On Tue, Sep 01, 2015 at 03:48:57PM -0400, Josef Bacik wrote:
> On 09/01/2015 03:05 PM, Omar Sandoval wrote:
> >From: Omar Sandoval <osan...@fb.com>
> >
> >The free space tree is updated in tandem with the extent tree. There are
> >only a handful of places where we need to hook in:
> >
> >1. Block group creation
> >2. Block group deletion
> >3. Delayed refs (extent creation and deletion)
> >4. Block group caching
> >
> >Signed-off-by: Omar Sandoval <osan...@fb.com>
> >---
> > fs/btrfs/extent-tree.c | 73
> > +++++++++++++++++++++++++++++++++++++++++++++++---
> > 1 file changed, 70 insertions(+), 3 deletions(-)
> >
> >diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
> >index 37179a569f40..3f10df3932f0 100644
> >--- a/fs/btrfs/extent-tree.c
> >+++ b/fs/btrfs/extent-tree.c
> >@@ -33,6 +33,7 @@
> > #include "raid56.h"
> > #include "locking.h"
> > #include "free-space-cache.h"
> >+#include "free-space-tree.h"
> > #include "math.h"
> > #include "sysfs.h"
> > #include "qgroup.h"
> >@@ -589,7 +590,41 @@ static int cache_block_group(struct
> >btrfs_block_group_cache *cache,
> > cache->cached = BTRFS_CACHE_FAST;
> > spin_unlock(&cache->lock);
> >
> >- if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
> >+ if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
> >+ if (load_cache_only) {
> >+ spin_lock(&cache->lock);
> >+ cache->caching_ctl = NULL;
> >+ cache->cached = BTRFS_CACHE_NO;
> >+ spin_unlock(&cache->lock);
> >+ wake_up(&caching_ctl->wait);
> >+ } else {
> >+ mutex_lock(&caching_ctl->mutex);
> >+ ret = load_free_space_tree(fs_info, cache);
> >+ if (ret) {
> >+ btrfs_warn(fs_info, "failed to load free space
> >tree for %llu: %d",
> >+ cache->key.objectid, ret);
> >+ spin_lock(&cache->lock);
> >+ cache->caching_ctl = NULL;
> >+ cache->cached = BTRFS_CACHE_ERROR;
> >+ spin_unlock(&cache->lock);
> >+ goto tree_out;
> >+ }
> >+
> >+ spin_lock(&cache->lock);
> >+ cache->caching_ctl = NULL;
> >+ cache->cached = BTRFS_CACHE_FINISHED;
> >+ cache->last_byte_to_unpin = (u64)-1;
> >+ caching_ctl->progress = (u64)-1;
> >+ spin_unlock(&cache->lock);
> >+ mutex_unlock(&caching_ctl->mutex);
> >+
> >+tree_out:
> >+ wake_up(&caching_ctl->wait);
> >+ put_caching_control(caching_ctl);
> >+ free_excluded_extents(fs_info->extent_root, cache);
> >+ return 0;
> >+ }
> >+ } else if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
> > mutex_lock(&caching_ctl->mutex);
>
> So the reason we have this load_cache_only thing is because the free space
> cache could be loaded almost instantaneously since it was contiguously
> allocated. This isn't the case with the free space tree, and although it is
> better than the no space cache way of doing things, we are still going to
> incur a bit of latency when seeking through a large free space tree. So
> break this out and make the caching kthread either do the old style load or
> load the free space tree. Then you can use the add free space helpers that
> will wake anybody up waiting on allocations and you incur less direct
> latency. Thanks,
>
> Josef
Okay, I'll do the load from caching_thread(). Do you think we're going
to need the need_resched() || rwsem_is_contended(commit_root) check and
retry for the free space tree like we do with the extent tree? It seems
like it could get complicated since we would need to worry about the
format changing underneath us.
--
Omar
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
