1 files changed, 205 insertions, 168 deletions

diff --git a/fs/nilfs2/mdt.c b/fs/nilfs2/mdt.c
index 47dd815433f..c4dcd1db57e 100644
--- a/fs/nilfs2/mdt.c
+++ b/fs/nilfs2/mdt.c
@@ -26,7 +26,9 @@
 #include <linux/writeback.h>
 #include <linux/backing-dev.h>
 #include <linux/swap.h>
+#include <linux/slab.h>
 #include "nilfs.h"
+#include "btnode.h"
 #include "segment.h"
 #include "page.h"
 #include "mdt.h"
@@ -34,7 +36,6 @@
 
 #define NILFS_MDT_MAX_RA_BLOCKS		(16 - 1)
 
-#define INIT_UNUSED_INODE_FIELDS
 
 static int
 nilfs_mdt_insert_new_block(struct inode *inode, unsigned long block,
@@ -57,15 +58,15 @@ nilfs_mdt_insert_new_block(struct inode *inode, unsigned long block,
 
 	set_buffer_mapped(bh);
 
-	kaddr = kmap_atomic(bh->b_page, KM_USER0);
+	kaddr = kmap_atomic(bh->b_page);
 	memset(kaddr + bh_offset(bh), 0, 1 << inode->i_blkbits);
 	if (init_block)
 		init_block(inode, bh, kaddr);
 	flush_dcache_page(bh->b_page);
-	kunmap_atomic(kaddr, KM_USER0);
+	kunmap_atomic(kaddr);
 
 	set_buffer_uptodate(bh);
-	nilfs_mark_buffer_dirty(bh);
+	mark_buffer_dirty(bh);
 	nilfs_mdt_mark_dirty(inode);
 	return 0;
 }
@@ -76,22 +77,11 @@ static int nilfs_mdt_create_block(struct inode *inode, unsigned long block,
 						     struct buffer_head *,
 						     void *))
 {
-	struct the_nilfs *nilfs = NILFS_MDT(inode)->mi_nilfs;
-	struct nilfs_sb_info *writer = NULL;
 	struct super_block *sb = inode->i_sb;
 	struct nilfs_transaction_info ti;
 	struct buffer_head *bh;
 	int err;
 
-	if (!sb) {
-		writer = nilfs_get_writer(nilfs);
-		if (!writer) {
-			err = -EROFS;
-			goto out;
-		}
-		sb = writer->s_super;
-	}
-
 	nilfs_transaction_begin(sb, &ti, 0);
 
 	err = -ENOMEM;
@@ -100,17 +90,14 @@ static int nilfs_mdt_create_block(struct inode *inode, unsigned long block,
 		goto failed_unlock;
 
 	err = -EEXIST;
-	if (buffer_uptodate(bh) || buffer_mapped(bh))
+	if (buffer_uptodate(bh))
 		goto failed_bh;
-#if 0
-	/* The uptodate flag is not protected by the page lock, but
-	   the mapped flag is.  Thus, we don't have to wait the buffer. */
+
 	wait_on_buffer(bh);
 	if (buffer_uptodate(bh))
 		goto failed_bh;
-#endif
 
-	bh->b_bdev = nilfs->ns_bdev;
+	bh->b_bdev = sb->s_bdev;
 	err = nilfs_mdt_insert_new_block(inode, block, bh, init_block);
 	if (likely(!err)) {
 		get_bh(bh);
@@ -127,9 +114,7 @@ static int nilfs_mdt_create_block(struct inode *inode, unsigned long block,
 		err = nilfs_transaction_commit(sb);
 	else
 		nilfs_transaction_abort(sb);
-	if (writer)
-		nilfs_put_writer(nilfs);
- out:
+
 	return err;
 }
 
@@ -138,7 +123,7 @@ nilfs_mdt_submit_block(struct inode *inode, unsigned long blkoff,
 		       int mode, struct buffer_head **out_bh)
 {
 	struct buffer_head *bh;
-	unsigned long blknum = 0;
+	__u64 blknum = 0;
 	int ret = -ENOMEM;
 
 	bh = nilfs_grab_buffer(inode, inode->i_mapping, blkoff, 0);
@@ -161,17 +146,13 @@ nilfs_mdt_submit_block(struct inode *inode, unsigned long blkoff,
 		unlock_buffer(bh);
 		goto out;
 	}
-	if (!buffer_mapped(bh)) { /* unused buffer */
-		ret = nilfs_bmap_lookup(NILFS_I(inode)->i_bmap, blkoff,
-					&blknum);
-		if (unlikely(ret)) {
-			unlock_buffer(bh);
-			goto failed_bh;
-		}
-		bh->b_bdev = NILFS_MDT(inode)->mi_nilfs->ns_bdev;
-		bh->b_blocknr = blknum;
-		set_buffer_mapped(bh);
+
+	ret = nilfs_bmap_lookup(NILFS_I(inode)->i_bmap, blkoff, &blknum);
+	if (unlikely(ret)) {
+		unlock_buffer(bh);
+		goto failed_bh;
 	}
+	map_bh(bh, inode->i_sb, (sector_t)blknum);
 
 	bh->b_end_io = end_buffer_read_sync;
 	get_bh(bh);
@@ -190,7 +171,7 @@ nilfs_mdt_submit_block(struct inode *inode, unsigned long blkoff,
 }
 
 static int nilfs_mdt_read_block(struct inode *inode, unsigned long block,
-				struct buffer_head **out_bh)
+				int readahead, struct buffer_head **out_bh)
 {
 	struct buffer_head *first_bh, *bh;
 	unsigned long blkoff;
@@ -204,16 +185,18 @@ static int nilfs_mdt_read_block(struct inode *inode, unsigned long block,
 	if (unlikely(err))
 		goto failed;
 
-	blkoff = block + 1;
-	for (i = 0; i < nr_ra_blocks; i++, blkoff++) {
-		err = nilfs_mdt_submit_block(inode, blkoff, READA, &bh);
-		if (likely(!err || err == -EEXIST))
-			brelse(bh);
-		else if (err != -EBUSY)
-			break; /* abort readahead if bmap lookup failed */
-
-		if (!buffer_locked(first_bh))
-			goto out_no_wait;
+	if (readahead) {
+		blkoff = block + 1;
+		for (i = 0; i < nr_ra_blocks; i++, blkoff++) {
+			err = nilfs_mdt_submit_block(inode, blkoff, READA, &bh);
+			if (likely(!err || err == -EEXIST))
+				brelse(bh);
+			else if (err != -EBUSY)
+				break;
+				/* abort readahead if bmap lookup failed */
+			if (!buffer_locked(first_bh))
+				goto out_no_wait;
+		}
 	}
 
 	wait_on_buffer(first_bh);
@@ -254,8 +237,6 @@ static int nilfs_mdt_read_block(struct inode *inode, unsigned long block,
  *
  * %-ENOENT - the specified block does not exist (hole block)
  *
- * %-EINVAL - bmap is broken. (the caller should call nilfs_error())
- *
  * %-EROFS - Read only filesystem (for create mode)
  */
 int nilfs_mdt_get_block(struct inode *inode, unsigned long blkoff, int create,
@@ -267,7 +248,7 @@ int nilfs_mdt_get_block(struct inode *inode, unsigned long blkoff, int create,
 
 	/* Should be rewritten with merging nilfs_mdt_read_block() */
  retry:
-	ret = nilfs_mdt_read_block(inode, blkoff, out_bh);
+	ret = nilfs_mdt_read_block(inode, blkoff, !create, out_bh);
 	if (!create || ret != -ENOENT)
 		return ret;
 
@@ -290,8 +271,6 @@ int nilfs_mdt_get_block(struct inode *inode, unsigned long blkoff, int create,
  * %-ENOMEM - Insufficient memory available.
  *
  * %-EIO - I/O error
- *
- * %-EINVAL - bmap is broken. (the caller should call nilfs_error())
  */
 int nilfs_mdt_delete_block(struct inode *inode, unsigned long block)
 {
@@ -299,7 +278,7 @@ int nilfs_mdt_delete_block(struct inode *inode, unsigned long block)
 	int err;
 
 	err = nilfs_bmap_delete(ii->i_bmap, block);
-	if (likely(!err)) {
+	if (!err || err == -ENOENT) {
 		nilfs_mdt_mark_dirty(inode);
 		nilfs_mdt_forget_block(inode, block);
 	}
@@ -367,18 +346,16 @@ int nilfs_mdt_forget_block(struct inode *inode, unsigned long block)
  * %-EIO - I/O error
  *
  * %-ENOENT - the specified block does not exist (hole block)
- *
- * %-EINVAL - bmap is broken. (the caller should call nilfs_error())
  */
 int nilfs_mdt_mark_block_dirty(struct inode *inode, unsigned long block)
 {
 	struct buffer_head *bh;
 	int err;
 
-	err = nilfs_mdt_read_block(inode, block, &bh);
+	err = nilfs_mdt_read_block(inode, block, 0, &bh);
 	if (unlikely(err))
 		return err;
-	nilfs_mark_buffer_dirty(bh);
+	mark_buffer_dirty(bh);
 	nilfs_mdt_mark_dirty(inode);
 	brelse(bh);
 	return 0;
@@ -398,131 +375,67 @@ int nilfs_mdt_fetch_dirty(struct inode *inode)
 static int
 nilfs_mdt_write_page(struct page *page, struct writeback_control *wbc)
 {
-	struct inode *inode = container_of(page->mapping,
-					   struct inode, i_data);
-	struct super_block *sb = inode->i_sb;
-	struct nilfs_sb_info *writer = NULL;
+	struct inode *inode = page->mapping->host;
+	struct super_block *sb;
 	int err = 0;
 
+	if (inode && (inode->i_sb->s_flags & MS_RDONLY)) {
+		/*
+		 * It means that filesystem was remounted in read-only
+		 * mode because of error or metadata corruption. But we
+		 * have dirty pages that try to be flushed in background.
+		 * So, here we simply discard this dirty page.
+		 */
+		nilfs_clear_dirty_page(page, false);
+		unlock_page(page);
+		return -EROFS;
+	}
+
 	redirty_page_for_writepage(wbc, page);
 	unlock_page(page);
 
-	if (page->mapping->assoc_mapping)
-		return 0; /* Do not request flush for shadow page cache */
-	if (!sb) {
-		writer = nilfs_get_writer(NILFS_MDT(inode)->mi_nilfs);
-		if (!writer)
-			return -EROFS;
-		sb = writer->s_super;
-	}
+	if (!inode)
+		return 0;
+
+	sb = inode->i_sb;
 
 	if (wbc->sync_mode == WB_SYNC_ALL)
 		err = nilfs_construct_segment(sb);
 	else if (wbc->for_reclaim)
 		nilfs_flush_segment(sb, inode->i_ino);
 
-	if (writer)
-		nilfs_put_writer(NILFS_MDT(inode)->mi_nilfs);
 	return err;
 }
 
 
-static struct address_space_operations def_mdt_aops = {
+static const struct address_space_operations def_mdt_aops = {
 	.writepage		= nilfs_mdt_write_page,
 };
 
-static struct inode_operations def_mdt_iops;
-static struct file_operations def_mdt_fops;
+static const struct inode_operations def_mdt_iops;
+static const struct file_operations def_mdt_fops;
 
-/*
- * NILFS2 uses pseudo inodes for meta data files such as DAT, cpfile, sufile,
- * ifile, or gcinodes.  This allows the B-tree code and segment constructor
- * to treat them like regular files, and this helps to simplify the
- * implementation.
- *   On the other hand, some of the pseudo inodes have an irregular point:
- * They don't have valid inode->i_sb pointer because their lifetimes are
- * longer than those of the super block structs; they may continue for
- * several consecutive mounts/umounts.  This would need discussions.
- */
-struct inode *
-nilfs_mdt_new_common(struct the_nilfs *nilfs, struct super_block *sb,
-		     ino_t ino, gfp_t gfp_mask)
-{
-	struct inode *inode = nilfs_alloc_inode(sb);
 
-	if (!inode)
-		return NULL;
-	else {
-		struct address_space * const mapping = &inode->i_data;
-		struct nilfs_mdt_info *mi = kzalloc(sizeof(*mi), GFP_NOFS);
-
-		if (!mi) {
-			nilfs_destroy_inode(inode);
-			return NULL;
-		}
-		mi->mi_nilfs = nilfs;
-		init_rwsem(&mi->mi_sem);
-
-		inode->i_sb = sb; /* sb may be NULL for some meta data files */
-		inode->i_blkbits = nilfs->ns_blocksize_bits;
-		inode->i_flags = 0;
-		atomic_set(&inode->i_count, 1);
-		inode->i_nlink = 1;
-		inode->i_ino = ino;
-		inode->i_mode = S_IFREG;
-		inode->i_private = mi;
-
-#ifdef INIT_UNUSED_INODE_FIELDS
-		atomic_set(&inode->i_writecount, 0);
-		inode->i_size = 0;
-		inode->i_blocks = 0;
-		inode->i_bytes = 0;
-		inode->i_generation = 0;
-#ifdef CONFIG_QUOTA
-		memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
-#endif
-		inode->i_pipe = NULL;
-		inode->i_bdev = NULL;
-		inode->i_cdev = NULL;
-		inode->i_rdev = 0;
-#ifdef CONFIG_SECURITY
-		inode->i_security = NULL;
-#endif
-		inode->dirtied_when = 0;
-
-		INIT_LIST_HEAD(&inode->i_list);
-		INIT_LIST_HEAD(&inode->i_sb_list);
-		inode->i_state = 0;
-#endif
-
-		spin_lock_init(&inode->i_lock);
-		mutex_init(&inode->i_mutex);
-		init_rwsem(&inode->i_alloc_sem);
-
-		mapping->host = NULL;  /* instead of inode */
-		mapping->flags = 0;
-		mapping_set_gfp_mask(mapping, gfp_mask);
-		mapping->assoc_mapping = NULL;
-		mapping->backing_dev_info = nilfs->ns_bdi;
-
-		inode->i_mapping = mapping;
-	}
+int nilfs_mdt_init(struct inode *inode, gfp_t gfp_mask, size_t objsz)
+{
+	struct nilfs_mdt_info *mi;
 
-	return inode;
-}
+	mi = kzalloc(max(sizeof(*mi), objsz), GFP_NOFS);
+	if (!mi)
+		return -ENOMEM;
 
-struct inode *nilfs_mdt_new(struct the_nilfs *nilfs, struct super_block *sb,
-			    ino_t ino, gfp_t gfp_mask)
-{
-	struct inode *inode = nilfs_mdt_new_common(nilfs, sb, ino, gfp_mask);
+	init_rwsem(&mi->mi_sem);
+	inode->i_private = mi;
 
-	if (!inode)
-		return NULL;
+	inode->i_mode = S_IFREG;
+	mapping_set_gfp_mask(inode->i_mapping, gfp_mask);
+	inode->i_mapping->backing_dev_info = inode->i_sb->s_bdi;
 
 	inode->i_op = &def_mdt_iops;
 	inode->i_fop = &def_mdt_fops;
 	inode->i_mapping->a_ops = &def_mdt_aops;
-	return inode;
+
+	return 0;
 }
 
 void nilfs_mdt_set_entry_size(struct inode *inode, unsigned entry_size,
@@ -535,29 +448,153 @@ void nilfs_mdt_set_entry_size(struct inode *inode, unsigned entry_size,
 	mi->mi_first_entry_offset = DIV_ROUND_UP(header_size, entry_size);
 }
 
-void nilfs_mdt_set_shadow(struct inode *orig, struct inode *shadow)
+/**
+ * nilfs_mdt_setup_shadow_map - setup shadow map and bind it to metadata file
+ * @inode: inode of the metadata file
+ * @shadow: shadow mapping
+ */
+int nilfs_mdt_setup_shadow_map(struct inode *inode,
+			       struct nilfs_shadow_map *shadow)
+{
+	struct nilfs_mdt_info *mi = NILFS_MDT(inode);
+	struct backing_dev_info *bdi = inode->i_sb->s_bdi;
+
+	INIT_LIST_HEAD(&shadow->frozen_buffers);
+	address_space_init_once(&shadow->frozen_data);
+	nilfs_mapping_init(&shadow->frozen_data, inode, bdi);
+	address_space_init_once(&shadow->frozen_btnodes);
+	nilfs_mapping_init(&shadow->frozen_btnodes, inode, bdi);
+	mi->mi_shadow = shadow;
+	return 0;
+}
+
+/**
+ * nilfs_mdt_save_to_shadow_map - copy bmap and dirty pages to shadow map
+ * @inode: inode of the metadata file
+ */
+int nilfs_mdt_save_to_shadow_map(struct inode *inode)
+{
+	struct nilfs_mdt_info *mi = NILFS_MDT(inode);
+	struct nilfs_inode_info *ii = NILFS_I(inode);
+	struct nilfs_shadow_map *shadow = mi->mi_shadow;
+	int ret;
+
+	ret = nilfs_copy_dirty_pages(&shadow->frozen_data, inode->i_mapping);
+	if (ret)
+		goto out;
+
+	ret = nilfs_copy_dirty_pages(&shadow->frozen_btnodes,
+				     &ii->i_btnode_cache);
+	if (ret)
+		goto out;
+
+	nilfs_bmap_save(ii->i_bmap, &shadow->bmap_store);
+ out:
+	return ret;
+}
+
+int nilfs_mdt_freeze_buffer(struct inode *inode, struct buffer_head *bh)
+{
+	struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
+	struct buffer_head *bh_frozen;
+	struct page *page;
+	int blkbits = inode->i_blkbits;
+
+	page = grab_cache_page(&shadow->frozen_data, bh->b_page->index);
+	if (!page)
+		return -ENOMEM;
+
+	if (!page_has_buffers(page))
+		create_empty_buffers(page, 1 << blkbits, 0);
+
+	bh_frozen = nilfs_page_get_nth_block(page, bh_offset(bh) >> blkbits);
+
+	if (!buffer_uptodate(bh_frozen))
+		nilfs_copy_buffer(bh_frozen, bh);
+	if (list_empty(&bh_frozen->b_assoc_buffers)) {
+		list_add_tail(&bh_frozen->b_assoc_buffers,
+			      &shadow->frozen_buffers);
+		set_buffer_nilfs_redirected(bh);
+	} else {
+		brelse(bh_frozen); /* already frozen */
+	}
+
+	unlock_page(page);
+	page_cache_release(page);
+	return 0;
+}
+
+struct buffer_head *
+nilfs_mdt_get_frozen_buffer(struct inode *inode, struct buffer_head *bh)
 {
-	shadow->i_mapping->assoc_mapping = orig->i_mapping;
-	NILFS_I(shadow)->i_btnode_cache.assoc_mapping =
-		&NILFS_I(orig)->i_btnode_cache;
+	struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
+	struct buffer_head *bh_frozen = NULL;
+	struct page *page;
+	int n;
+
+	page = find_lock_page(&shadow->frozen_data, bh->b_page->index);
+	if (page) {
+		if (page_has_buffers(page)) {
+			n = bh_offset(bh) >> inode->i_blkbits;
+			bh_frozen = nilfs_page_get_nth_block(page, n);
+		}
+		unlock_page(page);
+		page_cache_release(page);
+	}
+	return bh_frozen;
 }
 
-void nilfs_mdt_clear(struct inode *inode)
+static void nilfs_release_frozen_buffers(struct nilfs_shadow_map *shadow)
 {
+	struct list_head *head = &shadow->frozen_buffers;
+	struct buffer_head *bh;
+
+	while (!list_empty(head)) {
+		bh = list_first_entry(head, struct buffer_head,
+				      b_assoc_buffers);
+		list_del_init(&bh->b_assoc_buffers);
+		brelse(bh); /* drop ref-count to make it releasable */
+	}
+}
+
+/**
+ * nilfs_mdt_restore_from_shadow_map - restore dirty pages and bmap state
+ * @inode: inode of the metadata file
+ */
+void nilfs_mdt_restore_from_shadow_map(struct inode *inode)
+{
+	struct nilfs_mdt_info *mi = NILFS_MDT(inode);
 	struct nilfs_inode_info *ii = NILFS_I(inode);
+	struct nilfs_shadow_map *shadow = mi->mi_shadow;
+
+	down_write(&mi->mi_sem);
+
+	if (mi->mi_palloc_cache)
+		nilfs_palloc_clear_cache(inode);
 
-	invalidate_mapping_pages(inode->i_mapping, 0, -1);
-	truncate_inode_pages(inode->i_mapping, 0);
+	nilfs_clear_dirty_pages(inode->i_mapping, true);
+	nilfs_copy_back_pages(inode->i_mapping, &shadow->frozen_data);
 
-	nilfs_bmap_clear(ii->i_bmap);
-	nilfs_btnode_cache_clear(&ii->i_btnode_cache);
+	nilfs_clear_dirty_pages(&ii->i_btnode_cache, true);
+	nilfs_copy_back_pages(&ii->i_btnode_cache, &shadow->frozen_btnodes);
+
+	nilfs_bmap_restore(ii->i_bmap, &shadow->bmap_store);
+
+	up_write(&mi->mi_sem);
 }
 
-void nilfs_mdt_destroy(struct inode *inode)
+/**
+ * nilfs_mdt_clear_shadow_map - truncate pages in shadow map caches
+ * @inode: inode of the metadata file
+ */
+void nilfs_mdt_clear_shadow_map(struct inode *inode)
 {
-	struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
+	struct nilfs_mdt_info *mi = NILFS_MDT(inode);
+	struct nilfs_shadow_map *shadow = mi->mi_shadow;
 
-	kfree(mdi->mi_bgl); /* kfree(NULL) is safe */
-	kfree(mdi);
-	nilfs_destroy_inode(inode);
+	down_write(&mi->mi_sem);
+	nilfs_release_frozen_buffers(shadow);
+	truncate_inode_pages(&shadow->frozen_data, 0);
+	truncate_inode_pages(&shadow->frozen_btnodes, 0);
+	up_write(&mi->mi_sem);
 }