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.clone_children",
  58                        "cgroup.controllers",
  59                        "cgroup.subtree_control",
  60                        "cgroup.populated")
  61                 if (fchownat(fd, fn, uid_shift, uid_shift, 0) < 0)
  62                         log_full_errno(errno == ENOENT ? LOG_DEBUG :  LOG_WARNING, errno,
  63                                        "Failed to chown() cgroup file %s, ignoring: %m", fn);
  64
  65         return 0;
  66 }
  67
  68 int sync_cgroup(pid_t pid, bool unified_requested) {
  69         _cleanup_free_ char *cgroup = NULL;
  70         char tree[] = "/tmp/unifiedXXXXXX", pid_string[DECIMAL_STR_MAX(pid) + 1];
  71         bool undo_mount = false;
  72         const char *fn;
  73         int unified, r;
  74
  75         unified = cg_unified();
  76         if (unified < 0)
  77                 return log_error_errno(unified, "Failed to determine whether the unified hierachy is used: %m");
  78
  79         if ((unified > 0) == unified_requested)
  80                 return 0;
  81
  82         /* When the host uses the legacy cgroup setup, but the
  83          * container shall use the unified hierarchy, let's make sure
  84          * we copy the path from the name=systemd hierarchy into the
  85          * unified hierarchy. Similar for the reverse situation. */
  86
  87         r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup);
  88         if (r < 0)
  89                 return log_error_errno(r, "Failed to get control group of " PID_FMT ": %m", pid);
  90
  91         /* In order to access the unified hierarchy we need to mount it */
  92         if (!mkdtemp(tree))
  93                 return log_error_errno(errno, "Failed to generate temporary mount point for unified hierarchy: %m");
  94
  95         if (unified)
  96                 r = mount("cgroup", tree, "cgroup", MS_NOSUID|MS_NOEXEC|MS_NODEV, "none,name=systemd,xattr");
  97         else
  98                 r = mount("cgroup", tree, "cgroup", MS_NOSUID|MS_NOEXEC|MS_NODEV, "__DEVEL__sane_behavior");
  99         if (r < 0) {
 100                 r = log_error_errno(errno, "Failed to mount unified hierarchy: %m");
 101                 goto finish;
 102         }
 103
 104         undo_mount = true;
 105
 106         fn = strjoina(tree, cgroup, "/cgroup.procs");
 107         (void) mkdir_parents(fn, 0755);
 108
 109         sprintf(pid_string, PID_FMT, pid);
 110         r = write_string_file(fn, pid_string, 0);
 111         if (r < 0)
 112                 log_error_errno(r, "Failed to move process: %m");
 113
 114 finish:
 115         if (undo_mount)
 116                 (void) umount(tree);
 117
 118         (void) rmdir(tree);
 119         return r;
 120 }
 121
 122 int create_subcgroup(pid_t pid, bool unified_requested) {
 123         _cleanup_free_ char *cgroup = NULL;
 124         const char *child;
 125         int unified, r;
 126         CGroupMask supported;
 127
 128         /* In the unified hierarchy inner nodes may only only contain
 129          * subgroups, but not processes. Hence, if we running in the
 130          * unified hierarchy and the container does the same, and we
 131          * did not create a scope unit for the container move us and
 132          * the container into two separate subcgroups. */
 133
 134         if (!unified_requested)
 135                 return 0;
 136
 137         unified = cg_unified();
 138         if (unified < 0)
 139                 return log_error_errno(unified, "Failed to determine whether the unified hierachy is used: %m");
 140         if (unified == 0)
 141                 return 0;
 142
 143         r = cg_mask_supported(&supported);
 144         if (r < 0)
 145                 return log_error_errno(r, "Failed to determine supported controllers: %m");
 146
 147         r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &cgroup);
 148         if (r < 0)
 149                 return log_error_errno(r, "Failed to get our control group: %m");
 150
 151         child = strjoina(cgroup, "/payload");
 152         r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, child, pid);
 153         if (r < 0)
 154                 return log_error_errno(r, "Failed to create %s subcgroup: %m", child);
 155
 156         child = strjoina(cgroup, "/supervisor");
 157         r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, child, 0);
 158         if (r < 0)
 159                 return log_error_errno(r, "Failed to create %s subcgroup: %m", child);
 160
 161         /* Try to enable as many controllers as possible for the new payload. */
 162         (void) cg_enable_everywhere(supported, supported, cgroup);
 163         return 0;
 164 }