src/basic/cgroup-util.c

   1 /***
   2   This file is part of systemd.
   3
   4   Copyright 2010 Lennart Poettering
   5
   6   systemd is free software; you can redistribute it and/or modify it
   7   under the terms of the GNU Lesser General Public License as published by
   8   the Free Software Foundation; either version 2.1 of the License, or
   9   (at your option) any later version.
  10
  11   systemd is distributed in the hope that it will be useful, but
  12   WITHOUT ANY WARRANTY; without even the implied warranty of
  13   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  14   Lesser General Public License for more details.
  15
  16   You should have received a copy of the GNU Lesser General Public License
  17   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  18 ***/
  19
  20 #include <dirent.h>
  21 #include <errno.h>
  22 #include <ftw.h>
  23 #include <limits.h>
  24 #include <signal.h>
  25 #include <stddef.h>
  26 #include <stdlib.h>
  27 #include <string.h>
  28 #include <sys/stat.h>
  29 #include <sys/statfs.h>
  30 #include <sys/types.h>
  31 #include <sys/xattr.h>
  32 #include <unistd.h>
  33
  34 #include "alloc-util.h"
  35 #include "cgroup-util.h"
  36 #include "def.h"
  37 #include "dirent-util.h"
  38 #include "extract-word.h"
  39 #include "fd-util.h"
  40 #include "fileio.h"
  41 #include "format-util.h"
  42 #include "fs-util.h"
  43 #include "log.h"
  44 #include "login-util.h"
  45 #include "macro.h"
  46 #include "missing.h"
  47 #include "mkdir.h"
  48 #include "parse-util.h"
  49 #include "path-util.h"
  50 #include "proc-cmdline.h"
  51 #include "process-util.h"
  52 #include "set.h"
  53 #include "special.h"
  54 #include "stat-util.h"
  55 #include "stdio-util.h"
  56 #include "string-table.h"
  57 #include "string-util.h"
  58 #include "strv.h"
  59 #include "unit-name.h"
  60 #include "user-util.h"
  61
  62 int cg_enumerate_processes(const char *controller, const char *path, FILE **_f) {
  63         _cleanup_free_ char *fs = NULL;
  64         FILE *f;
  65         int r;
  66
  67         assert(_f);
  68
  69         r = cg_get_path(controller, path, "cgroup.procs", &fs);
  70         if (r < 0)
  71                 return r;
  72
  73         f = fopen(fs, "re");
  74         if (!f)
  75                 return -errno;
  76
  77         *_f = f;
  78         return 0;
  79 }
  80
  81 int cg_read_pid(FILE *f, pid_t *_pid) {
  82         unsigned long ul;
  83
  84         /* Note that the cgroup.procs might contain duplicates! See
  85          * cgroups.txt for details. */
  86
  87         assert(f);
  88         assert(_pid);
  89
  90         errno = 0;
  91         if (fscanf(f, "%lu", &ul) != 1) {
  92
  93                 if (feof(f))
  94                         return 0;
  95
  96                 return errno > 0 ? -errno : -EIO;
  97         }
  98
  99         if (ul <= 0)
 100                 return -EIO;
 101
 102         *_pid = (pid_t) ul;
 103         return 1;
 104 }
 105
 106 int cg_read_event(
 107                 const char *controller,
 108                 const char *path,
 109                 const char *event,
 110                 char **val) {
 111
 112         _cleanup_free_ char *events = NULL, *content = NULL;
 113         char *p, *line;
 114         int r;
 115
 116         r = cg_get_path(controller, path, "cgroup.events", &events);
 117         if (r < 0)
 118                 return r;
 119
 120         r = read_full_file(events, &content, NULL);
 121         if (r < 0)
 122                 return r;
 123
 124         p = content;
 125         while ((line = strsep(&p, "\n"))) {
 126                 char *key;
 127
 128                 key = strsep(&line, " ");
 129                 if (!key || !line)
 130                         return -EINVAL;
 131
 132                 if (strcmp(key, event))
 133                         continue;
 134
 135                 *val = strdup(line);
 136                 return 0;
 137         }
 138
 139         return -ENOENT;
 140 }
 141
 142 bool cg_ns_supported(void) {
 143         static thread_local int enabled = -1;
 144
 145         if (enabled >= 0)
 146                 return enabled;
 147
 148         if (access("/proc/self/ns/cgroup", F_OK) == 0)
 149                 enabled = 1;
 150         else
 151                 enabled = 0;
 152
 153         return enabled;
 154 }
 155
 156 int cg_enumerate_subgroups(const char *controller, const char *path, DIR **_d) {
 157         _cleanup_free_ char *fs = NULL;
 158         int r;
 159         DIR *d;
 160
 161         assert(_d);
 162
 163         /* This is not recursive! */
 164
 165         r = cg_get_path(controller, path, NULL, &fs);
 166         if (r < 0)
 167                 return r;
 168
 169         d = opendir(fs);
 170         if (!d)
 171                 return -errno;
 172
 173         *_d = d;
 174         return 0;
 175 }
 176
 177 int cg_read_subgroup(DIR *d, char **fn) {
 178         struct dirent *de;
 179
 180         assert(d);
 181         assert(fn);
 182
 183         FOREACH_DIRENT_ALL(de, d, return -errno) {
 184                 char *b;
 185
 186                 if (de->d_type != DT_DIR)
 187                         continue;
 188
 189                 if (dot_or_dot_dot(de->d_name))
 190                         continue;
 191
 192                 b = strdup(de->d_name);
 193                 if (!b)
 194                         return -ENOMEM;
 195
 196                 *fn = b;
 197                 return 1;
 198         }
 199
 200         return 0;
 201 }
 202
 203 int cg_rmdir(const char *controller, const char *path) {
 204         _cleanup_free_ char *p = NULL;
 205         int r;
 206
 207         r = cg_get_path(controller, path, NULL, &p);
 208         if (r < 0)
 209                 return r;
 210
 211         r = rmdir(p);
 212         if (r < 0 && errno != ENOENT)
 213                 return -errno;
 214
 215         r = cg_hybrid_unified();
 216         if (r < 0)
 217                 return r;
 218         if (r == 0)
 219                 return 0;
 220
 221         if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
 222                 r = cg_rmdir(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path);
 223                 if (r < 0)
 224                         log_warning_errno(r, "Failed to remove compat systemd cgroup %s: %m", path);
 225         }
 226
 227         return 0;
 228 }
 229
 230 int cg_kill(
 231                 const char *controller,
 232                 const char *path,
 233                 int sig,
 234                 CGroupFlags flags,
 235                 Set *s,
 236                 cg_kill_log_func_t log_kill,
 237                 void *userdata) {
 238
 239         _cleanup_set_free_ Set *allocated_set = NULL;
 240         bool done = false;
 241         int r, ret = 0;
 242         pid_t my_pid;
 243
 244         assert(sig >= 0);
 245
 246          /* Don't send SIGCONT twice. Also, SIGKILL always works even when process is suspended, hence don't send
 247           * SIGCONT on SIGKILL. */
 248         if (IN_SET(sig, SIGCONT, SIGKILL))
 249                 flags &= ~CGROUP_SIGCONT;
 250
 251         /* This goes through the tasks list and kills them all. This
 252          * is repeated until no further processes are added to the
 253          * tasks list, to properly handle forking processes */
 254
 255         if (!s) {
 256                 s = allocated_set = set_new(NULL);
 257                 if (!s)
 258                         return -ENOMEM;
 259         }
 260
 261         my_pid = getpid_cached();
 262
 263         do {
 264                 _cleanup_fclose_ FILE *f = NULL;
 265                 pid_t pid = 0;
 266                 done = true;
 267
 268                 r = cg_enumerate_processes(controller, path, &f);
 269                 if (r < 0) {
 270                         if (ret >= 0 && r != -ENOENT)
 271                                 return r;
 272
 273                         return ret;
 274                 }
 275
 276                 while ((r = cg_read_pid(f, &pid)) > 0) {
 277
 278                         if ((flags & CGROUP_IGNORE_SELF) && pid == my_pid)
 279                                 continue;
 280
 281                         if (set_get(s, PID_TO_PTR(pid)) == PID_TO_PTR(pid))
 282                                 continue;
 283
 284                         if (log_kill)
 285                                 log_kill(pid, sig, userdata);
 286
 287                         /* If we haven't killed this process yet, kill
 288                          * it */
 289                         if (kill(pid, sig) < 0) {
 290                                 if (ret >= 0 && errno != ESRCH)
 291                                         ret = -errno;
 292                         } else {
 293                                 if (flags & CGROUP_SIGCONT)
 294                                         (void) kill(pid, SIGCONT);
 295
 296                                 if (ret == 0)
 297                                         ret = 1;
 298                         }
 299
 300                         done = false;
 301
 302                         r = set_put(s, PID_TO_PTR(pid));
 303                         if (r < 0) {
 304                                 if (ret >= 0)
 305                                         return r;
 306
 307                                 return ret;
 308                         }
 309                 }
 310
 311                 if (r < 0) {
 312                         if (ret >= 0)
 313                                 return r;
 314
 315                         return ret;
 316                 }
 317
 318                 /* To avoid racing against processes which fork
 319                  * quicker than we can kill them we repeat this until
 320                  * no new pids need to be killed. */
 321
 322         } while (!done);
 323
 324         return ret;
 325 }
 326
 327 int cg_kill_recursive(
 328                 const char *controller,
 329                 const char *path,
 330                 int sig,
 331                 CGroupFlags flags,
 332                 Set *s,
 333                 cg_kill_log_func_t log_kill,
 334                 void *userdata) {
 335
 336         _cleanup_set_free_ Set *allocated_set = NULL;
 337         _cleanup_closedir_ DIR *d = NULL;
 338         int r, ret;
 339         char *fn;
 340
 341         assert(path);
 342         assert(sig >= 0);
 343
 344         if (!s) {
 345                 s = allocated_set = set_new(NULL);
 346                 if (!s)
 347                         return -ENOMEM;
 348         }
 349
 350         ret = cg_kill(controller, path, sig, flags, s, log_kill, userdata);
 351
 352         r = cg_enumerate_subgroups(controller, path, &d);
 353         if (r < 0) {
 354                 if (ret >= 0 && r != -ENOENT)
 355                         return r;
 356
 357                 return ret;
 358         }
 359
 360         while ((r = cg_read_subgroup(d, &fn)) > 0) {
 361                 _cleanup_free_ char *p = NULL;
 362
 363                 p = strjoin(path, "/", fn);
 364                 free(fn);
 365                 if (!p)
 366                         return -ENOMEM;
 367
 368                 r = cg_kill_recursive(controller, p, sig, flags, s, log_kill, userdata);
 369                 if (r != 0 && ret >= 0)
 370                         ret = r;
 371         }
 372         if (ret >= 0 && r < 0)
 373                 ret = r;
 374
 375         if (flags & CGROUP_REMOVE) {
 376                 r = cg_rmdir(controller, path);
 377                 if (r < 0 && ret >= 0 && r != -ENOENT && r != -EBUSY)
 378                         return r;
 379         }
 380
 381         return ret;
 382 }
 383
 384 int cg_migrate(
 385                 const char *cfrom,
 386                 const char *pfrom,
 387                 const char *cto,
 388                 const char *pto,
 389                 CGroupFlags flags) {
 390
 391         bool done = false;
 392         _cleanup_set_free_ Set *s = NULL;
 393         int r, ret = 0;
 394         pid_t my_pid;
 395
 396         assert(cfrom);
 397         assert(pfrom);
 398         assert(cto);
 399         assert(pto);
 400
 401         s = set_new(NULL);
 402         if (!s)
 403                 return -ENOMEM;
 404
 405         my_pid = getpid_cached();
 406
 407         do {
 408                 _cleanup_fclose_ FILE *f = NULL;
 409                 pid_t pid = 0;
 410                 done = true;
 411
 412                 r = cg_enumerate_processes(cfrom, pfrom, &f);
 413                 if (r < 0) {
 414                         if (ret >= 0 && r != -ENOENT)
 415                                 return r;
 416
 417                         return ret;
 418                 }
 419
 420                 while ((r = cg_read_pid(f, &pid)) > 0) {
 421
 422                         /* This might do weird stuff if we aren't a
 423                          * single-threaded program. However, we
 424                          * luckily know we are not */
 425                         if ((flags & CGROUP_IGNORE_SELF) && pid == my_pid)
 426                                 continue;
 427
 428                         if (set_get(s, PID_TO_PTR(pid)) == PID_TO_PTR(pid))
 429                                 continue;
 430
 431                         /* Ignore kernel threads. Since they can only
 432                          * exist in the root cgroup, we only check for
 433                          * them there. */
 434                         if (cfrom &&
 435                             (isempty(pfrom) || path_equal(pfrom, "/")) &&
 436                             is_kernel_thread(pid) > 0)
 437                                 continue;
 438
 439                         r = cg_attach(cto, pto, pid);
 440                         if (r < 0) {
 441                                 if (ret >= 0 && r != -ESRCH)
 442                                         ret = r;
 443                         } else if (ret == 0)
 444                                 ret = 1;
 445
 446                         done = false;
 447
 448                         r = set_put(s, PID_TO_PTR(pid));
 449                         if (r < 0) {
 450                                 if (ret >= 0)
 451                                         return r;
 452
 453                                 return ret;
 454                         }
 455                 }
 456
 457                 if (r < 0) {
 458                         if (ret >= 0)
 459                                 return r;
 460
 461                         return ret;
 462                 }
 463         } while (!done);
 464
 465         return ret;
 466 }
 467
 468 int cg_migrate_recursive(
 469                 const char *cfrom,
 470                 const char *pfrom,
 471                 const char *cto,
 472                 const char *pto,
 473                 CGroupFlags flags) {
 474
 475         _cleanup_closedir_ DIR *d = NULL;
 476         int r, ret = 0;
 477         char *fn;
 478
 479         assert(cfrom);
 480         assert(pfrom);
 481         assert(cto);
 482         assert(pto);
 483
 484         ret = cg_migrate(cfrom, pfrom, cto, pto, flags);
 485
 486         r = cg_enumerate_subgroups(cfrom, pfrom, &d);
 487         if (r < 0) {
 488                 if (ret >= 0 && r != -ENOENT)
 489                         return r;
 490
 491                 return ret;
 492         }
 493
 494         while ((r = cg_read_subgroup(d, &fn)) > 0) {
 495                 _cleanup_free_ char *p = NULL;
 496
 497                 p = strjoin(pfrom, "/", fn);
 498                 free(fn);
 499                 if (!p)
 500                         return -ENOMEM;
 501
 502                 r = cg_migrate_recursive(cfrom, p, cto, pto, flags);
 503                 if (r != 0 && ret >= 0)
 504                         ret = r;
 505         }
 506
 507         if (r < 0 && ret >= 0)
 508                 ret = r;
 509
 510         if (flags & CGROUP_REMOVE) {
 511                 r = cg_rmdir(cfrom, pfrom);
 512                 if (r < 0 && ret >= 0 && r != -ENOENT && r != -EBUSY)
 513                         return r;
 514         }
 515
 516         return ret;
 517 }
 518
 519 int cg_migrate_recursive_fallback(
 520                 const char *cfrom,
 521                 const char *pfrom,
 522                 const char *cto,
 523                 const char *pto,
 524                 CGroupFlags flags) {
 525
 526         int r;
 527
 528         assert(cfrom);
 529         assert(pfrom);
 530         assert(cto);
 531         assert(pto);
 532
 533         r = cg_migrate_recursive(cfrom, pfrom, cto, pto, flags);
 534         if (r < 0) {
 535                 char prefix[strlen(pto) + 1];
 536
 537                 /* This didn't work? Then let's try all prefixes of the destination */
 538
 539                 PATH_FOREACH_PREFIX(prefix, pto) {
 540                         int q;
 541
 542                         q = cg_migrate_recursive(cfrom, pfrom, cto, prefix, flags);
 543                         if (q >= 0)
 544                                 return q;
 545                 }
 546         }
 547
 548         return r;
 549 }
 550
 551 static const char *controller_to_dirname(const char *controller) {
 552         const char *e;
 553
 554         assert(controller);
 555
 556         /* Converts a controller name to the directory name below
 557          * /sys/fs/cgroup/ we want to mount it to. Effectively, this
 558          * just cuts off the name= prefixed used for named
 559          * hierarchies, if it is specified. */
 560
 561         if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
 562                 if (cg_hybrid_unified() > 0)
 563                         controller = SYSTEMD_CGROUP_CONTROLLER_HYBRID;
 564                 else
 565                         controller = SYSTEMD_CGROUP_CONTROLLER_LEGACY;
 566         }
 567
 568         e = startswith(controller, "name=");
 569         if (e)
 570                 return e;
 571
 572         return controller;
 573 }
 574
 575 static int join_path_legacy(const char *controller, const char *path, const char *suffix, char **fs) {
 576         const char *dn;
 577         char *t = NULL;
 578
 579         assert(fs);
 580         assert(controller);
 581
 582         dn = controller_to_dirname(controller);
 583
 584         if (isempty(path) && isempty(suffix))
 585                 t = strappend("/sys/fs/cgroup/", dn);
 586         else if (isempty(path))
 587                 t = strjoin("/sys/fs/cgroup/", dn, "/", suffix);
 588         else if (isempty(suffix))
 589                 t = strjoin("/sys/fs/cgroup/", dn, "/", path);
 590         else
 591                 t = strjoin("/sys/fs/cgroup/", dn, "/", path, "/", suffix);
 592         if (!t)
 593                 return -ENOMEM;
 594
 595         *fs = t;
 596         return 0;
 597 }
 598
 599 static int join_path_unified(const char *path, const char *suffix, char **fs) {
 600         char *t;
 601
 602         assert(fs);
 603
 604         if (isempty(path) && isempty(suffix))
 605                 t = strdup("/sys/fs/cgroup");
 606         else if (isempty(path))
 607                 t = strappend("/sys/fs/cgroup/", suffix);
 608         else if (isempty(suffix))
 609                 t = strappend("/sys/fs/cgroup/", path);
 610         else
 611                 t = strjoin("/sys/fs/cgroup/", path, "/", suffix);
 612         if (!t)
 613                 return -ENOMEM;
 614
 615         *fs = t;
 616         return 0;
 617 }
 618
 619 int cg_get_path(const char *controller, const char *path, const char *suffix, char **fs) {
 620         int r;
 621
 622         assert(fs);
 623
 624         if (!controller) {
 625                 char *t;
 626
 627                 /* If no controller is specified, we return the path
 628                  * *below* the controllers, without any prefix. */
 629
 630                 if (!path && !suffix)
 631                         return -EINVAL;
 632
 633                 if (!suffix)
 634                         t = strdup(path);
 635                 else if (!path)
 636                         t = strdup(suffix);
 637                 else
 638                         t = strjoin(path, "/", suffix);
 639                 if (!t)
 640                         return -ENOMEM;
 641
 642                 *fs = path_kill_slashes(t);
 643                 return 0;
 644         }
 645
 646         if (!cg_controller_is_valid(controller))
 647                 return -EINVAL;
 648
 649         r = cg_all_unified();
 650         if (r < 0)
 651                 return r;
 652         if (r > 0)
 653                 r = join_path_unified(path, suffix, fs);
 654         else
 655                 r = join_path_legacy(controller, path, suffix, fs);
 656         if (r < 0)
 657                 return r;
 658
 659         path_kill_slashes(*fs);
 660         return 0;
 661 }
 662
 663 static int controller_is_accessible(const char *controller) {
 664         int r;
 665
 666         assert(controller);
 667
 668         /* Checks whether a specific controller is accessible,
 669          * i.e. its hierarchy mounted. In the unified hierarchy all
 670          * controllers are considered accessible, except for the named
 671          * hierarchies */
 672
 673         if (!cg_controller_is_valid(controller))
 674                 return -EINVAL;
 675
 676         r = cg_all_unified();
 677         if (r < 0)
 678                 return r;
 679         if (r > 0) {
 680                 /* We don't support named hierarchies if we are using
 681                  * the unified hierarchy. */
 682
 683                 if (streq(controller, SYSTEMD_CGROUP_CONTROLLER))
 684                         return 0;
 685
 686                 if (startswith(controller, "name="))
 687                         return -EOPNOTSUPP;
 688
 689         } else {
 690                 const char *cc, *dn;
 691
 692                 dn = controller_to_dirname(controller);
 693                 cc = strjoina("/sys/fs/cgroup/", dn);
 694
 695                 if (laccess(cc, F_OK) < 0)
 696                         return -errno;
 697         }
 698
 699         return 0;
 700 }
 701
 702 int cg_get_path_and_check(const char *controller, const char *path, const char *suffix, char **fs) {
 703         int r;
 704
 705         assert(controller);
 706         assert(fs);
 707
 708         /* Check if the specified controller is actually accessible */
 709         r = controller_is_accessible(controller);
 710         if (r < 0)
 711                 return r;
 712
 713         return cg_get_path(controller, path, suffix, fs);
 714 }
 715
 716 static int trim_cb(const char *path, const struct stat *sb, int typeflag, struct FTW *ftwbuf) {
 717         assert(path);
 718         assert(sb);
 719         assert(ftwbuf);
 720
 721         if (typeflag != FTW_DP)
 722                 return 0;
 723
 724         if (ftwbuf->level < 1)
 725                 return 0;
 726
 727         (void) rmdir(path);
 728         return 0;
 729 }
 730
 731 int cg_trim(const char *controller, const char *path, bool delete_root) {
 732         _cleanup_free_ char *fs = NULL;
 733         int r = 0, q;
 734
 735         assert(path);
 736
 737         r = cg_get_path(controller, path, NULL, &fs);
 738         if (r < 0)
 739                 return r;
 740
 741         errno = 0;
 742         if (nftw(fs, trim_cb, 64, FTW_DEPTH|FTW_MOUNT|FTW_PHYS) != 0) {
 743                 if (errno == ENOENT)
 744                         r = 0;
 745                 else if (errno > 0)
 746                         r = -errno;
 747                 else
 748                         r = -EIO;
 749         }
 750
 751         if (delete_root) {
 752                 if (rmdir(fs) < 0 && errno != ENOENT)
 753                         return -errno;
 754         }
 755
 756         q = cg_hybrid_unified();
 757         if (q < 0)
 758                 return q;
 759         if (q > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
 760                 q = cg_trim(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, delete_root);
 761                 if (q < 0)
 762                         log_warning_errno(q, "Failed to trim compat systemd cgroup %s: %m", path);
 763         }
 764
 765         return r;
 766 }
 767
 768 int cg_create(const char *controller, const char *path) {
 769         _cleanup_free_ char *fs = NULL;
 770         int r;
 771
 772         r = cg_get_path_and_check(controller, path, NULL, &fs);
 773         if (r < 0)
 774                 return r;
 775
 776         r = mkdir_parents(fs, 0755);
 777         if (r < 0)
 778                 return r;
 779
 780         if (mkdir(fs, 0755) < 0) {
 781
 782                 if (errno == EEXIST)
 783                         return 0;
 784
 785                 return -errno;
 786         }
 787
 788         r = cg_hybrid_unified();
 789         if (r < 0)
 790                 return r;
 791
 792         if (r > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
 793                 r = cg_create(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path);
 794                 if (r < 0)
 795                         log_warning_errno(r, "Failed to create compat systemd cgroup %s: %m", path);
 796         }
 797
 798         return 1;
 799 }
 800
 801 int cg_create_and_attach(const char *controller, const char *path, pid_t pid) {
 802         int r, q;
 803
 804         assert(pid >= 0);
 805
 806         r = cg_create(controller, path);
 807         if (r < 0)
 808                 return r;
 809
 810         q = cg_attach(controller, path, pid);
 811         if (q < 0)
 812                 return q;
 813
 814         /* This does not remove the cgroup on failure */
 815         return r;
 816 }
 817
 818 int cg_attach(const char *controller, const char *path, pid_t pid) {
 819         _cleanup_free_ char *fs = NULL;
 820         char c[DECIMAL_STR_MAX(pid_t) + 2];
 821         int r;
 822
 823         assert(path);
 824         assert(pid >= 0);
 825
 826         r = cg_get_path_and_check(controller, path, "cgroup.procs", &fs);
 827         if (r < 0)
 828                 return r;
 829
 830         if (pid == 0)
 831                 pid = getpid_cached();
 832
 833         xsprintf(c, PID_FMT "\n", pid);
 834
 835         r = write_string_file(fs, c, 0);
 836         if (r < 0)
 837                 return r;
 838
 839         r = cg_hybrid_unified();
 840         if (r < 0)
 841                 return r;
 842
 843         if (r > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
 844                 r = cg_attach(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, pid);
 845                 if (r < 0)
 846                         log_warning_errno(r, "Failed to attach "PID_FMT" to compat systemd cgroup %s: %m", pid, path);
 847         }
 848
 849         return 0;
 850 }
 851
 852 int cg_attach_fallback(const char *controller, const char *path, pid_t pid) {
 853         int r;
 854
 855         assert(controller);
 856         assert(path);
 857         assert(pid >= 0);
 858
 859         r = cg_attach(controller, path, pid);
 860         if (r < 0) {
 861                 char prefix[strlen(path) + 1];
 862
 863                 /* This didn't work? Then let's try all prefixes of
 864                  * the destination */
 865
 866                 PATH_FOREACH_PREFIX(prefix, path) {
 867                         int q;
 868
 869                         q = cg_attach(controller, prefix, pid);
 870                         if (q >= 0)
 871                                 return q;
 872                 }
 873         }
 874
 875         return r;
 876 }
 877
 878 int cg_set_group_access(
 879                 const char *controller,
 880                 const char *path,
 881                 mode_t mode,
 882                 uid_t uid,
 883                 gid_t gid) {
 884
 885         _cleanup_free_ char *fs = NULL;
 886         int r;
 887
 888         if (mode == MODE_INVALID && uid == UID_INVALID && gid == GID_INVALID)
 889                 return 0;
 890
 891         if (mode != MODE_INVALID)
 892                 mode &= 0777;
 893
 894         r = cg_get_path(controller, path, NULL, &fs);
 895         if (r < 0)
 896                 return r;
 897
 898         r = chmod_and_chown(fs, mode, uid, gid);
 899         if (r < 0)
 900                 return r;
 901
 902         r = cg_hybrid_unified();
 903         if (r < 0)
 904                 return r;
 905         if (r > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
 906                 r = cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, mode, uid, gid);
 907                 if (r < 0)
 908                         log_warning_errno(r, "Failed to set group access on compat systemd cgroup %s: %m", path);
 909         }
 910
 911         return 0;
 912 }
 913
 914 int cg_set_task_access(
 915                 const char *controller,
 916                 const char *path,
 917                 mode_t mode,
 918                 uid_t uid,
 919                 gid_t gid) {
 920
 921         _cleanup_free_ char *fs = NULL, *procs = NULL;
 922         int r;
 923
 924         assert(path);
 925
 926         if (mode == MODE_INVALID && uid == UID_INVALID && gid == GID_INVALID)
 927                 return 0;
 928
 929         if (mode != MODE_INVALID)
 930                 mode &= 0666;
 931
 932         r = cg_get_path(controller, path, "cgroup.procs", &fs);
 933         if (r < 0)
 934                 return r;
 935
 936         r = chmod_and_chown(fs, mode, uid, gid);
 937         if (r < 0)
 938                 return r;
 939
 940         r = cg_unified_controller(controller);
 941         if (r < 0)
 942                 return r;
 943         if (r == 0) {
 944                 /* Compatibility, Always keep values for "tasks" in sync with
 945                  * "cgroup.procs" */
 946                 if (cg_get_path(controller, path, "tasks", &procs) >= 0)
 947                         (void) chmod_and_chown(procs, mode, uid, gid);
 948         }
 949
 950         r = cg_hybrid_unified();
 951         if (r < 0)
 952                 return r;
 953         if (r > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
 954                 r = cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, mode, uid, gid);
 955                 if (r < 0)
 956                         log_warning_errno(r, "Failed to set task access on compat systemd cgroup %s: %m", path);
 957         }
 958
 959         return 0;
 960 }
 961
 962 int cg_set_xattr(const char *controller, const char *path, const char *name, const void *value, size_t size, int flags) {
 963         _cleanup_free_ char *fs = NULL;
 964         int r;
 965
 966         assert(path);
 967         assert(name);
 968         assert(value || size <= 0);
 969
 970         r = cg_get_path(controller, path, NULL, &fs);
 971         if (r < 0)
 972                 return r;
 973
 974         if (setxattr(fs, name, value, size, flags) < 0)
 975                 return -errno;
 976
 977         return 0;
 978 }
 979
 980 int cg_get_xattr(const char *controller, const char *path, const char *name, void *value, size_t size) {
 981         _cleanup_free_ char *fs = NULL;
 982         ssize_t n;
 983         int r;
 984
 985         assert(path);
 986         assert(name);
 987
 988         r = cg_get_path(controller, path, NULL, &fs);
 989         if (r < 0)
 990                 return r;
 991
 992         n = getxattr(fs, name, value, size);
 993         if (n < 0)
 994                 return -errno;
 995
 996         return (int) n;
 997 }
 998
 999 int cg_pid_get_path(const char *controller, pid_t pid, char **path) {
1000         _cleanup_fclose_ FILE *f = NULL;
1001         char line[LINE_MAX];
1002         const char *fs, *controller_str;
1003         size_t cs = 0;
1004         int unified;
1005
1006         assert(path);
1007         assert(pid >= 0);
1008
1009         if (controller) {
1010                 if (!cg_controller_is_valid(controller))
1011                         return -EINVAL;
1012         } else
1013                 controller = SYSTEMD_CGROUP_CONTROLLER;
1014
1015         unified = cg_unified_controller(controller);
1016         if (unified < 0)
1017                 return unified;
1018         if (unified == 0) {
1019                 if (streq(controller, SYSTEMD_CGROUP_CONTROLLER))
1020                         controller_str = SYSTEMD_CGROUP_CONTROLLER_LEGACY;
1021                 else
1022                         controller_str = controller;
1023
1024                 cs = strlen(controller_str);
1025         }
1026
1027         fs = procfs_file_alloca(pid, "cgroup");
1028         f = fopen(fs, "re");
1029         if (!f)
1030                 return errno == ENOENT ? -ESRCH : -errno;
1031
1032         FOREACH_LINE(line, f, return -errno) {
1033                 char *e, *p;
1034
1035                 truncate_nl(line);
1036
1037                 if (unified) {
1038                         e = startswith(line, "0:");
1039                         if (!e)
1040                                 continue;
1041
1042                         e = strchr(e, ':');
1043                         if (!e)
1044                                 continue;
1045                 } else {
1046                         char *l;
1047                         size_t k;
1048                         const char *word, *state;
1049                         bool found = false;
1050
1051                         l = strchr(line, ':');
1052                         if (!l)
1053                                 continue;
1054
1055                         l++;
1056                         e = strchr(l, ':');
1057                         if (!e)
1058                                 continue;
1059
1060                         *e = 0;
1061                         FOREACH_WORD_SEPARATOR(word, k, l, ",", state) {
1062                                 if (k == cs && memcmp(word, controller_str, cs) == 0) {
1063                                         found = true;
1064                                         break;
1065                                 }
1066                         }
1067
1068                         if (!found)
1069                                 continue;
1070                 }
1071
1072                 p = strdup(e + 1);
1073                 if (!p)
1074                         return -ENOMEM;
1075
1076                 *path = p;
1077                 return 0;
1078         }
1079
1080         return -ENODATA;
1081 }
1082
1083 int cg_install_release_agent(const char *controller, const char *agent) {
1084         _cleanup_free_ char *fs = NULL, *contents = NULL;
1085         const char *sc;
1086         int r;
1087
1088         assert(agent);
1089
1090         r = cg_unified_controller(controller);
1091         if (r < 0)
1092                 return r;
1093         if (r > 0) /* doesn't apply to unified hierarchy */
1094                 return -EOPNOTSUPP;
1095
1096         r = cg_get_path(controller, NULL, "release_agent", &fs);
1097         if (r < 0)
1098                 return r;
1099
1100         r = read_one_line_file(fs, &contents);
1101         if (r < 0)
1102                 return r;
1103
1104         sc = strstrip(contents);
1105         if (isempty(sc)) {
1106                 r = write_string_file(fs, agent, 0);
1107                 if (r < 0)
1108                         return r;
1109         } else if (!path_equal(sc, agent))
1110                 return -EEXIST;
1111
1112         fs = mfree(fs);
1113         r = cg_get_path(controller, NULL, "notify_on_release", &fs);
1114         if (r < 0)
1115                 return r;
1116
1117         contents = mfree(contents);
1118         r = read_one_line_file(fs, &contents);
1119         if (r < 0)
1120                 return r;
1121
1122         sc = strstrip(contents);
1123         if (streq(sc, "0")) {
1124                 r = write_string_file(fs, "1", 0);
1125                 if (r < 0)
1126                         return r;
1127
1128                 return 1;
1129         }
1130
1131         if (!streq(sc, "1"))
1132                 return -EIO;
1133
1134         return 0;
1135 }
1136
1137 int cg_uninstall_release_agent(const char *controller) {
1138         _cleanup_free_ char *fs = NULL;
1139         int r;
1140
1141         r = cg_unified_controller(controller);
1142         if (r < 0)
1143                 return r;
1144         if (r > 0) /* Doesn't apply to unified hierarchy */
1145                 return -EOPNOTSUPP;
1146
1147         r = cg_get_path(controller, NULL, "notify_on_release", &fs);
1148         if (r < 0)
1149                 return r;
1150
1151         r = write_string_file(fs, "0", 0);
1152         if (r < 0)
1153                 return r;
1154
1155         fs = mfree(fs);
1156
1157         r = cg_get_path(controller, NULL, "release_agent", &fs);
1158         if (r < 0)
1159                 return r;
1160
1161         r = write_string_file(fs, "", 0);
1162         if (r < 0)
1163                 return r;
1164
1165         return 0;
1166 }
1167
1168 int cg_is_empty(const char *controller, const char *path) {
1169         _cleanup_fclose_ FILE *f = NULL;
1170         pid_t pid;
1171         int r;
1172
1173         assert(path);
1174
1175         r = cg_enumerate_processes(controller, path, &f);
1176         if (r == -ENOENT)
1177                 return 1;
1178         if (r < 0)
1179                 return r;
1180
1181         r = cg_read_pid(f, &pid);
1182         if (r < 0)
1183                 return r;
1184
1185         return r == 0;
1186 }
1187
1188 int cg_is_empty_recursive(const char *controller, const char *path) {
1189         int r;
1190
1191         assert(path);
1192
1193         /* The root cgroup is always populated */
1194         if (controller && (isempty(path) || path_equal(path, "/")))
1195                 return false;
1196
1197         r = cg_unified_controller(controller);
1198         if (r < 0)
1199                 return r;
1200         if (r > 0) {
1201                 _cleanup_free_ char *t = NULL;
1202
1203                 /* On the unified hierarchy we can check empty state
1204                  * via the "populated" attribute of "cgroup.events". */
1205
1206                 r = cg_read_event(controller, path, "populated", &t);
1207                 if (r < 0)
1208                         return r;
1209
1210                 return streq(t, "0");
1211         } else {
1212                 _cleanup_closedir_ DIR *d = NULL;
1213                 char *fn;
1214
1215                 r = cg_is_empty(controller, path);
1216                 if (r <= 0)
1217                         return r;
1218
1219                 r = cg_enumerate_subgroups(controller, path, &d);
1220                 if (r == -ENOENT)
1221                         return 1;
1222                 if (r < 0)
1223                         return r;
1224
1225                 while ((r = cg_read_subgroup(d, &fn)) > 0) {
1226                         _cleanup_free_ char *p = NULL;
1227
1228                         p = strjoin(path, "/", fn);
1229                         free(fn);
1230                         if (!p)
1231                                 return -ENOMEM;
1232
1233                         r = cg_is_empty_recursive(controller, p);
1234                         if (r <= 0)
1235                                 return r;
1236                 }
1237                 if (r < 0)
1238                         return r;
1239
1240                 return true;
1241         }
1242 }
1243
1244 int cg_split_spec(const char *spec, char **controller, char **path) {
1245         char *t = NULL, *u = NULL;
1246         const char *e;
1247
1248         assert(spec);
1249
1250         if (*spec == '/') {
1251                 if (!path_is_safe(spec))
1252                         return -EINVAL;
1253
1254                 if (path) {
1255                         t = strdup(spec);
1256                         if (!t)
1257                                 return -ENOMEM;
1258
1259                         *path = path_kill_slashes(t);
1260                 }
1261
1262                 if (controller)
1263                         *controller = NULL;
1264
1265                 return 0;
1266         }
1267
1268         e = strchr(spec, ':');
1269         if (!e) {
1270                 if (!cg_controller_is_valid(spec))
1271                         return -EINVAL;
1272
1273                 if (controller) {
1274                         t = strdup(spec);
1275                         if (!t)
1276                                 return -ENOMEM;
1277
1278                         *controller = t;
1279                 }
1280
1281                 if (path)
1282                         *path = NULL;
1283
1284                 return 0;
1285         }
1286
1287         t = strndup(spec, e-spec);
1288         if (!t)
1289                 return -ENOMEM;
1290         if (!cg_controller_is_valid(t)) {
1291                 free(t);
1292                 return -EINVAL;
1293         }
1294
1295         if (isempty(e+1))
1296                 u = NULL;
1297         else {
1298                 u = strdup(e+1);
1299                 if (!u) {
1300                         free(t);
1301                         return -ENOMEM;
1302                 }
1303
1304                 if (!path_is_safe(u) ||
1305                     !path_is_absolute(u)) {
1306                         free(t);
1307                         free(u);
1308                         return -EINVAL;
1309                 }
1310
1311                 path_kill_slashes(u);
1312         }
1313
1314         if (controller)
1315                 *controller = t;
1316         else
1317                 free(t);
1318
1319         if (path)
1320                 *path = u;
1321         else
1322                 free(u);
1323
1324         return 0;
1325 }
1326
1327 int cg_mangle_path(const char *path, char **result) {
1328         _cleanup_free_ char *c = NULL, *p = NULL;
1329         char *t;
1330         int r;
1331
1332         assert(path);
1333         assert(result);
1334
1335         /* First, check if it already is a filesystem path */
1336         if (path_startswith(path, "/sys/fs/cgroup")) {
1337
1338                 t = strdup(path);
1339                 if (!t)
1340                         return -ENOMEM;
1341
1342                 *result = path_kill_slashes(t);
1343                 return 0;
1344         }
1345
1346         /* Otherwise, treat it as cg spec */
1347         r = cg_split_spec(path, &c, &p);
1348         if (r < 0)
1349                 return r;
1350
1351         return cg_get_path(c ?: SYSTEMD_CGROUP_CONTROLLER, p ?: "/", NULL, result);
1352 }
1353
1354 int cg_get_root_path(char **path) {
1355         char *p, *e;
1356         int r;
1357
1358         assert(path);
1359
1360         r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 1, &p);
1361         if (r < 0)
1362                 return r;
1363
1364         e = endswith(p, "/" SPECIAL_INIT_SCOPE);
1365         if (!e)
1366                 e = endswith(p, "/" SPECIAL_SYSTEM_SLICE); /* legacy */
1367         if (!e)
1368                 e = endswith(p, "/system"); /* even more legacy */
1369         if (e)
1370                 *e = 0;
1371
1372         *path = p;
1373         return 0;
1374 }
1375
1376 int cg_shift_path(const char *cgroup, const char *root, const char **shifted) {
1377         _cleanup_free_ char *rt = NULL;
1378         char *p;
1379         int r;
1380
1381         assert(cgroup);
1382         assert(shifted);
1383
1384         if (!root) {
1385                 /* If the root was specified let's use that, otherwise
1386                  * let's determine it from PID 1 */
1387
1388                 r = cg_get_root_path(&rt);
1389                 if (r < 0)
1390                         return r;
1391
1392                 root = rt;
1393         }
1394
1395         p = path_startswith(cgroup, root);
1396         if (p && p > cgroup)
1397                 *shifted = p - 1;
1398         else
1399                 *shifted = cgroup;
1400
1401         return 0;
1402 }
1403
1404 int cg_pid_get_path_shifted(pid_t pid, const char *root, char **cgroup) {
1405         _cleanup_free_ char *raw = NULL;
1406         const char *c;
1407         int r;
1408
1409         assert(pid >= 0);
1410         assert(cgroup);
1411
1412         r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &raw);
1413         if (r < 0)
1414                 return r;
1415
1416         r = cg_shift_path(raw, root, &c);
1417         if (r < 0)
1418                 return r;
1419
1420         if (c == raw) {
1421                 *cgroup = raw;
1422                 raw = NULL;
1423         } else {
1424                 char *n;
1425
1426                 n = strdup(c);
1427                 if (!n)
1428                         return -ENOMEM;
1429
1430                 *cgroup = n;
1431         }
1432
1433         return 0;
1434 }
1435
1436 int cg_path_decode_unit(const char *cgroup, char **unit) {
1437         char *c, *s;
1438         size_t n;
1439
1440         assert(cgroup);
1441         assert(unit);
1442
1443         n = strcspn(cgroup, "/");
1444         if (n < 3)
1445                 return -ENXIO;
1446
1447         c = strndupa(cgroup, n);
1448         c = cg_unescape(c);
1449
1450         if (!unit_name_is_valid(c, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE))
1451                 return -ENXIO;
1452
1453         s = strdup(c);
1454         if (!s)
1455                 return -ENOMEM;
1456
1457         *unit = s;
1458         return 0;
1459 }
1460
1461 static bool valid_slice_name(const char *p, size_t n) {
1462
1463         if (!p)
1464                 return false;
1465
1466         if (n < strlen("x.slice"))
1467                 return false;
1468
1469         if (memcmp(p + n - 6, ".slice", 6) == 0) {
1470                 char buf[n+1], *c;
1471
1472                 memcpy(buf, p, n);
1473                 buf[n] = 0;
1474
1475                 c = cg_unescape(buf);
1476
1477                 return unit_name_is_valid(c, UNIT_NAME_PLAIN);
1478         }
1479
1480         return false;
1481 }
1482
1483 static const char *skip_slices(const char *p) {
1484         assert(p);
1485
1486         /* Skips over all slice assignments */
1487
1488         for (;;) {
1489                 size_t n;
1490
1491                 p += strspn(p, "/");
1492
1493                 n = strcspn(p, "/");
1494                 if (!valid_slice_name(p, n))
1495                         return p;
1496
1497                 p += n;
1498         }
1499 }
1500
1501 int cg_path_get_unit(const char *path, char **ret) {
1502         const char *e;
1503         char *unit;
1504         int r;
1505
1506         assert(path);
1507         assert(ret);
1508
1509         e = skip_slices(path);
1510
1511         r = cg_path_decode_unit(e, &unit);
1512         if (r < 0)
1513                 return r;
1514
1515         /* We skipped over the slices, don't accept any now */
1516         if (endswith(unit, ".slice")) {
1517                 free(unit);
1518                 return -ENXIO;
1519         }
1520
1521         *ret = unit;
1522         return 0;
1523 }
1524
1525 int cg_pid_get_unit(pid_t pid, char **unit) {
1526         _cleanup_free_ char *cgroup = NULL;
1527         int r;
1528
1529         assert(unit);
1530
1531         r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1532         if (r < 0)
1533                 return r;
1534
1535         return cg_path_get_unit(cgroup, unit);
1536 }
1537
1538 /**
1539  * Skip session-*.scope, but require it to be there.
1540  */
1541 static const char *skip_session(const char *p) {
1542         size_t n;
1543
1544         if (isempty(p))
1545                 return NULL;
1546
1547         p += strspn(p, "/");
1548
1549         n = strcspn(p, "/");
1550         if (n < strlen("session-x.scope"))
1551                 return NULL;
1552
1553         if (memcmp(p, "session-", 8) == 0 && memcmp(p + n - 6, ".scope", 6) == 0) {
1554                 char buf[n - 8 - 6 + 1];
1555
1556                 memcpy(buf, p + 8, n - 8 - 6);
1557                 buf[n - 8 - 6] = 0;
1558
1559                 /* Note that session scopes never need unescaping,
1560                  * since they cannot conflict with the kernel's own
1561                  * names, hence we don't need to call cg_unescape()
1562                  * here. */
1563
1564                 if (!session_id_valid(buf))
1565                         return false;
1566
1567                 p += n;
1568                 p += strspn(p, "/");
1569                 return p;
1570         }
1571
1572         return NULL;
1573 }
1574
1575 /**
1576  * Skip user@*.service, but require it to be there.
1577  */
1578 static const char *skip_user_manager(const char *p) {
1579         size_t n;
1580
1581         if (isempty(p))
1582                 return NULL;
1583
1584         p += strspn(p, "/");
1585
1586         n = strcspn(p, "/");
1587         if (n < strlen("user@x.service"))
1588                 return NULL;
1589
1590         if (memcmp(p, "user@", 5) == 0 && memcmp(p + n - 8, ".service", 8) == 0) {
1591                 char buf[n - 5 - 8 + 1];
1592
1593                 memcpy(buf, p + 5, n - 5 - 8);
1594                 buf[n - 5 - 8] = 0;
1595
1596                 /* Note that user manager services never need unescaping,
1597                  * since they cannot conflict with the kernel's own
1598                  * names, hence we don't need to call cg_unescape()
1599                  * here. */
1600
1601                 if (parse_uid(buf, NULL) < 0)
1602                         return NULL;
1603
1604                 p += n;
1605                 p += strspn(p, "/");
1606
1607                 return p;
1608         }
1609
1610         return NULL;
1611 }
1612
1613 static const char *skip_user_prefix(const char *path) {
1614         const char *e, *t;
1615
1616         assert(path);
1617
1618         /* Skip slices, if there are any */
1619         e = skip_slices(path);
1620
1621         /* Skip the user manager, if it's in the path now... */
1622         t = skip_user_manager(e);
1623         if (t)
1624                 return t;
1625
1626         /* Alternatively skip the user session if it is in the path... */
1627         return skip_session(e);
1628 }
1629
1630 int cg_path_get_user_unit(const char *path, char **ret) {
1631         const char *t;
1632
1633         assert(path);
1634         assert(ret);
1635
1636         t = skip_user_prefix(path);
1637         if (!t)
1638                 return -ENXIO;
1639
1640         /* And from here on it looks pretty much the same as for a
1641          * system unit, hence let's use the same parser from here
1642          * on. */
1643         return cg_path_get_unit(t, ret);
1644 }
1645
1646 int cg_pid_get_user_unit(pid_t pid, char **unit) {
1647         _cleanup_free_ char *cgroup = NULL;
1648         int r;
1649
1650         assert(unit);
1651
1652         r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1653         if (r < 0)
1654                 return r;
1655
1656         return cg_path_get_user_unit(cgroup, unit);
1657 }
1658
1659 int cg_path_get_machine_name(const char *path, char **machine) {
1660         _cleanup_free_ char *u = NULL;
1661         const char *sl;
1662         int r;
1663
1664         r = cg_path_get_unit(path, &u);
1665         if (r < 0)
1666                 return r;
1667
1668         sl = strjoina("/run/systemd/machines/unit:", u);
1669         return readlink_malloc(sl, machine);
1670 }
1671
1672 int cg_pid_get_machine_name(pid_t pid, char **machine) {
1673         _cleanup_free_ char *cgroup = NULL;
1674         int r;
1675
1676         assert(machine);
1677
1678         r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1679         if (r < 0)
1680                 return r;
1681
1682         return cg_path_get_machine_name(cgroup, machine);
1683 }
1684
1685 int cg_path_get_session(const char *path, char **session) {
1686         _cleanup_free_ char *unit = NULL;
1687         char *start, *end;
1688         int r;
1689
1690         assert(path);
1691
1692         r = cg_path_get_unit(path, &unit);
1693         if (r < 0)
1694                 return r;
1695
1696         start = startswith(unit, "session-");
1697         if (!start)
1698                 return -ENXIO;
1699         end = endswith(start, ".scope");
1700         if (!end)
1701                 return -ENXIO;
1702
1703         *end = 0;
1704         if (!session_id_valid(start))
1705                 return -ENXIO;
1706
1707         if (session) {
1708                 char *rr;
1709
1710                 rr = strdup(start);
1711                 if (!rr)
1712                         return -ENOMEM;
1713
1714                 *session = rr;
1715         }
1716
1717         return 0;
1718 }
1719
1720 int cg_pid_get_session(pid_t pid, char **session) {
1721         _cleanup_free_ char *cgroup = NULL;
1722         int r;
1723
1724         r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1725         if (r < 0)
1726                 return r;
1727
1728         return cg_path_get_session(cgroup, session);
1729 }
1730
1731 int cg_path_get_owner_uid(const char *path, uid_t *uid) {
1732         _cleanup_free_ char *slice = NULL;
1733         char *start, *end;
1734         int r;
1735
1736         assert(path);
1737
1738         r = cg_path_get_slice(path, &slice);
1739         if (r < 0)
1740                 return r;
1741
1742         start = startswith(slice, "user-");
1743         if (!start)
1744                 return -ENXIO;
1745         end = endswith(start, ".slice");
1746         if (!end)
1747                 return -ENXIO;
1748
1749         *end = 0;
1750         if (parse_uid(start, uid) < 0)
1751                 return -ENXIO;
1752
1753         return 0;
1754 }
1755
1756 int cg_pid_get_owner_uid(pid_t pid, uid_t *uid) {
1757         _cleanup_free_ char *cgroup = NULL;
1758         int r;
1759
1760         r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1761         if (r < 0)
1762                 return r;
1763
1764         return cg_path_get_owner_uid(cgroup, uid);
1765 }
1766
1767 int cg_path_get_slice(const char *p, char **slice) {
1768         const char *e = NULL;
1769
1770         assert(p);
1771         assert(slice);
1772
1773         /* Finds the right-most slice unit from the beginning, but
1774          * stops before we come to the first non-slice unit. */
1775
1776         for (;;) {
1777                 size_t n;
1778
1779                 p += strspn(p, "/");
1780
1781                 n = strcspn(p, "/");
1782                 if (!valid_slice_name(p, n)) {
1783
1784                         if (!e) {
1785                                 char *s;
1786
1787                                 s = strdup(SPECIAL_ROOT_SLICE);
1788                                 if (!s)
1789                                         return -ENOMEM;
1790
1791                                 *slice = s;
1792                                 return 0;
1793                         }
1794
1795                         return cg_path_decode_unit(e, slice);
1796                 }
1797
1798                 e = p;
1799                 p += n;
1800         }
1801 }
1802
1803 int cg_pid_get_slice(pid_t pid, char **slice) {
1804         _cleanup_free_ char *cgroup = NULL;
1805         int r;
1806
1807         assert(slice);
1808
1809         r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1810         if (r < 0)
1811                 return r;
1812
1813         return cg_path_get_slice(cgroup, slice);
1814 }
1815
1816 int cg_path_get_user_slice(const char *p, char **slice) {
1817         const char *t;
1818         assert(p);
1819         assert(slice);
1820
1821         t = skip_user_prefix(p);
1822         if (!t)
1823                 return -ENXIO;
1824
1825         /* And now it looks pretty much the same as for a system
1826          * slice, so let's just use the same parser from here on. */
1827         return cg_path_get_slice(t, slice);
1828 }
1829
1830 int cg_pid_get_user_slice(pid_t pid, char **slice) {
1831         _cleanup_free_ char *cgroup = NULL;
1832         int r;
1833
1834         assert(slice);
1835
1836         r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1837         if (r < 0)
1838                 return r;
1839
1840         return cg_path_get_user_slice(cgroup, slice);
1841 }
1842
1843 char *cg_escape(const char *p) {
1844         bool need_prefix = false;
1845
1846         /* This implements very minimal escaping for names to be used
1847          * as file names in the cgroup tree: any name which might
1848          * conflict with a kernel name or is prefixed with '_' is
1849          * prefixed with a '_'. That way, when reading cgroup names it
1850          * is sufficient to remove a single prefixing underscore if
1851          * there is one. */
1852
1853         /* The return value of this function (unlike cg_unescape())
1854          * needs free()! */
1855
1856         if (p[0] == 0 ||
1857             p[0] == '_' ||
1858             p[0] == '.' ||
1859             streq(p, "notify_on_release") ||
1860             streq(p, "release_agent") ||
1861             streq(p, "tasks") ||
1862             startswith(p, "cgroup."))
1863                 need_prefix = true;
1864         else {
1865                 const char *dot;
1866
1867                 dot = strrchr(p, '.');
1868                 if (dot) {
1869                         CGroupController c;
1870                         size_t l = dot - p;
1871
1872                         for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
1873                                 const char *n;
1874
1875                                 n = cgroup_controller_to_string(c);
1876
1877                                 if (l != strlen(n))
1878                                         continue;
1879
1880                                 if (memcmp(p, n, l) != 0)
1881                                         continue;
1882
1883                                 need_prefix = true;
1884                                 break;
1885                         }
1886                 }
1887         }
1888
1889         if (need_prefix)
1890                 return strappend("_", p);
1891
1892         return strdup(p);
1893 }
1894
1895 char *cg_unescape(const char *p) {
1896         assert(p);
1897
1898         /* The return value of this function (unlike cg_escape())
1899          * doesn't need free()! */
1900
1901         if (p[0] == '_')
1902                 return (char*) p+1;
1903
1904         return (char*) p;
1905 }
1906
1907 #define CONTROLLER_VALID                        \
1908         DIGITS LETTERS                          \
1909         "_"
1910
1911 bool cg_controller_is_valid(const char *p) {
1912         const char *t, *s;
1913
1914         if (!p)
1915                 return false;
1916
1917         if (streq(p, SYSTEMD_CGROUP_CONTROLLER))
1918                 return true;
1919
1920         s = startswith(p, "name=");
1921         if (s)
1922                 p = s;
1923
1924         if (*p == 0 || *p == '_')
1925                 return false;
1926
1927         for (t = p; *t; t++)
1928                 if (!strchr(CONTROLLER_VALID, *t))
1929                         return false;
1930
1931         if (t - p > FILENAME_MAX)
1932                 return false;
1933
1934         return true;
1935 }
1936
1937 int cg_slice_to_path(const char *unit, char **ret) {
1938         _cleanup_free_ char *p = NULL, *s = NULL, *e = NULL;
1939         const char *dash;
1940         int r;
1941
1942         assert(unit);
1943         assert(ret);
1944
1945         if (streq(unit, SPECIAL_ROOT_SLICE)) {
1946                 char *x;
1947
1948                 x = strdup("");
1949                 if (!x)
1950                         return -ENOMEM;
1951                 *ret = x;
1952                 return 0;
1953         }
1954
1955         if (!unit_name_is_valid(unit, UNIT_NAME_PLAIN))
1956                 return -EINVAL;
1957
1958         if (!endswith(unit, ".slice"))
1959                 return -EINVAL;
1960
1961         r = unit_name_to_prefix(unit, &p);
1962         if (r < 0)
1963                 return r;
1964
1965         dash = strchr(p, '-');
1966
1967         /* Don't allow initial dashes */
1968         if (dash == p)
1969                 return -EINVAL;
1970
1971         while (dash) {
1972                 _cleanup_free_ char *escaped = NULL;
1973                 char n[dash - p + sizeof(".slice")];
1974
1975                 /* Don't allow trailing or double dashes */
1976                 if (dash[1] == 0 || dash[1] == '-')
1977                         return -EINVAL;
1978
1979                 strcpy(stpncpy(n, p, dash - p), ".slice");
1980                 if (!unit_name_is_valid(n, UNIT_NAME_PLAIN))
1981                         return -EINVAL;
1982
1983                 escaped = cg_escape(n);
1984                 if (!escaped)
1985                         return -ENOMEM;
1986
1987                 if (!strextend(&s, escaped, "/", NULL))
1988                         return -ENOMEM;
1989
1990                 dash = strchr(dash+1, '-');
1991         }
1992
1993         e = cg_escape(unit);
1994         if (!e)
1995                 return -ENOMEM;
1996
1997         if (!strextend(&s, e, NULL))
1998                 return -ENOMEM;
1999
2000         *ret = s;
2001         s = NULL;
2002
2003         return 0;
2004 }
2005
2006 int cg_set_attribute(const char *controller, const char *path, const char *attribute, const char *value) {
2007         _cleanup_free_ char *p = NULL;
2008         int r;
2009
2010         r = cg_get_path(controller, path, attribute, &p);
2011         if (r < 0)
2012                 return r;
2013
2014         return write_string_file(p, value, 0);
2015 }
2016
2017 int cg_get_attribute(const char *controller, const char *path, const char *attribute, char **ret) {
2018         _cleanup_free_ char *p = NULL;
2019         int r;
2020
2021         r = cg_get_path(controller, path, attribute, &p);
2022         if (r < 0)
2023                 return r;
2024
2025         return read_one_line_file(p, ret);
2026 }
2027
2028 int cg_get_keyed_attribute(const char *controller, const char *path, const char *attribute, const char **keys, char **values) {
2029         _cleanup_free_ char *filename = NULL, *content = NULL;
2030         char *line, *p;
2031         int i, r;
2032
2033         for (i = 0; keys[i]; i++)
2034                 values[i] = NULL;
2035
2036         r = cg_get_path(controller, path, attribute, &filename);
2037         if (r < 0)
2038                 return r;
2039
2040         r = read_full_file(filename, &content, NULL);
2041         if (r < 0)
2042                 return r;
2043
2044         p = content;
2045         while ((line = strsep(&p, "\n"))) {
2046                 char *key;
2047
2048                 key = strsep(&line, " ");
2049
2050                 for (i = 0; keys[i]; i++) {
2051                         if (streq(key, keys[i])) {
2052                                 values[i] = strdup(line);
2053                                 break;
2054                         }
2055                 }
2056         }
2057
2058         for (i = 0; keys[i]; i++) {
2059                 if (!values[i]) {
2060                         for (i = 0; keys[i]; i++) {
2061                                 free(values[i]);
2062                                 values[i] = NULL;
2063                         }
2064                         return -ENOENT;
2065                 }
2066         }
2067
2068         return 0;
2069 }
2070
2071 int cg_create_everywhere(CGroupMask supported, CGroupMask mask, const char *path) {
2072         CGroupController c;
2073         int r;
2074
2075         /* This one will create a cgroup in our private tree, but also
2076          * duplicate it in the trees specified in mask, and remove it
2077          * in all others */
2078
2079         /* First create the cgroup in our own hierarchy. */
2080         r = cg_create(SYSTEMD_CGROUP_CONTROLLER, path);
2081         if (r < 0)
2082                 return r;
2083
2084         /* If we are in the unified hierarchy, we are done now */
2085         r = cg_all_unified();
2086         if (r < 0)
2087                 return r;
2088         if (r > 0)
2089                 return 0;
2090
2091         /* Otherwise, do the same in the other hierarchies */
2092         for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2093                 CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
2094                 const char *n;
2095
2096                 n = cgroup_controller_to_string(c);
2097
2098                 if (mask & bit)
2099                         (void) cg_create(n, path);
2100                 else if (supported & bit)
2101                         (void) cg_trim(n, path, true);
2102         }
2103
2104         return 0;
2105 }
2106
2107 int cg_attach_everywhere(CGroupMask supported, const char *path, pid_t pid, cg_migrate_callback_t path_callback, void *userdata) {
2108         CGroupController c;
2109         int r;
2110
2111         r = cg_attach(SYSTEMD_CGROUP_CONTROLLER, path, pid);
2112         if (r < 0)
2113                 return r;
2114
2115         r = cg_all_unified();
2116         if (r < 0)
2117                 return r;
2118         if (r > 0)
2119                 return 0;
2120
2121         for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2122                 CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
2123                 const char *p = NULL;
2124
2125                 if (!(supported & bit))
2126                         continue;
2127
2128                 if (path_callback)
2129                         p = path_callback(bit, userdata);
2130
2131                 if (!p)
2132                         p = path;
2133
2134                 (void) cg_attach_fallback(cgroup_controller_to_string(c), p, pid);
2135         }
2136
2137         return 0;
2138 }
2139
2140 int cg_attach_many_everywhere(CGroupMask supported, const char *path, Set* pids, cg_migrate_callback_t path_callback, void *userdata) {
2141         Iterator i;
2142         void *pidp;
2143         int r = 0;
2144
2145         SET_FOREACH(pidp, pids, i) {
2146                 pid_t pid = PTR_TO_PID(pidp);
2147                 int q;
2148
2149                 q = cg_attach_everywhere(supported, path, pid, path_callback, userdata);
2150                 if (q < 0 && r >= 0)
2151                         r = q;
2152         }
2153
2154         return r;
2155 }
2156
2157 int cg_migrate_everywhere(CGroupMask supported, const char *from, const char *to, cg_migrate_callback_t to_callback, void *userdata) {
2158         CGroupController c;
2159         int r = 0, q;
2160
2161         if (!path_equal(from, to))  {
2162                 r = cg_migrate_recursive(SYSTEMD_CGROUP_CONTROLLER, from, SYSTEMD_CGROUP_CONTROLLER, to, CGROUP_REMOVE);
2163                 if (r < 0)
2164                         return r;
2165         }
2166
2167         q = cg_all_unified();
2168         if (q < 0)
2169                 return q;
2170         if (q > 0)
2171                 return r;
2172
2173         for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2174                 CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
2175                 const char *p = NULL;
2176
2177                 if (!(supported & bit))
2178                         continue;
2179
2180                 if (to_callback)
2181                         p = to_callback(bit, userdata);
2182
2183                 if (!p)
2184                         p = to;
2185
2186                 (void) cg_migrate_recursive_fallback(SYSTEMD_CGROUP_CONTROLLER, to, cgroup_controller_to_string(c), p, 0);
2187         }
2188
2189         return 0;
2190 }
2191
2192 int cg_trim_everywhere(CGroupMask supported, const char *path, bool delete_root) {
2193         CGroupController c;
2194         int r, q;
2195
2196         r = cg_trim(SYSTEMD_CGROUP_CONTROLLER, path, delete_root);
2197         if (r < 0)
2198                 return r;
2199
2200         q = cg_all_unified();
2201         if (q < 0)
2202                 return q;
2203         if (q > 0)
2204                 return r;
2205
2206         for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2207                 CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
2208
2209                 if (!(supported & bit))
2210                         continue;
2211
2212                 (void) cg_trim(cgroup_controller_to_string(c), path, delete_root);
2213         }
2214
2215         return 0;
2216 }
2217
2218 int cg_mask_to_string(CGroupMask mask, char **ret) {
2219         const char *controllers[_CGROUP_CONTROLLER_MAX + 1];
2220         CGroupController c;
2221         int i = 0;
2222         char *s;
2223
2224         assert(ret);
2225
2226         if (mask == 0) {
2227                 *ret = NULL;
2228                 return 0;
2229         }
2230
2231         for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2232
2233                 if (!(mask & CGROUP_CONTROLLER_TO_MASK(c)))
2234                         continue;
2235
2236                 controllers[i++] = cgroup_controller_to_string(c);
2237                 controllers[i] = NULL;
2238         }
2239
2240         s = strv_join((char **)controllers, NULL);
2241         if (!s)
2242                 return -ENOMEM;
2243
2244         *ret = s;
2245         return 0;
2246 }
2247
2248 int cg_mask_from_string(const char *value, CGroupMask *mask) {
2249         assert(mask);
2250         assert(value);
2251
2252         for (;;) {
2253                 _cleanup_free_ char *n = NULL;
2254                 CGroupController v;
2255                 int r;
2256
2257                 r = extract_first_word(&value, &n, NULL, 0);
2258                 if (r < 0)
2259                         return r;
2260                 if (r == 0)
2261                         break;
2262
2263                 v = cgroup_controller_from_string(n);
2264                 if (v < 0)
2265                         continue;
2266
2267                 *mask |= CGROUP_CONTROLLER_TO_MASK(v);
2268         }
2269         return 0;
2270 }
2271
2272 int cg_mask_supported(CGroupMask *ret) {
2273         CGroupMask mask = 0;
2274         int r;
2275
2276         /* Determines the mask of supported cgroup controllers. Only
2277          * includes controllers we can make sense of and that are
2278          * actually accessible. */
2279
2280         r = cg_all_unified();
2281         if (r < 0)
2282                 return r;
2283         if (r > 0) {
2284                 _cleanup_free_ char *root = NULL, *controllers = NULL, *path = NULL;
2285
2286                 /* In the unified hierarchy we can read the supported
2287                  * and accessible controllers from a the top-level
2288                  * cgroup attribute */
2289
2290                 r = cg_get_root_path(&root);
2291                 if (r < 0)
2292                         return r;
2293
2294                 r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, root, "cgroup.controllers", &path);
2295                 if (r < 0)
2296                         return r;
2297
2298                 r = read_one_line_file(path, &controllers);
2299                 if (r < 0)
2300                         return r;
2301
2302                 r = cg_mask_from_string(controllers, &mask);
2303                 if (r < 0)
2304                         return r;
2305
2306                 /* Currently, we support the cpu, memory, io and pids
2307                  * controller in the unified hierarchy, mask
2308                  * everything else off. */
2309                 mask &= CGROUP_MASK_CPU | CGROUP_MASK_MEMORY | CGROUP_MASK_IO | CGROUP_MASK_PIDS;
2310
2311         } else {
2312                 CGroupController c;
2313
2314                 /* In the legacy hierarchy, we check whether which
2315                  * hierarchies are mounted. */
2316
2317                 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2318                         const char *n;
2319
2320                         n = cgroup_controller_to_string(c);
2321                         if (controller_is_accessible(n) >= 0)
2322                                 mask |= CGROUP_CONTROLLER_TO_MASK(c);
2323                 }
2324         }
2325
2326         *ret = mask;
2327         return 0;
2328 }
2329
2330 int cg_kernel_controllers(Set *controllers) {
2331         _cleanup_fclose_ FILE *f = NULL;
2332         char buf[LINE_MAX];
2333         int r;
2334
2335         assert(controllers);
2336
2337         /* Determines the full list of kernel-known controllers. Might
2338          * include controllers we don't actually support, arbitrary
2339          * named hierarchies and controllers that aren't currently
2340          * accessible (because not mounted). */
2341
2342         f = fopen("/proc/cgroups", "re");
2343         if (!f) {
2344                 if (errno == ENOENT)
2345                         return 0;
2346                 return -errno;
2347         }
2348
2349         /* Ignore the header line */
2350         (void) fgets(buf, sizeof(buf), f);
2351
2352         for (;;) {
2353                 char *controller;
2354                 int enabled = 0;
2355
2356                 errno = 0;
2357                 if (fscanf(f, "%ms %*i %*i %i", &controller, &enabled) != 2) {
2358
2359                         if (feof(f))
2360                                 break;
2361
2362                         if (ferror(f) && errno > 0)
2363                                 return -errno;
2364
2365                         return -EBADMSG;
2366                 }
2367
2368                 if (!enabled) {
2369                         free(controller);
2370                         continue;
2371                 }
2372
2373                 if (!cg_controller_is_valid(controller)) {
2374                         free(controller);
2375                         return -EBADMSG;
2376                 }
2377
2378                 r = set_consume(controllers, controller);
2379                 if (r < 0)
2380                         return r;
2381         }
2382
2383         return 0;
2384 }
2385
2386 static thread_local CGroupUnified unified_cache = CGROUP_UNIFIED_UNKNOWN;
2387
2388 /* The hybrid mode was initially implemented in v232 and simply mounted cgroup v2 on /sys/fs/cgroup/systemd.  This
2389  * unfortunately broke other tools (such as docker) which expected the v1 "name=systemd" hierarchy on
2390  * /sys/fs/cgroup/systemd.  From v233 and on, the hybrid mode mountnbs v2 on /sys/fs/cgroup/unified and maintains
2391  * "name=systemd" hierarchy on /sys/fs/cgroup/systemd for compatibility with other tools.
2392  *
2393  * To keep live upgrade working, we detect and support v232 layout.  When v232 layout is detected, to keep cgroup v2
2394  * process management but disable the compat dual layout, we return %true on
2395  * cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) and %false on cg_hybrid_unified().
2396  */
2397 static thread_local bool unified_systemd_v232;
2398
2399 static int cg_unified_update(void) {
2400
2401         struct statfs fs;
2402
2403         /* Checks if we support the unified hierarchy. Returns an
2404          * error when the cgroup hierarchies aren't mounted yet or we
2405          * have any other trouble determining if the unified hierarchy
2406          * is supported. */
2407
2408         if (unified_cache >= CGROUP_UNIFIED_NONE)
2409                 return 0;
2410
2411         if (statfs("/sys/fs/cgroup/", &fs) < 0)
2412                 return -errno;
2413
2414         if (F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC))
2415                 unified_cache = CGROUP_UNIFIED_ALL;
2416         else if (F_TYPE_EQUAL(fs.f_type, TMPFS_MAGIC)) {
2417                 if (statfs("/sys/fs/cgroup/unified/", &fs) == 0 &&
2418                     F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC)) {
2419                         unified_cache = CGROUP_UNIFIED_SYSTEMD;
2420                         unified_systemd_v232 = false;
2421                 } else if (statfs("/sys/fs/cgroup/systemd/", &fs) == 0 &&
2422                            F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC)) {
2423                         unified_cache = CGROUP_UNIFIED_SYSTEMD;
2424                         unified_systemd_v232 = true;
2425                 } else {
2426                         if (statfs("/sys/fs/cgroup/systemd/", &fs) < 0)
2427                                 return -errno;
2428                         if (!F_TYPE_EQUAL(fs.f_type, CGROUP_SUPER_MAGIC))
2429                                 return -ENOMEDIUM;
2430                         unified_cache = CGROUP_UNIFIED_NONE;
2431                 }
2432         } else
2433                 return -ENOMEDIUM;
2434
2435         return 0;
2436 }
2437
2438 int cg_unified_controller(const char *controller) {
2439         int r;
2440
2441         r = cg_unified_update();
2442         if (r < 0)
2443                 return r;
2444
2445         if (unified_cache == CGROUP_UNIFIED_NONE)
2446                 return false;
2447
2448         if (unified_cache >= CGROUP_UNIFIED_ALL)
2449                 return true;
2450
2451         return streq_ptr(controller, SYSTEMD_CGROUP_CONTROLLER);
2452 }
2453
2454 int cg_all_unified(void) {
2455         int r;
2456
2457         r = cg_unified_update();
2458         if (r < 0)
2459                 return r;
2460
2461         return unified_cache >= CGROUP_UNIFIED_ALL;
2462 }
2463
2464 int cg_hybrid_unified(void) {
2465         int r;
2466
2467         r = cg_unified_update();
2468         if (r < 0)
2469                 return r;
2470
2471         return unified_cache == CGROUP_UNIFIED_SYSTEMD && !unified_systemd_v232;
2472 }
2473
2474 int cg_unified_flush(void) {
2475         unified_cache = CGROUP_UNIFIED_UNKNOWN;
2476
2477         return cg_unified_update();
2478 }
2479
2480 int cg_enable_everywhere(CGroupMask supported, CGroupMask mask, const char *p) {
2481         _cleanup_free_ char *fs = NULL;
2482         CGroupController c;
2483         int r;
2484
2485         assert(p);
2486
2487         if (supported == 0)
2488                 return 0;
2489
2490         r = cg_all_unified();
2491         if (r < 0)
2492                 return r;
2493         if (r == 0) /* on the legacy hiearchy there's no joining of controllers defined */
2494                 return 0;
2495
2496         r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, p, "cgroup.subtree_control", &fs);
2497         if (r < 0)
2498                 return r;
2499
2500         for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2501                 CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
2502                 const char *n;
2503
2504                 if (!(supported & bit))
2505                         continue;
2506
2507                 n = cgroup_controller_to_string(c);
2508                 {
2509                         char s[1 + strlen(n) + 1];
2510
2511                         s[0] = mask & bit ? '+' : '-';
2512                         strcpy(s + 1, n);
2513
2514                         r = write_string_file(fs, s, 0);
2515                         if (r < 0)
2516                                 log_debug_errno(r, "Failed to enable controller %s for %s (%s): %m", n, p, fs);
2517                 }
2518         }
2519
2520         return 0;
2521 }
2522
2523 bool cg_is_unified_wanted(void) {
2524         static thread_local int wanted = -1;
2525         int r;
2526         bool b;
2527         const bool is_default = DEFAULT_HIERARCHY == CGROUP_UNIFIED_ALL;
2528
2529         /* If we have a cached value, return that. */
2530         if (wanted >= 0)
2531                 return wanted;
2532
2533         /* If the hierarchy is already mounted, then follow whatever
2534          * was chosen for it. */
2535         if (cg_unified_flush() >= 0)
2536                 return (wanted = unified_cache >= CGROUP_UNIFIED_ALL);
2537
2538         /* Otherwise, let's see what the kernel command line has to say.
2539          * Since checking is expensive, cache a non-error result. */
2540         r = proc_cmdline_get_bool("systemd.unified_cgroup_hierarchy", &b);
2541
2542         return (wanted = r > 0 ? b : is_default);
2543 }
2544
2545 bool cg_is_legacy_wanted(void) {
2546         static thread_local int wanted = -1;
2547
2548         /* If we have a cached value, return that. */
2549         if (wanted >= 0)
2550                 return wanted;
2551
2552         /* Check if we have cgroups2 already mounted. */
2553         if (cg_unified_flush() >= 0 &&
2554             unified_cache == CGROUP_UNIFIED_ALL)
2555                 return (wanted = false);
2556
2557         /* Otherwise, assume that at least partial legacy is wanted,
2558          * since cgroups2 should already be mounted at this point. */
2559         return (wanted = true);
2560 }
2561
2562 bool cg_is_hybrid_wanted(void) {
2563         static thread_local int wanted = -1;
2564         int r;
2565         bool b;
2566         const bool is_default = DEFAULT_HIERARCHY >= CGROUP_UNIFIED_SYSTEMD;
2567         /* We default to true if the default is "hybrid", obviously,
2568          * but also when the default is "unified", because if we get
2569          * called, it means that unified hierarchy was not mounted. */
2570
2571         /* If we have a cached value, return that. */
2572         if (wanted >= 0)
2573                 return wanted;
2574
2575         /* If the hierarchy is already mounted, then follow whatever
2576          * was chosen for it. */
2577         if (cg_unified_flush() >= 0 &&
2578             unified_cache == CGROUP_UNIFIED_ALL)
2579                 return (wanted = false);
2580
2581         /* Otherwise, let's see what the kernel command line has to say.
2582          * Since checking is expensive, cache a non-error result. */
2583         r = proc_cmdline_get_bool("systemd.legacy_systemd_cgroup_controller", &b);
2584
2585         /* The meaning of the kernel option is reversed wrt. to the return value
2586          * of this function, hence the negation. */
2587         return (wanted = r > 0 ? !b : is_default);
2588 }
2589
2590 int cg_weight_parse(const char *s, uint64_t *ret) {
2591         uint64_t u;
2592         int r;
2593
2594         if (isempty(s)) {
2595                 *ret = CGROUP_WEIGHT_INVALID;
2596                 return 0;
2597         }
2598
2599         r = safe_atou64(s, &u);
2600         if (r < 0)
2601                 return r;
2602
2603         if (u < CGROUP_WEIGHT_MIN || u > CGROUP_WEIGHT_MAX)
2604                 return -ERANGE;
2605
2606         *ret = u;
2607         return 0;
2608 }
2609
2610 const uint64_t cgroup_io_limit_defaults[_CGROUP_IO_LIMIT_TYPE_MAX] = {
2611         [CGROUP_IO_RBPS_MAX]    = CGROUP_LIMIT_MAX,
2612         [CGROUP_IO_WBPS_MAX]    = CGROUP_LIMIT_MAX,
2613         [CGROUP_IO_RIOPS_MAX]   = CGROUP_LIMIT_MAX,
2614         [CGROUP_IO_WIOPS_MAX]   = CGROUP_LIMIT_MAX,
2615 };
2616
2617 static const char* const cgroup_io_limit_type_table[_CGROUP_IO_LIMIT_TYPE_MAX] = {
2618         [CGROUP_IO_RBPS_MAX]    = "IOReadBandwidthMax",
2619         [CGROUP_IO_WBPS_MAX]    = "IOWriteBandwidthMax",
2620         [CGROUP_IO_RIOPS_MAX]   = "IOReadIOPSMax",
2621         [CGROUP_IO_WIOPS_MAX]   = "IOWriteIOPSMax",
2622 };
2623
2624 DEFINE_STRING_TABLE_LOOKUP(cgroup_io_limit_type, CGroupIOLimitType);
2625
2626 int cg_cpu_shares_parse(const char *s, uint64_t *ret) {
2627         uint64_t u;
2628         int r;
2629
2630         if (isempty(s)) {
2631                 *ret = CGROUP_CPU_SHARES_INVALID;
2632                 return 0;
2633         }
2634
2635         r = safe_atou64(s, &u);
2636         if (r < 0)
2637                 return r;
2638
2639         if (u < CGROUP_CPU_SHARES_MIN || u > CGROUP_CPU_SHARES_MAX)
2640                 return -ERANGE;
2641
2642         *ret = u;
2643         return 0;
2644 }
2645
2646 int cg_blkio_weight_parse(const char *s, uint64_t *ret) {
2647         uint64_t u;
2648         int r;
2649
2650         if (isempty(s)) {
2651                 *ret = CGROUP_BLKIO_WEIGHT_INVALID;
2652                 return 0;
2653         }
2654
2655         r = safe_atou64(s, &u);
2656         if (r < 0)
2657                 return r;
2658
2659         if (u < CGROUP_BLKIO_WEIGHT_MIN || u > CGROUP_BLKIO_WEIGHT_MAX)
2660                 return -ERANGE;
2661
2662         *ret = u;
2663         return 0;
2664 }
2665
2666 bool is_cgroup_fs(const struct statfs *s) {
2667         return is_fs_type(s, CGROUP_SUPER_MAGIC) ||
2668                is_fs_type(s, CGROUP2_SUPER_MAGIC);
2669 }
2670
2671 bool fd_is_cgroup_fs(int fd) {
2672         struct statfs s;
2673
2674         if (fstatfs(fd, &s) < 0)
2675                 return -errno;
2676
2677         return is_cgroup_fs(&s);
2678 }
2679
2680 static const char *cgroup_controller_table[_CGROUP_CONTROLLER_MAX] = {
2681         [CGROUP_CONTROLLER_CPU] = "cpu",
2682         [CGROUP_CONTROLLER_CPUACCT] = "cpuacct",
2683         [CGROUP_CONTROLLER_IO] = "io",
2684         [CGROUP_CONTROLLER_BLKIO] = "blkio",
2685         [CGROUP_CONTROLLER_MEMORY] = "memory",
2686         [CGROUP_CONTROLLER_DEVICES] = "devices",
2687         [CGROUP_CONTROLLER_PIDS] = "pids",
2688 };
2689
2690 DEFINE_STRING_TABLE_LOOKUP(cgroup_controller, CGroupController);