Re: [PATCH] scsi: scsi_debug: fix return checks for kcalloc

[George Kennedy]

Thanks Joe,

On 11/4/2021 12:04 PM, Joe Perches wrote:
> On Wed, 2021-11-03 at 14:01 -0500, George Kennedy wrote:
> > Change return checks from kcalloc() to now check for NULL and
> > ZERO_SIZE_PTR using the ZERO_OR_NULL_PTR macro or the following
> > crash can occur if ZERO_SIZE_PTR indicator is returned.
> >
> > BUG: KASAN: null-ptr-deref in memcpy include/linux/fortify-string.h:191 [inline]
> > BUG: KASAN: null-ptr-deref in sg_copy_buffer+0x138/0x240 lib/scatterlist.c:974
> > Write of size 4 at addr 0000000000000010 by task syz-executor.1/22789
> []
> > diff --git a/drivers/scsi/scsi_debug.c b/drivers/scsi/scsi_debug.c
> []
> > @@ -3909,7 +3909,7 @@ static int resp_comp_write(struct scsi_cmnd *scp,
> >   		return ret;
> >   	dnum = 2 * num;
> >   	arr = kcalloc(lb_size, dnum, GFP_ATOMIC);
> > -	if (NULL == arr) {
> > +	if (ZERO_OR_NULL_PTR(arr)) {
> >   		mk_sense_buffer(scp, ILLEGAL_REQUEST, INSUFF_RES_ASC,
> >   				INSUFF_RES_ASCQ);
> >   		return check_condition_result;
> This one isn't necessary as num is already tested for non-0 above
> this block.

The check for "num" preceding the above does this:

        if (0 == num)
                return 0;       /* degenerate case, not an error */

Shouldn't I use that same size check and "return 0" if size == zero in 
the other cases?

> > @@ -4265,7 +4265,7 @@ static int resp_verify(struct scsi_cmnd *scp, struct sdebug_dev_info *devip)
> >   		return ret;
> >   
> >   	arr = kcalloc(lb_size, vnum, GFP_ATOMIC);
> > -	if (!arr) {
> > +	if (ZERO_OR_NULL_PTR(arr)) {
> >   		mk_sense_buffer(scp, ILLEGAL_REQUEST, INSUFF_RES_ASC,
> >   				INSUFF_RES_ASCQ);
> >   		return check_condition_result;
> Here it's probably clearer code to test vnum == 0 before the kcalloc
> and return check_condition_result;
>
> > @@ -4334,7 +4334,7 @@ static int resp_report_zones(struct scsi_cmnd *scp,
> >   			    max_zones);
> >   
> >   	arr = kcalloc(RZONES_DESC_HD, alloc_len, GFP_ATOMIC);
> > -	if (!arr) {
> > +	if (ZERO_OR_NULL_PTR(arr)) {
> >   		mk_sense_buffer(scp, ILLEGAL_REQUEST, INSUFF_RES_ASC,
> >   				INSUFF_RES_ASCQ);
> >   		return check_condition_result;
> And here test alloc_len == 0 before the kcalloc.

Using your suggested fix (with return 0 instead of return 
check_condition_result) the patch would look like this:

diff --git a/drivers/scsi/scsi_debug.c b/drivers/scsi/scsi_debug.c
index 40b473e..93913d2 100644
--- a/drivers/scsi/scsi_debug.c
+++ b/drivers/scsi/scsi_debug.c
@@ -4258,6 +4258,8 @@ static int resp_verify(struct scsi_cmnd *scp, 
struct sdebug_dev_info *devip)
                mk_sense_invalid_opcode(scp);
                return check_condition_result;
        }
+       if (0 == vnum)
+               return 0;       /* degenerate case, not an error */
        a_num = is_bytchk3 ? 1 : vnum;
        /* Treat following check like one for read (i.e. no write) 
access */
        ret = check_device_access_params(scp, lba, a_num, false);
@@ -4321,6 +4323,8 @@ static int resp_report_zones(struct scsi_cmnd *scp,
        }
        zs_lba = get_unaligned_be64(cmd + 2);
        alloc_len = get_unaligned_be32(cmd + 10);
+       if (0 == alloc_len)
+               return 0;       /* degenerate case, not an error */
        rep_opts = cmd[14] & 0x3f;
        partial = cmd[14] & 0x80;


Does the above look ok?

George

>