On 6/21/23 7:40?AM, Jens Axboe wrote: > On 6/21/23 1:38?AM, Guangwu Zhang wrote: >> HI, >> Found the io_req_local_work_add error when run liburing testing. >> >> kernel repo : >> Merge branch 'for-6.5/block' into for-next >> * for-6.5/block: >> reiserfs: fix blkdev_put() warning from release_journal_dev() >> >> [ 1733.389012] BUG: KASAN: out-of-bounds in io_req_local_work_add+0x3b1/0x4a0 >> [ 1733.395900] Read of size 4 at addr ffff888133320458 by task >> iou-wrk-97057/97138 >> [ 1733.403205] >> [ 1733.404706] CPU: 4 PID: 97138 Comm: iou-wrk-97057 Kdump: loaded Not >> tainted 6.4.0-rc3.kasan+ #1 >> [ 1733.413404] Hardware name: Dell Inc. PowerEdge R740/06WXJT, BIOS >> 2.13.3 12/13/2021 >> [ 1733.420972] Call Trace: >> [ 1733.423425] <TASK> >> [ 1733.425533] dump_stack_lvl+0x33/0x50 >> [ 1733.429207] print_address_description.constprop.0+0x2c/0x3e0 >> [ 1733.434959] print_report+0xb5/0x270 >> [ 1733.438539] ? kasan_addr_to_slab+0x9/0xa0 >> [ 1733.442639] ? io_req_local_work_add+0x3b1/0x4a0 >> [ 1733.447258] kasan_report+0xcf/0x100 >> [ 1733.450839] ? io_req_local_work_add+0x3b1/0x4a0 >> [ 1733.455456] io_req_local_work_add+0x3b1/0x4a0 >> [ 1733.459903] ? __pfx_io_req_local_work_add+0x10/0x10 >> [ 1733.464871] ? __schedule+0x616/0x1530 >> [ 1733.468622] __io_req_task_work_add+0x1bc/0x270 >> [ 1733.473156] io_issue_sqe+0x55a/0xe80 >> [ 1733.476831] io_wq_submit_work+0x23e/0xa00 >> [ 1733.480930] io_worker_handle_work+0x2f5/0xa80 >> [ 1733.485384] io_wq_worker+0x6c5/0x9d0 >> [ 1733.489051] ? __pfx_io_wq_worker+0x10/0x10 >> [ 1733.493246] ? _raw_spin_lock_irq+0x82/0xe0 >> [ 1733.497430] ? __pfx_io_wq_worker+0x10/0x10 >> [ 1733.501616] ret_from_fork+0x29/0x50 >> [ 1733.505204] </TASK> >> [ 1733.507396] >> [ 1733.508894] Allocated by task 97057: >> [ 1733.512475] kasan_save_stack+0x1e/0x40 >> [ 1733.516313] kasan_set_track+0x21/0x30 >> [ 1733.520068] __kasan_slab_alloc+0x83/0x90 >> [ 1733.524080] kmem_cache_alloc_bulk+0x13a/0x1e0 >> [ 1733.528526] __io_alloc_req_refill+0x238/0x510 >> [ 1733.532971] io_submit_sqes+0x65a/0xcd0 >> [ 1733.536810] __do_sys_io_uring_enter+0x4e9/0x830 >> [ 1733.541430] do_syscall_64+0x59/0x90 >> [ 1733.545010] entry_SYSCALL_64_after_hwframe+0x72/0xdc >> [ 1733.550071] >> [ 1733.551571] The buggy address belongs to the object at ffff8881333203c0 >> [ 1733.551571] which belongs to the cache io_kiocb of size 224 >> [ 1733.563816] The buggy address is located 152 bytes inside of >> [ 1733.563816] 224-byte region [ffff8881333203c0, ffff8881333204a0) >> [ 1733.575544] >> [ 1733.577042] The buggy address belongs to the physical page: >> [ 1733.582617] page:00000000edbe178c refcount:1 mapcount:0 >> mapping:0000000000000000 index:0x0 pfn:0x133320 >> [ 1733.592011] head:00000000edbe178c order:1 entire_mapcount:0 >> nr_pages_mapped:0 pincount:0 >> [ 1733.600096] memcg:ffff88810cd49001 >> [ 1733.603501] flags: >> 0x17ffffc0010200(slab|head|node=0|zone=2|lastcpupid=0x1fffff) >> [ 1733.610896] page_type: 0xffffffff() >> [ 1733.614390] raw: 0017ffffc0010200 ffff888101222280 ffffea0004473900 >> 0000000000000002 >> [ 1733.622128] raw: 0000000000000000 0000000000190019 00000001ffffffff >> ffff88810cd49001 >> [ 1733.629866] page dumped because: kasan: bad access detected >> [ 1733.635439] >> [ 1733.636938] Memory state around the buggy address: >> [ 1733.641731] ffff888133320300: 00 00 00 00 00 00 00 00 00 00 00 00 >> fc fc fc fc >> [ 1733.648952] ffff888133320380: fc fc fc fc fc fc fc fc 00 00 00 00 >> 00 00 00 00 >> [ 1733.656169] >ffff888133320400: 00 00 00 00 00 00 00 00 00 00 00 00 >> 00 00 00 00 >> [ 1733.663389] ^ >> [ 1733.669743] ffff888133320480: 00 00 00 00 fc fc fc fc fc fc fc fc >> fc fc fc fc >> [ 1733.676961] ffff888133320500: 00 00 00 00 00 00 00 00 00 00 00 00 >> 00 00 00 00 > > I appreciate you running tests and sending in failures, but can you > please be more specific about what exactly was run? We seem to need to > do this dance every time, which is just wasting time. So: > > 1) What test triggered this? > 2) Was it invoked with any arguments? > > In general, a good bug report should include exactly HOW you ended > up there. Another thing that's really handy in bug reports like this is a disassemble of where the bad memory access occured. We have this: BUG: KASAN: out-of-bounds in io_req_local_work_add+0x3b1/0x4a0 and you can look that up in gdb, for example. -- Jens Axboe
