On 01/02/2021 16:49, Filipe Manana wrote:
> On Wed, Jan 27, 2021 at 6:48 PM Naohiro Aota <naohiro.aota@xxxxxxx> wrote:
>>
>> This is the 3/3 patch to enable tree-log on ZONED mode.
>>
>> The allocation order of nodes of "fs_info->log_root_tree" and nodes of
>> "root->log_root" is not the same as the writing order of them. So, the
>> writing causes unaligned write errors.
>>
>> This patch reorders the allocation of them by delaying allocation of the
>> root node of "fs_info->log_root_tree," so that the node buffers can go out
>> sequentially to devices.
>>
>> Reviewed-by: Josef Bacik <josef@xxxxxxxxxxxxxx>
>> Signed-off-by: Johannes Thumshirn <johannes.thumshirn@xxxxxxx>
>> Signed-off-by: Naohiro Aota <naohiro.aota@xxxxxxx>
>> ---
>> fs/btrfs/disk-io.c | 7 -------
>> fs/btrfs/tree-log.c | 24 ++++++++++++++++++------
>> 2 files changed, 18 insertions(+), 13 deletions(-)
>>
>> diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
>> index c3b5cfe4d928..d2b30716de84 100644
>> --- a/fs/btrfs/disk-io.c
>> +++ b/fs/btrfs/disk-io.c
>> @@ -1241,18 +1241,11 @@ int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
>> struct btrfs_fs_info *fs_info)
>> {
>> struct btrfs_root *log_root;
>> - int ret;
>>
>> log_root = alloc_log_tree(trans, fs_info);
>> if (IS_ERR(log_root))
>> return PTR_ERR(log_root);
>>
>> - ret = btrfs_alloc_log_tree_node(trans, log_root);
>> - if (ret) {
>> - btrfs_put_root(log_root);
>> - return ret;
>> - }
>> -
>> WARN_ON(fs_info->log_root_tree);
>> fs_info->log_root_tree = log_root;
>> return 0;
>> diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c
>> index 71a1c0b5bc26..d8315363dc1e 100644
>> --- a/fs/btrfs/tree-log.c
>> +++ b/fs/btrfs/tree-log.c
>> @@ -3159,6 +3159,16 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans,
>> list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
>> root_log_ctx.log_transid = log_root_tree->log_transid;
>>
>> + mutex_lock(&fs_info->tree_log_mutex);
>> + if (!log_root_tree->node) {
>> + ret = btrfs_alloc_log_tree_node(trans, log_root_tree);
>> + if (ret) {
>> + mutex_unlock(&fs_info->tree_log_mutex);
>> + goto out;
>> + }
>> + }
>> + mutex_unlock(&fs_info->tree_log_mutex);
>
> Hum, so this now has an impact for non-zoned mode.
>
> It reduces the parallelism between a previous transaction finishing
> its commit and an fsync started in the current transaction.
>
> A transaction commit releases fs_info->tree_log_mutex after it commits
> the super block.
>
> By taking that mutex here, we wait for the transaction commit to write
> the super blocks before we update the log root, start writeback of log
> tree nodes and wait for the writeback to complete.
>
> Before this change, we would do those 3 things before waiting for the
> previous transaction to commit the super blocks.
>
> So this undoes part of what was made by the following commit:
>
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=47876f7ceffa0e6af7476e052b3c061f1f2c1d9f
>
> Which landed in 5.10. This patch and the rest of the patchset was
> based on pre 5.10 code - was this missed because of that?
> Or is there some other reason to have to do things this way for non-zoned mode?
>
> I think we should preserve the behaviour for non-zoned mode - i.e. any
> reason why not allocating log_root_tree->node at the top of start
> log_trans(), while under the protection of tree_root->log_mutex?
>
> My impression is that this, and the other patch with the subject
> "btrfs: serialize log transaction on ZONED mode", are out of sync with
> the changes in 5.10.
>
> Thanks, sorry for the late review.
Yes this code has been under development for quite some time and we probably
missed this upstream update.
We will be sending a fix ASAP.
