]> err.no Git - linux-2.6/commitdiff
[PATCH] splice: speedup __generic_file_splice_read
authorJens Axboe <axboe@suse.de>
Tue, 11 Apr 2006 11:52:47 +0000 (13:52 +0200)
committerJens Axboe <axboe@suse.de>
Tue, 11 Apr 2006 11:52:47 +0000 (13:52 +0200)
Using find_get_page() is a lot faster than find_or_create_page(). This
gets splice a lot closer to sendfile() for fd -> socket transfers.

Signed-off-by: Jens Axboe <axboe@suse.de>
fs/splice.c

index c47b561edac099aaa0650328a0313f0070deb6b6..e30743c2c06a87810367015c4ea861efb119acf6 100644 (file)
@@ -240,7 +240,7 @@ __generic_file_splice_read(struct file *in, struct pipe_inode_info *pipe,
        struct page *pages[PIPE_BUFFERS];
        struct page *page;
        pgoff_t index;
-       int i;
+       int i, error;
 
        index = in->f_pos >> PAGE_CACHE_SHIFT;
        offset = in->f_pos & ~PAGE_CACHE_MASK;
@@ -260,32 +260,84 @@ __generic_file_splice_read(struct file *in, struct pipe_inode_info *pipe,
        /*
         * now fill in the holes
         */
+       error = 0;
        for (i = 0; i < nr_pages; i++, index++) {
+find_page:
                /*
-                * no page there, look one up / create it
+                * lookup the page for this index
                 */
-               page = find_or_create_page(mapping, index,
-                                                  mapping_gfp_mask(mapping));
-               if (!page)
-                       break;
+               page = find_get_page(mapping, index);
+               if (!page) {
+                       /*
+                        * If in nonblock mode then dont block on
+                        * readpage (we've kicked readahead so there
+                        * will be asynchronous progress):
+                        */
+                       if (flags & SPLICE_F_NONBLOCK)
+                               break;
+
+                       /*
+                        * page didn't exist, allocate one
+                        */
+                       page = page_cache_alloc_cold(mapping);
+                       if (!page)
+                               break;
+
+                       error = add_to_page_cache_lru(page, mapping, index,
+                                               mapping_gfp_mask(mapping));
+                       if (unlikely(error)) {
+                               page_cache_release(page);
+                               break;
+                       }
 
-               if (PageUptodate(page))
-                       unlock_page(page);
-               else {
-                       int error = mapping->a_ops->readpage(in, page);
+                       goto readpage;
+               }
+
+               /*
+                * If the page isn't uptodate, we may need to start io on it
+                */
+               if (!PageUptodate(page)) {
+                       lock_page(page);
+
+                       /*
+                        * page was truncated, stop here. if this isn't the
+                        * first page, we'll just complete what we already
+                        * added
+                        */
+                       if (!page->mapping) {
+                               unlock_page(page);
+                               page_cache_release(page);
+                               break;
+                       }
+                       /*
+                        * page was already under io and is now done, great
+                        */
+                       if (PageUptodate(page)) {
+                               unlock_page(page);
+                               goto fill_it;
+                       }
+
+readpage:
+                       /*
+                        * need to read in the page
+                        */
+                       error = mapping->a_ops->readpage(in, page);
 
                        if (unlikely(error)) {
                                page_cache_release(page);
+                               if (error == AOP_TRUNCATED_PAGE)
+                                       goto find_page;
                                break;
                        }
                }
+fill_it:
                pages[i] = page;
        }
 
        if (i)
                return move_to_pipe(pipe, pages, i, offset, len, flags);
 
-       return 0;
+       return error;
 }
 
 /**