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