[PATCH] splice: use mapping_gfp_mask

[Hugh Dickins <hugh@veritas.com>]

The loop block driver is careful to mask __GFP_IO|__GFP_FS out of its
mapping_gfp_mask, to avoid hangs under memory pressure.  But nowadays
it uses splice, usually going through __generic_file_splice_read.  That
must use mapping_gfp_mask instead of GFP_KERNEL to avoid those hangs.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
---
Ought to go into 2.6.25.  For 2.6.23 and 2.6.24 stable?  Well, I've
not actually seen this hang on any of these, though presumably it's
lurking there.  Where I did see it, and test the fix, was 2.6.25-rc5-mm1:
whose SLUB had a disturbing predilection (since corrected) for order-4
allocations, even when allocating radix tree nodes.

 fs/splice.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

--- 2.6.25-rc8/fs/splice.c	2008-03-05 10:47:40.000000000 +0000
+++ linux/fs/splice.c	2008-04-01 22:28:41.000000000 +0100
@@ -320,7 +320,7 @@ __generic_file_splice_read(struct file *
 				break;
 
 			error = add_to_page_cache_lru(page, mapping, index,
-					      GFP_KERNEL);
+						mapping_gfp_mask(mapping));
 			if (unlikely(error)) {
 				page_cache_release(page);
 				if (error == -EEXIST)