From 77b7aad195099e7c6da11e94b7fa6ef5e6fb0025 Mon Sep 17 00:00:00 2001 From: David Sterba Date: Wed, 9 Jan 2019 15:02:23 +0100 Subject: Revert "btrfs: balance dirty metadata pages in btrfs_finish_ordered_io" This reverts commit e73e81b6d0114d4a303205a952ab2e87c44bd279. This patch causes a few problems: - adds latency to btrfs_finish_ordered_io - as btrfs_finish_ordered_io is used for free space cache, generating more work from btrfs_btree_balance_dirty_nodelay could end up in the same workque, effectively deadlocking 12260 kworker/u96:16+btrfs-freespace-write D [<0>] balance_dirty_pages+0x6e6/0x7ad [<0>] balance_dirty_pages_ratelimited+0x6bb/0xa90 [<0>] btrfs_finish_ordered_io+0x3da/0x770 [<0>] normal_work_helper+0x1c5/0x5a0 [<0>] process_one_work+0x1ee/0x5a0 [<0>] worker_thread+0x46/0x3d0 [<0>] kthread+0xf5/0x130 [<0>] ret_from_fork+0x24/0x30 [<0>] 0xffffffffffffffff Transaction commit will wait on the freespace cache: 838 btrfs-transacti D [<0>] btrfs_start_ordered_extent+0x154/0x1e0 [<0>] btrfs_wait_ordered_range+0xbd/0x110 [<0>] __btrfs_wait_cache_io+0x49/0x1a0 [<0>] btrfs_write_dirty_block_groups+0x10b/0x3b0 [<0>] commit_cowonly_roots+0x215/0x2b0 [<0>] btrfs_commit_transaction+0x37e/0x910 [<0>] transaction_kthread+0x14d/0x180 [<0>] kthread+0xf5/0x130 [<0>] ret_from_fork+0x24/0x30 [<0>] 0xffffffffffffffff And then writepages ends up waiting on transaction commit: 9520 kworker/u96:13+flush-btrfs-1 D [<0>] wait_current_trans+0xac/0xe0 [<0>] start_transaction+0x21b/0x4b0 [<0>] cow_file_range_inline+0x10b/0x6b0 [<0>] cow_file_range.isra.69+0x329/0x4a0 [<0>] run_delalloc_range+0x105/0x3c0 [<0>] writepage_delalloc+0x119/0x180 [<0>] __extent_writepage+0x10c/0x390 [<0>] extent_write_cache_pages+0x26f/0x3d0 [<0>] extent_writepages+0x4f/0x80 [<0>] do_writepages+0x17/0x60 [<0>] __writeback_single_inode+0x59/0x690 [<0>] writeback_sb_inodes+0x291/0x4e0 [<0>] __writeback_inodes_wb+0x87/0xb0 [<0>] wb_writeback+0x3bb/0x500 [<0>] wb_workfn+0x40d/0x610 [<0>] process_one_work+0x1ee/0x5a0 [<0>] worker_thread+0x1e0/0x3d0 [<0>] kthread+0xf5/0x130 [<0>] ret_from_fork+0x24/0x30 [<0>] 0xffffffffffffffff Eventually, we have every process in the system waiting on balance_dirty_pages(), and nobody is able to make progress on page writeback. The original patch tried to fix an OOM condition, that happened on 4.4 but no success reproducing that on later kernels (4.19 and 4.20). This is more likely a problem in OOM itself. Link: https://lore.kernel.org/linux-btrfs/20180528054821.9092-1-ethanlien@synology.com/ Reported-by: Chris Mason CC: stable@vger.kernel.org # 4.18+ CC: ethanlien Signed-off-by: David Sterba --- fs/btrfs/inode.c | 3 --- 1 file changed, 3 deletions(-) (limited to 'fs/btrfs/inode.c') diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index 43eb4535319d..b6025b5d0b25 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -3129,9 +3129,6 @@ out: /* once for the tree */ btrfs_put_ordered_extent(ordered_extent); - /* Try to release some metadata so we don't get an OOM but don't wait */ - btrfs_btree_balance_dirty_nodelay(fs_info); - return ret; } -- cgit v1.2.3-55-g7522 From fd340d0f68cc87badfc9efcb226f23a5428826a0 Mon Sep 17 00:00:00 2001 From: Josef Bacik Date: Fri, 11 Jan 2019 10:21:02 -0500 Subject: btrfs: wakeup cleaner thread when adding delayed iput The cleaner thread usually takes care of delayed iputs, with the exception of the btrfs_end_transaction_throttle path. Delaying iputs means we are potentially delaying the eviction of an inode and it's respective space. The cleaner thread only gets woken up every 30 seconds, or when we require space. If there are a lot of inodes that need to be deleted we could induce a serious amount of latency while we wait for these inodes to be evicted. So instead wakeup the cleaner if it's not already awake to process any new delayed iputs we add to the list. If we suddenly need space we will less likely be backed up behind a bunch of inodes that are waiting to be deleted, and we could possibly free space before we need to get into the flushing logic which will save us some latency. Reviewed-by: Filipe Manana Signed-off-by: Josef Bacik Signed-off-by: David Sterba --- fs/btrfs/ctree.h | 3 +++ fs/btrfs/disk-io.c | 3 +++ fs/btrfs/inode.c | 2 ++ 3 files changed, 8 insertions(+) (limited to 'fs/btrfs/inode.c') diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h index 34019c8b6158..8b1d06fa222d 100644 --- a/fs/btrfs/ctree.h +++ b/fs/btrfs/ctree.h @@ -787,6 +787,9 @@ enum { * main phase. The fs_info::balance_ctl is initialized. */ BTRFS_FS_BALANCE_RUNNING, + + /* Indicate that the cleaner thread is awake and doing something. */ + BTRFS_FS_CLEANER_RUNNING, }; struct btrfs_fs_info { diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c index 18eefc5b2532..6a2a2a951705 100644 --- a/fs/btrfs/disk-io.c +++ b/fs/btrfs/disk-io.c @@ -1682,6 +1682,8 @@ static int cleaner_kthread(void *arg) while (1) { again = 0; + set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); + /* Make the cleaner go to sleep early. */ if (btrfs_need_cleaner_sleep(fs_info)) goto sleep; @@ -1728,6 +1730,7 @@ static int cleaner_kthread(void *arg) */ btrfs_delete_unused_bgs(fs_info); sleep: + clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); if (kthread_should_park()) kthread_parkme(); if (kthread_should_stop()) diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index b6025b5d0b25..5c349667c761 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -3251,6 +3251,8 @@ void btrfs_add_delayed_iput(struct inode *inode) ASSERT(list_empty(&binode->delayed_iput)); list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs); spin_unlock(&fs_info->delayed_iput_lock); + if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags)) + wake_up_process(fs_info->cleaner_kthread); } void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info) -- cgit v1.2.3-55-g7522