[PATCH -next] [SCSI] Fix NULL pointer dereference in handling for passthrough commands

[PATCH -next] [SCSI] Fix NULL pointer dereference in handling for passthrough commands

[Laibin Qiu]

In passthrough path. If the command size for ioctl request from userspace
is 0. The original process will get cmd_len from cmd->cmnd, but It has
not been assigned at this time. So it will trigger a NULL pointer BUG.

------------[ cut here ]------------
BUG: kernel NULL pointer dereference, address: 0000000000000000
PF: supervisor read access in kernel mode
PF: error_code(0x0000) - not-present page
RIP: 0010:scsi_queue_rq+0xcb2/0x12b0
Call Trace:
blk_mq_dispatch_rq_list+0x541/0xe90
__blk_mq_sched_dispatch_requests+0x1fe/0x2b0
blk_mq_sched_dispatch_requests+0xbf/0x130
__blk_mq_run_hw_queue+0x15b/0x230
__blk_mq_delay_run_hw_queue+0x18f/0x320
blk_mq_run_hw_queue+0x252/0x280
blk_mq_sched_insert_request+0x228/0x260
blk_execute_rq+0x111/0x160
sg_io+0x51a/0x740
scsi_cmd_ioctl+0x533/0x910
scsi_cmd_blk_ioctl+0xa1/0xb0
cdrom_ioctl+0x3f/0x2510
sr_block_ioctl+0x142/0x180
blkdev_ioctl+0x398/0x450
block_ioctl+0x6d/0x80
__se_sys_ioctl+0xd1/0x140
__x64_sys_ioctl+0x3f/0x50
do_syscall_64+0x37/0x50
entry_SYSCALL_64_after_hwframe+0x44/0xa9

We can trigger front BUG by ioctl blow.
------------[ cut here ]------------

sg_io_hdr_t *addr;

addr = malloc(sizeof(sg_io_hdr_t));
memset(addr, 0, sizeof(sg_io_hdr_t));
addr->interface_id = 'S';

fd = open(/dev/sr0, O_RDONLY); // open a CD_ROM dev

ioctl(fd, SG_IO, addr); // all zero sg_io_hdr_t will trigger this bug

Fixes: 2ceda20f0a99a ("scsi: core: Move command size detection out of
the fast path")
Signed-off-by: Laibin Qiu <qiulaibin@huawei.com>
---
 drivers/scsi/scsi_lib.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
index 572673873ddf..53b47a9103d3 100644
--- a/drivers/scsi/scsi_lib.c
+++ b/drivers/scsi/scsi_lib.c
@@ -1174,9 +1174,9 @@ static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
 	}
 
 	cmd->cmd_len = scsi_req(req)->cmd_len;
+	cmd->cmnd = scsi_req(req)->cmd;
 	if (cmd->cmd_len == 0)
 		cmd->cmd_len = scsi_command_size(cmd->cmnd);
-	cmd->cmnd = scsi_req(req)->cmd;
 	cmd->transfersize = blk_rq_bytes(req);
 	cmd->allowed = scsi_req(req)->retries;
 	return BLK_STS_OK;
-- 
2.22.0

Re: [PATCH -next] [SCSI] Fix NULL pointer dereference in handling for passthrough commands

[Bart Van Assche]

On 9/3/21 23:45, Laibin Qiu wrote:
> In passthrough path. If the command size for ioctl request from userspace
> is 0. The original process will get cmd_len from cmd->cmnd, but It has
> not been assigned at this time. So it will trigger a NULL pointer BUG.

A Friday night in the middle of the merge window is not the best time to 
post a kernel patch. Anyway:
  Reviewed-by: Bart Van Assche <bvanassche@acm.org>

Re: [PATCH -next] [SCSI] Fix NULL pointer dereference in handling for passthrough commands

[Bart Van Assche]

On 9/3/21 23:45, Laibin Qiu wrote:
>   	cmd->cmd_len = scsi_req(req)->cmd_len;
> +	cmd->cmnd = scsi_req(req)->cmd;
>   	if (cmd->cmd_len == 0)
>   		cmd->cmd_len = scsi_command_size(cmd->cmnd);
> -	cmd->cmnd = scsi_req(req)->cmd;
>   	cmd->transfersize = blk_rq_bytes(req);

Thinking further about this: is there any code left that depends on 
scsi_setup_scsi_cmnd() setting cmd->cmd_len? Can the cmd->cmd_len 
assignment be removed from scsi_setup_scsi_cmnd()?

Thanks,

Bart.

Re: [PATCH -next] [SCSI] Fix NULL pointer dereference in handling for passthrough commands

[Christoph Hellwig]

On Mon, Sep 06, 2021 at 05:40:35PM -0700, Bart Van Assche wrote:
> On 9/3/21 23:45, Laibin Qiu wrote:
>>   	cmd->cmd_len = scsi_req(req)->cmd_len;
>> +	cmd->cmnd = scsi_req(req)->cmd;
>>   	if (cmd->cmd_len == 0)
>>   		cmd->cmd_len = scsi_command_size(cmd->cmnd);
>> -	cmd->cmnd = scsi_req(req)->cmd;
>>   	cmd->transfersize = blk_rq_bytes(req);
>
> Thinking further about this: is there any code left that depends on 
> scsi_setup_scsi_cmnd() setting cmd->cmd_len? Can the cmd->cmd_len 
> assignment be removed from scsi_setup_scsi_cmnd()?

cmd_len should never be 0 now, so I think we can remove it.