linux-kernel - Re: [PATCH 03/23] vfs: Generalize the file

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: <20090602070642.GD31556@wotan.suse.de>
Date:	Tue, 2 Jun 2009 09:06:42 +0200
From:	Nick Piggin <npiggin@...e.de>
To:	"Eric W. Biederman" <ebiederm@...ssion.com>
Cc:	Al Viro <viro@...IV.linux.org.uk>, linux-kernel@...r.kernel.org,
	linux-pci@...r.kernel.org, linux-mm@...ck.org,
	linux-fsdevel@...r.kernel.org, Hugh Dickins <hugh@...itas.com>,
	Tejun Heo <tj@...nel.org>,
	Alexey Dobriyan <adobriyan@...il.com>,
	Linus Torvalds <torvalds@...ux-foundation.org>,
	Alan Cox <alan@...rguk.ukuu.org.uk>,
	Greg Kroah-Hartman <gregkh@...e.de>,
	Andrew Morton <akpm@...ux-foundation.org>,
	Christoph Hellwig <hch@...radead.org>,
	"Eric W. Biederman" <ebiederm@...stanetworks.com>
Subject: Re: [PATCH 03/23] vfs: Generalize the file_list

On Mon, Jun 01, 2009 at 02:50:28PM -0700, Eric W. Biederman wrote:
> From: Eric W. Biederman <ebiederm@...ssion.com>
> 
> In the implementation of revoke it is desirable to find all of the
> files we want to operation on.  Currently tty's and mark_files_ro
> use the file_list for this, and this patch generalizes the file
> list so it can be used more efficiently.
> 
> This patch starts by introducing struct file_list making the file
> list a first class object.  file_list_lock and file_list_unlock
> are modified to take this object, making it clear which file_list
> we intended to lock.
> 
> file_move is transformed into file_list_add taking a file_list and not
> allowing the movement of one file to another. __dentry_open
> is modified to support this by only adding normal files in open,
> special files have always been ignored when walking the file_list.
> __dentry_open skipping special files allows __ptmx_open and __tty_open
> to safely call file_add as they are adding the file to the file_list
> for the first time.
> 
> file_kill has been renamed file_list_del to make it clear what it is
> doing and to keep from confusing it with a more revoke like operation.
> 
> put_filp has been modified to not take file_list_del as we are never
> on a file_list when put_filp is called.
> 
> fs_may_remount_ro and mark_files_ro have been modified to walk the
> inode list to find all of the inodes and then to walk the file list
> on those inodes.  It can be a slightly longer walk as we frequently
> cache inodes that we do not have open but the overall complexity
> should be about the same,

Well not really. I have a couple of orders of magnitude more cached
inodes than open files here.


> these are slow path functions, and it
> gives us much greater flexibility overall.

Define flexibility. Walking the sb's file list and checking for
equality with the inode in question gives the same functionality,
just different performance profile.


> --- a/include/linux/fs.h
> +++ b/include/linux/fs.h
> @@ -699,6 +699,11 @@ static inline int mapping_writably_mapped(struct address_space *mapping)
>  	return mapping->i_mmap_writable != 0;
>  }
>  
> +struct file_list {
> +	spinlock_t		lock;
> +	struct list_head	list;
> +};
> +
>  /*
>   * Use sequence counter to get consistent i_size on 32-bit processors.
>   */
> @@ -764,6 +769,7 @@ struct inode {
>  	struct list_head	inotify_watches; /* watches on this inode */
>  	struct mutex		inotify_mutex;	/* protects the watches list */
>  #endif
> +	struct file_list	i_files;
>  
>  	unsigned long		i_state;
>  	unsigned long		dirtied_when;	/* jiffies of first dirtying */
> @@ -934,9 +940,15 @@ struct file {
>  	unsigned long f_mnt_write_state;
>  #endif
>  };
> -extern spinlock_t files_lock;
> -#define file_list_lock() spin_lock(&files_lock);
> -#define file_list_unlock() spin_unlock(&files_lock);
> +
> +static inline void file_list_lock(struct file_list *files)
> +{
> +	spin_lock(&files->lock);
> +}
> +static inline void file_list_unlock(struct file_list *files)
> +{
> +	spin_unlock(&files->lock);
> +}

I don't really like this. It's just a list head. Get rid of
all these wrappers and crap I'd say. In fact, starting with my
patch to unexport files_lock and remove these wrappers would
be reasonable, wouldn't it?

Increasing the size of the struct inode by 24 bytes hurts.
Even when you decrapify it and can reuse i_lock or something,
then it is still 16 bytes on 64-bit.

I haven't looked through all the patches... but this is to
speed up a slowpath operation, isn't it? Or does revoke
need to be especially performant?

So this patch is purely a perofrmance improvement? Then I think
it needs to be justified with numbers and the downsides (bloating
struct inode in particulra) to be changelogged.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/