Message-ID: <20250515165348.GO25655@frogsfrogsfrogs>
Date: Thu, 15 May 2025 09:53:48 -0700
From: "Darrick J. Wong" <djwong@...nel.org>
To: "Ritesh Harjani (IBM)" <ritesh.list@...il.com>
Cc: linux-ext4@...r.kernel.org, Theodore Ts'o <tytso@....edu>,
	Jan Kara <jack@...e.cz>, John Garry <john.g.garry@...cle.com>,
	Ojaswin Mujoo <ojaswin@...ux.ibm.com>,
	linux-fsdevel@...r.kernel.org
Subject: Re: [PATCH v4 5/7] ext4: Add multi-fsblock atomic write support with
 bigalloc

On Thu, May 15, 2025 at 08:15:37PM +0530, Ritesh Harjani (IBM) wrote:
> EXT4 supports bigalloc feature which allows the FS to work in size of
> clusters (group of blocks) rather than individual blocks. This patch
> adds atomic write support for bigalloc so that systems with bs = ps can
> also create FS using -
>     mkfs.ext4 -F -O bigalloc -b 4096 -C 16384 <dev>
> 
> With bigalloc ext4 can support multi-fsblock atomic writes. We will have to
> adjust ext4's atomic write unit max value to cluster size. This can then support
> atomic write of size anywhere between [blocksize, clustersize]. This
> patch adds the required changes to enable multi-fsblock atomic write
> support using bigalloc in the next patch.
> 
> In this patch for block allocation:
> we first query the underlying region of the requested range by calling
> ext4_map_blocks() call. Here are the various cases which we then handle
> depending upon the underlying mapping type:
> 1. If the underlying region for the entire requested range is a mapped extent,
>    then we don't call ext4_map_blocks() to allocate anything. We don't need to
>    even start the jbd2 txn in this case.
> 2. For an append write case, we create a mapped extent.
> 3. If the underlying region is entirely a hole, then we create an unwritten
>    extent for the requested range.
> 4. If the underlying region is a large unwritten extent, then we split the
>    extent into 2 unwritten extent of required size.
> 5. If the underlying region has any type of mixed mapping, then we call
>    ext4_map_blocks() in a loop to zero out the unwritten and the hole regions
>    within the requested range. This then provide a single mapped extent type
>    mapping for the requested range.
> 
> Note: We invoke ext4_map_blocks() in a loop with the EXT4_GET_BLOCKS_ZERO
> flag only when the underlying extent mapping of the requested range is
> not entirely a hole, an unwritten extent, or a fully mapped extent. That
> is, if the underlying region contains a mix of hole(s), unwritten
> extent(s), and mapped extent(s), we use this loop to ensure that all the
> short mappings are zeroed out. This guarantees that the entire requested
> range becomes a single, uniformly mapped extent. It is ok to do so
> because we know this is being done on a bigalloc enabled filesystem
> where the block bitmap represents the entire cluster unit.
> 
> Note having a single contiguous underlying region of type mapped,
> unwrittn or hole is not a problem. But the reason to avoid writing on
> top of mixed mapping region is because, atomic writes requires all or
> nothing should get written for the userspace pwritev2 request. So if at
> any point in time during the write if a crash or a sudden poweroff
> occurs, the region undergoing atomic write should read either complete
> old data or complete new data. But it should never have a mix of both
> old and new data.
> So, we first convert any mixed mapping region to a single contiguous
> mapped extent before any data gets written to it. This is because
> normally FS will only convert unwritten extents to written at the end of
> the write in ->end_io() call. And if we allow the writes over a mixed
> mapping and if a sudden power off happens in between, we will end up
> reading mix of new data (over mapped extents) and old data (over
> unwritten extents), because unwritten to written conversion never went
> through.
> So to avoid this and to avoid writes getting torned due to mixed
> mapping, we first allocate a single contiguous block mapping and then
> do the write.
> 
> Co-developed-by: Ojaswin Mujoo <ojaswin@...ux.ibm.com>
> Signed-off-by: Ojaswin Mujoo <ojaswin@...ux.ibm.com>
> Signed-off-by: Ritesh Harjani (IBM) <ritesh.list@...il.com>
> ---
>  fs/ext4/ext4.h    |   2 +
>  fs/ext4/extents.c |  87 ++++++++++++++++++++
>  fs/ext4/file.c    |   7 +-
>  fs/ext4/inode.c   | 200 +++++++++++++++++++++++++++++++++++++++++++++-
>  4 files changed, 291 insertions(+), 5 deletions(-)
> 
> diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
> index ef6cac6b4b4c..8eb1f332ee7d 100644
> --- a/fs/ext4/ext4.h
> +++ b/fs/ext4/ext4.h
> @@ -3728,6 +3728,8 @@ extern long ext4_fallocate(struct file *file, int mode, loff_t offset,
>  			  loff_t len);
>  extern int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
>  					  loff_t offset, ssize_t len);
> +extern int ext4_convert_unwritten_extents_atomic(handle_t *handle,
> +			struct inode *inode, loff_t offset, ssize_t len);
>  extern int ext4_convert_unwritten_io_end_vec(handle_t *handle,
>  					     ext4_io_end_t *io_end);
>  extern int ext4_map_blocks(handle_t *handle, struct inode *inode,
> diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
> index fa850f188d46..2967c74dabaf 100644
> --- a/fs/ext4/extents.c
> +++ b/fs/ext4/extents.c
> @@ -4792,6 +4792,93 @@ long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
>  	return ret;
>  }
>  
> +/*
> + * This function converts a range of blocks to written extents. The caller of
> + * this function will pass the start offset and the size. all unwritten extents
> + * within this range will be converted to written extents.
> + *
> + * This function is called from the direct IO end io call back function for
> + * atomic writes, to convert the unwritten extents after IO is completed.
> + *
> + * Note that the requirement for atomic writes is that all conversion should
> + * happen atomically in a single fs journal transaction. We mainly only allocate
> + * unwritten extents either on a hole on a pre-exiting unwritten extent range in
> + * ext4_map_blocks_atomic_write(). The only case where we can have multiple
> + * unwritten extents in a range [offset, offset+len) is when there is a split
> + * unwritten extent between two leaf nodes which was cached in extent status
> + * cache during ext4_iomap_alloc() time. That will allow
> + * ext4_map_blocks_atomic_write() to return the unwritten extent range w/o going
> + * into the slow path. That means we might need a loop for conversion of this
> + * unwritten extent split across leaf block within a single journal transaction.
> + * Split extents across leaf nodes is a rare case, but let's still handle that
> + * to meet the requirements of multi-fsblock atomic writes.
> + *
> + * Returns 0 on success.
> + */
> +int ext4_convert_unwritten_extents_atomic(handle_t *handle, struct inode *inode,
> +					  loff_t offset, ssize_t len)
> +{
> +	unsigned int max_blocks;
> +	int ret = 0, ret2 = 0, ret3 = 0;
> +	struct ext4_map_blocks map;
> +	unsigned int blkbits = inode->i_blkbits;
> +	unsigned int credits = 0;
> +	int flags = EXT4_GET_BLOCKS_IO_CONVERT_EXT;
> +
> +	map.m_lblk = offset >> blkbits;
> +	max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
> +
> +	if (!handle) {
> +		/*
> +		 * TODO: An optimization can be added later by having an extent
> +		 * status flag e.g. EXTENT_STATUS_SPLIT_LEAF. If we query that
> +		 * it can tell if the extent in the cache is a split extent.
> +		 * But for now let's assume pextents as 2 always.
> +		 */
> +		credits = ext4_meta_trans_blocks(inode, max_blocks, 2);
> +	}
> +
> +	if (credits) {
> +		handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, credits);
> +		if (IS_ERR(handle)) {
> +			ret = PTR_ERR(handle);
> +			return ret;
> +		}
> +	}
> +
> +	while (ret >= 0 && ret < max_blocks) {
> +		map.m_lblk += ret;
> +		map.m_len = (max_blocks -= ret);
> +		ret = ext4_map_blocks(handle, inode, &map, flags);
> +		if (ret != max_blocks)
> +			ext4_msg(inode->i_sb, KERN_INFO,
> +				     "inode #%lu: block %u: len %u: "
> +				     "split block mapping found for atomic write, "
> +				     "ret = %d",
> +				     inode->i_ino, map.m_lblk,
> +				     map.m_len, ret);
> +		if (ret <= 0)
> +			break;
> +	}
> +
> +	ret2 = ext4_mark_inode_dirty(handle, inode);
> +
> +	if (credits) {
> +		ret3 = ext4_journal_stop(handle);
> +		if (unlikely(ret3))
> +			ret2 = ret3;
> +	}
> +
> +	if (ret <= 0 || ret2)
> +		ext4_warning(inode->i_sb,
> +			     "inode #%lu: block %u: len %u: "
> +			     "returned %d or %d",
> +			     inode->i_ino, map.m_lblk,
> +			     map.m_len, ret, ret2);
> +
> +	return ret > 0 ? ret2 : ret;
> +}
> +
>  /*
>   * This function convert a range of blocks to written extents
>   * The caller of this function will pass the start offset and the size.
> diff --git a/fs/ext4/file.c b/fs/ext4/file.c
> index beb078ee4811..959328072c15 100644
> --- a/fs/ext4/file.c
> +++ b/fs/ext4/file.c
> @@ -377,7 +377,12 @@ static int ext4_dio_write_end_io(struct kiocb *iocb, ssize_t size,
>  	loff_t pos = iocb->ki_pos;
>  	struct inode *inode = file_inode(iocb->ki_filp);
>  
> -	if (!error && size && flags & IOMAP_DIO_UNWRITTEN)
> +
> +	if (!error && size && (flags & IOMAP_DIO_UNWRITTEN) &&
> +			(iocb->ki_flags & IOCB_ATOMIC))
> +		error = ext4_convert_unwritten_extents_atomic(NULL, inode, pos,
> +							      size);
> +	else if (!error && size && flags & IOMAP_DIO_UNWRITTEN)
>  		error = ext4_convert_unwritten_extents(NULL, inode, pos, size);
>  	if (error)
>  		return error;
> diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
> index 8b86b1a29bdc..13bc9f07ae7f 100644
> --- a/fs/ext4/inode.c
> +++ b/fs/ext4/inode.c
> @@ -3412,12 +3412,149 @@ static void ext4_set_iomap(struct inode *inode, struct iomap *iomap,
>  	}
>  }
>  
> +static int ext4_map_blocks_atomic_write_slow(handle_t *handle,
> +			struct inode *inode, struct ext4_map_blocks *map)
> +{
> +	ext4_lblk_t m_lblk = map->m_lblk;
> +	unsigned int m_len = map->m_len;
> +	unsigned int mapped_len = 0, m_flags = 0;
> +	ext4_fsblk_t next_pblk;
> +	bool check_next_pblk = false;
> +	int ret = 0;
> +
> +	WARN_ON_ONCE(!ext4_has_feature_bigalloc(inode->i_sb));
> +
> +	/*
> +	 * This is a slow path in case of mixed mapping. We use
> +	 * EXT4_GET_BLOCKS_CREATE_ZERO flag here to make sure we get a single
> +	 * contiguous mapped mapping. This will ensure any unwritten or hole
> +	 * regions within the requested range is zeroed out and we return
> +	 * a single contiguous mapped extent.
> +	 */
> +	m_flags = EXT4_GET_BLOCKS_CREATE_ZERO;
> +
> +	do {
> +		ret = ext4_map_blocks(handle, inode, map, m_flags);
> +		if (ret < 0 && ret != -ENOSPC)
> +			goto out_err;
> +		/*
> +		 * This should never happen, but let's return an error code to
> +		 * avoid an infinite loop in here.
> +		 */
> +		if (ret == 0) {
> +			ret = -EFSCORRUPTED;
> +			ext4_warning_inode(inode,
> +				"ext4_map_blocks() couldn't allocate blocks m_flags: 0x%x, ret:%d",
> +				m_flags, ret);
> +			goto out_err;
> +		}
> +		/*
> +		 * With bigalloc we should never get ENOSPC nor discontiguous
> +		 * physical extents.
> +		 */
> +		if ((check_next_pblk && next_pblk != map->m_pblk) ||
> +				ret == -ENOSPC) {
> +			ext4_warning_inode(inode,
> +				"Non-contiguous allocation detected: expected %llu, got %llu, "
> +				"or ext4_map_blocks() returned out of space ret: %d",
> +				next_pblk, map->m_pblk, ret);
> +			ret = -EFSCORRUPTED;
> +			goto out_err;
> +		}
> +		next_pblk = map->m_pblk + map->m_len;
> +		check_next_pblk = true;
> +
> +		mapped_len += map->m_len;
> +		map->m_lblk += map->m_len;
> +		map->m_len = m_len - mapped_len;
> +	} while (mapped_len < m_len);
> +
> +	/*
> +	 * We might have done some work in above loop, so we need to query the
> +	 * start of the physical extent, based on the origin m_lblk and m_len.
> +	 * Let's also ensure we were able to allocate the required range for
> +	 * mixed mapping case.
> +	 */
> +	map->m_lblk = m_lblk;
> +	map->m_len = m_len;
> +	map->m_flags = 0;
> +
> +	ret = ext4_map_blocks(handle, inode, map,
> +			      EXT4_GET_BLOCKS_QUERY_LAST_IN_LEAF);
> +	if (ret != m_len) {
> +		ext4_warning_inode(inode,
> +			"allocation failed for atomic write request m_lblk:%u, m_len:%u, ret:%d\n",
> +			m_lblk, m_len, ret);
> +		ret = -EINVAL;

When does this produce a short mapping?  In theory the cluster's already
allocated, right?  So this is (AFAICT) a handler for a "should never
happen" corner case, right?

> +	}
> +	return ret;
> +
> +out_err:
> +	/* reset map before returning an error */
> +	map->m_lblk = m_lblk;
> +	map->m_len = m_len;
> +	map->m_flags = 0;
> +	return ret;
> +}
> +
> +/*
> + * ext4_map_blocks_atomic: Helper routine to ensure the entire requested
> + * range in @map [lblk, lblk + len) is one single contiguous extent with no
> + * mixed mappings.
> + *
> + * We first use m_flags passed to us by our caller (ext4_iomap_alloc()).
> + * We only call EXT4_GET_BLOCKS_ZERO in the slow path, when the underlying
> + * physical extent for the requested range does not have a single contiguous
> + * mapping type i.e. (Hole, Mapped, or Unwritten) throughout.
> + * In that case we will loop over the requested range to allocate and zero out
> + * the unwritten / holes in between, to get a single mapped extent from
> + * [m_lblk, m_lblk +  m_len). Note that this is only possible because we know
> + * this can be called only with bigalloc enabled filesystem where the underlying
> + * cluster is already allocated. This avoids allocating discontiguous extents
> + * in the slow path due to multiple calls to ext4_map_blocks().
> + * The slow path is mostly non-performance critical path, so it should be ok to
> + * loop using ext4_map_blocks() with appropriate flags to allocate & zero the
> + * underlying short holes/unwritten extents within the requested range.
> + */
> +static int ext4_map_blocks_atomic_write(handle_t *handle, struct inode *inode,
> +				struct ext4_map_blocks *map, int m_flags,
> +				bool *force_commit)
> +{
> +	ext4_lblk_t m_lblk = map->m_lblk;
> +	unsigned int m_len = map->m_len;
> +	int ret = 0;
> +
> +	WARN_ON_ONCE(m_len > 1 && !ext4_has_feature_bigalloc(inode->i_sb));
> +
> +	ret = ext4_map_blocks(handle, inode, map, m_flags);
> +	if (ret < 0 || ret == m_len)
> +		goto out;
> +	/*
> +	 * This is a mixed mapping case where we were not able to allocate
> +	 * a single contiguous extent. In that case let's reset requested
> +	 * mapping and call the slow path.
> +	 */
> +	map->m_lblk = m_lblk;
> +	map->m_len = m_len;
> +	map->m_flags = 0;
> +
> +	/*
> +	 * slow path means we have mixed mapping, that means we will need
> +	 * to force txn commit.
> +	 */
> +	*force_commit = true;
> +	return ext4_map_blocks_atomic_write_slow(handle, inode, map);
> +out:
> +	return ret;
> +}
> +
>  static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map,
>  			    unsigned int flags)
>  {
>  	handle_t *handle;
>  	u8 blkbits = inode->i_blkbits;
>  	int ret, dio_credits, m_flags = 0, retries = 0;
> +	bool force_commit = false;
>  
>  	/*
>  	 * Trim the mapping request to the maximum value that we can map at
> @@ -3425,7 +3562,30 @@ static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map,
>  	 */
>  	if (map->m_len > DIO_MAX_BLOCKS)
>  		map->m_len = DIO_MAX_BLOCKS;
> -	dio_credits = ext4_chunk_trans_blocks(inode, map->m_len);
> +
> +	/*
> +	 * journal credits estimation for atomic writes. We call
> +	 * ext4_map_blocks(), to find if there could be a mixed mapping. If yes,
> +	 * then let's assume the no. of pextents required can be m_len i.e.
> +	 * every alternate block can be unwritten and hole.
> +	 */
> +	if (flags & IOMAP_ATOMIC) {
> +		unsigned int orig_mlen = map->m_len;
> +
> +		ret = ext4_map_blocks(NULL, inode, map, 0);
> +		if (ret < 0)
> +			return ret;
> +		if (map->m_len < orig_mlen) {
> +			map->m_len = orig_mlen;
> +			dio_credits = ext4_meta_trans_blocks(inode, orig_mlen,
> +							     map->m_len);
> +		} else {
> +			dio_credits = ext4_chunk_trans_blocks(inode,
> +							      map->m_len);
> +		}
> +	} else {
> +		dio_credits = ext4_chunk_trans_blocks(inode, map->m_len);
> +	}
>  
>  retry:
>  	/*
> @@ -3456,7 +3616,11 @@ static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map,
>  	else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
>  		m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT;
>  
> -	ret = ext4_map_blocks(handle, inode, map, m_flags);
> +	if (flags & IOMAP_ATOMIC)
> +		ret = ext4_map_blocks_atomic_write(handle, inode, map, m_flags,
> +						   &force_commit);
> +	else
> +		ret = ext4_map_blocks(handle, inode, map, m_flags);
>  
>  	/*
>  	 * We cannot fill holes in indirect tree based inodes as that could
> @@ -3470,6 +3634,14 @@ static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map,
>  	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
>  		goto retry;
>  
> +	if (ret > 0 && force_commit) {
> +		int ret2;
> +
> +		ret2 = ext4_force_commit(inode->i_sb);
> +		if (ret2)
> +			ret = ret2;

Nit: This could return ret2 directly instead of assigning it to ret and
letting it fall out.

But my bigger complaint is that you ought to leave a comment here along
the lines of:

		/*
		 * Someone forced us to commit the journal ahead of an
		 * IO operation so that the ondisk mapping state is
		 * consistent with the contents of the file data blocks.
		 * The commit failed, so we abort the whole IO.
		 */

so it's obvious why we got a mapping but are erroring out anyway.

If the answers to my questions are all 'yes' and the extra comment gets
added, then

Acked-by: "Darrick J. Wong" <djwong@...nel.org>

--D

> +	}
> +
>  	return ret;
>  }
>  
> @@ -3480,6 +3652,7 @@ static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
>  	int ret;
>  	struct ext4_map_blocks map;
>  	u8 blkbits = inode->i_blkbits;
> +	unsigned int orig_mlen;
>  
>  	if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
>  		return -EINVAL;
> @@ -3493,6 +3666,7 @@ static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
>  	map.m_lblk = offset >> blkbits;
>  	map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
>  			  EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
> +	orig_mlen = map.m_len;
>  
>  	if (flags & IOMAP_WRITE) {
>  		/*
> @@ -3503,8 +3677,16 @@ static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
>  		 */
>  		if (offset + length <= i_size_read(inode)) {
>  			ret = ext4_map_blocks(NULL, inode, &map, 0);
> -			if (ret > 0 && (map.m_flags & EXT4_MAP_MAPPED))
> -				goto out;
> +			/*
> +			 * For atomic writes the entire requested length should
> +			 * be mapped.
> +			 */
> +			if (map.m_flags & EXT4_MAP_MAPPED) {
> +				if ((!(flags & IOMAP_ATOMIC) && ret > 0) ||
> +				   (flags & IOMAP_ATOMIC && ret >= orig_mlen))
> +					goto out;
> +			}
> +			map.m_len = orig_mlen;
>  		}
>  		ret = ext4_iomap_alloc(inode, &map, flags);
>  	} else {
> @@ -3525,6 +3707,16 @@ static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
>  	 */
>  	map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len);
>  
> +	/*
> +	 * Before returning to iomap, let's ensure the allocated mapping
> +	 * covers the entire requested length for atomic writes.
> +	 */
> +	if (flags & IOMAP_ATOMIC) {
> +		if (map.m_len < (length >> blkbits)) {
> +			WARN_ON(1);
> +			return -EINVAL;
> +		}
> +	}
>  	ext4_set_iomap(inode, iomap, &map, offset, length, flags);
>  
>  	return 0;
> -- 
> 2.49.0
> 
> 

Powered by blists - more mailing lists