new(app_actions): add base_syscalls.repair option

See https://github.com/falcosecurity/falco/issues/2433

Co-authored-by: Jason Dellaluce <jasondellaluce@gmail.com>
Signed-off-by: Melissa Kilby <melissa.kilby.oss@gmail.com>

falco.yaml

@@ -460,66 +460,86 @@ metadata_download:
   watch_freq_sec: 1
 
 
-# base_syscalls ! Use with caution !
+# base_syscalls ! Use with caution, read carefully !
 #
 # --- [Description]
 #
-# With this option you are in full control of the total set of syscalls that
+# With this option you are in full control of the set of syscalls that
 # Falco will enable in the kernel for active tracing.
 
-# All syscalls and events from each enabled Falco rule will automatically be activated
+# All syscalls and events from each enabled Falco rule are activated
-# even when choosing this option. This option provides full end user control to specifically
+# even when choosing this option. This option allows you to define a
-# define a static set of base syscalls that will be activated in addition to the
+# set of base syscalls that will be activated in addition to the
 # syscalls defined in the rules.
 #
-# When using this option, Falco does not add any other syscalls that may be needed for
+# You may ask yourself why do we need to activate syscalls in addition to the rules?
-# Falco's state engine. The union of all syscalls from the rules (including resolved macros)
-# and the ones specified here compose the final set of syscalls that are traced in the kernel.
-# This puts the end user in the driver seat, but if not used correctly Falco logs may be
-# incomplete or wrong. This option however can be very useful to lower CPU utilization and
-# allowing you to tailor Falco to specific environments according to your
-# organization's threat model and security posture as well as cost budget.
-
-# !!! When NOT using this option, Falco defaults to adding a static set of syscalls in addition
-# to the rules system calls you need for Falco's state engine build-up and life-cycle management.
 #
+# Falco requires a set of syscalls to build up state in userspace. This is because for
+# example when spawning a new process or creating a network connection more than one syscall
+# is involved. Furthermore, properties of a process during its life time can be modified
+# by syscalls. Falco takes care of this by activating more syscalls than the ones defined
+# in the rules and by managing a smart process cache table in userspace.
+# Processes are purged when a process exits.
+#
+# Looking back to what this option does, it activates all syscalls from the rules
+# (including resolved macros) and the ones specified here.
+#
+# This puts the end user in the driver seat to tell Falco what it needs, but if not used correctly
+# Falco logs may be incomplete or wrong or Falco won't work at all. This option however can be
+# very useful to lower CPU utilization and allowing you to tailor Falco to specific environments
+# according to your organization's threat model and cost budget.
+#
+# !!! When NOT using this option, Falco defaults to adding a static set of syscalls in addition
+# to the rules system calls Falco needs for its state engine build-up and life-cycle management.
+#
+# If you like the recommendations below you can automate them via setting `base_syscalls.repair`
+# to true. `base_syscalls.repair` is an experimental alternative to Falco's default state engine
+# enforcement. `base_syscalls.repair` is designed to be the most resourceful option to ensure
+# Falco runs correctly while activating a most minimal set of additional syscalls.
 #
 # --- [Usage]
 #
 # List of system calls names (<syscall-name>) plus negative ("!<syscall-name>") notation supported.
 #
-# base_syscalls: [<syscall-name>, <syscall-name>, "!<syscall-name>"]
+# base_syscalls.repair: <bool>
+# base_syscalls.custom_set: [<syscall-name>, <syscall-name>, "!<syscall-name>"]
 #
 #
 # --- [Suggestions]
 #
-# Here are a few recommendations that may help you to use the full power of this option:
+# Here are a few recommendations that may help you.
+# Setting `base_syscalls.repair: true` automates the recommendations.
 #
 # Consider to at minimum add the following syscalls regardless of the syscalls used in the rules.
 #
 # [clone, clone3, fork, vfork, execve, execveat, close]
 #
-# This is because some Falco fields you may output for an execve* system call are retrieved
+# This is because some Falco fields for an execve* system call are retrieved
-# from the associated "clone", "clone3", "fork", "vfork" syscall when spawning a new process.
+# from the associated `clone`, `clone3`, `fork`, `vfork` syscall when spawning a
-# The "close" system call is used to purge file descriptors from Falco's internal
+# new process. The `close` system call is used to purge file descriptors from Falco's
-# thread / process cache table and therefore should always be added when you have rules around fds
+# internal thread / process cache table and should always be added when you have
+# rules around file descriptors.
 # (e.g. open, openat, openat2, socket, connect, accept, accept4 ... and many more)
 #
 # When network syscalls are used in rules we recommend to at minimum set
 #
 # [clone, clone3, fork, vfork, execve, execveat, close, socket, bind, getsockopt]
 #
-# It turns out that while you absolutely can log connect or accept* syscalls without the socket
+# It turns out that while you can log `connect` or `accept*` syscalls without the socket
 # system call, the log however would not contain the ip tuples.
-# For listen and accept* system calls you also need the bind system call.
+# For listen and accept* system calls you also need the `bind` system call.
-#
+#`
-# Lastly, if you care about the correct uid, gid or sid, pgid of a process when that process then
+# Lastly, if you care about the correct `uid`, `gid` or `sid`, `pgid of a process when the
-# opens a file or makes a network connection or any other action, consider also
+# running process opens a file or makes a network connection, consider adding the following syscalls:
-# adding the following syscalls:
 #
 # setresuid, setsid, setuid, setgid, setpgid, setresgid, setsid, capset, chdir, chroot, fchdir
 #
-# Only exclude syscalls, e.g. "!mprotect" if you need a fast deployment update (overriding rules),
+# We recommend to exclude syscalls, e.g. "!mprotect" only if you need a fast deployment update
-# else rather remove unwanted or not needed syscalls from the Falco rules.
+# (overriding rules), else remove unwanted syscalls from the Falco rules.
+#
+# Passing `-o "log_level=debug"` to Falco's cmd args during a dry-run will print the
+# final set of syscalls to STDOUT.
 
-base_syscalls: []
+base_syscalls:
+  repair: false
+  custom_set: []

userspace/falco/app/actions/configure_interesting_sets.cpp

@@ -85,13 +85,15 @@ static void select_event_set(falco::app::state& s, const libsinsp::events::set<p
 	/* USER OVERRIDE INPUT OPTION "base_syscalls". */
 	std::unordered_set<std::string> user_positive_names = {};
 	std::unordered_set<std::string> user_negative_names = {};
-	extract_base_syscalls_names(s.config->m_base_syscalls, user_positive_names, user_negative_names);
+	extract_base_syscalls_names(s.config->m_base_syscalls_custom_set, user_positive_names, user_negative_names);
 	auto user_positive_sc_set = libsinsp::events::names_to_sc_set(user_positive_names);
 	auto user_negative_sc_set = libsinsp::events::names_to_sc_set(user_negative_names);
 
-	if (!user_positive_sc_set.empty())
+	if (!user_positive_sc_set.empty() || s.config->m_base_syscalls_repair)
 	{
 		// user overrides base event set
+		// in case `user_positive_sc_set` is empty, but `base_syscalls.repair` is set
+		// this has the effect of clearing the default `sinsp_state_sc_set()`
 		base_sc_set = user_positive_sc_set;
 
 		// we re-transform from sc_set to names to make
@@ -111,6 +113,17 @@ static void select_event_set(falco::app::state& s, const libsinsp::events::set<p
 	// base events set (either the default or the user-defined one)
 	s.selected_sc_set = rules_sc_set.merge(base_sc_set);
 
+	if (s.config->m_base_syscalls_repair)
+	{
+		/* If base_syscalls.repair set enforce `libsinsp` state based on rules set
+		 * and merge with user supplied base_syscalls.custom_set
+		 *
+		 * Also applies when base_syscalls.custom_set empty but base_syscalls.repair set
+		 * effectively bypassing the default `libsinsp` state enforcement and using
+		 * `sinsp_repair_state_sc_set` instead. */
+		s.selected_sc_set = libsinsp::events::sinsp_repair_state_sc_set(rules_sc_set).merge(base_sc_set);
+	}
+
 	if (!user_negative_sc_set.empty())
 	{
 		/* Remove negative base_syscalls events. */

userspace/falco/configuration.cpp

@@ -332,8 +332,9 @@ void falco_configuration::load_yaml(const std::string& config_name, const yaml_h
 
 	m_syscall_drop_failed_exit = config.get_scalar<bool>("syscall_drop_failed_exit", false);
 
-	m_base_syscalls.clear();
+	m_base_syscalls_custom_set.clear();
-	config.get_sequence<std::unordered_set<std::string>>(m_base_syscalls, std::string("base_syscalls"));
+	config.get_sequence<std::unordered_set<std::string>>(m_base_syscalls_custom_set, std::string("base_syscalls.custom_set"));
+	m_base_syscalls_repair = config.get_scalar<bool>("base_syscalls.repair", false);
 
 	std::set<std::string> load_plugins;
 

userspace/falco/configuration.h

@@ -109,7 +109,8 @@ public:
 	bool m_syscall_drop_failed_exit;
 
 	// User supplied base_syscalls, overrides any Falco state engine enforcement.
-	std::unordered_set<std::string> m_base_syscalls;
+	std::unordered_set<std::string> m_base_syscalls_custom_set;
+	bool m_base_syscalls_repair;
 
 	std::vector<plugin_config> m_plugins;