Re: [PATCH] support polling of /proc/swaps

[Andrew Morton <akpm@linux-foundation.org>]

On Tue, 19 Oct 2010 11:19:16 +0200
Kay Sievers <kay.sievers@vrfy.org> wrote:

> From: Kay Sievers <kay.sievers@vrfy.org>
> Subject: support polling of /proc/swaps
> 
> System management wants to subscribe to changes in swap
> configuration. Make /proc/swaps pollable like /proc/mounts.
> 

It's a bit sad that we have to add quite a pile of infrastructure to
make a procfs file pollable.  I wonder if it's possible to provide some
core support for this, and reduce the amount of code at each particular
handler site.

Also, I wonder how we are to communicate the existence of this feature
to our users.  Nobody will look in Documentation/filesystems/.  Is
there a manpage?  Seems not...

> ---
>  mm/swapfile.c |   48 +++++++++++++++++++++++++++++++++++++++++++++++-
>  1 file changed, 47 insertions(+), 1 deletion(-)
> 
> --- a/mm/swapfile.c
> +++ b/mm/swapfile.c
> @@ -30,6 +30,7 @@
>  #include <linux/capability.h>
>  #include <linux/syscalls.h>
>  #include <linux/memcontrol.h>
> +#include <linux/poll.h>
>  
>  #include <asm/pgtable.h>
>  #include <asm/tlbflush.h>
> @@ -58,6 +59,9 @@ static struct swap_info_struct *swap_inf
>  
>  static DEFINE_MUTEX(swapon_mutex);
>  
> +static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
> +static int proc_poll_event;

Please pick a lock to protect proc_poll_event.  Then document that
locking here, when you also document proc_poll_event ;)

>  static inline unsigned char swap_count(unsigned char ent)
>  {
>  	return ent & ~SWAP_HAS_CACHE;	/* may include SWAP_HAS_CONT flag */
> @@ -1680,6 +1684,8 @@ SYSCALL_DEFINE1(swapoff, const char __us
>  	}
>  	filp_close(swap_file, NULL);
>  	err = 0;
> +	proc_poll_event++;

Then fix this race.

> +	wake_up_interruptible(&proc_poll_wait);
>  
>  out_dput:
>  	filp_close(victim, NULL);