18 #include "transaction.h"
43  * Transaction states and transitions
45  * No running transaction (fs tree blocks are not modified)
50  * Transaction N [[TRANS_STATE_RUNNING]]
52  * | New trans handles can be attached to transaction N by calling all
57  * |  transaction N
59  * Transaction N [[TRANS_STATE_COMMIT_PREP]]
62  * | the race and the rest will wait for the winner to commit the transaction.
64  * | The winner will wait for previous running transaction to completely finish
67  * Transaction N [[TRANS_STATE_COMMIT_START]]
72  * | - Wait for current transaction to be committed by others.
76  * | to this running transaction.
78  * | transaction N+1.
81  * |  Caller is chosen to commit transaction N, and all other trans handle
84  * Transaction N [[TRANS_STATE_COMMIT_DOING]]
86  * | The heavy lifting transaction work is started.
93  * | transaction to finish and attach to transaction N+1.
98  * Transaction N [[TRANS_STATE_UNBLOCKED]]
99  * |						    Transaction N+1
104  * | At this stage, new transaction is allowed to   |
113  * Transaction N [[TRANS_STATE_COMPLETED]]	    V
114  *   All tree blocks and super blocks are written.  Transaction N+1
115  *   This transaction is finished and all its	    [[TRANS_STATE_COMMIT_START]]
143 void btrfs_put_transaction(struct btrfs_transaction *transaction)  in btrfs_put_transaction()  argument
145 	WARN_ON(refcount_read(&transaction->use_count) == 0);  in btrfs_put_transaction()
146 	if (refcount_dec_and_test(&transaction->use_count)) {  in btrfs_put_transaction()
147 		BUG_ON(!list_empty(&transaction->list));  in btrfs_put_transaction()
149 				&transaction->delayed_refs.href_root.rb_root));  in btrfs_put_transaction()
151 				&transaction->delayed_refs.dirty_extent_root));  in btrfs_put_transaction()
152 		if (transaction->delayed_refs.pending_csums)  in btrfs_put_transaction()
153 			btrfs_err(transaction->fs_info,  in btrfs_put_transaction()
155 				  transaction->delayed_refs.pending_csums);  in btrfs_put_transaction()
158 		 * because the transaction was aborted and a commit did not  in btrfs_put_transaction()
163 		while (!list_empty(&transaction->deleted_bgs)) {  in btrfs_put_transaction()
166 			cache = list_first_entry(&transaction->deleted_bgs,  in btrfs_put_transaction()
173 		WARN_ON(!list_empty(&transaction->dev_update_list));  in btrfs_put_transaction()
174 		kfree(transaction);  in btrfs_put_transaction()
180 	struct btrfs_transaction *cur_trans = trans->transaction;  in switch_commit_roots()
185 	 * At this point no one can be using this transaction to modify any tree  in switch_commit_roots()
186 	 * and no one can start another transaction to modify any tree either.  in switch_commit_roots()
265  * either allocate a new transaction or hop into the existing one
302 	 * current transaction, and commit it. If there is no transaction, just  in join_transaction()
309 	 * JOIN_NOLOCK only happens during the transaction commit, so  in join_transaction()
324 		 * someone started a transaction after we unlocked.  Make sure  in join_transaction()
349 	 * commit the transaction.  in join_transaction()
362 	 * although the tree mod log is per file system and not per transaction,  in join_transaction()
363 	 * the log must never go across transaction boundaries.  in join_transaction()
367 		WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n");  in join_transaction()
369 		WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n");  in join_transaction()
400  * is properly recorded in a given transaction.  This is required to make sure
401  * the old root from before we joined the transaction is deleted when the
402  * transaction commits.
441 		 * this transaction.  in record_root_in_trans()
469 	struct btrfs_transaction *cur_trans = trans->transaction;  in btrfs_add_dropped_root()
471 	/* Add ourselves to the transaction dropped list */  in btrfs_add_dropped_root()
516 /* wait for commit against the current transaction to become unblocked
517  * when this is done, it is safe to start a new transaction, but the current
518  * transaction might not be fully on disk.
593 	 * Do the reservation before we join the transaction so we can do all  in start_transaction()
603 		 * transaction that could free this reserved space prematurely  in start_transaction()
613 		 * we only do 1 enospc flushing cycle per transaction start.  We  in start_transaction()
666 	 * If we are JOIN_NOLOCK we're already committing a transaction and  in start_transaction()
673 	 * transaction and commit it, so we needn't do sb_start_intwrite().   in start_transaction()
697 	h->transaction = cur_trans;  in start_transaction()
713 		trace_btrfs_space_reservation(fs_info, "transaction",  in start_transaction()
721 	 * Now that we have found a transaction to be a part of, convert the  in start_transaction()
722 	 * qgroup reservation from prealloc to pertrans. A different transaction  in start_transaction()
748 	 * transaction.  in start_transaction()
756 		 * The transaction handle is fully initialized and linked with  in start_transaction()
807  * Similar to regular join but it never starts a transaction when none is
810  * happen if the transaction commit already started but it's not yet in the
820  * btrfs_attach_transaction() - catch the running transaction
822  * It is used when we want to commit the current the transaction, but
826  * running transaction. But it is possible that the inactive transaction
828  * inactive transaction in the fs when -ENOENT is returned, you should
839  * btrfs_attach_transaction_barrier() - catch the running transaction
863 /* Wait for a transaction commit to reach at least the given state. */
889 		 * A transaction isn't really completed until all of the  in wait_for_commit()
892 		 * transaction. Wait for those.  in wait_for_commit()
918 		/* find specified transaction */  in btrfs_wait_for_commit()
935 		 * The specified transaction doesn't exist, or we  in btrfs_wait_for_commit()
944 		/* find newest transaction that is committing | committed */  in btrfs_wait_for_commit()
975 	struct btrfs_transaction *cur_trans = trans->transaction;  in btrfs_should_end_transaction()
1001 	trace_btrfs_space_reservation(fs_info, "transaction",  in btrfs_trans_release_metadata()
1012 	struct btrfs_transaction *cur_trans = trans->transaction;  in __btrfs_end_transaction()
1104 		 * the transaction (through extent_io_tree_release()).  in btrfs_write_marked_extents()
1127  * those extents are on disk for transaction or log commit.  We wait
1144 		 * When committing the transaction, we'll remove any entries  in __btrfs_wait_marked_extents()
1147 		 * concurrently - we do it only at transaction commit time when  in __btrfs_wait_marked_extents()
1208  * This function ensures such extents are persisted on disk for transaction or
1211  * @trans: transaction whose dirty pages we'd like to write
1217 	struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages;  in btrfs_write_and_wait_transaction()
1226 	extent_io_tree_release(&trans->transaction->dirty_pages);  in btrfs_write_and_wait_transaction()
1286 	struct list_head *dirty_bgs = &trans->transaction->dirty_bgs;  in commit_cowonly_roots()
1287 	struct list_head *io_bgs = &trans->transaction->io_bgs;  in commit_cowonly_roots()
1293 	 * At this point no one can be using this transaction to modify any tree  in commit_cowonly_roots()
1294 	 * and no one can start another transaction to modify any tree either.  in commit_cowonly_roots()
1296 	ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING);  in commit_cowonly_roots()
1330 			      &trans->transaction->switch_commits);  in commit_cowonly_roots()
1427 	 * At this point no one can be using this transaction to modify any tree  in commit_fs_roots()
1428 	 * and no one can start another transaction to modify any tree either.  in commit_fs_roots()
1430 	ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING);  in commit_fs_roots()
1468 					&trans->transaction->switch_commits);  in commit_fs_roots()
1529  * roots inside one transaction and write all btree into disk, to make
1590 	 * Now we do a simplified commit transaction, which will:  in qgroup_account_snapshot()
1608 			"Error while writing out transaction for qgroup");  in qgroup_account_snapshot()
1624  * transaction commit.  This does the actual creation.
1627  * If the error which may affect the commitment of the current transaction
1663 	 * We're inside a transaction and must make sure that any potential  in create_pending_snapshot()
1703 	trace_btrfs_space_reservation(fs_info, "transaction",  in create_pending_snapshot()
1742 	if (ret) {	/* Transaction aborted */  in create_pending_snapshot()
1910 	struct list_head *head = &trans->transaction->pending_snapshots;  in create_pending_snapshots()
1977 	/* Kick the transaction kthread. */  in btrfs_commit_transaction_async()
1981 	/* take transaction reference */  in btrfs_commit_transaction_async()
1982 	cur_trans = trans->transaction;  in btrfs_commit_transaction_async()
1988 	 * Wait for the current transaction commit to start and block  in btrfs_commit_transaction_async()
1989 	 * subsequent transaction joins  in btrfs_commit_transaction_async()
2001 	struct btrfs_transaction *cur_trans = trans->transaction;  in cleanup_transaction()
2010 	 * If the transaction is removed from the list, it means this  in cleanup_transaction()
2011 	 * transaction has been committed successfully, so it is impossible  in cleanup_transaction()
2032 	 * Now that we know no one else is still using the transaction we can  in cleanup_transaction()
2033 	 * remove the transaction from the list of transactions. This avoids  in cleanup_transaction()
2034 	 * the transaction kthread from cleaning up the transaction while some  in cleanup_transaction()
2036 	 * on things like log trees, as it forces the transaction kthread to  in cleanup_transaction()
2037 	 * wait for this transaction to be cleaned up by us.  in cleanup_transaction()
2043 	btrfs_cleanup_one_transaction(trans->transaction, fs_info);  in cleanup_transaction()
2063 	 * pauses scrub, then starts and commits a transaction before unpausing  in cleanup_transaction()
2064 	 * scrub. If the transaction commit is being done by the relocation  in cleanup_transaction()
2079  * transaction and remove them from the list
2100 	 * from already being in a transaction and our join_transaction doesn't  in btrfs_start_delalloc_flush()
2123  * Add a pending snapshot associated with the given transaction handle to the
2124  * respective handle. This must be called after the transaction commit started
2132 	struct btrfs_transaction *cur_trans = trans->transaction;  in add_pending_snapshot()
2155 	struct btrfs_transaction *cur_trans = trans->transaction;  in btrfs_commit_transaction()
2176 	 * We only want one transaction commit doing the flushing so we do not  in btrfs_commit_transaction()
2200 		 * will wait for the transaction to commit if it  in btrfs_commit_transaction()
2273 		 * The previous transaction was aborted and was already removed  in btrfs_commit_transaction()
2306 	 * The thread has started/joined the transaction thus it holds the  in btrfs_commit_transaction()
2326 	 * transaction. Otherwise if this transaction commits before the ordered  in btrfs_commit_transaction()
2336 	 * commit the transaction.  We could have started a join before setting  in btrfs_commit_transaction()
2345 	 * The thread has started/joined the transaction thus it holds the  in btrfs_commit_transaction()
2366 	 * do an async commit but somebody else started before the transaction  in btrfs_commit_transaction()
2416 	WARN_ON(cur_trans != trans->transaction);  in btrfs_commit_transaction()
2486 	 * Before changing the transaction state to TRANS_STATE_UNBLOCKED and  in btrfs_commit_transaction()
2489 	 * start a new transaction and commit a log tree before we commit our  in btrfs_commit_transaction()
2512 				      "Error while writing out transaction");  in btrfs_commit_transaction()
2585 	btrfs_warn(fs_info, "Skipping commit of aborted transaction.");  in btrfs_commit_transaction()
2643  * We only mark the transaction aborted and then set the file system read-only.
2649  * further cleanup. The transaction should complete as it normally would
2662 	WRITE_ONCE(trans->transaction->aborted, errno);  in __btrfs_abort_transaction()
2665 	/* Wake up anybody who may be waiting on this transaction */  in __btrfs_abort_transaction()