17 #include "transaction.h"
30  * Transaction states and transitions
32  * No running transaction (fs tree blocks are not modified)
37  * Transaction N [[TRANS_STATE_RUNNING]]
39  * | New trans handles can be attached to transaction N by calling all
44  * |  transaction N
46  * Transaction N [[TRANS_STATE_COMMIT_START]]
48  * | Will wait for previous running transaction to completely finish if there
54  * | - Wait for current transaction to be committed by others.
58  * | to this running transaction.
60  * | transaction N+1.
63  * |  Caller is chosen to commit transaction N, and all other trans handle
66  * Transaction N [[TRANS_STATE_COMMIT_DOING]]
68  * | The heavy lifting transaction work is started.
75  * | transaction to finish and attach to transaction N+1.
80  * Transaction N [[TRANS_STATE_UNBLOCKED]]
81  * |						    Transaction N+1
86  * | At this stage, new transaction is allowed to   |
95  * Transaction N [[TRANS_STATE_COMPLETED]]	    V
96  *   All tree blocks and super blocks are written.  Transaction N+1
97  *   This transaction is finished and all its	    [[TRANS_STATE_COMMIT_START]]
124 void btrfs_put_transaction(struct btrfs_transaction *transaction)  in btrfs_put_transaction()  argument
126 	WARN_ON(refcount_read(&transaction->use_count) == 0);  in btrfs_put_transaction()
127 	if (refcount_dec_and_test(&transaction->use_count)) {  in btrfs_put_transaction()
128 		BUG_ON(!list_empty(&transaction->list));  in btrfs_put_transaction()
130 				&transaction->delayed_refs.href_root.rb_root));  in btrfs_put_transaction()
132 				&transaction->delayed_refs.dirty_extent_root));  in btrfs_put_transaction()
133 		if (transaction->delayed_refs.pending_csums)  in btrfs_put_transaction()
134 			btrfs_err(transaction->fs_info,  in btrfs_put_transaction()
136 				  transaction->delayed_refs.pending_csums);  in btrfs_put_transaction()
139 		 * because the transaction was aborted and a commit did not  in btrfs_put_transaction()
144 		while (!list_empty(&transaction->deleted_bgs)) {  in btrfs_put_transaction()
147 			cache = list_first_entry(&transaction->deleted_bgs,  in btrfs_put_transaction()
154 		WARN_ON(!list_empty(&transaction->dev_update_list));  in btrfs_put_transaction()
155 		kfree(transaction);  in btrfs_put_transaction()
161 	struct btrfs_transaction *cur_trans = trans->transaction;  in switch_commit_roots()
166 	 * At this point no one can be using this transaction to modify any tree  in switch_commit_roots()
167 	 * and no one can start another transaction to modify any tree either.  in switch_commit_roots()
246  * either allocate a new transaction or hop into the existing one
282 	 * If we are ATTACH, we just want to catch the current transaction,  in join_transaction()
283 	 * and commit it. If there is no transaction, just return ENOENT.  in join_transaction()
289 	 * JOIN_NOLOCK only happens during the transaction commit, so  in join_transaction()
304 		 * someone started a transaction after we unlocked.  Make sure  in join_transaction()
329 	 * commit the transaction.  in join_transaction()
342 	 * although the tree mod log is per file system and not per transaction,  in join_transaction()
343 	 * the log must never go across transaction boundaries.  in join_transaction()
347 		WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n");  in join_transaction()
349 		WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n");  in join_transaction()
382  * is properly recorded in a given transaction.  This is required to make sure
383  * the old root from before we joined the transaction is deleted when the
384  * transaction commits.
423 		 * this transaction.  in record_root_in_trans()
451 	struct btrfs_transaction *cur_trans = trans->transaction;  in btrfs_add_dropped_root()
453 	/* Add ourselves to the transaction dropped list */  in btrfs_add_dropped_root()
498 /* wait for commit against the current transaction to become unblocked
499  * when this is done, it is safe to start a new transaction, but the current
500  * transaction might not be fully on disk.
575 	 * Do the reservation before we join the transaction so we can do all  in start_transaction()
590 		 * we only do 1 enospc flushing cycle per transaction start.  We  in start_transaction()
642 	 * If we are JOIN_NOLOCK we're already committing a transaction and  in start_transaction()
649 	 * transaction and commit it, so we needn't do sb_start_intwrite().   in start_transaction()
673 	h->transaction = cur_trans;  in start_transaction()
689 		trace_btrfs_space_reservation(fs_info, "transaction",  in start_transaction()
716 	 * transaction.  in start_transaction()
724 		 * The transaction handle is fully initialized and linked with  in start_transaction()
775  * Similar to regular join but it never starts a transaction when none is
785  * btrfs_attach_transaction() - catch the running transaction
787  * It is used when we want to commit the current the transaction, but
791  * running transaction. But it is possible that the inactive transaction
793  * inactive transaction in the fs when -ENOENT is returned, you should
804  * btrfs_attach_transaction_barrier() - catch the running transaction
823 /* Wait for a transaction commit to reach at least the given state. */
849 		 * A transaction isn't really completed until all of the  in wait_for_commit()
852 		 * transaction. Wait for those.  in wait_for_commit()
878 		/* find specified transaction */  in btrfs_wait_for_commit()
895 		 * The specified transaction doesn't exist, or we  in btrfs_wait_for_commit()
904 		/* find newest transaction that is committing | committed */  in btrfs_wait_for_commit()
944 	struct btrfs_transaction *cur_trans = trans->transaction;  in btrfs_should_end_transaction()
967 	trace_btrfs_space_reservation(fs_info, "transaction",  in btrfs_trans_release_metadata()
978 	struct btrfs_transaction *cur_trans = trans->transaction;  in __btrfs_end_transaction()
1071 		 * the transaction (through extent_io_tree_release()).  in btrfs_write_marked_extents()
1095  * those extents are on disk for transaction or log commit.  We wait
1112 		 * When committing the transaction, we'll remove any entries  in __btrfs_wait_marked_extents()
1115 		 * concurrently - we do it only at transaction commit time when  in __btrfs_wait_marked_extents()
1176  * This function ensures such extents are persisted on disk for transaction or
1179  * @trans: transaction whose dirty pages we'd like to write
1185 	struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages;  in btrfs_write_and_wait_transaction()
1194 	extent_io_tree_release(&trans->transaction->dirty_pages);  in btrfs_write_and_wait_transaction()
1254 	struct list_head *dirty_bgs = &trans->transaction->dirty_bgs;  in commit_cowonly_roots()
1255 	struct list_head *io_bgs = &trans->transaction->io_bgs;  in commit_cowonly_roots()
1261 	 * At this point no one can be using this transaction to modify any tree  in commit_cowonly_roots()
1262 	 * and no one can start another transaction to modify any tree either.  in commit_cowonly_roots()
1264 	ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING);  in commit_cowonly_roots()
1298 			      &trans->transaction->switch_commits);  in commit_cowonly_roots()
1395 	 * At this point no one can be using this transaction to modify any tree  in commit_fs_roots()
1396 	 * and no one can start another transaction to modify any tree either.  in commit_fs_roots()
1398 	ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING);  in commit_fs_roots()
1436 					&trans->transaction->switch_commits);  in commit_fs_roots()
1497  * roots inside one transaction and write all btree into disk, to make
1558 	 * Now we do a simplified commit transaction, which will:  in qgroup_account_snapshot()
1576 			"Error while writing out transaction for qgroup");  in qgroup_account_snapshot()
1592  * transaction commit.  This does the actual creation.
1595  * If the error which may affect the commitment of the current transaction
1657 	trace_btrfs_space_reservation(fs_info, "transaction",  in create_pending_snapshot()
1696 	if (ret) {	/* Transaction aborted */  in create_pending_snapshot()
1863 	struct list_head *head = &trans->transaction->pending_snapshots;  in create_pending_snapshots()
1930 	/* Kick the transaction kthread. */  in btrfs_commit_transaction_async()
1934 	/* take transaction reference */  in btrfs_commit_transaction_async()
1935 	cur_trans = trans->transaction;  in btrfs_commit_transaction_async()
1941 	 * Wait for the current transaction commit to start and block  in btrfs_commit_transaction_async()
1942 	 * subsequent transaction joins  in btrfs_commit_transaction_async()
1954 	struct btrfs_transaction *cur_trans = trans->transaction;  in cleanup_transaction()
1963 	 * If the transaction is removed from the list, it means this  in cleanup_transaction()
1964 	 * transaction has been committed successfully, so it is impossible  in cleanup_transaction()
1985 	 * Now that we know no one else is still using the transaction we can  in cleanup_transaction()
1986 	 * remove the transaction from the list of transactions. This avoids  in cleanup_transaction()
1987 	 * the transaction kthread from cleaning up the transaction while some  in cleanup_transaction()
1989 	 * on things like log trees, as it forces the transaction kthread to  in cleanup_transaction()
1990 	 * wait for this transaction to be cleaned up by us.  in cleanup_transaction()
1996 	btrfs_cleanup_one_transaction(trans->transaction, fs_info);  in cleanup_transaction()
2019  * transaction and remove them from the list
2040 	 * from already being in a transaction and our join_transaction doesn't  in btrfs_start_delalloc_flush()
2063  * Add a pending snapshot associated with the given transaction handle to the
2064  * respective handle. This must be called after the transaction commit started
2072 	struct btrfs_transaction *cur_trans = trans->transaction;  in add_pending_snapshot()
2095 	struct btrfs_transaction *cur_trans = trans->transaction;  in btrfs_commit_transaction()
2114 	 * We only want one transaction commit doing the flushing so we do not  in btrfs_commit_transaction()
2138 		 * will wait for the transaction to commit if it  in btrfs_commit_transaction()
2213 		 * The previous transaction was aborted and was already removed  in btrfs_commit_transaction()
2241 	 * The thread has started/joined the transaction thus it holds the  in btrfs_commit_transaction()
2261 	 * transaction. Otherwise if this transaction commits before the ordered  in btrfs_commit_transaction()
2271 	 * commit the transaction.  We could have started a join before setting  in btrfs_commit_transaction()
2280 	 * The thread has started/joined the transaction thus it holds the  in btrfs_commit_transaction()
2301 	 * do an async commit but somebody else started before the transaction  in btrfs_commit_transaction()
2351 	WARN_ON(cur_trans != trans->transaction);  in btrfs_commit_transaction()
2358 	 * Since the transaction is done, we can apply the pending changes  in btrfs_commit_transaction()
2359 	 * before the next transaction.  in btrfs_commit_transaction()
2427 	 * Before changing the transaction state to TRANS_STATE_UNBLOCKED and  in btrfs_commit_transaction()
2430 	 * start a new transaction and commit a log tree before we commit our  in btrfs_commit_transaction()
2448 				      "Error while writing out transaction");  in btrfs_commit_transaction()
2455 	 * in this transaction. So it's now safe to free the redirtyied extent  in btrfs_commit_transaction()
2528 	btrfs_warn(fs_info, "Skipping commit of aborted transaction.");  in btrfs_commit_transaction()