linux-kernel - Re: [syzbot] WARNING in mntput_no

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Sun, 4 Apr 2021 02:34:08 +0000
From:   Al Viro <viro@...iv.linux.org.uk>
To:     Christian Brauner <christian.brauner@...ntu.com>
Cc:     Jens Axboe <axboe@...nel.dk>,
        syzbot <syzbot+c88a7030da47945a3cc3@...kaller.appspotmail.com>,
        linux-fsdevel@...r.kernel.org, linux-kernel@...r.kernel.org,
        syzkaller-bugs@...glegroups.com, io-uring@...r.kernel.org
Subject: Re: [syzbot] WARNING in mntput_no_expire (2)

On Thu, Apr 01, 2021 at 07:11:12PM +0000, Al Viro wrote:

> > I _think_ I see what the issue is. It seems that an assumption made in
> > this commit might be wrong and we're missing a mnt_add_count() bump that
> > we would otherwise have gotten if we've moved the failure handling into
> > the unlazy helpers themselves.
> > 
> > Al, does that sound plausible?
> 
> mnt_add_count() on _what_?  Failure in legitimize_links() ends up with
> nd->path.mnt zeroed, in both callers.  So which vfsmount would be
> affected?

Could you turn that WARN_ON(count < 0) into
	if (WARN_ON(count < 0))
		printk(KERN_ERR "id = %d, dev = %s, count = %d\n",
				mnt->mnt_id,
				mnt->mnt_sb->s_id,
				count);
add system("cat /proc/self/mountinfo"); right after sandbox_common()
call and try to reproduce that?

I really wonder what mount is it happening to.  BTW, how painful would
it be to teach syzcaller to turn those cascades of
	NONFAILING(*(uint8_t*)0x20000080 = 0x12);
	NONFAILING(*(uint8_t*)0x20000081 = 0);
	NONFAILING(*(uint16_t*)0x20000082 = 0);
	NONFAILING(*(uint32_t*)0x20000084 = 0xffffff9c);
	NONFAILING(*(uint64_t*)0x20000088 = 0);
	NONFAILING(*(uint64_t*)0x20000090 = 0x20000180);
	NONFAILING(memcpy((void*)0x20000180, "./file0\000", 8));
	NONFAILING(*(uint32_t*)0x20000098 = 0);
	NONFAILING(*(uint32_t*)0x2000009c = 0x80);
	NONFAILING(*(uint64_t*)0x200000a0 = 0x23456);
	....
	NONFAILING(syz_io_uring_submit(r[1], r[2], 0x20000080, 0));
into something more readable?  Bloody annoyance every time...  Sure, I can
manually translate it into
	struct io_uring_sqe *sqe = (void *)0x20000080;
	char *s = (void *)0x20000180;
	memset(sqe, '\0', sizeof(*sqe));
	sqe->opcode = 0x12; // IORING_OP_OPENAT?
	sqe->fd = -100;	// AT_FDCWD?
	sqe->addr = s;
	strcpy(s, "./file0");
	sqe->open_flags = 0x80;	// O_EXCL???
	sqe->user_data = 0x23456;	// random tag?
	syz_io_uring_submit(r[1], r[2], (unsigned long)p, 0);
but it's really annoying as hell, especially since syz_io_uring_submit()
comes from syzcaller and the damn thing _knows_ that the third argument
is sodding io_uring_sqe, and never passed to anything other than
memcpy() in there, at that, so the exact address can't matter.

Incidentally, solitary O_EXCL (without O_CREAT) is... curious.  Does that
sucker still trigger without it?  I.e. with store to 0x2000009c replaced
with storing 0?