summaryrefslogtreecommitdiffstats
path: root/src/shared/cgroup-setup.c
blob: d415d13d41e6291e453e29a10dda1f8616e44ca9 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
/* SPDX-License-Identifier: LGPL-2.1-or-later */

#include <unistd.h>

#include "cgroup-setup.h"
#include "cgroup-util.h"
#include "errno-util.h"
#include "fd-util.h"
#include "fileio.h"
#include "fs-util.h"
#include "missing_threads.h"
#include "mkdir.h"
#include "parse-util.h"
#include "path-util.h"
#include "proc-cmdline.h"
#include "process-util.h"
#include "recurse-dir.h"
#include "stdio-util.h"
#include "string-util.h"
#include "user-util.h"
#include "virt.h"

static int cg_any_controller_used_for_v1(void) {
        _cleanup_free_ char *buf = NULL;
        _cleanup_strv_free_ char **lines = NULL;
        int r;

        r = read_full_virtual_file("/proc/cgroups", &buf, NULL);
        if (r < 0)
                return log_debug_errno(r, "Could not read /proc/cgroups, ignoring: %m");

        r = strv_split_newlines_full(&lines, buf, 0);
        if (r < 0)
                return r;

        /* The intention of this is to check if the fully unified cgroup tree setup is possible, meaning all
         * enabled kernel cgroup controllers are currently not in use by cgroup1.  For reference:
         * https://systemd.io/CGROUP_DELEGATION/#three-different-tree-setups-
         *
         * Note that this is typically only useful to check inside a container where we don't know what
         * cgroup tree setup is in use by the host; if the host is using legacy or hybrid, we can't use
         * unified since some or all controllers would be missing. This is not the best way to detect this,
         * as whatever container manager created our container should have mounted /sys/fs/cgroup
         * appropriately, but in case that wasn't done, we try to detect if it's possible for us to use
         * unified cgroups. */
        STRV_FOREACH(line, lines) {
                _cleanup_free_ char *name = NULL, *hierarchy_id = NULL, *num = NULL, *enabled = NULL;

                /* Skip header line */
                if (startswith(*line, "#"))
                        continue;

                const char *p = *line;
                r = extract_many_words(&p, NULL, 0, &name, &hierarchy_id, &num, &enabled);
                if (r < 0)
                        return log_debug_errno(r, "Error parsing /proc/cgroups line, ignoring: %m");
                else if (r < 4) {
                        log_debug("Invalid /proc/cgroups line, ignoring.");
                        continue;
                }

                /* Ignore disabled controllers. */
                if (streq(enabled, "0"))
                        continue;

                /* Ignore controllers we don't care about. */
                if (cgroup_controller_from_string(name) < 0)
                        continue;

                /* Since the unified cgroup doesn't use multiple hierarchies, if any controller has a
                 * non-zero hierarchy_id that means it's in use already in a legacy (or hybrid) cgroup v1
                 * hierarchy, and can't be used in a unified cgroup. */
                if (!streq(hierarchy_id, "0")) {
                        log_debug("Cgroup controller %s in use by legacy v1 hierarchy.", name);
                        return 1;
                }
        }

        return 0;
}

bool cg_is_unified_wanted(void) {
        static thread_local int wanted = -1;
        int r;

        /* If we have a cached value, return that. */
        if (wanted >= 0)
                return wanted;

        /* If the hierarchy is already mounted, then follow whatever was chosen for it. */
        r = cg_unified_cached(true);
        if (r >= 0)
                return (wanted = r >= CGROUP_UNIFIED_ALL);

        /* If we were explicitly passed systemd.unified_cgroup_hierarchy, respect that. */
        bool b;
        r = proc_cmdline_get_bool("systemd.unified_cgroup_hierarchy", /* flags = */ 0, &b);
        if (r > 0)
                return (wanted = b);

        /* If we passed cgroup_no_v1=all with no other instructions, it seems highly unlikely that we want to
         * use hybrid or legacy hierarchy. */
        _cleanup_free_ char *c = NULL;
        r = proc_cmdline_get_key("cgroup_no_v1", 0, &c);
        if (r > 0 && streq_ptr(c, "all"))
                return (wanted = true);

        /* If any controller is in use as v1, don't use unified. */
        return (wanted = (cg_any_controller_used_for_v1() <= 0));
}

bool cg_is_legacy_wanted(void) {
        static thread_local int wanted = -1;

        /* If we have a cached value, return that. */
        if (wanted >= 0)
                return wanted;

        /* Check if we have cgroup v2 already mounted. */
        if (cg_unified_cached(true) == CGROUP_UNIFIED_ALL)
                return (wanted = false);

        /* Otherwise, assume that at least partial legacy is wanted,
         * since cgroup v2 should already be mounted at this point. */
        return (wanted = true);
}

bool cg_is_hybrid_wanted(void) {
        static thread_local int wanted = -1;
        int r;

        /* If we have a cached value, return that. */
        if (wanted >= 0)
                return wanted;

        /* If the hierarchy is already mounted, then follow whatever was chosen for it. */
        if (cg_unified_cached(true) == CGROUP_UNIFIED_ALL)
                return (wanted = false);

        /* Otherwise, let's see what the kernel command line has to say.  Since checking is expensive, cache
         * a non-error result.
         * The meaning of the kernel option is reversed wrt. to the return value of this function, hence the
         * negation. */
        bool b;
        r = proc_cmdline_get_bool("systemd.legacy_systemd_cgroup_controller", /* flags = */ 0, &b);
        if (r > 0)
                return (wanted = !b);

        /* The default hierarchy is "unified". But if this is reached, it means that unified hierarchy was
         * not mounted, so return true too. */
        return (wanted = true);
}

bool cg_is_legacy_force_enabled(void) {
        bool force;

        if (!cg_is_legacy_wanted())
                return false;

        /* If in container, we have to follow host's cgroup hierarchy. */
        if (detect_container() > 0)
                return true;

        if (proc_cmdline_get_bool("SYSTEMD_CGROUP_ENABLE_LEGACY_FORCE", /* flags = */ 0, &force) < 0)
                return false;

        return force;
}

int cg_weight_parse(const char *s, uint64_t *ret) {
        uint64_t u;
        int r;

        assert(s);
        assert(ret);

        if (isempty(s)) {
                *ret = CGROUP_WEIGHT_INVALID;
                return 0;
        }

        r = safe_atou64(s, &u);
        if (r < 0)
                return r;

        if (u < CGROUP_WEIGHT_MIN || u > CGROUP_WEIGHT_MAX)
                return -ERANGE;

        *ret = u;
        return 0;
}

int cg_cpu_weight_parse(const char *s, uint64_t *ret) {
        assert(s);
        assert(ret);

        if (streq(s, "idle"))
                return *ret = CGROUP_WEIGHT_IDLE;

        return cg_weight_parse(s, ret);
}

int cg_cpu_shares_parse(const char *s, uint64_t *ret) {
        uint64_t u;
        int r;

        assert(s);
        assert(ret);

        if (isempty(s)) {
                *ret = CGROUP_CPU_SHARES_INVALID;
                return 0;
        }

        r = safe_atou64(s, &u);
        if (r < 0)
                return r;

        if (u < CGROUP_CPU_SHARES_MIN || u > CGROUP_CPU_SHARES_MAX)
                return -ERANGE;

        *ret = u;
        return 0;
}

int cg_blkio_weight_parse(const char *s, uint64_t *ret) {
        uint64_t u;
        int r;

        assert(s);
        assert(ret);

        if (isempty(s)) {
                *ret = CGROUP_BLKIO_WEIGHT_INVALID;
                return 0;
        }

        r = safe_atou64(s, &u);
        if (r < 0)
                return r;

        if (u < CGROUP_BLKIO_WEIGHT_MIN || u > CGROUP_BLKIO_WEIGHT_MAX)
                return -ERANGE;

        *ret = u;
        return 0;
}

static int trim_cb(
                RecurseDirEvent event,
                const char *path,
                int dir_fd,
                int inode_fd,
                const struct dirent *de,
                const struct statx *sx,
                void *userdata) {

        /* Failures to delete inner cgroup we ignore (but debug log in case error code is unexpected) */
        if (event == RECURSE_DIR_LEAVE &&
            de->d_type == DT_DIR &&
            unlinkat(dir_fd, de->d_name, AT_REMOVEDIR) < 0 &&
            !IN_SET(errno, ENOENT, ENOTEMPTY, EBUSY))
                log_debug_errno(errno, "Failed to trim inner cgroup %s, ignoring: %m", path);

        return RECURSE_DIR_CONTINUE;
}

int cg_trim(const char *controller, const char *path, bool delete_root) {
        _cleanup_free_ char *fs = NULL;
        int r, q;

        assert(controller);

        r = cg_get_path(controller, path, NULL, &fs);
        if (r < 0)
                return r;

        r = recurse_dir_at(
                        AT_FDCWD,
                        fs,
                        /* statx_mask = */ 0,
                        /* n_depth_max = */ UINT_MAX,
                        RECURSE_DIR_ENSURE_TYPE,
                        trim_cb,
                        /* userdata = */ NULL);
        if (r == -ENOENT) /* non-existing is the ultimate trimming, hence no error */
                r = 0;
        else if (r < 0)
                log_debug_errno(r, "Failed to trim subcgroups of '%s': %m", path);

        /* If we shall delete the top-level cgroup, then propagate the failure to do so (except if it is
         * already gone anyway). Also, let's debug log about this failure, except if the error code is an
         * expected one. */
        if (delete_root && !empty_or_root(path) &&
            rmdir(fs) < 0 && errno != ENOENT) {
                if (!IN_SET(errno, ENOTEMPTY, EBUSY))
                        log_debug_errno(errno, "Failed to trim cgroup '%s': %m", path);
                RET_GATHER(r, -errno);
        }

        q = cg_hybrid_unified();
        if (q < 0)
                return q;
        if (q > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER))
                (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, delete_root);

        return r;
}

/* Create a cgroup in the hierarchy of controller.
 * Returns 0 if the group already existed, 1 on success, negative otherwise.
 */
int cg_create(const char *controller, const char *path) {
        _cleanup_free_ char *fs = NULL;
        int r;

        assert(controller);

        r = cg_get_path_and_check(controller, path, NULL, &fs);
        if (r < 0)
                return r;

        r = mkdir_parents(fs, 0755);
        if (r < 0)
                return r;

        r = RET_NERRNO(mkdir(fs, 0755));
        if (r == -EEXIST)
                return 0;
        if (r < 0)
                return r;

        r = cg_hybrid_unified();
        if (r < 0)
                return r;
        if (r > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
                r = cg_create(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path);
                if (r < 0)
                        log_warning_errno(r, "Failed to create compat systemd cgroup '%s', ignoring: %m", path);
        }

        return 1;
}

int cg_attach(const char *controller, const char *path, pid_t pid) {
        _cleanup_free_ char *fs = NULL;
        char c[DECIMAL_STR_MAX(pid_t) + 2];
        int r;

        assert(controller);
        assert(path);
        assert(pid >= 0);

        r = cg_get_path_and_check(controller, path, "cgroup.procs", &fs);
        if (r < 0)
                return r;

        if (pid == 0)
                pid = getpid_cached();

        xsprintf(c, PID_FMT "\n", pid);

        r = write_string_file(fs, c, WRITE_STRING_FILE_DISABLE_BUFFER);
        if (r == -EOPNOTSUPP && cg_is_threaded(path) > 0)
                /* When the threaded mode is used, we cannot read/write the file. Let's return recognizable error. */
                return -EUCLEAN;
        if (r < 0)
                return r;

        r = cg_hybrid_unified();
        if (r < 0)
                return r;
        if (r > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
                r = cg_attach(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, pid);
                if (r < 0)
                        log_warning_errno(r, "Failed to attach "PID_FMT" to compat systemd cgroup '%s', ignoring: %m", pid, path);
        }

        return 0;
}

int cg_fd_attach(int fd, pid_t pid) {
        char c[DECIMAL_STR_MAX(pid_t) + 2];

        assert(fd >= 0);
        assert(pid >= 0);

        if (pid == 0)
                pid = getpid_cached();

        xsprintf(c, PID_FMT "\n", pid);

        return write_string_file_at(fd, "cgroup.procs", c, WRITE_STRING_FILE_DISABLE_BUFFER);
}

int cg_attach_fallback(const char *controller, const char *path, pid_t pid) {
        int r;

        assert(controller);
        assert(path);
        assert(pid >= 0);

        r = cg_attach(controller, path, pid);
        if (r < 0) {
                char prefix[strlen(path) + 1];

                /* This didn't work? Then let's try all prefixes of the destination */

                PATH_FOREACH_PREFIX(prefix, path) {
                        int q;

                        q = cg_attach(controller, prefix, pid);
                        if (q >= 0)
                                return q;
                }
        }

        return r;
}

int cg_create_and_attach(const char *controller, const char *path, pid_t pid) {
        int r, q;

        /* This does not remove the cgroup on failure */

        assert(pid >= 0);

        r = cg_create(controller, path);
        if (r < 0)
                return r;

        q = cg_attach(controller, path, pid);
        if (q < 0)
                return q;

        return r;
}

int cg_set_access(
                const char *controller,
                const char *path,
                uid_t uid,
                gid_t gid) {

        struct Attribute {
                const char *name;
                bool fatal;
        };

        /* cgroup v1, aka legacy/non-unified */
        static const struct Attribute legacy_attributes[] = {
                { "cgroup.procs",           true  },
                { "tasks",                  false },
                { "cgroup.clone_children",  false },
                {},
        };

        /* cgroup v2, aka unified */
        static const struct Attribute unified_attributes[] = {
                { "cgroup.procs",           true  },
                { "cgroup.subtree_control", true  },
                { "cgroup.threads",         false },
                { "memory.oom.group",       false },
                { "memory.reclaim",         false },
                {},
        };

        static const struct Attribute* const attributes[] = {
                [false] = legacy_attributes,
                [true]  = unified_attributes,
        };

        _cleanup_free_ char *fs = NULL;
        const struct Attribute *i;
        int r, unified;

        assert(path);

        if (uid == UID_INVALID && gid == GID_INVALID)
                return 0;

        unified = cg_unified_controller(controller);
        if (unified < 0)
                return unified;

        /* Configure access to the cgroup itself */
        r = cg_get_path(controller, path, NULL, &fs);
        if (r < 0)
                return r;

        r = chmod_and_chown(fs, 0755, uid, gid);
        if (r < 0)
                return r;

        /* Configure access to the cgroup's attributes */
        for (i = attributes[unified]; i->name; i++) {
                fs = mfree(fs);

                r = cg_get_path(controller, path, i->name, &fs);
                if (r < 0)
                        return r;

                r = chmod_and_chown(fs, 0644, uid, gid);
                if (r < 0) {
                        if (i->fatal)
                                return r;

                        log_debug_errno(r, "Failed to set access on cgroup %s, ignoring: %m", fs);
                }
        }

        if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
                r = cg_hybrid_unified();
                if (r < 0)
                        return r;
                if (r > 0) {
                        /* Always propagate access mode from unified to legacy controller */
                        r = cg_set_access(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, uid, gid);
                        if (r < 0)
                                log_debug_errno(r, "Failed to set access on compatibility systemd cgroup %s, ignoring: %m", path);
                }
        }

        return 0;
}

struct access_callback_data {
        uid_t uid;
        gid_t gid;
        int error;
};

static int access_callback(
                RecurseDirEvent event,
                const char *path,
                int dir_fd,
                int inode_fd,
                const struct dirent *de,
                const struct statx *sx,
                void *userdata) {

        if (!IN_SET(event, RECURSE_DIR_ENTER, RECURSE_DIR_ENTRY))
                return RECURSE_DIR_CONTINUE;

        struct access_callback_data *d = ASSERT_PTR(userdata);

        assert(path);
        assert(inode_fd >= 0);

        if (fchownat(inode_fd, "", d->uid, d->gid, AT_EMPTY_PATH) < 0)
                RET_GATHER(d->error, log_debug_errno(errno, "Failed to change ownership of '%s', ignoring: %m", path));

        return RECURSE_DIR_CONTINUE;
}

int cg_set_access_recursive(
                const char *controller,
                const char *path,
                uid_t uid,
                gid_t gid) {

        _cleanup_close_ int fd = -EBADF;
        _cleanup_free_ char *fs = NULL;
        int r;

        assert(controller);
        assert(path);

        /* A recursive version of cg_set_access(). But note that this one changes ownership of *all* files,
         * not just the allowlist that cg_set_access() uses. Use cg_set_access() on the cgroup you want to
         * delegate, and cg_set_access_recursive() for any subcgroups you might want to create below it. */

        if (!uid_is_valid(uid) && !gid_is_valid(gid))
                return 0;

        r = cg_get_path(controller, path, NULL, &fs);
        if (r < 0)
                return r;

        fd = open(fs, O_DIRECTORY|O_CLOEXEC);
        if (fd < 0)
                return -errno;

        struct access_callback_data d = {
                .uid = uid,
                .gid = gid,
        };

        r = recurse_dir(fd,
                        fs,
                        /* statx_mask= */ 0,
                        /* n_depth_max= */ UINT_MAX,
                        RECURSE_DIR_SAME_MOUNT|RECURSE_DIR_INODE_FD|RECURSE_DIR_TOPLEVEL,
                        access_callback,
                        &d);
        if (r < 0)
                return r;

        assert(d.error <= 0);
        return d.error;
}

int cg_migrate(
                const char *cfrom,
                const char *pfrom,
                const char *cto,
                const char *pto,
                CGroupFlags flags) {

        _cleanup_set_free_ Set *s = NULL;
        bool done;
        int r, ret = 0;

        assert(cfrom);
        assert(pfrom);
        assert(cto);
        assert(pto);

        do {
                _cleanup_fclose_ FILE *f = NULL;
                pid_t pid;

                done = true;

                r = cg_enumerate_processes(cfrom, pfrom, &f);
                if (r < 0)
                        return RET_GATHER(ret, r);

                while ((r = cg_read_pid(f, &pid, flags)) > 0) {
                        /* Throw an error if unmappable PIDs are in output, we can't migrate those. */
                        if (pid == 0)
                                return -EREMOTE;

                        /* This might do weird stuff if we aren't a single-threaded program. However, we
                         * luckily know we are. */
                        if (FLAGS_SET(flags, CGROUP_IGNORE_SELF) && pid == getpid_cached())
                                continue;

                        if (set_contains(s, PID_TO_PTR(pid)))
                                continue;

                        if (pid_is_kernel_thread(pid) > 0)
                                continue;

                        r = cg_attach(cto, pto, pid);
                        if (r < 0) {
                                if (r != -ESRCH)
                                        RET_GATHER(ret, r);
                        } else if (ret == 0)
                                ret = 1;

                        done = false;

                        r = set_ensure_put(&s, /* hash_ops = */ NULL, PID_TO_PTR(pid));
                        if (r < 0)
                                return RET_GATHER(ret, r);
                }
                if (r < 0)
                        return RET_GATHER(ret, r);
        } while (!done);

        return ret;
}

int cg_create_everywhere(CGroupMask supported, CGroupMask mask, const char *path) {
        CGroupController c;
        CGroupMask done;
        bool created;
        int r;

        /* This one will create a cgroup in our private tree, but also
         * duplicate it in the trees specified in mask, and remove it
         * in all others.
         *
         * Returns 0 if the group already existed in the systemd hierarchy,
         * 1 on success, negative otherwise.
         */

        /* First create the cgroup in our own hierarchy. */
        r = cg_create(SYSTEMD_CGROUP_CONTROLLER, path);
        if (r < 0)
                return r;
        created = r;

        /* If we are in the unified hierarchy, we are done now */
        r = cg_all_unified();
        if (r < 0)
                return r;
        if (r > 0)
                return created;

        supported &= CGROUP_MASK_V1;
        mask = CGROUP_MASK_EXTEND_JOINED(mask);
        done = 0;

        /* Otherwise, do the same in the other hierarchies */
        for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
                CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
                const char *n;

                if (!FLAGS_SET(supported, bit))
                        continue;

                if (FLAGS_SET(done, bit))
                        continue;

                n = cgroup_controller_to_string(c);
                if (FLAGS_SET(mask, bit))
                        (void) cg_create(n, path);

                done |= CGROUP_MASK_EXTEND_JOINED(bit);
        }

        return created;
}

int cg_attach_everywhere(CGroupMask supported, const char *path, pid_t pid) {
        int r;

        assert(path);
        assert(pid >= 0);

        r = cg_attach(SYSTEMD_CGROUP_CONTROLLER, path, pid);
        if (r < 0)
                return r;

        r = cg_all_unified();
        if (r < 0)
                return r;
        if (r > 0)
                return 0;

        supported &= CGROUP_MASK_V1;
        CGroupMask done = 0;

        for (CGroupController c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
                CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);

                if (!FLAGS_SET(supported, bit))
                        continue;

                if (FLAGS_SET(done, bit))
                        continue;

                (void) cg_attach_fallback(cgroup_controller_to_string(c), path, pid);
                done |= CGROUP_MASK_EXTEND_JOINED(bit);
        }

        return 0;
}

int cg_trim_everywhere(CGroupMask supported, const char *path, bool delete_root) {
        int r, q;

        assert(path);

        r = cg_trim(SYSTEMD_CGROUP_CONTROLLER, path, delete_root);
        if (r < 0)
                return r;

        q = cg_all_unified();
        if (q < 0)
                return q;
        if (q > 0)
                return r;

        return cg_trim_v1_controllers(supported, _CGROUP_MASK_ALL, path, delete_root);
}

int cg_enable_everywhere(
                CGroupMask supported,
                CGroupMask mask,
                const char *p,
                CGroupMask *ret_result_mask) {

        _cleanup_fclose_ FILE *f = NULL;
        _cleanup_free_ char *fs = NULL;
        CGroupController c;
        CGroupMask ret = 0;
        int r;

        assert(p);

        if (supported == 0) {
                if (ret_result_mask)
                        *ret_result_mask = 0;
                return 0;
        }

        r = cg_all_unified();
        if (r < 0)
                return r;
        if (r == 0) {
                /* On the legacy hierarchy there's no concept of "enabling" controllers in cgroups defined. Let's claim
                 * complete success right away. (If you wonder why we return the full mask here, rather than zero: the
                 * caller tends to use the returned mask later on to compare if all controllers where properly joined,
                 * and if not requeues realization. This use is the primary purpose of the return value, hence let's
                 * minimize surprises here and reduce triggers for re-realization by always saying we fully
                 * succeeded.) */
                if (ret_result_mask)
                        *ret_result_mask = mask & supported & CGROUP_MASK_V2; /* If you wonder why we mask this with
                                                                               * CGROUP_MASK_V2: The 'supported' mask
                                                                               * might contain pure-V1 or BPF
                                                                               * controllers, and we never want to
                                                                               * claim that we could enable those with
                                                                               * cgroup.subtree_control */
                return 0;
        }

        r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, p, "cgroup.subtree_control", &fs);
        if (r < 0)
                return r;

        for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
                CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
                const char *n;

                if (!FLAGS_SET(CGROUP_MASK_V2, bit))
                        continue;

                if (!FLAGS_SET(supported, bit))
                        continue;

                n = cgroup_controller_to_string(c);
                {
                        char s[1 + strlen(n) + 1];

                        s[0] = FLAGS_SET(mask, bit) ? '+' : '-';
                        strcpy(s + 1, n);

                        if (!f) {
                                f = fopen(fs, "we");
                                if (!f)
                                        return log_debug_errno(errno, "Failed to open cgroup.subtree_control file of %s: %m", p);
                        }

                        r = write_string_stream(f, s, WRITE_STRING_FILE_DISABLE_BUFFER);
                        if (r < 0) {
                                log_debug_errno(r, "Failed to %s controller %s for %s (%s): %m",
                                                FLAGS_SET(mask, bit) ? "enable" : "disable", n, p, fs);
                                clearerr(f);

                                /* If we can't turn off a controller, leave it on in the reported resulting mask. This
                                 * happens for example when we attempt to turn off a controller up in the tree that is
                                 * used down in the tree. */
                                if (!FLAGS_SET(mask, bit) && r == -EBUSY) /* You might wonder why we check for EBUSY
                                                                           * only here, and not follow the same logic
                                                                           * for other errors such as EINVAL or
                                                                           * EOPNOTSUPP or anything else. That's
                                                                           * because EBUSY indicates that the
                                                                           * controllers is currently enabled and
                                                                           * cannot be disabled because something down
                                                                           * the hierarchy is still using it. Any other
                                                                           * error most likely means something like "I
                                                                           * never heard of this controller" or
                                                                           * similar. In the former case it's hence
                                                                           * safe to assume the controller is still on
                                                                           * after the failed operation, while in the
                                                                           * latter case it's safer to assume the
                                                                           * controller is unknown and hence certainly
                                                                           * not enabled. */
                                        ret |= bit;
                        } else {
                                /* Otherwise, if we managed to turn on a controller, set the bit reflecting that. */
                                if (FLAGS_SET(mask, bit))
                                        ret |= bit;
                        }
                }
        }

        /* Let's return the precise set of controllers now enabled for the cgroup. */
        if (ret_result_mask)
                *ret_result_mask = ret;

        return 0;
}

int cg_migrate_recursive(
                const char *cfrom,
                const char *pfrom,
                const char *cto,
                const char *pto,
                CGroupFlags flags) {

        _cleanup_closedir_ DIR *d = NULL;
        int r, ret = 0;
        char *fn;

        assert(cfrom);
        assert(pfrom);
        assert(cto);
        assert(pto);

        ret = cg_migrate(cfrom, pfrom, cto, pto, flags);

        r = cg_enumerate_subgroups(cfrom, pfrom, &d);
        if (r < 0) {
                if (ret >= 0 && r != -ENOENT)
                        return r;

                return ret;
        }

        while ((r = cg_read_subgroup(d, &fn)) > 0) {
                _cleanup_free_ char *p = NULL;

                p = path_join(empty_to_root(pfrom), fn);
                free(fn);
                if (!p)
                        return -ENOMEM;

                r = cg_migrate_recursive(cfrom, p, cto, pto, flags);
                if (r != 0 && ret >= 0)
                        ret = r;
        }

        if (r < 0 && ret >= 0)
                ret = r;

        return ret;
}

int cg_migrate_recursive_fallback(
                const char *cfrom,
                const char *pfrom,
                const char *cto,
                const char *pto,
                CGroupFlags flags) {

        int r;

        assert(cfrom);
        assert(pfrom);
        assert(cto);
        assert(pto);

        r = cg_migrate_recursive(cfrom, pfrom, cto, pto, flags);
        if (r < 0) {
                char prefix[strlen(pto) + 1];

                /* This didn't work? Then let's try all prefixes of the destination */

                PATH_FOREACH_PREFIX(prefix, pto) {
                        int q;

                        q = cg_migrate_recursive(cfrom, pfrom, cto, prefix, flags);
                        if (q >= 0)
                                return q;
                }
        }

        return r;
}

int cg_migrate_v1_controllers(CGroupMask supported, CGroupMask mask, const char *from, cg_migrate_callback_t to_callback, void *userdata) {
        CGroupController c;
        CGroupMask done;
        int r = 0, q;

        assert(to_callback);

        supported &= CGROUP_MASK_V1;
        mask = CGROUP_MASK_EXTEND_JOINED(mask);
        done = 0;

        for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
                CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
                const char *to = NULL;

                if (!FLAGS_SET(supported, bit))
                        continue;

                if (FLAGS_SET(done, bit))
                        continue;

                if (!FLAGS_SET(mask, bit))
                        continue;

                to = to_callback(bit, userdata);

                /* Remember first error and try continuing */
                q = cg_migrate_recursive_fallback(SYSTEMD_CGROUP_CONTROLLER, from, cgroup_controller_to_string(c), to, 0);
                r = (r < 0) ? r : q;

                done |= CGROUP_MASK_EXTEND_JOINED(bit);
        }

        return r;
}

int cg_trim_v1_controllers(CGroupMask supported, CGroupMask mask, const char *path, bool delete_root) {
        CGroupController c;
        CGroupMask done;
        int r = 0, q;

        supported &= CGROUP_MASK_V1;
        mask = CGROUP_MASK_EXTEND_JOINED(mask);
        done = 0;

        for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
                CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);

                if (!FLAGS_SET(supported, bit))
                        continue;

                if (FLAGS_SET(done, bit))
                        continue;

                if (FLAGS_SET(mask, bit)) {
                        /* Remember first error and try continuing */
                        q = cg_trim(cgroup_controller_to_string(c), path, delete_root);
                        r = (r < 0) ? r : q;
                }
                done |= CGROUP_MASK_EXTEND_JOINED(bit);
        }

        return r;
}

int cg_install_release_agent(const char *controller, const char *agent) {
        _cleanup_free_ char *fs = NULL, *contents = NULL;
        const char *sc;
        int r;

        assert(agent);

        r = cg_unified_controller(controller);
        if (r < 0)
                return r;
        if (r > 0) /* doesn't apply to unified hierarchy */
                return -EOPNOTSUPP;

        r = cg_get_path(controller, NULL, "release_agent", &fs);
        if (r < 0)
                return r;

        r = read_one_line_file(fs, &contents);
        if (r < 0)
                return r;

        sc = strstrip(contents);
        if (isempty(sc)) {
                r = write_string_file(fs, agent, WRITE_STRING_FILE_DISABLE_BUFFER);
                if (r < 0)
                        return r;
        } else if (!path_equal(sc, agent))
                return -EEXIST;

        fs = mfree(fs);
        r = cg_get_path(controller, NULL, "notify_on_release", &fs);
        if (r < 0)
                return r;

        contents = mfree(contents);
        r = read_one_line_file(fs, &contents);
        if (r < 0)
                return r;

        sc = strstrip(contents);
        if (streq(sc, "0")) {
                r = write_string_file(fs, "1", WRITE_STRING_FILE_DISABLE_BUFFER);
                if (r < 0)
                        return r;

                return 1;
        }

        if (!streq(sc, "1"))
                return -EIO;

        return 0;
}

int cg_uninstall_release_agent(const char *controller) {
        _cleanup_free_ char *fs = NULL;
        int r;

        r = cg_unified_controller(controller);
        if (r < 0)
                return r;
        if (r > 0) /* Doesn't apply to unified hierarchy */
                return -EOPNOTSUPP;

        r = cg_get_path(controller, NULL, "notify_on_release", &fs);
        if (r < 0)
                return r;

        r = write_string_file(fs, "0", WRITE_STRING_FILE_DISABLE_BUFFER);
        if (r < 0)
                return r;

        fs = mfree(fs);

        r = cg_get_path(controller, NULL, "release_agent", &fs);
        if (r < 0)
                return r;

        r = write_string_file(fs, "", WRITE_STRING_FILE_DISABLE_BUFFER);
        if (r < 0)
                return r;

        return 0;
}