Re: [syzbot] [mm?] KCSAN: data-race in mas_wr_store_entry / mtree_range_walk (2)

["Liam R. Howlett" <Liam.Howlett@oracle.com>]

* Marco Elver <elver@google.com> [260417 20:26]:
> On Sat, 18 Apr 2026 at 01:51, 'Liam R. Howlett' via syzkaller-bugs
> <syzkaller-bugs@googlegroups.com> wrote:
> >
> > * syzbot <syzbot+38a879f4a73497f2dfef@syzkaller.appspotmail.com> [260417 05:12]:
> > > Hello,
> > >
> > > syzbot found the following issue on:
> > >
> > > HEAD commit:    1d51b370a0f8 Merge tag 'jfs-7.1' of github.com:kleikamp/li..
> > > git tree:       upstream
> > > console output: https://syzkaller.appspot.com/x/log.txt?x=117dc4ce580000
> > > kernel config:  https://syzkaller.appspot.com/x/.config?x=7f207c4b1fbf85a3
> > > dashboard link: https://syzkaller.appspot.com/bug?extid=38a879f4a73497f2dfef
> > > compiler:       Debian clang version 21.1.8 (++20251221033036+2078da43e25a-1~exp1~20251221153213.50), Debian LLD 21.1.8
> > >
> > > Unfortunately, I don't have any reproducer for this issue yet.
> >
> > ... and you won't.  This will work unless we tear aligned unsigned long
> > writes/reads.
> >
> > I'm debating marking these as data_race().  Marking them all as
> > READ_ONCE and this one write as WRITE_ONCE.  It seems overkill for
> > something that won't happen.
> >
> > Alternatively, I can move the slot store fast path to need an
> > allocation, but that's worse.
> 
> The writer:
> 
> >   rcu_assign_pointer(slots[offset + 1], wr_mas->entry);
> >   wr_mas->pivots[offset] = mas->index - 1;  // <-- stores pivots[offset]
> 
> The reader races here:
> 
> >   if (pivots[offset] >= mas->index) {   // <-- load pivots[offset]
> >       max = pivots[offset];  // <-- load pivots[offset] again
> >       break;
> >   }
> 
> The compiler is free to reload them as written. What if there's a
> concurrent update between the first and second load?

Then the benign race has happened.

Looking at [1], we see that care has been taken to limit the slot store
code to only !rcu mode, except for a subset of cases.  Digging through
the information in git will eventually lead you to this note Peng wrote:

commit 64891ba3e51fb841b0af70db029038eb93bd5a43
Author: Peng Zhang <zhangpeng.00@bytedance.com>
Date:   Wed Jun 28 15:36:57 2023 +0800

    maple_tree: add a fast path case in mas_wr_slot_store()
    
    When expanding a range in two directions, only partially overwriting the
    previous and next ranges, the number of entries will not be increased, so
    we can just update the pivots as a fast path. However, it may introduce
    potential risks in RCU mode, because it updates two pivots. We only
    enable it in non-RCU mode.
    
    Link: https://lkml.kernel.org/r/20230628073657.75314-5-zhangpeng.00@bytedance.com
    Signed-off-by: Peng Zhang <zhangpeng.00@bytedance.com>
    Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>
    Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

So, you can see that the author of the initial code did look at race
conditions.  I wanted to read the link for more information but that
link isn't working right now (403 error).

-----

Or, we can ask an LLM about it:

In mas_wr_store_type(), we only allow wr_slot_store in RCU mode for the narrow
case where wr_mas->offset_end - mas->offset == 1. That condition means the
update touches only one boundary between two adjacent ranges, so the in-place
mutation in mas_wr_slot_store() stays limited to a single slot/pivot boundary
update and is considered safe for lockless readers.

If the span is wider than that, we do not use in-place slot-store under RCU.
The broader in-place path in mas_wr_slot_store() is explicitly guarded with
WARN_ON_ONCE(mt_in_rcu(...)), and the store-type logic instead falls back to
node-replacement paths (wr_node_store/split/rebalance), which preserve RCU
reader safety by publishing a new node rather than mutating too much in place.

In non-RCU mode (!mt_in_rcu()), we allow the wider in-place cases because
readers are expected to be synchronized by locking, so the stricter
lockless-reader constraints do not apply.

-----

I am sorry, but I don't have time to work through the scenarios as this
is not an issue and I no longer have the time budget for mailing lists
as I once did.

If you can come up with a problem (and ideally a reproducer), then
please let me know.

[1]. https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/lib/maple_tree.c?id=c1f49dea2b8f335813d3b348fd39117fb8efb428#n3696


> 
> > > Downloadable assets:
> > > disk image: https://storage.googleapis.com/syzbot-assets/e08ff8d2b0e5/disk-1d51b370.raw.xz
> > > vmlinux: https://storage.googleapis.com/syzbot-assets/c11d4b098bbf/vmlinux-1d51b370.xz
> > > kernel image: https://storage.googleapis.com/syzbot-assets/6a4691f32e3d/bzImage-1d51b370.xz
> > >
> > > IMPORTANT: if you fix the issue, please add the following tag to the commit:
> > > Reported-by: syzbot+38a879f4a73497f2dfef@syzkaller.appspotmail.com
> > >
> > > ==================================================================
> > > BUG: KCSAN: data-race in mas_wr_store_entry / mtree_range_walk
> > >
> > > write to 0xffff888104f71d08 of 8 bytes by task 4757 on cpu 0:
> > >  mas_wr_slot_store lib/maple_tree.c:3232 [inline]
> > >  mas_wr_store_entry+0x3405/0x5ad0 lib/maple_tree.c:3528
> > >  mas_store_prealloc+0x43e/0x690 lib/maple_tree.c:4936
> > >  vma_iter_store_overwrite mm/vma.h:616 [inline]
> > >  commit_merge+0x6a1/0x720 mm/vma.c:766
> > >  vma_expand+0x301/0x460 mm/vma.c:1219
> > >  vma_merge_new_range+0x29c/0x320 mm/vma.c:1112
> > >  __mmap_region mm/vma.c:2766 [inline]
> > >  mmap_region+0x1073/0x2110 mm/vma.c:2856
> > >  do_mmap+0x9b2/0xbd0 mm/mmap.c:560
> > >  vm_mmap_pgoff+0x183/0x2d0 mm/util.c:581
> > >  ksys_mmap_pgoff+0xc1/0x310 mm/mmap.c:606
> > >  x64_sys_call+0x14df/0x3020 arch/x86/include/generated/asm/syscalls_64.h:10
> > >  do_syscall_x64 arch/x86/entry/syscall_64.c:63 [inline]
> > >  do_syscall_64+0x12c/0x3b0 arch/x86/entry/syscall_64.c:94
> > >  entry_SYSCALL_64_after_hwframe+0x77/0x7f
> > >
> > > read to 0xffff888104f71d08 of 8 bytes by task 4759 on cpu 1:
> > >  mtree_range_walk+0x1a6/0x490 lib/maple_tree.c:2032
> > >  mas_state_walk lib/maple_tree.c:2952 [inline]
> > >  mas_walk+0x1cc/0x370 lib/maple_tree.c:4366
> > >  lock_vma_under_rcu+0xc9/0x210 mm/mmap_lock.c:304
> > >  do_user_addr_fault+0x232/0x1050 arch/x86/mm/fault.c:1325
> > >  handle_page_fault arch/x86/mm/fault.c:1474 [inline]
> > >  exc_page_fault+0x62/0xa0 arch/x86/mm/fault.c:1527
> > >  asm_exc_page_fault+0x26/0x30 arch/x86/include/asm/idtentry.h:618
> > >
> > > value changed: 0x00007f68dc2a5fff -> 0x00007f68dc284fff
> > >
> > > Reported by Kernel Concurrency Sanitizer on:
> > > CPU: 1 UID: 0 PID: 4759 Comm: syz.5.348 Not tainted syzkaller #0 PREEMPT(full)
> > > Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 03/18/2026
> > > ==================================================================
> > > netlink: 64 bytes leftover after parsing attributes in process `syz.5.348'.
> > >
> > >
> > > ---
> > > This report is generated by a bot. It may contain errors.
> > > See https://goo.gl/tpsmEJ for more information about syzbot.
> > > syzbot engineers can be reached at syzkaller@googlegroups.com.
> > >
> > > syzbot will keep track of this issue. See:
> > > https://goo.gl/tpsmEJ#status for how to communicate with syzbot.
> > >
> > > If the report is already addressed, let syzbot know by replying with:
> > > #syz fix: exact-commit-title
> > >
> > > If you want to overwrite report's subsystems, reply with:
> > > #syz set subsystems: new-subsystem
> > > (See the list of subsystem names on the web dashboard)
> > >
> > > If the report is a duplicate of another one, reply with:
> > > #syz dup: exact-subject-of-another-report
> > >
> > > If you want to undo deduplication, reply with:
> > > #syz undup
>