On Sun, May 31, 2026 at 04:13:24PM -0400, Jon Maloy wrote: > conf_open_files() opens four file descriptors (fd_tap_listen, > fd_repair_listen, pidfile_fd, fd_control_listen) that are held for > the entire daemon lifetime. Because no close() call exists for them > anywhere, Coverity flags each as INCOMPLETE_DEALLOCATOR. This is > clearly a false positive, but we still want to get rid of this > warning. A bit frustrating that Coverity doesn't recognize an exit() will close all remaining files, but ok. I'm ok with close()ing these at exit... but some details remain. First, for pidfile_fd specifically, we shouldn't need that after we've written the pid. So we should just close() at that point, rather than waiting until exit. > > We now register the execution context so that passt_exit() can use > to close these descriptors before calling _exit(). All exit paths > (signal handler, die(), die_perror()) funnel through passt_exit(), > so this covers all cases. passt_exit() is the right place for this, but registering an actual global to point at a local that's sort of a global seems pretty perverse. It would make more sense to convert the main struct ctx itself to a global, rather than a main()-local. Note that I'm not suggesting changing all the places that use it (current or future) - I think it's still preferable for those (excepting I guess main() itself) to access it via a parameter. We can encourage that by using a bulky name for the global. I see that setting 'exit_cleanup_ctx' where you do also serves one other purpose - it stops early passt_exit()s from attempting to close() those fds before they're initialised, even to -1. We can fix that by using an initialiser on the global to set fd fields to -1 before main(), which is arguably more elegant than the cluster of assignments at the top of main() anyway. > > Signed-off-by: Jon Maloy > --- > log.h | 2 ++ > passt.c | 1 + > util.c | 22 ++++++++++++++++++++++ > 3 files changed, 25 insertions(+) > > diff --git a/log.h b/log.h > index 69cfb507..079f429c 100644 > --- a/log.h > +++ b/log.h > @@ -63,7 +63,9 @@ extern bool debug_flag; > /* This would make more sense in util.h, but because we use it in die(), that > * would cause awkward circular reference problems. > */ > +struct ctx; > void passt_exit(int status) __attribute__((noreturn)); > +void passt_exit_set_ctx(struct ctx *c); > > #define LOGFILE_SIZE_DEFAULT (1024 * 1024UL) > #define LOGFILE_CUT_RATIO 30 /* When full, cut ~30% size */ > diff --git a/passt.c b/passt.c > index b6fc12d4..ec6aa57a 100644 > --- a/passt.c > +++ b/passt.c > @@ -392,6 +392,7 @@ int main(int argc, char **argv) > sock_probe_features(&c); > > conf(&c, argc, argv); > + passt_exit_set_ctx(&c); > trace_init(c.trace); > > pasta_netns_quit_init(&c); > diff --git a/util.c b/util.c > index b64c29ed..f15b1f9a 100644 > --- a/util.c > +++ b/util.c > @@ -1097,6 +1097,17 @@ void abort_with_msg(const char *fmt, ...) > abort(); > } > > +static struct ctx *exit_cleanup_ctx; > + > +/** > + * passt_exit_set_ctx() - Register context for cleanup on exit > + * @c: Execution context > + */ > +void passt_exit_set_ctx(struct ctx *c) > +{ > + exit_cleanup_ctx = c; > +} > + > /** > * passt_exit() - Perform vital cleanup and exit > * > @@ -1108,6 +1119,17 @@ void abort_with_msg(const char *fmt, ...) > */ > void passt_exit(int status) > { > + if (exit_cleanup_ctx) { > + if (exit_cleanup_ctx->fd_tap_listen >= 0) > + close(exit_cleanup_ctx->fd_tap_listen); > + if (exit_cleanup_ctx->fd_repair_listen >= 0) > + close(exit_cleanup_ctx->fd_repair_listen); > + if (exit_cleanup_ctx->pidfile_fd >= 0) > + close(exit_cleanup_ctx->pidfile_fd); > + if (exit_cleanup_ctx->fd_control_listen >= 0) > + close(exit_cleanup_ctx->fd_control_listen); > + } > + > /* Make sure we don't leave the pcap file truncated */ > if (pcap_fd != -1 && fsync(pcap_fd)) > warn_perror("Failed to flush pcap file, it might be truncated"); > -- > 2.52.0 > -- David Gibson (he or they) | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you, not the other way | around. http://www.ozlabs.org/~dgibson