#include "transaction.h"
#include "locking.h"
#include "ref-cache.h"
+#include "tree-log.h"
static int total_trans = 0;
extern struct kmem_cache *btrfs_trans_handle_cachep;
}
}
+/*
+ * either allocate a new transaction or hop into the existing one
+ */
static noinline int join_transaction(struct btrfs_root *root)
{
struct btrfs_transaction *cur_trans;
return 0;
}
-static noinline int record_root_in_trans(struct btrfs_root *root)
+/*
+ * 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)
{
struct btrfs_dirty_root *dirty;
u64 running_trans_id = root->fs_info->running_transaction->transid;
spin_lock_init(&dirty->root->node_lock);
spin_lock_init(&dirty->root->list_lock);
mutex_init(&dirty->root->objectid_mutex);
+ mutex_init(&dirty->root->log_mutex);
INIT_LIST_HEAD(&dirty->root->dead_list);
dirty->root->node = root->commit_root;
dirty->root->commit_root = NULL;
return 0;
}
+/* wait for commit against the current transaction to become unblocked
+ * when this is done, it is safe to start a new transaction, but the current
+ * transaction might not be fully on disk.
+ */
static void wait_current_trans(struct btrfs_root *root)
{
struct btrfs_transaction *cur_trans;
}
}
-struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
+static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
int num_blocks, int wait)
{
struct btrfs_trans_handle *h =
int ret;
mutex_lock(&root->fs_info->trans_mutex);
- if ((wait == 1 && !root->fs_info->open_ioctl_trans) || wait == 2)
+ if (!root->fs_info->log_root_recovering &&
+ ((wait == 1 && !root->fs_info->open_ioctl_trans) || wait == 2))
wait_current_trans(root);
ret = join_transaction(root);
BUG_ON(ret);
- record_root_in_trans(root);
+ btrfs_record_root_in_trans(root);
h->transid = root->fs_info->running_transaction->transid;
h->transaction = root->fs_info->running_transaction;
h->blocks_reserved = num_blocks;
return start_transaction(r, num_blocks, 2);
}
-
+/* wait for a transaction commit to be fully complete */
static noinline int wait_for_commit(struct btrfs_root *root,
struct btrfs_transaction *commit)
{
return 0;
}
+/*
+ * 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)
{
struct btrfs_fs_info *info = root->fs_info;
return __btrfs_end_transaction(trans, root, 1);
}
-
-int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
- struct btrfs_root *root)
+/*
+ * when btree blocks are allocated, they have some corresponding bits set for
+ * them in one of two extent_io trees. This is used to make sure all of
+ * those extents are on disk for transaction or log commit
+ */
+int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
+ struct extent_io_tree *dirty_pages)
{
int ret;
int err = 0;
int werr = 0;
- struct extent_io_tree *dirty_pages;
struct page *page;
struct inode *btree_inode = root->fs_info->btree_inode;
u64 start = 0;
u64 end;
unsigned long index;
- if (!trans || !trans->transaction) {
- return filemap_write_and_wait(btree_inode->i_mapping);
- }
- dirty_pages = &trans->transaction->dirty_pages;
while(1) {
ret = find_first_extent_bit(dirty_pages, start, &start, &end,
EXTENT_DIRTY);
if (ret)
break;
while(start <= end) {
- if (btrfs_congested_async(root->fs_info, 0))
- congestion_wait(WRITE, HZ/10);
cond_resched();
index = start >> PAGE_CACHE_SHIFT;
start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
- page = find_lock_page(btree_inode->i_mapping, index);
+ page = find_get_page(btree_inode->i_mapping, index);
if (!page)
continue;
+
+ btree_lock_page_hook(page);
+ if (!page->mapping) {
+ unlock_page(page);
+ page_cache_release(page);
+ continue;
+ }
+
if (PageWriteback(page)) {
if (PageDirty(page))
wait_on_page_writeback(page);
if (!page)
continue;
if (PageDirty(page)) {
- lock_page(page);
+ btree_lock_page_hook(page);
+ wait_on_page_writeback(page);
err = write_one_page(page, 0);
if (err)
werr = err;
return werr;
}
+int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
+ struct btrfs_root *root)
+{
+ if (!trans || !trans->transaction) {
+ struct inode *btree_inode;
+ btree_inode = root->fs_info->btree_inode;
+ return filemap_write_and_wait(btree_inode->i_mapping);
+ }
+ return btrfs_write_and_wait_marked_extents(root,
+ &trans->transaction->dirty_pages);
+}
+
+/*
+ * this is used to update the root pointer in the tree of tree roots.
+ *
+ * But, in the case of the extent allocation tree, updating the root
+ * pointer may allocate blocks which may change the root of the extent
+ * allocation tree.
+ *
+ * So, this loops and repeats and makes sure the cowonly root didn't
+ * change while the root pointer was being updated in the metadata.
+ */
static int update_cowonly_root(struct btrfs_trans_handle *trans,
struct btrfs_root *root)
{
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) {
old_root_bytenr = btrfs_root_bytenr(&root->root_item);
if (old_root_bytenr == root->node->start)
root->node->start);
btrfs_set_root_level(&root->root_item,
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);
}
return 0;
}
+/*
+ * update all the cowonly tree roots on disk
+ */
int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
struct btrfs_root *root)
{
struct btrfs_fs_info *fs_info = root->fs_info;
struct list_head *next;
+ struct extent_buffer *eb;
+
+ btrfs_extent_post_op(trans, fs_info->tree_root);
+
+ eb = btrfs_lock_root_node(fs_info->tree_root);
+ btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb, 0);
+ btrfs_tree_unlock(eb);
+ free_extent_buffer(eb);
+
+ btrfs_extent_post_op(trans, fs_info->tree_root);
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);
}
return 0;
}
+/*
+ * dead roots are old snapshots that need to be deleted. This allocates
+ * a dirty root struct and adds it into the list of dead roots that need to
+ * be deleted
+ */
int btrfs_add_dead_root(struct btrfs_root *root, struct btrfs_root *latest)
{
struct btrfs_dirty_root *dirty;
return 0;
}
+/*
+ * at transaction commit time we need to schedule the old roots for
+ * deletion via btrfs_drop_snapshot. This runs through all the
+ * reference counted roots that were modified in the current
+ * transaction and puts them into the drop list
+ */
static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
struct radix_tree_root *radix,
struct list_head *list)
BUG_ON(!root->ref_tree);
dirty = root->dirty_root;
+ btrfs_free_log(trans, root);
+ btrfs_free_reloc_root(trans, root);
+
if (root->commit_root == root->node) {
WARN_ON(root->node->start !=
btrfs_root_bytenr(&root->root_item));
root->node->start);
btrfs_set_root_level(&root->root_item,
btrfs_header_level(root->node));
+ btrfs_set_root_generation(&root->root_item,
+ root->root_key.offset);
+
err = btrfs_insert_root(trans, root->fs_info->tree_root,
&root->root_key,
&root->root_item);
return err;
}
+/*
+ * defrag a given btree. If cacheonly == 1, this won't read from the disk,
+ * otherwise every leaf in the btree is read and defragged.
+ */
int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
{
struct btrfs_fs_info *info = root->fs_info;
return 0;
}
+/*
+ * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
+ * all of them
+ */
static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
struct list_head *list)
{
root = dirty->latest_root;
atomic_inc(&root->fs_info->throttles);
- mutex_lock(&root->fs_info->drop_mutex);
while(1) {
trans = btrfs_start_transaction(tree_root, 1);
+ mutex_lock(&root->fs_info->drop_mutex);
ret = btrfs_drop_snapshot(trans, dirty->root);
if (ret != -EAGAIN) {
break;
}
+ mutex_unlock(&root->fs_info->drop_mutex);
err = btrfs_update_root(trans,
tree_root,
ret = btrfs_end_transaction(trans, tree_root);
BUG_ON(ret);
- mutex_unlock(&root->fs_info->drop_mutex);
btrfs_btree_balance_dirty(tree_root, nr);
cond_resched();
- mutex_lock(&root->fs_info->drop_mutex);
}
BUG_ON(ret);
atomic_dec(&root->fs_info->throttles);
wake_up(&root->fs_info->transaction_throttle);
- mutex_lock(&root->fs_info->alloc_mutex);
num_bytes -= btrfs_root_used(&dirty->root->root_item);
bytes_used = btrfs_root_used(&root->root_item);
if (num_bytes) {
- record_root_in_trans(root);
+ btrfs_record_root_in_trans(root);
btrfs_set_root_used(&root->root_item,
bytes_used - num_bytes);
}
- mutex_unlock(&root->fs_info->alloc_mutex);
ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
if (ret) {
ret = btrfs_end_transaction(trans, tree_root);
BUG_ON(ret);
- ret = btrfs_remove_leaf_refs(root, max_useless);
+ ret = btrfs_remove_leaf_refs(root, max_useless, 0);
BUG_ON(ret);
free_extent_buffer(dirty->root->node);
return ret;
}
+/*
+ * new snapshots need to be created at a very specific time in the
+ * transaction commit. This does the actual creation
+ */
static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
struct btrfs_fs_info *fs_info,
struct btrfs_pending_snapshot *pending)
if (ret)
goto fail;
+ btrfs_record_root_in_trans(root);
+ btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
key.objectid = objectid;
- key.offset = 1;
+ key.offset = trans->transid;
btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
old = btrfs_lock_root_node(root);
btrfs_set_root_bytenr(new_root_item, tmp->start);
btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
+ btrfs_set_root_generation(new_root_item, trans->transid);
ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
new_root_item);
btrfs_tree_unlock(tmp);
return ret;
}
+/*
+ * create all the snapshots we've scheduled for creation
+ */
static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
struct btrfs_fs_info *fs_info)
{
int ret;
INIT_LIST_HEAD(&dirty_fs_roots);
-
mutex_lock(&root->fs_info->trans_mutex);
if (trans->transaction->in_commit) {
cur_trans = trans->transaction;
WARN_ON(cur_trans != trans->transaction);
+ /* btrfs_commit_tree_roots is responsible for getting the
+ * various roots consistent with each other. Every pointer
+ * in the tree of tree roots has to point to the most up to date
+ * root for every subvolume and other tree. So, we have to keep
+ * the tree logging code from jumping in and changing any
+ * of the trees.
+ *
+ * At this point in the commit, there can't be any tree-log
+ * writers, but a little lower down we drop the trans mutex
+ * and let new people in. By holding the tree_log_mutex
+ * from now until after the super is written, we avoid races
+ * with the tree-log code.
+ */
+ mutex_lock(&root->fs_info->tree_log_mutex);
+ /*
+ * keep tree reloc code from adding new reloc trees
+ */
+ mutex_lock(&root->fs_info->tree_reloc_mutex);
+
+
ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
&dirty_fs_roots);
BUG_ON(ret);
+ /* add_dirty_roots gets rid of all the tree log roots, it is now
+ * safe to free the root of tree log roots
+ */
+ btrfs_free_log_root_tree(trans, root->fs_info);
+
ret = btrfs_commit_tree_roots(trans, root);
BUG_ON(ret);
chunk_root->node->start);
btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
btrfs_header_level(chunk_root->node));
+ btrfs_set_super_chunk_root_generation(&root->fs_info->super_copy,
+ btrfs_header_generation(chunk_root->node));
+
+ if (!root->fs_info->log_root_recovering) {
+ btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
+ btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
+ }
+
memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
sizeof(root->fs_info->super_copy));
BUG_ON(ret);
write_ctree_super(trans, root);
- btrfs_finish_extent_commit(trans, root, pinned_copy);
- mutex_lock(&root->fs_info->trans_mutex);
+ /*
+ * the super is written, we can safely allow the tree-loggers
+ * to go about their business
+ */
+ mutex_unlock(&root->fs_info->tree_log_mutex);
+ btrfs_finish_extent_commit(trans, root, pinned_copy);
kfree(pinned_copy);
+ btrfs_drop_dead_reloc_roots(root);
+ mutex_unlock(&root->fs_info->tree_reloc_mutex);
+
+ 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);
return ret;
}
+/*
+ * interface function to delete all the snapshots we have scheduled for deletion
+ */
int btrfs_clean_old_snapshots(struct btrfs_root *root)
{
struct list_head dirty_roots;