lkml.org 
[lkml]   [2019]   [Dec]   [21]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: KASAN: use-after-free Read in io_wq_flush (2)
From
Date
On 12/21/19 7:30 AM, Hillf Danton wrote:
>
> On Fri, 20 Dec 2019 23:58:08 -0800
>> Hello,
>>
>> syzbot found the following crash on:
>>
>> HEAD commit: 7ddd09fc Add linux-next specific files for 20191220
>> git tree: linux-next
>> console output: https://syzkaller.appspot.com/x/log.txt?x=12e1823ee00000
>> kernel config: https://syzkaller.appspot.com/x/.config?x=f183b01c3088afc6
>> dashboard link: https://syzkaller.appspot.com/bug?extid=8e7705a7ae1bdce77c07
>> compiler: gcc (GCC) 9.0.0 20181231 (experimental)
>>
>> Unfortunately, I don't have any reproducer for this crash yet.
>>
>> IMPORTANT: if you fix the bug, please add the following tag to the commit:
>> Reported-by: syzbot+8e7705a7ae1bdce77c07@syzkaller.appspotmail.com
>>
>> ==================================================================
>> BUG: KASAN: use-after-free in io_wq_flush+0x1f7/0x210 fs/io-wq.c:1009
>> Read of size 8 at addr ffff8880a8453d00 by task kworker/0:1/12
>>
>> CPU: 0 PID: 12 Comm: kworker/0:1 Not tainted
>> 5.5.0-rc2-next-20191220-syzkaller #0
>> Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS
>> Google 01/01/2011
>> Workqueue: events io_ring_file_ref_switch
>> Call Trace:
>> __dump_stack lib/dump_stack.c:77 [inline]
>> dump_stack+0x197/0x210 lib/dump_stack.c:118
>> print_address_description.constprop.0.cold+0xd4/0x30b mm/kasan/report.c:374
>> __kasan_report.cold+0x1b/0x41 mm/kasan/report.c:506
>> kasan_report+0x12/0x20 mm/kasan/common.c:639
>> __asan_report_load8_noabort+0x14/0x20 mm/kasan/generic_report.c:135
>> io_wq_flush+0x1f7/0x210 fs/io-wq.c:1009
>> io_destruct_skb+0x8e/0xc0 fs/io_uring.c:4668
>> skb_release_head_state+0xeb/0x260 net/core/skbuff.c:652
>> skb_release_all+0x16/0x60 net/core/skbuff.c:663
>> __kfree_skb net/core/skbuff.c:679 [inline]
>> kfree_skb net/core/skbuff.c:697 [inline]
>> kfree_skb+0x101/0x420 net/core/skbuff.c:691
>> io_ring_file_put fs/io_uring.c:4836 [inline]
>> io_ring_file_ref_switch+0x68a/0xac0 fs/io_uring.c:4881
>> process_one_work+0x9af/0x1740 kernel/workqueue.c:2264
>> worker_thread+0x98/0xe40 kernel/workqueue.c:2410
>> kthread+0x361/0x430 kernel/kthread.c:255
>> ret_from_fork+0x24/0x30 arch/x86/entry/entry_64.S:352
>>
>> Allocated by task 9937:
>> save_stack+0x23/0x90 mm/kasan/common.c:72
>> set_track mm/kasan/common.c:80 [inline]
>> __kasan_kmalloc mm/kasan/common.c:513 [inline]
>> __kasan_kmalloc.constprop.0+0xcf/0xe0 mm/kasan/common.c:486
>> kasan_kmalloc+0x9/0x10 mm/kasan/common.c:527
>> kmem_cache_alloc_trace+0x158/0x790 mm/slab.c:3551
>> kmalloc include/linux/slab.h:555 [inline]
>> kzalloc include/linux/slab.h:669 [inline]
>> io_wq_create+0x52/0xa40 fs/io-wq.c:1024
>> io_sq_offload_start fs/io_uring.c:5244 [inline]
>> io_uring_create fs/io_uring.c:6002 [inline]
>> io_uring_setup+0xf4a/0x2080 fs/io_uring.c:6062
>> __do_sys_io_uring_setup fs/io_uring.c:6075 [inline]
>> __se_sys_io_uring_setup fs/io_uring.c:6072 [inline]
>> __x64_sys_io_uring_setup+0x54/0x80 fs/io_uring.c:6072
>> do_syscall_64+0xfa/0x790 arch/x86/entry/common.c:294
>> entry_SYSCALL_64_after_hwframe+0x49/0xbe
>>
>> Freed by task 9935:
>> save_stack+0x23/0x90 mm/kasan/common.c:72
>> set_track mm/kasan/common.c:80 [inline]
>> kasan_set_free_info mm/kasan/common.c:335 [inline]
>> __kasan_slab_free+0x102/0x150 mm/kasan/common.c:474
>> kasan_slab_free+0xe/0x10 mm/kasan/common.c:483
>> __cache_free mm/slab.c:3426 [inline]
>> kfree+0x10a/0x2c0 mm/slab.c:3757
>> io_wq_destroy+0x2ce/0x3c0 fs/io-wq.c:1116
>> io_finish_async+0x128/0x1b0 fs/io_uring.c:4657
>> io_ring_ctx_free fs/io_uring.c:5569 [inline]
>> io_ring_ctx_wait_and_kill+0x330/0x9a0 fs/io_uring.c:5644
>> io_uring_release+0x42/0x50 fs/io_uring.c:5652
>> __fput+0x2ff/0x890 fs/file_table.c:280
>> ____fput+0x16/0x20 fs/file_table.c:313
>> task_work_run+0x145/0x1c0 kernel/task_work.c:113
>> tracehook_notify_resume include/linux/tracehook.h:188 [inline]
>> exit_to_usermode_loop+0x316/0x380 arch/x86/entry/common.c:164
>> prepare_exit_to_usermode arch/x86/entry/common.c:195 [inline]
>> syscall_return_slowpath arch/x86/entry/common.c:278 [inline]
>> do_syscall_64+0x676/0x790 arch/x86/entry/common.c:304
>> entry_SYSCALL_64_after_hwframe+0x49/0xbe
>>
>> The buggy address belongs to the object at ffff8880a8453d00
>> which belongs to the cache kmalloc-192 of size 192
>> The buggy address is located 0 bytes inside of
>> 192-byte region [ffff8880a8453d00, ffff8880a8453dc0)
>> The buggy address belongs to the page:
>> page:ffffea0002a114c0 refcount:1 mapcount:0 mapping:ffff8880aa400000
>> index:0x0
>> raw: 00fffe0000000200 ffffea0002644808 ffffea0002482f08 ffff8880aa400000
>> raw: 0000000000000000 ffff8880a8453000 0000000100000010 0000000000000000
>> page dumped because: kasan: bad access detected
>>
>> Memory state around the buggy address:
>> ffff8880a8453c00: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
>> ffff8880a8453c80: 00 00 00 00 fc fc fc fc fc fc fc fc fc fc fc fc
>>> ffff8880a8453d00: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
>> ^
>> ffff8880a8453d80: fb fb fb fb fb fb fb fb fc fc fc fc fc fc fc fc
>> ffff8880a8453e00: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
>> ==================================================================
>
> Erase ctx's io_wq before destroying.
>
> --- a/fs/io_uring.c
> +++ b/fs/io_uring.c
> @@ -4651,12 +4651,13 @@ static void io_sq_thread_stop(struct io_
>
> static void io_finish_async(struct io_ring_ctx *ctx)
> {
> + struct io_wq *io_wq;
> +
> io_sq_thread_stop(ctx);
>
> - if (ctx->io_wq) {
> - io_wq_destroy(ctx->io_wq);
> - ctx->io_wq = NULL;
> - }
> + io_wq = xchg(&ctx->io_wq, NULL);
> + if (io_wq)
> + io_wq_destroy(io_wq);
> }
>
> #if defined(CONFIG_UNIX)

I actually think we can just kill the flush off the skb put path, it's
not needed.

--
Jens Axboe

\
 
 \ /
  Last update: 2019-12-21 15:39    [W:0.034 / U:2.152 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site