| Message ID | 1393487254-17978-2-git-send-email-bo.li.liu@oracle.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On 02/27/2014 03:47 PM, Liu Bo wrote: > Currently to check whether a directory has been created, we search > DIR_INDEX items one by one to check if children has been processed. > > Try to picture such a scenario: > . > |-- dir (ino X) > |-- foo_1 (ino X+1) > |-- ... > |-- foo_k (ino X+k) > > With the current way, we have to check all the k DIR_INDEX items > to find it is a fresh new one. > > So this introduced a rbtree to store those directories which are > created out of order, and in the above case, we just need an O(logn) > search instead of O(1) search. Just a reminder, we ususally call O(n) rather O(1) here. If we falls O(1) to O(logn)..things are becoming worse~~ Thanks, Wang > > Signed-off-by: Liu Bo <bo.li.liu@oracle.com> > --- > v2: fix wrong patch name. > > fs/btrfs/send.c | 87 ++++++++++++++++++++++++++++----------------------------- > 1 file changed, 43 insertions(+), 44 deletions(-) > > diff --git a/fs/btrfs/send.c b/fs/btrfs/send.c > index 33063d1..fcad93c 100644 > --- a/fs/btrfs/send.c > +++ b/fs/btrfs/send.c > @@ -175,6 +175,9 @@ struct send_ctx { > * own move/rename can be performed. > */ > struct rb_root waiting_dir_moves; > + > + /* directories which are created out of order, check did_create_dir() */ > + struct rb_root out_of_order; > }; > > struct pending_dir_move { > @@ -2494,56 +2497,40 @@ out: > */ > static int did_create_dir(struct send_ctx *sctx, u64 dir) > { > - int ret = 0; > - struct btrfs_path *path = NULL; > - struct btrfs_key key; > - struct btrfs_key found_key; > - struct btrfs_key di_key; > - struct extent_buffer *eb; > - struct btrfs_dir_item *di; > - int slot; > + struct rb_node **p = &sctx->out_of_order.rb_node; > + struct rb_node *parent = NULL; > + struct send_dir_node *entry = NULL; > + int cur_is_dir = !!(dir == sctx->cur_ino); > > - path = alloc_path_for_send(); > - if (!path) { > - ret = -ENOMEM; > - goto out; > - } > + verbose_printk("dir=%llu cur_ino=%llu send_progress=%llu\n", > + dir, sctx->cur_ino, sctx->send_progress); > > - key.objectid = dir; > - key.type = BTRFS_DIR_INDEX_KEY; > - key.offset = 0; > - while (1) { > - ret = btrfs_search_slot_for_read(sctx->send_root, &key, path, > - 1, 0); > - if (ret < 0) > - goto out; > - if (!ret) { > - eb = path->nodes[0]; > - slot = path->slots[0]; > - btrfs_item_key_to_cpu(eb, &found_key, slot); > - } > - if (ret || found_key.objectid != key.objectid || > - found_key.type != key.type) { > - ret = 0; > - goto out; > + while (*p) { > + parent = *p; > + entry = rb_entry(parent, struct send_dir_node, node); > + if (dir < entry->ino) { > + p = &(*p)->rb_left; > + } else if (dir > entry->ino) { > + p = &(*p)->rb_right; > + } else { > + if (cur_is_dir) { > + rb_erase(&entry->node, &sctx->out_of_order); > + kfree(entry); > + } > + return 1; > } > + } > > - di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item); > - btrfs_dir_item_key_to_cpu(eb, di, &di_key); > - > - if (di_key.type != BTRFS_ROOT_ITEM_KEY && > - di_key.objectid < sctx->send_progress) { > - ret = 1; > - goto out; > - } > + if (!cur_is_dir) { > + entry = kmalloc(sizeof(*entry), GFP_NOFS); > + if (!entry) > + return -ENOMEM; > + entry->ino = dir; > > - key.offset = found_key.offset + 1; > - btrfs_release_path(path); > + rb_link_node(&entry->node, parent, p); > + rb_insert_color(&entry->node, &sctx->out_of_order); > } > - > -out: > - btrfs_free_path(path); > - return ret; > + return 0; > } > > /* > @@ -5340,6 +5327,7 @@ long btrfs_ioctl_send(struct file *mnt_file, void __user *arg_) > > sctx->pending_dir_moves = RB_ROOT; > sctx->waiting_dir_moves = RB_ROOT; > + sctx->out_of_order = RB_ROOT; > > sctx->clone_roots = vzalloc(sizeof(struct clone_root) * > (arg->clone_sources_count + 1)); > @@ -5477,6 +5465,17 @@ out: > kfree(dm); > } > > + WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->out_of_order)); > + while (sctx && !RB_EMPTY_ROOT(&sctx->out_of_order)) { > + struct rb_node *n; > + struct send_dir_node *entry; > + > + n = rb_first(&sctx->out_of_order); > + entry = rb_entry(n, struct send_dir_node, node); > + rb_erase(&entry->node, &sctx->out_of_order); > + kfree(entry); > + } > + > if (sort_clone_roots) { > for (i = 0; i < sctx->clone_roots_cnt; i++) > btrfs_root_dec_send_in_progress( -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Thu, Feb 27, 2014 at 04:01:23PM +0800, Wang Shilong wrote: > On 02/27/2014 03:47 PM, Liu Bo wrote: > >Currently to check whether a directory has been created, we search > >DIR_INDEX items one by one to check if children has been processed. > > > >Try to picture such a scenario: > > . > > |-- dir (ino X) > > |-- foo_1 (ino X+1) > > |-- ... > > |-- foo_k (ino X+k) > > > >With the current way, we have to check all the k DIR_INDEX items > >to find it is a fresh new one. > > > >So this introduced a rbtree to store those directories which are > >created out of order, and in the above case, we just need an O(logn) > >search instead of O(1) search. > Just a reminder, we ususally call O(n) rather O(1) here. > If we falls O(1) to O(logn)..things are becoming worse~~ Good catch, my bad. thanks, -liubo -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/fs/btrfs/send.c b/fs/btrfs/send.c index 33063d1..fcad93c 100644 --- a/fs/btrfs/send.c +++ b/fs/btrfs/send.c @@ -175,6 +175,9 @@ struct send_ctx { * own move/rename can be performed. */ struct rb_root waiting_dir_moves; + + /* directories which are created out of order, check did_create_dir() */ + struct rb_root out_of_order; }; struct pending_dir_move { @@ -2494,56 +2497,40 @@ out: */ static int did_create_dir(struct send_ctx *sctx, u64 dir) { - int ret = 0; - struct btrfs_path *path = NULL; - struct btrfs_key key; - struct btrfs_key found_key; - struct btrfs_key di_key; - struct extent_buffer *eb; - struct btrfs_dir_item *di; - int slot; + struct rb_node **p = &sctx->out_of_order.rb_node; + struct rb_node *parent = NULL; + struct send_dir_node *entry = NULL; + int cur_is_dir = !!(dir == sctx->cur_ino); - path = alloc_path_for_send(); - if (!path) { - ret = -ENOMEM; - goto out; - } + verbose_printk("dir=%llu cur_ino=%llu send_progress=%llu\n", + dir, sctx->cur_ino, sctx->send_progress); - key.objectid = dir; - key.type = BTRFS_DIR_INDEX_KEY; - key.offset = 0; - while (1) { - ret = btrfs_search_slot_for_read(sctx->send_root, &key, path, - 1, 0); - if (ret < 0) - goto out; - if (!ret) { - eb = path->nodes[0]; - slot = path->slots[0]; - btrfs_item_key_to_cpu(eb, &found_key, slot); - } - if (ret || found_key.objectid != key.objectid || - found_key.type != key.type) { - ret = 0; - goto out; + while (*p) { + parent = *p; + entry = rb_entry(parent, struct send_dir_node, node); + if (dir < entry->ino) { + p = &(*p)->rb_left; + } else if (dir > entry->ino) { + p = &(*p)->rb_right; + } else { + if (cur_is_dir) { + rb_erase(&entry->node, &sctx->out_of_order); + kfree(entry); + } + return 1; } + } - di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item); - btrfs_dir_item_key_to_cpu(eb, di, &di_key); - - if (di_key.type != BTRFS_ROOT_ITEM_KEY && - di_key.objectid < sctx->send_progress) { - ret = 1; - goto out; - } + if (!cur_is_dir) { + entry = kmalloc(sizeof(*entry), GFP_NOFS); + if (!entry) + return -ENOMEM; + entry->ino = dir; - key.offset = found_key.offset + 1; - btrfs_release_path(path); + rb_link_node(&entry->node, parent, p); + rb_insert_color(&entry->node, &sctx->out_of_order); } - -out: - btrfs_free_path(path); - return ret; + return 0; } /* @@ -5340,6 +5327,7 @@ long btrfs_ioctl_send(struct file *mnt_file, void __user *arg_) sctx->pending_dir_moves = RB_ROOT; sctx->waiting_dir_moves = RB_ROOT; + sctx->out_of_order = RB_ROOT; sctx->clone_roots = vzalloc(sizeof(struct clone_root) * (arg->clone_sources_count + 1)); @@ -5477,6 +5465,17 @@ out: kfree(dm); } + WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->out_of_order)); + while (sctx && !RB_EMPTY_ROOT(&sctx->out_of_order)) { + struct rb_node *n; + struct send_dir_node *entry; + + n = rb_first(&sctx->out_of_order); + entry = rb_entry(n, struct send_dir_node, node); + rb_erase(&entry->node, &sctx->out_of_order); + kfree(entry); + } + if (sort_clone_roots) { for (i = 0; i < sctx->clone_roots_cnt; i++) btrfs_root_dec_send_in_progress(
Currently to check whether a directory has been created, we search DIR_INDEX items one by one to check if children has been processed. Try to picture such a scenario: . |-- dir (ino X) |-- foo_1 (ino X+1) |-- ... |-- foo_k (ino X+k) With the current way, we have to check all the k DIR_INDEX items to find it is a fresh new one. So this introduced a rbtree to store those directories which are created out of order, and in the above case, we just need an O(logn) search instead of O(1) search. Signed-off-by: Liu Bo <bo.li.liu@oracle.com> --- v2: fix wrong patch name. fs/btrfs/send.c | 87 ++++++++++++++++++++++++++++----------------------------- 1 file changed, 43 insertions(+), 44 deletions(-)