Merge pull request #691 from mihalicyn/psi_write_support

/proc/pressure/{cpu, io, memory} virtualization

Author: stgraber

.github/actions/build/action.yml

@@ -48,6 +48,7 @@ runs:
         meson setup build \
             -Ddocs=false \
             -Dtests=true \
+            -Dmocks=true \
             -Dinit-script=systemd \
             -Dprefix=/usr \
             -Db_sanitize=address,undefined

.github/workflows/tests.yml

@@ -38,6 +38,7 @@ jobs:
       - name: Test
         env:
           CC: ${{ matrix.compiler }}
+          LIBFUSE: ${{ matrix.fuse }}
         run: |
           echo "::group::Running the testsuite"
 

meson.build

@@ -69,6 +69,7 @@ conf.set_quoted('LXCFSTARGETDIR', join_paths(localstatedir, 'lib/lxcfs'))
 # Custom configuration.
 init_script = get_option('init-script')
 want_tests = get_option('tests')
+want_mocks = get_option('mocks')
 want_docs = get_option('docs')
 
 # Build flags.
@@ -232,12 +233,18 @@ liblxcfs_common_dependencies = declare_dependency(
         libfuse,
     ])
 
+liblxcfs_cargs = []
+if want_mocks == true
+	liblxcfs_cargs = ['-DPSITRIGGERTEST', '-DDEBUG']
+endif
+
 liblxcfs = shared_module(
     'lxcfs',
     liblxcfs_sources,
     dependencies: liblxcfs_common_dependencies,
     install: true,
-    install_dir: lxcfsdir)
+    install_dir: lxcfsdir,
+    c_args: liblxcfs_cargs)
 
 # Tests.
 test_programs = []
@@ -248,7 +255,7 @@ if want_tests == true
         dependencies: liblxcfs_common_dependencies,
         install: false,
         install_dir: lxcfsdir,
-        c_args: '-DRELOADTEST -DDEBUG')
+        c_args: ['-DRELOADTEST', '-DDEBUG'])
 endif
 
 # RPM spec.
@@ -306,6 +313,7 @@ status = [
     'lxcfs source root directory:	@0@'.format(project_source_root),
     'init system(s):		@0@'.format(init_script),
     'tests:				@0@'.format(want_tests),
+    'mocks:				@0@'.format(want_mocks),
     'documentation:			@0@'.format(want_docs),
 ]
 message('\n         '.join(status))

meson_options.txt

@@ -3,6 +3,9 @@
 option('tests', type : 'boolean', value: 'false',
        description : 'enable tests')
 
+option('mocks', type : 'boolean', value: 'false',
+       description : 'enable CI-only features (mocks some things in LXCFS for CI/testing purposes)')
+
 option('runtime-path', type : 'string', value : '/run',
        description : 'the runtime directory')
 

src/bindings.h

@@ -87,6 +87,16 @@ enum lxcfs_virt_t {
 #define LXCFS_TYPE_SYS(type) (type >= LXC_TYPE_SYS && type <= LXC_TYPE_SYS_DEVICES_SYSTEM_CPU_ONLINE)
 #define LXCFS_TYPE_OK(type) (type >= LXC_TYPE_CGDIR && type < LXC_TYPE_MAX)
 
+/*
+ * This signal will be used in /proc/pressure/{cpu, memory, io} poll()
+ * virtualization to notify a background thread that it's
+ * a time to finish.
+ *
+ * We don't need a handler for this, we want this signal to be
+ * ignored, but interrupt any interruptible syscall like poll().
+ */
+#define SIG_NOTIFY_POLL_WAKEUP (SIGRTMIN + 0)
+
 /*
  * This signal will be used to signal fuse request processing thread that
  * request was interrupted (FUSE_INTERRUPT came from the kernel).
@@ -106,13 +116,20 @@ extern int rwlock_rdlock_interruptible(pthread_rwlock_t *l);
 extern int rwlock_wrlock_interruptible(pthread_rwlock_t *l);
 
 struct file_info {
-	char *controller;
-	char *cgroup;
-	char *file;
+	union {
+		struct {
+			char *controller;
+			char *cgroup;
+			char *file;
+		};
+		struct {
+			void *private_data;
+		};
+	};
 	int type;
 	char *buf; /* unused */
 	int buflen;
-	int size; /*actual data size */
+	int size; /* actual data size */
 	int cached;
 };
 
@@ -125,17 +142,19 @@ struct lxcfs_opts {
 	 * and the use of bool instead of explicited __u32 and __u64 we can't.
 	 */
 	__u32 version;
-        // As of opts version 2.
-        char runtime_path[PATH_MAX];
+	// As of opts version 2.
+	char runtime_path[PATH_MAX];
 	bool zswap_off;
+	bool psi_poll_on;
 };
 
 typedef enum lxcfs_opt_t {
-	LXCFS_SWAP_ON	= 0,
-	LXCFS_PIDFD_ON	= 1,
-	LXCFS_CFS_ON	= 2,
-	LXCFS_ZSWAP_ON  = 3,
-	LXCFS_OPTS_MAX	= LXCFS_ZSWAP_ON,
+	LXCFS_SWAP_ON		= 0,
+	LXCFS_PIDFD_ON		= 1,
+	LXCFS_CFS_ON		= 2,
+	LXCFS_ZSWAP_ON		= 3,
+	LXCFS_PSI_POLL_ON	= 4,
+	LXCFS_OPTS_MAX		= LXCFS_PSI_POLL_ON,
 } lxcfs_opt_t;
 
 
@@ -172,6 +191,8 @@ static inline bool lxcfs_has_opt(struct lxcfs_opts *opts, lxcfs_opt_t opt)
 		if (opts->version >= 3 && !opts->zswap_off)
 			return liblxcfs_can_use_zswap();
 		return false;
+	case LXCFS_PSI_POLL_ON:
+		return (opts->version >= 4 && opts->psi_poll_on);
 	}
 
 	return false;

src/lxcfs.c

@@ -242,6 +242,8 @@ static void sigusr1_reload(int signo, siginfo_t *info, void *extra)
 	need_reload = 1;
 }
 
+static void sig_noop_handler(int signo, siginfo_t *info, void *extra) {}
+
 /* Functions to run the library methods */
 
 #define DEF_LIB_FS_OP(type, fsop)					\
@@ -276,8 +278,14 @@ DEF_LIB_FS_OP(sys  , read)
 					off_t offset, struct fuse_file_info *fi
 #define LIB_FS_write_OP_ARGS		path, buf, size, offset, fi
 DEF_LIB_FS_OP(cg   , write)
+DEF_LIB_FS_OP(proc , write)
 DEF_LIB_FS_OP(sys  , write)
 
+#define LIB_FS_poll_OP_ARGS_TYPE	const char *path, struct fuse_file_info *fi, \
+					struct fuse_pollhandle *ph, unsigned *reventsp
+#define LIB_FS_poll_OP_ARGS		path, fi, ph, reventsp
+DEF_LIB_FS_OP(proc , poll)
+
 #define LIB_FS_mkdir_OP_ARGS_TYPE	const char *path, mode_t mode
 #define LIB_FS_mkdir_OP_ARGS		path, mode
 DEF_LIB_FS_OP(cg, mkdir)
@@ -605,6 +613,13 @@ int lxcfs_write(const char *path, const char *buf, size_t size, off_t offset,
 		return ret;
 	}
 
+	if (LXCFS_TYPE_PROC(type)) {
+		up_users();
+		ret = do_proc_write(path, buf, size, offset, fi);
+		down_users();
+		return ret;
+	}
+
 	if (LXCFS_TYPE_SYS(type)) {
 		up_users();
 		ret = do_sys_write(path, buf, size, offset, fi);
@@ -615,6 +630,27 @@ int lxcfs_write(const char *path, const char *buf, size_t size, off_t offset,
 	return -EINVAL;
 }
 
+int lxcfs_poll(const char *path, struct fuse_file_info *fi,
+	       struct fuse_pollhandle *ph, unsigned *reventsp)
+{
+	int ret;
+	enum lxcfs_virt_t type;
+
+	type = file_info_type(fi);
+
+	if (LXCFS_TYPE_PROC(type)) {
+		up_users();
+		ret = do_proc_poll(path, fi, ph, reventsp);
+		down_users();
+		return ret;
+	}
+
+	/* default f_op->poll() behavior when not supported */
+	fuse_pollhandle_destroy(ph);
+	*reventsp = DEFAULT_POLLMASK;
+	return 0;
+}
+
 int lxcfs_readlink(const char *path, char *buf, size_t size)
 {
 	int ret;
@@ -842,6 +878,9 @@ const struct fuse_operations lxcfs_ops = {
 	.truncate	= lxcfs_truncate,
 	.write		= lxcfs_write,
 	.readlink	= lxcfs_readlink,
+#if HAVE_FUSE3
+	.poll		= lxcfs_poll,
+#endif
 
 	.create		= NULL,
 	.destroy	= NULL,
@@ -937,6 +976,7 @@ static const struct option long_options[] = {
 	{"enable-cfs",		no_argument,		0,	  0	},
 	{"enable-pidfd",	no_argument,		0,	  0	},
 	{"enable-cgroup",	no_argument,		0,	  0	},
+	{"enable-psi-poll",	no_argument,		0,	  0	},
 
 	{"pidfile",		required_argument,	0,	'p'	},
 	{"runtime-dir",		required_argument,	0,	  0	},
@@ -1011,7 +1051,8 @@ int main(int argc, char *argv[])
 	opts->zswap_off = false;
 	opts->use_pidfd = false;
 	opts->use_cfs = false;
-	opts->version = 3;
+	opts->psi_poll_on = false;
+	opts->version = 4;
 
 	while ((c = getopt_long(argc, argv, "dulfhvso:p:", long_options, &idx)) != -1) {
 		switch (c) {
@@ -1022,6 +1063,8 @@ int main(int argc, char *argv[])
 				opts->use_cfs = true;
 			else if (strcmp(long_options[idx].name, "enable-cgroup") == 0)
 				cgroup_is_enabled = true;
+			else if (strcmp(long_options[idx].name, "enable-psi-poll") == 0)
+				opts->psi_poll_on = true;
 			else if (strcmp(long_options[idx].name, "runtime-dir") == 0)
 				runtime_path_arg = optarg;
 			else
@@ -1189,6 +1232,11 @@ int main(int argc, char *argv[])
 	}
 #endif
 
+	if (install_signal_handler(SIG_NOTIFY_POLL_WAKEUP, sig_noop_handler)) {
+		lxcfs_error("%s - Failed to install SIG_NOTIFY_POLL_WAKEUP signal handler", strerror(errno));
+		goto out;
+	}
+
 	if (!pidfile) {
 		snprintf(pidfile_buf, sizeof(pidfile_buf), "%s%s", runtime_path, PID_FILE);
 		pidfile = pidfile_buf;

src/macro.h

@@ -20,6 +20,8 @@
 #define CGROUP2_SUPER_MAGIC 0x63677270
 #endif
 
+#define DEFAULT_POLLMASK (EPOLLIN | EPOLLOUT | EPOLLRDNORM | EPOLLWRNORM)
+
 #define lxcfs_debug_stream(stream, format, ...)                       \
 	do {                                                          \
 		fprintf(stream, "%s: %d: %s: " format "\n", __FILE__, \