From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mx0b-00082601.pphosted.com ([67.231.153.30]:57786 "EHLO mx0a-00082601.pphosted.com" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1751264AbcJGNRX (ORCPT ); Fri, 7 Oct 2016 09:17:23 -0400 Subject: Re: [PATCH 1/2] btrfs: try to satisfy metadata requests when every flush_space() returns To: Wang Xiaoguang , References: <1474441173-31049-1-git-send-email-wangxg.fnst@cn.fujitsu.com> CC: From: Josef Bacik Message-ID: <64702aa8-206f-d167-a4c7-0303864a1e23@fb.com> Date: Fri, 7 Oct 2016 09:16:55 -0400 MIME-Version: 1.0 In-Reply-To: <1474441173-31049-1-git-send-email-wangxg.fnst@cn.fujitsu.com> Content-Type: text/plain; charset="windows-1252"; format=flowed Sender: linux-btrfs-owner@vger.kernel.org List-ID: On 09/21/2016 02:59 AM, Wang Xiaoguang wrote: > In flush_space()->shrink_delalloc(), if can_overcommit() returns true, > though no bytes added to space_info, we still may satisfy some requests. > > Signed-off-by: Wang Xiaoguang > --- > fs/btrfs/extent-tree.c | 39 +++++++++++++++++++++++++++++++++++++++ > 1 file changed, 39 insertions(+) > > diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c > index 38c2df8..fdfc97f 100644 > --- a/fs/btrfs/extent-tree.c > +++ b/fs/btrfs/extent-tree.c > @@ -4960,6 +4960,43 @@ static void wake_all_tickets(struct list_head *head) > } > > /* > + * This function must be protected by btrfs_space_info's lock. > + */ > +static void try_to_wake_tickets(struct btrfs_root *root, > + struct btrfs_space_info *space_info) > +{ > + struct reserve_ticket *ticket; > + struct list_head *head = &space_info->priority_tickets; > + enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH; > + u64 used; > + > +again: > + while (!list_empty(head)) { > + ticket = list_first_entry(head, struct reserve_ticket, > + list); > + used = space_info->bytes_used + > + space_info->bytes_reserved + space_info->bytes_pinned + > + space_info->bytes_readonly + space_info->bytes_may_use; > + > + if (used + ticket->bytes <= space_info->total_bytes || > + can_overcommit(root, space_info, ticket->bytes, flush)) { > + space_info->bytes_may_use += ticket->bytes; > + list_del_init(&ticket->list); > + ticket->bytes = 0; > + space_info->tickets_id++; > + wake_up(&ticket->wait); > + } else > + return; > + } > + > + if (head == &space_info->priority_tickets) { > + head = &space_info->tickets; > + flush = BTRFS_RESERVE_FLUSH_ALL; > + goto again; > + } > +} > + > +/* > * This is for normal flushers, we can wait all goddamned day if we want to. We > * will loop and continuously try to flush as long as we are making progress. > * We count progress as clearing off tickets each time we have to loop. > @@ -4995,6 +5032,8 @@ static void btrfs_async_reclaim_metadata_space(struct work_struct *work) > ret = flush_space(fs_info->fs_root, space_info, to_reclaim, > to_reclaim, flush_state); > spin_lock(&space_info->lock); > + if (!ret) > + try_to_wake_tickets(fs_info->fs_root, space_info); > if (list_empty(&space_info->tickets)) { > space_info->flush = 0; > spin_unlock(&space_info->lock); > So first off I have no problems with this patch, it seems reasonable to me. However I'm curious to see where it helped. The only time can_overcommit() would suddenly start returning true where it wouldn't have before without actually adding space to the space_info would be when we've dropped an empty block group (or added a new drive) and suddenly fs_info->free_chunk_space is larger than it was. Is that what you were observing? Thanks, Josef