Re: [syzbot] [xfs?] WARNING in xfs_bmap_extents_to_btree

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Thu, Mar 30, 2023 at 06:25:37PM -0700, Darrick J. Wong wrote:
> On Fri, Mar 31, 2023 at 09:43:02AM +1100, Dave Chinner wrote:
> > On Thu, Mar 30, 2023 at 10:52:37AM +0200, Aleksandr Nogikh wrote:
> > > On Thu, Mar 30, 2023 at 3:27 AM 'Dave Chinner' via syzkaller-bugs
> > > <syzkaller-bugs@xxxxxxxxxxxxxxxx> wrote:
> > > >
> > > > On Tue, Mar 28, 2023 at 09:08:01PM -0700, syzbot wrote:
> > > > > Hello,
> > > > >
> > > > > syzbot found the following issue on:
> > > > >
> > > > > HEAD commit:    1e760fa3596e Merge tag 'gfs2-v6.3-rc3-fix' of git://git.ke..
> > > > > git tree:       upstream
> > > > > console output: https://syzkaller.appspot.com/x/log.txt?x=16f83651c80000
> > > > > kernel config:  https://syzkaller.appspot.com/x/.config?x=acdb62bf488a8fe5
> > > > > dashboard link: https://syzkaller.appspot.com/bug?extid=0c383e46e9b4827b01b1
> > > > > compiler:       Debian clang version 15.0.7, GNU ld (GNU Binutils for Debian) 2.35.2
> > > > >
> > > > > Unfortunately, I don't have any reproducer for this issue yet.
> > > > >
> > > > > Downloadable assets:
> > > > > disk image: https://storage.googleapis.com/syzbot-assets/17229b6e6fe0/disk-1e760fa3.raw.xz
> > > > > vmlinux: https://storage.googleapis.com/syzbot-assets/69b5d310fba0/vmlinux-1e760fa3.xz
> > > > > kernel image: https://storage.googleapis.com/syzbot-assets/0c65624aace9/bzImage-1e760fa3.xz
> > > > >
> > > > > IMPORTANT: if you fix the issue, please add the following tag to the commit:
> > > > > Reported-by: syzbot+0c383e46e9b4827b01b1@xxxxxxxxxxxxxxxxxxxxxxxxx
> > > > >
> > > > > ------------[ cut here ]------------
> > > > > WARNING: CPU: 1 PID: 24101 at fs/xfs/libxfs/xfs_bmap.c:660 xfs_bmap_extents_to_btree+0xe1b/0x1190
> > > >
> > > > Allocation got an unexpected ENOSPC when it was supposed to have a
> > > > valid reservation for the space. Likely because of an inconsistency
> > > > that had been induced into the filesystem where superblock space
> > > > accounting doesn't exactly match the AG space accounting and/or the
> > > > tracked free space.
> > > >
> > > > Given this is a maliciously corrupted filesystem image, this sort of
> > > > warning is expected and there's probably nothing we can do to avoid
> > > > it short of a full filesystem verification pass during mount.
> > > > That's not a viable solution, so I think we should just ignore
> > > > syzbot when it generates this sort of warning....
> > > 
> > > If it's not a warning about a kernel bug, then WARN_ON should probably
> > > be replaced by some more suitable reporting mechanism. Kernel coding
> > > style document explicitly says:
> > > 
> > > "WARN*() must not be used for a condition that is expected to trigger
> > > easily, for example, by user space actions.
> > 
> > That's exactly the case here. It should *never* happen in normal
> > production workloads, and it if does then we have the *potential*
> > for silent data loss occurring. That's *exactly* the sort of thing
> > we should be warning admins about in no uncertain terms.  Also, we
> > use WARN_ON_ONCE(), so it's not going to spam the logs.
> > 
> > syzbot is a malicious program - it is injecting broken stuff into
> > the kernel as root to try to trigger situations like this. That
> > doesn't make a warning it triggers bad or incorrect - syzbot is
> > pertubing tightly coupled structures in a way that makes the
> > information shared across those structures inconsistent and
> > eventually the code is going to trip over that inconsistency.
> > 
> > IOWs, once someone has used root permissions to mount a maliciously
> > crafted filesystem image, *all bets are off*. The machine is running
> > a potentially compromised kernel at this point. Hence it is almost
> > guaranteed that at some point the kernel is going to discover things
> > are *badly wrong* and start dumping "this should never happen!"
> > warnings into the logs. That's what the warnings are supposed to do,
> > and the fact that syzbot can trigger them doesn't make the warnings
> > wrong.
> > 
> > > pr_warn_once() is a
> > > possible alternative, if you need to notify the user of a problem."
> > > https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/Documentation/process/coding-style.rst?id=1e760fa3596e8c7f08412712c168288b79670d78#n1223
> > 
> > It is worth remembering that those are guidelines, not enforcable
> > rules and any experienced kernel developer will tell you the same
> > thing.  We know the guidelines, we know when to apply them, we know
> > there are cases that the guidelines simply can't, don't or won't
> > cover.
> 
> ...and perhaps the WARNs that can result from corrupted metadata should
> be changed to XFS_IS_CORRUPT() ?

Well, I think in the case it isn't -corrupt- metadata, more the case
that there is an inconsistency between different structures that are
internally consistent.

e.g. remove a free space extent from the freespace tree without
removing the space from the global free space counters.  Now
delalloc reservation is allowed by the global counters, but when we
got to allocate the extent - or the bmap btree block to index it -
we fail the allocation because the free space btrees are empty.

The allocation structures are not internally inconsistent or
corrupt, so it's done the right thing by returning ENOSPC. The
global counters are not obviously inconsistent or corrupt, either.
So it can be triggered by just the right sort of corruption at
exactly the right time (i.e at 100% ENOSPC), but the chances of this
convoluted set of circumstances happening in production systems is
pretty much infintesimal.

> We still get a kernel log about something going wrong, only now the
> report doesn't trigger everyone's WARN triggers, and we tell the user to
> go run xfs_repair.

I think that is exactly the wrong thing to do.

We have a history of this WARN firing as a result of software bugs
in XFS - typically a transaction space reservation or allocation
parameter setup issue - in which case a WARN_ON_ONCE is more
appropriate here than declaring the filesystem corrupt.

That's the bottom line - this specific WARN has been placed because
it is an indicator of a bug in the code, not because it is something
that occurs because of filesystem corruption. The WARN is an
indicator that the bug needs to be reported, not simply put back on
the user to clean up the mess and continue on blissfully unaware
that they tripped over a kernel bug rather than some nebulous,
unexplainable corruption.

syzbot being able to trip over it by corrupting the fs in just the
right way doesn't mean we should change it - syzbot is a malicious
attacker, not a production workload, and I really don't think we
should be changing warnings that we actually want users to report
just to shut up syzbot.

-Dave.
-- 
Dave Chinner
david@xxxxxxxxxxxxx



[Index of Archives]     [Linux Ext4 Filesystem]     [Union Filesystem]     [Filesystem Testing]     [Ceph Users]     [Ecryptfs]     [NTFS 3]     [AutoFS]     [Kernel Newbies]     [Share Photos]     [Security]     [Netfilter]     [Bugtraq]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux Cachefs]     [Reiser Filesystem]     [Linux RAID]     [NTFS 3]     [Samba]     [Device Mapper]     [CEPH Development]

  Powered by Linux