]> git.proxmox.com Git - mirror_lxc.git/blobdiff - src/lxc/start.c
network: Fixes bug in macvlan mode selection
[mirror_lxc.git] / src / lxc / start.c
index 967d23c1bbdd75886c5b5e06dd246f382290c25e..b26d0ce7a49fe99c7cf27507706d6e7e10f1d887 100644 (file)
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
-#define _GNU_SOURCE
-#include "config.h"
-
-#include <alloca.h>
+#ifndef _GNU_SOURCE
+#define _GNU_SOURCE 1
+#endif
 #include <dirent.h>
 #include <errno.h>
 #include <fcntl.h>
@@ -37,7 +36,6 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
-#include <unistd.h>
 #include <sys/file.h>
 #include <sys/mount.h>
 #include <sys/param.h>
 #include <sys/types.h>
 #include <sys/un.h>
 #include <sys/wait.h>
-
-#if HAVE_LIBCAP
-#include <sys/capability.h>
-#endif
-
-#if !HAVE_DECL_PR_CAPBSET_DROP
-#define PR_CAPBSET_DROP 24
-#endif
-
-#if !HAVE_DECL_PR_SET_NO_NEW_PRIVS
-#define PR_SET_NO_NEW_PRIVS 38
-#endif
-
-#if !HAVE_DECL_PR_GET_NO_NEW_PRIVS
-#define PR_GET_NO_NEW_PRIVS 39
-#endif
+#include <unistd.h>
 
 #include "af_unix.h"
 #include "caps.h"
 #include "commands.h"
 #include "commands_utils.h"
 #include "conf.h"
+#include "config.h"
 #include "confile_utils.h"
 #include "error.h"
+#include "file_utils.h"
 #include "list.h"
-#include "lsm/lsm.h"
 #include "log.h"
+#include "lsm/lsm.h"
 #include "lxccontainer.h"
 #include "lxclock.h"
 #include "lxcseccomp.h"
+#include "macro.h"
 #include "mainloop.h"
+#include "memory_utils.h"
 #include "monitor.h"
 #include "namespace.h"
 #include "network.h"
+#include "raw_syscalls.h"
 #include "start.h"
 #include "storage/storage.h"
 #include "storage/storage_utils.h"
 #include "sync.h"
+#include "syscall_wrappers.h"
 #include "terminal.h"
 #include "utils.h"
 
+#if HAVE_LIBCAP
+#include <sys/capability.h>
+#endif
+
 #ifndef HAVE_STRLCPY
 #include "include/strlcpy.h"
 #endif
@@ -104,16 +97,13 @@ static void lxc_destroy_container_on_signal(struct lxc_handler *handler,
 
 static void print_top_failing_dir(const char *path)
 {
+       __do_free char *copy = NULL;
        int ret;
-       size_t len;
-       char *copy, *e, *p, saved;
-
-       len = strlen(path);
-       copy = alloca(len + 1);
-       (void)strlcpy(copy, path, len + 1);
+       char *e, *p, saved;
 
+       copy = must_copy_string(path);
        p = copy;
-       e = copy + len;
+       e = copy + strlen(path);
 
        while (p < e) {
                while (p < e && *p == '/')
@@ -187,8 +177,6 @@ static bool lxc_try_preserve_namespaces(struct lxc_handler *handler,
 
                fd = lxc_try_preserve_ns(pid, ns_info[i].proc_name);
                if (fd < 0) {
-                       handler->nsfd[i] = -EBADF;
-
                        /* Do not fail to start container on kernels that do
                         * not support interacting with namespaces through
                         * /proc.
@@ -213,6 +201,38 @@ static inline bool match_stdfds(int fd)
        return (fd == STDIN_FILENO || fd == STDOUT_FILENO || fd == STDERR_FILENO);
 }
 
+#ifdef HAVE_DLOG
+static bool match_dlog_fds(struct dirent *direntp)
+{
+       char path[PATH_MAX] = {0};
+       char link[PATH_MAX] = {0};
+       ssize_t linklen;
+       int ret;
+
+       ret = snprintf(path, PATH_MAX, "/proc/self/fd/%s", direntp->d_name);
+       if (ret < 0 || ret >= PATH_MAX) {
+               ERROR("Failed to create file descriptor name");
+               return false;
+       }
+
+       linklen = readlink(path, link, PATH_MAX);
+       if (linklen < 0) {
+               SYSERROR("Failed to read link path - \"%s\"", path);
+               return false;
+       } else if (linklen >= PATH_MAX) {
+               ERROR("The name of link path is too long - \"%s\"", path);
+               return false;
+       }
+
+       if (strcmp(link, "/dev/log_main") == 0 ||
+           strcmp(link, "/dev/log_system") == 0 ||
+           strcmp(link, "/dev/log_radio") == 0)
+               return true;
+
+       return false;
+}
+#endif
+
 int lxc_check_inherited(struct lxc_conf *conf, bool closeall,
                        int *fds_to_ignore, size_t len_fds)
 {
@@ -280,6 +300,11 @@ restart:
                if (match_stdfds(fd))
                        continue;
 
+#ifdef HAVE_DLOG
+               if (match_dlog_fds(direntp))
+                       continue;
+
+#endif
                if (closeall) {
                        close(fd);
                        closedir(dir);
@@ -302,7 +327,6 @@ restart:
 static int setup_signal_fd(sigset_t *oldmask)
 {
        int ret;
-       int sig;
        sigset_t mask;
        const int signals[] = {SIGBUS, SIGILL, SIGSEGV, SIGWINCH};
 
@@ -311,7 +335,7 @@ static int setup_signal_fd(sigset_t *oldmask)
        if (ret < 0)
                return -EBADF;
 
-       for (sig = 0; sig < (sizeof(signals) / sizeof(signals[0])); sig++) {
+       for (int sig = 0; sig < (sizeof(signals) / sizeof(signals[0])); sig++) {
                ret = sigdelset(&mask, signals[sig]);
                if (ret < 0)
                        return -EBADF;
@@ -384,13 +408,15 @@ static int signal_handler(int fd, uint32_t events, void *data,
        if (siginfo.ssi_signo == SIGHUP) {
                kill(hdlr->pid, SIGTERM);
                INFO("Killing %d since terminal hung up", hdlr->pid);
-               return hdlr->init_died ? LXC_MAINLOOP_CLOSE : LXC_MAINLOOP_CONTINUE;
+               return hdlr->init_died ? LXC_MAINLOOP_CLOSE
+                                      : LXC_MAINLOOP_CONTINUE;
        }
 
        if (siginfo.ssi_signo != SIGCHLD) {
                kill(hdlr->pid, siginfo.ssi_signo);
                INFO("Forwarded signal %d to pid %d", siginfo.ssi_signo, hdlr->pid);
-               return hdlr->init_died ? LXC_MAINLOOP_CLOSE : LXC_MAINLOOP_CONTINUE;
+               return hdlr->init_died ? LXC_MAINLOOP_CLOSE
+                                      : LXC_MAINLOOP_CONTINUE;
        }
 
        /* More robustness, protect ourself from a SIGCHLD sent
@@ -399,18 +425,24 @@ static int signal_handler(int fd, uint32_t events, void *data,
        if (siginfo.ssi_pid != hdlr->pid) {
                NOTICE("Received %d from pid %d instead of container init %d",
                       siginfo.ssi_signo, siginfo.ssi_pid, hdlr->pid);
-               return hdlr->init_died ? LXC_MAINLOOP_CLOSE : LXC_MAINLOOP_CONTINUE;
+               return hdlr->init_died ? LXC_MAINLOOP_CLOSE
+                                      : LXC_MAINLOOP_CONTINUE;
        }
 
        if (siginfo.ssi_code == CLD_STOPPED) {
                INFO("Container init process was stopped");
-               return hdlr->init_died ? LXC_MAINLOOP_CLOSE : LXC_MAINLOOP_CONTINUE;
-       } else if (siginfo.ssi_code == CLD_CONTINUED) {
+               return hdlr->init_died ? LXC_MAINLOOP_CLOSE
+                                      : LXC_MAINLOOP_CONTINUE;
+       }
+
+       if (siginfo.ssi_code == CLD_CONTINUED) {
                INFO("Container init process was continued");
-               return hdlr->init_died ? LXC_MAINLOOP_CLOSE : LXC_MAINLOOP_CONTINUE;
+               return hdlr->init_died ? LXC_MAINLOOP_CLOSE
+                                      : LXC_MAINLOOP_CONTINUE;
        }
 
        DEBUG("Container init process %d exited", hdlr->pid);
+
        return LXC_MAINLOOP_CLOSE;
 }
 
@@ -420,7 +452,6 @@ int lxc_serve_state_clients(const char *name, struct lxc_handler *handler,
        size_t retlen;
        ssize_t ret;
        struct lxc_list *cur, *next;
-       struct lxc_state_client *client;
        struct lxc_msg msg = {.type = lxc_msg_state, .value = state};
 
        if (state == THAWED)
@@ -440,7 +471,7 @@ int lxc_serve_state_clients(const char *name, struct lxc_handler *handler,
                return -E2BIG;
 
        lxc_list_for_each_safe(cur, &handler->conf->state_clients, next) {
-               client = cur->elem;
+               struct lxc_state_client *client = cur->elem;
 
                if (client->states[state] == 0) {
                        TRACE("State %s not registered for state client %d",
@@ -451,16 +482,9 @@ int lxc_serve_state_clients(const char *name, struct lxc_handler *handler,
                TRACE("Sending state %s to state client %d",
                      lxc_state2str(state), client->clientfd);
 
-       again:
-               ret = send(client->clientfd, &msg, sizeof(msg), MSG_NOSIGNAL);
-               if (ret <= 0) {
-                       if (errno == EINTR) {
-                               TRACE("Caught EINTR; retrying");
-                               goto again;
-                       }
-
+               ret = lxc_send_nointr(client->clientfd, &msg, sizeof(msg), MSG_NOSIGNAL);
+               if (ret <= 0)
                        SYSERROR("Failed to send message to client");
-               }
 
                /* kick client from list */
                lxc_list_del(cur);
@@ -478,7 +502,7 @@ static int lxc_serve_state_socket_pair(const char *name,
 {
        ssize_t ret;
 
-       if (!handler->backgrounded ||
+       if (!handler->daemonize ||
             handler->state_socket_pair[1] < 0 ||
            state == STARTING)
                return 0;
@@ -567,6 +591,33 @@ int lxc_poll(const char *name, struct lxc_handler *handler)
                goto out_mainloop_console;
        }
 
+#if HAVE_DECL_SECCOMP_NOTIF_GET_FD
+       if (handler->conf->has_seccomp_notify &&
+           handler->conf->seccomp_notify_proxy_addr.sun_path[1] != '\0') {
+                __do_close_prot_errno int notify_fd = -EBADF;
+
+               notify_fd = lxc_unix_connect(&handler->conf->seccomp_notify_proxy_addr);
+               if (notify_fd < 0)
+                       goto out_mainloop_console;
+
+               /* 30 second timeout */
+               ret = lxc_socket_set_timeout(notify_fd, 30, 30);
+               if (ret)
+                       goto out_mainloop_console;
+
+               ret = lxc_mainloop_add_handler(&descr,
+                                              handler->conf->seccomp_notify_fd,
+                                              seccomp_notify_handler, handler);
+               if (ret < 0) {
+                       ERROR("Failed to add seccomp notify handler for %d to mainloop",
+                             handler->conf->seccomp_notify_fd);
+                       goto out_mainloop_console;
+               }
+
+               handler->conf->seccomp_notify_proxy_fd = move_fd(notify_fd);
+       }
+#endif
+
        if (has_console) {
                struct lxc_terminal *console = &handler->conf->console;
 
@@ -663,6 +714,9 @@ void lxc_free_handler(struct lxc_handler *handler)
        if (handler->state_socket_pair[1] >= 0)
                close(handler->state_socket_pair[1]);
 
+       if (handler->cgroup_ops)
+               cgroup_exit(handler->cgroup_ops);
+
        handler->conf = NULL;
        free(handler);
        handler = NULL;
@@ -742,6 +796,8 @@ int lxc_init(const char *name, struct lxc_handler *handler)
        const char *loglevel;
        struct lxc_conf *conf = handler->conf;
 
+       handler->monitor_pid = lxc_raw_getpid();
+
        lsm_init();
        TRACE("Initialized LSM");
 
@@ -856,7 +912,7 @@ int lxc_init(const char *name, struct lxc_handler *handler)
        }
        TRACE("Chowned console");
 
-       handler->cgroup_ops = cgroup_init(handler);
+       handler->cgroup_ops = cgroup_init(handler->conf);
        if (!handler->cgroup_ops) {
                ERROR("Failed to initialize cgroup driver");
                goto out_delete_terminal;
@@ -874,7 +930,8 @@ int lxc_init(const char *name, struct lxc_handler *handler)
        return 0;
 
 out_destroy_cgroups:
-       handler->cgroup_ops->destroy(handler->cgroup_ops, handler);
+       handler->cgroup_ops->payload_destroy(handler->cgroup_ops, handler);
+       handler->cgroup_ops->monitor_destroy(handler->cgroup_ops, handler);
 
 out_delete_terminal:
        lxc_terminal_delete(&handler->conf->console);
@@ -968,8 +1025,8 @@ void lxc_fini(const char *name, struct lxc_handler *handler)
 
        lsm_process_cleanup(handler->conf, handler->lxcpath);
 
-       cgroup_ops->destroy(cgroup_ops, handler);
-       cgroup_exit(cgroup_ops);
+       cgroup_ops->payload_destroy(cgroup_ops, handler);
+       cgroup_ops->monitor_destroy(cgroup_ops, handler);
 
        if (handler->conf->reboot == REBOOT_NONE) {
                /* For all new state clients simply close the command socket.
@@ -996,6 +1053,9 @@ void lxc_fini(const char *name, struct lxc_handler *handler)
                lxc_set_state(name, handler, STOPPED);
        }
 
+       /* Avoid lingering namespace references. */
+       lxc_put_nsfds(handler);
+
        ret = run_lxc_hooks(name, "post-stop", handler->conf, NULL);
        if (ret < 0) {
                ERROR("Failed to run lxc.hook.post-stop for container \"%s\"", name);
@@ -1061,14 +1121,17 @@ void lxc_abort(const char *name, struct lxc_handler *handler)
 
 static int do_start(void *data)
 {
+       struct lxc_handler *handler = data;
+       ATTR_UNUSED __do_close_prot_errno int data_sock0 = handler->data_sock[0],
+                                             data_sock1 = handler->data_sock[1];
        int ret;
        char path[PATH_MAX];
-       bool have_cap_setgid;
        uid_t new_uid;
        gid_t new_gid;
        struct lxc_list *iterator;
+       uid_t nsuid = 0;
+       gid_t nsgid = 0;
        int devnull_fd = -1;
-       struct lxc_handler *handler = data;
 
        lxc_sync_fini_parent(handler);
 
@@ -1078,7 +1141,7 @@ static int do_start(void *data)
         * exit before we set the pdeath signal leading to a unsupervized
         * container.
         */
-       ret = lxc_set_death_signal(SIGKILL);
+       ret = lxc_set_death_signal(SIGKILL, handler->monitor_pid);
        if (ret < 0) {
                SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL");
                goto out_warn_father;
@@ -1134,22 +1197,20 @@ static int do_start(void *data)
         * privilege over our namespace.
         */
        if (!lxc_list_empty(&handler->conf->id_map)) {
-               uid_t nsuid = (handler->conf->root_nsuid_map != NULL)
-                                 ? 0
-                                 : handler->conf->init_uid;
-               gid_t nsgid = (handler->conf->root_nsgid_map != NULL)
-                                 ? 0
-                                 : handler->conf->init_gid;
-
-               ret = lxc_switch_uid_gid(nsuid, nsgid);
-               if (ret < 0)
+               if (!handler->conf->root_nsuid_map)
+                       nsuid = handler->conf->init_uid;
+
+               if (!handler->conf->root_nsgid_map)
+                       nsgid = handler->conf->init_gid;
+
+               if (!lxc_switch_uid_gid(nsuid, nsgid))
                        goto out_warn_father;
 
                /* Drop groups only after we switched to a valid gid in the new
                 * user namespace.
                 */
-               ret = lxc_setgroups(0, NULL);
-               if (ret < 0 && (handler->am_root || errno != EPERM))
+               if (!lxc_setgroups(0, NULL) &&
+                   (handler->am_root || errno != EPERM))
                        goto out_warn_father;
 
                ret = prctl(PR_SET_DUMPABLE, prctl_arg(1), prctl_arg(0),
@@ -1158,7 +1219,7 @@ static int do_start(void *data)
                        goto out_warn_father;
 
                /* set{g,u}id() clears deathsignal */
-               ret = lxc_set_death_signal(SIGKILL);
+               ret = lxc_set_death_signal(SIGKILL, handler->monitor_pid);
                if (ret < 0) {
                        SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL");
                        goto out_warn_father;
@@ -1186,7 +1247,7 @@ static int do_start(void *data)
         * means that migration won't work, but at least we won't spew output
         * where it isn't wanted.
         */
-       if (handler->backgrounded && !handler->conf->autodev) {
+       if (handler->daemonize && !handler->conf->autodev) {
                ret = access(path, F_OK);
                if (ret != 0) {
                        devnull_fd = open_devnull();
@@ -1219,10 +1280,16 @@ static int do_start(void *data)
        if (handler->ns_clone_flags & CLONE_NEWCGROUP) {
                ret = unshare(CLONE_NEWCGROUP);
                if (ret < 0) {
-                       INFO("Failed to unshare CLONE_NEWCGROUP");
-                       goto out_warn_father;
+                       if (errno != EINVAL) {
+                               SYSERROR("Failed to unshare CLONE_NEWCGROUP");
+                               goto out_warn_father;
+                       }
+
+                       handler->ns_clone_flags &= ~CLONE_NEWCGROUP;
+                       SYSINFO("Kernel does not support CLONE_NEWCGROUP");
+               } else {
+                       INFO("Unshared CLONE_NEWCGROUP");
                }
-               INFO("Unshared CLONE_NEWCGROUP");
        }
 
        /* Add the requested environment variables to the current environment to
@@ -1240,8 +1307,6 @@ static int do_start(void *data)
 
        /* Setup the container, ip, names, utsname, ... */
        ret = lxc_setup(handler);
-       close(handler->data_sock[1]);
-       close(handler->data_sock[0]);
        if (ret < 0) {
                ERROR("Failed to setup container \"%s\"", handler->name);
                goto out_warn_father;
@@ -1274,7 +1339,7 @@ static int do_start(void *data)
         * make sure that that pty is stdin,stdout,stderr.
         */
         if (handler->conf->console.slave >= 0) {
-                if (handler->backgrounded || !handler->conf->is_execute)
+                if (handler->daemonize || !handler->conf->is_execute)
                         ret = set_stdfds(handler->conf->console.slave);
                 else
                         ret = lxc_terminal_set_stdfds(handler->conf->console.slave);
@@ -1292,6 +1357,20 @@ static int do_start(void *data)
        if (ret < 0)
                goto out_warn_father;
 
+#if HAVE_DECL_SECCOMP_NOTIF_GET_FD
+       if (handler->conf->has_seccomp_notify) {
+               ret = lxc_abstract_unix_send_fds(data_sock0,
+                                                &handler->conf->seccomp_notify_fd,
+                                                1, NULL, 0);
+               if (ret < 0) {
+                       SYSERROR("Failed to send seccomp notify fd to parent");
+                       goto out_warn_father;
+               }
+               close(handler->conf->seccomp_notify_fd);
+               handler->conf->seccomp_notify_fd = -EBADF;
+       }
+#endif
+
        ret = run_lxc_hooks(handler->name, "start", handler->conf, NULL);
        if (ret < 0) {
                ERROR("Failed to run lxc.hook.start for container \"%s\"",
@@ -1301,7 +1380,7 @@ static int do_start(void *data)
 
        close(handler->sigfd);
 
-       if (handler->conf->console.slave < 0 && handler->backgrounded) {
+       if (handler->conf->console.slave < 0 && handler->daemonize) {
                if (devnull_fd < 0) {
                        devnull_fd = open_devnull();
                        if (devnull_fd < 0)
@@ -1372,25 +1451,27 @@ static int do_start(void *data)
        new_uid = handler->conf->init_uid;
        new_gid = handler->conf->init_gid;
 
-       /* If we are in a new user namespace we already dropped all groups when
-       *  we switched to root in the new user namespace further above. Only
-       *  drop groups if we can, so ensure that we have necessary privilege.
-        */
-       #if HAVE_LIBCAP
-       have_cap_setgid = lxc_proc_cap_is_set(CAP_SETGID, CAP_EFFECTIVE);
-       #else
-       have_cap_setgid = false;
-       #endif
-       if (lxc_list_empty(&handler->conf->id_map) && have_cap_setgid) {
-               ret = lxc_setgroups(0, NULL);
-               if (ret < 0)
-                       goto out_warn_father;
-       }
+       /* Avoid unnecessary syscalls. */
+       if (new_uid == nsuid)
+               new_uid = LXC_INVALID_UID;
 
-       ret = lxc_switch_uid_gid(new_uid, new_gid);
-       if (ret < 0)
+       if (new_gid == nsgid)
+               new_gid = LXC_INVALID_GID;
+
+       if (!lxc_switch_uid_gid(new_uid, new_gid))
                goto out_warn_father;
 
+       /* If we are in a new user namespace we already dropped all groups when
+        * we switched to root in the new user namespace further above. Only
+        * drop groups if we can, so ensure that we have necessary privilege.
+        */
+       if (lxc_list_empty(&handler->conf->id_map))
+               #if HAVE_LIBCAP
+               if (lxc_proc_cap_is_set(CAP_SETGID, CAP_EFFECTIVE))
+               #endif
+                       if (!lxc_setgroups(0, NULL))
+                               goto out_warn_father;
+
        ret = lxc_ambient_caps_down();
        if (ret < 0) {
                ERROR("Failed to clear ambient capabilities");
@@ -1398,7 +1479,7 @@ static int do_start(void *data)
        }
 
        if (handler->conf->monitor_signal_pdeath != SIGKILL) {
-               ret = lxc_set_death_signal(handler->conf->monitor_signal_pdeath);
+               ret = lxc_set_death_signal(handler->conf->monitor_signal_pdeath, handler->monitor_pid);
                if (ret < 0) {
                        SYSERROR("Failed to set PR_SET_PDEATHSIG to %d",
                                 handler->conf->monitor_signal_pdeath);
@@ -1469,11 +1550,11 @@ int resolve_clone_flags(struct lxc_handler *handler)
        struct lxc_conf *conf = handler->conf;
 
        for (i = 0; i < LXC_NS_MAX; i++) {
-               if (conf->ns_keep != 0) {
-                       if ((conf->ns_keep & ns_info[i].clone_flag) == 0)
+               if (conf->ns_keep) {
+                       if (!(conf->ns_keep & ns_info[i].clone_flag))
                                handler->ns_clone_flags |= ns_info[i].clone_flag;
-               } else if (conf->ns_clone != 0) {
-                       if ((conf->ns_clone & ns_info[i].clone_flag) > 0)
+               } else if (conf->ns_clone) {
+                       if ((conf->ns_clone & ns_info[i].clone_flag))
                                handler->ns_clone_flags |= ns_info[i].clone_flag;
                } else {
                        if (i == LXC_NS_USER && lxc_list_empty(&handler->conf->id_map))
@@ -1505,7 +1586,7 @@ int resolve_clone_flags(struct lxc_handler *handler)
  * getpid() in the child would return the parent's pid. This is all fixed in
  * newer glibc versions where the getpid() cache is removed and the pid/tid is
  * not reset anymore.
- * However, if for whatever reason you - dear commiter - somehow need to get the
+ * However, if for whatever reason you - dear committer - somehow need to get the
  * pid of the dummy intermediate process for do_share_ns() you need to call
  * lxc_raw_getpid(). The next lxc_raw_clone() call does not employ CLONE_VM and
  * will be fine.
@@ -1543,75 +1624,6 @@ static inline int do_share_ns(void *arg)
        return 0;
 }
 
-static int lxc_setup_shmount(struct lxc_conf *conf)
-{
-       size_t len_cont;
-       char *full_cont_path;
-       int ret = -1;
-
-       /* Construct the shmount path under the container root. */
-       len_cont = strlen(conf->rootfs.mount) + 1 + strlen(conf->shmount.path_cont);
-       /* +1 for the terminating '\0' */
-       full_cont_path = malloc(len_cont + 1);
-       if (!full_cont_path) {
-               SYSERROR("Not enough memory");
-               return -ENOMEM;
-       }
-
-       ret = snprintf(full_cont_path, len_cont + 1, "%s/%s",
-                      conf->rootfs.mount, conf->shmount.path_cont);
-       if (ret < 0 || ret >= len_cont + 1) {
-               SYSERROR("Failed to create filename");
-               free(full_cont_path);
-               return -1;
-       }
-
-       /* Check if shmount point is already set up. */
-       if (is_shared_mountpoint(conf->shmount.path_host)) {
-               INFO("Path \"%s\" is already MS_SHARED. Reusing",
-                    conf->shmount.path_host);
-               free(full_cont_path);
-               return 0;
-       }
-
-       /* Create host and cont mount paths */
-       ret = mkdir_p(conf->shmount.path_host, 0711);
-       if (ret < 0 && errno != EEXIST) {
-               SYSERROR("Failed to create directory \"%s\"",
-                        conf->shmount.path_host);
-               free(full_cont_path);
-               return ret;
-       }
-
-       ret = mkdir_p(full_cont_path, 0711);
-       if (ret < 0 && errno != EEXIST) {
-               SYSERROR("Failed to create directory \"%s\"", full_cont_path);
-               free(full_cont_path);
-               return ret;
-       }
-
-       /* Prepare host mountpoint */
-       ret = mount("tmpfs", conf->shmount.path_host, "tmpfs", 0,
-                   "size=100k,mode=0711");
-       if (ret < 0) {
-               SYSERROR("Failed to mount \"%s\"", conf->shmount.path_host);
-               free(full_cont_path);
-               return ret;
-       }
-
-       ret = mount(conf->shmount.path_host, conf->shmount.path_host, "none",
-                   MS_REC | MS_SHARED, "");
-       if (ret < 0) {
-               SYSERROR("Failed to make shared \"%s\"", conf->shmount.path_host);
-               free(full_cont_path);
-               return ret;
-       }
-
-       INFO("Setup shared mount point \"%s\"", conf->shmount.path_host);
-       free(full_cont_path);
-       return 0;
-}
-
 /* lxc_spawn() performs crucial setup tasks and clone()s the new process which
  * exec()s the requested container binary.
  * Note that lxc_spawn() runs in the parent namespaces. Any operations performed
@@ -1621,6 +1633,7 @@ static int lxc_setup_shmount(struct lxc_conf *conf)
  */
 static int lxc_spawn(struct lxc_handler *handler)
 {
+       __do_close_prot_errno int data_sock0 = -EBADF, data_sock1 = -EBADF;
        int i, ret;
        char pidstr[20];
        bool wants_to_map_ids;
@@ -1651,30 +1664,14 @@ static int lxc_spawn(struct lxc_handler *handler)
 
        ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0,
                         handler->data_sock);
-       if (ret < 0) {
-               lxc_sync_fini(handler);
-               return -1;
-       }
+       if (ret < 0)
+               goto out_sync_fini;
+       data_sock0 = handler->data_sock[0];
+       data_sock1 = handler->data_sock[1];
 
        ret = resolve_clone_flags(handler);
-       if (ret < 0) {
-               lxc_sync_fini(handler);
-               return -1;
-       }
-
-       if (conf->shmount.path_host) {
-               if (!conf->shmount.path_cont) {
-                       lxc_sync_fini(handler);
-                       return -1;
-               }
-
-               ret = lxc_setup_shmount(conf);
-               if (ret < 0) {
-                       ERROR("Failed to setup shared mount point");
-                       lxc_sync_fini(handler);
-                       return -1;
-               }
-       }
+       if (ret < 0)
+               goto out_sync_fini;
 
        if (handler->ns_clone_flags & CLONE_NEWNET) {
                if (!lxc_list_empty(&conf->network)) {
@@ -1687,8 +1684,7 @@ static int lxc_spawn(struct lxc_handler *handler)
                        ret = lxc_find_gateway_addresses(handler);
                        if (ret < 0) {
                                ERROR("Failed to find gateway addresses");
-                               lxc_sync_fini(handler);
-                               return -1;
+                               goto out_sync_fini;
                        }
 
                        /* That should be done before the clone because we will
@@ -1702,12 +1698,7 @@ static int lxc_spawn(struct lxc_handler *handler)
                }
        }
 
-       if (!cgroup_init(handler)) {
-               ERROR("Failed initializing cgroup support");
-               goto out_delete_net;
-       }
-
-       if (!cgroup_ops->create(cgroup_ops, handler)) {
+       if (!cgroup_ops->payload_create(cgroup_ops, handler)) {
                ERROR("Failed creating cgroups");
                goto out_delete_net;
        }
@@ -1801,7 +1792,7 @@ static int lxc_spawn(struct lxc_handler *handler)
                goto out_delete_net;
        }
 
-       if (!cgroup_ops->enter(cgroup_ops, handler->pid))
+       if (!cgroup_ops->payload_enter(cgroup_ops, handler->pid))
                goto out_delete_net;
 
        if (!cgroup_ops->chown(cgroup_ops, handler->conf))
@@ -1819,12 +1810,10 @@ static int lxc_spawn(struct lxc_handler *handler)
                DEBUG("Preserved net namespace via fd %d", ret);
 
                ret = lxc_netns_set_nsid(handler->nsfd[LXC_NS_NET]);
-               if (ret < 0) {
-                       errno = -ret;
-                       SYSERROR("Failed to allocate new network namespace id");
-               } else {
+               if (ret < 0)
+                       SYSWARN("Failed to allocate new network namespace id");
+               else
                        TRACE("Allocated new network namespace id");
-               }
        }
 
        /* Create the network configuration. */
@@ -1930,7 +1919,7 @@ static int lxc_spawn(struct lxc_handler *handler)
        }
 
        /* Now all networks are created, network devices are moved into place,
-        * and the correct names and ifindeces in the respective namespaces have
+        * and the correct names and ifindices in the respective namespaces have
         * been recorded. The corresponding structs have now all been filled. So
         * log them for debugging purposes.
         */
@@ -1943,6 +1932,26 @@ static int lxc_spawn(struct lxc_handler *handler)
                goto out_delete_net;
        }
 
+#if HAVE_DECL_SECCOMP_NOTIF_GET_FD
+       if (handler->conf->has_seccomp_notify) {
+               ret = lxc_abstract_unix_recv_fds(handler->data_sock[1],
+                                                &handler->conf->seccomp_notify_fd,
+                                                1, NULL, 0);
+               if (ret < 0) {
+                       SYSERROR("Failed to receive seccomp notify fd from child");
+                       goto out_delete_net;
+               }
+
+               ret = seccomp_notif_alloc(&handler->conf->seccomp_notify_req,
+                                         &handler->conf->seccomp_notify_resp);
+               if (ret) {
+                       errno = ret;
+                       ret = -1;
+                       goto out_delete_net;
+               }
+       }
+#endif
+
        ret = handler->ops->post_start(handler, handler->data);
        if (ret < 0)
                goto out_abort;
@@ -1963,6 +1972,8 @@ out_delete_net:
 
 out_abort:
        lxc_abort(name, handler);
+
+out_sync_fini:
        lxc_sync_fini(handler);
        if (handler->pinfd >= 0) {
                close(handler->pinfd);
@@ -1974,10 +1985,11 @@ out_abort:
 
 int __lxc_start(const char *name, struct lxc_handler *handler,
                struct lxc_operations* ops, void *data, const char *lxcpath,
-               bool backgrounded, int *error_num)
+               bool daemonize, int *error_num)
 {
        int ret, status;
        struct lxc_conf *conf = handler->conf;
+       struct cgroup_ops *cgroup_ops;
 
        ret = lxc_init(name, handler);
        if (ret < 0) {
@@ -1986,10 +1998,24 @@ int __lxc_start(const char *name, struct lxc_handler *handler,
        }
        handler->ops = ops;
        handler->data = data;
-       handler->backgrounded = backgrounded;
+       handler->daemonize = daemonize;
+       cgroup_ops = handler->cgroup_ops;
 
        if (!attach_block_device(handler->conf)) {
                ERROR("Failed to attach block device");
+               ret = -1;
+               goto out_fini_nonet;
+       }
+
+       if (!cgroup_ops->monitor_create(cgroup_ops, handler)) {
+               ERROR("Failed to create monitor cgroup");
+               ret = -1;
+               goto out_fini_nonet;
+       }
+
+       if (!cgroup_ops->monitor_enter(cgroup_ops, handler->monitor_pid)) {
+               ERROR("Failed to enter monitor cgroup");
+               ret = -1;
                goto out_fini_nonet;
        }
 
@@ -2018,11 +2044,6 @@ int __lxc_start(const char *name, struct lxc_handler *handler,
                ERROR("Failed to spawn container \"%s\"", name);
                goto out_detach_blockdev;
        }
-       /* close parent side of data socket */
-       close(handler->data_sock[0]);
-       handler->data_sock[0] = -1;
-       close(handler->data_sock[1]);
-       handler->data_sock[1] = -1;
 
        handler->conf->reboot = REBOOT_NONE;
 
@@ -2034,6 +2055,7 @@ int __lxc_start(const char *name, struct lxc_handler *handler,
 
        if (!handler->init_died && handler->pid > 0) {
                ERROR("Child process is not killed");
+               ret = -1;
                goto out_abort;
        }
 
@@ -2122,20 +2144,20 @@ static struct lxc_operations start_ops = {
 };
 
 int lxc_start(const char *name, char *const argv[], struct lxc_handler *handler,
-             const char *lxcpath, bool backgrounded, int *error_num)
+             const char *lxcpath, bool daemonize, int *error_num)
 {
        struct start_args start_arg = {
                .argv = argv,
        };
 
        TRACE("Doing lxc_start");
-       return __lxc_start(name, handler, &start_ops, &start_arg, lxcpath, backgrounded, error_num);
+       return __lxc_start(name, handler, &start_ops, &start_arg, lxcpath, daemonize, error_num);
 }
 
 static void lxc_destroy_container_on_signal(struct lxc_handler *handler,
                                            const char *name)
 {
-       char destroy[MAXPATHLEN];
+       char destroy[PATH_MAX];
        struct lxc_container *c;
        int ret = 0;
        bool bret = true;
@@ -2149,8 +2171,8 @@ static void lxc_destroy_container_on_signal(struct lxc_handler *handler,
        }
        INFO("Destroyed rootfs for container \"%s\"", name);
 
-       ret = snprintf(destroy, MAXPATHLEN, "%s/%s", handler->lxcpath, name);
-       if (ret < 0 || ret >= MAXPATHLEN) {
+       ret = snprintf(destroy, PATH_MAX, "%s/%s", handler->lxcpath, name);
+       if (ret < 0 || ret >= PATH_MAX) {
                ERROR("Error destroying directory for container \"%s\"", name);
                return;
        }