Re: [PATCH RFC v4] nbd: skip queue limits update for size-only reconfigure

[Krystian Kaniewski <krystianmkaniewski@gmail.com>]

#syz upstream

On 7/23/2026 11:49 PM, syzbot wrote:
> Commit 242a49e5c878 ("nbd: freeze the queue for queue limits updates")
> correctly added queue freezing for live updates of real queue limits.
> However, nbd_set_size() is also used for capacity-only generic netlink
> updates, so the freeze is unnecessarily broad for the reported request.
>
> When an NBD server becomes unresponsive, it leaves I/O in flight. An
> unnecessary queue freeze during a size-only reconfigure will wait forever
> for these in-flight I/Os to complete while the global genl_mutex is held,
> leading to a global generic netlink stall.
>
> A size-only reconfigure can update config->bytesize and disk capacity
> directly because the effective block size and all other queue limits stay
> unchanged.
>
> This patch introduces nbd_size_set() to handle size-only updates directly
> without freezing the queue or committing queue limits. Capacity-only
> generic netlink reconfiguration no longer freezes the queue, while startup
> and real block-size changes continue to use the existing frozen limits
> update via nbd_set_size().
>
> Fixes: 242a49e5c878 ("nbd: freeze the queue for queue limits updates")
> Assisted-by: Gemini:gemini-3.5-flash Gemini:gemini-3.1-pro-preview syzbot
> Reported-by: syzbot <syzbot+f272bbfbf8498ddadea5@syzkaller.appspotmail.com>
> Closes: https://syzkaller.appspot.com/bug?extid=f272bbfbf8498ddadea5
> Link: https://syzkaller.appspot.com/ai_job?id=786b3bd0-fdad-4252-9e86-7a6a35c3b756
> To: "Jens Axboe" <axboe@kernel.dk>
> To: "Josef Bacik" <josef@toxicpanda.com>
> To: <linux-block@vger.kernel.org>
> To: <nbd@other.debian.org>
> To: "Christoph Hellwig" <hch@lst.de>
> Cc: <linux-kernel@vger.kernel.org>
>
> ---
> v4:
> - Removed stack traces from the commit message.
> - Replaced "system-wide deadlock" with "global generic netlink stall" in the commit message.
>
> v3:
> - Removed the unrelated decrement of nbd_total_devices in nbd_dev_remove().
> - Corrected the referenced commit title of 242a49e5c878 in the description.
> - Clarified that capacity-only generic netlink reconfiguration no longer freezes the queue, while startup and real block-size changes continue to do so.
> https://lore.kernel.org/all/130f544a-70fd-4ca0-91e8-08824170c91b@mail.kernel.org/T/
>
> v2:
> - Replaced the approach of using parallel netlink ops and custom timeouts with skipping queue limits updates for size-only reconfigurations.
> - Introduced nbd_size_set() and nbd_size_update() to update device capacity directly without freezing the queue when the block size does not change.
> - Retained queue freezing only for actual block-size changes.
> - Added decrement of nbd_total_devices in nbd_dev_remove().
> https://lore.kernel.org/all/514764a5-5acd-4e5d-a8eb-f40c403e7514@mail.kernel.org/T/
>
> v1:
> https://lore.kernel.org/all/69ff9a21-60be-4e97-bc8f-a59e739fd982@mail.kernel.org/T/
> ---
> diff --git a/drivers/block/nbd.c b/drivers/block/nbd.c
> index fe63f3c55..567b031ec 100644
> --- a/drivers/block/nbd.c
> +++ b/drivers/block/nbd.c
> @@ -331,6 +331,27 @@ static void nbd_mark_nsock_dead(struct nbd_device *nbd, struct nbd_sock *nsock,
>   	nsock->sent = 0;
>   }
>   
> +static void nbd_size_update(struct nbd_device *nbd)
> +{
> +	struct nbd_config *config = nbd->config;
> +
> +	if (max_part)
> +		set_bit(GD_NEED_PART_SCAN, &nbd->disk->state);
> +	if (!set_capacity_and_notify(nbd->disk, config->bytesize >> 9))
> +		kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
> +}
> +
> +static int nbd_size_set(struct nbd_device *nbd, loff_t bytesize)
> +{
> +	if (bytesize < 0)
> +		return -EINVAL;
> +
> +	nbd->config->bytesize = bytesize;
> +	if (nbd->pid)
> +		nbd_size_update(nbd);
> +	return 0;
> +}
> +
>   static int nbd_set_size(struct nbd_device *nbd, loff_t bytesize, loff_t blksize)
>   {
>   	struct queue_limits lim;
> @@ -375,10 +396,7 @@ static int nbd_set_size(struct nbd_device *nbd, loff_t bytesize, loff_t blksize)
>   	if (error)
>   		return error;
>   
> -	if (max_part)
> -		set_bit(GD_NEED_PART_SCAN, &nbd->disk->state);
> -	if (!set_capacity_and_notify(nbd->disk, bytesize >> 9))
> -		kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
> +	nbd_size_update(nbd);
>   	return 0;
>   }
>   
> @@ -2062,11 +2080,19 @@ static int nbd_genl_size_set(struct genl_info *info, struct nbd_device *nbd)
>   	if (info->attrs[NBD_ATTR_SIZE_BYTES])
>   		bytes = nla_get_u64(info->attrs[NBD_ATTR_SIZE_BYTES]);
>   
> -	if (info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES])
> +	if (info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]) {
>   		bsize = nla_get_u64(info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]);
> +		if (!bsize)
> +			bsize = 1u << NBD_DEF_BLKSIZE_BITS;
> +		if (blk_validate_block_size(bsize))
> +			return -EINVAL;
> +	}
>   
> -	if (bytes != config->bytesize || bsize != nbd_blksize(config))
> -		return nbd_set_size(nbd, bytes, bsize);
> +	if (bytes != config->bytesize || bsize != nbd_blksize(config)) {
> +		if (bsize != nbd_blksize(config))
> +			return nbd_set_size(nbd, bytes, bsize);
> +		return nbd_size_set(nbd, bytes);
> +	}
>   	return 0;
>   }
>   
>
>
> base-commit: 8cd9520d35a6c38db6567e97dd93b1f11f185dc6