Btrfs: verify csums on read

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

fs/btrfs/ctree.c

@@ -109,14 +109,14 @@ static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
 		return 1;
 	if (k1.objectid < k2->objectid)
 		return -1;
-	if (k1.flags > k2->flags)
-		return 1;
-	if (k1.flags < k2->flags)
-		return -1;
 	if (k1.offset > k2->offset)
 		return 1;
 	if (k1.offset < k2->offset)
 		return -1;
+	if (k1.flags > k2->flags)
+		return 1;
+	if (k1.flags < k2->flags)
+		return -1;
 	return 0;
 }
 
@@ -1165,7 +1165,6 @@ int btrfs_insert_empty_item(struct btrfs_trans_handle *trans, struct btrfs_root
 		BUG();
 	ret = btrfs_search_slot(trans, root, cpu_key, path, data_size, 1);
 	if (ret == 0) {
-		btrfs_release_path(root, path);
 		return -EEXIST;
 	}
 	if (ret < 0)

fs/btrfs/ctree.h

@@ -21,6 +21,9 @@ struct btrfs_transaction;
  */
 #define BTRFS_NAME_LEN 255
 
+/* 32 bytes in various csum fields */
+#define BTRFS_CSUM_SIZE 32
+
 /*
  * the key defines the order in the tree, and so it also defines (optimal)
  * block layout.  objectid corresonds to the inode number.  The flags
@@ -37,21 +40,21 @@ struct btrfs_transaction;
  */
 struct btrfs_disk_key {
 	__le64 objectid;
-	__le32 flags;
 	__le64 offset;
+	__le32 flags;
 } __attribute__ ((__packed__));
 
 struct btrfs_key {
 	u64 objectid;
-	u32 flags;
 	u64 offset;
+	u32 flags;
 } __attribute__ ((__packed__));
 
 /*
  * every tree block (leaf or node) starts with this header.
  */
 struct btrfs_header {
-	__le32 csum[8];
+	u8 csum[BTRFS_CSUM_SIZE];
 	u8 fsid[16]; /* FS specific uuid */
 	__le64 blocknr; /* which block this node is supposed to live in */
 	__le64 generation;
@@ -75,7 +78,7 @@ struct buffer_head;
  * it currently lacks any block count etc etc
  */
 struct btrfs_super_block {
-	__le32 csum[8];
+	u8 csum[BTRFS_CSUM_SIZE];
 	/* the first 3 fields must match struct btrfs_header */
 	u8 fsid[16];    /* FS specific uuid */
 	__le64 blocknr; /* this block number */
@@ -147,7 +150,7 @@ struct btrfs_extent_item {
 } __attribute__ ((__packed__));
 
 struct btrfs_inode_timespec {
-	__le32 sec;
+	__le64 sec;
 	__le32 nsec;
 } __attribute__ ((__packed__));
 
@@ -214,6 +217,10 @@ struct btrfs_file_extent_item {
 	__le64 num_blocks;
 } __attribute__ ((__packed__));
 
+struct btrfs_csum_item {
+	u8 csum[BTRFS_CSUM_SIZE];
+} __attribute__ ((__packed__));
+
 struct btrfs_inode_map_item {
 	struct btrfs_disk_key key;
 } __attribute__ ((__packed__));
@@ -283,27 +290,32 @@ struct btrfs_root {
  * a (hopefully) huge chunk of disk
  */
 #define BTRFS_EXTENT_DATA_KEY	4
+/*
+ * csum items have the checksums for data in the extents
+ */
+#define BTRFS_CSUM_ITEM_KEY	5
+
 /*
  * root items point to tree roots.  There are typically in the root
  * tree used by the super block to find all the other trees
  */
-#define BTRFS_ROOT_ITEM_KEY	5
+#define BTRFS_ROOT_ITEM_KEY	6
 /*
  * extent items are in the extent map tree.  These record which blocks
  * are used, and how many references there are to each block
  */
-#define BTRFS_EXTENT_ITEM_KEY	6
+#define BTRFS_EXTENT_ITEM_KEY	7
 
 /*
  * the inode map records which inode numbers are in use and where
  * they actually live on disk
  */
-#define BTRFS_INODE_MAP_ITEM_KEY 7
+#define BTRFS_INODE_MAP_ITEM_KEY 8
 /*
  * string items are for debugging.  They just store a short string of
  * data in the FS
  */
-#define BTRFS_STRING_ITEM_KEY	8
+#define BTRFS_STRING_ITEM_KEY	9
 
 static inline u64 btrfs_inode_generation(struct btrfs_inode_item *i)
 {
@@ -407,15 +419,15 @@ static inline void btrfs_set_inode_compat_flags(struct btrfs_inode_item *i,
 	i->compat_flags = cpu_to_le16(val);
 }
 
-static inline u32 btrfs_timespec_sec(struct btrfs_inode_timespec *ts)
+static inline u64 btrfs_timespec_sec(struct btrfs_inode_timespec *ts)
 {
-	return le32_to_cpu(ts->sec);
+	return le64_to_cpu(ts->sec);
 }
 
 static inline void btrfs_set_timespec_sec(struct btrfs_inode_timespec *ts,
-					  u32 val)
+					  u64 val)
 {
-	ts->sec = cpu_to_le32(val);
+	ts->sec = cpu_to_le64(val);
 }
 
 static inline u32 btrfs_timespec_nsec(struct btrfs_inode_timespec *ts)
@@ -429,8 +441,6 @@ static inline void btrfs_set_timespec_nsec(struct btrfs_inode_timespec *ts,
 	ts->nsec = cpu_to_le32(val);
 }
 
-
-
 static inline u64 btrfs_extent_owner(struct btrfs_extent_item *ei)
 {
 	return le64_to_cpu(ei->owner);
@@ -905,4 +915,11 @@ int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
 			     struct btrfs_root *root,
 			     struct btrfs_path *path, u64 objectid,
 			     u64 blocknr, int mod);
+int btrfs_csum_file_block(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root,
+			  u64 objectid, u64 offset,
+			  char *data, size_t len);
+int btrfs_csum_verify_file_block(struct btrfs_root *root,
+				 u64 objectid, u64 offset,
+				 char *data, size_t len);
 #endif

fs/btrfs/disk-io.c

@@ -127,31 +127,46 @@ static int btree_get_block(struct inode *inode, sector_t iblock,
 	return 0;
 }
 
-static int csum_tree_block(struct btrfs_root * root, struct buffer_head *bh,
-			    int verify)
+int btrfs_csum_data(struct btrfs_root * root, char *data, size_t len,
+		    char *result)
 {
-	struct btrfs_node *node = btrfs_buffer_node(bh);
 	struct scatterlist sg;
 	struct crypto_hash *tfm = root->fs_info->hash_tfm;
 	struct hash_desc desc;
 	int ret;
-	char result[32];
 
 	desc.tfm = tfm;
 	desc.flags = 0;
-	sg_init_one(&sg, bh->b_data + 32, bh->b_size - 32);
+	sg_init_one(&sg, data, len);
 	spin_lock(&root->fs_info->hash_lock);
-	ret = crypto_hash_digest(&desc, &sg, bh->b_size - 32, result);
+	ret = crypto_hash_digest(&desc, &sg, len, result);
 	spin_unlock(&root->fs_info->hash_lock);
 	if (ret) {
 		printk("sha256 digest failed\n");
 	}
+	return ret;
+}
+static int csum_tree_block(struct btrfs_root *root, struct buffer_head *bh,
+			   int verify)
+{
+	char result[BTRFS_CSUM_SIZE];
+	int ret;
+	struct btrfs_node *node;
+
+	ret = btrfs_csum_data(root, bh->b_data + BTRFS_CSUM_SIZE,
+			      bh->b_size - BTRFS_CSUM_SIZE, result);
+	if (ret)
+		return ret;
 	if (verify) {
-		if (memcmp(node->header.csum, result, sizeof(result)))
-			printk("csum verify failed on %Lu\n", bh->b_blocknr);
-		return -EINVAL;
-	} else
-		memcpy(node->header.csum, result, sizeof(node->header.csum));
+		if (memcmp(bh->b_data, result, BTRFS_CSUM_SIZE)) {
+			printk("checksum verify failed on %lu\n",
+			       bh->b_blocknr);
+			return 1;
+		}
+	} else {
+		node = btrfs_buffer_node(bh);
+		memcpy(&node->header.csum, result, BTRFS_CSUM_SIZE);
+	}
 	return 0;
 }
 

fs/btrfs/disk-io.h

@@ -39,4 +39,6 @@ void btrfs_block_release(struct btrfs_root *root, struct buffer_head *buf);
 int write_ctree_super(struct btrfs_trans_handle *trans,
 		      struct btrfs_root *root);
 struct buffer_head *btrfs_find_tree_block(struct btrfs_root *root, u64 blocknr);
+int btrfs_csum_data(struct btrfs_root * root, char *data, size_t len,
+		    char *result);
 #endif

fs/btrfs/file-item.c

@@ -57,3 +57,63 @@ int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
 	ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
 	return ret;
 }
+
+int btrfs_csum_file_block(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root,
+			  u64 objectid, u64 offset,
+			  char *data, size_t len)
+{
+	int ret;
+	struct btrfs_key file_key;
+	struct btrfs_path path;
+	struct btrfs_csum_item *item;
+
+	btrfs_init_path(&path);
+	file_key.objectid = objectid;
+	file_key.offset = offset;
+	file_key.flags = 0;
+	btrfs_set_key_type(&file_key, BTRFS_CSUM_ITEM_KEY);
+	ret = btrfs_insert_empty_item(trans, root, &path, &file_key,
+				      BTRFS_CSUM_SIZE);
+	if (ret != 0 && ret != -EEXIST)
+		goto fail;
+	item = btrfs_item_ptr(btrfs_buffer_leaf(path.nodes[0]), path.slots[0],
+			      struct btrfs_csum_item);
+	ret = 0;
+	ret = btrfs_csum_data(root, data, len, item->csum);
+	mark_buffer_dirty(path.nodes[0]);
+fail:
+	btrfs_release_path(root, &path);
+	return ret;
+}
+
+int btrfs_csum_verify_file_block(struct btrfs_root *root,
+				 u64 objectid, u64 offset,
+				 char *data, size_t len)
+{
+	int ret;
+	struct btrfs_key file_key;
+	struct btrfs_path path;
+	struct btrfs_csum_item *item;
+	char result[BTRFS_CSUM_SIZE];
+
+	btrfs_init_path(&path);
+	file_key.objectid = objectid;
+	file_key.offset = offset;
+	file_key.flags = 0;
+	btrfs_set_key_type(&file_key, BTRFS_CSUM_ITEM_KEY);
+	ret = btrfs_search_slot(NULL, root, &file_key, &path, 0, 0);
+	if (ret)
+		goto fail;
+	item = btrfs_item_ptr(btrfs_buffer_leaf(path.nodes[0]), path.slots[0],
+			      struct btrfs_csum_item);
+	ret = 0;
+	ret = btrfs_csum_data(root, data, len, result);
+	WARN_ON(ret);
+	if (memcmp(result, item->csum, BTRFS_CSUM_SIZE))
+		ret = 1;
+fail:
+	btrfs_release_path(root, &path);
+	return ret;
+}
+

fs/btrfs/super.c

@@ -249,15 +249,16 @@ static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
 	struct btrfs_key key;
 	struct btrfs_disk_key *found_key;
 	struct btrfs_leaf *leaf;
-	struct btrfs_file_extent_item *fi;
-	u64 extent_start;
-	u64 extent_num_blocks;
+	struct btrfs_file_extent_item *fi = NULL;
+	u64 extent_start = 0;
+	u64 extent_num_blocks = 0;
+	int found_extent;
 
 	/* FIXME, add redo link to tree so we don't leak on crash */
 	key.objectid = inode->i_ino;
 	key.offset = (u64)-1;
 	key.flags = 0;
-	btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
+	btrfs_set_key_type(&key, BTRFS_CSUM_ITEM_KEY);
 	while(1) {
 		btrfs_init_path(&path);
 		ret = btrfs_search_slot(trans, root, &key, &path, -1, 1);
@@ -273,25 +274,32 @@ static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
 		found_key = &leaf->items[path.slots[0]].key;
 		if (btrfs_disk_key_objectid(found_key) != inode->i_ino)
 			break;
-		if (btrfs_disk_key_type(found_key) != BTRFS_EXTENT_DATA_KEY)
+		if (btrfs_disk_key_type(found_key) != BTRFS_CSUM_ITEM_KEY &&
+		    btrfs_disk_key_type(found_key) != BTRFS_EXTENT_DATA_KEY)
 			break;
 		if (btrfs_disk_key_offset(found_key) < inode->i_size)
 			break;
-		fi = btrfs_item_ptr(btrfs_buffer_leaf(path.nodes[0]),
-				    path.slots[0],
-				    struct btrfs_file_extent_item);
-		extent_start = btrfs_file_extent_disk_blocknr(fi);
-		extent_num_blocks = btrfs_file_extent_disk_num_blocks(fi);
-		key.offset = btrfs_disk_key_offset(found_key) - 1;
+		if (btrfs_disk_key_type(found_key) == BTRFS_EXTENT_DATA_KEY) {
+			fi = btrfs_item_ptr(btrfs_buffer_leaf(path.nodes[0]),
+					    path.slots[0],
+					    struct btrfs_file_extent_item);
+			extent_start = btrfs_file_extent_disk_blocknr(fi);
+			extent_num_blocks =
+				btrfs_file_extent_disk_num_blocks(fi);
+			inode->i_blocks -=
+				btrfs_file_extent_num_blocks(fi) >> 9;
+			found_extent = 1;
+		} else {
+			found_extent = 0;
+		}
 		ret = btrfs_del_item(trans, root, &path);
 		BUG_ON(ret);
-		inode->i_blocks -= btrfs_file_extent_num_blocks(fi) >> 9;
 		btrfs_release_path(root, &path);
-		ret = btrfs_free_extent(trans, root, extent_start,
-					extent_num_blocks, 0);
-		BUG_ON(ret);
-		if (key.offset + 1 == 0)
-			break;
+		if (found_extent) {
+			ret = btrfs_free_extent(trans, root, extent_start,
+						extent_num_blocks, 0);
+			BUG_ON(ret);
+		}
 	}
 	btrfs_release_path(root, &path);
 	ret = 0;
@@ -975,10 +983,24 @@ static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
 	int err = 0;
 	int ret;
 	int this_write;
+	struct inode *inode = file->f_path.dentry->d_inode;
 
 	for (i = 0; i < num_pages; i++) {
 		offset = pos & (PAGE_CACHE_SIZE -1);
 		this_write = min(PAGE_CACHE_SIZE - offset, write_bytes);
+		/* FIXME, one block at a time */
+
+		mutex_lock(&root->fs_info->fs_mutex);
+		trans = btrfs_start_transaction(root, 1);
+		btrfs_csum_file_block(trans, root, inode->i_ino,
+				      pages[i]->index << PAGE_CACHE_SHIFT,
+				      kmap(pages[i]), PAGE_CACHE_SIZE);
+		kunmap(pages[i]);
+		SetPageChecked(pages[i]);
+		ret = btrfs_end_transaction(trans, root);
+		BUG_ON(ret);
+		mutex_unlock(&root->fs_info->fs_mutex);
+
 		ret = nobh_commit_write(file, pages[i], offset,
 					 offset + this_write);
 		pos += this_write;
@@ -1022,7 +1044,7 @@ static int prepare_pages(struct btrfs_trans_handle *trans,
 		this_write = min(PAGE_CACHE_SIZE - offset, write_bytes);
 		ret = nobh_prepare_write(pages[i], offset,
 					 offset + this_write,
-					 btrfs_get_block_lock);
+					 btrfs_get_block);
 		pos += this_write;
 		if (ret) {
 			err = ret;
@@ -1051,7 +1073,6 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 	size_t num_written = 0;
 	int err = 0;
 	int ret = 0;
-	struct btrfs_trans_handle *trans;
 	struct inode *inode = file->f_path.dentry->d_inode;
 	struct btrfs_root *root = btrfs_sb(inode->i_sb);
 	struct page *pages[1];
@@ -1077,25 +1098,18 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 		size_t write_bytes = min(count, PAGE_CACHE_SIZE - offset);
 		size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 					PAGE_CACHE_SHIFT;
-		mutex_lock(&root->fs_info->fs_mutex);
-		trans = btrfs_start_transaction(root, 1);
-
-		ret = prepare_pages(trans, root, file, pages, num_pages,
+		ret = prepare_pages(NULL, root, file, pages, num_pages,
 				    pos, write_bytes);
 		BUG_ON(ret);
 		ret = btrfs_copy_from_user(pos, num_pages,
 					   write_bytes, pages, buf);
 		BUG_ON(ret);
 
-		mutex_unlock(&root->fs_info->fs_mutex);
-
-		ret = dirty_and_release_pages(trans, root, file, pages,
+		ret = dirty_and_release_pages(NULL, root, file, pages,
 					      num_pages, pos, write_bytes);
 		BUG_ON(ret);
 		btrfs_drop_pages(pages, num_pages);
 
-		ret = btrfs_end_transaction(trans, root);
-
 		buf += write_bytes;
 		count -= write_bytes;
 		pos += write_bytes;
@@ -1111,6 +1125,118 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 	return num_written ? num_written : err;
 }
 
+static int btrfs_read_actor(read_descriptor_t *desc, struct page *page,
+			unsigned long offset, unsigned long size)
+{
+	char *kaddr;
+	unsigned long left, count = desc->count;
+
+	if (size > count)
+		size = count;
+
+	if (!PageChecked(page)) {
+		/* FIXME, do it per block */
+		struct btrfs_root *root = btrfs_sb(page->mapping->host->i_sb);
+		int ret = btrfs_csum_verify_file_block(root,
+					  page->mapping->host->i_ino,
+					  page->index << PAGE_CACHE_SHIFT,
+					  kmap(page), PAGE_CACHE_SIZE);
+		if (ret) {
+			printk("failed to verify ino %lu page %lu\n",
+			       page->mapping->host->i_ino,
+			       page->index);
+			memset(page_address(page), 0, PAGE_CACHE_SIZE);
+		}
+		SetPageChecked(page);
+		kunmap(page);
+	}
+	/*
+	 * Faults on the destination of a read are common, so do it before
+	 * taking the kmap.
+	 */
+	if (!fault_in_pages_writeable(desc->arg.buf, size)) {
+		kaddr = kmap_atomic(page, KM_USER0);
+		left = __copy_to_user_inatomic(desc->arg.buf,
+						kaddr + offset, size);
+		kunmap_atomic(kaddr, KM_USER0);
+		if (left == 0)
+			goto success;
+	}
+
+	/* Do it the slow way */
+	kaddr = kmap(page);
+	left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
+	kunmap(page);
+
+	if (left) {
+		size -= left;
+		desc->error = -EFAULT;
+	}
+success:
+	desc->count = count - size;
+	desc->written += size;
+	desc->arg.buf += size;
+	return size;
+}
+
+/**
+ * btrfs_file_aio_read - filesystem read routine
+ * @iocb:	kernel I/O control block
+ * @iov:	io vector request
+ * @nr_segs:	number of segments in the iovec
+ * @pos:	current file position
+ */
+static ssize_t btrfs_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
+				   unsigned long nr_segs, loff_t pos)
+{
+	struct file *filp = iocb->ki_filp;
+	ssize_t retval;
+	unsigned long seg;
+	size_t count;
+	loff_t *ppos = &iocb->ki_pos;
+
+	count = 0;
+	for (seg = 0; seg < nr_segs; seg++) {
+		const struct iovec *iv = &iov[seg];
+
+		/*
+		 * If any segment has a negative length, or the cumulative
+		 * length ever wraps negative then return -EINVAL.
+		 */
+		count += iv->iov_len;
+		if (unlikely((ssize_t)(count|iv->iov_len) < 0))
+			return -EINVAL;
+		if (access_ok(VERIFY_WRITE, iv->iov_base, iv->iov_len))
+			continue;
+		if (seg == 0)
+			return -EFAULT;
+		nr_segs = seg;
+		count -= iv->iov_len;	/* This segment is no good */
+		break;
+	}
+	retval = 0;
+	if (count) {
+		for (seg = 0; seg < nr_segs; seg++) {
+			read_descriptor_t desc;
+
+			desc.written = 0;
+			desc.arg.buf = iov[seg].iov_base;
+			desc.count = iov[seg].iov_len;
+			if (desc.count == 0)
+				continue;
+			desc.error = 0;
+			do_generic_file_read(filp, ppos, &desc,
+					     btrfs_read_actor);
+			retval += desc.written;
+			if (desc.error) {
+				retval = retval ?: desc.error;
+				break;
+			}
+		}
+	}
+	return retval;
+}
+
 static int btrfs_get_sb(struct file_system_type *fs_type,
 	int flags, const char *dev_name, void *data, struct vfsmount *mnt)
 {
@@ -1166,7 +1292,7 @@ static struct inode_operations btrfs_file_inode_operations = {
 static struct file_operations btrfs_file_operations = {
 	.llseek		= generic_file_llseek,
 	.read		= do_sync_read,
-	.aio_read       = generic_file_aio_read,
+	.aio_read       = btrfs_file_aio_read,
 	.write		= btrfs_file_write,
 	.mmap		= generic_file_mmap,
 	.open		= generic_file_open,