Re: dm: Forbid requeue of writes to zones

[Damien Le Moal <Damien.LeMoal@wdc.com>]

On 2021/06/04 7:17, Colin Ian King wrote:
> Hi,
> 
> Static analysis with Coverity on Linux next has found and issue in
> drivers/md/dm.c with the following commit:
> 
> commit 2c243153d1d4be4e23735cd10984ac17c7a54531
> Author: Damien Le Moal <damien.lemoal@wdc.com>
> Date:   Wed May 26 06:24:58 2021 +0900
> 
>     dm: Forbid requeue of writes to zones
> 
> The analysis is as follows:
> 
>  828 static void dec_pending(struct dm_io *io, blk_status_t error)
>  829 {
>  830        unsigned long flags;
>  831        blk_status_t io_error;
> 
>     1. var_decl: Declaring variable bio without initializer.

This one is related to #7.

> 
>  832        struct bio *bio;
>  833        struct mapped_device *md = io->md;
>  834
>  835        /* Push-back supersedes any I/O errors */
> 
>     2. Condition !!error, taking true branch.
> 
>  836        if (unlikely(error)) {
>  837                spin_lock_irqsave(&io->endio_lock, flags);
> 
>     3. Condition io->status == 11 /* (blk_status_t)11 */, taking false
> branch.
> 
>  838                if (!(io->status == BLK_STS_DM_REQUEUE &&
> __noflush_suspending(md)))
>  839                        io->status = error;
>  840                spin_unlock_irqrestore(&io->endio_lock, flags);
>  841        }
>  842

My patch does not touch these hunks. They are as is. So that is not new.

> 
>     4. Condition atomic_dec_and_test(&io->io_count), taking true branch.
> 
>  843        if (atomic_dec_and_test(&io->io_count)) {
> 
>     5. Condition io->status == 11 /* (blk_status_t)11 */, taking true
> branch.
> 
>  844                if (io->status == BLK_STS_DM_REQUEUE) {
>  845                        /*
>  846                         * Target requested pushing back the I/O.
>  847                         */
>  848                        spin_lock_irqsave(&md->deferred_lock, flags);
> 
>     6. Condition __noflush_suspending(md), taking true branch.
> 
>  849                        if (__noflush_suspending(md) &&

I do not understand this one, nor #4.

> 
> Uninitialized pointer read
>     7. uninit_use_in_call: Using uninitialized value bio when calling
> dm_is_zone_write.
> 
>  850                            !WARN_ON_ONCE(dm_is_zone_write(md, bio)))
>  851                                /* NOTE early return due to
> BLK_STS_DM_REQUEUE below */
>  852                                bio_list_add_head(&md->deferred,
> io->orig_bio);

The kernel build robot signaled this one already. Will send an incremental patch
asap today.

> 
> The pointer bio is not initialized and yet is being used in the call to
> function dm_is_zone_write where pointer bio is being accessed. I'm not
> sure what the original intent was, but this looks incorrect.
> 
> Colin
> 


-- 
Damien Le Moal
Western Digital Research