linux-kernel - Re: [PATCH 8/9] pstore/blk: use the normal block device I/O path

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: Windows password security audit tool. GUI, reports in PDF.

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

Message-ID: <202012011149.5650B9796@keescook>
Date:   Tue, 1 Dec 2020 11:52:09 -0800
From:   Kees Cook <keescook@...omium.org>
To:     Christoph Hellwig <hch@....de>
Cc:     Anton Vorontsov <anton@...msg.org>,
        Colin Cross <ccross@...roid.com>,
        Tony Luck <tony.luck@...el.com>,
        WeiXiong Liao <liaoweixiong@...winnertech.com>,
        linux-mtd@...ts.infradead.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH 8/9] pstore/blk: use the normal block device I/O path

On Fri, Oct 16, 2020 at 03:20:46PM +0200, Christoph Hellwig wrote:
> Stop poking into block layer internals and just open the block device
> file an use kernel_read and kernel_write on it.  Note that this means
> the transformation from name_to_dev_t can't be used anymore, and proper
> block device file names must be used.
> 
> Signed-off-by: Christoph Hellwig <hch@....de>
> [...]
> +
> +#ifdef MODULE
> +static int __init pstore_blk_init(void)
> +{
> +	return __pstore_blk_init(blkdev);
> +}
>  late_initcall(pstore_blk_init);
>  
>  static void __exit pstore_blk_exit(void)
>  {
> -	if (!psblk_bdev)
> +	if (!psblk_file)
>  		return;
>  	unregister_pstore_device(&pstore_blk_zone_ops);
> -	blkdev_put(psblk_bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
> +	fput(psblk_file);
>  }
>  module_exit(pstore_blk_exit);
> +#else /* MODULE */
> +/*
> + * During early boot the real root file system hasn't been mounted yet,
> + * and not device nodes are present yet.  Use the same scheme to find
> + * the device that we use for mounting the root file system.
> + */
> +void __init pstore_blk_early_init(void)
> +{
> +	const char devname[] = "/dev/pstore-blk";
> +	dev_t dev = name_to_dev_t(blkdev);
> +
> +	if (!dev)
> +		return;
> +	init_unlink(devname);
> +	init_mknod(devname, S_IFBLK | 0600, new_encode_dev(dev));
> +	__pstore_blk_init(devname);
> +}
> +#endif /* MODULE */

That the allowed naming conventions change based on _how_ pstore is
built seems very wrong to me.

While I do like the clean up to simplify the read/write activities, this
seems like totally the wrong approach here.

>  
>  /* get information of pstore/blk */
>  int pstore_blk_get_config(struct pstore_blk_config *info)
> diff --git a/include/linux/pstore_blk.h b/include/linux/pstore_blk.h
> index 0abd412a6cb3e3..7c06b9d6740e2a 100644
> --- a/include/linux/pstore_blk.h
> +++ b/include/linux/pstore_blk.h
> @@ -39,4 +39,6 @@ struct pstore_blk_config {
>   */
>  int pstore_blk_get_config(struct pstore_blk_config *info);
>  
> +void __init pstore_blk_early_init(void);
> +
>  #endif
> diff --git a/init/main.c b/init/main.c
> index 1af84337cb18d5..058cce64f70390 100644
> --- a/init/main.c
> +++ b/init/main.c
> @@ -98,6 +98,7 @@
>  #include <linux/mem_encrypt.h>
>  #include <linux/kcsan.h>
>  #include <linux/init_syscalls.h>
> +#include <linux/pstore_blk.h>
>  
>  #include <asm/io.h>
>  #include <asm/bugs.h>
> @@ -1524,6 +1525,9 @@ static noinline void __init kernel_init_freeable(void)
>  		prepare_namespace();
>  	}
>  
> +	if (IS_BUILTIN(CONFIG_PSTORE_BLK))
> +		pstore_blk_early_init();
> +

I hate this being a special-case in kernel_init. For ramoops, we use:

postcore_initcall(ramoops_init);

which is much better than open coding this here.

>  	/*
>  	 * Ok, we have completed the initial bootup, and
>  	 * we're essentially up and running. Get rid of the
> -- 
> 2.28.0
> 

-- 
Kees Cook