[v4] Btrfs: add support for fallocate's zero range operation
diff mbox

Message ID 20171104040747.8737-1-fdmanana@kernel.org
State New
Headers show

Commit Message

Filipe Manana Nov. 4, 2017, 4:07 a.m. UTC
From: Filipe Manana <fdmanana@suse.com>

This implements support the zero range operation of fallocate. For now
at least it's as simple as possible while reusing most of the existing
fallocate and hole punching infrastructure.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
---

V2: Removed double inode unlock on error path from failure to lock range.
V3: Factored common code to update isize and inode item into a helper
    function, plus some minor cleanup.
V4: Removed no longer lock_inode parameter as of V3.

 fs/btrfs/file.c | 338 +++++++++++++++++++++++++++++++++++++++++++++-----------
 1 file changed, 276 insertions(+), 62 deletions(-)

Comments

Nikolay Borisov Nov. 10, 2017, 4:43 p.m. UTC | #1
On  4.11.2017 06:07, fdmanana@kernel.org wrote:
> From: Filipe Manana <fdmanana@suse.com>
> 
> This implements support the zero range operation of fallocate. For now
> at least it's as simple as possible while reusing most of the existing
> fallocate and hole punching infrastructure.
> 
> Signed-off-by: Filipe Manana <fdmanana@suse.com>
> ---
> 
> V2: Removed double inode unlock on error path from failure to lock range.
> V3: Factored common code to update isize and inode item into a helper
>     function, plus some minor cleanup.
> V4: Removed no longer lock_inode parameter as of V3.
> 

When this gets merged into Linus' tree we'd need to update fallocate's
man page as well
--
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
David Sterba Jan. 5, 2018, 4:49 p.m. UTC | #2
On Sat, Nov 04, 2017 at 04:07:47AM +0000, fdmanana@kernel.org wrote:
> From: Filipe Manana <fdmanana@suse.com>
> 
> This implements support the zero range operation of fallocate. For now
> at least it's as simple as possible while reusing most of the existing
> fallocate and hole punching infrastructure.
> 
> Signed-off-by: Filipe Manana <fdmanana@suse.com>

FYI, I've added this patch to the rest of the 4.16 queue.
--
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

Patch
diff mbox

diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
index aafcc785f840..ea2e863eb540 100644
--- a/fs/btrfs/file.c
+++ b/fs/btrfs/file.c
@@ -2448,6 +2448,46 @@  static int find_first_non_hole(struct inode *inode, u64 *start, u64 *len)
 	return ret;
 }
 
+static int btrfs_punch_hole_lock_range(struct inode *inode,
+				       const u64 lockstart,
+				       const u64 lockend,
+				       struct extent_state **cached_state)
+{
+	while (1) {
+		struct btrfs_ordered_extent *ordered;
+		int ret;
+
+		truncate_pagecache_range(inode, lockstart, lockend);
+
+		lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
+				 cached_state);
+		ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
+
+		/*
+		 * We need to make sure we have no ordered extents in this range
+		 * and nobody raced in and read a page in this range, if we did
+		 * we need to try again.
+		 */
+		if ((!ordered ||
+		    (ordered->file_offset + ordered->len <= lockstart ||
+		     ordered->file_offset > lockend)) &&
+		     !btrfs_page_exists_in_range(inode, lockstart, lockend)) {
+			if (ordered)
+				btrfs_put_ordered_extent(ordered);
+			break;
+		}
+		if (ordered)
+			btrfs_put_ordered_extent(ordered);
+		unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
+				     lockend, cached_state, GFP_NOFS);
+		ret = btrfs_wait_ordered_range(inode, lockstart,
+					       lockend - lockstart + 1);
+		if (ret)
+			return ret;
+	}
+	return 0;
+}
+
 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
 {
 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
@@ -2564,38 +2604,11 @@  static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
 		goto out_only_mutex;
 	}
 
-	while (1) {
-		struct btrfs_ordered_extent *ordered;
-
-		truncate_pagecache_range(inode, lockstart, lockend);
-
-		lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
-				 &cached_state);
-		ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
-
-		/*
-		 * We need to make sure we have no ordered extents in this range
-		 * and nobody raced in and read a page in this range, if we did
-		 * we need to try again.
-		 */
-		if ((!ordered ||
-		    (ordered->file_offset + ordered->len <= lockstart ||
-		     ordered->file_offset > lockend)) &&
-		     !btrfs_page_exists_in_range(inode, lockstart, lockend)) {
-			if (ordered)
-				btrfs_put_ordered_extent(ordered);
-			break;
-		}
-		if (ordered)
-			btrfs_put_ordered_extent(ordered);
-		unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
-				     lockend, &cached_state, GFP_NOFS);
-		ret = btrfs_wait_ordered_range(inode, lockstart,
-					       lockend - lockstart + 1);
-		if (ret) {
-			inode_unlock(inode);
-			return ret;
-		}
+	ret = btrfs_punch_hole_lock_range(inode, lockstart, lockend,
+					  &cached_state);
+	if (ret) {
+		inode_unlock(inode);
+		goto out_only_mutex;
 	}
 
 	path = btrfs_alloc_path();
@@ -2804,6 +2817,217 @@  static int add_falloc_range(struct list_head *head, u64 start, u64 len)
 	return 0;
 }
 
+static int btrfs_fallocate_update_isize(struct inode *inode,
+					const u64 end,
+					const int mode)
+{
+	struct btrfs_trans_handle *trans;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	int ret;
+	int ret2;
+
+	if (mode & FALLOC_FL_KEEP_SIZE || end <= i_size_read(inode))
+		return 0;
+
+	trans = btrfs_start_transaction(root, 1);
+	if (IS_ERR(trans))
+		return PTR_ERR(trans);
+
+	inode->i_ctime = current_time(inode);
+	i_size_write(inode, end);
+	btrfs_ordered_update_i_size(inode, end, NULL);
+	ret = btrfs_update_inode(trans, root, inode);
+	ret2 = btrfs_end_transaction(trans);
+
+	return ret ? ret : ret2;
+}
+
+static int btrfs_zero_range_check_range_boundary(struct inode *inode,
+						 u64 offset)
+{
+	const u64 sectorsize = btrfs_inode_sectorsize(inode);
+	struct extent_map *em;
+	int ret = 0;
+
+	offset = round_down(offset, sectorsize);
+	em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, offset, sectorsize, 0);
+	if (IS_ERR(em))
+		return PTR_ERR(em);
+
+	if (em->block_start == EXTENT_MAP_HOLE)
+		ret = 1;
+
+	free_extent_map(em);
+	return ret;
+}
+
+static int btrfs_zero_range(struct inode *inode,
+			    loff_t offset,
+			    loff_t len,
+			    const int mode)
+{
+	struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
+	struct extent_map *em;
+	struct extent_changeset *data_reserved = NULL;
+	int ret;
+	u64 alloc_hint = 0;
+	const u64 sectorsize = btrfs_inode_sectorsize(inode);
+	u64 alloc_start = round_down(offset, sectorsize);
+	u64 alloc_end = round_up(offset + len, sectorsize);
+	u64 bytes_to_reserve = 0;
+	bool space_reserved = false;
+
+	inode_dio_wait(inode);
+
+	em = btrfs_get_extent(BTRFS_I(inode), NULL, 0,
+			      alloc_start, alloc_end - alloc_start, 0);
+	if (IS_ERR(em)) {
+		ret = PTR_ERR(em);
+		goto out;
+	}
+
+	/*
+	 * Avoid hole punching and extent allocation for some cases. More cases
+	 * could be considered, but these are unlikely common and we keep things
+	 * as simple as possible for now. Also, intentionally, if the target
+	 * range contains one or more prealloc extents together with regular
+	 * extents and holes, we drop all the existing extents and allocate a
+	 * new prealloc extent, so that we get a larger contiguous disk extent.
+	 */
+	if (em->start <= alloc_start &&
+	    test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
+		const u64 em_end = em->start + em->len;
+
+		if (em_end >= offset + len) {
+			/*
+			 * The whole range is already a prealloc extent,
+			 * do nothing except updating the inode's i_size if
+			 * needed.
+			 */
+			free_extent_map(em);
+			ret = btrfs_fallocate_update_isize(inode, offset + len,
+							   mode);
+			goto out;
+		}
+		/*
+		 * Part of the range is already a prealloc extent, so operate
+		 * only on the remaining part of the range.
+		 */
+		alloc_start = em_end;
+		ASSERT(IS_ALIGNED(alloc_start, sectorsize));
+		len = offset + len - alloc_start;
+		offset = alloc_start;
+		alloc_hint = em->block_start + em->len;
+	}
+	free_extent_map(em);
+
+	if (BTRFS_BYTES_TO_BLKS(fs_info, offset) ==
+	    BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)) {
+		em = btrfs_get_extent(BTRFS_I(inode), NULL, 0,
+				      alloc_start, sectorsize, 0);
+		if (IS_ERR(em)) {
+			ret = PTR_ERR(em);
+			goto out;
+		}
+
+		if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
+			free_extent_map(em);
+			ret = btrfs_fallocate_update_isize(inode, offset + len,
+							   mode);
+			goto out;
+		}
+		if (len < sectorsize && em->block_start != EXTENT_MAP_HOLE) {
+			free_extent_map(em);
+			ret = btrfs_truncate_block(inode, offset, len, 0);
+			if (!ret)
+				ret = btrfs_fallocate_update_isize(inode,
+								   offset + len,
+								   mode);
+			return ret;
+		}
+		free_extent_map(em);
+		alloc_start = round_down(offset, sectorsize);
+		alloc_end = alloc_start + sectorsize;
+		goto reserve_space;
+	}
+
+	alloc_start = round_up(offset, sectorsize);
+	alloc_end = round_down(offset + len, sectorsize);
+
+	/*
+	 * For unaligned ranges, check the pages at the boundaries, they might
+	 * map to an extent, in which case we need to partially zero them, or
+	 * they might map to a hole, in which case we need our allocation range
+	 * to cover them.
+	 */
+	if (!IS_ALIGNED(offset, sectorsize)) {
+		ret = btrfs_zero_range_check_range_boundary(inode, offset);
+		if (ret < 0)
+			goto out;
+		if (ret) {
+			alloc_start = round_down(offset, sectorsize);
+			ret = 0;
+		} else {
+			ret = btrfs_truncate_block(inode, offset, 0, 0);
+			if (ret)
+				goto out;
+		}
+	}
+
+	if (!IS_ALIGNED(offset + len, sectorsize)) {
+		ret = btrfs_zero_range_check_range_boundary(inode,
+							    offset + len);
+		if (ret < 0)
+			goto out;
+		if (ret) {
+			alloc_end = round_up(offset + len, sectorsize);
+			ret = 0;
+		} else {
+			ret = btrfs_truncate_block(inode, offset + len, 0, 1);
+			if (ret)
+				goto out;
+		}
+	}
+
+reserve_space:
+	if (alloc_start < alloc_end) {
+		struct extent_state *cached_state = NULL;
+		const u64 lockstart = alloc_start;
+		const u64 lockend = alloc_end - 1;
+
+		bytes_to_reserve = alloc_end - alloc_start;
+		ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
+						      bytes_to_reserve);
+		if (ret < 0)
+			goto out;
+		space_reserved = true;
+		ret = btrfs_qgroup_reserve_data(inode, &data_reserved,
+						alloc_start, bytes_to_reserve);
+		if (ret)
+			goto out;
+		ret = btrfs_punch_hole_lock_range(inode, lockstart, lockend,
+						  &cached_state);
+		if (ret)
+			goto out;
+		ret = btrfs_prealloc_file_range(inode, mode, alloc_start,
+						alloc_end - alloc_start,
+						i_blocksize(inode),
+						offset + len, &alloc_hint);
+		unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
+				     lockend, &cached_state, GFP_KERNEL);
+		/* btrfs_prealloc_file_range releases reserved space on error */
+		if (ret)
+			space_reserved = false;
+	}
+ out:
+	if (ret && space_reserved)
+		btrfs_free_reserved_data_space(inode, data_reserved,
+					       alloc_start, bytes_to_reserve);
+	extent_changeset_free(data_reserved);
+
+	return ret;
+}
+
 static long btrfs_fallocate(struct file *file, int mode,
 			    loff_t offset, loff_t len)
 {
@@ -2829,7 +3053,8 @@  static long btrfs_fallocate(struct file *file, int mode,
 	cur_offset = alloc_start;
 
 	/* Make sure we aren't being give some crap mode */
-	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
+	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
+		     FALLOC_FL_ZERO_RANGE))
 		return -EOPNOTSUPP;
 
 	if (mode & FALLOC_FL_PUNCH_HOLE)
@@ -2840,10 +3065,12 @@  static long btrfs_fallocate(struct file *file, int mode,
 	 *
 	 * For qgroup space, it will be checked later.
 	 */
-	ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
-			alloc_end - alloc_start);
-	if (ret < 0)
-		return ret;
+	if (!(mode & FALLOC_FL_ZERO_RANGE)) {
+		ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
+						      alloc_end - alloc_start);
+		if (ret < 0)
+			return ret;
+	}
 
 	inode_lock(inode);
 
@@ -2885,6 +3112,12 @@  static long btrfs_fallocate(struct file *file, int mode,
 	if (ret)
 		goto out;
 
+	if (mode & FALLOC_FL_ZERO_RANGE) {
+		ret = btrfs_zero_range(inode, offset, len, mode);
+		inode_unlock(inode);
+		return ret;
+	}
+
 	locked_end = alloc_end - 1;
 	while (1) {
 		struct btrfs_ordered_extent *ordered;
@@ -2980,37 +3213,18 @@  static long btrfs_fallocate(struct file *file, int mode,
 	if (ret < 0)
 		goto out_unlock;
 
-	if (actual_end > inode->i_size &&
-	    !(mode & FALLOC_FL_KEEP_SIZE)) {
-		struct btrfs_trans_handle *trans;
-		struct btrfs_root *root = BTRFS_I(inode)->root;
-
-		/*
-		 * We didn't need to allocate any more space, but we
-		 * still extended the size of the file so we need to
-		 * update i_size and the inode item.
-		 */
-		trans = btrfs_start_transaction(root, 1);
-		if (IS_ERR(trans)) {
-			ret = PTR_ERR(trans);
-		} else {
-			inode->i_ctime = current_time(inode);
-			i_size_write(inode, actual_end);
-			btrfs_ordered_update_i_size(inode, actual_end, NULL);
-			ret = btrfs_update_inode(trans, root, inode);
-			if (ret)
-				btrfs_end_transaction(trans);
-			else
-				ret = btrfs_end_transaction(trans);
-		}
-	}
+	/*
+	 * We didn't need to allocate any more space, but we still extended the
+	 * size of the file so we need to update i_size and the inode item.
+	 */
+	ret = btrfs_fallocate_update_isize(inode, actual_end, mode);
 out_unlock:
 	unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
 			     &cached_state, GFP_KERNEL);
 out:
 	inode_unlock(inode);
 	/* Let go of our reservation. */
-	if (ret != 0)
+	if (ret != 0 && !(mode & FALLOC_FL_ZERO_RANGE))
 		btrfs_free_reserved_data_space(inode, data_reserved,
 				alloc_start, alloc_end - cur_offset);
 	extent_changeset_free(data_reserved);