f2fs: readahead contiguous pages for restore_node_summary
authorChao Yu <chao2.yu@samsung.com>
Fri, 22 Nov 2013 07:48:54 +0000 (15:48 +0800)
committerJaegeuk Kim <jaegeuk.kim@samsung.com>
Mon, 23 Dec 2013 01:18:05 +0000 (10:18 +0900)
If cp has no CP_UMOUNT_FLAG, we will read all pages in whole node segment
one by one, it makes low performance. So let's merge contiguous pages and
readahead for better performance.

Signed-off-by: Chao Yu <chao2.yu@samsung.com>
[Jaegeuk Kim: adjust the new bio operations]
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
fs/f2fs/node.c

index 0e1a3df18e58c167344f7c31afce21e7e3c5cc3b..0855168af7d7d9be23d93fa093deaf30ab7f80db 100644 (file)
@@ -1565,47 +1565,84 @@ int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
        return 0;
 }
 
+/*
+ * ra_sum_pages() merge contiguous pages into one bio and submit.
+ * these pre-readed pages are linked in pages list.
+ */
+static int ra_sum_pages(struct f2fs_sb_info *sbi, struct list_head *pages,
+                               int start, int nrpages)
+{
+       struct page *page;
+       int page_idx = start;
+
+       for (; page_idx < start + nrpages; page_idx++) {
+               /* alloc temporal page for read node summary info*/
+               page = alloc_page(GFP_NOFS | __GFP_ZERO);
+               if (!page) {
+                       struct page *tmp;
+                       list_for_each_entry_safe(page, tmp, pages, lru) {
+                               list_del(&page->lru);
+                               unlock_page(page);
+                               __free_pages(page, 0);
+                       }
+                       return -ENOMEM;
+               }
+
+               lock_page(page);
+               page->index = page_idx;
+               list_add_tail(&page->lru, pages);
+       }
+
+       list_for_each_entry(page, pages, lru)
+               f2fs_submit_page_mbio(sbi, page, page->index, META, READ);
+
+       f2fs_submit_merged_bio(sbi, META, true, READ);
+       return 0;
+}
+
 int restore_node_summary(struct f2fs_sb_info *sbi,
                        unsigned int segno, struct f2fs_summary_block *sum)
 {
        struct f2fs_node *rn;
        struct f2fs_summary *sum_entry;
-       struct page *page;
+       struct page *page, *tmp;
        block_t addr;
-       int i, last_offset;
-
-       /* alloc temporal page for read node */
-       page = alloc_page(GFP_NOFS | __GFP_ZERO);
-       if (!page)
-               return -ENOMEM;
-       lock_page(page);
+       int bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
+       int i, last_offset, nrpages, err = 0;
+       LIST_HEAD(page_list);
 
        /* scan the node segment */
        last_offset = sbi->blocks_per_seg;
        addr = START_BLOCK(sbi, segno);
        sum_entry = &sum->entries[0];
 
-       for (i = 0; i < last_offset; i++, sum_entry++) {
-               /*
-                * In order to read next node page,
-                * we must clear PageUptodate flag.
-                */
-               ClearPageUptodate(page);
+       for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
+               nrpages = min(last_offset - i, bio_blocks);
 
-               if (f2fs_submit_page_bio(sbi, page, addr, READ_SYNC))
-                       goto out;
+               /* read ahead node pages */
+               err = ra_sum_pages(sbi, &page_list, addr, nrpages);
+               if (err)
+                       return err;
 
-               lock_page(page);
-               rn = F2FS_NODE(page);
-               sum_entry->nid = rn->footer.nid;
-               sum_entry->version = 0;
-               sum_entry->ofs_in_node = 0;
-               addr++;
+               list_for_each_entry_safe(page, tmp, &page_list, lru) {
+
+                       lock_page(page);
+                       if(PageUptodate(page)) {
+                               rn = F2FS_NODE(page);
+                               sum_entry->nid = rn->footer.nid;
+                               sum_entry->version = 0;
+                               sum_entry->ofs_in_node = 0;
+                               sum_entry++;
+                       } else {
+                               err = -EIO;
+                       }
+
+                       list_del(&page->lru);
+                       unlock_page(page);
+                       __free_pages(page, 0);
+               }
        }
-       unlock_page(page);
-out:
-       __free_pages(page, 0);
-       return 0;
+       return err;
 }
 
 static bool flush_nats_in_journal(struct f2fs_sb_info *sbi)