src/basic/cgroup-util.c

   1 /***
   2   This file is part of systemd.
   3
   4   Copyright 2010 Lennart Poettering
   5
   6   systemd is free software; you can redistribute it and/or modify it
   7   under the terms of the GNU Lesser General Public License as published by
   8   the Free Software Foundation; either version 2.1 of the License, or
   9   (at your option) any later version.
  10
  11   systemd is distributed in the hope that it will be useful, but
  12   WITHOUT ANY WARRANTY; without even the implied warranty of
  13   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  14   Lesser General Public License for more details.
  15
  16   You should have received a copy of the GNU Lesser General Public License
  17   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  18 ***/
  19
  20 #include <dirent.h>
  21 #include <errno.h>
  22 #include <ftw.h>
  23 #include <limits.h>
  24 #include <signal.h>
  25 #include <stddef.h>
  26 #include <stdlib.h>
  27 #include <string.h>
  28 #include <sys/stat.h>
  29 #include <sys/statfs.h>
  30 #include <sys/types.h>
  31 #include <sys/xattr.h>
  32 #include <unistd.h>
  33
  34 #include "alloc-util.h"
  35 #include "cgroup-util.h"
  36 #include "def.h"
  37 #include "dirent-util.h"
  38 #include "extract-word.h"
  39 #include "fd-util.h"
  40 #include "fileio.h"
  41 #include "format-util.h"
  42 #include "fs-util.h"
  43 #include "log.h"
  44 #include "login-util.h"
  45 #include "macro.h"
  46 #include "missing.h"
  47 #include "mkdir.h"
  48 #include "parse-util.h"
  49 #include "path-util.h"
  50 #include "proc-cmdline.h"
  51 #include "process-util.h"
  52 #include "set.h"
  53 #include "special.h"
  54 #include "stat-util.h"
  55 #include "stdio-util.h"
  56 #include "string-table.h"
  57 #include "string-util.h"
  58 #include "strv.h"
  59 #include "unit-name.h"
  60 #include "user-util.h"
  61
  62 int cg_enumerate_processes(const char *controller, const char *path, FILE **_f) {
  63         _cleanup_free_ char *fs = NULL;
  64         FILE *f;
  65         int r;
  66
  67         assert(_f);
  68
  69         r = cg_get_path(controller, path, "cgroup.procs", &fs);
  70         if (r < 0)
  71                 return r;
  72
  73         f = fopen(fs, "re");
  74         if (!f)
  75                 return -errno;
  76
  77         *_f = f;
  78         return 0;
  79 }
  80
  81 int cg_read_pid(FILE *f, pid_t *_pid) {
  82         unsigned long ul;
  83
  84         /* Note that the cgroup.procs might contain duplicates! See
  85          * cgroups.txt for details. */
  86
  87         assert(f);
  88         assert(_pid);
  89
  90         errno = 0;
  91         if (fscanf(f, "%lu", &ul) != 1) {
  92
  93                 if (feof(f))
  94                         return 0;
  95
  96                 return errno > 0 ? -errno : -EIO;
  97         }
  98
  99         if (ul <= 0)
 100                 return -EIO;
 101
 102         *_pid = (pid_t) ul;
 103         return 1;
 104 }
 105
 106 int cg_read_event(
 107                 const char *controller,
 108                 const char *path,
 109                 const char *event,
 110                 char **val) {
 111
 112         _cleanup_free_ char *events = NULL, *content = NULL;
 113         char *p, *line;
 114         int r;
 115
 116         r = cg_get_path(controller, path, "cgroup.events", &events);
 117         if (r < 0)
 118                 return r;
 119
 120         r = read_full_file(events, &content, NULL);
 121         if (r < 0)
 122                 return r;
 123
 124         p = content;
 125         while ((line = strsep(&p, "\n"))) {
 126                 char *key;
 127
 128                 key = strsep(&line, " ");
 129                 if (!key || !line)
 130                         return -EINVAL;
 131
 132                 if (strcmp(key, event))
 133                         continue;
 134
 135                 *val = strdup(line);
 136                 return 0;
 137         }
 138
 139         return -ENOENT;
 140 }
 141
 142 bool cg_ns_supported(void) {
 143         static thread_local int enabled = -1;
 144
 145         if (enabled >= 0)
 146                 return enabled;
 147
 148         if (access("/proc/self/ns/cgroup", F_OK) == 0)
 149                 enabled = 1;
 150         else
 151                 enabled = 0;
 152
 153         return enabled;
 154 }
 155
 156 int cg_enumerate_subgroups(const char *controller, const char *path, DIR **_d) {
 157         _cleanup_free_ char *fs = NULL;
 158         int r;
 159         DIR *d;
 160
 161         assert(_d);
 162
 163         /* This is not recursive! */
 164
 165         r = cg_get_path(controller, path, NULL, &fs);
 166         if (r < 0)
 167                 return r;
 168
 169         d = opendir(fs);
 170         if (!d)
 171                 return -errno;
 172
 173         *_d = d;
 174         return 0;
 175 }
 176
 177 int cg_read_subgroup(DIR *d, char **fn) {
 178         struct dirent *de;
 179
 180         assert(d);
 181         assert(fn);
 182
 183         FOREACH_DIRENT_ALL(de, d, return -errno) {
 184                 char *b;
 185
 186                 if (de->d_type != DT_DIR)
 187                         continue;
 188
 189                 if (dot_or_dot_dot(de->d_name))
 190                         continue;
 191
 192                 b = strdup(de->d_name);
 193                 if (!b)
 194                         return -ENOMEM;
 195
 196                 *fn = b;
 197                 return 1;
 198         }
 199
 200         return 0;
 201 }
 202
 203 int cg_rmdir(const char *controller, const char *path) {
 204         _cleanup_free_ char *p = NULL;
 205         int r;
 206
 207         r = cg_get_path(controller, path, NULL, &p);
 208         if (r < 0)
 209                 return r;
 210
 211         r = rmdir(p);
 212         if (r < 0 && errno != ENOENT)
 213                 return -errno;
 214
 215         r = cg_hybrid_unified();
 216         if (r < 0)
 217                 return r;
 218         if (r == 0)
 219                 return 0;
 220
 221         if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
 222                 r = cg_rmdir(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path);
 223                 if (r < 0)
 224                         log_warning_errno(r, "Failed to remove compat systemd cgroup %s: %m", path);
 225         }
 226
 227         return 0;
 228 }
 229
 230 int cg_kill(
 231                 const char *controller,
 232                 const char *path,
 233                 int sig,
 234                 CGroupFlags flags,
 235                 Set *s,
 236                 cg_kill_log_func_t log_kill,
 237                 void *userdata) {
 238
 239         _cleanup_set_free_ Set *allocated_set = NULL;
 240         bool done = false;
 241         int r, ret = 0;
 242         pid_t my_pid;
 243
 244         assert(sig >= 0);
 245
 246          /* Don't send SIGCONT twice. Also, SIGKILL always works even when process is suspended, hence don't send
 247           * SIGCONT on SIGKILL. */
 248         if (IN_SET(sig, SIGCONT, SIGKILL))
 249                 flags &= ~CGROUP_SIGCONT;
 250
 251         /* This goes through the tasks list and kills them all. This
 252          * is repeated until no further processes are added to the
 253          * tasks list, to properly handle forking processes */
 254
 255         if (!s) {
 256                 s = allocated_set = set_new(NULL);
 257                 if (!s)
 258                         return -ENOMEM;
 259         }
 260
 261         my_pid = getpid_cached();
 262
 263         do {
 264                 _cleanup_fclose_ FILE *f = NULL;
 265                 pid_t pid = 0;
 266                 done = true;
 267
 268                 r = cg_enumerate_processes(controller, path, &f);
 269                 if (r < 0) {
 270                         if (ret >= 0 && r != -ENOENT)
 271                                 return r;
 272
 273                         return ret;
 274                 }
 275
 276                 while ((r = cg_read_pid(f, &pid)) > 0) {
 277
 278                         if ((flags & CGROUP_IGNORE_SELF) && pid == my_pid)
 279                                 continue;
 280
 281                         if (set_get(s, PID_TO_PTR(pid)) == PID_TO_PTR(pid))
 282                                 continue;
 283
 284                         if (log_kill)
 285                                 log_kill(pid, sig, userdata);
 286
 287                         /* If we haven't killed this process yet, kill
 288                          * it */
 289                         if (kill(pid, sig) < 0) {
 290                                 if (ret >= 0 && errno != ESRCH)
 291                                         ret = -errno;
 292                         } else {
 293                                 if (flags & CGROUP_SIGCONT)
 294                                         (void) kill(pid, SIGCONT);
 295
 296                                 if (ret == 0)
 297                                         ret = 1;
 298                         }
 299
 300                         done = false;
 301
 302                         r = set_put(s, PID_TO_PTR(pid));
 303                         if (r < 0) {
 304                                 if (ret >= 0)
 305                                         return r;
 306
 307                                 return ret;
 308                         }
 309                 }
 310
 311                 if (r < 0) {
 312                         if (ret >= 0)
 313                                 return r;
 314
 315                         return ret;
 316                 }
 317
 318                 /* To avoid racing against processes which fork
 319                  * quicker than we can kill them we repeat this until
 320                  * no new pids need to be killed. */
 321
 322         } while (!done);
 323
 324         return ret;
 325 }
 326
 327 int cg_kill_recursive(
 328                 const char *controller,
 329                 const char *path,
 330                 int sig,
 331                 CGroupFlags flags,
 332                 Set *s,
 333                 cg_kill_log_func_t log_kill,
 334                 void *userdata) {
 335
 336         _cleanup_set_free_ Set *allocated_set = NULL;
 337         _cleanup_closedir_ DIR *d = NULL;
 338         int r, ret;
 339         char *fn;
 340
 341         assert(path);
 342         assert(sig >= 0);
 343
 344         if (!s) {
 345                 s = allocated_set = set_new(NULL);
 346                 if (!s)
 347                         return -ENOMEM;
 348         }
 349
 350         ret = cg_kill(controller, path, sig, flags, s, log_kill, userdata);
 351
 352         r = cg_enumerate_subgroups(controller, path, &d);
 353         if (r < 0) {
 354                 if (ret >= 0 && r != -ENOENT)
 355                         return r;
 356
 357                 return ret;
 358         }
 359
 360         while ((r = cg_read_subgroup(d, &fn)) > 0) {
 361                 _cleanup_free_ char *p = NULL;
 362
 363                 p = strjoin(path, "/", fn);
 364                 free(fn);
 365                 if (!p)
 366                         return -ENOMEM;
 367
 368                 r = cg_kill_recursive(controller, p, sig, flags, s, log_kill, userdata);
 369                 if (r != 0 && ret >= 0)
 370                         ret = r;
 371         }
 372         if (ret >= 0 && r < 0)
 373                 ret = r;
 374
 375         if (flags & CGROUP_REMOVE) {
 376                 r = cg_rmdir(controller, path);
 377                 if (r < 0 && ret >= 0 && !IN_SET(r, -ENOENT, -EBUSY))
 378                         return r;
 379         }
 380
 381         return ret;
 382 }
 383
 384 int cg_migrate(
 385                 const char *cfrom,
 386                 const char *pfrom,
 387                 const char *cto,
 388                 const char *pto,
 389                 CGroupFlags flags) {
 390
 391         bool done = false;
 392         _cleanup_set_free_ Set *s = NULL;
 393         int r, ret = 0;
 394         pid_t my_pid;
 395
 396         assert(cfrom);
 397         assert(pfrom);
 398         assert(cto);
 399         assert(pto);
 400
 401         s = set_new(NULL);
 402         if (!s)
 403                 return -ENOMEM;
 404
 405         my_pid = getpid_cached();
 406
 407         do {
 408                 _cleanup_fclose_ FILE *f = NULL;
 409                 pid_t pid = 0;
 410                 done = true;
 411
 412                 r = cg_enumerate_processes(cfrom, pfrom, &f);
 413                 if (r < 0) {
 414                         if (ret >= 0 && r != -ENOENT)
 415                                 return r;
 416
 417                         return ret;
 418                 }
 419
 420                 while ((r = cg_read_pid(f, &pid)) > 0) {
 421
 422                         /* This might do weird stuff if we aren't a
 423                          * single-threaded program. However, we
 424                          * luckily know we are not */
 425                         if ((flags & CGROUP_IGNORE_SELF) && pid == my_pid)
 426                                 continue;
 427
 428                         if (set_get(s, PID_TO_PTR(pid)) == PID_TO_PTR(pid))
 429                                 continue;
 430
 431                         /* Ignore kernel threads. Since they can only
 432                          * exist in the root cgroup, we only check for
 433                          * them there. */
 434                         if (cfrom &&
 435                             (isempty(pfrom) || path_equal(pfrom, "/")) &&
 436                             is_kernel_thread(pid) > 0)
 437                                 continue;
 438
 439                         r = cg_attach(cto, pto, pid);
 440                         if (r < 0) {
 441                                 if (ret >= 0 && r != -ESRCH)
 442                                         ret = r;
 443                         } else if (ret == 0)
 444                                 ret = 1;
 445
 446                         done = false;
 447
 448                         r = set_put(s, PID_TO_PTR(pid));
 449                         if (r < 0) {
 450                                 if (ret >= 0)
 451                                         return r;
 452
 453                                 return ret;
 454                         }
 455                 }
 456
 457                 if (r < 0) {
 458                         if (ret >= 0)
 459                                 return r;
 460
 461                         return ret;
 462                 }
 463         } while (!done);
 464
 465         return ret;
 466 }
 467
 468 int cg_migrate_recursive(
 469                 const char *cfrom,
 470                 const char *pfrom,
 471                 const char *cto,
 472                 const char *pto,
 473                 CGroupFlags flags) {
 474
 475         _cleanup_closedir_ DIR *d = NULL;
 476         int r, ret = 0;
 477         char *fn;
 478
 479         assert(cfrom);
 480         assert(pfrom);
 481         assert(cto);
 482         assert(pto);
 483
 484         ret = cg_migrate(cfrom, pfrom, cto, pto, flags);
 485
 486         r = cg_enumerate_subgroups(cfrom, pfrom, &d);
 487         if (r < 0) {
 488                 if (ret >= 0 && r != -ENOENT)
 489                         return r;
 490
 491                 return ret;
 492         }
 493
 494         while ((r = cg_read_subgroup(d, &fn)) > 0) {
 495                 _cleanup_free_ char *p = NULL;
 496
 497                 p = strjoin(pfrom, "/", fn);
 498                 free(fn);
 499                 if (!p)
 500                         return -ENOMEM;
 501
 502                 r = cg_migrate_recursive(cfrom, p, cto, pto, flags);
 503                 if (r != 0 && ret >= 0)
 504                         ret = r;
 505         }
 506
 507         if (r < 0 && ret >= 0)
 508                 ret = r;
 509
 510         if (flags & CGROUP_REMOVE) {
 511                 r = cg_rmdir(cfrom, pfrom);
 512                 if (r < 0 && ret >= 0 && !IN_SET(r, -ENOENT, -EBUSY))
 513                         return r;
 514         }
 515
 516         return ret;
 517 }
 518
 519 int cg_migrate_recursive_fallback(
 520                 const char *cfrom,
 521                 const char *pfrom,
 522                 const char *cto,
 523                 const char *pto,
 524                 CGroupFlags flags) {
 525
 526         int r;
 527
 528         assert(cfrom);
 529         assert(pfrom);
 530         assert(cto);
 531         assert(pto);
 532
 533         r = cg_migrate_recursive(cfrom, pfrom, cto, pto, flags);
 534         if (r < 0) {
 535                 char prefix[strlen(pto) + 1];
 536
 537                 /* This didn't work? Then let's try all prefixes of the destination */
 538
 539                 PATH_FOREACH_PREFIX(prefix, pto) {
 540                         int q;
 541
 542                         q = cg_migrate_recursive(cfrom, pfrom, cto, prefix, flags);
 543                         if (q >= 0)
 544                                 return q;
 545                 }
 546         }
 547
 548         return r;
 549 }
 550
 551 static const char *controller_to_dirname(const char *controller) {
 552         const char *e;
 553
 554         assert(controller);
 555
 556         /* Converts a controller name to the directory name below
 557          * /sys/fs/cgroup/ we want to mount it to. Effectively, this
 558          * just cuts off the name= prefixed used for named
 559          * hierarchies, if it is specified. */
 560
 561         if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
 562                 if (cg_hybrid_unified() > 0)
 563                         controller = SYSTEMD_CGROUP_CONTROLLER_HYBRID;
 564                 else
 565                         controller = SYSTEMD_CGROUP_CONTROLLER_LEGACY;
 566         }
 567
 568         e = startswith(controller, "name=");
 569         if (e)
 570                 return e;
 571
 572         return controller;
 573 }
 574
 575 static int join_path_legacy(const char *controller, const char *path, const char *suffix, char **fs) {
 576         const char *dn;
 577         char *t = NULL;
 578
 579         assert(fs);
 580         assert(controller);
 581
 582         dn = controller_to_dirname(controller);
 583
 584         if (isempty(path) && isempty(suffix))
 585                 t = strappend("/sys/fs/cgroup/", dn);
 586         else if (isempty(path))
 587                 t = strjoin("/sys/fs/cgroup/", dn, "/", suffix);
 588         else if (isempty(suffix))
 589                 t = strjoin("/sys/fs/cgroup/", dn, "/", path);
 590         else
 591                 t = strjoin("/sys/fs/cgroup/", dn, "/", path, "/", suffix);
 592         if (!t)
 593                 return -ENOMEM;
 594
 595         *fs = t;
 596         return 0;
 597 }
 598
 599 static int join_path_unified(const char *path, const char *suffix, char **fs) {
 600         char *t;
 601
 602         assert(fs);
 603
 604         if (isempty(path) && isempty(suffix))
 605                 t = strdup("/sys/fs/cgroup");
 606         else if (isempty(path))
 607                 t = strappend("/sys/fs/cgroup/", suffix);
 608         else if (isempty(suffix))
 609                 t = strappend("/sys/fs/cgroup/", path);
 610         else
 611                 t = strjoin("/sys/fs/cgroup/", path, "/", suffix);
 612         if (!t)
 613                 return -ENOMEM;
 614
 615         *fs = t;
 616         return 0;
 617 }
 618
 619 int cg_get_path(const char *controller, const char *path, const char *suffix, char **fs) {
 620         int r;
 621
 622         assert(fs);
 623
 624         if (!controller) {
 625                 char *t;
 626
 627                 /* If no controller is specified, we return the path
 628                  * *below* the controllers, without any prefix. */
 629
 630                 if (!path && !suffix)
 631                         return -EINVAL;
 632
 633                 if (!suffix)
 634                         t = strdup(path);
 635                 else if (!path)
 636                         t = strdup(suffix);
 637                 else
 638                         t = strjoin(path, "/", suffix);
 639                 if (!t)
 640                         return -ENOMEM;
 641
 642                 *fs = path_kill_slashes(t);
 643                 return 0;
 644         }
 645
 646         if (!cg_controller_is_valid(controller))
 647                 return -EINVAL;
 648
 649         r = cg_all_unified();
 650         if (r < 0)
 651                 return r;
 652         if (r > 0)
 653                 r = join_path_unified(path, suffix, fs);
 654         else
 655                 r = join_path_legacy(controller, path, suffix, fs);
 656         if (r < 0)
 657                 return r;
 658
 659         path_kill_slashes(*fs);
 660         return 0;
 661 }
 662
 663 static int controller_is_accessible(const char *controller) {
 664         int r;
 665
 666         assert(controller);
 667
 668         /* Checks whether a specific controller is accessible,
 669          * i.e. its hierarchy mounted. In the unified hierarchy all
 670          * controllers are considered accessible, except for the named
 671          * hierarchies */
 672
 673         if (!cg_controller_is_valid(controller))
 674                 return -EINVAL;
 675
 676         r = cg_all_unified();
 677         if (r < 0)
 678                 return r;
 679         if (r > 0) {
 680                 /* We don't support named hierarchies if we are using
 681                  * the unified hierarchy. */
 682
 683                 if (streq(controller, SYSTEMD_CGROUP_CONTROLLER))
 684                         return 0;
 685
 686                 if (startswith(controller, "name="))
 687                         return -EOPNOTSUPP;
 688
 689         } else {
 690                 const char *cc, *dn;
 691
 692                 dn = controller_to_dirname(controller);
 693                 cc = strjoina("/sys/fs/cgroup/", dn);
 694
 695                 if (laccess(cc, F_OK) < 0)
 696                         return -errno;
 697         }
 698
 699         return 0;
 700 }
 701
 702 int cg_get_path_and_check(const char *controller, const char *path, const char *suffix, char **fs) {
 703         int r;
 704
 705         assert(controller);
 706         assert(fs);
 707
 708         /* Check if the specified controller is actually accessible */
 709         r = controller_is_accessible(controller);
 710         if (r < 0)
 711                 return r;
 712
 713         return cg_get_path(controller, path, suffix, fs);
 714 }
 715
 716 static int trim_cb(const char *path, const struct stat *sb, int typeflag, struct FTW *ftwbuf) {
 717         assert(path);
 718         assert(sb);
 719         assert(ftwbuf);
 720
 721         if (typeflag != FTW_DP)
 722                 return 0;
 723
 724         if (ftwbuf->level < 1)
 725                 return 0;
 726
 727         (void) rmdir(path);
 728         return 0;
 729 }
 730
 731 int cg_trim(const char *controller, const char *path, bool delete_root) {
 732         _cleanup_free_ char *fs = NULL;
 733         int r = 0, q;
 734
 735         assert(path);
 736
 737         r = cg_get_path(controller, path, NULL, &fs);
 738         if (r < 0)
 739                 return r;
 740
 741         errno = 0;
 742         if (nftw(fs, trim_cb, 64, FTW_DEPTH|FTW_MOUNT|FTW_PHYS) != 0) {
 743                 if (errno == ENOENT)
 744                         r = 0;
 745                 else if (errno > 0)
 746                         r = -errno;
 747                 else
 748                         r = -EIO;
 749         }
 750
 751         if (delete_root) {
 752                 if (rmdir(fs) < 0 && errno != ENOENT)
 753                         return -errno;
 754         }
 755
 756         q = cg_hybrid_unified();
 757         if (q < 0)
 758                 return q;
 759         if (q > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
 760                 q = cg_trim(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, delete_root);
 761                 if (q < 0)
 762                         log_warning_errno(q, "Failed to trim compat systemd cgroup %s: %m", path);
 763         }
 764
 765         return r;
 766 }
 767
 768 int cg_create(const char *controller, const char *path) {
 769         _cleanup_free_ char *fs = NULL;
 770         int r;
 771
 772         r = cg_get_path_and_check(controller, path, NULL, &fs);
 773         if (r < 0)
 774                 return r;
 775
 776         r = mkdir_parents(fs, 0755);
 777         if (r < 0)
 778                 return r;
 779
 780         if (mkdir(fs, 0755) < 0) {
 781
 782                 if (errno == EEXIST)
 783                         return 0;
 784
 785                 return -errno;
 786         }
 787
 788         r = cg_hybrid_unified();
 789         if (r < 0)
 790                 return r;
 791
 792         if (r > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
 793                 r = cg_create(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path);
 794                 if (r < 0)
 795                         log_warning_errno(r, "Failed to create compat systemd cgroup %s: %m", path);
 796         }
 797
 798         return 1;
 799 }
 800
 801 int cg_create_and_attach(const char *controller, const char *path, pid_t pid) {
 802         int r, q;
 803
 804         assert(pid >= 0);
 805
 806         r = cg_create(controller, path);
 807         if (r < 0)
 808                 return r;
 809
 810         q = cg_attach(controller, path, pid);
 811         if (q < 0)
 812                 return q;
 813
 814         /* This does not remove the cgroup on failure */
 815         return r;
 816 }
 817
 818 int cg_attach(const char *controller, const char *path, pid_t pid) {
 819         _cleanup_free_ char *fs = NULL;
 820         char c[DECIMAL_STR_MAX(pid_t) + 2];
 821         int r;
 822
 823         assert(path);
 824         assert(pid >= 0);
 825
 826         r = cg_get_path_and_check(controller, path, "cgroup.procs", &fs);
 827         if (r < 0)
 828                 return r;
 829
 830         if (pid == 0)
 831                 pid = getpid_cached();
 832
 833         xsprintf(c, PID_FMT "\n", pid);
 834
 835         r = write_string_file(fs, c, 0);
 836         if (r < 0)
 837                 return r;
 838
 839         r = cg_hybrid_unified();
 840         if (r < 0)
 841                 return r;
 842
 843         if (r > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
 844                 r = cg_attach(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, pid);
 845                 if (r < 0)
 846                         log_warning_errno(r, "Failed to attach "PID_FMT" to compat systemd cgroup %s: %m", pid, path);
 847         }
 848
 849         return 0;
 850 }
 851
 852 int cg_attach_fallback(const char *controller, const char *path, pid_t pid) {
 853         int r;
 854
 855         assert(controller);
 856         assert(path);
 857         assert(pid >= 0);
 858
 859         r = cg_attach(controller, path, pid);
 860         if (r < 0) {
 861                 char prefix[strlen(path) + 1];
 862
 863                 /* This didn't work? Then let's try all prefixes of
 864                  * the destination */
 865
 866                 PATH_FOREACH_PREFIX(prefix, path) {
 867                         int q;
 868
 869                         q = cg_attach(controller, prefix, pid);
 870                         if (q >= 0)
 871                                 return q;
 872                 }
 873         }
 874
 875         return r;
 876 }
 877
 878 int cg_set_group_access(
 879                 const char *controller,
 880                 const char *path,
 881                 mode_t mode,
 882                 uid_t uid,
 883                 gid_t gid) {
 884
 885         _cleanup_free_ char *fs = NULL;
 886         int r;
 887
 888         if (mode == MODE_INVALID && uid == UID_INVALID && gid == GID_INVALID)
 889                 return 0;
 890
 891         if (mode != MODE_INVALID)
 892                 mode &= 0777;
 893
 894         r = cg_get_path(controller, path, NULL, &fs);
 895         if (r < 0)
 896                 return r;
 897
 898         r = chmod_and_chown(fs, mode, uid, gid);
 899         if (r < 0)
 900                 return r;
 901
 902         r = cg_hybrid_unified();
 903         if (r < 0)
 904                 return r;
 905         if (r > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
 906                 r = cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, mode, uid, gid);
 907                 if (r < 0)
 908                         log_debug_errno(r, "Failed to set group access on compatibility systemd cgroup %s, ignoring: %m", path);
 909         }
 910
 911         return 0;
 912 }
 913
 914 int cg_set_task_access(
 915                 const char *controller,
 916                 const char *path,
 917                 mode_t mode,
 918                 uid_t uid,
 919                 gid_t gid) {
 920
 921         _cleanup_free_ char *fs = NULL;
 922         int r;
 923
 924         assert(path);
 925
 926         if (mode == MODE_INVALID && uid == UID_INVALID && gid == GID_INVALID)
 927                 return 0;
 928
 929         if (mode != MODE_INVALID)
 930                 mode &= 0666;
 931
 932         /* For both the legacy and unified hierarchies, "cgroup.procs" is the main entry point for PIDs */
 933         r = cg_get_path(controller, path, "cgroup.procs", &fs);
 934         if (r < 0)
 935                 return r;
 936
 937         r = chmod_and_chown(fs, mode, uid, gid);
 938         if (r < 0)
 939                 return r;
 940
 941         r = cg_unified_controller(controller);
 942         if (r < 0)
 943                 return r;
 944         if (r == 0) {
 945                 const char *fn;
 946
 947                 /* Compatibility: on cgroupsv1 always keep values for the legacy files "tasks" and
 948                  * "cgroup.clone_children" in sync with "cgroup.procs". Since this is legacy stuff, we don't care if
 949                  * this fails. */
 950
 951                 FOREACH_STRING(fn,
 952                                "tasks",
 953                                "cgroup.clone_children") {
 954
 955                         fs = mfree(fs);
 956
 957                         r = cg_get_path(controller, path, fn, &fs);
 958                         if (r < 0)
 959                                 log_debug_errno(r, "Failed to get path for %s of %s, ignoring: %m", fn, path);
 960
 961                         r = chmod_and_chown(fs, mode, uid, gid);
 962                         if (r < 0)
 963                                 log_debug_errno(r, "Failed to to change ownership/access mode for %s of %s, ignoring: %m", fn, path);
 964                 }
 965         } else {
 966                 /* On the unified controller, we want to permit subtree controllers too. */
 967
 968                 fs = mfree(fs);
 969                 r = cg_get_path(controller, path, "cgroup.subtree_control", &fs);
 970                 if (r < 0)
 971                         return r;
 972
 973                 r = chmod_and_chown(fs, mode, uid, gid);
 974                 if (r < 0)
 975                         return r;
 976         }
 977
 978         r = cg_hybrid_unified();
 979         if (r < 0)
 980                 return r;
 981         if (r > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
 982                 /* Always propagate access mode from unified to legacy controller */
 983
 984                 r = cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, mode, uid, gid);
 985                 if (r < 0)
 986                         log_debug_errno(r, "Failed to set task access on compatibility systemd cgroup %s, ignoring: %m", path);
 987         }
 988
 989         return 0;
 990 }
 991
 992 int cg_set_xattr(const char *controller, const char *path, const char *name, const void *value, size_t size, int flags) {
 993         _cleanup_free_ char *fs = NULL;
 994         int r;
 995
 996         assert(path);
 997         assert(name);
 998         assert(value || size <= 0);
 999
1000         r = cg_get_path(controller, path, NULL, &fs);
1001         if (r < 0)
1002                 return r;
1003
1004         if (setxattr(fs, name, value, size, flags) < 0)
1005                 return -errno;
1006
1007         return 0;
1008 }
1009
1010 int cg_get_xattr(const char *controller, const char *path, const char *name, void *value, size_t size) {
1011         _cleanup_free_ char *fs = NULL;
1012         ssize_t n;
1013         int r;
1014
1015         assert(path);
1016         assert(name);
1017
1018         r = cg_get_path(controller, path, NULL, &fs);
1019         if (r < 0)
1020                 return r;
1021
1022         n = getxattr(fs, name, value, size);
1023         if (n < 0)
1024                 return -errno;
1025
1026         return (int) n;
1027 }
1028
1029 int cg_pid_get_path(const char *controller, pid_t pid, char **path) {
1030         _cleanup_fclose_ FILE *f = NULL;
1031         char line[LINE_MAX];
1032         const char *fs, *controller_str;
1033         size_t cs = 0;
1034         int unified;
1035
1036         assert(path);
1037         assert(pid >= 0);
1038
1039         if (controller) {
1040                 if (!cg_controller_is_valid(controller))
1041                         return -EINVAL;
1042         } else
1043                 controller = SYSTEMD_CGROUP_CONTROLLER;
1044
1045         unified = cg_unified_controller(controller);
1046         if (unified < 0)
1047                 return unified;
1048         if (unified == 0) {
1049                 if (streq(controller, SYSTEMD_CGROUP_CONTROLLER))
1050                         controller_str = SYSTEMD_CGROUP_CONTROLLER_LEGACY;
1051                 else
1052                         controller_str = controller;
1053
1054                 cs = strlen(controller_str);
1055         }
1056
1057         fs = procfs_file_alloca(pid, "cgroup");
1058         f = fopen(fs, "re");
1059         if (!f)
1060                 return errno == ENOENT ? -ESRCH : -errno;
1061
1062         FOREACH_LINE(line, f, return -errno) {
1063                 char *e, *p;
1064
1065                 truncate_nl(line);
1066
1067                 if (unified) {
1068                         e = startswith(line, "0:");
1069                         if (!e)
1070                                 continue;
1071
1072                         e = strchr(e, ':');
1073                         if (!e)
1074                                 continue;
1075                 } else {
1076                         char *l;
1077                         size_t k;
1078                         const char *word, *state;
1079                         bool found = false;
1080
1081                         l = strchr(line, ':');
1082                         if (!l)
1083                                 continue;
1084
1085                         l++;
1086                         e = strchr(l, ':');
1087                         if (!e)
1088                                 continue;
1089
1090                         *e = 0;
1091                         FOREACH_WORD_SEPARATOR(word, k, l, ",", state) {
1092                                 if (k == cs && memcmp(word, controller_str, cs) == 0) {
1093                                         found = true;
1094                                         break;
1095                                 }
1096                         }
1097
1098                         if (!found)
1099                                 continue;
1100                 }
1101
1102                 p = strdup(e + 1);
1103                 if (!p)
1104                         return -ENOMEM;
1105
1106                 *path = p;
1107                 return 0;
1108         }
1109
1110         return -ENODATA;
1111 }
1112
1113 int cg_install_release_agent(const char *controller, const char *agent) {
1114         _cleanup_free_ char *fs = NULL, *contents = NULL;
1115         const char *sc;
1116         int r;
1117
1118         assert(agent);
1119
1120         r = cg_unified_controller(controller);
1121         if (r < 0)
1122                 return r;
1123         if (r > 0) /* doesn't apply to unified hierarchy */
1124                 return -EOPNOTSUPP;
1125
1126         r = cg_get_path(controller, NULL, "release_agent", &fs);
1127         if (r < 0)
1128                 return r;
1129
1130         r = read_one_line_file(fs, &contents);
1131         if (r < 0)
1132                 return r;
1133
1134         sc = strstrip(contents);
1135         if (isempty(sc)) {
1136                 r = write_string_file(fs, agent, 0);
1137                 if (r < 0)
1138                         return r;
1139         } else if (!path_equal(sc, agent))
1140                 return -EEXIST;
1141
1142         fs = mfree(fs);
1143         r = cg_get_path(controller, NULL, "notify_on_release", &fs);
1144         if (r < 0)
1145                 return r;
1146
1147         contents = mfree(contents);
1148         r = read_one_line_file(fs, &contents);
1149         if (r < 0)
1150                 return r;
1151
1152         sc = strstrip(contents);
1153         if (streq(sc, "0")) {
1154                 r = write_string_file(fs, "1", 0);
1155                 if (r < 0)
1156                         return r;
1157
1158                 return 1;
1159         }
1160
1161         if (!streq(sc, "1"))
1162                 return -EIO;
1163
1164         return 0;
1165 }
1166
1167 int cg_uninstall_release_agent(const char *controller) {
1168         _cleanup_free_ char *fs = NULL;
1169         int r;
1170
1171         r = cg_unified_controller(controller);
1172         if (r < 0)
1173                 return r;
1174         if (r > 0) /* Doesn't apply to unified hierarchy */
1175                 return -EOPNOTSUPP;
1176
1177         r = cg_get_path(controller, NULL, "notify_on_release", &fs);
1178         if (r < 0)
1179                 return r;
1180
1181         r = write_string_file(fs, "0", 0);
1182         if (r < 0)
1183                 return r;
1184
1185         fs = mfree(fs);
1186
1187         r = cg_get_path(controller, NULL, "release_agent", &fs);
1188         if (r < 0)
1189                 return r;
1190
1191         r = write_string_file(fs, "", 0);
1192         if (r < 0)
1193                 return r;
1194
1195         return 0;
1196 }
1197
1198 int cg_is_empty(const char *controller, const char *path) {
1199         _cleanup_fclose_ FILE *f = NULL;
1200         pid_t pid;
1201         int r;
1202
1203         assert(path);
1204
1205         r = cg_enumerate_processes(controller, path, &f);
1206         if (r == -ENOENT)
1207                 return 1;
1208         if (r < 0)
1209                 return r;
1210
1211         r = cg_read_pid(f, &pid);
1212         if (r < 0)
1213                 return r;
1214
1215         return r == 0;
1216 }
1217
1218 int cg_is_empty_recursive(const char *controller, const char *path) {
1219         int r;
1220
1221         assert(path);
1222
1223         /* The root cgroup is always populated */
1224         if (controller && (isempty(path) || path_equal(path, "/")))
1225                 return false;
1226
1227         r = cg_unified_controller(controller);
1228         if (r < 0)
1229                 return r;
1230         if (r > 0) {
1231                 _cleanup_free_ char *t = NULL;
1232
1233                 /* On the unified hierarchy we can check empty state
1234                  * via the "populated" attribute of "cgroup.events". */
1235
1236                 r = cg_read_event(controller, path, "populated", &t);
1237                 if (r < 0)
1238                         return r;
1239
1240                 return streq(t, "0");
1241         } else {
1242                 _cleanup_closedir_ DIR *d = NULL;
1243                 char *fn;
1244
1245                 r = cg_is_empty(controller, path);
1246                 if (r <= 0)
1247                         return r;
1248
1249                 r = cg_enumerate_subgroups(controller, path, &d);
1250                 if (r == -ENOENT)
1251                         return 1;
1252                 if (r < 0)
1253                         return r;
1254
1255                 while ((r = cg_read_subgroup(d, &fn)) > 0) {
1256                         _cleanup_free_ char *p = NULL;
1257
1258                         p = strjoin(path, "/", fn);
1259                         free(fn);
1260                         if (!p)
1261                                 return -ENOMEM;
1262
1263                         r = cg_is_empty_recursive(controller, p);
1264                         if (r <= 0)
1265                                 return r;
1266                 }
1267                 if (r < 0)
1268                         return r;
1269
1270                 return true;
1271         }
1272 }
1273
1274 int cg_split_spec(const char *spec, char **controller, char **path) {
1275         char *t = NULL, *u = NULL;
1276         const char *e;
1277
1278         assert(spec);
1279
1280         if (*spec == '/') {
1281                 if (!path_is_safe(spec))
1282                         return -EINVAL;
1283
1284                 if (path) {
1285                         t = strdup(spec);
1286                         if (!t)
1287                                 return -ENOMEM;
1288
1289                         *path = path_kill_slashes(t);
1290                 }
1291
1292                 if (controller)
1293                         *controller = NULL;
1294
1295                 return 0;
1296         }
1297
1298         e = strchr(spec, ':');
1299         if (!e) {
1300                 if (!cg_controller_is_valid(spec))
1301                         return -EINVAL;
1302
1303                 if (controller) {
1304                         t = strdup(spec);
1305                         if (!t)
1306                                 return -ENOMEM;
1307
1308                         *controller = t;
1309                 }
1310
1311                 if (path)
1312                         *path = NULL;
1313
1314                 return 0;
1315         }
1316
1317         t = strndup(spec, e-spec);
1318         if (!t)
1319                 return -ENOMEM;
1320         if (!cg_controller_is_valid(t)) {
1321                 free(t);
1322                 return -EINVAL;
1323         }
1324
1325         if (isempty(e+1))
1326                 u = NULL;
1327         else {
1328                 u = strdup(e+1);
1329                 if (!u) {
1330                         free(t);
1331                         return -ENOMEM;
1332                 }
1333
1334                 if (!path_is_safe(u) ||
1335                     !path_is_absolute(u)) {
1336                         free(t);
1337                         free(u);
1338                         return -EINVAL;
1339                 }
1340
1341                 path_kill_slashes(u);
1342         }
1343
1344         if (controller)
1345                 *controller = t;
1346         else
1347                 free(t);
1348
1349         if (path)
1350                 *path = u;
1351         else
1352                 free(u);
1353
1354         return 0;
1355 }
1356
1357 int cg_mangle_path(const char *path, char **result) {
1358         _cleanup_free_ char *c = NULL, *p = NULL;
1359         char *t;
1360         int r;
1361
1362         assert(path);
1363         assert(result);
1364
1365         /* First, check if it already is a filesystem path */
1366         if (path_startswith(path, "/sys/fs/cgroup")) {
1367
1368                 t = strdup(path);
1369                 if (!t)
1370                         return -ENOMEM;
1371
1372                 *result = path_kill_slashes(t);
1373                 return 0;
1374         }
1375
1376         /* Otherwise, treat it as cg spec */
1377         r = cg_split_spec(path, &c, &p);
1378         if (r < 0)
1379                 return r;
1380
1381         return cg_get_path(c ?: SYSTEMD_CGROUP_CONTROLLER, p ?: "/", NULL, result);
1382 }
1383
1384 int cg_get_root_path(char **path) {
1385         char *p, *e;
1386         int r;
1387
1388         assert(path);
1389
1390         r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 1, &p);
1391         if (r < 0)
1392                 return r;
1393
1394         e = endswith(p, "/" SPECIAL_INIT_SCOPE);
1395         if (!e)
1396                 e = endswith(p, "/" SPECIAL_SYSTEM_SLICE); /* legacy */
1397         if (!e)
1398                 e = endswith(p, "/system"); /* even more legacy */
1399         if (e)
1400                 *e = 0;
1401
1402         *path = p;
1403         return 0;
1404 }
1405
1406 int cg_shift_path(const char *cgroup, const char *root, const char **shifted) {
1407         _cleanup_free_ char *rt = NULL;
1408         char *p;
1409         int r;
1410
1411         assert(cgroup);
1412         assert(shifted);
1413
1414         if (!root) {
1415                 /* If the root was specified let's use that, otherwise
1416                  * let's determine it from PID 1 */
1417
1418                 r = cg_get_root_path(&rt);
1419                 if (r < 0)
1420                         return r;
1421
1422                 root = rt;
1423         }
1424
1425         p = path_startswith(cgroup, root);
1426         if (p && p > cgroup)
1427                 *shifted = p - 1;
1428         else
1429                 *shifted = cgroup;
1430
1431         return 0;
1432 }
1433
1434 int cg_pid_get_path_shifted(pid_t pid, const char *root, char **cgroup) {
1435         _cleanup_free_ char *raw = NULL;
1436         const char *c;
1437         int r;
1438
1439         assert(pid >= 0);
1440         assert(cgroup);
1441
1442         r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &raw);
1443         if (r < 0)
1444                 return r;
1445
1446         r = cg_shift_path(raw, root, &c);
1447         if (r < 0)
1448                 return r;
1449
1450         if (c == raw) {
1451                 *cgroup = raw;
1452                 raw = NULL;
1453         } else {
1454                 char *n;
1455
1456                 n = strdup(c);
1457                 if (!n)
1458                         return -ENOMEM;
1459
1460                 *cgroup = n;
1461         }
1462
1463         return 0;
1464 }
1465
1466 int cg_path_decode_unit(const char *cgroup, char **unit) {
1467         char *c, *s;
1468         size_t n;
1469
1470         assert(cgroup);
1471         assert(unit);
1472
1473         n = strcspn(cgroup, "/");
1474         if (n < 3)
1475                 return -ENXIO;
1476
1477         c = strndupa(cgroup, n);
1478         c = cg_unescape(c);
1479
1480         if (!unit_name_is_valid(c, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE))
1481                 return -ENXIO;
1482
1483         s = strdup(c);
1484         if (!s)
1485                 return -ENOMEM;
1486
1487         *unit = s;
1488         return 0;
1489 }
1490
1491 static bool valid_slice_name(const char *p, size_t n) {
1492
1493         if (!p)
1494                 return false;
1495
1496         if (n < strlen("x.slice"))
1497                 return false;
1498
1499         if (memcmp(p + n - 6, ".slice", 6) == 0) {
1500                 char buf[n+1], *c;
1501
1502                 memcpy(buf, p, n);
1503                 buf[n] = 0;
1504
1505                 c = cg_unescape(buf);
1506
1507                 return unit_name_is_valid(c, UNIT_NAME_PLAIN);
1508         }
1509
1510         return false;
1511 }
1512
1513 static const char *skip_slices(const char *p) {
1514         assert(p);
1515
1516         /* Skips over all slice assignments */
1517
1518         for (;;) {
1519                 size_t n;
1520
1521                 p += strspn(p, "/");
1522
1523                 n = strcspn(p, "/");
1524                 if (!valid_slice_name(p, n))
1525                         return p;
1526
1527                 p += n;
1528         }
1529 }
1530
1531 int cg_path_get_unit(const char *path, char **ret) {
1532         const char *e;
1533         char *unit;
1534         int r;
1535
1536         assert(path);
1537         assert(ret);
1538
1539         e = skip_slices(path);
1540
1541         r = cg_path_decode_unit(e, &unit);
1542         if (r < 0)
1543                 return r;
1544
1545         /* We skipped over the slices, don't accept any now */
1546         if (endswith(unit, ".slice")) {
1547                 free(unit);
1548                 return -ENXIO;
1549         }
1550
1551         *ret = unit;
1552         return 0;
1553 }
1554
1555 int cg_pid_get_unit(pid_t pid, char **unit) {
1556         _cleanup_free_ char *cgroup = NULL;
1557         int r;
1558
1559         assert(unit);
1560
1561         r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1562         if (r < 0)
1563                 return r;
1564
1565         return cg_path_get_unit(cgroup, unit);
1566 }
1567
1568 /**
1569  * Skip session-*.scope, but require it to be there.
1570  */
1571 static const char *skip_session(const char *p) {
1572         size_t n;
1573
1574         if (isempty(p))
1575                 return NULL;
1576
1577         p += strspn(p, "/");
1578
1579         n = strcspn(p, "/");
1580         if (n < strlen("session-x.scope"))
1581                 return NULL;
1582
1583         if (memcmp(p, "session-", 8) == 0 && memcmp(p + n - 6, ".scope", 6) == 0) {
1584                 char buf[n - 8 - 6 + 1];
1585
1586                 memcpy(buf, p + 8, n - 8 - 6);
1587                 buf[n - 8 - 6] = 0;
1588
1589                 /* Note that session scopes never need unescaping,
1590                  * since they cannot conflict with the kernel's own
1591                  * names, hence we don't need to call cg_unescape()
1592                  * here. */
1593
1594                 if (!session_id_valid(buf))
1595                         return false;
1596
1597                 p += n;
1598                 p += strspn(p, "/");
1599                 return p;
1600         }
1601
1602         return NULL;
1603 }
1604
1605 /**
1606  * Skip user@*.service, but require it to be there.
1607  */
1608 static const char *skip_user_manager(const char *p) {
1609         size_t n;
1610
1611         if (isempty(p))
1612                 return NULL;
1613
1614         p += strspn(p, "/");
1615
1616         n = strcspn(p, "/");
1617         if (n < strlen("user@x.service"))
1618                 return NULL;
1619
1620         if (memcmp(p, "user@", 5) == 0 && memcmp(p + n - 8, ".service", 8) == 0) {
1621                 char buf[n - 5 - 8 + 1];
1622
1623                 memcpy(buf, p + 5, n - 5 - 8);
1624                 buf[n - 5 - 8] = 0;
1625
1626                 /* Note that user manager services never need unescaping,
1627                  * since they cannot conflict with the kernel's own
1628                  * names, hence we don't need to call cg_unescape()
1629                  * here. */
1630
1631                 if (parse_uid(buf, NULL) < 0)
1632                         return NULL;
1633
1634                 p += n;
1635                 p += strspn(p, "/");
1636
1637                 return p;
1638         }
1639
1640         return NULL;
1641 }
1642
1643 static const char *skip_user_prefix(const char *path) {
1644         const char *e, *t;
1645
1646         assert(path);
1647
1648         /* Skip slices, if there are any */
1649         e = skip_slices(path);
1650
1651         /* Skip the user manager, if it's in the path now... */
1652         t = skip_user_manager(e);
1653         if (t)
1654                 return t;
1655
1656         /* Alternatively skip the user session if it is in the path... */
1657         return skip_session(e);
1658 }
1659
1660 int cg_path_get_user_unit(const char *path, char **ret) {
1661         const char *t;
1662
1663         assert(path);
1664         assert(ret);
1665
1666         t = skip_user_prefix(path);
1667         if (!t)
1668                 return -ENXIO;
1669
1670         /* And from here on it looks pretty much the same as for a
1671          * system unit, hence let's use the same parser from here
1672          * on. */
1673         return cg_path_get_unit(t, ret);
1674 }
1675
1676 int cg_pid_get_user_unit(pid_t pid, char **unit) {
1677         _cleanup_free_ char *cgroup = NULL;
1678         int r;
1679
1680         assert(unit);
1681
1682         r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1683         if (r < 0)
1684                 return r;
1685
1686         return cg_path_get_user_unit(cgroup, unit);
1687 }
1688
1689 int cg_path_get_machine_name(const char *path, char **machine) {
1690         _cleanup_free_ char *u = NULL;
1691         const char *sl;
1692         int r;
1693
1694         r = cg_path_get_unit(path, &u);
1695         if (r < 0)
1696                 return r;
1697
1698         sl = strjoina("/run/systemd/machines/unit:", u);
1699         return readlink_malloc(sl, machine);
1700 }
1701
1702 int cg_pid_get_machine_name(pid_t pid, char **machine) {
1703         _cleanup_free_ char *cgroup = NULL;
1704         int r;
1705
1706         assert(machine);
1707
1708         r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1709         if (r < 0)
1710                 return r;
1711
1712         return cg_path_get_machine_name(cgroup, machine);
1713 }
1714
1715 int cg_path_get_session(const char *path, char **session) {
1716         _cleanup_free_ char *unit = NULL;
1717         char *start, *end;
1718         int r;
1719
1720         assert(path);
1721
1722         r = cg_path_get_unit(path, &unit);
1723         if (r < 0)
1724                 return r;
1725
1726         start = startswith(unit, "session-");
1727         if (!start)
1728                 return -ENXIO;
1729         end = endswith(start, ".scope");
1730         if (!end)
1731                 return -ENXIO;
1732
1733         *end = 0;
1734         if (!session_id_valid(start))
1735                 return -ENXIO;
1736
1737         if (session) {
1738                 char *rr;
1739
1740                 rr = strdup(start);
1741                 if (!rr)
1742                         return -ENOMEM;
1743
1744                 *session = rr;
1745         }
1746
1747         return 0;
1748 }
1749
1750 int cg_pid_get_session(pid_t pid, char **session) {
1751         _cleanup_free_ char *cgroup = NULL;
1752         int r;
1753
1754         r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1755         if (r < 0)
1756                 return r;
1757
1758         return cg_path_get_session(cgroup, session);
1759 }
1760
1761 int cg_path_get_owner_uid(const char *path, uid_t *uid) {
1762         _cleanup_free_ char *slice = NULL;
1763         char *start, *end;
1764         int r;
1765
1766         assert(path);
1767
1768         r = cg_path_get_slice(path, &slice);
1769         if (r < 0)
1770                 return r;
1771
1772         start = startswith(slice, "user-");
1773         if (!start)
1774                 return -ENXIO;
1775         end = endswith(start, ".slice");
1776         if (!end)
1777                 return -ENXIO;
1778
1779         *end = 0;
1780         if (parse_uid(start, uid) < 0)
1781                 return -ENXIO;
1782
1783         return 0;
1784 }
1785
1786 int cg_pid_get_owner_uid(pid_t pid, uid_t *uid) {
1787         _cleanup_free_ char *cgroup = NULL;
1788         int r;
1789
1790         r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1791         if (r < 0)
1792                 return r;
1793
1794         return cg_path_get_owner_uid(cgroup, uid);
1795 }
1796
1797 int cg_path_get_slice(const char *p, char **slice) {
1798         const char *e = NULL;
1799
1800         assert(p);
1801         assert(slice);
1802
1803         /* Finds the right-most slice unit from the beginning, but
1804          * stops before we come to the first non-slice unit. */
1805
1806         for (;;) {
1807                 size_t n;
1808
1809                 p += strspn(p, "/");
1810
1811                 n = strcspn(p, "/");
1812                 if (!valid_slice_name(p, n)) {
1813
1814                         if (!e) {
1815                                 char *s;
1816
1817                                 s = strdup(SPECIAL_ROOT_SLICE);
1818                                 if (!s)
1819                                         return -ENOMEM;
1820
1821                                 *slice = s;
1822                                 return 0;
1823                         }
1824
1825                         return cg_path_decode_unit(e, slice);
1826                 }
1827
1828                 e = p;
1829                 p += n;
1830         }
1831 }
1832
1833 int cg_pid_get_slice(pid_t pid, char **slice) {
1834         _cleanup_free_ char *cgroup = NULL;
1835         int r;
1836
1837         assert(slice);
1838
1839         r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1840         if (r < 0)
1841                 return r;
1842
1843         return cg_path_get_slice(cgroup, slice);
1844 }
1845
1846 int cg_path_get_user_slice(const char *p, char **slice) {
1847         const char *t;
1848         assert(p);
1849         assert(slice);
1850
1851         t = skip_user_prefix(p);
1852         if (!t)
1853                 return -ENXIO;
1854
1855         /* And now it looks pretty much the same as for a system
1856          * slice, so let's just use the same parser from here on. */
1857         return cg_path_get_slice(t, slice);
1858 }
1859
1860 int cg_pid_get_user_slice(pid_t pid, char **slice) {
1861         _cleanup_free_ char *cgroup = NULL;
1862         int r;
1863
1864         assert(slice);
1865
1866         r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1867         if (r < 0)
1868                 return r;
1869
1870         return cg_path_get_user_slice(cgroup, slice);
1871 }
1872
1873 char *cg_escape(const char *p) {
1874         bool need_prefix = false;
1875
1876         /* This implements very minimal escaping for names to be used
1877          * as file names in the cgroup tree: any name which might
1878          * conflict with a kernel name or is prefixed with '_' is
1879          * prefixed with a '_'. That way, when reading cgroup names it
1880          * is sufficient to remove a single prefixing underscore if
1881          * there is one. */
1882
1883         /* The return value of this function (unlike cg_unescape())
1884          * needs free()! */
1885
1886         if (IN_SET(p[0], 0, '_', '.') ||
1887             streq(p, "notify_on_release") ||
1888             streq(p, "release_agent") ||
1889             streq(p, "tasks") ||
1890             startswith(p, "cgroup."))
1891                 need_prefix = true;
1892         else {
1893                 const char *dot;
1894
1895                 dot = strrchr(p, '.');
1896                 if (dot) {
1897                         CGroupController c;
1898                         size_t l = dot - p;
1899
1900                         for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
1901                                 const char *n;
1902
1903                                 n = cgroup_controller_to_string(c);
1904
1905                                 if (l != strlen(n))
1906                                         continue;
1907
1908                                 if (memcmp(p, n, l) != 0)
1909                                         continue;
1910
1911                                 need_prefix = true;
1912                                 break;
1913                         }
1914                 }
1915         }
1916
1917         if (need_prefix)
1918                 return strappend("_", p);
1919
1920         return strdup(p);
1921 }
1922
1923 char *cg_unescape(const char *p) {
1924         assert(p);
1925
1926         /* The return value of this function (unlike cg_escape())
1927          * doesn't need free()! */
1928
1929         if (p[0] == '_')
1930                 return (char*) p+1;
1931
1932         return (char*) p;
1933 }
1934
1935 #define CONTROLLER_VALID                        \
1936         DIGITS LETTERS                          \
1937         "_"
1938
1939 bool cg_controller_is_valid(const char *p) {
1940         const char *t, *s;
1941
1942         if (!p)
1943                 return false;
1944
1945         if (streq(p, SYSTEMD_CGROUP_CONTROLLER))
1946                 return true;
1947
1948         s = startswith(p, "name=");
1949         if (s)
1950                 p = s;
1951
1952         if (IN_SET(*p, 0, '_'))
1953                 return false;
1954
1955         for (t = p; *t; t++)
1956                 if (!strchr(CONTROLLER_VALID, *t))
1957                         return false;
1958
1959         if (t - p > FILENAME_MAX)
1960                 return false;
1961
1962         return true;
1963 }
1964
1965 int cg_slice_to_path(const char *unit, char **ret) {
1966         _cleanup_free_ char *p = NULL, *s = NULL, *e = NULL;
1967         const char *dash;
1968         int r;
1969
1970         assert(unit);
1971         assert(ret);
1972
1973         if (streq(unit, SPECIAL_ROOT_SLICE)) {
1974                 char *x;
1975
1976                 x = strdup("");
1977                 if (!x)
1978                         return -ENOMEM;
1979                 *ret = x;
1980                 return 0;
1981         }
1982
1983         if (!unit_name_is_valid(unit, UNIT_NAME_PLAIN))
1984                 return -EINVAL;
1985
1986         if (!endswith(unit, ".slice"))
1987                 return -EINVAL;
1988
1989         r = unit_name_to_prefix(unit, &p);
1990         if (r < 0)
1991                 return r;
1992
1993         dash = strchr(p, '-');
1994
1995         /* Don't allow initial dashes */
1996         if (dash == p)
1997                 return -EINVAL;
1998
1999         while (dash) {
2000                 _cleanup_free_ char *escaped = NULL;
2001                 char n[dash - p + sizeof(".slice")];
2002
2003                 /* Don't allow trailing or double dashes */
2004                 if (IN_SET(dash[1], 0, '-'))
2005                         return -EINVAL;
2006
2007                 strcpy(stpncpy(n, p, dash - p), ".slice");
2008                 if (!unit_name_is_valid(n, UNIT_NAME_PLAIN))
2009                         return -EINVAL;
2010
2011                 escaped = cg_escape(n);
2012                 if (!escaped)
2013                         return -ENOMEM;
2014
2015                 if (!strextend(&s, escaped, "/", NULL))
2016                         return -ENOMEM;
2017
2018                 dash = strchr(dash+1, '-');
2019         }
2020
2021         e = cg_escape(unit);
2022         if (!e)
2023                 return -ENOMEM;
2024
2025         if (!strextend(&s, e, NULL))
2026                 return -ENOMEM;
2027
2028         *ret = s;
2029         s = NULL;
2030
2031         return 0;
2032 }
2033
2034 int cg_set_attribute(const char *controller, const char *path, const char *attribute, const char *value) {
2035         _cleanup_free_ char *p = NULL;
2036         int r;
2037
2038         r = cg_get_path(controller, path, attribute, &p);
2039         if (r < 0)
2040                 return r;
2041
2042         return write_string_file(p, value, 0);
2043 }
2044
2045 int cg_get_attribute(const char *controller, const char *path, const char *attribute, char **ret) {
2046         _cleanup_free_ char *p = NULL;
2047         int r;
2048
2049         r = cg_get_path(controller, path, attribute, &p);
2050         if (r < 0)
2051                 return r;
2052
2053         return read_one_line_file(p, ret);
2054 }
2055
2056 int cg_get_keyed_attribute(const char *controller, const char *path, const char *attribute, const char **keys, char **values) {
2057         _cleanup_free_ char *filename = NULL, *content = NULL;
2058         char *line, *p;
2059         int i, r;
2060
2061         for (i = 0; keys[i]; i++)
2062                 values[i] = NULL;
2063
2064         r = cg_get_path(controller, path, attribute, &filename);
2065         if (r < 0)
2066                 return r;
2067
2068         r = read_full_file(filename, &content, NULL);
2069         if (r < 0)
2070                 return r;
2071
2072         p = content;
2073         while ((line = strsep(&p, "\n"))) {
2074                 char *key;
2075
2076                 key = strsep(&line, " ");
2077
2078                 for (i = 0; keys[i]; i++) {
2079                         if (streq(key, keys[i])) {
2080                                 values[i] = strdup(line);
2081                                 break;
2082                         }
2083                 }
2084         }
2085
2086         for (i = 0; keys[i]; i++) {
2087                 if (!values[i]) {
2088                         for (i = 0; keys[i]; i++) {
2089                                 free(values[i]);
2090                                 values[i] = NULL;
2091                         }
2092                         return -ENOENT;
2093                 }
2094         }
2095
2096         return 0;
2097 }
2098
2099 int cg_create_everywhere(CGroupMask supported, CGroupMask mask, const char *path) {
2100         CGroupController c;
2101         int r;
2102
2103         /* This one will create a cgroup in our private tree, but also
2104          * duplicate it in the trees specified in mask, and remove it
2105          * in all others */
2106
2107         /* First create the cgroup in our own hierarchy. */
2108         r = cg_create(SYSTEMD_CGROUP_CONTROLLER, path);
2109         if (r < 0)
2110                 return r;
2111
2112         /* If we are in the unified hierarchy, we are done now */
2113         r = cg_all_unified();
2114         if (r < 0)
2115                 return r;
2116         if (r > 0)
2117                 return 0;
2118
2119         /* Otherwise, do the same in the other hierarchies */
2120         for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2121                 CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
2122                 const char *n;
2123
2124                 n = cgroup_controller_to_string(c);
2125
2126                 if (mask & bit)
2127                         (void) cg_create(n, path);
2128                 else if (supported & bit)
2129                         (void) cg_trim(n, path, true);
2130         }
2131
2132         return 0;
2133 }
2134
2135 int cg_attach_everywhere(CGroupMask supported, const char *path, pid_t pid, cg_migrate_callback_t path_callback, void *userdata) {
2136         CGroupController c;
2137         int r;
2138
2139         r = cg_attach(SYSTEMD_CGROUP_CONTROLLER, path, pid);
2140         if (r < 0)
2141                 return r;
2142
2143         r = cg_all_unified();
2144         if (r < 0)
2145                 return r;
2146         if (r > 0)
2147                 return 0;
2148
2149         for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2150                 CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
2151                 const char *p = NULL;
2152
2153                 if (!(supported & bit))
2154                         continue;
2155
2156                 if (path_callback)
2157                         p = path_callback(bit, userdata);
2158
2159                 if (!p)
2160                         p = path;
2161
2162                 (void) cg_attach_fallback(cgroup_controller_to_string(c), p, pid);
2163         }
2164
2165         return 0;
2166 }
2167
2168 int cg_attach_many_everywhere(CGroupMask supported, const char *path, Set* pids, cg_migrate_callback_t path_callback, void *userdata) {
2169         Iterator i;
2170         void *pidp;
2171         int r = 0;
2172
2173         SET_FOREACH(pidp, pids, i) {
2174                 pid_t pid = PTR_TO_PID(pidp);
2175                 int q;
2176
2177                 q = cg_attach_everywhere(supported, path, pid, path_callback, userdata);
2178                 if (q < 0 && r >= 0)
2179                         r = q;
2180         }
2181
2182         return r;
2183 }
2184
2185 int cg_migrate_everywhere(CGroupMask supported, const char *from, const char *to, cg_migrate_callback_t to_callback, void *userdata) {
2186         CGroupController c;
2187         int r = 0, q;
2188
2189         if (!path_equal(from, to))  {
2190                 r = cg_migrate_recursive(SYSTEMD_CGROUP_CONTROLLER, from, SYSTEMD_CGROUP_CONTROLLER, to, CGROUP_REMOVE);
2191                 if (r < 0)
2192                         return r;
2193         }
2194
2195         q = cg_all_unified();
2196         if (q < 0)
2197                 return q;
2198         if (q > 0)
2199                 return r;
2200
2201         for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2202                 CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
2203                 const char *p = NULL;
2204
2205                 if (!(supported & bit))
2206                         continue;
2207
2208                 if (to_callback)
2209                         p = to_callback(bit, userdata);
2210
2211                 if (!p)
2212                         p = to;
2213
2214                 (void) cg_migrate_recursive_fallback(SYSTEMD_CGROUP_CONTROLLER, to, cgroup_controller_to_string(c), p, 0);
2215         }
2216
2217         return 0;
2218 }
2219
2220 int cg_trim_everywhere(CGroupMask supported, const char *path, bool delete_root) {
2221         CGroupController c;
2222         int r, q;
2223
2224         r = cg_trim(SYSTEMD_CGROUP_CONTROLLER, path, delete_root);
2225         if (r < 0)
2226                 return r;
2227
2228         q = cg_all_unified();
2229         if (q < 0)
2230                 return q;
2231         if (q > 0)
2232                 return r;
2233
2234         for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2235                 CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
2236
2237                 if (!(supported & bit))
2238                         continue;
2239
2240                 (void) cg_trim(cgroup_controller_to_string(c), path, delete_root);
2241         }
2242
2243         return 0;
2244 }
2245
2246 int cg_mask_to_string(CGroupMask mask, char **ret) {
2247         const char *controllers[_CGROUP_CONTROLLER_MAX + 1];
2248         CGroupController c;
2249         int i = 0;
2250         char *s;
2251
2252         assert(ret);
2253
2254         if (mask == 0) {
2255                 *ret = NULL;
2256                 return 0;
2257         }
2258
2259         for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2260
2261                 if (!(mask & CGROUP_CONTROLLER_TO_MASK(c)))
2262                         continue;
2263
2264                 controllers[i++] = cgroup_controller_to_string(c);
2265                 controllers[i] = NULL;
2266         }
2267
2268         s = strv_join((char **)controllers, NULL);
2269         if (!s)
2270                 return -ENOMEM;
2271
2272         *ret = s;
2273         return 0;
2274 }
2275
2276 int cg_mask_from_string(const char *value, CGroupMask *mask) {
2277         assert(mask);
2278         assert(value);
2279
2280         for (;;) {
2281                 _cleanup_free_ char *n = NULL;
2282                 CGroupController v;
2283                 int r;
2284
2285                 r = extract_first_word(&value, &n, NULL, 0);
2286                 if (r < 0)
2287                         return r;
2288                 if (r == 0)
2289                         break;
2290
2291                 v = cgroup_controller_from_string(n);
2292                 if (v < 0)
2293                         continue;
2294
2295                 *mask |= CGROUP_CONTROLLER_TO_MASK(v);
2296         }
2297         return 0;
2298 }
2299
2300 int cg_mask_supported(CGroupMask *ret) {
2301         CGroupMask mask = 0;
2302         int r;
2303
2304         /* Determines the mask of supported cgroup controllers. Only
2305          * includes controllers we can make sense of and that are
2306          * actually accessible. */
2307
2308         r = cg_all_unified();
2309         if (r < 0)
2310                 return r;
2311         if (r > 0) {
2312                 _cleanup_free_ char *root = NULL, *controllers = NULL, *path = NULL;
2313
2314                 /* In the unified hierarchy we can read the supported
2315                  * and accessible controllers from a the top-level
2316                  * cgroup attribute */
2317
2318                 r = cg_get_root_path(&root);
2319                 if (r < 0)
2320                         return r;
2321
2322                 r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, root, "cgroup.controllers", &path);
2323                 if (r < 0)
2324                         return r;
2325
2326                 r = read_one_line_file(path, &controllers);
2327                 if (r < 0)
2328                         return r;
2329
2330                 r = cg_mask_from_string(controllers, &mask);
2331                 if (r < 0)
2332                         return r;
2333
2334                 /* Currently, we support the cpu, memory, io and pids
2335                  * controller in the unified hierarchy, mask
2336                  * everything else off. */
2337                 mask &= CGROUP_MASK_CPU | CGROUP_MASK_MEMORY | CGROUP_MASK_IO | CGROUP_MASK_PIDS;
2338
2339         } else {
2340                 CGroupController c;
2341
2342                 /* In the legacy hierarchy, we check whether which
2343                  * hierarchies are mounted. */
2344
2345                 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2346                         const char *n;
2347
2348                         n = cgroup_controller_to_string(c);
2349                         if (controller_is_accessible(n) >= 0)
2350                                 mask |= CGROUP_CONTROLLER_TO_MASK(c);
2351                 }
2352         }
2353
2354         *ret = mask;
2355         return 0;
2356 }
2357
2358 int cg_kernel_controllers(Set *controllers) {
2359         _cleanup_fclose_ FILE *f = NULL;
2360         int r;
2361
2362         assert(controllers);
2363
2364         /* Determines the full list of kernel-known controllers. Might
2365          * include controllers we don't actually support, arbitrary
2366          * named hierarchies and controllers that aren't currently
2367          * accessible (because not mounted). */
2368
2369         f = fopen("/proc/cgroups", "re");
2370         if (!f) {
2371                 if (errno == ENOENT)
2372                         return 0;
2373                 return -errno;
2374         }
2375
2376         /* Ignore the header line */
2377         (void) read_line(f, (size_t) -1, NULL);
2378
2379         for (;;) {
2380                 char *controller;
2381                 int enabled = 0;
2382
2383                 errno = 0;
2384                 if (fscanf(f, "%ms %*i %*i %i", &controller, &enabled) != 2) {
2385
2386                         if (feof(f))
2387                                 break;
2388
2389                         if (ferror(f) && errno > 0)
2390                                 return -errno;
2391
2392                         return -EBADMSG;
2393                 }
2394
2395                 if (!enabled) {
2396                         free(controller);
2397                         continue;
2398                 }
2399
2400                 if (!cg_controller_is_valid(controller)) {
2401                         free(controller);
2402                         return -EBADMSG;
2403                 }
2404
2405                 r = set_consume(controllers, controller);
2406                 if (r < 0)
2407                         return r;
2408         }
2409
2410         return 0;
2411 }
2412
2413 static thread_local CGroupUnified unified_cache = CGROUP_UNIFIED_UNKNOWN;
2414
2415 /* The hybrid mode was initially implemented in v232 and simply mounted cgroup v2 on /sys/fs/cgroup/systemd.  This
2416  * unfortunately broke other tools (such as docker) which expected the v1 "name=systemd" hierarchy on
2417  * /sys/fs/cgroup/systemd.  From v233 and on, the hybrid mode mountnbs v2 on /sys/fs/cgroup/unified and maintains
2418  * "name=systemd" hierarchy on /sys/fs/cgroup/systemd for compatibility with other tools.
2419  *
2420  * To keep live upgrade working, we detect and support v232 layout.  When v232 layout is detected, to keep cgroup v2
2421  * process management but disable the compat dual layout, we return %true on
2422  * cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) and %false on cg_hybrid_unified().
2423  */
2424 static thread_local bool unified_systemd_v232;
2425
2426 static int cg_unified_update(void) {
2427
2428         struct statfs fs;
2429
2430         /* Checks if we support the unified hierarchy. Returns an
2431          * error when the cgroup hierarchies aren't mounted yet or we
2432          * have any other trouble determining if the unified hierarchy
2433          * is supported. */
2434
2435         if (unified_cache >= CGROUP_UNIFIED_NONE)
2436                 return 0;
2437
2438         if (statfs("/sys/fs/cgroup/", &fs) < 0)
2439                 return -errno;
2440
2441         if (F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC))
2442                 unified_cache = CGROUP_UNIFIED_ALL;
2443         else if (F_TYPE_EQUAL(fs.f_type, TMPFS_MAGIC)) {
2444                 if (statfs("/sys/fs/cgroup/unified/", &fs) == 0 &&
2445                     F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC)) {
2446                         unified_cache = CGROUP_UNIFIED_SYSTEMD;
2447                         unified_systemd_v232 = false;
2448                 } else if (statfs("/sys/fs/cgroup/systemd/", &fs) == 0 &&
2449                            F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC)) {
2450                         unified_cache = CGROUP_UNIFIED_SYSTEMD;
2451                         unified_systemd_v232 = true;
2452                 } else {
2453                         if (statfs("/sys/fs/cgroup/systemd/", &fs) < 0)
2454                                 return -errno;
2455                         if (!F_TYPE_EQUAL(fs.f_type, CGROUP_SUPER_MAGIC))
2456                                 return -ENOMEDIUM;
2457                         unified_cache = CGROUP_UNIFIED_NONE;
2458                 }
2459         } else {
2460                 log_debug("Unknown filesystem type %llx mounted on /sys/fs/cgroup.",
2461                           (unsigned long long) fs.f_type);
2462                 return -ENOMEDIUM;
2463         }
2464
2465         return 0;
2466 }
2467
2468 int cg_unified_controller(const char *controller) {
2469         int r;
2470
2471         r = cg_unified_update();
2472         if (r < 0)
2473                 return r;
2474
2475         if (unified_cache == CGROUP_UNIFIED_NONE)
2476                 return false;
2477
2478         if (unified_cache >= CGROUP_UNIFIED_ALL)
2479                 return true;
2480
2481         return streq_ptr(controller, SYSTEMD_CGROUP_CONTROLLER);
2482 }
2483
2484 int cg_all_unified(void) {
2485         int r;
2486
2487         r = cg_unified_update();
2488         if (r < 0)
2489                 return r;
2490
2491         return unified_cache >= CGROUP_UNIFIED_ALL;
2492 }
2493
2494 int cg_hybrid_unified(void) {
2495         int r;
2496
2497         r = cg_unified_update();
2498         if (r < 0)
2499                 return r;
2500
2501         return unified_cache == CGROUP_UNIFIED_SYSTEMD && !unified_systemd_v232;
2502 }
2503
2504 int cg_unified_flush(void) {
2505         unified_cache = CGROUP_UNIFIED_UNKNOWN;
2506
2507         return cg_unified_update();
2508 }
2509
2510 int cg_enable_everywhere(CGroupMask supported, CGroupMask mask, const char *p) {
2511         _cleanup_free_ char *fs = NULL;
2512         CGroupController c;
2513         int r;
2514
2515         assert(p);
2516
2517         if (supported == 0)
2518                 return 0;
2519
2520         r = cg_all_unified();
2521         if (r < 0)
2522                 return r;
2523         if (r == 0) /* on the legacy hiearchy there's no joining of controllers defined */
2524                 return 0;
2525
2526         r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, p, "cgroup.subtree_control", &fs);
2527         if (r < 0)
2528                 return r;
2529
2530         for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
2531                 CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
2532                 const char *n;
2533
2534                 if (!(supported & bit))
2535                         continue;
2536
2537                 n = cgroup_controller_to_string(c);
2538                 {
2539                         char s[1 + strlen(n) + 1];
2540
2541                         s[0] = mask & bit ? '+' : '-';
2542                         strcpy(s + 1, n);
2543
2544                         r = write_string_file(fs, s, 0);
2545                         if (r < 0)
2546                                 log_debug_errno(r, "Failed to enable controller %s for %s (%s): %m", n, p, fs);
2547                 }
2548         }
2549
2550         return 0;
2551 }
2552
2553 bool cg_is_unified_wanted(void) {
2554         static thread_local int wanted = -1;
2555         int r;
2556         bool b;
2557         const bool is_default = DEFAULT_HIERARCHY == CGROUP_UNIFIED_ALL;
2558
2559         /* If we have a cached value, return that. */
2560         if (wanted >= 0)
2561                 return wanted;
2562
2563         /* If the hierarchy is already mounted, then follow whatever
2564          * was chosen for it. */
2565         if (cg_unified_flush() >= 0)
2566                 return (wanted = unified_cache >= CGROUP_UNIFIED_ALL);
2567
2568         /* Otherwise, let's see what the kernel command line has to say.
2569          * Since checking is expensive, cache a non-error result. */
2570         r = proc_cmdline_get_bool("systemd.unified_cgroup_hierarchy", &b);
2571
2572         return (wanted = r > 0 ? b : is_default);
2573 }
2574
2575 bool cg_is_legacy_wanted(void) {
2576         static thread_local int wanted = -1;
2577
2578         /* If we have a cached value, return that. */
2579         if (wanted >= 0)
2580                 return wanted;
2581
2582         /* Check if we have cgroups2 already mounted. */
2583         if (cg_unified_flush() >= 0 &&
2584             unified_cache == CGROUP_UNIFIED_ALL)
2585                 return (wanted = false);
2586
2587         /* Otherwise, assume that at least partial legacy is wanted,
2588          * since cgroups2 should already be mounted at this point. */
2589         return (wanted = true);
2590 }
2591
2592 bool cg_is_hybrid_wanted(void) {
2593         static thread_local int wanted = -1;
2594         int r;
2595         bool b;
2596         const bool is_default = DEFAULT_HIERARCHY >= CGROUP_UNIFIED_SYSTEMD;
2597         /* We default to true if the default is "hybrid", obviously,
2598          * but also when the default is "unified", because if we get
2599          * called, it means that unified hierarchy was not mounted. */
2600
2601         /* If we have a cached value, return that. */
2602         if (wanted >= 0)
2603                 return wanted;
2604
2605         /* If the hierarchy is already mounted, then follow whatever
2606          * was chosen for it. */
2607         if (cg_unified_flush() >= 0 &&
2608             unified_cache == CGROUP_UNIFIED_ALL)
2609                 return (wanted = false);
2610
2611         /* Otherwise, let's see what the kernel command line has to say.
2612          * Since checking is expensive, cache a non-error result. */
2613         r = proc_cmdline_get_bool("systemd.legacy_systemd_cgroup_controller", &b);
2614
2615         /* The meaning of the kernel option is reversed wrt. to the return value
2616          * of this function, hence the negation. */
2617         return (wanted = r > 0 ? !b : is_default);
2618 }
2619
2620 int cg_weight_parse(const char *s, uint64_t *ret) {
2621         uint64_t u;
2622         int r;
2623
2624         if (isempty(s)) {
2625                 *ret = CGROUP_WEIGHT_INVALID;
2626                 return 0;
2627         }
2628
2629         r = safe_atou64(s, &u);
2630         if (r < 0)
2631                 return r;
2632
2633         if (u < CGROUP_WEIGHT_MIN || u > CGROUP_WEIGHT_MAX)
2634                 return -ERANGE;
2635
2636         *ret = u;
2637         return 0;
2638 }
2639
2640 const uint64_t cgroup_io_limit_defaults[_CGROUP_IO_LIMIT_TYPE_MAX] = {
2641         [CGROUP_IO_RBPS_MAX]    = CGROUP_LIMIT_MAX,
2642         [CGROUP_IO_WBPS_MAX]    = CGROUP_LIMIT_MAX,
2643         [CGROUP_IO_RIOPS_MAX]   = CGROUP_LIMIT_MAX,
2644         [CGROUP_IO_WIOPS_MAX]   = CGROUP_LIMIT_MAX,
2645 };
2646
2647 static const char* const cgroup_io_limit_type_table[_CGROUP_IO_LIMIT_TYPE_MAX] = {
2648         [CGROUP_IO_RBPS_MAX]    = "IOReadBandwidthMax",
2649         [CGROUP_IO_WBPS_MAX]    = "IOWriteBandwidthMax",
2650         [CGROUP_IO_RIOPS_MAX]   = "IOReadIOPSMax",
2651         [CGROUP_IO_WIOPS_MAX]   = "IOWriteIOPSMax",
2652 };
2653
2654 DEFINE_STRING_TABLE_LOOKUP(cgroup_io_limit_type, CGroupIOLimitType);
2655
2656 int cg_cpu_shares_parse(const char *s, uint64_t *ret) {
2657         uint64_t u;
2658         int r;
2659
2660         if (isempty(s)) {
2661                 *ret = CGROUP_CPU_SHARES_INVALID;
2662                 return 0;
2663         }
2664
2665         r = safe_atou64(s, &u);
2666         if (r < 0)
2667                 return r;
2668
2669         if (u < CGROUP_CPU_SHARES_MIN || u > CGROUP_CPU_SHARES_MAX)
2670                 return -ERANGE;
2671
2672         *ret = u;
2673         return 0;
2674 }
2675
2676 int cg_blkio_weight_parse(const char *s, uint64_t *ret) {
2677         uint64_t u;
2678         int r;
2679
2680         if (isempty(s)) {
2681                 *ret = CGROUP_BLKIO_WEIGHT_INVALID;
2682                 return 0;
2683         }
2684
2685         r = safe_atou64(s, &u);
2686         if (r < 0)
2687                 return r;
2688
2689         if (u < CGROUP_BLKIO_WEIGHT_MIN || u > CGROUP_BLKIO_WEIGHT_MAX)
2690                 return -ERANGE;
2691
2692         *ret = u;
2693         return 0;
2694 }
2695
2696 bool is_cgroup_fs(const struct statfs *s) {
2697         return is_fs_type(s, CGROUP_SUPER_MAGIC) ||
2698                is_fs_type(s, CGROUP2_SUPER_MAGIC);
2699 }
2700
2701 bool fd_is_cgroup_fs(int fd) {
2702         struct statfs s;
2703
2704         if (fstatfs(fd, &s) < 0)
2705                 return -errno;
2706
2707         return is_cgroup_fs(&s);
2708 }
2709
2710 static const char *cgroup_controller_table[_CGROUP_CONTROLLER_MAX] = {
2711         [CGROUP_CONTROLLER_CPU] = "cpu",
2712         [CGROUP_CONTROLLER_CPUACCT] = "cpuacct",
2713         [CGROUP_CONTROLLER_IO] = "io",
2714         [CGROUP_CONTROLLER_BLKIO] = "blkio",
2715         [CGROUP_CONTROLLER_MEMORY] = "memory",
2716         [CGROUP_CONTROLLER_DEVICES] = "devices",
2717         [CGROUP_CONTROLLER_PIDS] = "pids",
2718 };
2719
2720 DEFINE_STRING_TABLE_LOOKUP(cgroup_controller, CGroupController);