ext4: use bio layer instead of buffer layer in mpage_da_submit_io

Call the block I/O layer directly instad of going through the buffer
layer.  This should give us much better performance and scalability,
as well as lowering our CPU utilization when doing buffered writeback.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>

fs/ext4/Makefile

@@ -4,7 +4,7 @@
 
 obj-$(CONFIG_EXT4_FS) += ext4.o
 
-ext4-y	:= balloc.o bitmap.o dir.o file.o fsync.o ialloc.o inode.o \
+ext4-y	:= balloc.o bitmap.o dir.o file.o fsync.o ialloc.o inode.o page-io.o \
 		ioctl.o namei.o super.o symlink.o hash.o resize.o extents.o \
 		ext4_jbd2.o migrate.o mballoc.o block_validity.o move_extent.o
 

fs/ext4/ext4.h

@@ -168,7 +168,20 @@ struct mpage_da_data {
 	int pages_written;
 	int retval;
 };
-#define	EXT4_IO_UNWRITTEN	0x1
+
+/*
+ * Flags for ext4_io_end->flags
+ */
+#define	EXT4_IO_END_UNWRITTEN	0x0001
+#define EXT4_IO_END_ERROR	0x0002
+
+struct ext4_io_page {
+	struct page	*p_page;
+	int		p_count;
+};
+
+#define MAX_IO_PAGES 128
+
 typedef struct ext4_io_end {
 	struct list_head	list;		/* per-file finished IO list */
 	struct inode		*inode;		/* file being written to */
@@ -179,8 +192,18 @@ typedef struct ext4_io_end {
 	struct work_struct	work;		/* data work queue */
 	struct kiocb		*iocb;		/* iocb struct for AIO */
 	int			result;		/* error value for AIO */
+	int			num_io_pages;
+	struct ext4_io_page	*pages[MAX_IO_PAGES];
 } ext4_io_end_t;
 
+struct ext4_io_submit {
+	int			io_op;
+	struct bio		*io_bio;
+	ext4_io_end_t		*io_end;
+	struct ext4_io_page	*io_page;
+	sector_t		io_next_block;
+};
+
 /*
  * Special inodes numbers
  */
@@ -2044,6 +2067,17 @@ extern int ext4_move_extents(struct file *o_filp, struct file *d_filp,
 			     __u64 start_orig, __u64 start_donor,
 			     __u64 len, __u64 *moved_len);
 
+/* page-io.c */
+extern int __init init_ext4_pageio(void);
+extern void exit_ext4_pageio(void);
+extern void ext4_free_io_end(ext4_io_end_t *io);
+extern ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags);
+extern int ext4_end_io_nolock(ext4_io_end_t *io);
+extern void ext4_io_submit(struct ext4_io_submit *io);
+extern int ext4_bio_write_page(struct ext4_io_submit *io,
+			       struct page *page,
+			       int len,
+			       struct writeback_control *wbc);
 
 /* BH_Uninit flag: blocks are allocated but uninitialized on disk */
 enum ext4_state_bits {

fs/ext4/extents.c

@@ -3202,7 +3202,7 @@ ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
 		 * completed
 		 */
 		if (io)
-			io->flag = EXT4_IO_UNWRITTEN;
+			io->flag = EXT4_IO_END_UNWRITTEN;
 		else
 			ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
 		if (ext4_should_dioread_nolock(inode))
@@ -3494,7 +3494,7 @@ int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
 		 */
 		if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
 			if (io)
-				io->flag = EXT4_IO_UNWRITTEN;
+				io->flag = EXT4_IO_END_UNWRITTEN;
 			else
 				ext4_set_inode_state(inode,
 						     EXT4_STATE_DIO_UNWRITTEN);

fs/ext4/inode.c

@@ -2016,8 +2016,10 @@ static int mpage_da_submit_io(struct mpage_da_data *mpd,
 	struct buffer_head *bh, *page_bufs = NULL;
 	int journal_data = ext4_should_journal_data(inode);
 	sector_t pblock = 0, cur_logical = 0;
+	struct ext4_io_submit io_submit;
 
 	BUG_ON(mpd->next_page <= mpd->first_page);
+	memset(&io_submit, 0, sizeof(io_submit));
 	/*
 	 * We need to start from the first_page to the next_page - 1
 	 * to make sure we also write the mapped dirty buffer_heads.
@@ -2109,16 +2111,16 @@ static int mpage_da_submit_io(struct mpage_da_data *mpd,
 				/* mark the buffer_heads as dirty & uptodate */
 				block_commit_write(page, 0, len);
 
-			if (journal_data && PageChecked(page))
+			/*
+			 * Delalloc doesn't support data journalling,
+			 * but eventually maybe we'll lift this
+			 * restriction.
+			 */
+			if (unlikely(journal_data && PageChecked(page)))
 				err = __ext4_journalled_writepage(page, len);
-			else if (buffer_uninit(page_bufs)) {
-				ext4_set_bh_endio(page_bufs, inode);
-				err = block_write_full_page_endio(page,
-					noalloc_get_block_write,
-					mpd->wbc, ext4_end_io_buffer_write);
-			} else
-				err = block_write_full_page(page,
-					    noalloc_get_block_write, mpd->wbc);
+			else
+				err = ext4_bio_write_page(&io_submit, page,
+							  len, mpd->wbc);
 
 			if (!err)
 				mpd->pages_written++;
@@ -2131,6 +2133,7 @@ static int mpage_da_submit_io(struct mpage_da_data *mpd,
 		}
 		pagevec_release(&pvec);
 	}
+	ext4_io_submit(&io_submit);
 	return ret;
 }
 
@@ -3426,15 +3429,6 @@ ext4_readpages(struct file *file, struct address_space *mapping,
 	return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
 }
 
-static void ext4_free_io_end(ext4_io_end_t *io)
-{
-	BUG_ON(!io);
-	if (io->page)
-		put_page(io->page);
-	iput(io->inode);
-	kfree(io);
-}
-
 static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
 {
 	struct buffer_head *head, *bh;
@@ -3639,68 +3633,6 @@ static void dump_completed_IO(struct inode * inode)
 #endif
 }
 
-/*
- * check a range of space and convert unwritten extents to written.
- */
-static int ext4_end_io_nolock(ext4_io_end_t *io)
-{
-	struct inode *inode = io->inode;
-	loff_t offset = io->offset;
-	ssize_t size = io->size;
-	int ret = 0;
-
-	ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
-		   "list->prev 0x%p\n",
-	           io, inode->i_ino, io->list.next, io->list.prev);
-
-	if (list_empty(&io->list))
-		return ret;
-
-	if (io->flag != EXT4_IO_UNWRITTEN)
-		return ret;
-
-	ret = ext4_convert_unwritten_extents(inode, offset, size);
-	if (ret < 0) {
-		printk(KERN_EMERG "%s: failed to convert unwritten"
-			"extents to written extents, error is %d"
-			" io is still on inode %lu aio dio list\n",
-                       __func__, ret, inode->i_ino);
-		return ret;
-	}
-
-	if (io->iocb)
-		aio_complete(io->iocb, io->result, 0);
-	/* clear the DIO AIO unwritten flag */
-	io->flag = 0;
-	return ret;
-}
-
-/*
- * work on completed aio dio IO, to convert unwritten extents to extents
- */
-static void ext4_end_io_work(struct work_struct *work)
-{
-	ext4_io_end_t		*io = container_of(work, ext4_io_end_t, work);
-	struct inode		*inode = io->inode;
-	struct ext4_inode_info	*ei = EXT4_I(inode);
-	unsigned long		flags;
-	int			ret;
-
-	mutex_lock(&inode->i_mutex);
-	ret = ext4_end_io_nolock(io);
-	if (ret < 0) {
-		mutex_unlock(&inode->i_mutex);
-		return;
-	}
-
-	spin_lock_irqsave(&ei->i_completed_io_lock, flags);
-	if (!list_empty(&io->list))
-		list_del_init(&io->list);
-	spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
-	mutex_unlock(&inode->i_mutex);
-	ext4_free_io_end(io);
-}
-
 /*
  * This function is called from ext4_sync_file().
  *
@@ -3756,28 +3688,6 @@ int flush_completed_IO(struct inode *inode)
 	return (ret2 < 0) ? ret2 : 0;
 }
 
-static ext4_io_end_t *ext4_init_io_end (struct inode *inode, gfp_t flags)
-{
-	ext4_io_end_t *io = NULL;
-
-	io = kmalloc(sizeof(*io), flags);
-
-	if (io) {
-		igrab(inode);
-		io->inode = inode;
-		io->flag = 0;
-		io->offset = 0;
-		io->size = 0;
-		io->page = NULL;
-		io->iocb = NULL;
-		io->result = 0;
-		INIT_WORK(&io->work, ext4_end_io_work);
-		INIT_LIST_HEAD(&io->list);
-	}
-
-	return io;
-}
-
 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
 			    ssize_t size, void *private, int ret,
 			    bool is_async)
@@ -3797,7 +3707,7 @@ static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
 		  size);
 
 	/* if not aio dio with unwritten extents, just free io and return */
-	if (io_end->flag != EXT4_IO_UNWRITTEN){
+	if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
 		ext4_free_io_end(io_end);
 		iocb->private = NULL;
 out:
@@ -3842,7 +3752,7 @@ static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
 		goto out;
 	}
 
-	io_end->flag = EXT4_IO_UNWRITTEN;
+	io_end->flag = EXT4_IO_END_UNWRITTEN;
 	inode = io_end->inode;
 
 	/* Add the io_end to per-inode completed io list*/

fs/ext4/super.c

@@ -4769,9 +4769,12 @@ static int __init init_ext4_fs(void)
 	int err;
 
 	ext4_check_flag_values();
-	err = init_ext4_system_zone();
+	err = init_ext4_pageio();
 	if (err)
 		return err;
+	err = init_ext4_system_zone();
+	if (err)
+		goto out5;
 	ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
 	if (!ext4_kset)
 		goto out4;
@@ -4812,6 +4815,8 @@ static int __init init_ext4_fs(void)
 	kset_unregister(ext4_kset);
 out4:
 	exit_ext4_system_zone();
+out5:
+	exit_ext4_pageio();
 	return err;
 }
 
@@ -4827,6 +4832,7 @@ static void __exit exit_ext4_fs(void)
 	remove_proc_entry("fs/ext4", NULL);
 	kset_unregister(ext4_kset);
 	exit_ext4_system_zone();
+	exit_ext4_pageio();
 }
 
 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");