Btrfs: Retry metadata reads in the face of checksum failures

Signed-off-by: Chris Mason <chris.mason@oracle.com>

fs/btrfs/ctree.c

@@ -569,7 +569,16 @@ static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
 static int noinline check_block(struct btrfs_root *root,
 				struct btrfs_path *path, int level)
 {
-	return 0;
+	u64 found_start;
+	if (btrfs_header_level(path->nodes[level]) != level)
+	    printk("warning: bad level %Lu wanted %d found %d\n",
+		   path->nodes[level]->start, level,
+		   btrfs_header_level(path->nodes[level]));
+	found_start = btrfs_header_bytenr(path->nodes[level]);
+	if (found_start != path->nodes[level]->start) {
+	    printk("warning: bad bytentr %Lu found %Lu\n",
+		   path->nodes[level]->start, found_start);
+	}
 #if 0
 	struct extent_buffer *buf = path->nodes[level];
 

fs/btrfs/disk-io.c

@@ -156,7 +156,6 @@ static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 			memcpy(&found, result, BTRFS_CRC32_SIZE);
 
 			read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
-			WARN_ON(1);
 			printk("btrfs: %s checksum verify failed on %llu "
 			       "wanted %X found %X from_this_trans %d "
 			       "level %d\n",
@@ -171,6 +170,40 @@ static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 	return 0;
 }
 
+static int btree_read_extent_buffer_pages(struct btrfs_root *root,
+					  struct extent_buffer *eb,
+					  u64 start)
+{
+	struct extent_io_tree *io_tree;
+	int ret;
+	int num_copies = 0;
+	int mirror_num = 0;
+
+	io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
+	while (1) {
+		ret = read_extent_buffer_pages(io_tree, eb, start, 1,
+					       btree_get_extent, mirror_num);
+		if (!ret) {
+			if (mirror_num)
+printk("good read %Lu mirror %d total %d\n", eb->start, mirror_num, num_copies);
+			return ret;
+		}
+		num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
+					      eb->start, eb->len);
+printk("failed to read %Lu mirror %d total %d\n", eb->start, mirror_num, num_copies);
+		if (num_copies == 1) {
+printk("reading %Lu failed only one copy\n", eb->start);
+			return ret;
+		}
+		mirror_num++;
+		if (mirror_num > num_copies) {
+printk("bailing at mirror %d of %d\n", mirror_num, num_copies);
+			return ret;
+		}
+	}
+printk("read extent buffer page last\n");
+	return -EIO;
+}
 
 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 {
@@ -180,6 +213,8 @@ int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 	int found_level;
 	unsigned long len;
 	struct extent_buffer *eb;
+	int ret;
+
 	tree = &BTRFS_I(page->mapping->host)->io_tree;
 
 	if (page->private == EXTENT_PAGE_PRIVATE)
@@ -191,8 +226,8 @@ int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 		WARN_ON(1);
 	}
 	eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
-	read_extent_buffer_pages(tree, eb, start + PAGE_CACHE_SIZE, 1,
-				 btree_get_extent);
+	ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE);
+	BUG_ON(ret);
 	btrfs_clear_buffer_defrag(eb);
 	found_start = btrfs_header_bytenr(eb);
 	if (found_start != start) {
@@ -240,7 +275,7 @@ int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 	unsigned long len;
 	struct extent_buffer *eb;
 	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
-	int ret;
+	int ret = 0;
 
 	tree = &BTRFS_I(page->mapping->host)->io_tree;
 	if (page->private == EXTENT_PAGE_PRIVATE)
@@ -252,25 +287,26 @@ int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 		WARN_ON(1);
 	}
 	eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
-	read_extent_buffer_pages(tree, eb, start + PAGE_CACHE_SIZE, 1,
-				 btree_get_extent);
+
 	btrfs_clear_buffer_defrag(eb);
 	found_start = btrfs_header_bytenr(eb);
 	if (found_start != start) {
-		printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
-		       start, found_start, len);
-		WARN_ON(1);
+printk("bad start on %Lu found %Lu\n", eb->start, found_start);
+		ret = -EIO;
 		goto err;
 	}
 	if (eb->first_page != page) {
 		printk("bad first page %lu %lu\n", eb->first_page->index,
 		       page->index);
 		WARN_ON(1);
+		ret = -EIO;
 		goto err;
 	}
 	found_level = btrfs_header_level(eb);
 
 	ret = csum_tree_block(root, eb, 1);
+	if (ret)
+		ret = -EIO;
 
 	end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 	end = eb->start + end - 1;
@@ -278,7 +314,7 @@ int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 err:
 	free_extent_buffer(eb);
 out:
-	return 0;
+	return ret;
 }
 
 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
@@ -329,7 +365,8 @@ int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 	return 0;
 }
 
-static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio)
+static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
+				 int mirror_num)
 {
 	struct btrfs_root *root = BTRFS_I(inode)->root;
 	u64 offset;
@@ -338,7 +375,7 @@ static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio)
 	offset = bio->bi_sector << 9;
 
 	if (rw & (1 << BIO_RW)) {
-		return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio);
+		return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num);
 	}
 
 	ret = btrfs_bio_wq_end_io(root->fs_info, bio, 1);
@@ -349,7 +386,7 @@ static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio)
 		submit_bio(rw, bio);
 		return 0;
 	}
-	return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio);
+	return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num);
 }
 
 static int btree_writepage(struct page *page, struct writeback_control *wbc)
@@ -459,7 +496,7 @@ int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize)
 	if (!buf)
 		return 0;
 	read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
-				 buf, 0, 0, btree_get_extent);
+				 buf, 0, 0, btree_get_extent, 0);
 	free_extent_buffer(buf);
 	return ret;
 }
@@ -522,8 +559,7 @@ struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 	if (!buf)
 		return NULL;
 
-	ret = read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree, buf, 0,
-				       1, btree_get_extent);
+	ret = btree_read_extent_buffer_pages(root, buf, 0);
 
 	if (ret == 0) {
 		buf->flags |= EXTENT_UPTODATE;
@@ -1366,10 +1402,8 @@ int btrfs_clear_buffer_defrag(struct extent_buffer *buf)
 int btrfs_read_buffer(struct extent_buffer *buf)
 {
 	struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
-	struct inode *btree_inode = root->fs_info->btree_inode;
 	int ret;
-	ret = read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
-					buf, 0, 1, btree_get_extent);
+	ret = btree_read_extent_buffer_pages(root, buf, 0);
 	if (ret == 0) {
 		buf->flags |= EXTENT_UPTODATE;
 	}

fs/btrfs/extent_io.c

@@ -1546,7 +1546,7 @@ static int end_bio_extent_readpage(struct bio *bio,
 				    !(state->state & EXTENT_LOCKED))
 					state = NULL;
 			}
-			if (!state) {
+			if (!state && uptodate) {
 				spin_unlock_irqrestore(&tree->lock, flags);
 				set_extent_uptodate(tree, start, end,
 						    GFP_ATOMIC);
@@ -1567,8 +1567,10 @@ static int end_bio_extent_readpage(struct bio *bio,
 			} else {
 				state = NULL;
 			}
-			set_state_cb(tree, clear, EXTENT_UPTODATE);
-			clear->state |= EXTENT_UPTODATE;
+			if (uptodate) {
+				set_state_cb(tree, clear, EXTENT_UPTODATE);
+				clear->state |= EXTENT_UPTODATE;
+			}
 			clear_state_bit(tree, clear, EXTENT_LOCKED,
 					1, 0);
 			if (cur == start)
@@ -1685,7 +1687,7 @@ extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
 	return bio;
 }
 
-static int submit_one_bio(int rw, struct bio *bio)
+static int submit_one_bio(int rw, struct bio *bio, int mirror_num)
 {
 	u64 maxsector;
 	int ret = 0;
@@ -1722,7 +1724,8 @@ static int submit_one_bio(int rw, struct bio *bio)
 		WARN_ON(1);
 	}
 	if (tree->ops && tree->ops->submit_bio_hook)
-		tree->ops->submit_bio_hook(page->mapping->host, rw, bio);
+		tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
+					   mirror_num);
 	else
 		submit_bio(rw, bio);
 	if (bio_flagged(bio, BIO_EOPNOTSUPP))
@@ -1737,7 +1740,8 @@ static int submit_extent_page(int rw, struct extent_io_tree *tree,
 			      struct block_device *bdev,
 			      struct bio **bio_ret,
 			      unsigned long max_pages,
-			      bio_end_io_t end_io_func)
+			      bio_end_io_t end_io_func,
+			      int mirror_num)
 {
 	int ret = 0;
 	struct bio *bio;
@@ -1749,7 +1753,7 @@ static int submit_extent_page(int rw, struct extent_io_tree *tree,
 		    (tree->ops && tree->ops->merge_bio_hook &&
 		     tree->ops->merge_bio_hook(page, offset, size, bio)) ||
 		    bio_add_page(bio, page, size, offset) < size) {
-			ret = submit_one_bio(rw, bio);
+			ret = submit_one_bio(rw, bio, mirror_num);
 			bio = NULL;
 		} else {
 			return 0;
@@ -1769,7 +1773,7 @@ static int submit_extent_page(int rw, struct extent_io_tree *tree,
 	if (bio_ret) {
 		*bio_ret = bio;
 	} else {
-		ret = submit_one_bio(rw, bio);
+		ret = submit_one_bio(rw, bio, mirror_num);
 	}
 
 	return ret;
@@ -1798,7 +1802,7 @@ void set_page_extent_head(struct page *page, unsigned long len)
 static int __extent_read_full_page(struct extent_io_tree *tree,
 				   struct page *page,
 				   get_extent_t *get_extent,
-				   struct bio **bio)
+				   struct bio **bio, int mirror_num)
 {
 	struct inode *inode = page->mapping->host;
 	u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
@@ -1901,7 +1905,7 @@ static int __extent_read_full_page(struct extent_io_tree *tree,
 			ret = submit_extent_page(READ, tree, page,
 					 sector, iosize, page_offset,
 					 bdev, bio, nr,
-					 end_bio_extent_readpage);
+					 end_bio_extent_readpage, mirror_num);
 		}
 		if (ret)
 			SetPageError(page);
@@ -1923,9 +1927,9 @@ int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
 	struct bio *bio = NULL;
 	int ret;
 
-	ret = __extent_read_full_page(tree, page, get_extent, &bio);
+	ret = __extent_read_full_page(tree, page, get_extent, &bio, 0);
 	if (bio)
-		submit_one_bio(READ, bio);
+		submit_one_bio(READ, bio, 0);
 	return ret;
 }
 EXPORT_SYMBOL(extent_read_full_page);
@@ -2077,7 +2081,7 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
 			ret = submit_extent_page(WRITE, tree, page, sector,
 						 iosize, page_offset, bdev,
 						 &epd->bio, max_nr,
-						 end_bio_extent_writepage);
+						 end_bio_extent_writepage, 0);
 			if (ret)
 				SetPageError(page);
 		}
@@ -2244,7 +2248,7 @@ int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
 
 	write_cache_pages(mapping, &wbc_writepages, __extent_writepage, &epd);
 	if (epd.bio) {
-		submit_one_bio(WRITE, epd.bio);
+		submit_one_bio(WRITE, epd.bio, 0);
 	}
 	return ret;
 }
@@ -2265,7 +2269,7 @@ int extent_writepages(struct extent_io_tree *tree,
 
 	ret = write_cache_pages(mapping, wbc, __extent_writepage, &epd);
 	if (epd.bio) {
-		submit_one_bio(WRITE, epd.bio);
+		submit_one_bio(WRITE, epd.bio, 0);
 	}
 	return ret;
 }
@@ -2297,7 +2301,8 @@ int extent_readpages(struct extent_io_tree *tree,
 			page_cache_get(page);
 			if (!pagevec_add(&pvec, page))
 				__pagevec_lru_add(&pvec);
-			__extent_read_full_page(tree, page, get_extent, &bio);
+			__extent_read_full_page(tree, page, get_extent,
+						&bio, 0);
 		}
 		page_cache_release(page);
 	}
@@ -2305,7 +2310,7 @@ int extent_readpages(struct extent_io_tree *tree,
 		__pagevec_lru_add(&pvec);
 	BUG_ON(!list_empty(pages));
 	if (bio)
-		submit_one_bio(READ, bio);
+		submit_one_bio(READ, bio, 0);
 	return 0;
 }
 EXPORT_SYMBOL(extent_readpages);
@@ -2430,7 +2435,7 @@ int extent_prepare_write(struct extent_io_tree *tree,
 			ret = submit_extent_page(READ, tree, page,
 					 sector, iosize, page_offset, em->bdev,
 					 NULL, 1,
-					 end_bio_extent_preparewrite);
+					 end_bio_extent_preparewrite, 0);
 			iocount++;
 			block_start = block_start + iosize;
 		} else {
@@ -2696,6 +2701,7 @@ struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
 		mark_page_accessed(page0);
 		set_page_extent_mapped(page0);
 		set_page_extent_head(page0, len);
+		uptodate = PageUptodate(page0);
 	} else {
 		i = 0;
 	}
@@ -3006,7 +3012,7 @@ EXPORT_SYMBOL(extent_buffer_uptodate);
 int read_extent_buffer_pages(struct extent_io_tree *tree,
 			     struct extent_buffer *eb,
 			     u64 start, int wait,
-			     get_extent_t *get_extent)
+			     get_extent_t *get_extent, int mirror_num)
 {
 	unsigned long i;
 	unsigned long start_i;
@@ -3062,8 +3068,10 @@ int read_extent_buffer_pages(struct extent_io_tree *tree,
 		if (!PageUptodate(page)) {
 			if (start_i == 0)
 				inc_all_pages = 1;
+			ClearPageError(page);
 			err = __extent_read_full_page(tree, page,
-						      get_extent, &bio);
+						      get_extent, &bio,
+						      mirror_num);
 			if (err) {
 				ret = err;
 			}
@@ -3073,7 +3081,7 @@ int read_extent_buffer_pages(struct extent_io_tree *tree,
 	}
 
 	if (bio)
-		submit_one_bio(READ, bio);
+		submit_one_bio(READ, bio, mirror_num);
 
 	if (ret || !wait) {
 		return ret;

fs/btrfs/extent_io.h

@@ -27,7 +27,8 @@ struct extent_state;
 struct extent_io_ops {
 	int (*fill_delalloc)(struct inode *inode, u64 start, u64 end);
 	int (*writepage_io_hook)(struct page *page, u64 start, u64 end);
-	int (*submit_bio_hook)(struct inode *inode, int rw, struct bio *bio);
+	int (*submit_bio_hook)(struct inode *inode, int rw, struct bio *bio,
+			       int mirror_num);
 	int (*merge_bio_hook)(struct page *page, unsigned long offset,
 			      size_t size, struct bio *bio);
 	int (*readpage_io_hook)(struct page *page, u64 start, u64 end);
@@ -172,7 +173,7 @@ struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
 void free_extent_buffer(struct extent_buffer *eb);
 int read_extent_buffer_pages(struct extent_io_tree *tree,
 			     struct extent_buffer *eb, u64 start, int wait,
-			     get_extent_t *get_extent);
+			     get_extent_t *get_extent, int mirror_num);
 
 static inline void extent_buffer_get(struct extent_buffer *eb)
 {

fs/btrfs/inode.c

@@ -314,7 +314,7 @@ int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
 	map_tree = &root->fs_info->mapping_tree;
 	map_length = length;
 	ret = btrfs_map_block(map_tree, READ, logical,
-			      &map_length, NULL);
+			      &map_length, NULL, 0);
 
 	if (map_length < length + size) {
 		return 1;
@@ -322,7 +322,8 @@ int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
 	return 0;
 }
 
-int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio)
+int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
+			  int mirror_num)
 {
 	struct btrfs_root *root = BTRFS_I(inode)->root;
 	struct btrfs_trans_handle *trans;
@@ -347,7 +348,7 @@ int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio)
 	BUG_ON(ret);
 	mutex_unlock(&root->fs_info->fs_mutex);
 mapit:
-	return btrfs_map_bio(root, rw, bio);
+	return btrfs_map_bio(root, rw, bio, mirror_num);
 }
 
 int btrfs_readpage_io_hook(struct page *page, u64 start, u64 end)

fs/btrfs/volumes.c

@@ -788,9 +788,31 @@ void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
 	}
 }
 
+int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
+{
+	struct extent_map *em;
+	struct map_lookup *map;
+	struct extent_map_tree *em_tree = &map_tree->map_tree;
+	int ret;
+
+	spin_lock(&em_tree->lock);
+	em = lookup_extent_mapping(em_tree, logical, len);
+	BUG_ON(!em);
+
+	BUG_ON(em->start > logical || em->start + em->len < logical);
+	map = (struct map_lookup *)em->bdev;
+	if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
+		ret = map->num_stripes;
+	else
+		ret = 1;
+	free_extent_map(em);
+	spin_unlock(&em_tree->lock);
+	return ret;
+}
+
 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
 		    u64 logical, u64 *length,
-		    struct btrfs_multi_bio **multi_ret)
+		    struct btrfs_multi_bio **multi_ret, int mirror_num)
 {
 	struct extent_map *em;
 	struct map_lookup *map;
@@ -822,6 +844,9 @@ again:
 	map = (struct map_lookup *)em->bdev;
 	offset = logical - em->start;
 
+	if (mirror_num > map->num_stripes)
+		mirror_num = 0;
+
 	/* if our multi bio struct is too small, back off and try again */
 	if (multi_ret && (rw & (1 << BIO_RW)) &&
 	    stripes_allocated < map->num_stripes &&
@@ -862,7 +887,9 @@ again:
 	if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
 		if (rw & (1 << BIO_RW))
 			multi->num_stripes = map->num_stripes;
-		else {
+		else if (mirror_num) {
+			stripe_index = mirror_num - 1;
+		} else {
 			int i;
 			u64 least = (u64)-1;
 			struct btrfs_device *cur;
@@ -880,6 +907,8 @@ again:
 	} else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
 		if (rw & (1 << BIO_RW))
 			multi->num_stripes = map->num_stripes;
+		else if (mirror_num)
+			stripe_index = mirror_num - 1;
 	} else {
 		/*
 		 * after this do_div call, stripe_nr is the number of stripes
@@ -938,7 +967,8 @@ static int end_bio_multi_stripe(struct bio *bio,
 #endif
 }
 
-int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio)
+int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
+		  int mirror_num)
 {
 	struct btrfs_mapping_tree *map_tree;
 	struct btrfs_device *dev;
@@ -960,7 +990,8 @@ int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio)
 	map_tree = &root->fs_info->mapping_tree;
 	map_length = length;
 
-	ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi);
+	ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
+			      mirror_num);
 	BUG_ON(ret);
 
 	total_devs = multi->num_stripes;

fs/btrfs/volumes.h

@@ -93,7 +93,7 @@ int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
 			   u64 owner, u64 num_bytes, u64 *start);
 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
 		    u64 logical, u64 *length,
-		    struct btrfs_multi_bio **multi_ret);
+		    struct btrfs_multi_bio **multi_ret, int mirror_num);
 int btrfs_read_sys_array(struct btrfs_root *root);
 int btrfs_read_chunk_tree(struct btrfs_root *root);
 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
@@ -101,7 +101,8 @@ int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
 		      u64 *num_bytes, u64 type);
 void btrfs_mapping_init(struct btrfs_mapping_tree *tree);
 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree);
-int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio);
+int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
+		  int mirror_num);
 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf);
 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 		       int flags, void *holder);
@@ -112,4 +113,5 @@ int btrfs_add_device(struct btrfs_trans_handle *trans,
 		     struct btrfs_root *root,
 		     struct btrfs_device *device);
 int btrfs_cleanup_fs_uuids(void);
+int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len);
 #endif