ext4: avoid deadlock in fs reclaim with page writeback

Ext4 has a filesystem wide lock protecting ext4_writepages() calls to
avoid races with switching of journalled data flag or inode format. This
lock can however cause a deadlock like:

CPU0                            CPU1

ext4_writepages()
  percpu_down_read(sbi->s_writepages_rwsem);
                                ext4_change_inode_journal_flag()
                                  percpu_down_write(sbi->s_writepages_rwsem);
                                    - blocks, all readers block from now on
  ext4_do_writepages()
    ext4_init_io_end()
      kmem_cache_zalloc(io_end_cachep, GFP_KERNEL)
        fs_reclaim frees dentry...
          dentry_unlink_inode()
            iput() - last ref =>
              iput_final() - inode dirty =>
                write_inode_now()...
                  ext4_writepages() tries to acquire sbi->s_writepages_rwsem
                    and blocks forever

Make sure we cannot recurse into filesystem reclaim from writeback code
to avoid the deadlock.

Reported-by: [email protected]
Link: https://lore.kernel.org/all/[email protected]
Fixes: c8585c6 ("ext4: fix races between changing inode journal mode and ext4_writepages")
CC: [email protected]
Signed-off-by: Jan Kara <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Theodore Ts'o <[email protected]>

Author: jankara

fs/ext4/ext4.h

@@ -1684,6 +1684,30 @@ static inline struct ext4_inode_info *EXT4_I(struct inode *inode)
 	return container_of(inode, struct ext4_inode_info, vfs_inode);
 }
 
+static inline int ext4_writepages_down_read(struct super_block *sb)
+{
+	percpu_down_read(&EXT4_SB(sb)->s_writepages_rwsem);
+	return memalloc_nofs_save();
+}
+
+static inline void ext4_writepages_up_read(struct super_block *sb, int ctx)
+{
+	memalloc_nofs_restore(ctx);
+	percpu_up_read(&EXT4_SB(sb)->s_writepages_rwsem);
+}
+
+static inline int ext4_writepages_down_write(struct super_block *sb)
+{
+	percpu_down_write(&EXT4_SB(sb)->s_writepages_rwsem);
+	return memalloc_nofs_save();
+}
+
+static inline void ext4_writepages_up_write(struct super_block *sb, int ctx)
+{
+	memalloc_nofs_restore(ctx);
+	percpu_up_write(&EXT4_SB(sb)->s_writepages_rwsem);
+}
+
 static inline int ext4_valid_inum(struct super_block *sb, unsigned long ino)
 {
 	return ino == EXT4_ROOT_INO ||

fs/ext4/inode.c

@@ -2783,11 +2783,12 @@ static int ext4_writepages(struct address_space *mapping,
 		.can_map = 1,
 	};
 	int ret;
+	int alloc_ctx;
 
 	if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 		return -EIO;
 
-	percpu_down_read(&EXT4_SB(sb)->s_writepages_rwsem);
+	alloc_ctx = ext4_writepages_down_read(sb);
 	ret = ext4_do_writepages(&mpd);
 	/*
 	 * For data=journal writeback we could have come across pages marked
@@ -2796,7 +2797,7 @@ static int ext4_writepages(struct address_space *mapping,
 	 */
 	if (!ret && mpd.journalled_more_data)
 		ret = ext4_do_writepages(&mpd);
-	percpu_up_read(&EXT4_SB(sb)->s_writepages_rwsem);
+	ext4_writepages_up_read(sb, alloc_ctx);
 
 	return ret;
 }
@@ -2824,17 +2825,18 @@ static int ext4_dax_writepages(struct address_space *mapping,
 	long nr_to_write = wbc->nr_to_write;
 	struct inode *inode = mapping->host;
 	struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
+	int alloc_ctx;
 
 	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 		return -EIO;
 
-	percpu_down_read(&sbi->s_writepages_rwsem);
+	alloc_ctx = ext4_writepages_down_read(inode->i_sb);
 	trace_ext4_writepages(inode, wbc);
 
 	ret = dax_writeback_mapping_range(mapping, sbi->s_daxdev, wbc);
 	trace_ext4_writepages_result(inode, wbc, ret,
 				     nr_to_write - wbc->nr_to_write);
-	percpu_up_read(&sbi->s_writepages_rwsem);
+	ext4_writepages_up_read(inode->i_sb, alloc_ctx);
 	return ret;
 }
 
@@ -5928,7 +5930,7 @@ int ext4_change_inode_journal_flag(struct inode *inode, int val)
 	journal_t *journal;
 	handle_t *handle;
 	int err;
-	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
+	int alloc_ctx;
 
 	/*
 	 * We have to be very careful here: changing a data block's
@@ -5966,7 +5968,7 @@ int ext4_change_inode_journal_flag(struct inode *inode, int val)
 		}
 	}
 
-	percpu_down_write(&sbi->s_writepages_rwsem);
+	alloc_ctx = ext4_writepages_down_write(inode->i_sb);
 	jbd2_journal_lock_updates(journal);
 
 	/*
@@ -5983,15 +5985,15 @@ int ext4_change_inode_journal_flag(struct inode *inode, int val)
 		err = jbd2_journal_flush(journal, 0);
 		if (err < 0) {
 			jbd2_journal_unlock_updates(journal);
-			percpu_up_write(&sbi->s_writepages_rwsem);
+			ext4_writepages_up_write(inode->i_sb, alloc_ctx);
 			return err;
 		}
 		ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
 	}
 	ext4_set_aops(inode);
 
 	jbd2_journal_unlock_updates(journal);
-	percpu_up_write(&sbi->s_writepages_rwsem);
+	ext4_writepages_up_write(inode->i_sb, alloc_ctx);
 
 	if (val)
 		filemap_invalidate_unlock(inode->i_mapping);

fs/ext4/migrate.c

@@ -408,7 +408,6 @@ static int free_ext_block(handle_t *handle, struct inode *inode)
 
 int ext4_ext_migrate(struct inode *inode)
 {
-	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 	handle_t *handle;
 	int retval = 0, i;
 	__le32 *i_data;
@@ -418,6 +417,7 @@ int ext4_ext_migrate(struct inode *inode)
 	unsigned long max_entries;
 	__u32 goal, tmp_csum_seed;
 	uid_t owner[2];
+	int alloc_ctx;
 
 	/*
 	 * If the filesystem does not support extents, or the inode
@@ -434,7 +434,7 @@ int ext4_ext_migrate(struct inode *inode)
 		 */
 		return retval;
 
-	percpu_down_write(&sbi->s_writepages_rwsem);
+	alloc_ctx = ext4_writepages_down_write(inode->i_sb);
 
 	/*
 	 * Worst case we can touch the allocation bitmaps and a block
@@ -586,7 +586,7 @@ int ext4_ext_migrate(struct inode *inode)
 	unlock_new_inode(tmp_inode);
 	iput(tmp_inode);
 out_unlock:
-	percpu_up_write(&sbi->s_writepages_rwsem);
+	ext4_writepages_up_write(inode->i_sb, alloc_ctx);
 	return retval;
 }
 
@@ -605,6 +605,7 @@ int ext4_ind_migrate(struct inode *inode)
 	ext4_fsblk_t			blk;
 	handle_t			*handle;
 	int				ret, ret2 = 0;
+	int				alloc_ctx;
 
 	if (!ext4_has_feature_extents(inode->i_sb) ||
 	    (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
@@ -621,7 +622,7 @@ int ext4_ind_migrate(struct inode *inode)
 	if (test_opt(inode->i_sb, DELALLOC))
 		ext4_alloc_da_blocks(inode);
 
-	percpu_down_write(&sbi->s_writepages_rwsem);
+	alloc_ctx = ext4_writepages_down_write(inode->i_sb);
 
 	handle = ext4_journal_start(inode, EXT4_HT_MIGRATE, 1);
 	if (IS_ERR(handle)) {
@@ -665,6 +666,6 @@ int ext4_ind_migrate(struct inode *inode)
 	ext4_journal_stop(handle);
 	up_write(&EXT4_I(inode)->i_data_sem);
 out_unlock:
-	percpu_up_write(&sbi->s_writepages_rwsem);
+	ext4_writepages_up_write(inode->i_sb, alloc_ctx);
 	return ret;
 }