Patchwork vmsplice: relax alignement requirements for SPLICE_F_GIFT

login
register
mail settings
Submitter Eric Dumazet
Date April 4, 2012, 9:08 a.m.
Message ID <1333530531.18626.550.camel@edumazet-glaptop>
Download mbox | patch
Permalink /patch/150686/
State Not Applicable
Delegated to: David Miller
Headers show

Comments

Eric Dumazet - April 4, 2012, 9:08 a.m.
It seems there is no fundamental reason to limit vmsplice()
SPLICE_F_GIFT to page aligned chunks.

All helpers are prepared to cope with offsets in page.

This limitation makes vmsplice() API very impractical in the zero-copy
land.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Tom Herbert <therbert@google.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: David Miller <davem@davemloft.net>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Hugh Dickins <hughd@google.com>
Cc: Changli Gao <xiaosuo@gmail.com>
Cc: Miklos Szeredi <mszeredi@suse.cz>
---
 fs/splice.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)



--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Jens Axboe - April 20, 2012, 8:19 a.m.
On Wed, Apr 04 2012, Eric Dumazet wrote:
> It seems there is no fundamental reason to limit vmsplice()
> SPLICE_F_GIFT to page aligned chunks.
> 
> All helpers are prepared to cope with offsets in page.
> 
> This limitation makes vmsplice() API very impractical in the zero-copy
> land.

Sorry for the slow reply on this one. It's been a long time since I
wrote that code... If I recall correctly, the reason it currently
requires alignment, is that a gifted page is no longer yours, by
definition. So the alignment check was to ensure that we gift full
pages. But if the caller screws that up, I guess it's the callers
problem and it's a shame to limit the API in this way.

I will apply it for 3.5.
Eric Dumazet - April 20, 2012, 8:35 a.m.
On Fri, 2012-04-20 at 10:19 +0200, Jens Axboe wrote:
> On Wed, Apr 04 2012, Eric Dumazet wrote:
> > It seems there is no fundamental reason to limit vmsplice()
> > SPLICE_F_GIFT to page aligned chunks.
> > 
> > All helpers are prepared to cope with offsets in page.
> > 
> > This limitation makes vmsplice() API very impractical in the zero-copy
> > land.
> 
> Sorry for the slow reply on this one. It's been a long time since I
> wrote that code... If I recall correctly, the reason it currently
> requires alignment, is that a gifted page is no longer yours, by
> definition. So the alignment check was to ensure that we gift full
> pages. But if the caller screws that up, I guess it's the callers
> problem and it's a shame to limit the API in this way.
> 
> I will apply it for 3.5.
> 

Thanks Jens 


--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch

diff --git a/fs/splice.c b/fs/splice.c
index 5f883de..5417aa3 100644
--- a/fs/splice.c
+++ b/fs/splice.c
@@ -1385,7 +1385,7 @@  static long do_splice(struct file *in, loff_t __user *off_in,
  */
 static int get_iovec_page_array(const struct iovec __user *iov,
 				unsigned int nr_vecs, struct page **pages,
-				struct partial_page *partial, int aligned,
+				struct partial_page *partial, bool aligned,
 				unsigned int pipe_buffers)
 {
 	int buffers = 0, error = 0;
@@ -1623,7 +1623,7 @@  static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov,
 		return -ENOMEM;
 
 	spd.nr_pages = get_iovec_page_array(iov, nr_segs, spd.pages,
-					    spd.partial, flags & SPLICE_F_GIFT,
+					    spd.partial, false,
 					    pipe->buffers);
 	if (spd.nr_pages <= 0)
 		ret = spd.nr_pages;