@@ -1016,6 +1016,7 @@ struct btrfs_fs_info {
/* used to keep from writing metadata until there is a nice batch */
struct percpu_counter dirty_metadata_bytes;
struct percpu_counter delalloc_bytes;
+ struct percpu_counter odirect_bytes;
s32 dirty_metadata_batch;
s32 delalloc_batch;
@@ -2641,11 +2641,17 @@ int open_ctree(struct super_block *sb,
goto fail;
}
- ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
+ ret = percpu_counter_init(&fs_info->odirect_bytes, 0, GFP_KERNEL);
if (ret) {
err = ret;
goto fail_srcu;
}
+
+ ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
+ if (ret) {
+ err = ret;
+ goto fail_odirect_bytes;
+ }
fs_info->dirty_metadata_batch = PAGE_SIZE *
(1 + ilog2(nr_cpu_ids));
@@ -3344,6 +3350,8 @@ int open_ctree(struct super_block *sb,
percpu_counter_destroy(&fs_info->delalloc_bytes);
fail_dirty_metadata_bytes:
percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
+fail_odirect_bytes:
+ percpu_counter_destroy(&fs_info->odirect_bytes);
fail_srcu:
cleanup_srcu_struct(&fs_info->subvol_srcu);
fail:
@@ -4025,6 +4033,10 @@ void close_ctree(struct btrfs_fs_info *fs_info)
percpu_counter_sum(&fs_info->delalloc_bytes));
}
+ if (percpu_counter_sum(&fs_info->odirect_bytes))
+ btrfs_info(fs_info, "at unmount odirect count %lld",
+ percpu_counter_sum(&fs_info->odirect_bytes));
+
btrfs_sysfs_remove_mounted(fs_info);
btrfs_sysfs_remove_fsid(fs_info->fs_devices);
@@ -4056,6 +4068,7 @@ void close_ctree(struct btrfs_fs_info *fs_info)
percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
percpu_counter_destroy(&fs_info->delalloc_bytes);
+ percpu_counter_destroy(&fs_info->odirect_bytes);
percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
cleanup_srcu_struct(&fs_info->subvol_srcu);
@@ -4727,6 +4727,7 @@ static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
struct btrfs_space_info *space_info;
struct btrfs_trans_handle *trans;
u64 delalloc_bytes;
+ u64 odirect_bytes;
u64 async_pages;
u64 items;
long time_left;
@@ -4742,7 +4743,9 @@ static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
delalloc_bytes = percpu_counter_sum_positive(
&fs_info->delalloc_bytes);
- if (delalloc_bytes == 0) {
+ odirect_bytes = percpu_counter_sum_positive(
+ &fs_info->odirect_bytes);
+ if (delalloc_bytes == 0 && odirect_bytes == 0) {
if (trans)
return;
if (wait_ordered)
@@ -4750,8 +4753,16 @@ static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
return;
}
+ /*
+ * If we are doing more ordered than delalloc we need to just wait on
+ * ordered extents, otherwise we'll waste time trying to flush delalloc
+ * that likely won't give us the space back we need.
+ */
+ if (odirect_bytes > delalloc_bytes)
+ wait_ordered = true;
+
loops = 0;
- while (delalloc_bytes && loops < 3) {
+ while ((delalloc_bytes || odirect_bytes) && loops < 3) {
nr_pages = min(delalloc_bytes, to_reclaim) >> PAGE_SHIFT;
/*
@@ -4801,6 +4812,8 @@ static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
}
delalloc_bytes = percpu_counter_sum_positive(
&fs_info->delalloc_bytes);
+ odirect_bytes = percpu_counter_sum_positive(
+ &fs_info->odirect_bytes);
}
}
@@ -194,8 +194,11 @@ static int __btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
set_bit(type, &entry->flags);
- if (dio)
+ if (dio) {
+ percpu_counter_add_batch(&fs_info->odirect_bytes, len,
+ fs_info->delalloc_batch);
set_bit(BTRFS_ORDERED_DIRECT, &entry->flags);
+ }
/* one ref for the tree */
refcount_set(&entry->refs, 1);
@@ -468,6 +471,10 @@ void btrfs_remove_ordered_extent(struct inode *inode,
if (root != fs_info->tree_root)
btrfs_delalloc_release_metadata(btrfs_inode, entry->len, false);
+ if (test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
+ percpu_counter_add_batch(&fs_info->odirect_bytes, -entry->len,
+ fs_info->delalloc_batch);
+
tree = &btrfs_inode->ordered_tree;
spin_lock_irq(&tree->lock);
node = &entry->rb_node;
When diagnosing a slowdown of generic/224 I noticed we were not doing anything when calling into shrink_delalloc(). This is because all writes in 224 are O_DIRECT, not delalloc, and thus our delalloc_bytes counter is 0, which short circuits most of the work inside of shrink_delalloc(). However O_DIRECT writes still consume metadata resources and generate ordered extents, which we can still wait on. Fix this by tracking outstandingn odirect write bytes, and use this as well as the delalloc byts counter to decide if we need to lookup and wait on any ordered extents. If we have more odirect writes than delalloc bytes we'll go ahead and wait on any ordered extents irregardless of our flush state as flushing delalloc is likely to not gain us anything. Signed-off-by: Josef Bacik <josef@toxicpanda.com> --- v1->v2: - updated the changelog fs/btrfs/ctree.h | 1 + fs/btrfs/disk-io.c | 15 ++++++++++++++- fs/btrfs/extent-tree.c | 17 +++++++++++++++-- fs/btrfs/ordered-data.c | 9 ++++++++- 4 files changed, 38 insertions(+), 4 deletions(-)