Re: [PATCH 3/7] minix: don't flush page immediately for DIRSYNC directories

From: Al Viro <viro@zeniv.linux.org.uk>
To: Christoph Hellwig <hch@lst.de>
Cc: Matthew Wilcox <willy@infradead.org>,
	Andrew Morton <akpm@linux-foundation.org>,
	Chris Mason <clm@fb.com>, Josef Bacik <josef@toxicpanda.com>,
	David Sterba <dsterba@suse.com>,
	Dave Kleikamp <shaggy@kernel.org>, Mark Fasheh <mark@fasheh.com>,
	Joel Becker <jlbec@evilplan.org>,
	Joseph Qi <joseph.qi@linux.alibaba.com>,
	Evgeniy Dushistov <dushistov@mail.ru>,
	linux-btrfs@vger.kernel.org,
	jfs-discussion@lists.sourceforge.net, ocfs2-devel@oss.oracle.com,
	linux-fsdevel@vger.kernel.org, linux-mm@kvack.org
Subject: Re: [PATCH 3/7] minix: don't flush page immediately for DIRSYNC directories
Date: Wed, 11 Jan 2023 04:58:43 +0000	[thread overview]
Message-ID: <Y75CAwtM1gE1sevy@ZenIV> (raw)
In-Reply-To: <20230111042641.GA15181@lst.de>

On Wed, Jan 11, 2023 at 05:26:41AM +0100, Christoph Hellwig wrote:
> On Wed, Jan 11, 2023 at 02:20:41AM +0000, Al Viro wrote:
> > More seriously, all those ..._set_link() need to return an error and their
> > callers (..._rename()) need to deal with failures.
> 
> That's actually what I did yesterday:
> 
> http://git.infradead.org/users/hch/misc.git/shortlog/refs/heads/remove-write_one_page

ext2 also has that bug.  As well as "need to check for delete_entry errors"
one (also in ext2_rename()).

Completely untested patch follows:

diff --git a/fs/ext2/dir.c b/fs/ext2/dir.c
index e5cbc27ba459..b38fab33cd0d 100644
--- a/fs/ext2/dir.c
+++ b/fs/ext2/dir.c
@@ -461,7 +461,7 @@ static int ext2_handle_dirsync(struct inode *dir)
 	return err;
 }
 
-void ext2_set_link(struct inode *dir, struct ext2_dir_entry_2 *de,
+int ext2_set_link(struct inode *dir, struct ext2_dir_entry_2 *de,
 		   struct page *page, void *page_addr, struct inode *inode,
 		   int update_times)
 {
@@ -480,7 +480,7 @@ void ext2_set_link(struct inode *dir, struct ext2_dir_entry_2 *de,
 		dir->i_mtime = dir->i_ctime = current_time(dir);
 	EXT2_I(dir)->i_flags &= ~EXT2_BTREE_FL;
 	mark_inode_dirty(dir);
-	ext2_handle_dirsync(dir);
+	return ext2_handle_dirsync(dir);
 }
 
 /*
diff --git a/fs/ext2/ext2.h b/fs/ext2/ext2.h
index 28de11a22e5f..95c083bb1b7c 100644
--- a/fs/ext2/ext2.h
+++ b/fs/ext2/ext2.h
@@ -734,7 +734,7 @@ extern int ext2_delete_entry(struct ext2_dir_entry_2 *dir, struct page *page,
 			     char *kaddr);
 extern int ext2_empty_dir (struct inode *);
 extern struct ext2_dir_entry_2 *ext2_dotdot(struct inode *dir, struct page **p, void **pa);
-extern void ext2_set_link(struct inode *, struct ext2_dir_entry_2 *, struct page *, void *,
+extern int ext2_set_link(struct inode *, struct ext2_dir_entry_2 *, struct page *, void *,
 			  struct inode *, int);
 static inline void ext2_put_page(struct page *page, void *page_addr)
 {
diff --git a/fs/ext2/namei.c b/fs/ext2/namei.c
index c056957221a2..5e3397680faa 100644
--- a/fs/ext2/namei.c
+++ b/fs/ext2/namei.c
@@ -370,8 +370,10 @@ static int ext2_rename (struct user_namespace * mnt_userns,
 			err = PTR_ERR(new_de);
 			goto out_dir;
 		}
-		ext2_set_link(new_dir, new_de, new_page, page_addr, old_inode, 1);
+		err = ext2_set_link(new_dir, new_de, new_page, page_addr, old_inode, 1);
 		ext2_put_page(new_page, page_addr);
+		if (err)
+			goto out_dir;
 		new_inode->i_ctime = current_time(new_inode);
 		if (dir_de)
 			drop_nlink(new_inode);
@@ -391,7 +393,9 @@ static int ext2_rename (struct user_namespace * mnt_userns,
 	old_inode->i_ctime = current_time(old_inode);
 	mark_inode_dirty(old_inode);
 
-	ext2_delete_entry(old_de, old_page, old_page_addr);
+	err = ext2_delete_entry(old_de, old_page, old_page_addr);
+	if (err)
+		goto out_dir;
 
 	if (dir_de) {
 		if (old_dir != new_dir)
