[PATCH v4 00/10] btrfs: defrag: rework to support sector perfect defrag

[PATCH v4 00/10] btrfs: defrag: rework to support sector perfect defrag

[Qu Wenruo]

This branch is based on subpage RW branch, as the last patch needs to
enable defrag support for subpage cases.

But despite that one, all other patches can be applied on current
misc-next.

[BACKGROUND]
In subpage rw branch, we disable defrag completely due to the fact that
current code can only work on page basis.

This could lead to problems like btrfs/062 crash.

Thus this patchset will make defrag to work on both regular and subpage
sectorsize.

[SOLUTION]
To defrag a file range, what we do is pretty much like buffered write,
except we don't really write any new data to page cache, but just mark
the range dirty.

Then let later writeback to merge the range into a larger extent.

But current defrag code is working on per-page basis, not per-sector,
thus we have to refactor it a little to make it to work properly for
subpage.

This patch will separate the code into 3 layers:
Layer 0:	btrfs_defrag_file()
		The defrag entrace
		Just do proper inode lock and split the file into
		page aligned 256K clusters to defrag

Layer 1:	defrag_one_cluster()
		Will collect the initial targets file extents, and pass
		each continuous target to defrag_one_range()

Layer 2:	defrag_one_range()
		Will prepare the needed page and extent locking.
		Then re-check the range for real target list, as initial
		target list is not consistent as it doesn't hage
		page/extent locking to prevent hole punching.

Layer 3:	defrag_one_locked_target()
		The real work, to make the extent range defrag and
		update involved page status

[BEHAVIOR CHANGE]
In the refactor, there is one behavior change:

- Defraged sector counter is based on the initial target list
  This is mostly to avoid the paremters to be passed too deep into
  defrag_one_locked_target().
  Considering the accounting is not that important, we can afford some
  difference.

[PATCH STRUTURE]
Patch 01~03:	Small independent refactor to improve readability
Patch 04~08:	Implement the more readable and subpage friendly defrag
Patch 09:	Cleanup of old infrastruture
Patch 10:	Enable defrag for subpage case

Now both regular sectorsize and subpage sectorsize can pass defrag test
group.

Although the tests no longer crashes, I still see random test failure on
64K page size with 4K sectorsize, reporting nbytes mismatch for fsstress
+ defrag workload.

So please don't merge the branch into v5.14 merge window at least.

This normally means defrag has defragged some hole without setting it
with DELALLOC_NEW.
But I haven't find any location which makes it possible, as the new
defrag_collect_targets() call timing (with both page and extent locked,
no ordered extents) should be as safe as the original one.

Thus I'm wondering if the check timing is even correct for the existing
code.
But at least I can't reproduce the nbytes problem on x86_64, thus no yet
sure what the real cause is.

[CHANGELOG]
v2:
- Make sure we won't defrag hole
  This is done by re-collect the target list after have page and extent
  locked. So that we can have a consistent view of the extent map.

- Add a new layer to avoid variable naming bugs
  Since we need to handle real target list inside defrag_one_range(),
  and that function has parameters like "start" and "len", while inside
  the loop we need things like "entry->start" and "entry->len", it has
  already caused hard to debug bugs during development.

  Thus introduce a new layer, defrag_one_ragen() to prepare pages/extent
  lock then pass the entry to defrag_one_locked_target().

v3:
- Fix extent_state leak
  Now we pass the @cached_state to defrag_one_locked_target() other
  than allowing it to allocate a new one.

  This can be reproduced by enabling "TEST_FS_MODULE_RELOAD" environment
  variable for fstests and run "-g defrag" group.

- Fix a random hang in test cases like btrfs/062
  Since defrag_one_range() will lock the extent range first, then
  call defrag_collect_targets(), which will in turn call
  defrag_lookup_extent() and lock extent range again.

  This will cause a dead lock, and this only happens when the
  extent range is smaller than the original locked range.
  Thus sometimes the test can pass, but sometimes it can hang.

  Fix it by teaching defrag_collect_targets() and defrag_lookup_extent()
  to skip extent lock for certain call sites.
  Thus this needs some loops for btrfs/062.
  The hang possibility is around 1/2 ~ 1/3 when run in a loop.

v4:
- Fix a callsite which can cause hang due to extent locking
  The call site is defrag_lookup_extent() in defrag_collect_tagets(),
  which has one call site called with extent lock hold.
  Thus it also need to pass the @locked parameter

- Fix a typo
  "defraged" -> "defragged"

Qu Wenruo (10):
  btrfs: defrag: pass file_ra_state instead of file for
    btrfs_defrag_file()
  btrfs: defrag: extract the page preparation code into one helper
  btrfs: defrag: replace hard coded PAGE_SIZE to sectorsize for
    defrag_lookup_extent()
  btrfs: defrag: introduce a new helper to collect target file extents
  btrfs: defrag: introduce a helper to defrag a continuous prepared
    range
  btrfs: defrag: introduce a helper to defrag a range
  btrfs: defrag: introduce a new helper to defrag one cluster
  btrfs: defrag: use defrag_one_cluster() to implement
    btrfs_defrag_file()
  btrfs: defrag: remove the old infrastructure
  btrfs: defrag: enable defrag for subpage case

 fs/btrfs/ctree.h |   4 +-
 fs/btrfs/ioctl.c | 908 ++++++++++++++++++++++-------------------------
 2 files changed, 426 insertions(+), 486 deletions(-)

-- 
2.32.0

[PATCH v4 01/10] btrfs: defrag: pass file_ra_state instead of file for btrfs_defrag_file()

[Qu Wenruo]

Currently btrfs_defrag_file() accepts both "struct inode" and "struct
file" as parameter, which can't be more confusing.
As we can easily grab "struct inode" from "struct file" using
file_inode() helper.

The reason why we need "struct file" is just to re-use its f_ra.

This patch will change this by passing "struct file_ra_state" parameter,
so that it's more clear what we really want.

Since we're here, also add some comment on the function
btrfs_defrag_file() to make later reader less miserable.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
---
 fs/btrfs/ctree.h |  4 ++--
 fs/btrfs/ioctl.c | 27 ++++++++++++++++++---------
 2 files changed, 20 insertions(+), 11 deletions(-)

diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 45798e68331a..918df8877b45 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -3212,9 +3212,9 @@ int btrfs_fileattr_set(struct user_namespace *mnt_userns,
 int btrfs_ioctl_get_supported_features(void __user *arg);
 void btrfs_sync_inode_flags_to_i_flags(struct inode *inode);
 int __pure btrfs_is_empty_uuid(u8 *uuid);
-int btrfs_defrag_file(struct inode *inode, struct file *file,
+int btrfs_defrag_file(struct inode *inode, struct file_ra_state *ra,
 		      struct btrfs_ioctl_defrag_range_args *range,
-		      u64 newer_than, unsigned long max_pages);
+		      u64 newer_than, unsigned long max_to_defrag);
 void btrfs_get_block_group_info(struct list_head *groups_list,
 				struct btrfs_ioctl_space_info *space);
 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index 11adf4670c55..05af6f5ff3ff 100644
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -1350,13 +1350,22 @@ static int cluster_pages_for_defrag(struct inode *inode,
 
 }
 
-int btrfs_defrag_file(struct inode *inode, struct file *file,
+/*
+ * Btrfs entrace for defrag.
+ *
+ * @inode:	   Inode to be defragged
+ * @ra:		   Readahead state. If NULL, one will be allocated at runtime.
+ * @range:	   Defrag options including range and flags.
+ * @newer_than:	   Minimal transid to defrag
+ * @max_to_defrag: Max number of sectors to be defragged, if 0, the whole inode
+ *		   will be defragged.
+ */
+int btrfs_defrag_file(struct inode *inode, struct file_ra_state *ra,
 		      struct btrfs_ioctl_defrag_range_args *range,
 		      u64 newer_than, unsigned long max_to_defrag)
 {
 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 	struct btrfs_root *root = BTRFS_I(inode)->root;
-	struct file_ra_state *ra = NULL;
 	unsigned long last_index;
 	u64 isize = i_size_read(inode);
 	u64 last_len = 0;
@@ -1374,6 +1383,7 @@ int btrfs_defrag_file(struct inode *inode, struct file *file,
 	u64 new_align = ~((u64)SZ_128K - 1);
 	struct page **pages = NULL;
 	bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
+	bool ra_allocated = false;
 
 	if (isize == 0)
 		return 0;
@@ -1392,16 +1402,15 @@ int btrfs_defrag_file(struct inode *inode, struct file *file,
 		extent_thresh = SZ_256K;
 
 	/*
-	 * If we were not given a file, allocate a readahead context. As
+	 * If we were not given a ra, allocate a readahead context. As
 	 * readahead is just an optimization, defrag will work without it so
 	 * we don't error out.
 	 */
-	if (!file) {
+	if (!ra) {
+		ra_allocated = true;
 		ra = kzalloc(sizeof(*ra), GFP_KERNEL);
 		if (ra)
 			file_ra_state_init(ra, inode->i_mapping);
-	} else {
-		ra = &file->f_ra;
 	}
 
 	pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
@@ -1483,7 +1492,7 @@ int btrfs_defrag_file(struct inode *inode, struct file *file,
 			ra_index = max(i, ra_index);
 			if (ra)
 				page_cache_sync_readahead(inode->i_mapping, ra,
-						file, ra_index, cluster);
+						NULL, ra_index, cluster);
 			ra_index += cluster;
 		}
 
@@ -1554,7 +1563,7 @@ int btrfs_defrag_file(struct inode *inode, struct file *file,
 		BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
 		btrfs_inode_unlock(inode, 0);
 	}
-	if (!file)
+	if (ra_allocated)
 		kfree(ra);
 	kfree(pages);
 	return ret;
@@ -3076,7 +3085,7 @@ static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
 			/* the rest are all set to zero by kzalloc */
 			range->len = (u64)-1;
 		}
-		ret = btrfs_defrag_file(file_inode(file), file,
+		ret = btrfs_defrag_file(file_inode(file), &file->f_ra,
 					range, BTRFS_OLDEST_GENERATION, 0);
 		if (ret > 0)
 			ret = 0;
-- 
2.32.0

[PATCH v4 02/10] btrfs: defrag: extract the page preparation code into one helper

[Qu Wenruo]

In cluster_pages_for_defrag(), we have complex code block inside one
for() loop.

The code block is to prepare one page for defrag, this will ensure:
- The page is locked and set up properly
- No ordered extent exists in the page range
- The page is uptodate
- The writeback has finished

This behavior is pretty common and will be reused by later defrag
rework.

So extract the code into its own helper, defrag_prepare_one_page(), for
later usage, and cleanup the code by a little.

Since we're here, also make the page check to be subpage compatible,
which means we will also check page::private, not only page->mapping.

Signed-off-by: Qu Wenruo <wqu@suse.com>
---
 fs/btrfs/ioctl.c | 151 +++++++++++++++++++++++++++--------------------
 1 file changed, 86 insertions(+), 65 deletions(-)

diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index 05af6f5ff3ff..a06ceb8fdf28 100644
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -1143,6 +1143,89 @@ static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
 	return ret;
 }
 
+/*
+ * Prepare one page to be defragged.
+ *
+ * This will ensure:
+ * - Returned page is locked and has been set up properly
+ * - No ordered extent exists in the page
+ * - The page is uptodate
+ * - The writeback has finished
+ */
+static struct page *defrag_prepare_one_page(struct btrfs_inode *inode,
+					    unsigned long index)
+{
+	struct address_space *mapping = inode->vfs_inode.i_mapping;
+	gfp_t mask = btrfs_alloc_write_mask(mapping);
+	u64 page_start = index << PAGE_SHIFT;
+	u64 page_end = page_start + PAGE_SIZE - 1;
+	struct extent_state *cached_state = NULL;
+	struct page *page;
+	int ret;
+
+again:
+	page = find_or_create_page(mapping, index, mask);
+	if (!page)
+		return ERR_PTR(-ENOMEM);
+
+	ret = set_page_extent_mapped(page);
+	if (ret < 0) {
+		unlock_page(page);
+		put_page(page);
+		return ERR_PTR(ret);
+	}
+
+	/* Wait for any exists ordered extent in the range */
+	while (1) {
+		struct btrfs_ordered_extent *ordered;
+
+		lock_extent_bits(&inode->io_tree, page_start, page_end,
+				 &cached_state);
+		ordered = btrfs_lookup_ordered_range(inode, page_start,
+						     PAGE_SIZE);
+		unlock_extent_cached(&inode->io_tree, page_start, page_end,
+				     &cached_state);
+		if (!ordered)
+			break;
+
+		unlock_page(page);
+		btrfs_start_ordered_extent(ordered, 1);
+		btrfs_put_ordered_extent(ordered);
+		lock_page(page);
+		/*
+		 * we unlocked the page above, so we need check if
+		 * it was released or not.
+		 */
+		if (page->mapping != mapping || !PagePrivate(page)) {
+			unlock_page(page);
+			put_page(page);
+			goto again;
+		}
+	}
+
+	/*
+	 * Now the page range has no ordered extent any more.
+	 * Read the page to make it uptodate.
+	 */
+	if (!PageUptodate(page)) {
+		btrfs_readpage(NULL, page);
+		lock_page(page);
+		if (page->mapping != mapping || !PagePrivate(page)) {
+			unlock_page(page);
+			put_page(page);
+			goto again;
+		}
+		if (!PageUptodate(page)) {
+			unlock_page(page);
+			put_page(page);
+			return ERR_PTR(-EIO);
+		}
+	}
+
+	wait_on_page_writeback(page);
+	return page;
+}
+
 /*
  * it doesn't do much good to defrag one or two pages
  * at a time.  This pulls in a nice chunk of pages
@@ -1170,11 +1253,8 @@ static int cluster_pages_for_defrag(struct inode *inode,
 	int ret;
 	int i;
 	int i_done;
-	struct btrfs_ordered_extent *ordered;
 	struct extent_state *cached_state = NULL;
-	struct extent_io_tree *tree;
 	struct extent_changeset *data_reserved = NULL;
-	gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
 
 	file_end = (isize - 1) >> PAGE_SHIFT;
 	if (!isize || start_index > file_end)
@@ -1187,68 +1267,16 @@ static int cluster_pages_for_defrag(struct inode *inode,
 	if (ret)
 		return ret;
 	i_done = 0;
-	tree = &BTRFS_I(inode)->io_tree;
 
 	/* step one, lock all the pages */
 	for (i = 0; i < page_cnt; i++) {
 		struct page *page;
-again:
-		page = find_or_create_page(inode->i_mapping,
-					   start_index + i, mask);
-		if (!page)
-			break;
 
-		ret = set_page_extent_mapped(page);
-		if (ret < 0) {
-			unlock_page(page);
-			put_page(page);
+		page = defrag_prepare_one_page(BTRFS_I(inode), start_index + i);
+		if (IS_ERR(page)) {
+			ret = PTR_ERR(page);
 			break;
 		}
-
-		page_start = page_offset(page);
-		page_end = page_start + PAGE_SIZE - 1;
-		while (1) {
-			lock_extent_bits(tree, page_start, page_end,
-					 &cached_state);
-			ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode),
-							      page_start);
-			unlock_extent_cached(tree, page_start, page_end,
-					     &cached_state);
-			if (!ordered)
-				break;
-
-			unlock_page(page);
-			btrfs_start_ordered_extent(ordered, 1);
-			btrfs_put_ordered_extent(ordered);
-			lock_page(page);
-			/*
-			 * we unlocked the page above, so we need check if
-			 * it was released or not.
-			 */
-			if (page->mapping != inode->i_mapping) {
-				unlock_page(page);
-				put_page(page);
-				goto again;
-			}
-		}
-
-		if (!PageUptodate(page)) {
-			btrfs_readpage(NULL, page);
-			lock_page(page);
-			if (!PageUptodate(page)) {
-				unlock_page(page);
-				put_page(page);
-				ret = -EIO;
-				break;
-			}
-		}
-
-		if (page->mapping != inode->i_mapping) {
-			unlock_page(page);
-			put_page(page);
-			goto again;
-		}
-
 		pages[i] = page;
 		i_done++;
 	}
@@ -1258,13 +1286,6 @@ static int cluster_pages_for_defrag(struct inode *inode,
 	if (!(inode->i_sb->s_flags & SB_ACTIVE))
 		goto out;
 
-	/*
-	 * so now we have a nice long stream of locked
-	 * and up to date pages, lets wait on them
-	 */
-	for (i = 0; i < i_done; i++)
-		wait_on_page_writeback(pages[i]);
-
 	page_start = page_offset(pages[0]);
 	page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
 
-- 
2.32.0

[PATCH v4 03/10] btrfs: defrag: replace hard coded PAGE_SIZE to sectorsize for defrag_lookup_extent()

[Qu Wenruo]

When testing subpage defrag support, I always find some strange inode
nbytes error, after a lot of debugging, it turns out that
defrag_lookup_extent() is using PAGE_SIZE as size for
lookup_extent_mapping().

Since lookup_extent_mapping() is calling __lookup_extent_mapping() with
@strict == 1, this means any extent map smaller than one page will be
ignored, prevent subpage defrag to grab a correct extent map.

There are quite some PAGE_SIZE usage in ioctl.c, but most of them are
correct usages, and can be one of the following cases:

- ioctl structure size check
  We want ioctl structure be contained inside one page.

- real page opeartions

There is one special case left untouched, check_defrag_in_cache().
This function will later be removed completely, thus there is not much
meaning to change it now.

Signed-off-by: Qu Wenruo <wqu@suse.com>
---
 fs/btrfs/ioctl.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index a06ceb8fdf28..096157ee30f1 100644
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -1037,23 +1037,24 @@ static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 	struct extent_map *em;
-	u64 len = PAGE_SIZE;
+	const u32 sectorsize = BTRFS_I(inode)->root->fs_info->sectorsize;
 
 	/*
 	 * hopefully we have this extent in the tree already, try without
 	 * the full extent lock
 	 */
 	read_lock(&em_tree->lock);
-	em = lookup_extent_mapping(em_tree, start, len);
+	em = lookup_extent_mapping(em_tree, start, sectorsize);
 	read_unlock(&em_tree->lock);
 
 	if (!em) {
 		struct extent_state *cached = NULL;
-		u64 end = start + len - 1;
+		u64 end = start + sectorsize - 1;
 
 		/* get the big lock and read metadata off disk */
 		lock_extent_bits(io_tree, start, end, &cached);
-		em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
+		em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start,
+				      sectorsize);
 		unlock_extent_cached(io_tree, start, end, &cached);
 
 		if (IS_ERR(em))
-- 
2.32.0

[PATCH v4 04/10] btrfs: defrag: introduce a new helper to collect target file extents

[Qu Wenruo]

Introduce a new helper, defrag_collect_targets(), to collect all
possible targets to be defragged.

This function will not consider things like max_sectors_to_defrag, thus
caller should be responsible to ensure we don't exceed the limit.

This function will be the first stage of later defrag rework.

Signed-off-by: Qu Wenruo <wqu@suse.com>
---
 fs/btrfs/ioctl.c | 120 +++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 120 insertions(+)

diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index 096157ee30f1..0bfd13b959ed 100644
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -1372,6 +1372,126 @@ static int cluster_pages_for_defrag(struct inode *inode,
 
 }
 
+struct defrag_target_range {
+	struct list_head list;
+	u64 start;
+	u64 len;
+};
+
+/*
+ * Helper to collect all valid target extents.
+ *
+ * @start:	   The file offset to lookup
+ * @len:	   The length to lookup
+ * @extent_thresh: File extent size threshold, any extent size >= this value
+ *		   will be ignored
+ * @newer_than:    Only defrag extents newer than this value
+ * @do_compress:   Whether the defrag is doing compression
+ *		   If true, @extent_thresh will be ignored and all regular
+ *		   file extents meeting @newer_than will be targets.
+ * @target_list:   The list of targets file extents
+ */
+static int defrag_collect_targets(struct btrfs_inode *inode,
+				  u64 start, u64 len, u32 extent_thresh,
+				  u64 newer_than, bool do_compress,
+				  struct list_head *target_list)
+{
+	u64 cur = start;
+	int ret = 0;
+
+	while (cur < start + len) {
+		struct extent_map *em;
+		struct defrag_target_range *new;
+		bool next_mergeable = true;
+		u64 range_len;
+
+		em = defrag_lookup_extent(&inode->vfs_inode, cur);
+		if (!em)
+			break;
+
+		/* Skip hole/inline/preallocated extents */
+		if (em->block_start >= EXTENT_MAP_LAST_BYTE ||
+		    test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
+			goto next;
+
+		/* Skip older extent */
+		if (em->generation < newer_than)
+			goto next;
+
+		/*
+		 * For do_compress case, we want to compress all valid file
+		 * extents, thus no @extent_thresh or mergeable check.
+		 */
+		if (do_compress)
+			goto add;
+
+		/* Skip too large extent */
+		if (em->len >= extent_thresh)
+			goto next;
+
+		next_mergeable = defrag_check_next_extent(&inode->vfs_inode, em);
+		if (!next_mergeable) {
+			struct defrag_target_range *last;
+
+			/* Empty target list, no way to merge with last entry */
+			if (list_empty(target_list))
+				goto next;
+			last = list_entry(target_list->prev,
+					struct defrag_target_range, list);
+			/* Not mergeable with last entry */
+			if (last->start + last->len != cur)
+				goto next;
+
+			/* Mergeable, fall through to add it to @target_list. */
+		}
+
+add:
+		range_len = min(extent_map_end(em), start + len) - cur;
+		/*
+		 * This one is a good target, check if it can be merged into
+		 * last range of the target list
+		 */
+		if (!list_empty(target_list)) {
+			struct defrag_target_range *last;
+
+			last = list_entry(target_list->prev,
+					struct defrag_target_range, list);
+			ASSERT(last->start + last->len <= cur);
+			if (last->start + last->len == cur) {
+				/* Mergeable, enlarge the last entry */
+				last->len += range_len;
+				goto next;
+			}
+			/* Fall through to allocate a new entry */
+		}
+
+		/* Allocate new defrag_target_range */
+		new = kmalloc(sizeof(*new), GFP_NOFS);
+		if (!new) {
+			free_extent_map(em);
+			ret = -ENOMEM;
+			break;
+		}
+		new->start = cur;
+		new->len = range_len;
+		list_add_tail(&new->list, target_list);
+
+next:
+		cur = extent_map_end(em);
+		free_extent_map(em);
+	}
+	if (ret < 0) {
+		struct defrag_target_range *entry;
+		struct defrag_target_range *tmp;
+
+		list_for_each_entry_safe(entry, tmp, target_list, list) {
+			list_del_init(&entry->list);
+			kfree(entry);
+		}
+	}
+	return ret;
+}
+
 /*
  * Btrfs entrace for defrag.
  *
-- 
2.32.0

[PATCH v4 05/10] btrfs: defrag: introduce a helper to defrag a continuous prepared range

[Qu Wenruo]

A new helper, defrag_one_locked_target(), introduced to do the real part
of defrag.

The caller needs to ensure both page and extents bits are locked, and no
ordered extent for the range, and all writeback is finished.

The core defrag part is pretty straight-forward:

- Reserve space
- Set extent bits to defrag
- Update involved pages to be dirty

Signed-off-by: Qu Wenruo <wqu@suse.com>
---
 fs/btrfs/ioctl.c | 56 ++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 56 insertions(+)

diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index 0bfd13b959ed..423c880e42d4 100644
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -47,6 +47,7 @@
 #include "space-info.h"
 #include "delalloc-space.h"
 #include "block-group.h"
+#include "subpage.h"
 
 #ifdef CONFIG_64BIT
 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
@@ -1492,6 +1493,61 @@ static int defrag_collect_targets(struct btrfs_inode *inode,
 	return ret;
 }
 
+#define CLUSTER_SIZE	(SZ_256K)
+
+/*
+ * Defrag one continuous target range.
+ *
+ * @inode:	Target inode
+ * @target:	Target range to defrag
+ * @pages:	Locked pages covering the defrag range
+ * @nr_pages:	Number of locked pages
+ *
+ * Caller should ensure:
+ *
+ * - Pages are prepared
+ *   Pages should be locked, no ordered extent in the pages range,
+ *   no writeback.
+ *
+ * - Extent bits are locked
+ */
+static int defrag_one_locked_target(struct btrfs_inode *inode,
+				    struct defrag_target_range *target,
+				    struct page **pages, int nr_pages,
+				    struct extent_state **cached_state)
+{
+	struct btrfs_fs_info *fs_info = inode->root->fs_info;
+	struct extent_changeset *data_reserved = NULL;
+	const u64 start = target->start;
+	const u64 len = target->len;
+	unsigned long last_index = (start + len - 1) >> PAGE_SHIFT;
+	unsigned long start_index = start >> PAGE_SHIFT;
+	unsigned long first_index = page_index(pages[0]);
+	int ret = 0;
+	int i;
+
+	ASSERT(last_index - first_index + 1 <= nr_pages);
+
+	ret = btrfs_delalloc_reserve_space(inode, &data_reserved, start, len);
+	if (ret < 0)
+		return ret;
+	clear_extent_bit(&inode->io_tree, start, start + len - 1,
+			 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
+			 EXTENT_DEFRAG, 0, 0, cached_state);
+	set_extent_defrag(&inode->io_tree, start, start + len - 1,
+			  cached_state);
+
+	/* Update the page status */
+	for (i = start_index - first_index; i <= last_index - first_index;
+	     i++) {
+		ClearPageChecked(pages[i]);
+		btrfs_page_clamp_set_dirty(fs_info, pages[i], start, len);
+	}
+	btrfs_delalloc_release_extents(inode, len);
+	extent_changeset_free(data_reserved);
+	return ret;
+}
+
 /*
  * Btrfs entrace for defrag.
  *
-- 
2.32.0

[PATCH v4 06/10] btrfs: defrag: introduce a helper to defrag a range

[Qu Wenruo]

A new helper, defrag_one_range(), is introduced to defrag one range.

This function will mostly prepare the needed pages and extent status for
defrag_one_locked_target().

As we can only have a consistent view of extent map with page and
extent bits locked, we need to re-check the range passed in to get a
real target list for defrag_one_locked_target().

Since defrag_collect_targets() will call defrag_lookup_extent() and lock
extent range, we also need to teach those two functions to skip extent
lock.
Thus new parameter, @locked, is introruced to skip extent lock if the
caller has already locked the range.

Signed-off-by: Qu Wenruo <wqu@suse.com>
---
 fs/btrfs/ioctl.c | 102 ++++++++++++++++++++++++++++++++++++++++++-----
 1 file changed, 92 insertions(+), 10 deletions(-)

diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index 423c880e42d4..7ca013781e69 100644
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -1033,7 +1033,8 @@ static int find_new_extents(struct btrfs_root *root,
 	return -ENOENT;
 }
 
-static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
+static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start,
+					       bool locked)
 {
 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
@@ -1053,10 +1054,12 @@ static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 		u64 end = start + sectorsize - 1;
 
 		/* get the big lock and read metadata off disk */
-		lock_extent_bits(io_tree, start, end, &cached);
+		if (!locked)
+			lock_extent_bits(io_tree, start, end, &cached);
 		em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start,
 				      sectorsize);
-		unlock_extent_cached(io_tree, start, end, &cached);
+		if (!locked)
+			unlock_extent_cached(io_tree, start, end, &cached);
 
 		if (IS_ERR(em))
 			return NULL;
@@ -1065,7 +1068,8 @@ static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 	return em;
 }
 
-static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
+static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em,
+				     bool locked)
 {
 	struct extent_map *next;
 	bool ret = true;
@@ -1074,7 +1078,7 @@ static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
 	if (em->start + em->len >= i_size_read(inode))
 		return false;
 
-	next = defrag_lookup_extent(inode, em->start + em->len);
+	next = defrag_lookup_extent(inode, em->start + em->len, locked);
 	if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
 		ret = false;
 	else if ((em->block_start + em->block_len == next->block_start) &&
@@ -1103,7 +1107,7 @@ static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
 
 	*skip = 0;
 
-	em = defrag_lookup_extent(inode, start);
+	em = defrag_lookup_extent(inode, start, false);
 	if (!em)
 		return 0;
 
@@ -1116,7 +1120,7 @@ static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
 	if (!*defrag_end)
 		prev_mergeable = false;
 
-	next_mergeable = defrag_check_next_extent(inode, em);
+	next_mergeable = defrag_check_next_extent(inode, em, false);
 	/*
 	 * we hit a real extent, if it is big or the next extent is not a
 	 * real extent, don't bother defragging it
@@ -1390,12 +1394,13 @@ struct defrag_target_range {
  * @do_compress:   Whether the defrag is doing compression
  *		   If true, @extent_thresh will be ignored and all regular
  *		   file extents meeting @newer_than will be targets.
+ * @locked:	   If the range has already hold extent lock
  * @target_list:   The list of targets file extents
  */
 static int defrag_collect_targets(struct btrfs_inode *inode,
 				  u64 start, u64 len, u32 extent_thresh,
 				  u64 newer_than, bool do_compress,
-				  struct list_head *target_list)
+				  bool locked, struct list_head *target_list)
 {
 	u64 cur = start;
 	int ret = 0;
@@ -1406,7 +1411,7 @@ static int defrag_collect_targets(struct btrfs_inode *inode,
 		bool next_mergeable = true;
 		u64 range_len;
 
-		em = defrag_lookup_extent(&inode->vfs_inode, cur);
+		em = defrag_lookup_extent(&inode->vfs_inode, cur, locked);
 		if (!em)
 			break;
 
@@ -1430,7 +1435,8 @@ static int defrag_collect_targets(struct btrfs_inode *inode,
 		if (em->len >= extent_thresh)
 			goto next;
 
-		next_mergeable = defrag_check_next_extent(&inode->vfs_inode, em);
+		next_mergeable = defrag_check_next_extent(&inode->vfs_inode, em,
+							  locked);
 		if (!next_mergeable) {
 			struct defrag_target_range *last;
 
@@ -1548,6 +1554,82 @@ static int defrag_one_locked_target(struct btrfs_inode *inode,
 	return ret;
 }
 
+static int defrag_one_range(struct btrfs_inode *inode,
+			    u64 start, u32 len,
+			    u32 extent_thresh, u64 newer_than,
+			    bool do_compress)
+{
+	struct extent_state *cached_state = NULL;
+	struct defrag_target_range *entry;
+	struct defrag_target_range *tmp;
+	LIST_HEAD(target_list);
+	struct page **pages;
+	const u32 sectorsize = inode->root->fs_info->sectorsize;
+	unsigned long last_index = (start + len - 1) >> PAGE_SHIFT;
+	unsigned long start_index = start >> PAGE_SHIFT;
+	unsigned int nr_pages = last_index - start_index + 1;
+	int ret = 0;
+	int i;
+
+	ASSERT(nr_pages <= CLUSTER_SIZE / PAGE_SIZE);
+	ASSERT(IS_ALIGNED(start, sectorsize) && IS_ALIGNED(len, sectorsize));
+
+	pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
+	if (!pages)
+		return -ENOMEM;
+
+	/* Prepare all pages */
+	for (i = 0; i < nr_pages; i++) {
+		pages[i] = defrag_prepare_one_page(inode, start_index + i);
+		if (IS_ERR(pages[i])) {
+			ret = PTR_ERR(pages[i]);
+			pages[i] = NULL;
+			goto free_pages;
+		}
+	}
+	/* Also lock the pages range */
+	lock_extent_bits(&inode->io_tree, start_index << PAGE_SHIFT,
+			 (last_index << PAGE_SHIFT) + PAGE_SIZE - 1,
+			 &cached_state);
+	/*
+	 * Now we have a consistent view about the extent map, re-check
+	 * which range really needs to be defragged.
+	 *
+	 * And this time we have extent locked already, pass @locked = true
+	 * so that we won't re-lock the extent range and cause deadlock.
+	 */
+	ret = defrag_collect_targets(inode, start, len, extent_thresh,
+				     newer_than, do_compress, true,
+				     &target_list);
+	if (ret < 0)
+		goto unlock_extent;
+
+	list_for_each_entry(entry, &target_list, list) {
+		ret = defrag_one_locked_target(inode, entry, pages, nr_pages,
+					       &cached_state);
+		if (ret < 0)
+			break;
+	}
+
+	list_for_each_entry_safe(entry, tmp, &target_list, list) {
+		list_del_init(&entry->list);
+		kfree(entry);
+	}
+unlock_extent:
+	unlock_extent_cached(&inode->io_tree, start_index << PAGE_SHIFT,
+			     (last_index << PAGE_SHIFT) + PAGE_SIZE - 1,
+			     &cached_state);
+free_pages:
+	for (i = 0; i < nr_pages; i++) {
+		if (pages[i]) {
+			unlock_page(pages[i]);
+			put_page(pages[i]);
+		}
+	}
+	kfree(pages);
+	return ret;
+}
+
 /*
  * Btrfs entrace for defrag.
  *
-- 
2.32.0

[PATCH v4 07/10] btrfs: defrag: introduce a new helper to defrag one cluster

[Qu Wenruo]

This new helper, defrag_one_cluster(), will defrag one cluster (at most
256K) by:

- Collect all initial targets

- Kick in readahead when possible

- Call defrag_one_range() on each initial target
  With some extra range clamping.

- Update @sectors_defragged parameter

This involves one behavior change, the defragged sectors accounting is
no longer as accurate as old behavior, as the initial targets are not
consistent.

We can have new holes punched inside the initial target, and we will
skip such holes later.
But the defragged sectors accounting doesn't need to be that accurate
anyway, thus I don't want to pass those extra accounting burden into
defrag_one_range().

Signed-off-by: Qu Wenruo <wqu@suse.com>
---
 fs/btrfs/ioctl.c | 56 ++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 56 insertions(+)

diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index 7ca013781e69..c7f41b8f313d 100644
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -1630,6 +1630,62 @@ static int defrag_one_range(struct btrfs_inode *inode,
 	return ret;
 }
 
+static int defrag_one_cluster(struct btrfs_inode *inode,
+			      struct file_ra_state *ra,
+			      u64 start, u32 len, u32 extent_thresh,
+			      u64 newer_than, bool do_compress,
+			      unsigned long *sectors_defragged,
+			      unsigned long max_sectors)
+{
+	const u32 sectorsize = inode->root->fs_info->sectorsize;
+	struct defrag_target_range *entry;
+	struct defrag_target_range *tmp;
+	LIST_HEAD(target_list);
+	int ret;
+
+	BUILD_BUG_ON(!IS_ALIGNED(CLUSTER_SIZE, PAGE_SIZE));
+	ret = defrag_collect_targets(inode, start, len, extent_thresh,
+				     newer_than, do_compress, false,
+				     &target_list);
+	if (ret < 0)
+		goto out;
+
+	list_for_each_entry(entry, &target_list, list) {
+		u32 range_len = entry->len;
+
+		/* Reached the limit */
+		if (max_sectors && max_sectors == *sectors_defragged)
+			break;
+
+		if (max_sectors)
+			range_len = min_t(u32, range_len,
+				(max_sectors - *sectors_defragged) * sectorsize);
+
+		if (ra)
+			page_cache_sync_readahead(inode->vfs_inode.i_mapping,
+				ra, NULL, entry->start >> PAGE_SHIFT,
+				((entry->start + range_len - 1) >> PAGE_SHIFT) -
+				(entry->start >> PAGE_SHIFT) + 1);
+		/*
+		 * Here we may not defrag any range if holes are punched before
+		 * we locked the pages.
+		 * But that's fine, it only affects the @sectors_defragged
+		 * accounting.
+		 */
+		ret = defrag_one_range(inode, entry->start, range_len,
+					extent_thresh, newer_than, do_compress);
+		if (ret < 0)
+			break;
+		*sectors_defragged += range_len;
+	}
+out:
+	list_for_each_entry_safe(entry, tmp, &target_list, list) {
+		list_del_init(&entry->list);
+		kfree(entry);
+	}
+	return ret;
+}
+
 /*
  * Btrfs entrace for defrag.
  *
-- 
2.32.0

[PATCH v4 08/10] btrfs: defrag: use defrag_one_cluster() to implement btrfs_defrag_file()

[Qu Wenruo]

The function defrag_one_cluster() is able to defrag one range well
enough, we only need to do prepration for it, including:

- Clamp and align the defrag range
- Exclude invalid cases
- Proper inode locking

The old infrastructures will not be removed in this patch, as it would
be too noisy to review.

Signed-off-by: Qu Wenruo <wqu@suse.com>
---
 fs/btrfs/ioctl.c | 204 +++++++++++++----------------------------------
 1 file changed, 55 insertions(+), 149 deletions(-)

diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index c7f41b8f313d..9bd062c92fcd 100644
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -1701,25 +1701,15 @@ int btrfs_defrag_file(struct inode *inode, struct file_ra_state *ra,
 		      u64 newer_than, unsigned long max_to_defrag)
 {
 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
-	struct btrfs_root *root = BTRFS_I(inode)->root;
-	unsigned long last_index;
+	unsigned long sectors_defragged = 0;
 	u64 isize = i_size_read(inode);
-	u64 last_len = 0;
-	u64 skip = 0;
-	u64 defrag_end = 0;
-	u64 newer_off = range->start;
-	unsigned long i;
-	unsigned long ra_index = 0;
-	int ret;
-	int defrag_count = 0;
-	int compress_type = BTRFS_COMPRESS_ZLIB;
-	u32 extent_thresh = range->extent_thresh;
-	unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
-	unsigned long cluster = max_cluster;
-	u64 new_align = ~((u64)SZ_128K - 1);
-	struct page **pages = NULL;
+	u64 cur;
+	u64 last_byte;
 	bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
 	bool ra_allocated = false;
+	int compress_type = BTRFS_COMPRESS_ZLIB;
+	int ret;
+	u32 extent_thresh = range->extent_thresh;
 
 	if (isize == 0)
 		return 0;
@@ -1737,6 +1727,14 @@ int btrfs_defrag_file(struct inode *inode, struct file_ra_state *ra,
 	if (extent_thresh == 0)
 		extent_thresh = SZ_256K;
 
+	if (range->start + range->len > range->start) {
+		/* Got a specific range */
+		last_byte = min(isize, range->start + range->len) - 1;
+	} else {
+		/* Defrag until file end */
+		last_byte = isize - 1;
+	}
+
 	/*
 	 * If we were not given a ra, allocate a readahead context. As
 	 * readahead is just an optimization, defrag will work without it so
@@ -1749,159 +1747,67 @@ int btrfs_defrag_file(struct inode *inode, struct file_ra_state *ra,
 			file_ra_state_init(ra, inode->i_mapping);
 	}
 
-	pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
-	if (!pages) {
-		ret = -ENOMEM;
-		goto out_ra;
-	}
+	/* Align the range */
+	cur = round_down(range->start, fs_info->sectorsize);
+	last_byte = round_up(last_byte, fs_info->sectorsize) - 1;
 
-	/* find the last page to defrag */
-	if (range->start + range->len > range->start) {
-		last_index = min_t(u64, isize - 1,
-			 range->start + range->len - 1) >> PAGE_SHIFT;
-	} else {
-		last_index = (isize - 1) >> PAGE_SHIFT;
-	}
-
-	if (newer_than) {
-		ret = find_new_extents(root, inode, newer_than,
-				       &newer_off, SZ_64K);
-		if (!ret) {
-			range->start = newer_off;
-			/*
-			 * we always align our defrag to help keep
-			 * the extents in the file evenly spaced
-			 */
-			i = (newer_off & new_align) >> PAGE_SHIFT;
-		} else
-			goto out_ra;
-	} else {
-		i = range->start >> PAGE_SHIFT;
-	}
-	if (!max_to_defrag)
-		max_to_defrag = last_index - i + 1;
-
-	/*
-	 * make writeback starts from i, so the defrag range can be
-	 * written sequentially.
-	 */
-	if (i < inode->i_mapping->writeback_index)
-		inode->i_mapping->writeback_index = i;
-
-	while (i <= last_index && defrag_count < max_to_defrag &&
-	       (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
-		/*
-		 * make sure we stop running if someone unmounts
-		 * the FS
-		 */
-		if (!(inode->i_sb->s_flags & SB_ACTIVE))
-			break;
-
-		if (btrfs_defrag_cancelled(fs_info)) {
-			btrfs_debug(fs_info, "defrag_file cancelled");
-			ret = -EAGAIN;
-			goto error;
-		}
-
-		if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
-					 extent_thresh, &last_len, &skip,
-					 &defrag_end, do_compress)){
-			unsigned long next;
-			/*
-			 * the should_defrag function tells us how much to skip
-			 * bump our counter by the suggested amount
-			 */
-			next = DIV_ROUND_UP(skip, PAGE_SIZE);
-			i = max(i + 1, next);
-			continue;
-		}
+	while (cur < last_byte) {
+		u64 cluster_end;
 
-		if (!newer_than) {
-			cluster = (PAGE_ALIGN(defrag_end) >>
-				   PAGE_SHIFT) - i;
-			cluster = min(cluster, max_cluster);
-		} else {
-			cluster = max_cluster;
-		}
+		/* The cluster size 256K should always be page aligned */
+		BUILD_BUG_ON(!IS_ALIGNED(CLUSTER_SIZE, PAGE_SIZE));
 
-		if (i + cluster > ra_index) {
-			ra_index = max(i, ra_index);
-			if (ra)
-				page_cache_sync_readahead(inode->i_mapping, ra,
-						NULL, ra_index, cluster);
-			ra_index += cluster;
-		}
+		/* We want the cluster ends at page boundary when possible */
+		cluster_end = (((cur >> PAGE_SHIFT) +
+			       (SZ_256K >> PAGE_SHIFT)) << PAGE_SHIFT) - 1;
+		cluster_end = min(cluster_end, last_byte);
 
 		btrfs_inode_lock(inode, 0);
 		if (IS_SWAPFILE(inode)) {
 			ret = -ETXTBSY;
-		} else {
-			if (do_compress)
-				BTRFS_I(inode)->defrag_compress = compress_type;
-			ret = cluster_pages_for_defrag(inode, pages, i, cluster);
+			btrfs_inode_unlock(inode, 0);
+			break;
 		}
-		if (ret < 0) {
+		if (!(inode->i_sb->s_flags & SB_ACTIVE)) {
 			btrfs_inode_unlock(inode, 0);
-			goto out_ra;
+			break;
 		}
-
-		defrag_count += ret;
-		balance_dirty_pages_ratelimited(inode->i_mapping);
+		if (do_compress)
+			BTRFS_I(inode)->defrag_compress = compress_type;
+		ret = defrag_one_cluster(BTRFS_I(inode), ra, cur,
+				cluster_end + 1 - cur, extent_thresh,
+				newer_than, do_compress,
+				&sectors_defragged, max_to_defrag);
 		btrfs_inode_unlock(inode, 0);
-
-		if (newer_than) {
-			if (newer_off == (u64)-1)
-				break;
-
-			if (ret > 0)
-				i += ret;
-
-			newer_off = max(newer_off + 1,
-					(u64)i << PAGE_SHIFT);
-
-			ret = find_new_extents(root, inode, newer_than,
-					       &newer_off, SZ_64K);
-			if (!ret) {
-				range->start = newer_off;
-				i = (newer_off & new_align) >> PAGE_SHIFT;
-			} else {
-				break;
-			}
-		} else {
-			if (ret > 0) {
-				i += ret;
-				last_len += ret << PAGE_SHIFT;
-			} else {
-				i++;
-				last_len = 0;
-			}
-		}
+		if (ret < 0)
+			break;
+		cur = cluster_end + 1;
 	}
 
-	ret = defrag_count;
-error:
-	if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
-		filemap_flush(inode->i_mapping);
-		if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
-			     &BTRFS_I(inode)->runtime_flags))
+	if (ra_allocated)
+		kfree(ra);
+	if (sectors_defragged) {
+		/*
+		 * We have defragged some sectors, for compression case
+		 * they need to be written back immediately.
+		 */
+		if (range->flags & BTRFS_DEFRAG_RANGE_START_IO) {
 			filemap_flush(inode->i_mapping);
+			if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
+				     &BTRFS_I(inode)->runtime_flags))
+				filemap_flush(inode->i_mapping);
+		}
+		if (range->compress_type == BTRFS_COMPRESS_LZO)
+			btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
+		else if (range->compress_type == BTRFS_COMPRESS_ZSTD)
+			btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
+		ret = sectors_defragged;
 	}
-
-	if (range->compress_type == BTRFS_COMPRESS_LZO) {
-		btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
-	} else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
-		btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
-	}
-
-out_ra:
 	if (do_compress) {
 		btrfs_inode_lock(inode, 0);
 		BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
 		btrfs_inode_unlock(inode, 0);
 	}
-	if (ra_allocated)
-		kfree(ra);
-	kfree(pages);
 	return ret;
 }
 
-- 
2.32.0

[PATCH v4 09/10] btrfs: defrag: remove the old infrastructure

[Qu Wenruo]

Now the old infrastructure can all be removed.

Signed-off-by: Qu Wenruo <wqu@suse.com>
---
 fs/btrfs/ioctl.c | 305 -----------------------------------------------
 1 file changed, 305 deletions(-)

diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index 9bd062c92fcd..446b2336725c 100644
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -933,106 +933,6 @@ static noinline int btrfs_mksnapshot(const struct path *parent,
 	return ret;
 }
 
-/*
- * When we're defragging a range, we don't want to kick it off again
- * if it is really just waiting for delalloc to send it down.
- * If we find a nice big extent or delalloc range for the bytes in the
- * file you want to defrag, we return 0 to let you know to skip this
- * part of the file
- */
-static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
-{
-	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
-	struct extent_map *em = NULL;
-	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
-	u64 end;
-
-	read_lock(&em_tree->lock);
-	em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
-	read_unlock(&em_tree->lock);
-
-	if (em) {
-		end = extent_map_end(em);
-		free_extent_map(em);
-		if (end - offset > thresh)
-			return 0;
-	}
-	/* if we already have a nice delalloc here, just stop */
-	thresh /= 2;
-	end = count_range_bits(io_tree, &offset, offset + thresh,
-			       thresh, EXTENT_DELALLOC, 1);
-	if (end >= thresh)
-		return 0;
-	return 1;
-}
-
-/*
- * helper function to walk through a file and find extents
- * newer than a specific transid, and smaller than thresh.
- *
- * This is used by the defragging code to find new and small
- * extents
- */
-static int find_new_extents(struct btrfs_root *root,
-			    struct inode *inode, u64 newer_than,
-			    u64 *off, u32 thresh)
-{
-	struct btrfs_path *path;
-	struct btrfs_key min_key;
-	struct extent_buffer *leaf;
-	struct btrfs_file_extent_item *extent;
-	int type;
-	int ret;
-	u64 ino = btrfs_ino(BTRFS_I(inode));
-
-	path = btrfs_alloc_path();
-	if (!path)
-		return -ENOMEM;
-
-	min_key.objectid = ino;
-	min_key.type = BTRFS_EXTENT_DATA_KEY;
-	min_key.offset = *off;
-
-	while (1) {
-		ret = btrfs_search_forward(root, &min_key, path, newer_than);
-		if (ret != 0)
-			goto none;
-process_slot:
-		if (min_key.objectid != ino)
-			goto none;
-		if (min_key.type != BTRFS_EXTENT_DATA_KEY)
-			goto none;
-
-		leaf = path->nodes[0];
-		extent = btrfs_item_ptr(leaf, path->slots[0],
-					struct btrfs_file_extent_item);
-
-		type = btrfs_file_extent_type(leaf, extent);
-		if (type == BTRFS_FILE_EXTENT_REG &&
-		    btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
-		    check_defrag_in_cache(inode, min_key.offset, thresh)) {
-			*off = min_key.offset;
-			btrfs_free_path(path);
-			return 0;
-		}
-
-		path->slots[0]++;
-		if (path->slots[0] < btrfs_header_nritems(leaf)) {
-			btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
-			goto process_slot;
-		}
-
-		if (min_key.offset == (u64)-1)
-			goto none;
-
-		min_key.offset++;
-		btrfs_release_path(path);
-	}
-none:
-	btrfs_free_path(path);
-	return -ENOENT;
-}
-
 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start,
 					       bool locked)
 {
@@ -1089,66 +989,6 @@ static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em,
 	return ret;
 }
 
-static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
-			       u64 *last_len, u64 *skip, u64 *defrag_end,
-			       int compress)
-{
-	struct extent_map *em;
-	int ret = 1;
-	bool next_mergeable = true;
-	bool prev_mergeable = true;
-
-	/*
-	 * make sure that once we start defragging an extent, we keep on
-	 * defragging it
-	 */
-	if (start < *defrag_end)
-		return 1;
-
-	*skip = 0;
-
-	em = defrag_lookup_extent(inode, start, false);
-	if (!em)
-		return 0;
-
-	/* this will cover holes, and inline extents */
-	if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
-		ret = 0;
-		goto out;
-	}
-
-	if (!*defrag_end)
-		prev_mergeable = false;
-
-	next_mergeable = defrag_check_next_extent(inode, em, false);
-	/*
-	 * we hit a real extent, if it is big or the next extent is not a
-	 * real extent, don't bother defragging it
-	 */
-	if (!compress && (*last_len == 0 || *last_len >= thresh) &&
-	    (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
-		ret = 0;
-out:
-	/*
-	 * last_len ends up being a counter of how many bytes we've defragged.
-	 * every time we choose not to defrag an extent, we reset *last_len
-	 * so that the next tiny extent will force a defrag.
-	 *
-	 * The end result of this is that tiny extents before a single big
-	 * extent will force at least part of that big extent to be defragged.
-	 */
-	if (ret) {
-		*defrag_end = extent_map_end(em);
-	} else {
-		*last_len = 0;
-		*skip = extent_map_end(em);
-		*defrag_end = 0;
-	}
-
-	free_extent_map(em);
-	return ret;
-}
-
 /*
  * Prepare one page to be defragged.
  *
@@ -1232,151 +1072,6 @@ static struct page *defrag_prepare_one_page(struct btrfs_inode *inode,
 	return page;
 }
 
-/*
- * it doesn't do much good to defrag one or two pages
- * at a time.  This pulls in a nice chunk of pages
- * to COW and defrag.
- *
- * It also makes sure the delalloc code has enough
- * dirty data to avoid making new small extents as part
- * of the defrag
- *
- * It's a good idea to start RA on this range
- * before calling this.
- */
-static int cluster_pages_for_defrag(struct inode *inode,
-				    struct page **pages,
-				    unsigned long start_index,
-				    unsigned long num_pages)
-{
-	unsigned long file_end;
-	u64 isize = i_size_read(inode);
-	u64 page_start;
-	u64 page_end;
-	u64 page_cnt;
-	u64 start = (u64)start_index << PAGE_SHIFT;
-	u64 search_start;
-	int ret;
-	int i;
-	int i_done;
-	struct extent_state *cached_state = NULL;
-	struct extent_changeset *data_reserved = NULL;
-
-	file_end = (isize - 1) >> PAGE_SHIFT;
-	if (!isize || start_index > file_end)
-		return 0;
-
-	page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
-
-	ret = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
-			start, page_cnt << PAGE_SHIFT);
-	if (ret)
-		return ret;
-	i_done = 0;
-
-	/* step one, lock all the pages */
-	for (i = 0; i < page_cnt; i++) {
-		struct page *page;
-
-		page = defrag_prepare_one_page(BTRFS_I(inode), start_index + i);
-		if (IS_ERR(page)) {
-			ret = PTR_ERR(page);
-			break;
-		}
-		pages[i] = page;
-		i_done++;
-	}
-	if (!i_done || ret)
-		goto out;
-
-	if (!(inode->i_sb->s_flags & SB_ACTIVE))
-		goto out;
-
-	page_start = page_offset(pages[0]);
-	page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
-
-	lock_extent_bits(&BTRFS_I(inode)->io_tree,
-			 page_start, page_end - 1, &cached_state);
-
-	/*
-	 * When defragmenting we skip ranges that have holes or inline extents,
-	 * (check should_defrag_range()), to avoid unnecessary IO and wasting
-	 * space. At btrfs_defrag_file(), we check if a range should be defragged
-	 * before locking the inode and then, if it should, we trigger a sync
-	 * page cache readahead - we lock the inode only after that to avoid
-	 * blocking for too long other tasks that possibly want to operate on
-	 * other file ranges. But before we were able to get the inode lock,
-	 * some other task may have punched a hole in the range, or we may have
-	 * now an inline extent, in which case we should not defrag. So check
-	 * for that here, where we have the inode and the range locked, and bail
-	 * out if that happened.
-	 */
-	search_start = page_start;
-	while (search_start < page_end) {
-		struct extent_map *em;
-
-		em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, search_start,
-				      page_end - search_start);
-		if (IS_ERR(em)) {
-			ret = PTR_ERR(em);
-			goto out_unlock_range;
-		}
-		if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
-			free_extent_map(em);
-			/* Ok, 0 means we did not defrag anything */
-			ret = 0;
-			goto out_unlock_range;
-		}
-		search_start = extent_map_end(em);
-		free_extent_map(em);
-	}
-
-	clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
-			  page_end - 1, EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
-			  EXTENT_DEFRAG, 0, 0, &cached_state);
-
-	if (i_done != page_cnt) {
-		spin_lock(&BTRFS_I(inode)->lock);
-		btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
-		spin_unlock(&BTRFS_I(inode)->lock);
-		btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
-				start, (page_cnt - i_done) << PAGE_SHIFT, true);
-	}
-
-
-	set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
-			  &cached_state);
-
-	unlock_extent_cached(&BTRFS_I(inode)->io_tree,
-			     page_start, page_end - 1, &cached_state);
-
-	for (i = 0; i < i_done; i++) {
-		clear_page_dirty_for_io(pages[i]);
-		ClearPageChecked(pages[i]);
-		set_page_dirty(pages[i]);
-		unlock_page(pages[i]);
-		put_page(pages[i]);
-	}
-	btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
-	extent_changeset_free(data_reserved);
-	return i_done;
-
-out_unlock_range:
-	unlock_extent_cached(&BTRFS_I(inode)->io_tree,
-			     page_start, page_end - 1, &cached_state);
-out:
-	for (i = 0; i < i_done; i++) {
-		unlock_page(pages[i]);
-		put_page(pages[i]);
-	}
-	btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
-			start, page_cnt << PAGE_SHIFT, true);
-	btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
-	extent_changeset_free(data_reserved);
-	return ret;
-
-}
-
 struct defrag_target_range {
 	struct list_head list;
 	u64 start;
-- 
2.32.0

[PATCH v4 10/10] btrfs: defrag: enable defrag for subpage case

[Qu Wenruo]

With the new infrasturture which has taken subpage into consideration,
now we should be safe to allow defrag to work for subpage case.

Signed-off-by: Qu Wenruo <wqu@suse.com>
---
 fs/btrfs/ioctl.c | 6 ------
 1 file changed, 6 deletions(-)

diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index 446b2336725c..edf597ad515f 100644
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -2974,12 +2974,6 @@ static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
 		goto out;
 	}
 
-	/* Subpage defrag will be supported in later commits */
-	if (root->fs_info->sectorsize < PAGE_SIZE) {
-		ret = -ENOTTY;
-		goto out;
-	}
-
 	switch (inode->i_mode & S_IFMT) {
 	case S_IFDIR:
 		if (!capable(CAP_SYS_ADMIN)) {
-- 
2.32.0

Re: [PATCH v4 00/10] btrfs: defrag: rework to support sector perfect defrag

[riteshh]

On 21/06/10 01:09PM, Qu Wenruo wrote:
> This branch is based on subpage RW branch, as the last patch needs to
> enable defrag support for subpage cases.
>
> But despite that one, all other patches can be applied on current
> misc-next.
>
> [BACKGROUND]
> In subpage rw branch, we disable defrag completely due to the fact that
> current code can only work on page basis.
>
> This could lead to problems like btrfs/062 crash.
>
> Thus this patchset will make defrag to work on both regular and subpage
> sectorsize.
>
> [SOLUTION]
> To defrag a file range, what we do is pretty much like buffered write,
> except we don't really write any new data to page cache, but just mark
> the range dirty.
>
> Then let later writeback to merge the range into a larger extent.
>
> But current defrag code is working on per-page basis, not per-sector,
> thus we have to refactor it a little to make it to work properly for
> subpage.
>
> This patch will separate the code into 3 layers:
> Layer 0:	btrfs_defrag_file()
> 		The defrag entrace
> 		Just do proper inode lock and split the file into
> 		page aligned 256K clusters to defrag
>
> Layer 1:	defrag_one_cluster()
> 		Will collect the initial targets file extents, and pass
> 		each continuous target to defrag_one_range()
>
> Layer 2:	defrag_one_range()
> 		Will prepare the needed page and extent locking.
> 		Then re-check the range for real target list, as initial
> 		target list is not consistent as it doesn't hage
> 		page/extent locking to prevent hole punching.
>
> Layer 3:	defrag_one_locked_target()
> 		The real work, to make the extent range defrag and
> 		update involved page status
>
> [BEHAVIOR CHANGE]
> In the refactor, there is one behavior change:
>
> - Defraged sector counter is based on the initial target list
>   This is mostly to avoid the paremters to be passed too deep into
>   defrag_one_locked_target().
>   Considering the accounting is not that important, we can afford some
>   difference.
>
> [PATCH STRUTURE]
> Patch 01~03:	Small independent refactor to improve readability
> Patch 04~08:	Implement the more readable and subpage friendly defrag
> Patch 09:	Cleanup of old infrastruture
> Patch 10:	Enable defrag for subpage case
>
> Now both regular sectorsize and subpage sectorsize can pass defrag test
> group.
>
> Although the tests no longer crashes, I still see random test failure on
> 64K page size with 4K sectorsize, reporting nbytes mismatch for fsstress
> + defrag workload.
>
> So please don't merge the branch into v5.14 merge window at least.
>
> This normally means defrag has defragged some hole without setting it
> with DELALLOC_NEW.
> But I haven't find any location which makes it possible, as the new
> defrag_collect_targets() call timing (with both page and extent locked,
> no ordered extents) should be as safe as the original one.
>
> Thus I'm wondering if the check timing is even correct for the existing
> code.
> But at least I can't reproduce the nbytes problem on x86_64, thus no yet
> sure what the real cause is.

Hi Qu,

Sorry for getting back late on this. I was using my test machines for other
tests hence this got delayed.

I did test this whole patch series from your github https://github.com/adam900710/linux/commits/defrag_refactor.
I didn't find any crashes (as you also mentioned). But some tests did fail e.g. btrfs/072.

Is this(btrfs/072) what you are referring to in above (random test failues with subpage
sectorsize)?

Please do let me know if you would like me to try any of the fstest in loop for
reproducing any error. Or if there is any other tests that you would like me to
run?

1. output from btrfs/072 test.
	_check_btrfs_filesystem: filesystem on /dev/vdc is inconsistent
	*** fsck.btrfs output ***
	[1/7] checking root items
	[2/7] checking extents
	[3/7] checking free space cache
	btrfs: csum mismatch on free space cache
	failed to load free space cache for block group 22020096
	btrfs: csum mismatch on free space cache
	failed to load free space cache for block group 1095761920
	btrfs: csum mismatch on free space cache
	failed to load free space cache for block group 1364197376
	[4/7] checking fs roots
	root 5 inode 323 errors 400, nbytes wrong
	ERROR: errors found in fs roots
	Opening filesystem to check...
	Checking filesystem on /dev/vdc
	UUID: b638db59-981c-4502-831e-038930e65cf4
	found 28282880 bytes used, error(s) found
	total csum bytes: 15196
	total tree bytes: 442368
	total fs tree bytes: 331776
	total extent tree bytes: 36864
	btree space waste bytes: 163191
	file data blocks allocated: 37101568
	 referenced 27430912
	*** end fsck.btrfs output

2. btrfs/124 (- I guess this we have seen in past w/o your changes too).
	[ 2565.969989] BTRFS info (device vdc): balance: start -d -m -s
	[ 2565.970727] BTRFS info (device vdc): relocating block group 1513488384 flags data
	[ 2570.693594] BTRFS error (device vdc): bad tree block start, want 31821824 have 0
	[ 2570.725804] BTRFS info (device vdc): read error corrected: ino 0 off 31821824 (dev /dev/vdi sector 21192)
	[ 2581.579592] BTRFS info (device vdc): found 5 extents, stage: move data extents
	[ 2581.818799] BTRFS info (device vdc): found 5 extents, stage: update data pointers
	[ 2582.086796] BTRFS info (device vdc): relocating block group 976617472 flags data
	[ 2598.759264] BTRFS info (device vdc): found 6 extents, stage: move data extents
	[ 2598.991059] BTRFS info (device vdc): found 6 extents, stage: update data pointers
	[ 2599.274253] BTRFS info (device vdc): relocating block group 943063040 flags system
	[ 2599.525191] BTRFS info (device vdc): relocating block group 674627584 flags metadata
	[ 2599.788734] BTRFS info (device vdc): relocating block group 298844160 flags data|raid1
	[ 2609.746593] BTRFS info (device vdc): found 4 extents, stage: move data extents
	[ 2609.989663] BTRFS info (device vdc): found 3 extents, stage: update data pointers
	[ 2610.234642] BTRFS info (device vdc): relocating block group 30408704 flags metadata|raid1
	[ 2610.234889] ------------[ cut here ]------------
	[ 2610.234970] BTRFS: Transaction aborted (error -28)
	[ 2610.235172] WARNING: CPU: 3 PID: 26186 at fs/btrfs/extent-tree.c:2163 btrfs_run_delayed_refs+0x108/0x330
	[ 2610.235364] Modules linked in:
	[ 2610.235426] CPU: 3 PID: 26186 Comm: btrfs Not tainted 5.13.0-rc2-00393-gc0e4bc6f271f #43
	[ 2610.235547] NIP:  c0000000009f0bb8 LR: c0000000009f0bb4 CTR: c000000000e5fb30
	[ 2610.235665] REGS: c000000078fb7320 TRAP: 0700   Not tainted  (5.13.0-rc2-00393-gc0e4bc6f271f)
	[ 2610.235802] MSR:  800000000282b033 <SF,VEC,VSX,EE,FP,ME,IR,DR,RI,LE>  CR: 48004222  XER: 20000000
	[ 2610.235961] CFAR: c0000000001cea40 IRQMASK: 0
	               GPR00: c0000000009f0bb4 c000000078fb75c0 c000000001c5dc00 0000000000000026
	               GPR04: 00000000fffff5ff c000000078fb7280 0000000000000027 c0000000ff507e98
	               GPR08: 0000000000000023 0000000000000000 c00000000c286580 c000000001a11050
	               GPR12: 0000000000004400 c00000003ffeae00 0000000000000002 c000000045e4e088
	               GPR16: c000000045e4e000 c000000001334eb0 c000000012654000 0000000001d00000
	               GPR20: c000000012654000 0000000000000e5c 0000000000000000 c000000001662dc8
	               GPR24: 0000000000000014 c000000009ec9800 c000000009ec9b48 c0000000410f6e08
	               GPR28: 0000000000000000 0000000000000001 c000000012654000 ffffffffffffffe4
	[ 2610.237012] NIP [c0000000009f0bb8] btrfs_run_delayed_refs+0x108/0x330
	[ 2610.237111] LR [c0000000009f0bb4] btrfs_run_delayed_refs+0x104/0x330
	[ 2610.237213] Call Trace:
	[ 2610.237254] [c000000078fb75c0] [c0000000009f0bb4] btrfs_run_delayed_refs+0x104/0x330 (unreliable)
	[ 2610.237404] [c000000078fb7680] [c000000000a0c4e4] btrfs_commit_transaction+0xf4/0x1330
	[ 2610.237533] [c000000078fb7770] [c000000000a96e74] prepare_to_relocate+0x104/0x140
	[ 2610.237657] [c000000078fb77a0] [c000000000a9d524] relocate_block_group+0x74/0x5f0
	[ 2610.237770] [c000000078fb7860] [c000000000a9dcd0] btrfs_relocate_block_group+0x230/0x4a0
	[ 2610.237886] [c000000078fb7920] [c000000000a56a60] btrfs_relocate_chunk+0x80/0x1c0
	[ 2610.238002] [c000000078fb79a0] [c000000000a57d3c] btrfs_balance+0x103c/0x1560
	[ 2610.238117] [c000000078fb7b30] [c000000000a6ab98] btrfs_ioctl_balance+0x2d8/0x450
	[ 2610.238232] [c000000078fb7b90] [c000000000a6eb90] btrfs_ioctl+0x1df0/0x3be0
	[ 2610.238328] [c000000078fb7d10] [c00000000058e048] sys_ioctl+0xa8/0x120
	[ 2610.238435] [c000000078fb7d60] [c000000000036524] system_call_exception+0x3d4/0x410
	[ 2610.238551] [c000000078fb7e10] [c00000000000d45c] system_call_common+0xec/0x278
	[ 2610.238666] --- interrupt: c00 at 0x7ffff7be2990

3. There were other failues, but I guess those were either due to mount failure
   or since no compression support yet for subpage blocksize.

-ritesh

Re: [PATCH v4 00/10] btrfs: defrag: rework to support sector perfect defrag

[Qu Wenruo]

On 2021/6/23 下午2:56, riteshh wrote:
> On 21/06/10 01:09PM, Qu Wenruo wrote:
[...]
>>
>> Thus I'm wondering if the check timing is even correct for the existing
>> code.
>> But at least I can't reproduce the nbytes problem on x86_64, thus no yet
>> sure what the real cause is.
> 
> Hi Qu,
> 
> Sorry for getting back late on this. I was using my test machines for other
> tests hence this got delayed.

No need to hurry, this branch is mostly in archive mode, no active 
development working on that yet.

But still, thank you for the testing.

> 
> I did test this whole patch series from your github https://github.com/adam900710/linux/commits/defrag_refactor.
> I didn't find any crashes (as you also mentioned). But some tests did fail e.g. btrfs/072.
> 
> Is this(btrfs/072) what you are referring to in above (random test failues with subpage
> sectorsize)?

Exactly the same thing.

> 
> Please do let me know if you would like me to try any of the fstest in loop for
> reproducing any error. Or if there is any other tests that you would like me to
> run?
> 
> 1. output from btrfs/072 test.
> 	_check_btrfs_filesystem: filesystem on /dev/vdc is inconsistent
> 	*** fsck.btrfs output ***
> 	[1/7] checking root items
> 	[2/7] checking extents
> 	[3/7] checking free space cache
> 	btrfs: csum mismatch on free space cache
> 	failed to load free space cache for block group 22020096
> 	btrfs: csum mismatch on free space cache
> 	failed to load free space cache for block group 1095761920
> 	btrfs: csum mismatch on free space cache
> 	failed to load free space cache for block group 1364197376
> 	[4/7] checking fs roots
> 	root 5 inode 323 errors 400, nbytes wrong

Yep, and I even know how to solve it if you really want.

The diff should solve it pretty well:

diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index edf597ad515f..40d6a46e43f3 100644
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -1235,8 +1235,9 @@ static int defrag_one_locked_target(struct 
btrfs_inode *inode,
         clear_extent_bit(&inode->io_tree, start, start + len - 1,
                          EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
                          EXTENT_DEFRAG, 0, 0, cached_state);
-       set_extent_defrag(&inode->io_tree, start, start + len - 1,
-                         cached_state);
+       ret = btrfs_set_extent_delalloc(inode, start, start + len - 1,
+                                       EXTENT_DEFRAG, cached_state);
+       ASSERT(!ret);

         /* Update the page status */
         for (i = start_index - first_index; i <= last_index - first_index;

But I don't know why the timing calling set_extent_defrag() is not safe 
enough.


Currently the branch is not a high priority thing, thus feel free to 
ignore the branch and above diff for now.


> 	ERROR: errors found in fs roots
> 	Opening filesystem to check...
> 	Checking filesystem on /dev/vdc
> 	UUID: b638db59-981c-4502-831e-038930e65cf4
> 	found 28282880 bytes used, error(s) found
> 	total csum bytes: 15196
> 	total tree bytes: 442368
> 	total fs tree bytes: 331776
> 	total extent tree bytes: 36864
> 	btree space waste bytes: 163191
> 	file data blocks allocated: 37101568
> 	 referenced 27430912
> 	*** end fsck.btrfs output
> 
> 2. btrfs/124 (- I guess this we have seen in past w/o your changes too).
> 	[ 2565.969989] BTRFS info (device vdc): balance: start -d -m -s
> 	[ 2565.970727] BTRFS info (device vdc): relocating block group 1513488384 flags data
> 	[ 2570.693594] BTRFS error (device vdc): bad tree block start, want 31821824 have 0
> 	[ 2570.725804] BTRFS info (device vdc): read error corrected: ino 0 off 31821824 (dev /dev/vdi sector 21192)
> 	[ 2581.579592] BTRFS info (device vdc): found 5 extents, stage: move data extents
> 	[ 2581.818799] BTRFS info (device vdc): found 5 extents, stage: update data pointers
> 	[ 2582.086796] BTRFS info (device vdc): relocating block group 976617472 flags data
> 	[ 2598.759264] BTRFS info (device vdc): found 6 extents, stage: move data extents
> 	[ 2598.991059] BTRFS info (device vdc): found 6 extents, stage: update data pointers
> 	[ 2599.274253] BTRFS info (device vdc): relocating block group 943063040 flags system
> 	[ 2599.525191] BTRFS info (device vdc): relocating block group 674627584 flags metadata
> 	[ 2599.788734] BTRFS info (device vdc): relocating block group 298844160 flags data|raid1
> 	[ 2609.746593] BTRFS info (device vdc): found 4 extents, stage: move data extents
> 	[ 2609.989663] BTRFS info (device vdc): found 3 extents, stage: update data pointers
> 	[ 2610.234642] BTRFS info (device vdc): relocating block group 30408704 flags metadata|raid1
> 	[ 2610.234889] ------------[ cut here ]------------
> 	[ 2610.234970] BTRFS: Transaction aborted (error -28)

Yep, I can also see this on our Power8 VM, without my patches.

> 	[ 2610.235172] WARNING: CPU: 3 PID: 26186 at fs/btrfs/extent-tree.c:2163 btrfs_run_delayed_refs+0x108/0x330
> 	[ 2610.235364] Modules linked in:
> 	[ 2610.235426] CPU: 3 PID: 26186 Comm: btrfs Not tainted 5.13.0-rc2-00393-gc0e4bc6f271f #43
> 	[ 2610.235547] NIP:  c0000000009f0bb8 LR: c0000000009f0bb4 CTR: c000000000e5fb30
> 	[ 2610.235665] REGS: c000000078fb7320 TRAP: 0700   Not tainted  (5.13.0-rc2-00393-gc0e4bc6f271f)
> 	[ 2610.235802] MSR:  800000000282b033 &lt;SF,VEC,VSX,EE,FP,ME,IR,DR,RI,LE&gt;  CR: 48004222  XER: 20000000
> 	[ 2610.235961] CFAR: c0000000001cea40 IRQMASK: 0
> 	               GPR00: c0000000009f0bb4 c000000078fb75c0 c000000001c5dc00 0000000000000026
> 	               GPR04: 00000000fffff5ff c000000078fb7280 0000000000000027 c0000000ff507e98
> 	               GPR08: 0000000000000023 0000000000000000 c00000000c286580 c000000001a11050
> 	               GPR12: 0000000000004400 c00000003ffeae00 0000000000000002 c000000045e4e088
> 	               GPR16: c000000045e4e000 c000000001334eb0 c000000012654000 0000000001d00000
> 	               GPR20: c000000012654000 0000000000000e5c 0000000000000000 c000000001662dc8
> 	               GPR24: 0000000000000014 c000000009ec9800 c000000009ec9b48 c0000000410f6e08
> 	               GPR28: 0000000000000000 0000000000000001 c000000012654000 ffffffffffffffe4
> 	[ 2610.237012] NIP [c0000000009f0bb8] btrfs_run_delayed_refs+0x108/0x330
> 	[ 2610.237111] LR [c0000000009f0bb4] btrfs_run_delayed_refs+0x104/0x330
> 	[ 2610.237213] Call Trace:
> 	[ 2610.237254] [c000000078fb75c0] [c0000000009f0bb4] btrfs_run_delayed_refs+0x104/0x330 (unreliable)
> 	[ 2610.237404] [c000000078fb7680] [c000000000a0c4e4] btrfs_commit_transaction+0xf4/0x1330
> 	[ 2610.237533] [c000000078fb7770] [c000000000a96e74] prepare_to_relocate+0x104/0x140
> 	[ 2610.237657] [c000000078fb77a0] [c000000000a9d524] relocate_block_group+0x74/0x5f0
> 	[ 2610.237770] [c000000078fb7860] [c000000000a9dcd0] btrfs_relocate_block_group+0x230/0x4a0
> 	[ 2610.237886] [c000000078fb7920] [c000000000a56a60] btrfs_relocate_chunk+0x80/0x1c0
> 	[ 2610.238002] [c000000078fb79a0] [c000000000a57d3c] btrfs_balance+0x103c/0x1560
> 	[ 2610.238117] [c000000078fb7b30] [c000000000a6ab98] btrfs_ioctl_balance+0x2d8/0x450
> 	[ 2610.238232] [c000000078fb7b90] [c000000000a6eb90] btrfs_ioctl+0x1df0/0x3be0
> 	[ 2610.238328] [c000000078fb7d10] [c00000000058e048] sys_ioctl+0xa8/0x120
> 	[ 2610.238435] [c000000078fb7d60] [c000000000036524] system_call_exception+0x3d4/0x410
> 	[ 2610.238551] [c000000078fb7e10] [c00000000000d45c] system_call_common+0xec/0x278
> 	[ 2610.238666] --- interrupt: c00 at 0x7ffff7be2990
> 
> 3. There were other failues, but I guess those were either due to mount failure
>     or since no compression support yet for subpage blocksize.

Yep, RAID56 mount/convert failures, and compression related output mismatch.

Thanks,
Qu
> 
> -ritesh
> 

Re: [PATCH v4 02/10] btrfs: defrag: extract the page preparation code into one helper

[David Sterba]

On Thu, Jun 10, 2021 at 01:09:09PM +0800, Qu Wenruo wrote:
> In cluster_pages_for_defrag(), we have complex code block inside one
> for() loop.
> 
> The code block is to prepare one page for defrag, this will ensure:
> - The page is locked and set up properly
> - No ordered extent exists in the page range
> - The page is uptodate
> - The writeback has finished
> 
> This behavior is pretty common and will be reused by later defrag
> rework.
> 
> So extract the code into its own helper, defrag_prepare_one_page(), for
> later usage, and cleanup the code by a little.
> 
> Since we're here, also make the page check to be subpage compatible,
> which means we will also check page::private, not only page->mapping.
> 
> Signed-off-by: Qu Wenruo <wqu@suse.com>
> ---
>  fs/btrfs/ioctl.c | 151 +++++++++++++++++++++++++++--------------------
>  1 file changed, 86 insertions(+), 65 deletions(-)
> 
> diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
> index 05af6f5ff3ff..a06ceb8fdf28 100644
> --- a/fs/btrfs/ioctl.c
> +++ b/fs/btrfs/ioctl.c
> @@ -1143,6 +1143,89 @@ static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
>  	return ret;
>  }
>  
> +/*
> + * Prepare one page to be defragged.
> + *
> + * This will ensure:
> + * - Returned page is locked and has been set up properly
> + * - No ordered extent exists in the page
> + * - The page is uptodate
> + * - The writeback has finished
> + */
> +static struct page *defrag_prepare_one_page(struct btrfs_inode *inode,
> +					    unsigned long index)
> +{
> +	struct address_space *mapping = inode->vfs_inode.i_mapping;
> +	gfp_t mask = btrfs_alloc_write_mask(mapping);
> +	u64 page_start = index << PAGE_SHIFT;

page index must be either pgoff_t or u64

Re: [PATCH v4 02/10] btrfs: defrag: extract the page preparation code into one helper

[David Sterba]

On Thu, Jun 10, 2021 at 01:09:09PM +0800, Qu Wenruo wrote:
> +static struct page *defrag_prepare_one_page(struct btrfs_inode *inode,
> +					    unsigned long index)
> +{
> +	struct address_space *mapping = inode->vfs_inode.i_mapping;
> +	gfp_t mask = btrfs_alloc_write_mask(mapping);
> +	u64 page_start = index << PAGE_SHIFT;
> +	u64 page_end = page_start + PAGE_SIZE - 1;
> +	struct extent_state *cached_state = NULL;
> +	struct page *page;
> +	int ret;
> +
> +again:
> +	page = find_or_create_page(mapping, index, mask);
> +	if (!page)
> +		return ERR_PTR(-ENOMEM);
> +
> +	ret = set_page_extent_mapped(page);
> +	if (ret < 0) {
> +		unlock_page(page);
> +		put_page(page);
> +		return ERR_PTR(ret);
> +	}
> +
> +	/* Wait for any exists ordered extent in the range */
> +	while (1) {
> +		struct btrfs_ordered_extent *ordered;
> +
> +		lock_extent_bits(&inode->io_tree, page_start, page_end,
> +				 &cached_state);
> +		ordered = btrfs_lookup_ordered_range(inode, page_start,
> +						     PAGE_SIZE);
> +		unlock_extent_cached(&inode->io_tree, page_start, page_end,
> +				     &cached_state);
> +		if (!ordered)
> +			break;
> +
> +		unlock_page(page);
> +		btrfs_start_ordered_extent(ordered, 1);
> +		btrfs_put_ordered_extent(ordered);
> +		lock_page(page);
> +		/*
> +		 * we unlocked the page above, so we need check if
> +		 * it was released or not.
> +		 */
> +		if (page->mapping != mapping || !PagePrivate(page)) {
> +			unlock_page(page);
> +			put_page(page);
> +			goto again;
> +		}
> +	}
> +
> +	/*
> +	 * Now the page range has no ordered extent any more.
> +	 * Read the page to make it uptodate.
> +	 */
> +	if (!PageUptodate(page)) {
> +		btrfs_readpage(NULL, page);
> +		lock_page(page);
> +		if (page->mapping != mapping || !PagePrivate(page)) {
> +			unlock_page(page);
> +			put_page(page);
> +			goto again;
> +		}
> +		if (!PageUptodate(page)) {
> +			unlock_page(page);
> +			put_page(page);
> +			return ERR_PTR(-EIO);
> +		}
> +	}
> +
> +	wait_on_page_writeback(page);

Wait for one page

> +	return page;
> +}
> +
>  /*
>   * it doesn't do much good to defrag one or two pages
>   * at a time.  This pulls in a nice chunk of pages
> @@ -1170,11 +1253,8 @@ static int cluster_pages_for_defrag(struct inode *inode,
>  	int ret;
>  	int i;
>  	int i_done;
> -	struct btrfs_ordered_extent *ordered;
>  	struct extent_state *cached_state = NULL;
> -	struct extent_io_tree *tree;
>  	struct extent_changeset *data_reserved = NULL;
> -	gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
>  
>  	file_end = (isize - 1) >> PAGE_SHIFT;
>  	if (!isize || start_index > file_end)
> @@ -1187,68 +1267,16 @@ static int cluster_pages_for_defrag(struct inode *inode,
>  	if (ret)
>  		return ret;
>  	i_done = 0;
> -	tree = &BTRFS_I(inode)->io_tree;
>  
>  	/* step one, lock all the pages */
>  	for (i = 0; i < page_cnt; i++) {
>  		struct page *page;
> -again:
> -		page = find_or_create_page(inode->i_mapping,
> -					   start_index + i, mask);
> -		if (!page)
> -			break;
>  
> -		ret = set_page_extent_mapped(page);
> -		if (ret < 0) {
> -			unlock_page(page);
> -			put_page(page);
> +		page = defrag_prepare_one_page(BTRFS_I(inode), start_index + i);

The loop goes one page a time

> +		if (IS_ERR(page)) {
> +			ret = PTR_ERR(page);
>  			break;
>  		}
> -
> -		page_start = page_offset(page);
> -		page_end = page_start + PAGE_SIZE - 1;
> -		while (1) {
> -			lock_extent_bits(tree, page_start, page_end,
> -					 &cached_state);
> -			ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode),
> -							      page_start);
> -			unlock_extent_cached(tree, page_start, page_end,
> -					     &cached_state);
> -			if (!ordered)
> -				break;
> -
> -			unlock_page(page);
> -			btrfs_start_ordered_extent(ordered, 1);
> -			btrfs_put_ordered_extent(ordered);
> -			lock_page(page);
> -			/*
> -			 * we unlocked the page above, so we need check if
> -			 * it was released or not.
> -			 */
> -			if (page->mapping != inode->i_mapping) {
> -				unlock_page(page);
> -				put_page(page);
> -				goto again;
> -			}
> -		}
> -
> -		if (!PageUptodate(page)) {
> -			btrfs_readpage(NULL, page);
> -			lock_page(page);
> -			if (!PageUptodate(page)) {
> -				unlock_page(page);
> -				put_page(page);
> -				ret = -EIO;
> -				break;
> -			}
> -		}
> -
> -		if (page->mapping != inode->i_mapping) {
> -			unlock_page(page);
> -			put_page(page);
> -			goto again;
> -		}
> -
>  		pages[i] = page;
>  		i_done++;
>  	}
> @@ -1258,13 +1286,6 @@ static int cluster_pages_for_defrag(struct inode *inode,
>  	if (!(inode->i_sb->s_flags & SB_ACTIVE))
>  		goto out;
>  
> -	/*
> -	 * so now we have a nice long stream of locked
> -	 * and up to date pages, lets wait on them
> -	 */
> -	for (i = 0; i < i_done; i++)
> -		wait_on_page_writeback(pages[i]);

So here it's making a big difference: originally, lots of pages have
been started and then waited for, while you change it to start+wait for
each page, that does not seem right.

> -
>  	page_start = page_offset(pages[0]);
>  	page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
>  
> -- 
> 2.32.0

Re: [PATCH v4 02/10] btrfs: defrag: extract the page preparation code into one helper

[David Sterba]

On Thu, Jun 10, 2021 at 01:09:09PM +0800, Qu Wenruo wrote:
> In cluster_pages_for_defrag(), we have complex code block inside one
> for() loop.
> 
> The code block is to prepare one page for defrag, this will ensure:
> - The page is locked and set up properly
> - No ordered extent exists in the page range
> - The page is uptodate
> - The writeback has finished
> 
> This behavior is pretty common and will be reused by later defrag
> rework.
> 
> So extract the code into its own helper, defrag_prepare_one_page(), for
> later usage, and cleanup the code by a little.
> 
> Since we're here, also make the page check to be subpage compatible,
> which means we will also check page::private, not only page->mapping.

Please split this change, it's unnoticeable in the refactoring.

Re: [PATCH v4 02/10] btrfs: defrag: extract the page preparation code into one helper

[Qu Wenruo]

On 2021/7/2 上午12:58, David Sterba wrote:
> On Thu, Jun 10, 2021 at 01:09:09PM +0800, Qu Wenruo wrote:
>> In cluster_pages_for_defrag(), we have complex code block inside one
>> for() loop.
>>
>> The code block is to prepare one page for defrag, this will ensure:
>> - The page is locked and set up properly
>> - No ordered extent exists in the page range
>> - The page is uptodate
>> - The writeback has finished
>>
>> This behavior is pretty common and will be reused by later defrag
>> rework.
>>
>> So extract the code into its own helper, defrag_prepare_one_page(), for
>> later usage, and cleanup the code by a little.
>>
>> Since we're here, also make the page check to be subpage compatible,
>> which means we will also check page::private, not only page->mapping.
> 
> Please split this change, it's unnoticeable in the refactoring.
> 
Sure, but the defrag patchset is now in very low priority mode, thus it 
will be a long time before you see the next update...

(And I hope I didn't forgot those comments before next update)

Thanks,
Qu