[PATCH v3 00/12] Improvements to static checker invocation
While working on pesto, I ran into a number of awkward errors with the static checkers. This series reworks the invocation of the checkers in a way that will let us deal with that. As a bonus, it also gives us static checking for passt-repair. It also a number of other cleanups to the Makefile that seemed natural along the way. v3: - Rework changes to $(FLAGS) so they're much less likely to have side effects we're not ready for. v2: - Fixed nasty test failure in test/build/build.py David Gibson (12): Makefile: Use make variables for static checker configuration Makefile: Make conditional definition of $(BIN) clearer Makefile: Use common binary compilation rule Makefile: Remove unhelpful $(HEADERS) variable Makefile: Add header dependencies for secondary binaries Makefile: Split $(FLAGS) into cpp and cc components cppcheck, clang-tidy: Static checkers don't need non-preprocessor flags Makefile: Split static checker targets passt-repair: Split out inotify handling to its own function passt-repair: Simplify construction of Unix path from inotify passt-repair: Run static checkers pesto: Run static checkers on pesto sources Makefile | 135 ++++++++++++++++++++++++-------------- linux_dep.h | 2 +- passt-repair.c | 171 +++++++++++++++++++++++++++---------------------- pesto.c | 1 - 4 files changed, 182 insertions(+), 127 deletions(-) -- 2.54.0
Our cppcheck and clang-tidy rules don't really follow normal Makefile
conventions. Usually any commands other than the very basics have their
binary specified in a variable so it can be overridden on the command line
if they're in an unusual location. Implement that for $(CPPCHECK) and
$(CLANG_TIDY)
Likewise flags to tools usually have their own Make variable. Do the same
with $(CLANG_TIDY_FLAGS) and $(CPPCHECK_FLAGS). Note that these only have
the options specifically for the static checker, not compiler flags which
we are also supplying to the static checker - those are derived from
FLAGS / CFLAGS / CPPFLAGS as before.
As part of that we change the probing for --check-level=exhaustive from
being run as part of the cppcheck target, to being run when we build the
CPPCHECK_FLAGS variable. That doesn't make any real difference now, but
will make things nicer if we need multiple cppcheck targets in future (e.g.
for passt-repair).
Signed-off-by: David Gibson
Currently we pass all our compiler flags to clang-tidy, except -pie, which
it won't accept. In fact in order to run the checker, we only need the
preprocessor flags. Simplify the command line by passing only those.
For cppcheck we already filter out just -D options from the compiler flags.
Simplify this by only passing preprocessor flags, now that we've split
those out into their own variables. Furthermore, one of cppcheck's
features which we're currently not exploiting is to check multiple / all
preprocessor option combinations in a single pass. Therefore, pass only
$(BASE_CPPFLAGS), which contains the mandatory options with which we can't
compile at all.
While we're there remove a redundant $^ that slipped in at some point.
Signed-off-by: David Gibson
The list of binaries is dependent on the target architecture, because
x86_64 adds the passt.avx2 and pasta.avx2 binaries. Make this more
obvious by defining BIN in common, then augmenting it in the x86_64
case.
Signed-off-by: David Gibson
The $(FLAGS) variable contains mandatory compiler flags that should not be
overridden. However, it contains a mixture of flags for the preprocessor
and for the compiler proper. That's causing some inconvenience for other
Makefile cleanups, so split it into $(BASE_CPPFLAGS) and $(BASE_CFLAGS)
variables.
Signed-off-by: David Gibson
Currently we have a single 'cppcheck' and 'clang-tidy' target which checks
passt. However, it doesn't check the additional binaries, qrap and
passt-repair. In preparation for running the static checkers on those as
well, split the targets into a top-level rule and a pattern rule which we
will be able to reuse.
Signed-off-by: David Gibson
PASST_HEADERS contains all the headers used by passt, which we use in
various dependencies. However, qrap and passt-repair each use several
headers which we don't have dependencies for. Add handling for this to the
Makefile.
Signed-off-by: David Gibson
Update the Makefile to run both clang-tidy and cppcheck on pesto as well
as on passt and passt-repair. This requires a couple of secondary
corrections:
* pesto.c had an inline suppression that is no longer correct now that
the protocol version has been bumped to 1. Remove it.
* We were globally suppressing the unusedStructMember because it
hit many false positives on both passt and passt-repair. It doesn't
in pesto, meaning it instead creates an unusedSuppression warning.
Apply the suppression as a flag override for passt and passt-repair,
instead of globally.
Signed-off-by: David Gibson
passt-repair can operate two ways: either it can be given an explicit
socket path to connect to, or it can be given a directory. In the second
mode, it will wait for a socket to appear in that directory before
connecting to it.
That waiting involves some inotify logic that is essentially unrelated to
the rest of the code. However, it's currently inline in main() making that
very long. Moreover, the block handling inotify shadows several variables
used in the rest of main() which will make static checkers complain once
we get them running on passt-repair.
Address this by moving the inotify handling into its own function.
Signed-off-by: David Gibson
When passt-repair is invoked with a directory name, it waits for a Unix
socket to appear in that directory. We need to build the Unix path name
from the given directory, plus the stem file name from the inotify event.
Currently, we build that path into a temporary buffer of size PATH_MAX,
then move it into the smaller buffer inside the Unix sockaddr. There's no
particular reason for this two step process, we can build the address
directly within the sockaddr_un. This will give a slightly different error
if the constructed path exceeds the maximum length of a Unix address, but
it will fail either way so it doesn't really matter.
Signed-off-by: David Gibson
Run the static checkers, cppcheck and clang-tidy on passt-repair as well
as on passt proper. This shows up handful of remaining minor warnings,
which we correct.
Signed-off-by: David Gibson
On Tue, 12 May 2026 15:52:50 +1000
David Gibson
The $(FLAGS) variable contains mandatory compiler flags that should not be overridden. However, it contains a mixture of flags for the preprocessor and for the compiler proper. That's causing some inconvenience for other Makefile cleanups, so split it into $(BASE_CPPFLAGS) and $(BASE_CFLAGS) variables.
Signed-off-by: David Gibson
--- Makefile | 40 ++++++++++++++++++++++++---------------- 1 file changed, 24 insertions(+), 16 deletions(-) diff --git a/Makefile b/Makefile index a8f8d06e..697f229f 100644 --- a/Makefile +++ b/Makefile @@ -30,12 +30,17 @@ ifeq ($(shell $(CC) -O2 -dM -E - < /dev/null 2>&1 | grep ' _FORTIFY_SOURCE ' > / FORTIFY_FLAG := -D_FORTIFY_SOURCE=2 endif
-FLAGS := -Wall -Wextra -Wno-format-zero-length -Wformat-security -FLAGS += -pedantic -std=c11 -D_XOPEN_SOURCE=700 -D_GNU_SOURCE -FLAGS += $(FORTIFY_FLAG) -O2 -pie -fPIE -FLAGS += -DPAGE_SIZE=$(shell getconf PAGE_SIZE) -FLAGS += -DVERSION=\"$(VERSION)\" -FLAGS += -DDUAL_STACK_SOCKETS=$(DUAL_STACK_SOCKETS) +# Mandatory preprocessor flags that won't be overridden with $(CPPFLAGS) +# FIXME: Could some of these be default, rather than required? +BASE_CPPFLAGS := -D_XOPEN_SOURCE=700 -D_GNU_SOURCE $(FORTIFY_FLAG) +BASE_CPPFLAGS += -DPAGE_SIZE=$(shell getconf PAGE_SIZE) +BASE_CPPFLAGS += -DVERSION=\"$(VERSION)\" +BASE_CPPFLAGS += -DDUAL_STACK_SOCKETS=$(DUAL_STACK_SOCKETS) + +# Mandatory compiler flags that won't be overridden with $(CFLAGS) +# FIXME: Could some of these be default, rather than required? +BASE_CFLAGS := -std=c11 -pie -fPIE -O2 +BASE_CFLAGS += -pedantic -Wall -Wextra -Wno-format-zero-length -Wformat-security
This new version of the series looks good to me in general (minus potential concern reported below), and everything seems to work on Debian and Fedora, but I would still like to try things out on Alpine or Void Linux because musl might cause surprises. I haven't got to it yet. Meanwhile, regarding these FIXME comments: I think it *is* currently possible to override those flags (with different values for the same options), and overriding -D_FORTIFY_SOURCE on openSUSE (I haven't tried right now) was the initial motivation behind FLAGS. That is, the overriding role of CFLAGS seems to be preserved for these BASE_* flags as well, because $CFLAGS is given to the compiler after $BASE_CPPFLAGS, $CPPFLAGS, and $BASE_CFLAGS. So, in this sense, I would already call them "default" flags. If that's the case, I think it's fine. Otherwise we need to find another solution at least for the short term. By the way, if it helps addressing those comments at some point (I would apply anyway this series meanwhile if I don't find breakages, because not being able to run static checkers automatically on pesto is pretty nasty), out of those flags: * -D_XOPEN_SOURCE, -D_GNU_SOURCE, and -DPAGE_SIZE are strictly required to build (at least in some environments) * -D_FORTIFY_SOURCE, -pie, -fPIE are not required to build but they are critical for security * -DVERSION is not required to build but makes things confusing and issues hard to debug because the version (usually supplied by the distribution) isn't reported in logs and logs of other tools - -DDUAL_STACK_SOCKETS doesn't seem to be used anymore starting from commit b8d4fac6a2e7 ("util, pif: Replace sock_l4() with pif_sock_l4()")... was it intended, actually? - -std=c11 is strictly required to ensure we build things correctly - -O2 is optional, but dropping it (by default) might require annoying adjustments in distributions - -pedantic, -Wall, -Wextra, -Wno-format-zero-length, -Wformat-security are all optional and useful for development (including distribution development), and might be security relevant in some cases -- Stefano
participants (2)
-
David Gibson
-
Stefano Brivio