From: Cyrill Gorcunov <gorcunov@gmail.com> To: tml <tarantool-patches@dev.tarantool.org> Subject: [Tarantool-patches] [PATCH 4/4] box/txn: fix nil dereference in txn_rollback Date: Mon, 17 Feb 2020 18:59:53 +0300 [thread overview] Message-ID: <20200217155953.25803-5-gorcunov@gmail.com> (raw) In-Reply-To: <20200217155953.25803-1-gorcunov@gmail.com> In case if we get failed in allocation of new journal entry the txn_rollback will try to derefernce nil pointer | txn_write | fiber_set_txn(fiber(), NULL); // zap fiber's storage.txn | txn_write_to_wal(txn); | journal_entry_new(..., txn_entry_done_cb, ...) | if (req == NULL) | txn_rollback(txn); | assert(txn == in_txn()); // in_txn()=nil, triggers This is because there are two call site: - when transaction is complete the wal engine will call txn_entry_complete_cb completion handler and since it is async in terms of threads (wal is a separate thread) it setup the txn it processes into a fiber's storage thus it expects the current storage is nil - in turn error may happen and we need to run a rollback procedure, there fiber's storage keeps current txn; Taking this into account we clean txn inside txn_write_to_wal right before journal_write is called and restore it back on error. This allows the caller code to run txn_rollback in a more consistent way. https://github.com/tarantool/tarantool/issues/4776 Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com> --- src/box/txn.c | 27 +++++++++++++++++++++------ 1 file changed, 21 insertions(+), 6 deletions(-) diff --git a/src/box/txn.c b/src/box/txn.c index a4ca48224..68365ea0b 100644 --- a/src/box/txn.c +++ b/src/box/txn.c @@ -495,10 +495,8 @@ txn_write_to_wal(struct txn *txn) &txn->region, txn_entry_complete_cb, txn); - if (req == NULL) { - txn_rollback(txn); + if (req == NULL) return -1; - } struct txn_stmt *stmt; struct xrow_header **remote_row = req->rows; @@ -519,8 +517,20 @@ txn_write_to_wal(struct txn *txn) assert(remote_row == req->rows + txn->n_applier_rows); assert(local_row == remote_row + txn->n_new_rows); - /* Send the entry to the journal. */ + /* + * Queue the entry for processing in journal + * engine. The semantics of complete_cb implies + * that fiber's txn (kept in storage) is nil + * becase WAL is a separet thread, for this + * sake we zap it here. + * + * Still this is messy since the caller runs + * txn_rollback if something bad happened. Thus + * restore the former txn on error path. + */ + fiber_set_txn(fiber(), NULL); if (journal_write(req) < 0) { + fiber_set_txn(fiber(), txn); diag_set(ClientError, ER_WAL_IO); diag_log(); return -1; @@ -583,8 +593,13 @@ txn_write(struct txn *txn) fiber_set_txn(fiber(), NULL); return 0; } - fiber_set_txn(fiber(), NULL); - return txn_write_to_wal(txn); + + if (txn_write_to_wal(txn) != 0) { + txn_rollback(txn); + return -1; + } + + return 0; } int -- 2.20.1
next prev parent reply other threads:[~2020-02-17 16:00 UTC|newest] Thread overview: 15+ messages / expand[flat|nested] mbox.gz Atom feed top 2020-02-17 15:59 [Tarantool-patches] [PATCH 0/4] box/txn: fix nil dereference on txn error path Cyrill Gorcunov 2020-02-17 15:59 ` [Tarantool-patches] [PATCH 1/4] box/txn: fix void args mess Cyrill Gorcunov 2020-02-17 17:13 ` Konstantin Osipov 2020-02-17 15:59 ` [Tarantool-patches] [PATCH 2/4] box/journal: sanitize completion naming Cyrill Gorcunov 2020-02-17 17:14 ` Konstantin Osipov 2020-02-17 15:59 ` [Tarantool-patches] [PATCH 3/4] box/txn: rename txn_entry_done_cb to txn_entry_complete_cb Cyrill Gorcunov 2020-02-17 17:14 ` Konstantin Osipov 2020-02-17 15:59 ` Cyrill Gorcunov [this message] 2020-02-17 17:19 ` [Tarantool-patches] [PATCH 4/4] box/txn: fix nil dereference in txn_rollback Konstantin Osipov 2020-02-17 19:05 ` Cyrill Gorcunov 2020-02-17 20:01 ` Konstantin Osipov 2020-02-17 20:01 ` Konstantin Osipov 2020-02-17 20:05 ` Cyrill Gorcunov 2020-02-17 20:34 ` Cyrill Gorcunov 2020-02-17 17:25 ` Georgy Kirichenko
Reply instructions: You may reply publicly to this message via plain-text email using any one of the following methods: * Save the following mbox file, import it into your mail client, and reply-to-all from there: mbox Avoid top-posting and favor interleaved quoting: https://en.wikipedia.org/wiki/Posting_style#Interleaved_style * Reply using the --to, --cc, and --in-reply-to switches of git-send-email(1): git send-email \ --in-reply-to=20200217155953.25803-5-gorcunov@gmail.com \ --to=gorcunov@gmail.com \ --cc=tarantool-patches@dev.tarantool.org \ --subject='Re: [Tarantool-patches] [PATCH 4/4] box/txn: fix nil dereference in txn_rollback' \ /path/to/YOUR_REPLY https://kernel.org/pub/software/scm/git/docs/git-send-email.html * If your mail client supports setting the In-Reply-To header via mailto: links, try the mailto: link
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox