#include "ref-cache.h"
#include "tree-log.h"
-extern struct kmem_cache *btrfs_trans_handle_cachep;
-extern struct kmem_cache *btrfs_transaction_cachep;
-
#define BTRFS_ROOT_TRANS_TAG 0
static noinline void put_transaction(struct btrfs_transaction *transaction)
GFP_NOFS);
BUG_ON(!cur_trans);
root->fs_info->generation++;
- root->fs_info->last_alloc = 0;
- root->fs_info->last_data_alloc = 0;
cur_trans->num_writers = 1;
cur_trans->num_joined = 0;
cur_trans->transid = root->fs_info->generation;
cur_trans->use_count = 1;
cur_trans->commit_done = 0;
cur_trans->start_time = get_seconds();
+
+ cur_trans->delayed_refs.root.rb_node = NULL;
+ cur_trans->delayed_refs.num_entries = 0;
+ cur_trans->delayed_refs.num_heads_ready = 0;
+ cur_trans->delayed_refs.num_heads = 0;
+ cur_trans->delayed_refs.flushing = 0;
+ cur_trans->delayed_refs.run_delayed_start = 0;
+ spin_lock_init(&cur_trans->delayed_refs.lock);
+
INIT_LIST_HEAD(&cur_trans->pending_snapshots);
list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
extent_io_tree_init(&cur_trans->dirty_pages,
}
/*
- * this does all the record keeping required to make sure that a
- * reference counted root is properly recorded in a given transaction.
- * This is required to make sure the old root from before we joined the transaction
- * is deleted when the transaction commits
+ * this does all the record keeping required to make sure that a reference
+ * counted root is properly recorded in a given transaction. This is required
+ * to make sure the old root from before we joined the transaction is deleted
+ * when the transaction commits
*/
noinline int btrfs_record_root_in_trans(struct btrfs_root *root)
{
if (cur_trans && cur_trans->blocked) {
DEFINE_WAIT(wait);
cur_trans->use_count++;
- while(1) {
+ while (1) {
prepare_to_wait(&root->fs_info->transaction_wait, &wait,
TASK_UNINTERRUPTIBLE);
if (cur_trans->blocked) {
h->transaction = root->fs_info->running_transaction;
h->blocks_reserved = num_blocks;
h->blocks_used = 0;
- h->block_group = NULL;
+ h->block_group = 0;
h->alloc_exclude_nr = 0;
h->alloc_exclude_start = 0;
+ h->delayed_ref_updates = 0;
+
root->fs_info->running_transaction->use_count++;
mutex_unlock(&root->fs_info->trans_mutex);
return h;
{
DEFINE_WAIT(wait);
mutex_lock(&root->fs_info->trans_mutex);
- while(!commit->commit_done) {
+ while (!commit->commit_done) {
prepare_to_wait(&commit->commit_wait, &wait,
TASK_UNINTERRUPTIBLE);
if (commit->commit_done)
}
/*
- * rate limit against the drop_snapshot code. This helps to slow down new operations
- * if the drop_snapshot code isn't able to keep up.
+ * rate limit against the drop_snapshot code. This helps to slow down new
+ * operations if the drop_snapshot code isn't able to keep up.
*/
static void throttle_on_drops(struct btrfs_root *root)
{
if (!root->fs_info->open_ioctl_trans)
wait_current_trans(root);
mutex_unlock(&root->fs_info->trans_mutex);
-
throttle_on_drops(root);
}
{
struct btrfs_transaction *cur_trans;
struct btrfs_fs_info *info = root->fs_info;
+ int count = 0;
+
+ while (count < 4) {
+ unsigned long cur = trans->delayed_ref_updates;
+ trans->delayed_ref_updates = 0;
+ if (cur &&
+ trans->transaction->delayed_refs.num_heads_ready > 64) {
+ trans->delayed_ref_updates = 0;
+
+ /*
+ * do a full flush if the transaction is trying
+ * to close
+ */
+ if (trans->transaction->delayed_refs.flushing)
+ cur = 0;
+ btrfs_run_delayed_refs(trans, root, cur);
+ } else {
+ break;
+ }
+ count++;
+ }
mutex_lock(&info->trans_mutex);
cur_trans = info->running_transaction;
u64 end;
unsigned long index;
- while(1) {
+ while (1) {
ret = find_first_extent_bit(dirty_pages, start, &start, &end,
EXTENT_DIRTY);
if (ret)
break;
- while(start <= end) {
+ while (start <= end) {
cond_resched();
index = start >> PAGE_CACHE_SHIFT;
page_cache_release(page);
}
}
- while(1) {
+ while (1) {
ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
EXTENT_DIRTY);
if (ret)
break;
clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
- while(start <= end) {
+ while (start <= end) {
index = start >> PAGE_CACHE_SHIFT;
start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
page = find_get_page(btree_inode->i_mapping, index);
u64 old_root_bytenr;
struct btrfs_root *tree_root = root->fs_info->tree_root;
- btrfs_extent_post_op(trans, root);
btrfs_write_dirty_block_groups(trans, root);
- btrfs_extent_post_op(trans, root);
- while(1) {
+ ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
+ BUG_ON(ret);
+
+ while (1) {
old_root_bytenr = btrfs_root_bytenr(&root->root_item);
if (old_root_bytenr == root->node->start)
break;
btrfs_header_level(root->node));
btrfs_set_root_generation(&root->root_item, trans->transid);
- btrfs_extent_post_op(trans, root);
-
ret = btrfs_update_root(trans, tree_root,
&root->root_key,
&root->root_item);
BUG_ON(ret);
btrfs_write_dirty_block_groups(trans, root);
- btrfs_extent_post_op(trans, root);
+
+ ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
+ BUG_ON(ret);
}
return 0;
}
struct btrfs_fs_info *fs_info = root->fs_info;
struct list_head *next;
struct extent_buffer *eb;
+ int ret;
- btrfs_extent_post_op(trans, fs_info->tree_root);
+ ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
+ BUG_ON(ret);
eb = btrfs_lock_root_node(fs_info->tree_root);
- btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb, 0);
+ btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb);
btrfs_tree_unlock(eb);
free_extent_buffer(eb);
- btrfs_extent_post_op(trans, fs_info->tree_root);
+ ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
+ BUG_ON(ret);
- while(!list_empty(&fs_info->dirty_cowonly_roots)) {
+ while (!list_empty(&fs_info->dirty_cowonly_roots)) {
next = fs_info->dirty_cowonly_roots.next;
list_del_init(next);
root = list_entry(next, struct btrfs_root, dirty_list);
update_cowonly_root(trans, root);
+
+ ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
+ BUG_ON(ret);
}
return 0;
}
int err = 0;
u32 refs;
- while(1) {
+ while (1) {
ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
ARRAY_SIZE(gang),
BTRFS_ROOT_TRANS_TAG);
}
/*
+ * when dropping snapshots, we generate a ton of delayed refs, and it makes
+ * sense not to join the transaction while it is trying to flush the current
+ * queue of delayed refs out.
+ *
+ * This is used by the drop snapshot code only
+ */
+static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info)
+{
+ DEFINE_WAIT(wait);
+
+ mutex_lock(&info->trans_mutex);
+ while (info->running_transaction &&
+ info->running_transaction->delayed_refs.flushing) {
+ prepare_to_wait(&info->transaction_wait, &wait,
+ TASK_UNINTERRUPTIBLE);
+ mutex_unlock(&info->trans_mutex);
+ schedule();
+ mutex_lock(&info->trans_mutex);
+ finish_wait(&info->transaction_wait, &wait);
+ }
+ mutex_unlock(&info->trans_mutex);
+ return 0;
+}
+
+/*
* Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
* all of them
*/
int ret = 0;
int err;
- while(!list_empty(list)) {
+ while (!list_empty(list)) {
struct btrfs_root *root;
dirty = list_entry(list->prev, struct btrfs_dirty_root, list);
root = dirty->latest_root;
atomic_inc(&root->fs_info->throttles);
- while(1) {
+ while (1) {
+ /*
+ * we don't want to jump in and create a bunch of
+ * delayed refs if the transaction is starting to close
+ */
+ wait_transaction_pre_flush(tree_root->fs_info);
trans = btrfs_start_transaction(tree_root, 1);
+
+ /*
+ * we've joined a transaction, make sure it isn't
+ * closing right now
+ */
+ if (trans->transaction->delayed_refs.flushing) {
+ btrfs_end_transaction(trans, tree_root);
+ continue;
+ }
+
mutex_lock(&root->fs_info->drop_mutex);
ret = btrfs_drop_snapshot(trans, dirty->root);
- if (ret != -EAGAIN) {
+ if (ret != -EAGAIN)
break;
- }
mutex_unlock(&root->fs_info->drop_mutex);
err = btrfs_update_root(trans,
num_bytes -= btrfs_root_used(&dirty->root->root_item);
bytes_used = btrfs_root_used(&root->root_item);
if (num_bytes) {
+ mutex_lock(&root->fs_info->trans_mutex);
btrfs_record_root_in_trans(root);
+ mutex_unlock(&root->fs_info->trans_mutex);
btrfs_set_root_used(&root->root_item,
bytes_used - num_bytes);
}
btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
old = btrfs_lock_root_node(root);
- btrfs_cow_block(trans, root, old, NULL, 0, &old, 0);
+ btrfs_cow_block(trans, root, old, NULL, 0, &old);
btrfs_copy_root(trans, root, old, &tmp, objectid);
btrfs_tree_unlock(old);
parent_inode = pending->dentry->d_parent->d_inode;
parent_root = BTRFS_I(parent_inode)->root;
- trans = btrfs_start_transaction(parent_root, 1);
+ trans = btrfs_join_transaction(parent_root, 1);
/*
* insert the directory item
if (ret)
goto fail;
+ btrfs_i_size_write(parent_inode, parent_inode->i_size + namelen * 2);
+ ret = btrfs_update_inode(trans, parent_root, parent_inode);
+ BUG_ON(ret);
+
/* add the backref first */
ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root,
pending->root_key.objectid,
{
struct btrfs_pending_snapshot *pending;
struct list_head *head = &trans->transaction->pending_snapshots;
- struct list_head *cur;
int ret;
- list_for_each(cur, head) {
- pending = list_entry(cur, struct btrfs_pending_snapshot, list);
+ list_for_each_entry(pending, head, list) {
ret = create_pending_snapshot(trans, fs_info, pending);
BUG_ON(ret);
}
struct list_head *head = &trans->transaction->pending_snapshots;
int ret;
- while(!list_empty(head)) {
+ while (!list_empty(head)) {
pending = list_entry(head->next,
struct btrfs_pending_snapshot, list);
ret = finish_pending_snapshot(fs_info, pending);
struct extent_io_tree *pinned_copy;
DEFINE_WAIT(wait);
int ret;
+ int should_grow = 0;
+ unsigned long now = get_seconds();
+ int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
+
+ btrfs_run_ordered_operations(root, 0);
+
+ /* make a pass through all the delayed refs we have so far
+ * any runnings procs may add more while we are here
+ */
+ ret = btrfs_run_delayed_refs(trans, root, 0);
+ BUG_ON(ret);
+
+ cur_trans = trans->transaction;
+ /*
+ * set the flushing flag so procs in this transaction have to
+ * start sending their work down.
+ */
+ cur_trans->delayed_refs.flushing = 1;
+
+ ret = btrfs_run_delayed_refs(trans, root, 0);
+ BUG_ON(ret);
- INIT_LIST_HEAD(&dirty_fs_roots);
mutex_lock(&root->fs_info->trans_mutex);
- if (trans->transaction->in_commit) {
- cur_trans = trans->transaction;
- trans->transaction->use_count++;
+ INIT_LIST_HEAD(&dirty_fs_roots);
+ if (cur_trans->in_commit) {
+ cur_trans->use_count++;
mutex_unlock(&root->fs_info->trans_mutex);
btrfs_end_transaction(trans, root);
trans->transaction->in_commit = 1;
trans->transaction->blocked = 1;
- cur_trans = trans->transaction;
if (cur_trans->list.prev != &root->fs_info->trans_list) {
prev_trans = list_entry(cur_trans->list.prev,
struct btrfs_transaction, list);
}
}
+ if (now < cur_trans->start_time || now - cur_trans->start_time < 1)
+ should_grow = 1;
+
do {
int snap_pending = 0;
joined = cur_trans->num_joined;
if (cur_trans->num_writers > 1)
timeout = MAX_SCHEDULE_TIMEOUT;
- else
+ else if (should_grow)
timeout = 1;
mutex_unlock(&root->fs_info->trans_mutex);
- if (snap_pending) {
+ if (flush_on_commit || snap_pending) {
+ if (flush_on_commit)
+ btrfs_start_delalloc_inodes(root);
ret = btrfs_wait_ordered_extents(root, 1);
BUG_ON(ret);
}
- schedule_timeout(timeout);
+ /*
+ * rename don't use btrfs_join_transaction, so, once we
+ * set the transaction to blocked above, we aren't going
+ * to get any new ordered operations. We can safely run
+ * it here and no for sure that nothing new will be added
+ * to the list
+ */
+ btrfs_run_ordered_operations(root, 1);
+
+ smp_mb();
+ if (cur_trans->num_writers > 1 || should_grow)
+ schedule_timeout(timeout);
mutex_lock(&root->fs_info->trans_mutex);
finish_wait(&cur_trans->writer_wait, &wait);
} while (cur_trans->num_writers > 1 ||
- (cur_trans->num_joined != joined));
+ (should_grow && cur_trans->num_joined != joined));
ret = create_pending_snapshots(trans, root->fs_info);
BUG_ON(ret);
+ ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
+ BUG_ON(ret);
+
WARN_ON(cur_trans != trans->transaction);
/* btrfs_commit_tree_roots is responsible for getting the
btrfs_copy_pinned(root, pinned_copy);
trans->transaction->blocked = 0;
+
wake_up(&root->fs_info->transaction_throttle);
wake_up(&root->fs_info->transaction_wait);
mutex_lock(&root->fs_info->trans_mutex);
cur_trans->commit_done = 1;
+
root->fs_info->last_trans_committed = cur_trans->transid;
wake_up(&cur_trans->commit_wait);
kmem_cache_free(btrfs_trans_handle_cachep, trans);
- if (root->fs_info->closing) {
+ if (root->fs_info->closing)
drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
- }
return ret;
}