Merge pull request #2010 from tanyifeng/set_oom_score_adj

doc/lxc.container.conf.sgml.in

@@ -339,6 +339,30 @@ Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
     </refsect2>
 
     <refsect2>
+      <title>Proc</title>
+      <para>
+        Configure proc filesystem for the container.
+      </para>
+      <variablelist>
+        <varlistentry>
+          <term>
+            <option>lxc.proc.[proc file name]</option>
+          </term>
+          <listitem>
+            <para>
+              Specify the proc file name to be set. The file name available
+              are those listed under /proc/PID/.
+              Example:
+            </para>
+            <programlisting>
+              lxc.proc.oom_score_adj = 10
+            </programlisting>
+          </listitem>
+        </varlistentry>
+      </variablelist>
+    </refsect2>
+
+    <refsect2>
       <title>Ephemeral</title>
       <para>
         Allows one to specify whether a container will be destroyed on shutdown.

src/lxc/conf.c

@@ -2398,6 +2398,38 @@ int setup_sysctl_parameters(struct lxc_list *sysctls)
 	return 0;
 }
 
+int setup_proc_filesystem(struct lxc_list *procs, pid_t pid)
+{
+	struct lxc_list *it;
+	struct lxc_proc *elem;
+	char *tmp = NULL;
+	char filename[MAXPATHLEN] = {0};
+	int ret = 0;
+
+	lxc_list_for_each(it, procs) {
+		elem = it->elem;
+		tmp = lxc_string_replace(".", "/", elem->filename);
+		if (!tmp) {
+			ERROR("Failed to replace key %s", elem->filename);
+			return -1;
+		}
+
+		ret = snprintf(filename, sizeof(filename), "/proc/%d/%s", pid, tmp);
+		free(tmp);
+		if (ret < 0 || (size_t)ret >= sizeof(filename)) {
+			ERROR("Error setting up proc filesystem path");
+			return -1;
+		}
+
+		ret = lxc_write_to_file(filename, elem->value, strlen(elem->value), false);
+		if (ret < 0) {
+			ERROR("Failed to setup proc filesystem %s to %s", elem->filename, elem->value);
+			return -1;
+		}
+	}
+	return 0;
+}
+
 static char *default_rootfs_mount = LXCROOTFSMOUNT;
 
 struct lxc_conf *lxc_conf_init(void)
@@ -2449,6 +2481,7 @@ struct lxc_conf *lxc_conf_init(void)
 	lxc_list_init(&new->environment);
 	lxc_list_init(&new->limits);
 	lxc_list_init(&new->sysctls);
+	lxc_list_init(&new->procs);
 	for (i = 0; i < NUM_LXC_HOOKS; i++)
 		lxc_list_init(&new->hooks[i]);
 	lxc_list_init(&new->groups);
@@ -3384,6 +3417,33 @@ int lxc_clear_sysctls(struct lxc_conf *c, const char *key)
 	return 0;
 }
 
+int lxc_clear_procs(struct lxc_conf *c, const char *key)
+{
+	struct lxc_list *it,*next;
+	bool all = false;
+	const char *k = NULL;
+
+	if (strcmp(key, "lxc.proc") == 0)
+		all = true;
+	else if (strncmp(key, "lxc.proc.", sizeof("lxc.proc.") - 1) == 0)
+		k = key + sizeof("lxc.proc.") - 1;
+	else
+		return -1;
+
+	lxc_list_for_each_safe(it, &c->procs, next) {
+		struct lxc_proc *proc = it->elem;
+		if (!all && strcmp(proc->filename, k) != 0)
+			continue;
+		lxc_list_del(it);
+		free(proc->filename);
+		free(proc->value);
+		free(proc);
+		free(it);
+	}
+
+	return 0;
+}
+
 int lxc_clear_groups(struct lxc_conf *c)
 {
 	struct lxc_list *it,*next;
@@ -3524,6 +3584,7 @@ void lxc_conf_free(struct lxc_conf *conf)
 	lxc_clear_environment(conf);
 	lxc_clear_limits(conf, "lxc.prlimit");
 	lxc_clear_sysctls(conf, "lxc.sysctl");
+	lxc_clear_procs(conf, "lxc.proc");
 	free(conf->cgroup_meta.dir);
 	free(conf->cgroup_meta.controllers);
 	free(conf);

src/lxc/conf.h

@@ -106,6 +106,16 @@ struct lxc_sysctl {
 };
 
 /*
+ * Defines a structure to configure proc filesystem at runtime.
+ * @filename : the proc filesystem will be configured without the "lxc.proc" prefix
+ * @value    : the value to set
+ */
+struct lxc_proc {
+	char *filename;
+	char *value;
+};
+
+/*
  * id_map is an id map entry.  Form in confile is:
  * lxc.idmap = u 0    9800 100
  * lxc.idmap = u 1000 9900 100
@@ -383,6 +393,9 @@ struct lxc_conf {
 
 	/* sysctls */
 	struct lxc_list sysctls;
+
+	/* procs */
+	struct lxc_list procs;
 };
 
 int write_id_mapping(enum idtype idtype, pid_t pid, const char *buf,
@@ -443,5 +456,7 @@ extern int run_script(const char *name, const char *section, const char *script,
 extern int in_caplist(int cap, struct lxc_list *caps);
 extern int setup_sysctl_parameters(struct lxc_list *sysctls);
 extern int lxc_clear_sysctls(struct lxc_conf *c, const char *key);
+extern int setup_proc_filesystem(struct lxc_list *procs, pid_t pid);
+extern int lxc_clear_procs(struct lxc_conf *c, const char *key);
 
 #endif /* __LXC_CONF_H */

src/lxc/confile.c

@@ -142,6 +142,7 @@ lxc_config_define(tty_max);
 lxc_config_define(tty_dir);
 lxc_config_define(uts_name);
 lxc_config_define(sysctl);
+lxc_config_define(proc);
 
 static struct lxc_config_t config[] = {
                                            /* REMOVE in LXC 3.0 */
@@ -243,6 +244,7 @@ static struct lxc_config_t config[] = {
 	{ "lxc.tty.max",                   false,                  set_config_tty_max,                     get_config_tty_max,                     clr_config_tty_max,                   },
 	{ "lxc.uts.name",                  false,                  set_config_uts_name,                    get_config_uts_name,                    clr_config_uts_name,                  },
 	{ "lxc.sysctl",                    false,                  set_config_sysctl,                      get_config_sysctl,                      clr_config_sysctl,                    },
+	{ "lxc.proc",                      false,                  set_config_proc,                        get_config_proc,                        clr_config_proc,                      },
 
 	/* [START]: REMOVE IN LXC 3.0 */
 	{ "lxc.pts",                       true,                   set_config_pty_max,                     get_config_pty_max,                     clr_config_pty_max,                   },
@@ -1562,6 +1564,55 @@ on_error:
 	return -1;
 }
 
+static int set_config_proc(const char *key, const char *value,
+			    struct lxc_conf *lxc_conf, void *data)
+{
+	const char *subkey;
+	struct lxc_list *proclist = NULL;
+	struct lxc_proc *procelem = NULL;
+
+	if (lxc_config_value_empty(value))
+		return clr_config_proc(key, lxc_conf, NULL);
+
+	if (strncmp(key, "lxc.proc.", sizeof("lxc.proc.") -1) != 0)
+		return -1;
+
+	subkey = key + sizeof("lxc.proc.") - 1;
+	if (*subkey == '\0')
+		return -EINVAL;
+
+	proclist = malloc(sizeof(*proclist));
+	if (!proclist)
+		goto on_error;
+
+	procelem = malloc(sizeof(*procelem));
+	if (!procelem)
+		goto on_error;
+	memset(procelem, 0, sizeof(*procelem));
+
+	procelem->filename = strdup(subkey);
+	procelem->value = strdup(value);
+
+	if (!procelem->filename || !procelem->value)
+		goto on_error;
+
+	proclist->elem = procelem;
+
+	lxc_list_add_tail(&lxc_conf->procs, proclist);
+
+	return 0;
+
+on_error:
+	free(proclist);
+	if (procelem) {
+		free(procelem->filename);
+		free(procelem->value);
+		free(procelem);
+	}
+
+	return -1;
+}
+
 static int set_config_idmaps(const char *key, const char *value,
 			     struct lxc_conf *lxc_conf, void *data)
 {
@@ -3414,6 +3465,40 @@ static int get_config_sysctl(const char *key, char *retv, int inlen,
 	return fulllen;
 }
 
+static int get_config_proc(const char *key, char *retv, int inlen,
+			    struct lxc_conf *c, void *data)
+{
+	struct lxc_list *it;
+	int len;
+	int fulllen = 0;
+	bool get_all = false;
+
+	if (!retv)
+		inlen = 0;
+	else
+		memset(retv, 0, inlen);
+
+	if (strcmp(key, "lxc.proc") == 0)
+		get_all = true;
+	else if (strncmp(key, "lxc.proc.", sizeof("lxc.proc.") - 1) == 0)
+		key += sizeof("lxc.proc.") - 1;
+	else
+		return -1;
+
+	lxc_list_for_each(it, &c->procs) {
+		struct lxc_proc *proc = it->elem;
+
+		if (get_all) {
+			strprint(retv, inlen, "lxc.proc.%s = %s\n",
+				 proc->filename, proc->value);
+		} else if (strcmp(proc->filename, key) == 0) {
+			strprint(retv, inlen, "%s", proc->value);
+		}
+	}
+
+	return fulllen;
+}
+
 static int get_config_noop(const char *key, char *retv, int inlen,
 			   struct lxc_conf *c, void *data)
 {
@@ -3797,6 +3882,12 @@ static inline int clr_config_sysctl(const char *key, struct lxc_conf *c,
 	return lxc_clear_sysctls(c, key);
 }
 
+static inline int clr_config_proc(const char *key, struct lxc_conf *c,
+				   void *data)
+{
+	return lxc_clear_procs(c, key);
+}
+
 static inline int clr_config_includefiles(const char *key, struct lxc_conf *c,
 					  void *data)
 {

src/lxc/start.c

@@ -1386,6 +1386,12 @@ static int lxc_spawn(struct lxc_handler *handler)
 		goto out_delete_net;
 	}
 
+	if (!lxc_list_empty(&conf->procs)) {
+		ret = setup_proc_filesystem(&conf->procs, handler->pid);
+		if (ret < 0)
+			goto out_delete_net;
+	}
+
 	/* Tell the child to continue its initialization. We'll get
 	 * LXC_SYNC_CGROUP when it is ready for us to setup cgroups.
 	 */

src/tests/get_item.c

@@ -309,6 +309,73 @@ int main(int argc, char *argv[])
 	}
 	printf("lxc.sysctl returned %d %s\n", ret, v3);
 
+#define PROC_OOM_SCORE_ADJ "lxc.proc.oom_score_adj = 10\n"
+#define ALL_PROCS "lxc.proc.setgroups = allow\n" PROC_OOM_SCORE_ADJ
+
+	ret = c->get_config_item(c, "lxc.proc", v3, 2047);
+	if (ret != 0) {
+		fprintf(stderr, "%d: get_config_item(proc) returned %d\n", __LINE__, ret);
+		goto out;
+	}
+
+	if (!c->set_config_item(c, "lxc.proc.setgroups", "allow")) {
+		fprintf(stderr, "%d: failed to set lxc.proc.setgroups\n", __LINE__);
+		goto out;
+	}
+
+	ret = c->get_config_item(c, "lxc.proc.setgroups", v2, 255);
+	if (ret < 0) {
+		fprintf(stderr, "%d: get_config_item(lxc.proc.setgroups) returned %d\n", __LINE__, ret);
+		goto out;
+	}
+	if (strcmp(v2, "allow")) {
+		fprintf(stderr, "%d: lxc.proc.setgroups returned wrong value: %d %s not 10\n", __LINE__, ret, v2);
+		goto out;
+	}
+	printf("lxc.proc.setgroups returned %d %s\n", ret, v2);
+
+	if (!c->set_config_item(c, "lxc.proc.oom_score_adj", "10")) {
+		fprintf(stderr, "%d: failed to set lxc.proc.oom_score_adj\n", __LINE__);
+		goto out;
+	}
+
+	ret = c->get_config_item(c, "lxc.proc.oom_score_adj", v2, 255);
+	if (ret < 0) {
+		fprintf(stderr, "%d: get_config_item(lxc.proc.oom_score_adj) returned %d\n", __LINE__, ret);
+		goto out;
+	}
+	if (strcmp(v2, "10")) {
+		fprintf(stderr, "%d: lxc.proc.oom_score_adj returned wrong value: %d %s not 10\n", __LINE__, ret, v2);
+		goto out;
+	}
+	printf("lxc.proc.oom_score_adj returned %d %s\n", ret, v2);
+
+	ret = c->get_config_item(c, "lxc.proc", v3, 2047);
+	if (ret != sizeof(ALL_PROCS)-1) {
+		fprintf(stderr, "%d: get_config_item(proc) returned %d\n", __LINE__, ret);
+		goto out;
+	}
+	if (strcmp(v3, ALL_PROCS)) {
+		fprintf(stderr, "%d: lxc.proc returned wrong value: %d %s not %d %s\n", __LINE__, ret, v3, (int)sizeof(ALL_PROCS) - 1, ALL_PROCS);
+		goto out;
+	}
+	printf("lxc.proc returned %d %s\n", ret, v3);
+
+	if (!c->clear_config_item(c, "lxc.proc.setgroups")) {
+		fprintf(stderr, "%d: failed clearing lxc.proc.setgroups\n", __LINE__);
+		goto out;
+	}
+	ret = c->get_config_item(c, "lxc.proc", v3, 2047);
+	if (ret < 0) {
+		fprintf(stderr, "%d: get_config_item(proc) returned %d\n", __LINE__, ret);
+		goto out;
+	}
+	if (strcmp(v3, PROC_OOM_SCORE_ADJ)) {
+		fprintf(stderr, "%d: lxc.proc returned wrong value: %d %s not %d %s\n", __LINE__, ret, v3, (int)sizeof(PROC_OOM_SCORE_ADJ) - 1, PROC_OOM_SCORE_ADJ);
+		goto out;
+	}
+	printf("lxc.proc returned %d %s\n", ret, v3);
+
 	if (!c->set_config_item(c, "lxc.aa_profile", "unconfined")) {
 		fprintf(stderr, "%d: failed to set aa_profile\n", __LINE__);
 		goto out;

src/tests/parse_config_file.c

@@ -929,6 +929,13 @@ int main(int argc, char *argv[])
 		goto non_test_error;
 	}
 
+	/* lxc.proc */
+	if (set_get_compare_clear_save_load(c, "lxc.proc.oom_score_adj", "10", tmpf,
+					    true) < 0) {
+		lxc_error("%s\n", "lxc.proc.oom_score_adj");
+		goto non_test_error;
+	}
+
 	/* REMOVE IN LXC 3.0
 	   legacy lxc.limit.* key
 	 */