0000000000000000
>>>> [  143.003477] page dumped because: kasan: bad access detected
>>>> [  143.004222]
>>>> [  143.004433] Memory state around the buggy address:
>>>> [  143.005077]  ffff88810306d700: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
>>>> [  143.006040]  ffff88810306d780: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
>>>> [  143.007012] >ffff88810306d800: fa fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
>>>> [  143.007981]                                                     ^
>>>> [  143.008795]  ffff88810306d880: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
>>>> [  143.009764]  ffff88810306d900: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
>>>> [  143.010731] ==================================================================
>>>>
>>>> This is because 'q_usage_counter' will not hold when bio_endio() is
>>>> called from error path, thus bio_endio() can concurrent with
>>>> blk_cleanup_queue():
>>>
>>> What is the exact error path? We actually grabs one ref of q_usage_counter
>>> during submitting bio, so the issue should have been fixed by not
>>> releasing the refcount early in the error path? Or the refcnt isn't grabbed
>>> yet when handling the error?
>>>
>>
>> Hi,
>>
>> We found at least two places:
>>
>> The first is error path from submit_bio_checks(), and we succeed to
>> construct repoducer here.
>>
>> The second is from bio_queue_enter(), bio_endio() will be called if
>> blk_queue_enter() failed.
> 
> OK, both should be addressed by the following simple patch:
> 
> 
> diff --git a/block/bio.c b/block/bio.c
> index 5df3dd282e40..a6fb6a0b4295 100644
> --- a/block/bio.c
> +++ b/block/bio.c
> @@ -1466,7 +1466,7 @@ void bio_endio(struct bio *bio)
>   	if (!bio_integrity_endio(bio))
>   		return;
>   
> -	if (bio->bi_bdev)
> +	if (bio->bi_bdev && bio_flagged(bio, BIO_TRACKED))
>   		rq_qos_done_bio(bio->bi_bdev->bd_disk->queue, bio);
>   
>   	if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) {
> 

Hi,

This patch looks good to me

Thanks,
Kuai