Re: [PATCH 1/2] btrfs: tree-checker: Add checker for variable length item

[Nikolay Borisov <nborisov@suse.com>]

On  2.11.2017 14:48, Qu Wenruo wrote:
> 
> 
> On 2017年11月02日 20:12, Nikolay Borisov wrote:
>>
>>
>> On  1.11.2017 14:14, Qu Wenruo wrote:
>>> For the following types, we have items with variable length:
>>> (With BTRFS_ prefix and _KEY suffix snipped)
>>>
>>> DIR_ITEM
>>> DIR_INDEX
>>> XATTR_ITEM
>>> INODE_REF
>>> INODE_EXTREF
>>> ROOT_REF
>>> ROOT_BACKREF
>>>
>>> They all use @name_len to indicate their name length, and XATTR_ITEM has
>>> extra @data_len to indicate it data length.
>>>
>>> Despite their variable length, it's also possible to have several such
>>> structure inside one item.
>>>
>>> This patch will add checker to ensure:
>>>
>>> 1) No structure header and its data cross item boundary
>>> 2) Except XATTR_ITEM, no structure should have non-zero @data_len
>>>
>>> This checker is especially useful to avoid possible access beyond
>>> boundary for fuzzed image.
>>>
>>> Signed-off-by: Qu Wenruo <wqu@suse.com>
>>> ---
>>>  fs/btrfs/tree-checker.c | 123 ++++++++++++++++++++++++++++++++++++++++++++++++
>>>  1 file changed, 123 insertions(+)
>>>
>>> diff --git a/fs/btrfs/tree-checker.c b/fs/btrfs/tree-checker.c
>>> index 114fc5f0ecc5..f26e86fcbd74 100644
>>> --- a/fs/btrfs/tree-checker.c
>>> +++ b/fs/btrfs/tree-checker.c
>>> @@ -222,6 +222,120 @@ static int check_csum_item(struct btrfs_root *root, struct extent_buffer *leaf,
>>>  	return 0;
>>>  }
>>>  
>>> +static u32 get_header_size(u8 type)
>>> +{
>>> +	switch (type) {
>>> +	case BTRFS_DIR_ITEM_KEY:
>>> +	case BTRFS_DIR_INDEX_KEY:
>>> +	case BTRFS_XATTR_ITEM_KEY:
>>> +		return sizeof(struct btrfs_dir_item);
>>> +	case BTRFS_INODE_REF_KEY:
>>> +		return sizeof(struct btrfs_inode_ref);
>>> +	case BTRFS_INODE_EXTREF_KEY:
>>> +		return sizeof(struct btrfs_inode_extref);
>>> +	case BTRFS_ROOT_REF_KEY:
>>> +	case BTRFS_ROOT_BACKREF_KEY:
>>> +		return sizeof(struct btrfs_root_ref);
>>> +	}
>>> +	WARN_ON(1);
>>> +	return 0;
>>> +}
>>> +
>>> +static u16 get_header_namelen(struct extent_buffer *leaf, u8 type,
>>> +			      u32 header_offset)
>>> +{
>>> +	/*
>>> +	 * @header_offset is offset starts after leaf header, while the
>>> +	 * accessors expects offset starts from leaf header.
>>> +	 * Sowe need to adds LEAF_DATA_OFFSET here
>>> +	 */
>>> +	unsigned long leaf_offset = header_offset + BTRFS_LEAF_DATA_OFFSET;
>>> +
>>> +	switch (type) {
>>> +	case BTRFS_DIR_ITEM_KEY:
>>> +	case BTRFS_DIR_INDEX_KEY:
>>> +	case BTRFS_XATTR_ITEM_KEY:
>>> +		return btrfs_dir_name_len(leaf, (void *)leaf_offset);
>>> +	case BTRFS_INODE_REF_KEY:
>>> +		return btrfs_inode_ref_name_len(leaf, (void *)leaf_offset);
>>> +	case BTRFS_INODE_EXTREF_KEY:
>>> +		return btrfs_inode_extref_name_len(leaf, (void *)leaf_offset);
>>> +	case BTRFS_ROOT_REF_KEY:
>>> +	case BTRFS_ROOT_BACKREF_KEY:
>>> +		return btrfs_root_ref_name_len(leaf, (void *)leaf_offset);
>>> +	}
>>> +	WARN_ON(1);
>>> +	return 0;
>>> +}
>>> +
>>> +static u16 get_header_datalen(struct extent_buffer *leaf, u8 type,
>>> +			      unsigned long header_offset)
>>> +{
>>> +	/* Same as get_header_namelen */
>>> +	unsigned long leaf_offset = header_offset + BTRFS_LEAF_DATA_OFFSET;
>>> +
>>> +	switch (type) {
>>> +	case BTRFS_DIR_ITEM_KEY:
>>> +	case BTRFS_DIR_INDEX_KEY:
>>> +	case BTRFS_XATTR_ITEM_KEY:
>>> +		return btrfs_dir_data_len(leaf, (void *)leaf_offset);
>>> +	}
>>> +	return 0;
>>> +}
>>> +
>>> +/*
>>> + * For items with variable length, normally with namelen and tailing data.
>>> + * Like INODE_REF or XATTR
>>> + */
>>> +static int check_variable_length_item(struct btrfs_root *root,
>>> +				      struct extent_buffer *leaf,
>>> +				      struct btrfs_key *key, int slot)
>>> +{
>>
>> One more thing - you are only validating the boundaries of such variable
>> length items, so make this specific. I.e rename the function to
>> something like:
> 
> If you follow the naming schema in tree-checker, you'll find that we (at
> least myself) is naming these function always using "check_" prefix, so
> "validate_" prefix seems less consistent.
> 
> Further more, the naming has its level"
> leaf (all items boundary already checked)
> |- leaf_item (hub wrapper)
>    |- csum_item
>    |- extent_data_item
> 
> So here check_<something>_item follows the original level according to
> the naming schema.
> 
> Further more, although we're checking boundary, the truth is, we are
> check the name_len/data_len *members*, so the check_<something>_item
> still makes sence.
> 
> If really want to change the name, I prefer some naming can show that
> we're checking several items which share the same variable length property.
> 
> So at least, none of the alternative seems to fit the schema.

The reason why I wanted this function renamed is that we have this
function which performs only some checks and then we have the
verify_dir_item which performs different checks for the same item. So
why can't those function be turned into one ? I'm not too hung up on the
actual naming!

> 
> Thanks,
> Qu
> 
>>
>> validate_variable_boundaries
>> check_variable_length_item_boundary
>> check_item_boundaries
>>
>>> +	u8 type = key->type;
>>> +	u32 item_start = btrfs_item_offset_nr(leaf, slot);
>>> +	u32 item_end = btrfs_item_end_nr(leaf, slot);
>>> +	u32 header_size = get_header_size(type);
>>> +	u32 total_size;
>>> +	u32 cur = item_start;
>>> +
>>> +	while (cur < item_end) {
>>> +		u32 namelen;
>>> +		u32 datalen;
>>> +
>>> +		/* header itself should not cross item boundary */
>>> +		if (cur + header_size > item_end) {
>>> +			generic_err(root, leaf, slot,
>>> +				"structure header crosses item boundary, have %u expect (%u, %u]",
>>> +				cur + header_size, cur, item_end);
>>> +			return -EUCLEAN;
>>> +		}
>>> +
>>> +		namelen = get_header_namelen(leaf, type, cur);
>>> +		datalen = get_header_datalen(leaf, type, cur);
>>> +
>>> +		/* Only XATTR can own data */
>>> +		if (type != BTRFS_XATTR_ITEM_KEY && datalen) {
>>> +			generic_err(root, leaf, slot,
>>> +				"item has invalid data len, have %u expect 0",
>>> +				datalen);
>>> +			return -EUCLEAN;
>>> +		}
>>> +
>>> +		total_size = header_size + namelen + datalen;
>>> +
>>> +		/* header and name/data should not cross item boundary */
>>> +		if (cur + total_size > item_end) {
>>> +			generic_err(root, leaf, slot,
>>> +				"structure data crosses item boundary, have %u expect (%u, %u]",
>>> +				cur + total_size, cur + header_size, item_end);
>>> +			return -EUCLEAN;
>>> +		}
>>> +
>>> +		cur += total_size;
>>> +	}
>>> +	return 0;
>>> +}
>>> +
>>>  /*
>>>   * Common point to switch the item-specific validation.
>>>   */
>>> @@ -238,6 +352,15 @@ static int check_leaf_item(struct btrfs_root *root,
>>>  	case BTRFS_EXTENT_CSUM_KEY:
>>>  		ret = check_csum_item(root, leaf, key, slot);
>>>  		break;
>>> +	case BTRFS_DIR_ITEM_KEY:
>>> +	case BTRFS_XATTR_ITEM_KEY:
>>> +	case BTRFS_DIR_INDEX_KEY:
>>> +	case BTRFS_INODE_REF_KEY:
>>> +	case BTRFS_INODE_EXTREF_KEY:
>>> +	case BTRFS_ROOT_REF_KEY:
>>> +	case BTRFS_ROOT_BACKREF_KEY:
>>> +		ret = check_variable_length_item(root, leaf, key, slot);
>>> +		break;
>>>  	}
>>>  	return ret;
>>>  }
>>>
>> --
>> To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>
>