Re: memory leak in inotify_update_watch

[Dmitry Vyukov <dvyukov@google.com>]

On Tue, Jul 7, 2020 at 8:17 PM Catalin Marinas <catalin.marinas@arm.com> wrote:
>
> On Tue, Jul 07, 2020 at 05:24:11PM +0200, Jan Kara wrote:
> > On Mon 06-07-20 08:42:24, syzbot wrote:
> > > syzbot found the following crash on:
> > >
> > > HEAD commit:    7cc2a8ea Merge tag 'block-5.8-2020-07-01' of git://git.ker..
> > > git tree:       upstream
> > > console output: https://syzkaller.appspot.com/x/log.txt?x=17644c05100000
> > > kernel config:  https://syzkaller.appspot.com/x/.config?x=5ee23b9caef4e07a
> > > dashboard link: https://syzkaller.appspot.com/bug?extid=dec34b033b3479b9ef13
> > > compiler:       gcc (GCC) 10.1.0-syz 20200507
> > > syz repro:      https://syzkaller.appspot.com/x/repro.syz?x=1478a67b100000
> > >
> > > IMPORTANT: if you fix the bug, please add the following tag to the commit:
> > > Reported-by: syzbot+dec34b033b3479b9ef13@syzkaller.appspotmail.com
> > >
> > > BUG: memory leak
> > > unreferenced object 0xffff888115db8480 (size 576):
> > >   comm "systemd-udevd", pid 11037, jiffies 4295104591 (age 56.960s)
> > >   hex dump (first 32 bytes):
> > >     00 04 00 00 00 00 00 00 80 fd e8 15 81 88 ff ff  ................
> > >     a0 02 dd 20 81 88 ff ff b0 81 d0 09 81 88 ff ff  ... ............
> > >   backtrace:
> > >     [<00000000288c0066>] radix_tree_node_alloc.constprop.0+0xc1/0x140 lib/radix-tree.c:252
> > >     [<00000000f80ba6a7>] idr_get_free+0x231/0x3b0 lib/radix-tree.c:1505
> > >     [<00000000ec9ab938>] idr_alloc_u32+0x91/0x120 lib/idr.c:46
> > >     [<00000000aea98d29>] idr_alloc_cyclic+0x84/0x110 lib/idr.c:125
> > >     [<00000000dbad44a4>] inotify_add_to_idr fs/notify/inotify/inotify_user.c:365 [inline]
> > >     [<00000000dbad44a4>] inotify_new_watch fs/notify/inotify/inotify_user.c:578 [inline]
> > >     [<00000000dbad44a4>] inotify_update_watch+0x1af/0x2d0 fs/notify/inotify/inotify_user.c:617
> > >     [<00000000e141890d>] __do_sys_inotify_add_watch fs/notify/inotify/inotify_user.c:755 [inline]
> > >     [<00000000e141890d>] __se_sys_inotify_add_watch fs/notify/inotify/inotify_user.c:698 [inline]
> > >     [<00000000e141890d>] __x64_sys_inotify_add_watch+0x12f/0x180 fs/notify/inotify/inotify_user.c:698
> > >     [<00000000d872d7cc>] do_syscall_64+0x4c/0xe0 arch/x86/entry/common.c:359
> > >     [<000000005c62d8da>] entry_SYSCALL_64_after_hwframe+0x44/0xa9
> >
> > I've been looking into this for a while and I don't think this is related
> > to inotify at all. Firstly the reproducer looks totally benign:
> >
> > prlimit64(0x0, 0xe, &(0x7f0000000280)={0x9, 0x8d}, 0x0)
> > sched_setattr(0x0, &(0x7f00000000c0)={0x38, 0x2, 0x0, 0x0, 0x9}, 0x0)
> > vmsplice(0xffffffffffffffff, 0x0, 0x0, 0x0)
> > perf_event_open(0x0, 0x0, 0xffffffffffffffff, 0xffffffffffffffff, 0x0)
> > clone(0x20000103, 0x0, 0xfffffffffffffffe, 0x0, 0xffffffffffffffff)
> > syz_mount_image$vfat(0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0)
> >
> > So we seem to set SCHED_RR class and prio 9 to itself, the rest of syscalls
> > seem to be invalid and should fail. Secondly, the kernel log shows that we
> > hit OOM killer frequently and after one of these kills, many leaked objects
> > (among them this radix tree node from inotify idr) are reported. I'm not
> > sure if it could be the leak detector getting confused (e.g. because it got
> > ENOMEM at some point) or something else... Catalin, any idea?
>
> Kmemleak never performs well under heavy load. Normally you'd need to
> let the system settle for a bit before checking whether the leaks are
> still reported. The issue is caused by the memory scanning not stopping
> the whole machine, so pointers may be hidden in registers on different
> CPUs (list insertion/deletion for example causes transient kmemleak
> confusion).
>
> I think the syzkaller guys tried a year or so ago to run it in parallel
> with kmemleak and gave up shortly. The proposal was to add a "stopscan"
> command to kmemleak which would do this under stop_machine(). However,
> no-one got to implementing it.
>
> So, in this case, does the leak still appear with the reproducer, once
> the system went idle?

Hi Catalin,

This report came from syzbot, so obviously we did not give up :)

We don't run scanning in parallel with fuzzing and do a very intricate
multi-step dance to overcome false positives:
https://github.com/google/syzkaller/blob/5962a2dc88f6511b77100acdf687c1088f253f6b/executor/common_linux.h#L3407-L3478
and only report leaks that are reproducible.
So far I have not seen any noticable amount of false positives, and
you can see 70 already fixed leaks here:
https://syzkaller.appspot.com/upstream/fixed?manager=ci-upstream-gce-leak
https://syzkaller.appspot.com/upstream?manager=ci-upstream-gce-leak