For indirect block mapping if the i_block > max supported block in inode
then ext4_ind_map_blocks may return a -EIO error. But in case of fiemap
this could be a valid query to ext4_map_blocks.
So in case if !create then return 0. This also makes ext4_warning to
ext4_debug in ext4_block_to_path() for the same reason.
Signed-off-by: Ritesh Harjani <[email protected]>
---
fs/ext4/indirect.c | 11 +++++++++--
1 file changed, 9 insertions(+), 2 deletions(-)
diff --git a/fs/ext4/indirect.c b/fs/ext4/indirect.c
index 3a4ab70fe9e0..e1ab495dd900 100644
--- a/fs/ext4/indirect.c
+++ b/fs/ext4/indirect.c
@@ -102,7 +102,11 @@ static int ext4_block_to_path(struct inode *inode,
offsets[n++] = i_block & (ptrs - 1);
final = ptrs;
} else {
- ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
+ /*
+ * It's not yet an error to just query beyond max
+ * block in inode. Fiemap callers may do so.
+ */
+ ext4_debug("block %lu > max in inode %lu",
i_block + direct_blocks +
indirect_blocks + double_blocks, inode->i_ino);
}
@@ -537,8 +541,11 @@ int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
depth = ext4_block_to_path(inode, map->m_lblk, offsets,
&blocks_to_boundary);
- if (depth == 0)
+ if (depth == 0) {
+ if (!(flags & EXT4_GET_BLOCKS_CREATE))
+ err = 0;
goto out;
+ }
partial = ext4_get_branch(inode, depth, offsets, chain, &err);
--
2.21.0
On Wed 26-02-20 15:27:06, Ritesh Harjani wrote:
> For indirect block mapping if the i_block > max supported block in inode
> then ext4_ind_map_blocks may return a -EIO error. But in case of fiemap
> this could be a valid query to ext4_map_blocks.
> So in case if !create then return 0. This also makes ext4_warning to
> ext4_debug in ext4_block_to_path() for the same reason.
>
> Signed-off-by: Ritesh Harjani <[email protected]>
Hmm, won't it be cleaner to just handle this in ext4_iomap_begin_report()?
We do trim map.m_len there anyway so it is only logical to trim it to
proper value supported by the inode on-disk format... BTW, note we have
sbi->s_bitmap_maxbytes value already computed in the superblock...
Honza
> ---
> fs/ext4/indirect.c | 11 +++++++++--
> 1 file changed, 9 insertions(+), 2 deletions(-)
>
> diff --git a/fs/ext4/indirect.c b/fs/ext4/indirect.c
> index 3a4ab70fe9e0..e1ab495dd900 100644
> --- a/fs/ext4/indirect.c
> +++ b/fs/ext4/indirect.c
> @@ -102,7 +102,11 @@ static int ext4_block_to_path(struct inode *inode,
> offsets[n++] = i_block & (ptrs - 1);
> final = ptrs;
> } else {
> - ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
> + /*
> + * It's not yet an error to just query beyond max
> + * block in inode. Fiemap callers may do so.
> + */
> + ext4_debug("block %lu > max in inode %lu",
> i_block + direct_blocks +
> indirect_blocks + double_blocks, inode->i_ino);
> }
> @@ -537,8 +541,11 @@ int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
> depth = ext4_block_to_path(inode, map->m_lblk, offsets,
> &blocks_to_boundary);
>
> - if (depth == 0)
> + if (depth == 0) {
> + if (!(flags & EXT4_GET_BLOCKS_CREATE))
> + err = 0;
> goto out;
> + }
>
> partial = ext4_get_branch(inode, depth, offsets, chain, &err);
>
> --
> 2.21.0
>
--
Jan Kara <[email protected]>
SUSE Labs, CR
On 2/26/20 6:09 PM, Jan Kara wrote:
> On Wed 26-02-20 15:27:06, Ritesh Harjani wrote:
>> For indirect block mapping if the i_block > max supported block in inode
>> then ext4_ind_map_blocks may return a -EIO error. But in case of fiemap
>> this could be a valid query to ext4_map_blocks.
>> So in case if !create then return 0. This also makes ext4_warning to
>> ext4_debug in ext4_block_to_path() for the same reason.
>>
>> Signed-off-by: Ritesh Harjani <[email protected]>
>
> Hmm, won't it be cleaner to just handle this in ext4_iomap_begin_report()?
> We do trim map.m_len there anyway so it is only logical to trim it to
> proper value supported by the inode on-disk format... BTW, note we have
> sbi->s_bitmap_maxbytes value already computed in the superblock...
hmm. Yes, thanks for the pointers. Let me check this again.
-ritesh
>
> Honza
>
>> ---
>> fs/ext4/indirect.c | 11 +++++++++--
>> 1 file changed, 9 insertions(+), 2 deletions(-)
>>
>> diff --git a/fs/ext4/indirect.c b/fs/ext4/indirect.c
>> index 3a4ab70fe9e0..e1ab495dd900 100644
>> --- a/fs/ext4/indirect.c
>> +++ b/fs/ext4/indirect.c
>> @@ -102,7 +102,11 @@ static int ext4_block_to_path(struct inode *inode,
>> offsets[n++] = i_block & (ptrs - 1);
>> final = ptrs;
>> } else {
>> - ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
>> + /*
>> + * It's not yet an error to just query beyond max
>> + * block in inode. Fiemap callers may do so.
>> + */
>> + ext4_debug("block %lu > max in inode %lu",
>> i_block + direct_blocks +
>> indirect_blocks + double_blocks, inode->i_ino);
>> }
>> @@ -537,8 +541,11 @@ int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
>> depth = ext4_block_to_path(inode, map->m_lblk, offsets,
>> &blocks_to_boundary);
>>
>> - if (depth == 0)
>> + if (depth == 0) {
>> + if (!(flags & EXT4_GET_BLOCKS_CREATE))
>> + err = 0;
>> goto out;
>> + }
>>
>> partial = ext4_get_branch(inode, depth, offsets, chain, &err);
>>
>> --
>> 2.21.0
>>
On Wed, Feb 26, 2020 at 03:27:06PM +0530, Ritesh Harjani wrote:
> For indirect block mapping if the i_block > max supported block in inode
> then ext4_ind_map_blocks may return a -EIO error. But in case of fiemap
> this could be a valid query to ext4_map_blocks.
> So in case if !create then return 0. This also makes ext4_warning to
> ext4_debug in ext4_block_to_path() for the same reason.
>
> Signed-off-by: Ritesh Harjani <[email protected]>
> ---
> fs/ext4/indirect.c | 11 +++++++++--
> 1 file changed, 9 insertions(+), 2 deletions(-)
>
> diff --git a/fs/ext4/indirect.c b/fs/ext4/indirect.c
> index 3a4ab70fe9e0..e1ab495dd900 100644
> --- a/fs/ext4/indirect.c
> +++ b/fs/ext4/indirect.c
> @@ -102,7 +102,11 @@ static int ext4_block_to_path(struct inode *inode,
> offsets[n++] = i_block & (ptrs - 1);
> final = ptrs;
> } else {
> - ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
> + /*
> + * It's not yet an error to just query beyond max
> + * block in inode. Fiemap callers may do so.
> + */
> + ext4_debug("block %lu > max in inode %lu",
> i_block + direct_blocks +
> indirect_blocks + double_blocks, inode->i_ino);
Does that mean fiemap callers can spamflood dmesg with this message just
by setting the query start range to a huge value?
--D
> }
> @@ -537,8 +541,11 @@ int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
> depth = ext4_block_to_path(inode, map->m_lblk, offsets,
> &blocks_to_boundary);
>
> - if (depth == 0)
> + if (depth == 0) {
> + if (!(flags & EXT4_GET_BLOCKS_CREATE))
> + err = 0;
> goto out;
> + }
>
> partial = ext4_get_branch(inode, depth, offsets, chain, &err);
>
> --
> 2.21.0
>
On 2/26/20 9:41 PM, Darrick J. Wong wrote:
> On Wed, Feb 26, 2020 at 03:27:06PM +0530, Ritesh Harjani wrote:
>> For indirect block mapping if the i_block > max supported block in inode
>> then ext4_ind_map_blocks may return a -EIO error. But in case of fiemap
>> this could be a valid query to ext4_map_blocks.
>> So in case if !create then return 0. This also makes ext4_warning to
>> ext4_debug in ext4_block_to_path() for the same reason.
>>
>> Signed-off-by: Ritesh Harjani <[email protected]>
>> ---
>> fs/ext4/indirect.c | 11 +++++++++--
>> 1 file changed, 9 insertions(+), 2 deletions(-)
>>
>> diff --git a/fs/ext4/indirect.c b/fs/ext4/indirect.c
>> index 3a4ab70fe9e0..e1ab495dd900 100644
>> --- a/fs/ext4/indirect.c
>> +++ b/fs/ext4/indirect.c
>> @@ -102,7 +102,11 @@ static int ext4_block_to_path(struct inode *inode,
>> offsets[n++] = i_block & (ptrs - 1);
>> final = ptrs;
>> } else {
>> - ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
>> + /*
>> + * It's not yet an error to just query beyond max
>> + * block in inode. Fiemap callers may do so.
>> + */
>> + ext4_debug("block %lu > max in inode %lu",
>> i_block + direct_blocks +
>> indirect_blocks + double_blocks, inode->i_ino);
>
> Does that mean fiemap callers can spamflood dmesg with this message just
> by setting the query start range to a huge value?
Not in the old implementation. But This could happen with indirect
block mapping with new implementation in iomap (as there is no check in
place before calling ext4_map_blocks()).
Previously __generic_block_fiemap() used to not query beyond
i_size_read(), so we were safe there.
So yes now as Jan also suggested, will add a check in place in
ext4_iomap_begin_report() itself, so that this flooding wont happen.
Thanks for the review!!
-ritesh
>
> --D
>
>> }
>> @@ -537,8 +541,11 @@ int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
>> depth = ext4_block_to_path(inode, map->m_lblk, offsets,
>> &blocks_to_boundary);
>>
>> - if (depth == 0)
>> + if (depth == 0) {
>> + if (!(flags & EXT4_GET_BLOCKS_CREATE))
>> + err = 0;
>> goto out;
>> + }
>>
>> partial = ext4_get_branch(inode, depth, offsets, chain, &err);
>>
>> --
>> 2.21.0
>>