f2fs: fix SSA updates resulting in corruption

The f2fs_collapse_range and f2fs_insert_range changes the block addresses
directly. But that can cause uncovered SSA updates.
In that case, we need to give up to change the block addresses and do buffered
writes to keep filesystem consistency.

Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

1 files changed, 93 insertions, 112 deletions

diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c
index 6d3cfd55c3e4..8d5583b21e5d 100644
--- a/fs/f2fs/file.c
+++ b/fs/f2fs/file.c
@@ -826,86 +826,100 @@ static int punch_hole(struct inode *inode, loff_t offset, loff_t len)
 	return ret;
 }
 
-static int f2fs_do_collapse(struct inode *inode, pgoff_t start, pgoff_t end)
+static int __exchange_data_block(struct inode *inode, pgoff_t src,
+					pgoff_t dst, bool full)
 {
 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 	struct dnode_of_data dn;
-	pgoff_t nrpages = (i_size_read(inode) + PAGE_SIZE - 1) / PAGE_SIZE;
-	int ret = 0;
-
-	for (; end < nrpages; start++, end++) {
-		block_t new_addr, old_addr;
-
-		f2fs_lock_op(sbi);
+	block_t new_addr;
+	bool do_replace = false;
+	int ret;
 
-		set_new_dnode(&dn, inode, NULL, NULL, 0);
-		ret = get_dnode_of_data(&dn, end, LOOKUP_NODE_RA);
-		if (ret && ret != -ENOENT) {
-			goto out;
-		} else if (ret == -ENOENT) {
-			new_addr = NULL_ADDR;
-		} else {
-			new_addr = dn.data_blkaddr;
-			truncate_data_blocks_range(&dn, 1);
-			f2fs_put_dnode(&dn);
+	set_new_dnode(&dn, inode, NULL, NULL, 0);
+	ret = get_dnode_of_data(&dn, src, LOOKUP_NODE_RA);
+	if (ret && ret != -ENOENT) {
+		return ret;
+	} else if (ret == -ENOENT) {
+		new_addr = NULL_ADDR;
+	} else {
+		new_addr = dn.data_blkaddr;
+		if (!is_checkpointed_data(sbi, new_addr)) {
+			dn.data_blkaddr = NULL_ADDR;
+			/* do not invalidate this block address */
+			set_data_blkaddr(&dn);
+			f2fs_update_extent_cache(&dn);
+			do_replace = true;
 		}
+		f2fs_put_dnode(&dn);
+	}
 
-		if (new_addr == NULL_ADDR) {
-			set_new_dnode(&dn, inode, NULL, NULL, 0);
-			ret = get_dnode_of_data(&dn, start, LOOKUP_NODE_RA);
-			if (ret && ret != -ENOENT) {
-				goto out;
-			} else if (ret == -ENOENT) {
-				f2fs_unlock_op(sbi);
-				continue;
-			}
+	if (new_addr == NULL_ADDR)
+		return full ? truncate_hole(inode, dst, dst + 1) : 0;
 
-			if (dn.data_blkaddr == NULL_ADDR) {
-				f2fs_put_dnode(&dn);
-				f2fs_unlock_op(sbi);
-				continue;
-			} else {
-				truncate_data_blocks_range(&dn, 1);
-			}
+	if (do_replace) {
+		struct page *ipage = get_node_page(sbi, inode->i_ino);
+		struct node_info ni;
 
-			f2fs_put_dnode(&dn);
-		} else {
-			struct page *ipage;
+		if (IS_ERR(ipage)) {
+			ret = PTR_ERR(ipage);
+			goto err_out;
+		}
 
-			ipage = get_node_page(sbi, inode->i_ino);
-			if (IS_ERR(ipage)) {
-				ret = PTR_ERR(ipage);
-				goto out;
-			}
+		set_new_dnode(&dn, inode, ipage, NULL, 0);
+		ret = f2fs_reserve_block(&dn, dst);
+		if (ret)
+			goto err_out;
 
-			set_new_dnode(&dn, inode, ipage, NULL, 0);
-			ret = f2fs_reserve_block(&dn, start);
-			if (ret)
-				goto out;
+		truncate_data_blocks_range(&dn, 1);
 
-			old_addr = dn.data_blkaddr;
-			if (old_addr != NEW_ADDR && new_addr == NEW_ADDR) {
-				dn.data_blkaddr = NULL_ADDR;
-				f2fs_update_extent_cache(&dn);
-				invalidate_blocks(sbi, old_addr);
+		get_node_info(sbi, dn.nid, &ni);
+		f2fs_replace_block(sbi, &dn, dn.data_blkaddr, new_addr,
+				ni.version, true);
+		f2fs_put_dnode(&dn);
+	} else {
+		struct page *psrc, *pdst;
+
+		psrc = get_lock_data_page(inode, src);
+		if (IS_ERR(psrc))
+			return PTR_ERR(psrc);
+		pdst = get_new_data_page(inode, NULL, dst, false);
+		if (IS_ERR(pdst)) {
+			f2fs_put_page(psrc, 1);
+			return PTR_ERR(pdst);
+		}
+		f2fs_copy_page(psrc, pdst);
+		set_page_dirty(pdst);
+		f2fs_put_page(pdst, 1);
+		f2fs_put_page(psrc, 1);
 
-				dn.data_blkaddr = new_addr;
-				set_data_blkaddr(&dn);
-			} else if (new_addr != NEW_ADDR) {
-				struct node_info ni;
+		return truncate_hole(inode, src, src + 1);
+	}
+	return 0;
 
-				get_node_info(sbi, dn.nid, &ni);
-				f2fs_replace_block(sbi, &dn, old_addr, new_addr,
-							ni.version, true);
-			}
+err_out:
+	if (!get_dnode_of_data(&dn, src, LOOKUP_NODE)) {
+		dn.data_blkaddr = new_addr;
+		set_data_blkaddr(&dn);
+		f2fs_update_extent_cache(&dn);
+		f2fs_put_dnode(&dn);
+	}
+	return ret;
+}
 
-			f2fs_put_dnode(&dn);
-		}
+static int f2fs_do_collapse(struct inode *inode, pgoff_t start, pgoff_t end)
+{
+	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
+	pgoff_t nrpages = (i_size_read(inode) + PAGE_SIZE - 1) / PAGE_SIZE;
+	int ret = 0;
+
+	for (; end < nrpages; start++, end++) {
+		f2fs_balance_fs(sbi);
+		f2fs_lock_op(sbi);
+		ret = __exchange_data_block(inode, end, start, true);
 		f2fs_unlock_op(sbi);
+		if (ret)
+			break;
 	}
-	return 0;
-out:
-	f2fs_unlock_op(sbi);
 	return ret;
 }
 
@@ -944,7 +958,12 @@ static int f2fs_collapse_range(struct inode *inode, loff_t offset, loff_t len)
 	if (ret)
 		return ret;
 
+	/* write out all moved pages, if possible */
+	filemap_write_and_wait_range(inode->i_mapping, offset, LLONG_MAX);
+	truncate_pagecache(inode, offset);
+
 	new_size = i_size_read(inode) - len;
+	truncate_pagecache(inode, new_size);
 
 	ret = truncate_blocks(inode, new_size, true);
 	if (!ret)
@@ -1067,7 +1086,7 @@ static int f2fs_insert_range(struct inode *inode, loff_t offset, loff_t len)
 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 	pgoff_t pg_start, pg_end, delta, nrpages, idx;
 	loff_t new_size;
-	int ret;
+	int ret = 0;
 
 	new_size = i_size_read(inode) + len;
 	if (new_size > inode->i_sb->s_maxbytes)
@@ -1105,57 +1124,19 @@ static int f2fs_insert_range(struct inode *inode, loff_t offset, loff_t len)
 	nrpages = (i_size_read(inode) + PAGE_SIZE - 1) / PAGE_SIZE;
 
 	for (idx = nrpages - 1; idx >= pg_start && idx != -1; idx--) {
-		struct dnode_of_data dn;
-		struct page *ipage;
-		block_t new_addr, old_addr;
-
 		f2fs_lock_op(sbi);
-
-		set_new_dnode(&dn, inode, NULL, NULL, 0);
-		ret = get_dnode_of_data(&dn, idx, LOOKUP_NODE_RA);
-		if (ret && ret != -ENOENT) {
-			goto out;
-		} else if (ret == -ENOENT) {
-			goto next;
-		} else if (dn.data_blkaddr == NULL_ADDR) {
-			f2fs_put_dnode(&dn);
-			goto next;
-		} else {
-			new_addr = dn.data_blkaddr;
-			truncate_data_blocks_range(&dn, 1);
-			f2fs_put_dnode(&dn);
-		}
-
-		ipage = get_node_page(sbi, inode->i_ino);
-		if (IS_ERR(ipage)) {
-			ret = PTR_ERR(ipage);
-			goto out;
-		}
-
-		set_new_dnode(&dn, inode, ipage, NULL, 0);
-		ret = f2fs_reserve_block(&dn, idx + delta);
-		if (ret)
-			goto out;
-
-		old_addr = dn.data_blkaddr;
-		f2fs_bug_on(sbi, old_addr != NEW_ADDR);
-
-		if (new_addr != NEW_ADDR) {
-			struct node_info ni;
-
-			get_node_info(sbi, dn.nid, &ni);
-			f2fs_replace_block(sbi, &dn, old_addr, new_addr,
-							ni.version, true);
-		}
-		f2fs_put_dnode(&dn);
-next:
+		ret = __exchange_data_block(inode, idx, idx + delta, false);
 		f2fs_unlock_op(sbi);
+		if (ret)
+			break;
 	}
 
-	i_size_write(inode, new_size);
-	return 0;
-out:
-	f2fs_unlock_op(sbi);
+	/* write out all moved pages, if possible */
+	filemap_write_and_wait_range(inode->i_mapping, offset, LLONG_MAX);
+	truncate_pagecache(inode, offset);
+
+	if (!ret)
+		i_size_write(inode, new_size);
 	return ret;
 }