[PATCH v2 0/3] btrfs-progs: mkfs: make sure we can clean up all temporary chunks

[PATCH v2 0/3] btrfs-progs: mkfs: make sure we can clean up all temporary chunks

[Qu Wenruo]

There is a bug report that with certain mkfs options, mkfs.btrfs may
fail to cleanup some temporary chunks, leading to "btrfs filesystem df"
warning about multiple profiles:

  WARNING: Multiple block group profiles detected, see 'man btrfs(5)'.
  WARNING:   Metadata: single, raid1 

The easiest way to reproduce is "mkfs.btrfs -f -R free-space-tree -m dup
-d dup".

It turns out that, the old _recow_root() can not handle tree levels > 0,
while with newer free space tree creation timing, the free space tree
can reach level 1 or higher.

To fix the problem, Patch 2 will do the proper full tree re-CoW, with
extra transaction commitment to make sure all free space tree get
re-CoWed.

The 3rd patch will do the extra verification during mkfs-tests.

The first patch is just to fix a confusing parameter which also caused
u64 -> int width reduction and can be problematic in the future.

Changelog:
v2:
- Remove a duplicated recow_roots() call in create_raid_groups()
  This call makes no difference as we will later commit transaction
  and manually call recow_roots() again.
  Remove such duplicated call to save some time.

- Replace the btrfs_next_sibling_tree_block() with btrfs_next_leaf()
  Since we're always handling leaves, there is no need for
  btrfs_next_sibling_tree_block()

- Work around a kernel bug which may cause false alerts
  For single device RAID0, btrfs kernel is not respecting it, and will
  allocate new chunks using SINGLE instead.
  This can be very noisy and cause false alerts, and not always
  reproducible, depending on how fast kernel creates new chunks.

  Work around it by mounting the RO before calling "btrfs fi df".

  The kernel bug needs to be investigated and fixed.


Qu Wenruo (3):
  btrfs-progs: rename @data parameter to @profile in extent allocation
    path
  btrfs-progs: mkfs: recow all tree blocks properly
  btrfs-progs: mfks-tests: make sure mkfs.btrfs cleans up temporary
    chunks

 kernel-shared/extent-tree.c                 | 26 +++---
 mkfs/main.c                                 | 90 ++++++++++++++++++---
 tests/mkfs-tests/001-basic-profiles/test.sh | 16 +++-
 3 files changed, 104 insertions(+), 28 deletions(-)

-- 
2.33.0

[PATCH v2 1/3] btrfs-progs: rename @data parameter to @profile in extent allocation path

[Qu Wenruo]

In function btrfs_reserve_extent(), we call find_free_extent() passing
"u64 profile" into "int data".

This is definitely a width reduction, but when looking further into the
code, it's more serious than that, in fact the "int data" parameter is
not really to indicate whether it's data extent, but really a block
group profile (with block group type).

This is not only width reduction, but also confusing.

Thankfully so for we don't have any BLOCK_GROUP bits beyond 32 bits, so
the width reduction is not causing a big problem.

This patch will rename the "int data" parameter to a more proper one,
"u64 profile" in all involved call paths.

Signed-off-by: Qu Wenruo <wqu@suse.com>
---
 kernel-shared/extent-tree.c | 26 +++++++++++++-------------
 1 file changed, 13 insertions(+), 13 deletions(-)

diff --git a/kernel-shared/extent-tree.c b/kernel-shared/extent-tree.c
index 9c6d17a52a24..8e0614e033fa 100644
--- a/kernel-shared/extent-tree.c
+++ b/kernel-shared/extent-tree.c
@@ -54,7 +54,7 @@ static int __free_extent(struct btrfs_trans_handle *trans,
 			 u64 owner_offset, int refs_to_drop);
 static struct btrfs_block_group *
 btrfs_find_block_group(struct btrfs_root *root, struct btrfs_block_group
-		       *hint, u64 search_start, int data, int owner);
+		       *hint, u64 search_start, u64 profile, int owner);
 
 static int remove_sb_from_cache(struct btrfs_root *root,
 				struct btrfs_block_group *cache)
@@ -264,7 +264,7 @@ static int block_group_bits(struct btrfs_block_group *cache, u64 bits)
 
 static int noinline find_search_start(struct btrfs_root *root,
 			      struct btrfs_block_group **cache_ret,
-			      u64 *start_ret, int num, int data)
+			      u64 *start_ret, int num, u64 profile)
 {
 	int ret;
 	struct btrfs_block_group *cache = *cache_ret;
@@ -282,7 +282,7 @@ again:
 		goto out;
 
 	last = max(search_start, cache->start);
-	if (cache->ro || !block_group_bits(cache, data))
+	if (cache->ro || !block_group_bits(cache, profile))
 		goto new_group;
 
 	if (btrfs_is_zoned(root->fs_info)) {
@@ -339,7 +339,7 @@ wrapped:
 
 static struct btrfs_block_group *
 btrfs_find_block_group(struct btrfs_root *root, struct btrfs_block_group
-		       *hint, u64 search_start, int data, int owner)
+		       *hint, u64 search_start, u64 profile, int owner)
 {
 	struct btrfs_block_group *cache;
 	struct btrfs_block_group *found_group = NULL;
@@ -357,7 +357,7 @@ btrfs_find_block_group(struct btrfs_root *root, struct btrfs_block_group
 	if (search_start) {
 		struct btrfs_block_group *shint;
 		shint = btrfs_lookup_block_group(info, search_start);
-		if (shint && !shint->ro && block_group_bits(shint, data)) {
+		if (shint && !shint->ro && block_group_bits(shint, profile)) {
 			used = shint->used;
 			if (used + shint->pinned <
 			    div_factor(shint->length, factor)) {
@@ -365,7 +365,7 @@ btrfs_find_block_group(struct btrfs_root *root, struct btrfs_block_group
 			}
 		}
 	}
-	if (hint && !hint->ro && block_group_bits(hint, data)) {
+	if (hint && !hint->ro && block_group_bits(hint, profile)) {
 		used = hint->used;
 		if (used + hint->pinned <
 		    div_factor(hint->length, factor)) {
@@ -390,7 +390,7 @@ again:
 		last = cache->start + cache->length;
 		used = cache->used;
 
-		if (!cache->ro && block_group_bits(cache, data)) {
+		if (!cache->ro && block_group_bits(cache, profile)) {
 			if (full_search)
 				free_check = cache->length;
 			else
@@ -2177,7 +2177,7 @@ static int noinline find_free_extent(struct btrfs_trans_handle *trans,
 				     u64 search_start, u64 search_end,
 				     u64 hint_byte, struct btrfs_key *ins,
 				     u64 exclude_start, u64 exclude_nr,
-				     int data)
+				     u64 profile)
 {
 	int ret;
 	u64 orig_search_start = search_start;
@@ -2198,11 +2198,11 @@ static int noinline find_free_extent(struct btrfs_trans_handle *trans,
 		if (!block_group)
 			hint_byte = search_start;
 		block_group = btrfs_find_block_group(root, block_group,
-						     hint_byte, data, 1);
+						     hint_byte, profile, 1);
 	} else {
 		block_group = btrfs_find_block_group(root,
 						     trans->block_group,
-						     search_start, data, 1);
+						     search_start, profile, 1);
 	}
 
 	total_needed += empty_size;
@@ -2217,7 +2217,7 @@ check_failed:
 						       orig_search_start);
 	}
 	ret = find_search_start(root, &block_group, &search_start,
-				total_needed, data);
+				total_needed, profile);
 	if (ret)
 		goto new_group;
 
@@ -2255,7 +2255,7 @@ check_failed:
 		goto new_group;
 	}
 
-	if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
+	if (!(profile & BTRFS_BLOCK_GROUP_DATA)) {
 		if (check_crossing_stripes(info, ins->objectid, num_bytes)) {
 			struct btrfs_block_group *bg_cache;
 			u64 bg_offset;
@@ -2295,7 +2295,7 @@ new_group:
 	}
 	cond_resched();
 	block_group = btrfs_find_block_group(root, block_group,
-					     search_start, data, 0);
+					     search_start, profile, 0);
 	goto check_failed;
 
 error:
-- 
2.33.0

[PATCH v2 2/3] btrfs-progs: mkfs: recow all tree blocks properly

[Qu Wenruo]

[BUG]
Since btrfs-progs v5.14, mkfs.btrfs no longer cleans up the temporary
SINGLE metadata chunks if "-R free-space-tree" is specified:

 $ mkfs.btrfs  -f -R free-space-tree -m dup -d dup /dev/test/test
 $ btrfs ins dump-tree -t chunk /dev/test/test | grep "type METADATA"
		length 8388608 owner 2 stripe_len 65536 type METADATA
		length 268435456 owner 2 stripe_len 65536 type METADATA|DUP

[CAUSE]
Since commit 4b6cf2a3eb78 ("btrfs-progs: mkfs: generate free space tree
at make_btrfs() time"), free space tree is created when the temporary
btrfs image is created.

This behavior itself has no problem at all.

The problem happens when "-m DUP -d DUP" (or other profiles) is
specified.

This makes btrfs to create extra chunks, enlarging free space tree so
that it can be as high as level 1.

During mkfs, we rely on recow_roots() to re-CoW all tree blocks to the
newly allocated chunks.

But __recow_root() can only handle tree root at level 0, as it forces
root node to be CoWed, not bothering the children leaves/nodes.

This makes part of the free space cache tree still live on the old
temporary chunks, leaving later cleanup_temp_chunks() unable to delete
temporary SINGLE chunks.

[FIX]
Rework __recow_root() to do a proper CoW of the whole tree.

But above rework is not enough, as if a free space tree block is
allocated during current transaction, but before new chunks added.
Then the reworked __recow_root() can't CoW it, as btrfs_search_slot()
won't CoW a tree block allocated in current transaction.

So this patch will also commit current transaction before calling
recow_roots(), to force us to re-cow all tree blocks.

This shouldn't be a problem, as at the time of calling, we should have
less than a dozen tree blocks, thus there won't be a performance impact.

Reported-by: FireFish5000 <firefish5000@gmail.com>
Fixes: 4b6cf2a3eb78 ("btrfs-progs: mkfs: generate free space tree at make_btrfs() time")
Signed-off-by: Qu Wenruo <wqu@suse.com>
---
 mkfs/main.c | 90 +++++++++++++++++++++++++++++++++++++++++++++--------
 1 file changed, 77 insertions(+), 13 deletions(-)

diff --git a/mkfs/main.c b/mkfs/main.c
index 11a0989be281..2e3d1bf69629 100644
--- a/mkfs/main.c
+++ b/mkfs/main.c
@@ -210,21 +210,59 @@ err:
 }
 
 static int __recow_root(struct btrfs_trans_handle *trans,
-			 struct btrfs_root *root)
+			struct btrfs_root *root)
 {
-	struct extent_buffer *tmp;
+	struct btrfs_path path;
+	struct btrfs_key key;
 	int ret;
 
-	if (trans->transid != btrfs_root_generation(&root->root_item)) {
-		extent_buffer_get(root->node);
-		ret = __btrfs_cow_block(trans, root, root->node,
-					NULL, 0, &tmp, 0, 0);
-		if (ret)
-			return ret;
-		free_extent_buffer(tmp);
-	}
+	btrfs_init_path(&path);
+	key.objectid = 0;
+	key.type = 0;
+	key.offset = 0;
 
-	return 0;
+	/* Get a path to the most-left leaves */
+	ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
+	if (ret < 0)
+		return ret;
+
+	while (true) {
+		struct btrfs_key found_key;
+
+		/*
+		 * Our parent nodes must be no newer than the leaf, thus
+		 * if the leaf is as new as the trans, no need to re-cow.
+		 */
+		if (btrfs_header_generation(path.nodes[0]) == trans->transid)
+			goto next;
+
+		/*
+		 * Grab the key of current tree block and do a CoW search to
+		 * the current tree block.
+		 */
+		btrfs_item_key_to_cpu(path.nodes[0], &key, 0);
+		btrfs_release_path(&path);
+
+		/* This will ensure this leaf and all its parent get CoWed */
+		ret = btrfs_search_slot(trans, root, &key, &path, 0, 1);
+		if (ret < 0)
+			goto out;
+		ret = 0;
+		btrfs_item_key_to_cpu(path.nodes[0], &found_key, 0);
+		ASSERT(btrfs_comp_cpu_keys(&key, &found_key) == 0);
+
+next:
+		ret = btrfs_next_leaf(root, &path);
+		if (ret < 0)
+			goto out;
+		if (ret > 0) {
+			ret = 0;
+			goto out;
+		}
+	}
+out:
+	btrfs_release_path(&path);
+	return ret;
 }
 
 static int recow_roots(struct btrfs_trans_handle *trans,
@@ -305,7 +343,7 @@ static int create_raid_groups(struct btrfs_trans_handle *trans,
 			      u64 metadata_profile, bool mixed,
 			      struct mkfs_allocation *allocation)
 {
-	int ret;
+	int ret = 0;
 
 	if (metadata_profile) {
 		u64 meta_flags = BTRFS_BLOCK_GROUP_METADATA;
@@ -332,7 +370,6 @@ static int create_raid_groups(struct btrfs_trans_handle *trans,
 		if (ret)
 			return ret;
 	}
-	ret = recow_roots(trans, root);
 
 	return ret;
 }
@@ -1479,6 +1516,33 @@ raid_groups:
 		goto out;
 	}
 
+	/*
+	 * Commit current trans so we can cow all existing tree blocks
+	 * to newly created raid groups.
+	 * As currently we use btrfs_search_slot() to CoW tree blocks in
+	 * recow_roots(), if a tree block is already modified in current trans,
+	 * it won't be re-CoWed, thus it will stay in temporary chunks.
+	 */
+	ret = btrfs_commit_transaction(trans, root);
+	if (ret) {
+		errno = -ret;
+		error("unable to commit transaction before recowing trees: %m");
+		goto out;
+	}
+	trans = btrfs_start_transaction(root, 1);
+	if (IS_ERR(trans)) {
+		errno = -PTR_ERR(trans);
+		error("failed to start transaction: %m");
+		goto error;
+	}
+	/* CoW all tree blocks to newly created chunks */
+	ret = recow_roots(trans, root);
+	if (ret) {
+		errno = -ret;
+		error("unable to CoW tree blocks to new profiles: %m");
+		goto out;
+	}
+
 	ret = create_data_reloc_tree(trans);
 	if (ret) {
 		error("unable to create data reloc tree: %d", ret);
-- 
2.33.0

[PATCH v2 3/3] btrfs-progs: mfks-tests: make sure mkfs.btrfs cleans up temporary chunks

[Qu Wenruo]

Since current "btrfs filesystem df" command will warn if there are
multiple profiles of the same type, it's a good way to detect left-over
temporary chunks.

This patch will enhance the existing mkfs-tests/001-basic-profiles test
case to also check for the warning messages, to make sure mkfs.btrfs has
properly cleaned up all temporary chunks.

There is a special workaround newly implemented in test_get_info(), as
recent kernel introduced single device RAID0 support, which is no
different than SINGLE.

But for single device RAID0, kernel may choose to preallocate new chunks
with SINGLE profile, causing false alerts.

Work around this kernel bug by mounting the btrfs read-only to prevent
preallocating new chunks.

Signed-off-by: Qu Wenruo <wqu@suse.com>
---
 tests/mkfs-tests/001-basic-profiles/test.sh | 16 ++++++++++++++--
 1 file changed, 14 insertions(+), 2 deletions(-)

diff --git a/tests/mkfs-tests/001-basic-profiles/test.sh b/tests/mkfs-tests/001-basic-profiles/test.sh
index b3ba50d71ddc..0be199749864 100755
--- a/tests/mkfs-tests/001-basic-profiles/test.sh
+++ b/tests/mkfs-tests/001-basic-profiles/test.sh
@@ -11,10 +11,22 @@ setup_root_helper
 
 test_get_info()
 {
+	tmp_out=$(mktemp --tmpdir btrfs-progs-mkfs-tests-get-info.XXXXXX)
 	run_check $SUDO_HELPER "$TOP/btrfs" inspect-internal dump-super "$dev1"
 	run_check $SUDO_HELPER "$TOP/btrfs" check "$dev1"
-	run_check $SUDO_HELPER mount "$dev1" "$TEST_MNT"
-	run_check "$TOP/btrfs" filesystem df "$TEST_MNT"
+
+	btrfs ins dump-tree -t chunk "$dev1" >> "$RESULTS"
+
+	# Work around a kernel bug that kernel will treat SINGLE and single
+	# device RAID0 as the same.
+	# Thus kernel may create new SINGLE chunks, causing extra warning
+	# when testing single device RAID0.
+	run_check $SUDO_HELPER mount -o ro "$dev1" "$TEST_MNT"
+	if grep -q "Multiple block group profiles detected" "$tmp_out"; then
+		rm -- "$tmp_out"
+		_fail "temporary chunks are not properly cleaned up"
+	fi
+	rm -- "$tmp_out"
 	run_check $SUDO_HELPER "$TOP/btrfs" filesystem usage "$TEST_MNT"
 	run_check $SUDO_HELPER "$TOP/btrfs" device usage "$TEST_MNT"
 	run_check $SUDO_HELPER umount "$TEST_MNT"
-- 
2.33.0

Re: [PATCH v2 0/3] btrfs-progs: mkfs: make sure we can clean up all temporary chunks

[Nikolay Borisov]

On 11.10.21 г. 15:06, Qu Wenruo wrote:
> There is a bug report that with certain mkfs options, mkfs.btrfs may
> fail to cleanup some temporary chunks, leading to "btrfs filesystem df"
> warning about multiple profiles:
> 
>   WARNING: Multiple block group profiles detected, see 'man btrfs(5)'.
>   WARNING:   Metadata: single, raid1 
> 
> The easiest way to reproduce is "mkfs.btrfs -f -R free-space-tree -m dup
> -d dup".
> 
> It turns out that, the old _recow_root() can not handle tree levels > 0,
> while with newer free space tree creation timing, the free space tree
> can reach level 1 or higher.
> 
> To fix the problem, Patch 2 will do the proper full tree re-CoW, with
> extra transaction commitment to make sure all free space tree get
> re-CoWed.
> 
> The 3rd patch will do the extra verification during mkfs-tests.
> 
> The first patch is just to fix a confusing parameter which also caused
> u64 -> int width reduction and can be problematic in the future.
> 
> Changelog:
> v2:
> - Remove a duplicated recow_roots() call in create_raid_groups()
>   This call makes no difference as we will later commit transaction
>   and manually call recow_roots() again.
>   Remove such duplicated call to save some time.
> 
> - Replace the btrfs_next_sibling_tree_block() with btrfs_next_leaf()
>   Since we're always handling leaves, there is no need for
>   btrfs_next_sibling_tree_block()
> 
> - Work around a kernel bug which may cause false alerts
>   For single device RAID0, btrfs kernel is not respecting it, and will
>   allocate new chunks using SINGLE instead.
>   This can be very noisy and cause false alerts, and not always
>   reproducible, depending on how fast kernel creates new chunks.
> 
>   Work around it by mounting the RO before calling "btrfs fi df".
> 
>   The kernel bug needs to be investigated and fixed.
It's better to see the kernel bug fixed rather than papering over it.

> 
> 
> Qu Wenruo (3):
>   btrfs-progs: rename @data parameter to @profile in extent allocation
>     path
>   btrfs-progs: mkfs: recow all tree blocks properly
>   btrfs-progs: mfks-tests: make sure mkfs.btrfs cleans up temporary
>     chunks
> 
>  kernel-shared/extent-tree.c                 | 26 +++---
>  mkfs/main.c                                 | 90 ++++++++++++++++++---
>  tests/mkfs-tests/001-basic-profiles/test.sh | 16 +++-
>  3 files changed, 104 insertions(+), 28 deletions(-)
> 

Re: [PATCH v2 0/3] btrfs-progs: mkfs: make sure we can clean up all temporary chunks

[Qu Wenruo]

On 2021/10/11 20:10, Nikolay Borisov wrote:
>
>
> On 11.10.21 г. 15:06, Qu Wenruo wrote:
>> There is a bug report that with certain mkfs options, mkfs.btrfs may
>> fail to cleanup some temporary chunks, leading to "btrfs filesystem df"
>> warning about multiple profiles:
>>
>>    WARNING: Multiple block group profiles detected, see 'man btrfs(5)'.
>>    WARNING:   Metadata: single, raid1
>>
>> The easiest way to reproduce is "mkfs.btrfs -f -R free-space-tree -m dup
>> -d dup".
>>
>> It turns out that, the old _recow_root() can not handle tree levels > 0,
>> while with newer free space tree creation timing, the free space tree
>> can reach level 1 or higher.
>>
>> To fix the problem, Patch 2 will do the proper full tree re-CoW, with
>> extra transaction commitment to make sure all free space tree get
>> re-CoWed.
>>
>> The 3rd patch will do the extra verification during mkfs-tests.
>>
>> The first patch is just to fix a confusing parameter which also caused
>> u64 -> int width reduction and can be problematic in the future.
>>
>> Changelog:
>> v2:
>> - Remove a duplicated recow_roots() call in create_raid_groups()
>>    This call makes no difference as we will later commit transaction
>>    and manually call recow_roots() again.
>>    Remove such duplicated call to save some time.
>>
>> - Replace the btrfs_next_sibling_tree_block() with btrfs_next_leaf()
>>    Since we're always handling leaves, there is no need for
>>    btrfs_next_sibling_tree_block()
>>
>> - Work around a kernel bug which may cause false alerts
>>    For single device RAID0, btrfs kernel is not respecting it, and will
>>    allocate new chunks using SINGLE instead.
>>    This can be very noisy and cause false alerts, and not always
>>    reproducible, depending on how fast kernel creates new chunks.
>>
>>    Work around it by mounting the RO before calling "btrfs fi df".
>>
>>    The kernel bug needs to be investigated and fixed.
> It's better to see the kernel bug fixed rather than papering over it.

That's for sure.

Just get overloaded by so many small bugs in one day.

Will investigate and fix the bug soon.

For the test case itself, mounting with RO in fact makes sense, we just
want to the initial chunk layout created by mkfs.

If later we choose to compare the total chunk size against the reported
values, such RO mount is a hard requirement to avoid chunk preallocation.

Thanks,
Qu
>
>>
>>
>> Qu Wenruo (3):
>>    btrfs-progs: rename @data parameter to @profile in extent allocation
>>      path
>>    btrfs-progs: mkfs: recow all tree blocks properly
>>    btrfs-progs: mfks-tests: make sure mkfs.btrfs cleans up temporary
>>      chunks
>>
>>   kernel-shared/extent-tree.c                 | 26 +++---
>>   mkfs/main.c                                 | 90 ++++++++++++++++++---
>>   tests/mkfs-tests/001-basic-profiles/test.sh | 16 +++-
>>   3 files changed, 104 insertions(+), 28 deletions(-)
>>

Re: [PATCH v2 0/3] btrfs-progs: mkfs: make sure we can clean up all temporary chunks

[David Sterba]

On Mon, Oct 11, 2021 at 08:06:47PM +0800, Qu Wenruo wrote:
> There is a bug report that with certain mkfs options, mkfs.btrfs may
> fail to cleanup some temporary chunks, leading to "btrfs filesystem df"
> warning about multiple profiles:
> 
>   WARNING: Multiple block group profiles detected, see 'man btrfs(5)'.
>   WARNING:   Metadata: single, raid1 
> 
> The easiest way to reproduce is "mkfs.btrfs -f -R free-space-tree -m dup
> -d dup".
> 
> It turns out that, the old _recow_root() can not handle tree levels > 0,
> while with newer free space tree creation timing, the free space tree
> can reach level 1 or higher.
> 
> To fix the problem, Patch 2 will do the proper full tree re-CoW, with
> extra transaction commitment to make sure all free space tree get
> re-CoWed.

The extra commit breaks assumptions of test misc/038 that looks up the
backup roots in particular slot(s). I already had to fix it once due to
the additional commit with free space tree. Now it broke again, the test
is too fragle, I'm not sure we want to keep doing the whack-a-mole.

Re: [PATCH v2 2/3] btrfs-progs: mkfs: recow all tree blocks properly

[David Sterba]

On Mon, Oct 11, 2021 at 08:06:49PM +0800, Qu Wenruo wrote:
> [BUG]
> Since btrfs-progs v5.14, mkfs.btrfs no longer cleans up the temporary
> SINGLE metadata chunks if "-R free-space-tree" is specified:
> 
>  $ mkfs.btrfs  -f -R free-space-tree -m dup -d dup /dev/test/test
>  $ btrfs ins dump-tree -t chunk /dev/test/test | grep "type METADATA"
> 		length 8388608 owner 2 stripe_len 65536 type METADATA
> 		length 268435456 owner 2 stripe_len 65536 type METADATA|DUP
> 
> [CAUSE]
> Since commit 4b6cf2a3eb78 ("btrfs-progs: mkfs: generate free space tree
> at make_btrfs() time"), free space tree is created when the temporary
> btrfs image is created.
> 
> This behavior itself has no problem at all.
> 
> The problem happens when "-m DUP -d DUP" (or other profiles) is
> specified.
> 
> This makes btrfs to create extra chunks, enlarging free space tree so
> that it can be as high as level 1.
> 
> During mkfs, we rely on recow_roots() to re-CoW all tree blocks to the

Please spell it as COW (not CoW or cow) when it means copy-on-write.

Re: [PATCH v2 3/3] btrfs-progs: mfks-tests: make sure mkfs.btrfs cleans up temporary chunks

[David Sterba]

On Mon, Oct 11, 2021 at 08:06:50PM +0800, Qu Wenruo wrote:
> Since current "btrfs filesystem df" command will warn if there are
> multiple profiles of the same type, it's a good way to detect left-over
> temporary chunks.
> 
> This patch will enhance the existing mkfs-tests/001-basic-profiles test
> case to also check for the warning messages, to make sure mkfs.btrfs has
> properly cleaned up all temporary chunks.
> 
> There is a special workaround newly implemented in test_get_info(), as
> recent kernel introduced single device RAID0 support, which is no
> different than SINGLE.
> 
> But for single device RAID0, kernel may choose to preallocate new chunks
> with SINGLE profile, causing false alerts.
> 
> Work around this kernel bug by mounting the btrfs read-only to prevent
> preallocating new chunks.
> 
> Signed-off-by: Qu Wenruo <wqu@suse.com>
> ---
>  tests/mkfs-tests/001-basic-profiles/test.sh | 16 ++++++++++++++--
>  1 file changed, 14 insertions(+), 2 deletions(-)
> 
> diff --git a/tests/mkfs-tests/001-basic-profiles/test.sh b/tests/mkfs-tests/001-basic-profiles/test.sh
> index b3ba50d71ddc..0be199749864 100755
> --- a/tests/mkfs-tests/001-basic-profiles/test.sh
> +++ b/tests/mkfs-tests/001-basic-profiles/test.sh
> @@ -11,10 +11,22 @@ setup_root_helper
>  
>  test_get_info()
>  {
> +	tmp_out=$(mktemp --tmpdir btrfs-progs-mkfs-tests-get-info.XXXXXX)

Local variables should be declared

	local tmp_out

>  	run_check $SUDO_HELPER "$TOP/btrfs" inspect-internal dump-super "$dev1"
>  	run_check $SUDO_HELPER "$TOP/btrfs" check "$dev1"
> -	run_check $SUDO_HELPER mount "$dev1" "$TEST_MNT"
> -	run_check "$TOP/btrfs" filesystem df "$TEST_MNT"
> +
> +	btrfs ins dump-tree -t chunk "$dev1" >> "$RESULTS"

Using $RESULTS is not recommended, the same can be achieved by

	run_check btrfs...

Also you should run the "$TOP/btrfs" binary and use the full name of the
subcommands, ie. 'inspect-internal'.

> +	# Work around a kernel bug that kernel will treat SINGLE and single
> +	# device RAID0 as the same.
> +	# Thus kernel may create new SINGLE chunks, causing extra warning
> +	# when testing single device RAID0.
> +	run_check $SUDO_HELPER mount -o ro "$dev1" "$TEST_MNT"
> +	if grep -q "Multiple block group profiles detected" "$tmp_out"; then
> +		rm -- "$tmp_out"
> +		_fail "temporary chunks are not properly cleaned up"
> +	fi
> +	rm -- "$tmp_out"

So to be able to run the testsuite on unfixed kernels the workaround
makes sense.

>  	run_check $SUDO_HELPER "$TOP/btrfs" filesystem usage "$TEST_MNT"
>  	run_check $SUDO_HELPER "$TOP/btrfs" device usage "$TEST_MNT"
>  	run_check $SUDO_HELPER umount "$TEST_MNT"
> -- 
> 2.33.0

Re: [PATCH v2 0/3] btrfs-progs: mkfs: make sure we can clean up all temporary chunks

[David Sterba]

On Mon, Oct 11, 2021 at 04:05:18PM +0200, David Sterba wrote:
> On Mon, Oct 11, 2021 at 08:06:47PM +0800, Qu Wenruo wrote:
> > There is a bug report that with certain mkfs options, mkfs.btrfs may
> > fail to cleanup some temporary chunks, leading to "btrfs filesystem df"
> > warning about multiple profiles:
> > 
> >   WARNING: Multiple block group profiles detected, see 'man btrfs(5)'.
> >   WARNING:   Metadata: single, raid1 
> > 
> > The easiest way to reproduce is "mkfs.btrfs -f -R free-space-tree -m dup
> > -d dup".
> > 
> > It turns out that, the old _recow_root() can not handle tree levels > 0,
> > while with newer free space tree creation timing, the free space tree
> > can reach level 1 or higher.
> > 
> > To fix the problem, Patch 2 will do the proper full tree re-CoW, with
> > extra transaction commitment to make sure all free space tree get
> > re-CoWed.
> 
> The extra commit breaks assumptions of test misc/038 that looks up the
> backup roots in particular slot(s). I already had to fix it once due to
> the additional commit with free space tree. Now it broke again, the test
> is too fragle, I'm not sure we want to keep doing the whack-a-mole.

I've check it with the v2, no change, so I'll disable the test.

Re: [PATCH v2 0/3] btrfs-progs: mkfs: make sure we can clean up all temporary chunks

[David Sterba]

On Mon, Oct 11, 2021 at 08:06:47PM +0800, Qu Wenruo wrote:
> There is a bug report that with certain mkfs options, mkfs.btrfs may
> fail to cleanup some temporary chunks, leading to "btrfs filesystem df"
> warning about multiple profiles:
> 
>   WARNING: Multiple block group profiles detected, see 'man btrfs(5)'.
>   WARNING:   Metadata: single, raid1 
> 
> The easiest way to reproduce is "mkfs.btrfs -f -R free-space-tree -m dup
> -d dup".
> 
> It turns out that, the old _recow_root() can not handle tree levels > 0,
> while with newer free space tree creation timing, the free space tree
> can reach level 1 or higher.
> 
> To fix the problem, Patch 2 will do the proper full tree re-CoW, with
> extra transaction commitment to make sure all free space tree get
> re-CoWed.
> 
> The 3rd patch will do the extra verification during mkfs-tests.
> 
> The first patch is just to fix a confusing parameter which also caused
> u64 -> int width reduction and can be problematic in the future.
> 
> Changelog:
> v2:
> - Remove a duplicated recow_roots() call in create_raid_groups()
>   This call makes no difference as we will later commit transaction
>   and manually call recow_roots() again.
>   Remove such duplicated call to save some time.
> 
> - Replace the btrfs_next_sibling_tree_block() with btrfs_next_leaf()
>   Since we're always handling leaves, there is no need for
>   btrfs_next_sibling_tree_block()
> 
> - Work around a kernel bug which may cause false alerts
>   For single device RAID0, btrfs kernel is not respecting it, and will
>   allocate new chunks using SINGLE instead.
>   This can be very noisy and cause false alerts, and not always
>   reproducible, depending on how fast kernel creates new chunks.
> 
>   Work around it by mounting the RO before calling "btrfs fi df".
> 
>   The kernel bug needs to be investigated and fixed.
> 
> 
> Qu Wenruo (3):
>   btrfs-progs: rename @data parameter to @profile in extent allocation
>     path
>   btrfs-progs: mkfs: recow all tree blocks properly
>   btrfs-progs: mfks-tests: make sure mkfs.btrfs cleans up temporary
>     chunks

I've replaced patches 2 and 3, thanks. No need to resend with the fixes
I mentioned, that's for future patches.

Re: [PATCH v2 3/3] btrfs-progs: mfks-tests: make sure mkfs.btrfs cleans up temporary chunks

[Qu Wenruo]

On 2021/10/11 22:38, David Sterba wrote:
> On Mon, Oct 11, 2021 at 08:06:50PM +0800, Qu Wenruo wrote:
>> Since current "btrfs filesystem df" command will warn if there are
>> multiple profiles of the same type, it's a good way to detect left-over
>> temporary chunks.
>>
>> This patch will enhance the existing mkfs-tests/001-basic-profiles test
>> case to also check for the warning messages, to make sure mkfs.btrfs has
>> properly cleaned up all temporary chunks.
>>
>> There is a special workaround newly implemented in test_get_info(), as
>> recent kernel introduced single device RAID0 support, which is no
>> different than SINGLE.
>>
>> But for single device RAID0, kernel may choose to preallocate new chunks
>> with SINGLE profile, causing false alerts.
>>
>> Work around this kernel bug by mounting the btrfs read-only to prevent
>> preallocating new chunks.
>>
>> Signed-off-by: Qu Wenruo <wqu@suse.com>
>> ---
>>   tests/mkfs-tests/001-basic-profiles/test.sh | 16 ++++++++++++++--
>>   1 file changed, 14 insertions(+), 2 deletions(-)
>>
>> diff --git a/tests/mkfs-tests/001-basic-profiles/test.sh b/tests/mkfs-tests/001-basic-profiles/test.sh
>> index b3ba50d71ddc..0be199749864 100755
>> --- a/tests/mkfs-tests/001-basic-profiles/test.sh
>> +++ b/tests/mkfs-tests/001-basic-profiles/test.sh
>> @@ -11,10 +11,22 @@ setup_root_helper
>>   
>>   test_get_info()
>>   {
>> +	tmp_out=$(mktemp --tmpdir btrfs-progs-mkfs-tests-get-info.XXXXXX)
> 
> Local variables should be declared
> 
> 	local tmp_out
> 
>>   	run_check $SUDO_HELPER "$TOP/btrfs" inspect-internal dump-super "$dev1"
>>   	run_check $SUDO_HELPER "$TOP/btrfs" check "$dev1"
>> -	run_check $SUDO_HELPER mount "$dev1" "$TEST_MNT"
>> -	run_check "$TOP/btrfs" filesystem df "$TEST_MNT"
>> +
>> +	btrfs ins dump-tree -t chunk "$dev1" >> "$RESULTS"
> 
> Using $RESULTS is not recommended, the same can be achieved by
> 
> 	run_check btrfs...
> 
> Also you should run the "$TOP/btrfs" binary and use the full name of the
> subcommands, ie. 'inspect-internal'.

Oh, that's the debug output I forgot to remove.

Please remove the "ins dump-tree" call...

Thanks,
Qu

> 
>> +	# Work around a kernel bug that kernel will treat SINGLE and single
>> +	# device RAID0 as the same.
>> +	# Thus kernel may create new SINGLE chunks, causing extra warning
>> +	# when testing single device RAID0.
>> +	run_check $SUDO_HELPER mount -o ro "$dev1" "$TEST_MNT"
>> +	if grep -q "Multiple block group profiles detected" "$tmp_out"; then
>> +		rm -- "$tmp_out"
>> +		_fail "temporary chunks are not properly cleaned up"
>> +	fi
>> +	rm -- "$tmp_out"
> 
> So to be able to run the testsuite on unfixed kernels the workaround
> makes sense.
> 
>>   	run_check $SUDO_HELPER "$TOP/btrfs" filesystem usage "$TEST_MNT"
>>   	run_check $SUDO_HELPER "$TOP/btrfs" device usage "$TEST_MNT"
>>   	run_check $SUDO_HELPER umount "$TEST_MNT"
>> -- 
>> 2.33.0
> 

Re: [PATCH v2 0/3] btrfs-progs: mkfs: make sure we can clean up all temporary chunks

[Qu Wenruo]

On 2021/10/11 22:39, David Sterba wrote:
> On Mon, Oct 11, 2021 at 04:05:18PM +0200, David Sterba wrote:
>> On Mon, Oct 11, 2021 at 08:06:47PM +0800, Qu Wenruo wrote:
>>> There is a bug report that with certain mkfs options, mkfs.btrfs may
>>> fail to cleanup some temporary chunks, leading to "btrfs filesystem df"
>>> warning about multiple profiles:
>>>
>>>    WARNING: Multiple block group profiles detected, see 'man btrfs(5)'.
>>>    WARNING:   Metadata: single, raid1
>>>
>>> The easiest way to reproduce is "mkfs.btrfs -f -R free-space-tree -m dup
>>> -d dup".
>>>
>>> It turns out that, the old _recow_root() can not handle tree levels > 0,
>>> while with newer free space tree creation timing, the free space tree
>>> can reach level 1 or higher.
>>>
>>> To fix the problem, Patch 2 will do the proper full tree re-CoW, with
>>> extra transaction commitment to make sure all free space tree get
>>> re-CoWed.
>>
>> The extra commit breaks assumptions of test misc/038 that looks up the
>> backup roots in particular slot(s). I already had to fix it once due to
>> the additional commit with free space tree. Now it broke again, the test
>> is too fragle, I'm not sure we want to keep doing the whack-a-mole.
> 
> I've check it with the v2, no change, so I'll disable the test.
> 

I'll update the test case to make it more flex to handle the extra commits.

The proper way to test the behavior is not to use the hardcoded slot 
number, but auto-detect the slot.

Thanks,
Qu

Re: [PATCH v2 0/3] btrfs-progs: mkfs: make sure we can clean up all temporary chunks

[Qu Wenruo]

On 2021/10/11 20:14, Qu Wenruo wrote:
>
>
> On 2021/10/11 20:10, Nikolay Borisov wrote:
>>
>>
>> On 11.10.21 г. 15:06, Qu Wenruo wrote:
>>> There is a bug report that with certain mkfs options, mkfs.btrfs may
>>> fail to cleanup some temporary chunks, leading to "btrfs filesystem df"
>>> warning about multiple profiles:
>>>
>>>    WARNING: Multiple block group profiles detected, see 'man btrfs(5)'.
>>>    WARNING:   Metadata: single, raid1
>>>
>>> The easiest way to reproduce is "mkfs.btrfs -f -R free-space-tree -m dup
>>> -d dup".
>>>
>>> It turns out that, the old _recow_root() can not handle tree levels > 0,
>>> while with newer free space tree creation timing, the free space tree
>>> can reach level 1 or higher.
>>>
>>> To fix the problem, Patch 2 will do the proper full tree re-CoW, with
>>> extra transaction commitment to make sure all free space tree get
>>> re-CoWed.
>>>
>>> The 3rd patch will do the extra verification during mkfs-tests.
>>>
>>> The first patch is just to fix a confusing parameter which also caused
>>> u64 -> int width reduction and can be problematic in the future.
>>>
>>> Changelog:
>>> v2:
>>> - Remove a duplicated recow_roots() call in create_raid_groups()
>>>    This call makes no difference as we will later commit transaction
>>>    and manually call recow_roots() again.
>>>    Remove such duplicated call to save some time.
>>>
>>> - Replace the btrfs_next_sibling_tree_block() with btrfs_next_leaf()
>>>    Since we're always handling leaves, there is no need for
>>>    btrfs_next_sibling_tree_block()
>>>
>>> - Work around a kernel bug which may cause false alerts
>>>    For single device RAID0, btrfs kernel is not respecting it, and will
>>>    allocate new chunks using SINGLE instead.
>>>    This can be very noisy and cause false alerts, and not always
>>>    reproducible, depending on how fast kernel creates new chunks.
>>>
>>>    Work around it by mounting the RO before calling "btrfs fi df".
>>>
>>>    The kernel bug needs to be investigated and fixed.
>> It's better to see the kernel bug fixed rather than papering over it.

The truth is, this is more like a kernel behavior change.

Before commit b2f78e88052b ("btrfs: allow degenerate raid0/raid10"),
kernel can only create RAID0 chunks with at least two devices.

Thus older kernel (when tested under my host, it's still v5.14) will
create SINGLE chunk as it has no other choice.

So false alert.

Thanks,
Qu

>
> That's for sure.
>
> Just get overloaded by so many small bugs in one day.
>
> Will investigate and fix the bug soon.
>
> For the test case itself, mounting with RO in fact makes sense, we just
> want to the initial chunk layout created by mkfs.
>
> If later we choose to compare the total chunk size against the reported
> values, such RO mount is a hard requirement to avoid chunk preallocation.
>
> Thanks,
> Qu
>>
>>>
>>>
>>> Qu Wenruo (3):
>>>    btrfs-progs: rename @data parameter to @profile in extent allocation
>>>      path
>>>    btrfs-progs: mkfs: recow all tree blocks properly
>>>    btrfs-progs: mfks-tests: make sure mkfs.btrfs cleans up temporary
>>>      chunks
>>>
>>>   kernel-shared/extent-tree.c                 | 26 +++---
>>>   mkfs/main.c                                 | 90 ++++++++++++++++++---
>>>   tests/mkfs-tests/001-basic-profiles/test.sh | 16 +++-
>>>   3 files changed, 104 insertions(+), 28 deletions(-)
>>>