linux-kernel - Re: [PATCH V4 1/8] null_blk: allow REQ_OP_WRITE

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Wed, 30 Nov 2022 07:59:32 +0000
From:   Shinichiro Kawasaki <shinichiro.kawasaki@....com>
To:     Chaitanya Kulkarni <kch@...dia.com>
CC:     "linux-block@...r.kernel.org" <linux-block@...r.kernel.org>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        "axboe@...nel.dk" <axboe@...nel.dk>,
        "damien.lemoal@...nsource.wdc.com" <damien.lemoal@...nsource.wdc.com>,
        "bvanassche@....org" <bvanassche@....org>,
        "vincent.fu@...sung.com" <vincent.fu@...sung.com>
Subject: Re: [PATCH V4 1/8] null_blk: allow REQ_OP_WRITE_ZEROES

On Nov 29, 2022 / 15:28, Chaitanya Kulkarni wrote:
> Add a helper function to enable the REQ_OP_WRITE_ZEROES operations
> when null_blk.
> 
> Since write-zeroes is a non-trivial I/O operation we need this to
> add a blktest so we can test the non-trivial I/O path from the
> application to the block layer.
> 
> Signed-off-by: Chaitanya Kulkarni <kch@...dia.com>

This motivation sounds good. I tried this patch. With a quick test it looks
working good for me. Please find minor comments in line.

[...]

> +static void null_zero_sector(struct nullb_device *d, sector_t sect,
> +			     sector_t nr_sects, bool cache)
> +{
> +	struct radix_tree_root *root = cache ? &d->cache : &d->data;
> +	struct nullb_page *t_page;
> +	unsigned int offset;
> +	void *dest;
> +
> +	t_page = radix_tree_lookup(root, sect >> PAGE_SECTORS_SHIFT);
> +	if (!t_page)
> +		return;
> +
> +	offset = (sect & SECTOR_MASK) << SECTOR_SHIFT;
> +	dest = kmap_atomic(t_page->page);
> +	memset(dest + offset, 0, SECTOR_SIZE * nr_sects);
> +	kunmap_atomic(dest);
> +}

Did you consider to call null_lookup_page() for __null_lookup_page() from
null_zero_sector()? It may simplify this function a bit.

> +
>  static struct nullb_page *null_radix_tree_insert(struct nullb *nullb, u64 idx,
>  	struct nullb_page *t_page, bool is_cache)
>  {
> @@ -1186,6 +1211,27 @@ blk_status_t null_handle_discard(struct nullb_device *dev,
>  	return BLK_STS_OK;
>  }
>  
> +static blk_status_t null_handle_write_zeroes(struct nullb_device *dev,
> +					     sector_t sector, sector_t nr_sectors)
> +{
> +	unsigned int bytes_left = nr_sectors << 9;
> +	struct nullb *nullb = dev->nullb;
> +	size_t curr_bytes;
> +
> +	spin_lock_irq(&nullb->lock);
> +	while (bytes_left > 0) {
> +		curr_bytes = min_t(size_t, bytes_left, nullb->dev->blocksize);
> +		nr_sectors = curr_bytes >> SECTOR_SHIFT;
> +		null_zero_sector(nullb->dev, sector, nr_sectors, false);
> +		if (null_cache_active(nullb))
> +			null_zero_sector(nullb->dev, sector, nr_sectors, true);
> +		sector += nr_sectors;
> +		bytes_left -= curr_bytes;
> +	}
> +	spin_unlock_irq(&nullb->lock);
> +	return BLK_STS_OK;
> +}
> +
>  static int null_handle_flush(struct nullb *nullb)
>  {
>  	int err;
> @@ -1352,6 +1398,9 @@ static inline blk_status_t null_handle_memory_backed(struct nullb_cmd *cmd,
>  	if (op == REQ_OP_DISCARD)
>  		return null_handle_discard(dev, sector, nr_sectors);
>  
> +	if (op == REQ_OP_WRITE_ZEROES)
> +		return null_handle_write_zeroes(dev, sector, nr_sectors);
> +
>  	if (dev->queue_mode == NULL_Q_BIO)
>  		err = null_handle_bio(cmd);
>  	else
> @@ -1800,6 +1849,13 @@ static void null_config_discard(struct nullb *nullb)
>  	blk_queue_max_discard_sectors(nullb->q, UINT_MAX >> 9);
>  }
>  
> +static void null_config_write_zeroes(struct nullb *nullb)
> +{
> +	if (!nullb->dev->write_zeroes)
> +		return;
> +	blk_queue_max_write_zeroes_sectors(nullb->q, UINT_MAX >> 9);

Just comment: this value UINT_MAX >> 9 sounds a bit weird, but probably ok. This
value was introduced by commit 306eb6b4ad4f ("nullb: support discard") to call
blk_queue_max_discard_sectors(). I guess you chose the same value for write
zeroes.

> +}
> +
>  static const struct block_device_operations null_bio_ops = {
>  	.owner		= THIS_MODULE,
>  	.submit_bio	= null_submit_bio,
> @@ -2111,6 +2167,7 @@ static int null_add_dev(struct nullb_device *dev)
>  		blk_queue_virt_boundary(nullb->q, PAGE_SIZE - 1);
>  
>  	null_config_discard(nullb);
> +	null_config_write_zeroes(nullb);
>  
>  	if (config_item_name(&dev->item)) {
>  		/* Use configfs dir name as the device name */
> diff --git a/drivers/block/null_blk/null_blk.h b/drivers/block/null_blk/null_blk.h
> index 94ff68052b1e..2c0c9c29158f 100644
> --- a/drivers/block/null_blk/null_blk.h
> +++ b/drivers/block/null_blk/null_blk.h
> @@ -111,6 +111,7 @@ struct nullb_device {
>  	bool power; /* power on/off the device */
>  	bool memory_backed; /* if data is stored in memory */
>  	bool discard; /* if support discard */
> +	bool write_zeroes; /* if support write_zeroes */
>  	bool zoned; /* if device is zoned */
>  	bool virt_boundary; /* virtual boundary on/off for the device */
>  	bool no_sched; /* no IO scheduler for the device */
> -- 
> 2.29.0
> 

-- 
Shin'ichiro Kawasaki