src/nspawn/nspawn-cgroup.c

   1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
   2
   3 /***
   4   This file is part of systemd.
   5
   6   Copyright 2015 Lennart Poettering
   7
   8   systemd is free software; you can redistribute it and/or modify it
   9   under the terms of the GNU Lesser General Public License as published by
  10   the Free Software Foundation; either version 2.1 of the License, or
  11   (at your option) any later version.
  12
  13   systemd is distributed in the hope that it will be useful, but
  14   WITHOUT ANY WARRANTY; without even the implied warranty of
  15   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  16   Lesser General Public License for more details.
  17
  18   You should have received a copy of the GNU Lesser General Public License
  19   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  20 ***/
  21
  22 #include <sys/mount.h>
  23
  24 #include "alloc-util.h"
  25 #include "cgroup-util.h"
  26 #include "fd-util.h"
  27 #include "fileio.h"
  28 #include "mkdir.h"
  29 #include "nspawn-cgroup.h"
  30 #include "string-util.h"
  31 #include "strv.h"
  32 #include "util.h"
  33
  34 int chown_cgroup(pid_t pid, uid_t uid_shift) {
  35         _cleanup_free_ char *path = NULL, *fs = NULL;
  36         _cleanup_close_ int fd = -1;
  37         const char *fn;
  38         int r;
  39
  40         r = cg_pid_get_path(NULL, pid, &path);
  41         if (r < 0)
  42                 return log_error_errno(r, "Failed to get container cgroup path: %m");
  43
  44         r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, path, NULL, &fs);
  45         if (r < 0)
  46                 return log_error_errno(r, "Failed to get file system path for container cgroup: %m");
  47
  48         fd = open(fs, O_RDONLY|O_CLOEXEC|O_DIRECTORY);
  49         if (fd < 0)
  50                 return log_error_errno(errno, "Failed to open %s: %m", fs);
  51
  52         FOREACH_STRING(fn,
  53                        ".",
  54                        "tasks",
  55                        "notify_on_release",
  56                        "cgroup.procs",
  57                        "cgroup.events",
  58                        "cgroup.clone_children",
  59                        "cgroup.controllers",
  60                        "cgroup.subtree_control",
  61                        "cgroup.populated")
  62                 if (fchownat(fd, fn, uid_shift, uid_shift, 0) < 0)
  63                         log_full_errno(errno == ENOENT ? LOG_DEBUG :  LOG_WARNING, errno,
  64                                        "Failed to chown() cgroup file %s, ignoring: %m", fn);
  65
  66         return 0;
  67 }
  68
  69 int sync_cgroup(pid_t pid, bool unified_requested) {
  70         _cleanup_free_ char *cgroup = NULL;
  71         char tree[] = "/tmp/unifiedXXXXXX", pid_string[DECIMAL_STR_MAX(pid) + 1];
  72         bool undo_mount = false;
  73         const char *fn;
  74         int unified, r;
  75
  76         unified = cg_unified();
  77         if (unified < 0)
  78                 return log_error_errno(unified, "Failed to determine whether the unified hierachy is used: %m");
  79
  80         if ((unified > 0) == unified_requested)
  81                 return 0;
  82
  83         /* When the host uses the legacy cgroup setup, but the
  84          * container shall use the unified hierarchy, let's make sure
  85          * we copy the path from the name=systemd hierarchy into the
  86          * unified hierarchy. Similar for the reverse situation. */
  87
  88         r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup);
  89         if (r < 0)
  90                 return log_error_errno(r, "Failed to get control group of " PID_FMT ": %m", pid);
  91
  92         /* In order to access the unified hierarchy we need to mount it */
  93         if (!mkdtemp(tree))
  94                 return log_error_errno(errno, "Failed to generate temporary mount point for unified hierarchy: %m");
  95
  96         if (unified)
  97                 r = mount("cgroup", tree, "cgroup", MS_NOSUID|MS_NOEXEC|MS_NODEV, "none,name=systemd,xattr");
  98         else
  99                 r = mount("cgroup", tree, "cgroup", MS_NOSUID|MS_NOEXEC|MS_NODEV, "__DEVEL__sane_behavior");
 100         if (r < 0) {
 101                 r = log_error_errno(errno, "Failed to mount unified hierarchy: %m");
 102                 goto finish;
 103         }
 104
 105         undo_mount = true;
 106
 107         fn = strjoina(tree, cgroup, "/cgroup.procs");
 108         (void) mkdir_parents(fn, 0755);
 109
 110         sprintf(pid_string, PID_FMT, pid);
 111         r = write_string_file(fn, pid_string, 0);
 112         if (r < 0)
 113                 log_error_errno(r, "Failed to move process: %m");
 114
 115 finish:
 116         if (undo_mount)
 117                 (void) umount(tree);
 118
 119         (void) rmdir(tree);
 120         return r;
 121 }
 122
 123 int create_subcgroup(pid_t pid, bool unified_requested) {
 124         _cleanup_free_ char *cgroup = NULL;
 125         const char *child;
 126         int unified, r;
 127         CGroupMask supported;
 128
 129         /* In the unified hierarchy inner nodes may only only contain
 130          * subgroups, but not processes. Hence, if we running in the
 131          * unified hierarchy and the container does the same, and we
 132          * did not create a scope unit for the container move us and
 133          * the container into two separate subcgroups. */
 134
 135         if (!unified_requested)
 136                 return 0;
 137
 138         unified = cg_unified();
 139         if (unified < 0)
 140                 return log_error_errno(unified, "Failed to determine whether the unified hierachy is used: %m");
 141         if (unified == 0)
 142                 return 0;
 143
 144         r = cg_mask_supported(&supported);
 145         if (r < 0)
 146                 return log_error_errno(r, "Failed to determine supported controllers: %m");
 147
 148         r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &cgroup);
 149         if (r < 0)
 150                 return log_error_errno(r, "Failed to get our control group: %m");
 151
 152         child = strjoina(cgroup, "/payload");
 153         r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, child, pid);
 154         if (r < 0)
 155                 return log_error_errno(r, "Failed to create %s subcgroup: %m", child);
 156
 157         child = strjoina(cgroup, "/supervisor");
 158         r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, child, 0);
 159         if (r < 0)
 160                 return log_error_errno(r, "Failed to create %s subcgroup: %m", child);
 161
 162         /* Try to enable as many controllers as possible for the new payload. */
 163         (void) cg_enable_everywhere(supported, supported, cgroup);
 164         return 0;
 165 }