linux-kernel - [BUG] scsi: megaraid_sas: possible use-after-free caused by bad error handling in megasas_probe

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>] [day] [month] [year] [list]

Message-Id: <20220906063318.2745858-1-r33s3n6@gmail.com>
Date:   Tue,  6 Sep 2022 14:33:18 +0800
From:   Zixuan Fu <r33s3n6@...il.com>
To:     kashyap.desai@...adcom.com, sumit.saxena@...adcom.com,
        shivasharan.srikanteshwara@...adcom.com, jejb@...ux.ibm.com,
        martin.petersen@...cle.com
Cc:     megaraidlinux.pdl@...adcom.com, linux-scsi@...r.kernel.org,
        linux-kernel@...r.kernel.org, baijiaju1990@...il.com,
        TOTE Robot <oslab@...nghua.edu.cn>
Subject: [BUG] scsi: megaraid_sas: possible use-after-free caused by bad error handling in megasas_probe_one()

Hello,

Our fault injection tool finds a possible use-after-free in the 
megaraid_sas driver in Linux 5.10.0:

In the file drivers/scsi/megaraid/megaraid_sas_base.c:
In megasas_io_attach(), the call to scsi_add_host() may fail: 
6814:	if (scsi_add_host(host, &instance->pdev->dev)) {
            ...
6818:		return -ENODEV;
6819:	}

This error is then propagated to its caller megasas_probe_one().
7414:	if (megasas_io_attach(instance))
7415:   	goto fail_io_attach;

In error handling code of megasas_probe_one(), it calls scsi_host_put():
7457:   scsi_host_put(host);

The function scsi_host_put() calls scsi_host_dev_release() to free `host`,
which contains a variable `instance`.

But megasas_probe_one() calls megasas_init_fw() before:
7372: 	if (megasas_init_fw(instance))

In megasas_init_fw(), it starts a timer:
6369:   megasas_start_timer(instance);

And megasas_probe_one() does nothing about it in error handling code. When
the timer expires, it accesses `instance`, causing a use-after-free bug.

I am not quite sure how to fix this possible bug. Any feedback would be
appreciated, thanks!

Reported-by: TOTE Robot <oslab@...nghua.edu.cn>

Best wishes,
Zixuan Fu