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