f2fs: reduce unncessary locking pages during read

This patch reduces redundant locking and unlocking pages during read operations.
In f2fs_readpage, let's use wait_on_page_locked() instead of lock_page.
And then, when we need to modify any data finally, let's lock the page so that
we can avoid lock contention.

[readpage rule]
- The f2fs_readpage returns unlocked page, or released page too in error cases.
- Its caller should handle read error, -EIO, after locking the page, which
  indicates read completion.
- Its caller should check PageUptodate after grab_cache_page.

Signed-off-by: Changman Lee <cm224.lee@samsung.com>
Reviewed-by: Namjae Jeon <namjae.jeon@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>

fs/f2fs/checkpoint.c

@@ -57,13 +57,15 @@ repeat:
 		cond_resched();
 		goto repeat;
 	}
-	if (f2fs_readpage(sbi, page, index, READ_SYNC)) {
+	if (PageUptodate(page))
-		f2fs_put_page(page, 1);
+		goto out;
-		goto repeat;
-	}
-	mark_page_accessed(page);
 
-	/* We do not allow returning an errorneous page */
+	if (f2fs_readpage(sbi, page, index, READ_SYNC))
+		goto repeat;
+
+	lock_page(page);
+out:
+	mark_page_accessed(page);
 	return page;
 }
 

fs/f2fs/data.c

@@ -199,15 +199,20 @@ struct page *find_data_page(struct inode *inode, pgoff_t index)
 	if (!page)
 		return ERR_PTR(-ENOMEM);
 
-	err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
+	if (PageUptodate(page)) {
-	if (err) {
-		f2fs_put_page(page, 1);
-		return ERR_PTR(err);
-	}
 		unlock_page(page);
 		return page;
 	}
 
+	err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
+	wait_on_page_locked(page);
+	if (!PageUptodate(page)) {
+		f2fs_put_page(page, 0);
+		return ERR_PTR(-EIO);
+	}
+	return page;
+}
+
 /*
  * If it tries to access a hole, return an error.
  * Because, the callers, functions in dir.c and GC, should be able to know
@@ -241,9 +246,13 @@ struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
 	BUG_ON(dn.data_blkaddr == NULL_ADDR);
 
 	err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
-	if (err) {
+	if (err)
-		f2fs_put_page(page, 1);
 		return ERR_PTR(err);
+
+	lock_page(page);
+	if (!PageUptodate(page)) {
+		f2fs_put_page(page, 1);
+		return ERR_PTR(-EIO);
 	}
 	return page;
 }
@@ -283,14 +292,17 @@ struct page *get_new_data_page(struct inode *inode, pgoff_t index,
 
 	if (dn.data_blkaddr == NEW_ADDR) {
 		zero_user_segment(page, 0, PAGE_CACHE_SIZE);
+		SetPageUptodate(page);
 	} else {
 		err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
-		if (err) {
+		if (err)
-			f2fs_put_page(page, 1);
 			return ERR_PTR(err);
+		lock_page(page);
+		if (!PageUptodate(page)) {
+			f2fs_put_page(page, 1);
+			return ERR_PTR(-EIO);
 		}
 	}
-	SetPageUptodate(page);
 
 	if (new_i_size &&
 		i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
@@ -325,22 +337,14 @@ static void read_end_io(struct bio *bio, int err)
 
 /*
  * Fill the locked page with data located in the block address.
- * Read operation is synchronous, and caller must unlock the page.
+ * Return unlocked page.
  */
 int f2fs_readpage(struct f2fs_sb_info *sbi, struct page *page,
 					block_t blk_addr, int type)
 {
 	struct block_device *bdev = sbi->sb->s_bdev;
-	bool sync = (type == READ_SYNC);
 	struct bio *bio;
 
-	/* This page can be already read by other threads */
-	if (PageUptodate(page)) {
-		if (!sync)
-			unlock_page(page);
-		return 0;
-	}
-
 	down_read(&sbi->bio_sem);
 
 	/* Allocate a new bio */
@@ -354,18 +358,12 @@ int f2fs_readpage(struct f2fs_sb_info *sbi, struct page *page,
 		kfree(bio->bi_private);
 		bio_put(bio);
 		up_read(&sbi->bio_sem);
+		f2fs_put_page(page, 1);
 		return -EFAULT;
 	}
 
 	submit_bio(type, bio);
 	up_read(&sbi->bio_sem);
-
-	/* wait for read completion if sync */
-	if (sync) {
-		lock_page(page);
-		if (PageError(page))
-			return -EIO;
-	}
 	return 0;
 }
 
@@ -636,18 +634,22 @@ static int f2fs_write_begin(struct file *file, struct address_space *mapping,
 
 		/* Reading beyond i_size is simple: memset to zero */
 		zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
-		return 0;
+		goto out;
 	}
 
 	if (dn.data_blkaddr == NEW_ADDR) {
 		zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 	} else {
 		err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
-		if (err) {
+		if (err)
-			f2fs_put_page(page, 1);
 			return err;
+		lock_page(page);
+		if (!PageUptodate(page)) {
+			f2fs_put_page(page, 1);
+			return -EIO;
 		}
 	}
+out:
 	SetPageUptodate(page);
 	clear_cold_data(page);
 	return 0;

fs/f2fs/node.c

@@ -100,10 +100,13 @@ static void ra_nat_pages(struct f2fs_sb_info *sbi, int nid)
 		page = grab_cache_page(mapping, index);
 		if (!page)
 			continue;
-		if (f2fs_readpage(sbi, page, index, READ)) {
+		if (PageUptodate(page)) {
 			f2fs_put_page(page, 1);
 			continue;
 		}
+		if (f2fs_readpage(sbi, page, index, READ))
+			continue;
+
 		f2fs_put_page(page, 0);
 	}
 }
@@ -851,8 +854,16 @@ static int read_node_page(struct page *page, int type)
 
 	get_node_info(sbi, page->index, &ni);
 
-	if (ni.blk_addr == NULL_ADDR)
+	if (ni.blk_addr == NULL_ADDR) {
+		f2fs_put_page(page, 1);
 		return -ENOENT;
+	}
+
+	if (PageUptodate(page)) {
+		unlock_page(page);
+		return 0;
+	}
+
 	return f2fs_readpage(sbi, page, ni.blk_addr, type);
 }
 
@@ -865,18 +876,17 @@ void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
 	struct page *apage;
 
 	apage = find_get_page(mapping, nid);
-	if (apage && PageUptodate(apage))
+	if (apage && PageUptodate(apage)) {
-		goto release_out;
+		f2fs_put_page(apage, 0);
+		return;
+	}
 	f2fs_put_page(apage, 0);
 
 	apage = grab_cache_page(mapping, nid);
 	if (!apage)
 		return;
 
-	if (read_node_page(apage, READA))
+	if (read_node_page(apage, READA) == 0)
-		unlock_page(apage);
-
-release_out:
 		f2fs_put_page(apage, 0);
 	return;
 }
@@ -892,11 +902,14 @@ struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
 		return ERR_PTR(-ENOMEM);
 
 	err = read_node_page(page, READ_SYNC);
-	if (err) {
+	if (err)
-		f2fs_put_page(page, 1);
 		return ERR_PTR(err);
-	}
 
+	lock_page(page);
+	if (!PageUptodate(page)) {
+		f2fs_put_page(page, 1);
+		return ERR_PTR(-EIO);
+	}
 	BUG_ON(nid != nid_of_node(page));
 	mark_page_accessed(page);
 	return page;
@@ -928,11 +941,8 @@ repeat:
 		goto page_hit;
 
 	err = read_node_page(page, READ_SYNC);
-	unlock_page(page);
+	if (err)
-	if (err) {
-		f2fs_put_page(page, 0);
 		return ERR_PTR(err);
-	}
 
 	/* Then, try readahead for siblings of the desired node */
 	end = start + MAX_RA_NODE;
@@ -957,6 +967,7 @@ page_hit:
 		f2fs_put_page(page, 1);
 		goto repeat;
 	}
+	mark_page_accessed(page);
 	return page;
 }
 
@@ -1473,23 +1484,24 @@ int restore_node_summary(struct f2fs_sb_info *sbi,
 	sum_entry = &sum->entries[0];
 
 	for (i = 0; i < last_offset; i++, sum_entry++) {
-		if (f2fs_readpage(sbi, page, addr, READ_SYNC))
-			goto out;
-
-		rn = (struct f2fs_node *)page_address(page);
-		sum_entry->nid = rn->footer.nid;
-		sum_entry->version = 0;
-		sum_entry->ofs_in_node = 0;
-		addr++;
-
 		/*
 		 * In order to read next node page,
 		 * we must clear PageUptodate flag.
 		 */
 		ClearPageUptodate(page);
+
+		if (f2fs_readpage(sbi, page, addr, READ_SYNC))
+			goto out;
+
+		lock_page(page);
+		rn = (struct f2fs_node *)page_address(page);
+		sum_entry->nid = rn->footer.nid;
+		sum_entry->version = 0;
+		sum_entry->ofs_in_node = 0;
+		addr++;
 	}
-out:
 	unlock_page(page);
+out:
 	__free_pages(page, 0);
 	return 0;
 }

fs/f2fs/recovery.c

@@ -112,11 +112,16 @@ static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head)
 	while (1) {
 		struct fsync_inode_entry *entry;
 
-		if (f2fs_readpage(sbi, page, blkaddr, READ_SYNC))
+		err = f2fs_readpage(sbi, page, blkaddr, READ_SYNC);
+		if (err)
 			goto out;
 
-		if (cp_ver != cpver_of_node(page))
+		lock_page(page);
-			goto out;
+
+		if (cp_ver != cpver_of_node(page)) {
+			err = -EINVAL;
+			goto unlock_out;
+		}
 
 		if (!is_fsync_dnode(page))
 			goto next;
@@ -131,7 +136,7 @@ static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head)
 			if (IS_INODE(page) && is_dent_dnode(page)) {
 				if (recover_inode_page(sbi, page)) {
 					err = -ENOMEM;
-					goto out;
+					goto unlock_out;
 				}
 			}
 
@@ -139,14 +144,14 @@ static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head)
 			entry = kmem_cache_alloc(fsync_entry_slab, GFP_NOFS);
 			if (!entry) {
 				err = -ENOMEM;
-				goto out;
+				goto unlock_out;
 			}
 
 			entry->inode = f2fs_iget(sbi->sb, ino_of_node(page));
 			if (IS_ERR(entry->inode)) {
 				err = PTR_ERR(entry->inode);
 				kmem_cache_free(fsync_entry_slab, entry);
-				goto out;
+				goto unlock_out;
 			}
 
 			list_add_tail(&entry->list, head);
@@ -155,15 +160,15 @@ static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head)
 		if (IS_INODE(page)) {
 			err = recover_inode(entry->inode, page);
 			if (err)
-				goto out;
+				goto unlock_out;
 		}
 next:
 		/* check next segment */
 		blkaddr = next_blkaddr_of_node(page);
-		ClearPageUptodate(page);
 	}
-out:
+unlock_out:
 	unlock_page(page);
+out:
 	__free_pages(page, 0);
 	return err;
 }
@@ -319,8 +324,10 @@ static void recover_data(struct f2fs_sb_info *sbi,
 		if (f2fs_readpage(sbi, page, blkaddr, READ_SYNC))
 			goto out;
 
+		lock_page(page);
+
 		if (cp_ver != cpver_of_node(page))
-			goto out;
+			goto unlock_out;
 
 		entry = get_fsync_inode(head, ino_of_node(page));
 		if (!entry)
@@ -336,10 +343,10 @@ static void recover_data(struct f2fs_sb_info *sbi,
 next:
 		/* check next segment */
 		blkaddr = next_blkaddr_of_node(page);
-		ClearPageUptodate(page);
 	}
-out:
+unlock_out:
 	unlock_page(page);
+out:
 	__free_pages(page, 0);
 
 	allocate_new_segments(sbi);