Re: [PATCH 3/4] scsi: core: Cap shost max_sectors according to DMA optimum mapping limits

[John Garry <john.garry@huawei.com>]

On 23/05/2022 12:08, Dan Carpenter wrote:

Thanks for the report

> 50b6cb3516365c Dexuan Cui        2021-10-07  224  	/* Use min_t(int, ...) in case shost->can_queue exceeds SHRT_MAX */
> 50b6cb3516365c Dexuan Cui        2021-10-07  225  	shost->cmd_per_lun = min_t(int, shost->cmd_per_lun,
> ea2f0f77538c50 John Garry        2021-05-19  226  				   shost->can_queue);
> ea2f0f77538c50 John Garry        2021-05-19  227
> 2ad7ba6ca08593 John Garry        2022-05-20 @228  	if (dma_dev->dma_mask) {
>                                                              ^^^^^^^^^^^^^^^^^

I knew that we fixed up dma_dev to be non-NULL, but I thought it was 
earlier in this function...

> The patch adds a new unchecked dereference
> 
> 2ad7ba6ca08593 John Garry        2022-05-20  229  		shost->max_sectors = min_t(unsigned int, shost->max_sectors,
> 2ad7ba6ca08593 John Garry        2022-05-20  230  				dma_opt_mapping_size(dma_dev) >> SECTOR_SHIFT);
> 2ad7ba6ca08593 John Garry        2022-05-20  231  	}
> 2ad7ba6ca08593 John Garry        2022-05-20  232
> 0a6ac4ee7c2109 Christoph Hellwig 2017-01-03  233  	error = scsi_init_sense_cache(shost);
> 0a6ac4ee7c2109 Christoph Hellwig 2017-01-03  234  	if (error)
> 0a6ac4ee7c2109 Christoph Hellwig 2017-01-03  235  		goto fail;
> 0a6ac4ee7c2109 Christoph Hellwig 2017-01-03  236
> d285203cf647d7 Christoph Hellwig 2014-01-17  237  	error = scsi_mq_setup_tags(shost);
> 542bd1377a9630 James Bottomley   2008-04-21  238  	if (error)
> 542bd1377a9630 James Bottomley   2008-04-21  239  		goto fail;
> d285203cf647d7 Christoph Hellwig 2014-01-17  240
> ^1da177e4c3f41 Linus Torvalds    2005-04-16  241  	if (!shost->shost_gendev.parent)
> ^1da177e4c3f41 Linus Torvalds    2005-04-16  242  		shost->shost_gendev.parent = dev ? dev : &platform_bus;
> 3c8d9a957d0ae6 James Bottomley   2012-05-04 @243  	if (!dma_dev)
>                                                              ^^^^^^^^

Cheers,
John