Incorrectly handled transaction restarts can be a source of heisenbugs; add a mode where we randomly inject them to shake them out.
Signed-off-by: Kent Overstreet <[email protected]> --- fs/bcachefs/Kconfig | 7 +++++++ fs/bcachefs/btree_iter.c | 31 ++++++++++++++++++++++++++++++- fs/bcachefs/btree_iter.h | 12 ++++++++++++ fs/bcachefs/btree_trans_commit.c | 4 ++++ fs/bcachefs/btree_types.h | 3 +++ 5 files changed, 56 insertions(+), 1 deletion(-) diff --git a/fs/bcachefs/Kconfig b/fs/bcachefs/Kconfig index a61b22c142a1..75aaa59e3baf 100644 --- a/fs/bcachefs/Kconfig +++ b/fs/bcachefs/Kconfig @@ -60,6 +60,13 @@ config BCACHEFS_DEBUG The resulting code will be significantly slower than normal; you probably shouldn't select this option unless you're a developer. +config BCACHEFS_INJECT_TRANSACTION_RESTARTS + bool "Randomly inject transaction restarts" + depends on BCACHEFS_DEBUG + help + Randomly inject transaction restarts in a few core paths - may have a + significant performance penalty + config BCACHEFS_TESTS bool "bcachefs unit and performance tests" depends on BCACHEFS_FS diff --git a/fs/bcachefs/btree_iter.c b/fs/bcachefs/btree_iter.c index 15ac72b1af51..2ef3ac463f66 100644 --- a/fs/bcachefs/btree_iter.c +++ b/fs/bcachefs/btree_iter.c @@ -2292,6 +2292,12 @@ struct bkey_s_c bch2_btree_iter_peek_upto(struct btree_iter *iter, struct bpos e bch2_trans_verify_not_unlocked(trans); EBUG_ON((iter->flags & BTREE_ITER_filter_snapshots) && bkey_eq(end, POS_MAX)); + ret = trans_maybe_inject_restart(trans, _RET_IP_); + if (unlikely(ret)) { + k = bkey_s_c_err(ret); + goto out_no_locked; + } + if (iter->update_path) { bch2_path_put_nokeep(trans, iter->update_path, iter->flags & BTREE_ITER_intent); @@ -2472,6 +2478,10 @@ struct bkey_s_c bch2_btree_iter_peek_prev(struct btree_iter *iter) bch2_btree_iter_verify(iter); bch2_btree_iter_verify_entry_exit(iter); + ret = trans_maybe_inject_restart(trans, _RET_IP_); + if (unlikely(ret)) + return bkey_s_c_err(ret); + if (iter->flags & BTREE_ITER_filter_snapshots) search_key.snapshot = U32_MAX; @@ -2606,6 +2616,12 @@ struct bkey_s_c bch2_btree_iter_peek_slot(struct btree_iter *iter) bch2_btree_iter_verify_entry_exit(iter); EBUG_ON(btree_iter_path(trans, iter)->level && (iter->flags & BTREE_ITER_with_key_cache)); + ret = trans_maybe_inject_restart(trans, _RET_IP_); + if (unlikely(ret)) { + k = bkey_s_c_err(ret); + goto out_no_locked; + } + /* extents can't span inode numbers: */ if ((iter->flags & BTREE_ITER_is_extents) && unlikely(iter->pos.offset == KEY_OFFSET_MAX)) { @@ -2951,6 +2967,10 @@ void *__bch2_trans_kmalloc(struct btree_trans *trans, size_t size) WARN_ON_ONCE(new_bytes > BTREE_TRANS_MEM_MAX); + ret = trans_maybe_inject_restart(trans, _RET_IP_); + if (ret) + return ERR_PTR(ret); + struct btree_transaction_stats *s = btree_trans_stats(trans); s->max_mem = max(s->max_mem, new_bytes); @@ -3008,7 +3028,8 @@ void *__bch2_trans_kmalloc(struct btree_trans *trans, size_t size) if (old_bytes) { trace_and_count(c, trans_restart_mem_realloced, trans, _RET_IP_, new_bytes); - return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_mem_realloced)); + return ERR_PTR(btree_trans_restart_ip(trans, + BCH_ERR_transaction_restart_mem_realloced, _RET_IP_)); } out_change_top: p = trans->mem + trans->mem_top; @@ -3116,6 +3137,14 @@ u32 bch2_trans_begin(struct btree_trans *trans) trans->last_begin_ip = _RET_IP_; +#ifdef CONFIG_BCACHEFS_INJECT_TRANSACTION_RESTARTS + if (trans->restarted) { + trans->restart_count_this_trans++; + } else { + trans->restart_count_this_trans = 0; + } +#endif + trans_set_locked(trans); if (trans->restarted) { diff --git a/fs/bcachefs/btree_iter.h b/fs/bcachefs/btree_iter.h index 36899c6b134e..286c11b0949c 100644 --- a/fs/bcachefs/btree_iter.h +++ b/fs/bcachefs/btree_iter.h @@ -363,6 +363,18 @@ static int btree_trans_restart(struct btree_trans *trans, int err) return btree_trans_restart_ip(trans, err, _THIS_IP_); } +static inline int trans_maybe_inject_restart(struct btree_trans *trans, unsigned long ip) +{ +#ifdef CONFIG_BCACHEFS_INJECT_TRANSACTION_RESTARTS + if (!(ktime_get_ns() & ~(~0ULL << min(63, (10 + trans->restart_count_this_trans))))) { + trace_and_count(trans->c, trans_restart_injected, trans, ip); + return btree_trans_restart_ip(trans, + BCH_ERR_transaction_restart_fault_inject, ip); + } +#endif + return 0; +} + bool bch2_btree_node_upgrade(struct btree_trans *, struct btree_path *, unsigned); diff --git a/fs/bcachefs/btree_trans_commit.c b/fs/bcachefs/btree_trans_commit.c index b47f11881fe4..766c56cfcd7c 100644 --- a/fs/bcachefs/btree_trans_commit.c +++ b/fs/bcachefs/btree_trans_commit.c @@ -1027,6 +1027,10 @@ int __bch2_trans_commit(struct btree_trans *trans, unsigned flags) bch2_trans_verify_not_unlocked(trans); bch2_trans_verify_not_in_restart(trans); + ret = trans_maybe_inject_restart(trans, _RET_IP_); + if (unlikely(ret)) + goto out_reset; + if (!trans->nr_updates && !trans->journal_entries_u64s) goto out_reset; diff --git a/fs/bcachefs/btree_types.h b/fs/bcachefs/btree_types.h index baab5288ecc9..3229de2c335c 100644 --- a/fs/bcachefs/btree_types.h +++ b/fs/bcachefs/btree_types.h @@ -509,6 +509,9 @@ struct btree_trans { bool notrace_relock_fail:1; enum bch_errcode restarted:16; u32 restart_count; +#ifdef CONFIG_BCACHEFS_INJECT_TRANSACTION_RESTARTS + u32 restart_count_this_trans; +#endif u64 last_begin_time; unsigned long last_begin_ip; -- 2.45.2
