f2fs: change the current atomic write way

Current atomic write has three major issues like below.
 - keeps the updates in non-reclaimable memory space and they are even
   hard to be migrated, which is not good for contiguous memory
   allocation.
 - disk spaces used for atomic files cannot be garbage collected, so
   this makes it difficult for the filesystem to be defragmented.
 - If atomic write operations hit the threshold of either memory usage
   or garbage collection failure count, All the atomic write operations
   will fail immediately.

To resolve the issues, I will keep a COW inode internally for all the
updates to be flushed from memory, when we need to flush them out in a
situation like high memory pressure. These COW inodes will be tagged
as orphan inodes to be reclaimed in case of sudden power-cut or system
failure during atomic writes.

Signed-off-by: Daeho Jeong <daehojeong@google.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

1 files changed, 115 insertions, 65 deletions

diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
index 9a1a526f2092..8763a4690aaf 100644
--- a/fs/f2fs/data.c
+++ b/fs/f2fs/data.c
@@ -69,8 +69,7 @@ static bool __is_cp_guaranteed(struct page *page)
 
 	if (f2fs_is_compressed_page(page))
 		return false;
-	if ((S_ISREG(inode->i_mode) &&
-			(f2fs_is_atomic_file(inode) || IS_NOQUOTA(inode))) ||
+	if ((S_ISREG(inode->i_mode) && IS_NOQUOTA(inode)) ||
 			page_private_gcing(page))
 		return true;
 	return false;
@@ -2563,7 +2562,12 @@ int f2fs_do_write_data_page(struct f2fs_io_info *fio)
 	bool ipu_force = false;
 	int err = 0;
 
-	set_new_dnode(&dn, inode, NULL, NULL, 0);
+	/* Use COW inode to make dnode_of_data for atomic write */
+	if (f2fs_is_atomic_file(inode))
+		set_new_dnode(&dn, F2FS_I(inode)->cow_inode, NULL, NULL, 0);
+	else
+		set_new_dnode(&dn, inode, NULL, NULL, 0);
+
 	if (need_inplace_update(fio) &&
 			f2fs_lookup_extent_cache(inode, page->index, &ei)) {
 		fio->old_blkaddr = ei.blk + page->index - ei.fofs;
@@ -2600,6 +2604,7 @@ got_it:
 		err = -EFSCORRUPTED;
 		goto out_writepage;
 	}
+
 	/*
 	 * If current allocation needs SSR,
 	 * it had better in-place writes for updated data.
@@ -3313,6 +3318,100 @@ unlock_out:
 	return err;
 }
 
+static int __find_data_block(struct inode *inode, pgoff_t index,
+				block_t *blk_addr)
+{
+	struct dnode_of_data dn;
+	struct page *ipage;
+	struct extent_info ei = {0, };
+	int err = 0;
+
+	ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
+	if (IS_ERR(ipage))
+		return PTR_ERR(ipage);
+
+	set_new_dnode(&dn, inode, ipage, ipage, 0);
+
+	if (f2fs_lookup_extent_cache(inode, index, &ei)) {
+		dn.data_blkaddr = ei.blk + index - ei.fofs;
+	} else {
+		/* hole case */
+		err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
+		if (err) {
+			dn.data_blkaddr = NULL_ADDR;
+			err = 0;
+		}
+	}
+	*blk_addr = dn.data_blkaddr;
+	f2fs_put_dnode(&dn);
+	return err;
+}
+
+static int __reserve_data_block(struct inode *inode, pgoff_t index,
+				block_t *blk_addr, bool *node_changed)
+{
+	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
+	struct dnode_of_data dn;
+	struct page *ipage;
+	int err = 0;
+
+	f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, true);
+
+	ipage = f2fs_get_node_page(sbi, inode->i_ino);
+	if (IS_ERR(ipage)) {
+		err = PTR_ERR(ipage);
+		goto unlock_out;
+	}
+	set_new_dnode(&dn, inode, ipage, ipage, 0);
+
+	err = f2fs_get_block(&dn, index);
+
+	*blk_addr = dn.data_blkaddr;
+	*node_changed = dn.node_changed;
+	f2fs_put_dnode(&dn);
+
+unlock_out:
+	f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, false);
+	return err;
+}
+
+static int prepare_atomic_write_begin(struct f2fs_sb_info *sbi,
+			struct page *page, loff_t pos, unsigned int len,
+			block_t *blk_addr, bool *node_changed)
+{
+	struct inode *inode = page->mapping->host;
+	struct inode *cow_inode = F2FS_I(inode)->cow_inode;
+	pgoff_t index = page->index;
+	int err = 0;
+	block_t ori_blk_addr;
+
+	/* If pos is beyond the end of file, reserve a new block in COW inode */
+	if ((pos & PAGE_MASK) >= i_size_read(inode))
+		return __reserve_data_block(cow_inode, index, blk_addr,
+					node_changed);
+
+	/* Look for the block in COW inode first */
+	err = __find_data_block(cow_inode, index, blk_addr);
+	if (err)
+		return err;
+	else if (*blk_addr != NULL_ADDR)
+		return 0;
+
+	/* Look for the block in the original inode */
+	err = __find_data_block(inode, index, &ori_blk_addr);
+	if (err)
+		return err;
+
+	/* Finally, we should reserve a new block in COW inode for the update */
+	err = __reserve_data_block(cow_inode, index, blk_addr, node_changed);
+	if (err)
+		return err;
+
+	if (ori_blk_addr != NULL_ADDR)
+		*blk_addr = ori_blk_addr;
+	return 0;
+}
+
 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
 		loff_t pos, unsigned len, unsigned flags,
 		struct page **pagep, void **fsdata)
@@ -3321,7 +3420,7 @@ static int f2fs_write_begin(struct file *file, struct address_space *mapping,
 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 	struct page *page = NULL;
 	pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
-	bool need_balance = false, drop_atomic = false;
+	bool need_balance = false;
 	block_t blkaddr = NULL_ADDR;
 	int err = 0;
 
@@ -3332,14 +3431,6 @@ static int f2fs_write_begin(struct file *file, struct address_space *mapping,
 		goto fail;
 	}
 
-	if ((f2fs_is_atomic_file(inode) &&
-			!f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
-			is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
-		err = -ENOMEM;
-		drop_atomic = true;
-		goto fail;
-	}
-
 	/*
 	 * We should check this at this moment to avoid deadlock on inode page
 	 * and #0 page. The locking rule for inline_data conversion should be:
@@ -3387,7 +3478,11 @@ repeat:
 
 	*pagep = page;
 
-	err = prepare_write_begin(sbi, page, pos, len,
+	if (f2fs_is_atomic_file(inode))
+		err = prepare_atomic_write_begin(sbi, page, pos, len,
+					&blkaddr, &need_balance);
+	else
+		err = prepare_write_begin(sbi, page, pos, len,
 					&blkaddr, &need_balance);
 	if (err)
 		goto fail;
@@ -3443,8 +3538,6 @@ repeat:
 fail:
 	f2fs_put_page(page, 1);
 	f2fs_write_failed(inode, pos + len);
-	if (drop_atomic)
-		f2fs_drop_inmem_pages_all(sbi, false);
 	return err;
 }
 
@@ -3488,8 +3581,12 @@ static int f2fs_write_end(struct file *file,
 	set_page_dirty(page);
 
 	if (pos + copied > i_size_read(inode) &&
-	    !f2fs_verity_in_progress(inode))
+	    !f2fs_verity_in_progress(inode)) {
 		f2fs_i_size_write(inode, pos + copied);
+		if (f2fs_is_atomic_file(inode))
+			f2fs_i_size_write(F2FS_I(inode)->cow_inode,
+					pos + copied);
+	}
 unlock_out:
 	f2fs_put_page(page, 1);
 	f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
@@ -3522,9 +3619,6 @@ void f2fs_invalidate_folio(struct folio *folio, size_t offset, size_t length)
 			inode->i_ino == F2FS_COMPRESS_INO(sbi))
 		clear_page_private_data(&folio->page);
 
-	if (page_private_atomic(&folio->page))
-		return f2fs_drop_inmem_page(inode, &folio->page);
-
 	folio_detach_private(folio);
 }
 
@@ -3534,10 +3628,6 @@ int f2fs_release_page(struct page *page, gfp_t wait)
 	if (PageDirty(page))
 		return 0;
 
-	/* This is atomic written page, keep Private */
-	if (page_private_atomic(page))
-		return 0;
-
 	if (test_opt(F2FS_P_SB(page), COMPRESS_CACHE)) {
 		struct inode *inode = page->mapping->host;
 
@@ -3563,18 +3653,6 @@ static bool f2fs_dirty_data_folio(struct address_space *mapping,
 		folio_mark_uptodate(folio);
 	BUG_ON(folio_test_swapcache(folio));
 
-	if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
-		if (!page_private_atomic(&folio->page)) {
-			f2fs_register_inmem_page(inode, &folio->page);
-			return true;
-		}
-		/*
-		 * Previously, this page has been registered, we just
-		 * return here.
-		 */
-		return false;
-	}
-
 	if (!folio_test_dirty(folio)) {
 		filemap_dirty_folio(mapping, folio);
 		f2fs_update_dirty_folio(inode, folio);
@@ -3654,42 +3732,14 @@ out:
 int f2fs_migrate_page(struct address_space *mapping,
 		struct page *newpage, struct page *page, enum migrate_mode mode)
 {
-	int rc, extra_count;
-	struct f2fs_inode_info *fi = F2FS_I(mapping->host);
-	bool atomic_written = page_private_atomic(page);
+	int rc, extra_count = 0;
 
 	BUG_ON(PageWriteback(page));
 
-	/* migrating an atomic written page is safe with the inmem_lock hold */
-	if (atomic_written) {
-		if (mode != MIGRATE_SYNC)
-			return -EBUSY;
-		if (!mutex_trylock(&fi->inmem_lock))
-			return -EAGAIN;
-	}
-
-	/* one extra reference was held for atomic_write page */
-	extra_count = atomic_written ? 1 : 0;
 	rc = migrate_page_move_mapping(mapping, newpage,
 				page, extra_count);
-	if (rc != MIGRATEPAGE_SUCCESS) {
-		if (atomic_written)
-			mutex_unlock(&fi->inmem_lock);
+	if (rc != MIGRATEPAGE_SUCCESS)
 		return rc;
-	}
-
-	if (atomic_written) {
-		struct inmem_pages *cur;
-
-		list_for_each_entry(cur, &fi->inmem_pages, list)
-			if (cur->page == page) {
-				cur->page = newpage;
-				break;
-			}
-		mutex_unlock(&fi->inmem_lock);
-		put_page(page);
-		get_page(newpage);
-	}
 
 	/* guarantee to start from no stale private field */
 	set_page_private(newpage, 0);