Re: panic in range_tree_seg64_compare()

Matthew Macy wrote:
> On Thu, Aug 27, 2020 at 6:34 PM Yuri Pankov <ypankov_at_xsmail.com> wrote:
>>
>> Yet another issue I'm seeing after last update (currently running
>> r364870), hit it 2 times today:
>>
>> Fatal trap 12: page fault while in kernel mode
>> cpuid = 19; apic id = 0d
>> fault virtual address   = 0xfffff819e2ecdc40
>> fault code              = supervisor read data, page not present
>> instruction pointer     = 0x20:0xffffffff8277fa64
>> stack pointer           = 0x28:0xfffffe01f9ff2d90
>> frame pointer           = 0x28:0xfffffe01f9ff2d90
>> code segment            = base 0x0, limit 0xfffff, type 0x1b
>>                           = DPL 0, pres 1, long 1, def32 0, gran 1
>> processor eflags        = interrupt enabled, resume, IOPL = 0
>> current process         = 48792 (blk-3:0-0)
>> trap number             = 12
>> panic: page fault
>> cpuid = 19
>> time = 1598577675
>> KDB: stack backtrace:
>> db_trace_self_wrapper() at db_trace_self_wrapper+0x2b/frame
>> 0xfffffe01f9ff2a40
>> vpanic() at vpanic+0x182/frame 0xfffffe01f9ff2a90
>> panic() at panic+0x43/frame 0xfffffe01f9ff2af0
>> trap_fatal() at trap_fatal+0x387/frame 0xfffffe01f9ff2b50
>> trap_pfault() at trap_pfault+0x97/frame 0xfffffe01f9ff2bb0
>> trap() at trap+0x2ab/frame 0xfffffe01f9ff2cc0
>> calltrap() at calltrap+0x8/frame 0xfffffe01f9ff2cc0
>> --- trap 0xc, rip = 0xffffffff8277fa64, rsp = 0xfffffe01f9ff2d90, rbp =
>> 0xfffffe01f9ff2d90 ---
>> range_tree_seg64_compare() at range_tree_seg64_compare+0x4/frame
>> 0xfffffe01f9ff2d90
>> zfs_btree_find() at zfs_btree_find+0x1bd/frame 0xfffffe01f9ff2df0
>> range_tree_find_impl() at range_tree_find_impl+0x6e/frame 0xfffffe01f9ff2e30
>> range_tree_find() at range_tree_find+0x1c/frame 0xfffffe01f9ff2e70
>> range_tree_contains() at range_tree_contains+0x9/frame 0xfffffe01f9ff2e80
>> dnode_block_freed() at dnode_block_freed+0x11d/frame 0xfffffe01f9ff2eb0
>> dbuf_read() at dbuf_read+0x70c/frame 0xfffffe01f9ff2fc0
>> dmu_buf_hold_array_by_dnode() at dmu_buf_hold_array_by_dnode+0x164/frame
>> 0xfffffe01f9ff3030
>> dmu_read_impl() at dmu_read_impl+0xce/frame 0xfffffe01f9ff30c0
>> dmu_read() at dmu_read+0x45/frame 0xfffffe01f9ff3100
>> zvol_geom_bio_strategy() at zvol_geom_bio_strategy+0x2aa/frame
>> 0xfffffe01f9ff3180
>> g_io_request() at g_io_request+0x2df/frame 0xfffffe01f9ff31b0
>> g_dev_strategy() at g_dev_strategy+0x155/frame 0xfffffe01f9ff31e0
>> physio() at physio+0x4f8/frame 0xfffffe01f9ff3270
>> devfs_read_f() at devfs_read_f+0xde/frame 0xfffffe01f9ff32d0
>> dofileread() at dofileread+0x81/frame 0xfffffe01f9ff3320
>> kern_preadv() at kern_preadv+0x62/frame 0xfffffe01f9ff3360
>> sys_preadv() at sys_preadv+0x39/frame 0xfffffe01f9ff3390
>> amd64_syscall() at amd64_syscall+0x140/frame 0xfffffe01f9ff34b0
>> fast_syscall_common() at fast_syscall_common+0xf8/frame 0xfffffe01f9ff34b0
>> --- syscall (289, FreeBSD ELF64, sys_preadv), rip = 0x8006fd89a, rsp =
>> 0x7fffdfdfcf18, rbp = 0x7fffdfdfcfc0 ---
>> Uptime: 4h13m43s
> 
> 
>>
>> Guessing on zvol_geom_bio_strategy(), it's volmode=dev zvol I'm using
>> for bhyve VM.  Anything known?
> 
> Not really. A reproduction scenario would be very helpful. This was
> seen once by someone at iX - I committed some additional asserts to
> the truenas tree, but haven't heard further.
> 
> +++ b/module/zfs/dbuf.c
> _at__at_ -3192,7 +3192,7 _at__at_
> dbuf_dirty_leaf_with_existing_frontend(dbuf_dirty_state_t *dds)
>                           * scheduled its write with its buffer, we must
>                           * disassociate by replacing the frontend.
>                           */
> -                       ASSERT(db->db_state & (DB_READ|DB_PARTIAL));
> +                       ASSERT3U(db->db_state, &, (DB_READ|DB_PARTIAL));
>                          ASSERT3U(db->db_dirtycnt, ==, 1);
>                          dbuf_dirty_set_data(dds);
>                  } else {
> _at__at_ -3238,18 +3238,24 _at__at_ dbuf_dirty_record_create_leaf(dbuf_dirty_state_t *dds)
> 
>          dr = dbuf_dirty_record_create(dds);
> 
> +       /*
> +        * XXX - convert to ASSERT after dn_free_ranges fix
> +        */
> +       VERIFY(db->db_level == 0);
> +       VERIFY(db->db_blkid != DMU_BONUS_BLKID);

Can't find context for both chunks, there are simply no such functions 
in sys/contrib/openzfs/module/zfs/dbuf.c, and yes, note that I'm running 
the in-tree version.