[PATCH] btrfs: Use default subvolume as filesystem root

[PATCH] btrfs: Use default subvolume as filesystem root

[Matwey V. Kornilov]

BTRFS volume consists of a number of subvolumes which can be mounted separately
from each other. The top-level subvolume always exists even if no subvolumes
were created manually. A subvolume can be denoted as the default subvolume i.e.
the subvolume which is mounted by default.

The default "default subvolume" is the top-level one, but this is far from the
common practices used in the wild. For instance, openSUSE provides an OS
snapshot/rollback feature based on BTRFS. To achieve this, the actual OS root
filesystem is located into a separate subvolume which is "default" but not
"top-level". That means that the /boot/dtb/ directory is also located inside
this default subvolume instead of top-level one.

However, the existing btrfs u-boot driver always uses the top-level subvolume
as the filesystem root. This behaviour 1) is inconsistent with

    mount /dev/sda1 /target

command, which mount the default subvolume 2) leads to the issues when
/boot/dtb cannot be found properly (see the reference).

This patch uses the default subvolume as the filesystem root to overcome
mentioned issues.

Reference: https://bugzilla.suse.com/show_bug.cgi?id=1185656
Signed-off-by: Matwey V. Kornilov <matwey.kornilov@gmail.com>
---
 fs/btrfs/disk-io.c | 38 +++++++++++++++++++++++++++++++++++---
 1 file changed, 35 insertions(+), 3 deletions(-)

diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 349411c3cc..12f9579fcf 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -804,6 +804,30 @@ static int setup_root_or_create_block(struct btrfs_fs_info *fs_info,
 	return 0;
 }
 
+static int get_default_subvolume(struct btrfs_fs_info *fs_info,
+				 struct btrfs_key *key_ret)
+{
+	struct btrfs_root *root = fs_info->tree_root;
+	struct btrfs_dir_item *dir_item;
+	struct btrfs_path path;
+	int ret = 0;
+
+	btrfs_init_path(&path);
+
+	dir_item = btrfs_lookup_dir_item(NULL, root, &path,
+					 BTRFS_ROOT_TREE_DIR_OBJECTID,
+					 "default", 7, 0);
+	if (IS_ERR(dir_item)) {
+		ret = PTR_ERR(dir_item);
+		goto out;
+	}
+
+	btrfs_dir_item_key_to_cpu(path.nodes[0], dir_item, key_ret);
+out:
+	btrfs_release_path(&path);
+	return ret;
+}
+
 int btrfs_setup_all_roots(struct btrfs_fs_info *fs_info)
 {
 	struct btrfs_super_block *sb = fs_info->super_copy;
@@ -833,9 +857,17 @@ int btrfs_setup_all_roots(struct btrfs_fs_info *fs_info)
 
 	fs_info->last_trans_committed = generation;
 
-	key.objectid = BTRFS_FS_TREE_OBJECTID;
-	key.type = BTRFS_ROOT_ITEM_KEY;
-	key.offset = (u64)-1;
+	ret = get_default_subvolume(fs_info, &key);
+	if (ret) {
+		/*
+		 * The default dir item isn't there. Linux kernel behaviour is
+		 * to silently use the top-level subvolume in this case.
+		 */
+		key.objectid = BTRFS_FS_TREE_OBJECTID;
+		key.type = BTRFS_ROOT_ITEM_KEY;
+		key.offset = (u64)-1;
+	}
+
 	fs_info->fs_root = btrfs_read_fs_root(fs_info, &key);
 
 	if (IS_ERR(fs_info->fs_root))
-- 
2.31.1

Re: [PATCH] btrfs: Use default subvolume as filesystem root

[Qu Wenruo]

On 2021/8/2 上午4:52, Matwey V. Kornilov wrote:
> BTRFS volume consists of a number of subvolumes which can be mounted separately
> from each other. The top-level subvolume always exists even if no subvolumes
> were created manually. A subvolume can be denoted as the default subvolume i.e.
> the subvolume which is mounted by default.
> 
> The default "default subvolume" is the top-level one, but this is far from the
> common practices used in the wild. For instance, openSUSE provides an OS
> snapshot/rollback feature based on BTRFS. To achieve this, the actual OS root
> filesystem is located into a separate subvolume which is "default" but not
> "top-level". That means that the /boot/dtb/ directory is also located inside
> this default subvolume instead of top-level one.
> 
> However, the existing btrfs u-boot driver always uses the top-level subvolume
> as the filesystem root. This behaviour 1) is inconsistent with
> 
>      mount /dev/sda1 /target
> 
> command, which mount the default subvolume 2) leads to the issues when
> /boot/dtb cannot be found properly (see the reference).

I also noticed the problem in the past, but forgot to fix it....

> 
> This patch uses the default subvolume as the filesystem root to overcome
> mentioned issues.
> 
> Reference: https://bugzilla.suse.com/show_bug.cgi?id=1185656
> Signed-off-by: Matwey V. Kornilov <matwey.kornilov@gmail.com>

Reviewed-by: Qu Wenruo <wqu@suse.com>

Thanks,
Qu

> ---
>   fs/btrfs/disk-io.c | 38 +++++++++++++++++++++++++++++++++++---
>   1 file changed, 35 insertions(+), 3 deletions(-)
> 
> diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
> index 349411c3cc..12f9579fcf 100644
> --- a/fs/btrfs/disk-io.c
> +++ b/fs/btrfs/disk-io.c
> @@ -804,6 +804,30 @@ static int setup_root_or_create_block(struct btrfs_fs_info *fs_info,
>   	return 0;
>   }
>   
> +static int get_default_subvolume(struct btrfs_fs_info *fs_info,
> +				 struct btrfs_key *key_ret)
> +{
> +	struct btrfs_root *root = fs_info->tree_root;
> +	struct btrfs_dir_item *dir_item;
> +	struct btrfs_path path;
> +	int ret = 0;
> +
> +	btrfs_init_path(&path);
> +
> +	dir_item = btrfs_lookup_dir_item(NULL, root, &path,
> +					 BTRFS_ROOT_TREE_DIR_OBJECTID,
> +					 "default", 7, 0);
> +	if (IS_ERR(dir_item)) {
> +		ret = PTR_ERR(dir_item);
> +		goto out;
> +	}
> +
> +	btrfs_dir_item_key_to_cpu(path.nodes[0], dir_item, key_ret);
> +out:
> +	btrfs_release_path(&path);
> +	return ret;
> +}
> +
>   int btrfs_setup_all_roots(struct btrfs_fs_info *fs_info)
>   {
>   	struct btrfs_super_block *sb = fs_info->super_copy;
> @@ -833,9 +857,17 @@ int btrfs_setup_all_roots(struct btrfs_fs_info *fs_info)
>   
>   	fs_info->last_trans_committed = generation;
>   
> -	key.objectid = BTRFS_FS_TREE_OBJECTID;
> -	key.type = BTRFS_ROOT_ITEM_KEY;
> -	key.offset = (u64)-1;
> +	ret = get_default_subvolume(fs_info, &key);
> +	if (ret) {
> +		/*
> +		 * The default dir item isn't there. Linux kernel behaviour is
> +		 * to silently use the top-level subvolume in this case.
> +		 */
> +		key.objectid = BTRFS_FS_TREE_OBJECTID;
> +		key.type = BTRFS_ROOT_ITEM_KEY;
> +		key.offset = (u64)-1;
> +	}
> +
>   	fs_info->fs_root = btrfs_read_fs_root(fs_info, &key);
>   
>   	if (IS_ERR(fs_info->fs_root))
> 

Re: [PATCH] btrfs: Use default subvolume as filesystem root

[Matthias Brugger]

Hi Tom,

On 02/08/2021 01:06, Qu Wenruo wrote:
> 
> 
> On 2021/8/2 上午4:52, Matwey V. Kornilov wrote:
>> BTRFS volume consists of a number of subvolumes which can be mounted separately
>> from each other. The top-level subvolume always exists even if no subvolumes
>> were created manually. A subvolume can be denoted as the default subvolume i.e.
>> the subvolume which is mounted by default.
>>
>> The default "default subvolume" is the top-level one, but this is far from the
>> common practices used in the wild. For instance, openSUSE provides an OS
>> snapshot/rollback feature based on BTRFS. To achieve this, the actual OS root
>> filesystem is located into a separate subvolume which is "default" but not
>> "top-level". That means that the /boot/dtb/ directory is also located inside
>> this default subvolume instead of top-level one.
>>
>> However, the existing btrfs u-boot driver always uses the top-level subvolume
>> as the filesystem root. This behaviour 1) is inconsistent with
>>
>>      mount /dev/sda1 /target
>>
>> command, which mount the default subvolume 2) leads to the issues when
>> /boot/dtb cannot be found properly (see the reference).
> 
> I also noticed the problem in the past, but forgot to fix it....
> 
>>
>> This patch uses the default subvolume as the filesystem root to overcome
>> mentioned issues.
>>
>> Reference: https://bugzilla.suse.com/show_bug.cgi?id=1185656
>> Signed-off-by: Matwey V. Kornilov <matwey.kornilov@gmail.com>
> 
> Reviewed-by: Qu Wenruo <wqu@suse.com>
> 

I can see that this patch is marked in your patchwork queue as "Need Review /
ACK". Qu is one of our core btrfs developer who reviewed the patch. Apart from
that we have it running on openSUSE on top of v2021.07 for some time without any
issues.

Would it be possible to merge this for v2021.10 or do you see any blocker here?

Regards,
Matthias

> Thanks,
> Qu
> 
>> ---
>>   fs/btrfs/disk-io.c | 38 +++++++++++++++++++++++++++++++++++---
>>   1 file changed, 35 insertions(+), 3 deletions(-)
>>
>> diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
>> index 349411c3cc..12f9579fcf 100644
>> --- a/fs/btrfs/disk-io.c
>> +++ b/fs/btrfs/disk-io.c
>> @@ -804,6 +804,30 @@ static int setup_root_or_create_block(struct
>> btrfs_fs_info *fs_info,
>>       return 0;
>>   }
>>   +static int get_default_subvolume(struct btrfs_fs_info *fs_info,
>> +                 struct btrfs_key *key_ret)
>> +{
>> +    struct btrfs_root *root = fs_info->tree_root;
>> +    struct btrfs_dir_item *dir_item;
>> +    struct btrfs_path path;
>> +    int ret = 0;
>> +
>> +    btrfs_init_path(&path);
>> +
>> +    dir_item = btrfs_lookup_dir_item(NULL, root, &path,
>> +                     BTRFS_ROOT_TREE_DIR_OBJECTID,
>> +                     "default", 7, 0);
>> +    if (IS_ERR(dir_item)) {
>> +        ret = PTR_ERR(dir_item);
>> +        goto out;
>> +    }
>> +
>> +    btrfs_dir_item_key_to_cpu(path.nodes[0], dir_item, key_ret);
>> +out:
>> +    btrfs_release_path(&path);
>> +    return ret;
>> +}
>> +
>>   int btrfs_setup_all_roots(struct btrfs_fs_info *fs_info)
>>   {
>>       struct btrfs_super_block *sb = fs_info->super_copy;
>> @@ -833,9 +857,17 @@ int btrfs_setup_all_roots(struct btrfs_fs_info *fs_info)
>>         fs_info->last_trans_committed = generation;
>>   -    key.objectid = BTRFS_FS_TREE_OBJECTID;
>> -    key.type = BTRFS_ROOT_ITEM_KEY;
>> -    key.offset = (u64)-1;
>> +    ret = get_default_subvolume(fs_info, &key);
>> +    if (ret) {
>> +        /*
>> +         * The default dir item isn't there. Linux kernel behaviour is
>> +         * to silently use the top-level subvolume in this case.
>> +         */
>> +        key.objectid = BTRFS_FS_TREE_OBJECTID;
>> +        key.type = BTRFS_ROOT_ITEM_KEY;
>> +        key.offset = (u64)-1;
>> +    }
>> +
>>       fs_info->fs_root = btrfs_read_fs_root(fs_info, &key);
>>         if (IS_ERR(fs_info->fs_root))
>>
> 

Re: [PATCH] btrfs: Use default subvolume as filesystem root

[Tom Rini]

[-- Attachment #1: Type: text/plain, Size: 2467 bytes --]

On Wed, Sep 01, 2021 at 01:28:30PM +0200, Matthias Brugger wrote:
> Hi Tom,
> 
> On 02/08/2021 01:06, Qu Wenruo wrote:
> > 
> > 
> > On 2021/8/2 上午4:52, Matwey V. Kornilov wrote:
> >> BTRFS volume consists of a number of subvolumes which can be mounted separately
> >> from each other. The top-level subvolume always exists even if no subvolumes
> >> were created manually. A subvolume can be denoted as the default subvolume i.e.
> >> the subvolume which is mounted by default.
> >>
> >> The default "default subvolume" is the top-level one, but this is far from the
> >> common practices used in the wild. For instance, openSUSE provides an OS
> >> snapshot/rollback feature based on BTRFS. To achieve this, the actual OS root
> >> filesystem is located into a separate subvolume which is "default" but not
> >> "top-level". That means that the /boot/dtb/ directory is also located inside
> >> this default subvolume instead of top-level one.
> >>
> >> However, the existing btrfs u-boot driver always uses the top-level subvolume
> >> as the filesystem root. This behaviour 1) is inconsistent with
> >>
> >>      mount /dev/sda1 /target
> >>
> >> command, which mount the default subvolume 2) leads to the issues when
> >> /boot/dtb cannot be found properly (see the reference).
> > 
> > I also noticed the problem in the past, but forgot to fix it....
> > 
> >>
> >> This patch uses the default subvolume as the filesystem root to overcome
> >> mentioned issues.
> >>
> >> Reference: https://bugzilla.suse.com/show_bug.cgi?id=1185656
> >> Signed-off-by: Matwey V. Kornilov <matwey.kornilov@gmail.com>
> > 
> > Reviewed-by: Qu Wenruo <wqu@suse.com>
> > 
> 
> I can see that this patch is marked in your patchwork queue as "Need Review /
> ACK". Qu is one of our core btrfs developer who reviewed the patch. Apart from
> that we have it running on openSUSE on top of v2021.07 for some time without any
> issues.

Ah, yup.  Qu is one of the people I do look for to have reviewed a btrfs
patch before I apply it (and I throw things under Need Review / ACK as a
note-to-self to make sure a patch does have one, when I can expect one,
before applying, FWIW).

> Would it be possible to merge this for v2021.10 or do you see any blocker here?

I think I had mentally filed it was feature not bugfix and was going to
hold off, but since you're asking, yes, I can grab it for this release.
Thanks!

-- 
Tom

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 659 bytes --]

Re: [PATCH] btrfs: Use default subvolume as filesystem root

[Matthias Brugger]

On 01/09/2021 13:36, Tom Rini wrote:
> On Wed, Sep 01, 2021 at 01:28:30PM +0200, Matthias Brugger wrote:
>> Hi Tom,
>>
>> On 02/08/2021 01:06, Qu Wenruo wrote:
>>>
>>>
>>> On 2021/8/2 上午4:52, Matwey V. Kornilov wrote:
>>>> BTRFS volume consists of a number of subvolumes which can be mounted separately
>>>> from each other. The top-level subvolume always exists even if no subvolumes
>>>> were created manually. A subvolume can be denoted as the default subvolume i.e.
>>>> the subvolume which is mounted by default.
>>>>
>>>> The default "default subvolume" is the top-level one, but this is far from the
>>>> common practices used in the wild. For instance, openSUSE provides an OS
>>>> snapshot/rollback feature based on BTRFS. To achieve this, the actual OS root
>>>> filesystem is located into a separate subvolume which is "default" but not
>>>> "top-level". That means that the /boot/dtb/ directory is also located inside
>>>> this default subvolume instead of top-level one.
>>>>
>>>> However, the existing btrfs u-boot driver always uses the top-level subvolume
>>>> as the filesystem root. This behaviour 1) is inconsistent with
>>>>
>>>>      mount /dev/sda1 /target
>>>>
>>>> command, which mount the default subvolume 2) leads to the issues when
>>>> /boot/dtb cannot be found properly (see the reference).
>>>
>>> I also noticed the problem in the past, but forgot to fix it....
>>>
>>>>
>>>> This patch uses the default subvolume as the filesystem root to overcome
>>>> mentioned issues.
>>>>
>>>> Reference: https://bugzilla.suse.com/show_bug.cgi?id=1185656
>>>> Signed-off-by: Matwey V. Kornilov <matwey.kornilov@gmail.com>
>>>
>>> Reviewed-by: Qu Wenruo <wqu@suse.com>
>>>
>>
>> I can see that this patch is marked in your patchwork queue as "Need Review /
>> ACK". Qu is one of our core btrfs developer who reviewed the patch. Apart from
>> that we have it running on openSUSE on top of v2021.07 for some time without any
>> issues.
> 
> Ah, yup.  Qu is one of the people I do look for to have reviewed a btrfs
> patch before I apply it (and I throw things under Need Review / ACK as a
> note-to-self to make sure a patch does have one, when I can expect one,
> before applying, FWIW).
> 
>> Would it be possible to merge this for v2021.10 or do you see any blocker here?
> 
> I think I had mentally filed it was feature not bugfix and was going to
> hold off, but since you're asking, yes, I can grab it for this release.
> Thanks!
> 

It's a bugfix, as loading files from BTRFS, at least in openSUSE does not work
without it.

Thanks for the quick answer!
Matthias

Re: [PATCH] btrfs: Use default subvolume as filesystem root

[Qu Wenruo]

On 2021/9/1 下午9:48, Matthias Brugger wrote:
> 
> 
> On 01/09/2021 13:36, Tom Rini wrote:
>> On Wed, Sep 01, 2021 at 01:28:30PM +0200, Matthias Brugger wrote:
>>> Hi Tom,
>>>
>>> On 02/08/2021 01:06, Qu Wenruo wrote:
>>>>
>>>>
>>>> On 2021/8/2 上午4:52, Matwey V. Kornilov wrote:
>>>>> BTRFS volume consists of a number of subvolumes which can be mounted separately
>>>>> from each other. The top-level subvolume always exists even if no subvolumes
>>>>> were created manually. A subvolume can be denoted as the default subvolume i.e.
>>>>> the subvolume which is mounted by default.
>>>>>
>>>>> The default "default subvolume" is the top-level one, but this is far from the
>>>>> common practices used in the wild. For instance, openSUSE provides an OS
>>>>> snapshot/rollback feature based on BTRFS. To achieve this, the actual OS root
>>>>> filesystem is located into a separate subvolume which is "default" but not
>>>>> "top-level". That means that the /boot/dtb/ directory is also located inside
>>>>> this default subvolume instead of top-level one.
>>>>>
>>>>> However, the existing btrfs u-boot driver always uses the top-level subvolume
>>>>> as the filesystem root. This behaviour 1) is inconsistent with
>>>>>
>>>>>       mount /dev/sda1 /target
>>>>>
>>>>> command, which mount the default subvolume 2) leads to the issues when
>>>>> /boot/dtb cannot be found properly (see the reference).
>>>>
>>>> I also noticed the problem in the past, but forgot to fix it....
>>>>
>>>>>
>>>>> This patch uses the default subvolume as the filesystem root to overcome
>>>>> mentioned issues.
>>>>>
>>>>> Reference: https://bugzilla.suse.com/show_bug.cgi?id=1185656
>>>>> Signed-off-by: Matwey V. Kornilov <matwey.kornilov@gmail.com>
>>>>
>>>> Reviewed-by: Qu Wenruo <wqu@suse.com>
>>>>
>>>
>>> I can see that this patch is marked in your patchwork queue as "Need Review /
>>> ACK". Qu is one of our core btrfs developer who reviewed the patch. Apart from
>>> that we have it running on openSUSE on top of v2021.07 for some time without any
>>> issues.
>>
>> Ah, yup.  Qu is one of the people I do look for to have reviewed a btrfs
>> patch before I apply it (and I throw things under Need Review / ACK as a
>> note-to-self to make sure a patch does have one, when I can expect one,
>> before applying, FWIW).
>>
>>> Would it be possible to merge this for v2021.10 or do you see any blocker here?
>>
>> I think I had mentally filed it was feature not bugfix and was going to
>> hold off, but since you're asking, yes, I can grab it for this release.
>> Thanks!
>>
> 
> It's a bugfix, as loading files from BTRFS, at least in openSUSE does not work
> without it.

Just to be extra accurate, the original code always accesses subvol 5 as 
the entrance point, thus for openSUSE or any other distros setting other 
default subvolume, this will cause behavior difference between kernel 
and u-boot, thus unable to find the correct /boot path.

This problem is cause by my rework using btrfs-progs code. All my fault.

Btrfs-progs never needs to bother the default subvolume, as it has no 
way to "mount" the fs nor explore the directories/files structure.

Thus it's a regression, and I'm not sure if we should use Fixes: tag.

Just in case, the fixes tag is:

Fixes: f06bfcf54d0e ("fs: btrfs: Crossport open_ctree_fs_info() from 
btrfs-progs")

Thank all guys involved in this case.
Qu

> 
> Thanks for the quick answer!
> Matthias
> 

Re: [PATCH] btrfs: Use default subvolume as filesystem root

[Tom Rini]

[-- Attachment #1: Type: text/plain, Size: 1463 bytes --]

On Sun, Aug 01, 2021 at 11:52:16PM +0300, Matwey V. Kornilov wrote:

> BTRFS volume consists of a number of subvolumes which can be mounted separately
> from each other. The top-level subvolume always exists even if no subvolumes
> were created manually. A subvolume can be denoted as the default subvolume i.e.
> the subvolume which is mounted by default.
> 
> The default "default subvolume" is the top-level one, but this is far from the
> common practices used in the wild. For instance, openSUSE provides an OS
> snapshot/rollback feature based on BTRFS. To achieve this, the actual OS root
> filesystem is located into a separate subvolume which is "default" but not
> "top-level". That means that the /boot/dtb/ directory is also located inside
> this default subvolume instead of top-level one.
> 
> However, the existing btrfs u-boot driver always uses the top-level subvolume
> as the filesystem root. This behaviour 1) is inconsistent with
> 
>     mount /dev/sda1 /target
> 
> command, which mount the default subvolume 2) leads to the issues when
> /boot/dtb cannot be found properly (see the reference).
> 
> This patch uses the default subvolume as the filesystem root to overcome
> mentioned issues.
> 
> Reference: https://bugzilla.suse.com/show_bug.cgi?id=1185656
> Signed-off-by: Matwey V. Kornilov <matwey.kornilov@gmail.com>
> Reviewed-by: Qu Wenruo <wqu@suse.com>

Applied to u-boot/master, thanks!

-- 
Tom

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 659 bytes --]