From 7f3a5eb70e476708551445219f7aceb7afc0a4a0 Mon Sep 17 00:00:00 2001 From: Evgeny Vereshchagin Date: Tue, 5 May 2020 11:10:56 +0000 Subject: [PATCH] build-system: bring back 'nonnull' Having taken a look at https://github.com/systemd/systemd/runs/645252074?check_suite_focus=true where fuzz-journal-remote failed with ``` AddressSanitizer:DEADLYSIGNAL ================================================================= ==16==ERROR: AddressSanitizer: SEGV on unknown address 0x000000000000 (pc 0x7f864f98948e bp 0x7ffde5c6b7c0 sp 0x7ffde5c6b560 T0) ==16==The signal is caused by a READ memory access. ==16==Hint: address points to the zero page. SCARINESS: 10 (null-deref) #0 0x7f864f98948e in output_short /work/build/../../src/systemd/src/shared/logs-show.c #1 0x7f864f984624 in show_journal_entry /work/build/../../src/systemd/src/shared/logs-show.c:1154:15 #2 0x7f864f984b63 in show_journal /work/build/../../src/systemd/src/shared/logs-show.c:1239:21 #3 0x4cabab in LLVMFuzzerTestOneInput /work/build/../../src/systemd/src/fuzz/fuzz-journal-remote.c:67:21 #4 0x51fd16 in fuzzer::Fuzzer::ExecuteCallback(unsigned char const*, unsigned long) /src/libfuzzer/FuzzerLoop.cpp:556:15 #5 0x51c330 in fuzzer::Fuzzer::RunOne(unsigned char const*, unsigned long, bool, fuzzer::InputInfo*, bool*) /src/libfuzzer/FuzzerLoop.cpp:470:3 #6 0x523700 in fuzzer::Fuzzer::ReadAndExecuteSeedCorpora(std::__1::vector >&) /src/libfuzzer/FuzzerLoop.cpp:765:7 #7 0x5246cd in fuzzer::Fuzzer::Loop(std::__1::vector >&) /src/libfuzzer/FuzzerLoop.cpp:792:3 #8 0x4de3d1 in fuzzer::FuzzerDriver(int*, char***, int (*)(unsigned char const*, unsigned long)) /src/libfuzzer/FuzzerDriver.cpp:824:6 #9 0x4cfb47 in main /src/libfuzzer/FuzzerMain.cpp:19:10 #10 0x7f864e69782f in __libc_start_main (/lib/x86_64-linux-gnu/libc.so.6+0x2082f) #11 0x41f2a8 in _start (out/fuzz-journal-remote+0x41f2a8) AddressSanitizer can not provide additional info. SUMMARY: AddressSanitizer: SEGV /work/build/../../src/systemd/src/shared/logs-show.c in output_short ==16==ABORTING MS: 0 ; base unit: 0000000000000000000000000000000000000000 0x44,0x3d,0xa,0x5f,0x5f,0x52,0x45,0x41,0x4c,0x54,0x49,0x4d,0x45,0x5f,0x54,0x49,0x4d,0x45,0x53,0x54,0x41,0x4d,0x50,0x3d,0x31,0xa,0xa, D=\x0a__REALTIME_TIMESTAMP=1\x0a\x0a artifact_prefix='./'; Test unit written to ./crash-d635b9dd31cceff3c912fd45e1a58d7e90f0ad73 Base64: RD0KX19SRUFMVElNRV9USU1FU1RBTVA9MQoK ``` I was wondering why it hadn't been caught by the compiler even though clang should have failed to compile it with ``` ../src/shared/logs-show.c:624:25: warning: null passed to a callee that requires a non-null argument [-Wnonnull] print_multiline(f, 4 + fieldlen + 1, 0, OUTPUT_FULL_WIDTH, 0, false, ^ ../src/shared/logs-show.c:161:24: note: callee declares array parameter as static here size_t highlight[static 2]) { ^ ~~~~~~~~~~ ../src/shared/logs-show.c:1239:21: warning: null passed to a callee that requires a non-null argument [-Wnonnull] r = show_journal_entry(f, j, mode, n_columns, flags, NULL, NULL, ellipsized); ^ ~~~~ ../src/shared/logs-show.c:1133:30: note: callee declares array parameter as static here const size_t highlight[static 2], ^ ~~~~~~~~~~ 2 warnings generated. ``` Given that judging by https://github.com/systemd/systemd/issues/13039 it doesn't seem to be the first time issues like that have been missed I think it would be better to turn nonnull on and get around false positives on a case-by-case basis with DISABLE_WARNING_NONNULL .. REENABLE_WARNING Reopens https://github.com/systemd/systemd/issues/6119 --- meson.build | 3 --- 1 file changed, 3 deletions(-) diff --git a/meson.build b/meson.build index fef3c27cf4..5e06f19c9c 100644 --- a/meson.build +++ b/meson.build @@ -377,9 +377,6 @@ possible_cc_flags = [ '-Wno-error=#warnings', # clang '-Wno-string-plus-int', # clang - # work-around for gcc 7.1 turning this on on its own. - '-Wno-error=nonnull', - # Disable -Wmaybe-uninitialized, since it's noisy on gcc 8 with # optimizations enabled, producing essentially false positives. '-Wno-maybe-uninitialized',