linux-kernel - Re: [PATCH RESEND] scsi: sg: Prevent potential double frees in sg driver

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 for Android: free password hash cracker in your pocket

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

Date:   Tue, 2 Oct 2018 09:34:33 +1000
From:   Douglas Gilbert <dgilbert@...erlog.com>
To:     Evan Green <evgreen@...omium.org>,
        "James E.J. Bottomley" <jejb@...ux.vnet.ibm.com>,
        "Martin K. Petersen" <martin.petersen@...cle.com>,
        linux-scsi@...r.kernel.org, linux-kernel@...r.kernel.org
Cc:     Nick Desaulniers <ndesaulniers@...gle.com>,
        Christoph Hellwig <hch@...radead.org>,
        Robb Glasser <rglasser@...gle.com>, stable@...r.kernel.org
Subject: Re: [PATCH RESEND] scsi: sg: Prevent potential double frees in sg
 driver

On 2018-10-02 02:15 AM, Evan Green wrote:
> From: Robb Glasser <rglasser@...gle.com>
> 
> sg_ioctl could be spammed by requests, leading to a double free in
> __free_pages. This protects the entry points of sg_ioctl where the
> memory could be corrupted by a double call to __free_pages if multiple
> requests are happening concurrently.

Hi,
I  don't like this patch. I would like to see the trace for the double
call to the __free_pages you are referring too. A test program that
show the fault, perhaps?

I have test code to "spam" the sg driver and have not seen a double
__free_pages that you refer to (see sg3_utils package version 1.44,
testing/sg_tst_async.cpp).

Currently I am dusting off 20 years of "laparoscopic" patches to the sg
driver that have made a bit of a mess of the naming and comments. Also
the 16 outstanding requests per file descriptor limit is being removed.
Then I want to add the SG_IOSUBMIT and SG_IORECEIVE ioctls proposed by
Linus Torvalds two week ago.

Executive summary: nak, without further information

Doug Gilbert

> Signed-off-by: Robb Glasser <rglasser@...gle.com>
> Signed-off-by: Nick Desaulniers <ndesaulniers@...gle.com>
> Signed-off-by: Evan Green <evgreen@...omium.org>
> Cc: stable@...r.kernel.org
> 
> ---
> Reposting this patch from last summer, as it looks like it fell in between
> the cracks.
> 
>   drivers/scsi/sg.c | 2 ++
>   1 file changed, 2 insertions(+)
> 
> diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c
> index 8a254bb46a9b..25579d8a16b5 100644
> --- a/drivers/scsi/sg.c
> +++ b/drivers/scsi/sg.c
> @@ -924,8 +924,10 @@ sg_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
>   			return -ENXIO;
>   		if (!access_ok(VERIFY_WRITE, p, SZ_SG_IO_HDR))
>   			return -EFAULT;
> +		mutex_lock(&sfp->parentdp->open_rel_lock);
>   		result = sg_new_write(sfp, filp, p, SZ_SG_IO_HDR,
>   				 1, read_only, 1, &srp);
> +		mutex_unlock(&sfp->parentdp->open_rel_lock);
>   		if (result < 0)
>   			return result;
>   		result = wait_event_interruptible(sfp->read_wait,
>