src/shared/cgroup-util.c

   1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
   2
   3 /***
   4   This file is part of systemd.
   5
   6   Copyright 2010 Lennart Poettering
   7
   8   systemd is free software; you can redistribute it and/or modify it
   9   under the terms of the GNU Lesser General Public License as published by
  10   the Free Software Foundation; either version 2.1 of the License, or
  11   (at your option) any later version.
  12
  13   systemd is distributed in the hope that it will be useful, but
  14   WITHOUT ANY WARRANTY; without even the implied warranty of
  15   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  16   Lesser General Public License for more details.
  17
  18   You should have received a copy of the GNU Lesser General Public License
  19   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  20 ***/
  21
  22 #include <errno.h>
  23 #include <unistd.h>
  24 #include <signal.h>
  25 #include <string.h>
  26 #include <stdlib.h>
  27 #include <dirent.h>
  28 #include <sys/stat.h>
  29 #include <sys/types.h>
  30 #include <ftw.h>
  31
  32 #include "cgroup-util.h"
  33 #include "log.h"
  34 #include "set.h"
  35 #include "macro.h"
  36 #include "util.h"
  37 #include "path-util.h"
  38 #include "strv.h"
  39 #include "unit-name.h"
  40 #include "fileio.h"
  41
  42 int cg_enumerate_processes(const char *controller, const char *path, FILE **_f) {
  43         _cleanup_free_ char *fs = NULL;
  44         FILE *f;
  45         int r;
  46
  47         assert(_f);
  48
  49         r = cg_get_path(controller, path, "cgroup.procs", &fs);
  50         if (r < 0)
  51                 return r;
  52
  53         f = fopen(fs, "re");
  54         if (!f)
  55                 return -errno;
  56
  57         *_f = f;
  58         return 0;
  59 }
  60
  61 int cg_enumerate_tasks(const char *controller, const char *path, FILE **_f) {
  62         _cleanup_free_ char *fs = NULL;
  63         FILE *f;
  64         int r;
  65
  66         assert(_f);
  67
  68         r = cg_get_path(controller, path, "tasks", &fs);
  69         if (r < 0)
  70                 return r;
  71
  72         f = fopen(fs, "re");
  73         if (!f)
  74                 return -errno;
  75
  76         *_f = f;
  77         return 0;
  78 }
  79
  80 int cg_read_pid(FILE *f, pid_t *_pid) {
  81         unsigned long ul;
  82
  83         /* Note that the cgroup.procs might contain duplicates! See
  84          * cgroups.txt for details. */
  85
  86         assert(f);
  87         assert(_pid);
  88
  89         errno = 0;
  90         if (fscanf(f, "%lu", &ul) != 1) {
  91
  92                 if (feof(f))
  93                         return 0;
  94
  95                 return errno ? -errno : -EIO;
  96         }
  97
  98         if (ul <= 0)
  99                 return -EIO;
 100
 101         *_pid = (pid_t) ul;
 102         return 1;
 103 }
 104
 105 int cg_enumerate_subgroups(const char *controller, const char *path, DIR **_d) {
 106         _cleanup_free_ char *fs = NULL;
 107         int r;
 108         DIR *d;
 109
 110         assert(_d);
 111
 112         /* This is not recursive! */
 113
 114         r = cg_get_path(controller, path, NULL, &fs);
 115         if (r < 0)
 116                 return r;
 117
 118         d = opendir(fs);
 119         if (!d)
 120                 return -errno;
 121
 122         *_d = d;
 123         return 0;
 124 }
 125
 126 int cg_read_subgroup(DIR *d, char **fn) {
 127         struct dirent *de;
 128
 129         assert(d);
 130         assert(fn);
 131
 132         FOREACH_DIRENT(de, d, return -errno) {
 133                 char *b;
 134
 135                 if (de->d_type != DT_DIR)
 136                         continue;
 137
 138                 if (streq(de->d_name, ".") ||
 139                     streq(de->d_name, ".."))
 140                         continue;
 141
 142                 b = strdup(de->d_name);
 143                 if (!b)
 144                         return -ENOMEM;
 145
 146                 *fn = b;
 147                 return 1;
 148         }
 149
 150         return 0;
 151 }
 152
 153 int cg_rmdir(const char *controller, const char *path, bool honour_sticky) {
 154         _cleanup_free_ char *p = NULL;
 155         int r;
 156
 157         r = cg_get_path(controller, path, NULL, &p);
 158         if (r < 0)
 159                 return r;
 160
 161         if (honour_sticky) {
 162                 char *tasks;
 163
 164                 /* If the sticky bit is set don't remove the directory */
 165
 166                 tasks = strappend(p, "/tasks");
 167                 if (!tasks)
 168                         return -ENOMEM;
 169
 170                 r = file_is_priv_sticky(tasks);
 171                 free(tasks);
 172
 173                 if (r > 0)
 174                         return 0;
 175         }
 176
 177         r = rmdir(p);
 178         if (r < 0 && errno != ENOENT)
 179                 return -errno;
 180
 181         return 0;
 182 }
 183
 184 int cg_kill(const char *controller, const char *path, int sig, bool sigcont, bool ignore_self, Set *s) {
 185         _cleanup_set_free_ Set *allocated_set = NULL;
 186         bool done = false;
 187         int r, ret = 0;
 188         pid_t my_pid;
 189
 190         assert(sig >= 0);
 191
 192         /* This goes through the tasks list and kills them all. This
 193          * is repeated until no further processes are added to the
 194          * tasks list, to properly handle forking processes */
 195
 196         if (!s) {
 197                 s = allocated_set = set_new(trivial_hash_func, trivial_compare_func);
 198                 if (!s)
 199                         return -ENOMEM;
 200         }
 201
 202         my_pid = getpid();
 203
 204         do {
 205                 _cleanup_fclose_ FILE *f = NULL;
 206                 pid_t pid = 0;
 207                 done = true;
 208
 209                 r = cg_enumerate_processes(controller, path, &f);
 210                 if (r < 0) {
 211                         if (ret >= 0 && r != -ENOENT)
 212                                 return r;
 213
 214                         return ret;
 215                 }
 216
 217                 while ((r = cg_read_pid(f, &pid)) > 0) {
 218
 219                         if (ignore_self && pid == my_pid)
 220                                 continue;
 221
 222                         if (set_get(s, LONG_TO_PTR(pid)) == LONG_TO_PTR(pid))
 223                                 continue;
 224
 225                         /* If we haven't killed this process yet, kill
 226                          * it */
 227                         if (kill(pid, sig) < 0) {
 228                                 if (ret >= 0 && errno != ESRCH)
 229                                         ret = -errno;
 230                         } else if (ret == 0) {
 231
 232                                 if (sigcont)
 233                                         kill(pid, SIGCONT);
 234
 235                                 ret = 1;
 236                         }
 237
 238                         done = false;
 239
 240                         r = set_put(s, LONG_TO_PTR(pid));
 241                         if (r < 0) {
 242                                 if (ret >= 0)
 243                                         return r;
 244
 245                                 return ret;
 246                         }
 247                 }
 248
 249                 if (r < 0) {
 250                         if (ret >= 0)
 251                                 return r;
 252
 253                         return ret;
 254                 }
 255
 256                 /* To avoid racing against processes which fork
 257                  * quicker than we can kill them we repeat this until
 258                  * no new pids need to be killed. */
 259
 260         } while (!done);
 261
 262         return ret;
 263 }
 264
 265 int cg_kill_recursive(const char *controller, const char *path, int sig, bool sigcont, bool ignore_self, bool rem, Set *s) {
 266         _cleanup_set_free_ Set *allocated_set = NULL;
 267         _cleanup_closedir_ DIR *d = NULL;
 268         int r, ret = 0;
 269         char *fn;
 270
 271         assert(path);
 272         assert(sig >= 0);
 273
 274         if (!s) {
 275                 s = allocated_set = set_new(trivial_hash_func, trivial_compare_func);
 276                 if (!s)
 277                         return -ENOMEM;
 278         }
 279
 280         ret = cg_kill(controller, path, sig, sigcont, ignore_self, s);
 281
 282         r = cg_enumerate_subgroups(controller, path, &d);
 283         if (r < 0) {
 284                 if (ret >= 0 && r != -ENOENT)
 285                         return r;
 286
 287                 return ret;
 288         }
 289
 290         while ((r = cg_read_subgroup(d, &fn)) > 0) {
 291                 _cleanup_free_ char *p = NULL;
 292
 293                 p = strjoin(path, "/", fn, NULL);
 294                 free(fn);
 295                 if (!p)
 296                         return -ENOMEM;
 297
 298                 r = cg_kill_recursive(controller, p, sig, sigcont, ignore_self, rem, s);
 299                 if (ret >= 0 && r != 0)
 300                         ret = r;
 301         }
 302
 303         if (ret >= 0 && r < 0)
 304                 ret = r;
 305
 306         if (rem) {
 307                 r = cg_rmdir(controller, path, true);
 308                 if (r < 0 && ret >= 0 && r != -ENOENT && r != -EBUSY)
 309                         return r;
 310         }
 311
 312         return ret;
 313 }
 314
 315 int cg_kill_recursive_and_wait(const char *controller, const char *path, bool rem) {
 316         unsigned i;
 317
 318         assert(path);
 319
 320         /* This safely kills all processes; first it sends a SIGTERM,
 321          * then checks 8 times after 200ms whether the group is now
 322          * empty, then kills everything that is left with SIGKILL and
 323          * finally checks 5 times after 200ms each whether the group
 324          * is finally empty. */
 325
 326         for (i = 0; i < 15; i++) {
 327                 int sig, r;
 328
 329                 if (i <= 0)
 330                         sig = SIGTERM;
 331                 else if (i == 9)
 332                         sig = SIGKILL;
 333                 else
 334                         sig = 0;
 335
 336                 r = cg_kill_recursive(controller, path, sig, true, true, rem, NULL);
 337                 if (r <= 0)
 338                         return r;
 339
 340                 usleep(200 * USEC_PER_MSEC);
 341         }
 342
 343         return 0;
 344 }
 345
 346 int cg_migrate(const char *cfrom, const char *pfrom, const char *cto, const char *pto, bool ignore_self) {
 347         bool done = false;
 348         _cleanup_set_free_ Set *s = NULL;
 349         int r, ret = 0;
 350         pid_t my_pid;
 351
 352         assert(cfrom);
 353         assert(pfrom);
 354         assert(cto);
 355         assert(pto);
 356
 357         s = set_new(trivial_hash_func, trivial_compare_func);
 358         if (!s)
 359                 return -ENOMEM;
 360
 361         my_pid = getpid();
 362
 363         do {
 364                 _cleanup_fclose_ FILE *f = NULL;
 365                 pid_t pid = 0;
 366                 done = true;
 367
 368                 r = cg_enumerate_tasks(cfrom, pfrom, &f);
 369                 if (r < 0) {
 370                         if (ret >= 0 && r != -ENOENT)
 371                                 return r;
 372
 373                         return ret;
 374                 }
 375
 376                 while ((r = cg_read_pid(f, &pid)) > 0) {
 377
 378                         /* This might do weird stuff if we aren't a
 379                          * single-threaded program. However, we
 380                          * luckily know we are not */
 381                         if (ignore_self && pid == my_pid)
 382                                 continue;
 383
 384                         if (set_get(s, LONG_TO_PTR(pid)) == LONG_TO_PTR(pid))
 385                                 continue;
 386
 387                         r = cg_attach(cto, pto, pid);
 388                         if (r < 0) {
 389                                 if (ret >= 0 && r != -ESRCH)
 390                                         ret = r;
 391                         } else if (ret == 0)
 392                                 ret = 1;
 393
 394                         done = false;
 395
 396                         r = set_put(s, LONG_TO_PTR(pid));
 397                         if (r < 0) {
 398                                 if (ret >= 0)
 399                                         return r;
 400
 401                                 return ret;
 402                         }
 403                 }
 404
 405                 if (r < 0) {
 406                         if (ret >= 0)
 407                                 return r;
 408
 409                         return ret;
 410                 }
 411         } while (!done);
 412
 413         return ret;
 414 }
 415
 416 int cg_migrate_recursive(const char *cfrom, const char *pfrom, const char *cto, const char *pto, bool ignore_self, bool rem) {
 417         _cleanup_closedir_ DIR *d = NULL;
 418         int r, ret = 0;
 419         char *fn;
 420
 421         assert(cfrom);
 422         assert(pfrom);
 423         assert(cto);
 424         assert(pto);
 425
 426         ret = cg_migrate(cfrom, pfrom, cto, pto, ignore_self);
 427
 428         r = cg_enumerate_subgroups(cfrom, pfrom, &d);
 429         if (r < 0) {
 430                 if (ret >= 0 && r != -ENOENT)
 431                         return r;
 432
 433                 return ret;
 434         }
 435
 436         while ((r = cg_read_subgroup(d, &fn)) > 0) {
 437                 _cleanup_free_ char *p = NULL;
 438
 439                 p = strjoin(pfrom, "/", fn, NULL);
 440                 free(fn);
 441                 if (!p) {
 442                         if (ret >= 0)
 443                                 return -ENOMEM;
 444
 445                         return ret;
 446                 }
 447
 448                 r = cg_migrate_recursive(cfrom, p, cto, pto, ignore_self, rem);
 449                 if (r != 0 && ret >= 0)
 450                         ret = r;
 451         }
 452
 453         if (r < 0 && ret >= 0)
 454                 ret = r;
 455
 456         if (rem) {
 457                 r = cg_rmdir(cfrom, pfrom, true);
 458                 if (r < 0 && ret >= 0 && r != -ENOENT && r != -EBUSY)
 459                         return r;
 460         }
 461
 462         return ret;
 463 }
 464
 465 static const char *normalize_controller(const char *controller) {
 466
 467         assert(controller);
 468
 469         if (streq(controller, SYSTEMD_CGROUP_CONTROLLER))
 470                 return "systemd";
 471         else if (startswith(controller, "name="))
 472                 return controller + 5;
 473         else
 474                 return controller;
 475 }
 476
 477 static int join_path(const char *controller, const char *path, const char *suffix, char **fs) {
 478         char *t = NULL;
 479
 480         if (controller) {
 481                 if (path && suffix)
 482                         t = strjoin("/sys/fs/cgroup/", controller, "/", path, "/", suffix, NULL);
 483                 else if (path)
 484                         t = strjoin("/sys/fs/cgroup/", controller, "/", path, NULL);
 485                 else if (suffix)
 486                         t = strjoin("/sys/fs/cgroup/", controller, "/", suffix, NULL);
 487                 else
 488                         t = strappend("/sys/fs/cgroup/", controller);
 489         } else {
 490                 if (path && suffix)
 491                         t = strjoin(path, "/", suffix, NULL);
 492                 else if (path)
 493                         t = strdup(path);
 494                 else
 495                         return -EINVAL;
 496         }
 497
 498         if (!t)
 499                 return -ENOMEM;
 500
 501         path_kill_slashes(t);
 502
 503         *fs = t;
 504         return 0;
 505 }
 506
 507 int cg_get_path(const char *controller, const char *path, const char *suffix, char **fs) {
 508         const char *p;
 509         static __thread bool good = false;
 510
 511         assert(fs);
 512
 513         if (_unlikely_(!good)) {
 514                 int r;
 515
 516                 r = path_is_mount_point("/sys/fs/cgroup", false);
 517                 if (r <= 0)
 518                         return r < 0 ? r : -ENOENT;
 519
 520                 /* Cache this to save a few stat()s */
 521                 good = true;
 522         }
 523
 524         p = controller ? normalize_controller(controller) : NULL;
 525
 526         return join_path(p, path, suffix, fs);
 527 }
 528
 529 static int check_hierarchy(const char *p) {
 530         char *cc;
 531
 532         assert(p);
 533
 534         /* Check if this controller actually really exists */
 535         cc = alloca(sizeof("/sys/fs/cgroup/") + strlen(p));
 536         strcpy(stpcpy(cc, "/sys/fs/cgroup/"), p);
 537         if (access(cc, F_OK) < 0)
 538                 return -errno;
 539
 540         return 0;
 541 }
 542
 543 int cg_get_path_and_check(const char *controller, const char *path, const char *suffix, char **fs) {
 544         const char *p;
 545         int r;
 546
 547         assert(fs);
 548
 549         if (isempty(controller))
 550                 return -EINVAL;
 551
 552         /* Normalize the controller syntax */
 553         p = normalize_controller(controller);
 554
 555         /* Check if this controller actually really exists */
 556         r = check_hierarchy(p);
 557         if (r < 0)
 558                 return r;
 559
 560         return join_path(p, path, suffix, fs);
 561 }
 562
 563 static int trim_cb(const char *path, const struct stat *sb, int typeflag, struct FTW *ftwbuf) {
 564         char *p;
 565         bool is_sticky;
 566
 567         if (typeflag != FTW_DP)
 568                 return 0;
 569
 570         if (ftwbuf->level < 1)
 571                 return 0;
 572
 573         p = strappend(path, "/tasks");
 574         if (!p) {
 575                 errno = ENOMEM;
 576                 return 1;
 577         }
 578
 579         is_sticky = file_is_priv_sticky(p) > 0;
 580         free(p);
 581
 582         if (is_sticky)
 583                 return 0;
 584
 585         rmdir(path);
 586         return 0;
 587 }
 588
 589 int cg_trim(const char *controller, const char *path, bool delete_root) {
 590         _cleanup_free_ char *fs = NULL;
 591         int r = 0;
 592
 593         assert(path);
 594
 595         r = cg_get_path(controller, path, NULL, &fs);
 596         if (r < 0)
 597                 return r;
 598
 599         errno = 0;
 600         if (nftw(fs, trim_cb, 64, FTW_DEPTH|FTW_MOUNT|FTW_PHYS) != 0)
 601                 r = errno ? -errno : -EIO;
 602
 603         if (delete_root) {
 604                 bool is_sticky;
 605                 char *p;
 606
 607                 p = strappend(fs, "/tasks");
 608                 if (!p)
 609                         return -ENOMEM;
 610
 611                 is_sticky = file_is_priv_sticky(p) > 0;
 612                 free(p);
 613
 614                 if (!is_sticky)
 615                         if (rmdir(fs) < 0 && errno != ENOENT && r == 0)
 616                                 return -errno;
 617         }
 618
 619         return r;
 620 }
 621
 622 int cg_delete(const char *controller, const char *path) {
 623         _cleanup_free_ char *parent = NULL;
 624         int r;
 625
 626         assert(path);
 627
 628         r = path_get_parent(path, &parent);
 629         if (r < 0)
 630                 return r;
 631
 632         r = cg_migrate_recursive(controller, path, controller, parent, false, true);
 633         return r == -ENOENT ? 0 : r;
 634 }
 635
 636 int cg_attach(const char *controller, const char *path, pid_t pid) {
 637         _cleanup_free_ char *fs = NULL;
 638         char c[DECIMAL_STR_MAX(pid_t) + 2];
 639         int r;
 640
 641         assert(path);
 642         assert(pid >= 0);
 643
 644         r = cg_get_path_and_check(controller, path, "tasks", &fs);
 645         if (r < 0)
 646                 return r;
 647
 648         if (pid == 0)
 649                 pid = getpid();
 650
 651         snprintf(c, sizeof(c), "%lu\n", (unsigned long) pid);
 652
 653         return write_string_file(fs, c);
 654 }
 655
 656 int cg_set_group_access(
 657                 const char *controller,
 658                 const char *path,
 659                 mode_t mode,
 660                 uid_t uid,
 661                 gid_t gid) {
 662
 663         _cleanup_free_ char *fs = NULL;
 664         int r;
 665
 666         assert(path);
 667
 668         if (mode != (mode_t) -1)
 669                 mode &= 0777;
 670
 671         r = cg_get_path(controller, path, NULL, &fs);
 672         if (r < 0)
 673                 return r;
 674
 675         return chmod_and_chown(fs, mode, uid, gid);
 676 }
 677
 678 int cg_set_task_access(
 679                 const char *controller,
 680                 const char *path,
 681                 mode_t mode,
 682                 uid_t uid,
 683                 gid_t gid,
 684                 int sticky) {
 685
 686         _cleanup_free_ char *fs = NULL, *procs = NULL;
 687         int r;
 688
 689         assert(path);
 690
 691         if (mode == (mode_t) -1 && uid == (uid_t) -1 && gid == (gid_t) -1 && sticky < 0)
 692                 return 0;
 693
 694         if (mode != (mode_t) -1)
 695                 mode &= 0666;
 696
 697         r = cg_get_path(controller, path, "tasks", &fs);
 698         if (r < 0)
 699                 return r;
 700
 701         if (sticky >= 0 && mode != (mode_t) -1)
 702                 /* Both mode and sticky param are passed */
 703                 mode |= (sticky ? S_ISVTX : 0);
 704         else if ((sticky >= 0 && mode == (mode_t) -1) ||
 705                  (mode != (mode_t) -1 && sticky < 0)) {
 706                 struct stat st;
 707
 708                 /* Only one param is passed, hence read the current
 709                  * mode from the file itself */
 710
 711                 r = lstat(fs, &st);
 712                 if (r < 0)
 713                         return -errno;
 714
 715                 if (mode == (mode_t) -1)
 716                         /* No mode set, we just shall set the sticky bit */
 717                         mode = (st.st_mode & ~S_ISVTX) | (sticky ? S_ISVTX : 0);
 718                 else
 719                         /* Only mode set, leave sticky bit untouched */
 720                         mode = (st.st_mode & ~0777) | mode;
 721         }
 722
 723         r = chmod_and_chown(fs, mode, uid, gid);
 724         if (r < 0)
 725                 return r;
 726
 727         /* Always keep values for "cgroup.procs" in sync with "tasks" */
 728         r = cg_get_path(controller, path, "cgroup.procs", &procs);
 729         if (r < 0)
 730                 return r;
 731
 732         return chmod_and_chown(procs, mode, uid, gid);
 733 }
 734
 735 int cg_pid_get_path(const char *controller, pid_t pid, char **path) {
 736         char fs[sizeof("/proc/") - 1 + DECIMAL_STR_MAX(pid_t) + sizeof("/cgroup")];
 737         _cleanup_fclose_ FILE *f = NULL;
 738         char line[LINE_MAX];
 739         size_t cs;
 740
 741         assert(path);
 742         assert(pid >= 0);
 743
 744         if (!controller)
 745                 controller = SYSTEMD_CGROUP_CONTROLLER;
 746
 747         if (pid == 0)
 748                 pid = getpid();
 749
 750         sprintf(fs, "/proc/%lu/cgroup", (unsigned long) pid);
 751         f = fopen(fs, "re");
 752         if (!f)
 753                 return errno == ENOENT ? -ESRCH : -errno;
 754
 755         cs = strlen(controller);
 756
 757         FOREACH_LINE(line, f, return -errno) {
 758                 char *l, *p;
 759
 760                 truncate_nl(line);
 761
 762                 l = strchr(line, ':');
 763                 if (!l)
 764                         continue;
 765
 766                 l++;
 767                 if (!strneq(l, controller, cs))
 768                         continue;
 769
 770                 if (l[cs] != ':')
 771                         continue;
 772
 773                 p = strdup(l + cs + 1);
 774                 if (!p)
 775                         return -ENOMEM;
 776
 777                 *path = p;
 778                 return 0;
 779         }
 780
 781         return -ENOENT;
 782 }
 783
 784 int cg_install_release_agent(const char *controller, const char *agent) {
 785         _cleanup_free_ char *fs = NULL, *contents = NULL;
 786         char *sc;
 787         int r;
 788
 789         assert(agent);
 790
 791         r = cg_get_path(controller, NULL, "release_agent", &fs);
 792         if (r < 0)
 793                 return r;
 794
 795         r = read_one_line_file(fs, &contents);
 796         if (r < 0)
 797                 return r;
 798
 799         sc = strstrip(contents);
 800         if (sc[0] == 0) {
 801                 r = write_string_file(fs, agent);
 802                 if (r < 0)
 803                         return r;
 804         } else if (!streq(sc, agent))
 805                 return -EEXIST;
 806
 807         free(fs);
 808         fs = NULL;
 809         r = cg_get_path(controller, NULL, "notify_on_release", &fs);
 810         if (r < 0)
 811                 return r;
 812
 813         free(contents);
 814         contents = NULL;
 815         r = read_one_line_file(fs, &contents);
 816         if (r < 0)
 817                 return r;
 818
 819         sc = strstrip(contents);
 820         if (streq(sc, "0")) {
 821                 r = write_string_file(fs, "1");
 822                 if (r < 0)
 823                         return r;
 824
 825                 return 1;
 826         }
 827
 828         if (!streq(sc, "1"))
 829                 return -EIO;
 830
 831         return 0;
 832 }
 833
 834 int cg_is_empty(const char *controller, const char *path, bool ignore_self) {
 835         _cleanup_fclose_ FILE *f = NULL;
 836         pid_t pid = 0, self_pid;
 837         bool found = false;
 838         int r;
 839
 840         assert(path);
 841
 842         r = cg_enumerate_tasks(controller, path, &f);
 843         if (r < 0)
 844                 return r == -ENOENT ? 1 : r;
 845
 846         self_pid = getpid();
 847
 848         while ((r = cg_read_pid(f, &pid)) > 0) {
 849
 850                 if (ignore_self && pid == self_pid)
 851                         continue;
 852
 853                 found = true;
 854                 break;
 855         }
 856
 857         if (r < 0)
 858                 return r;
 859
 860         return !found;
 861 }
 862
 863 int cg_is_empty_by_spec(const char *spec, bool ignore_self) {
 864         _cleanup_free_ char *controller = NULL, *path = NULL;
 865         int r;
 866
 867         assert(spec);
 868
 869         r = cg_split_spec(spec, &controller, &path);
 870         if (r < 0)
 871                 return r;
 872
 873         return cg_is_empty(controller, path, ignore_self);
 874 }
 875
 876 int cg_is_empty_recursive(const char *controller, const char *path, bool ignore_self) {
 877         _cleanup_closedir_ DIR *d = NULL;
 878         char *fn;
 879         int r;
 880
 881         assert(path);
 882
 883         r = cg_is_empty(controller, path, ignore_self);
 884         if (r <= 0)
 885                 return r;
 886
 887         r = cg_enumerate_subgroups(controller, path, &d);
 888         if (r < 0)
 889                 return r == -ENOENT ? 1 : r;
 890
 891         while ((r = cg_read_subgroup(d, &fn)) > 0) {
 892                 _cleanup_free_ char *p = NULL;
 893
 894                 p = strjoin(path, "/", fn, NULL);
 895                 free(fn);
 896                 if (!p)
 897                         return -ENOMEM;
 898
 899                 r = cg_is_empty_recursive(controller, p, ignore_self);
 900                 if (r <= 0)
 901                         return r;
 902         }
 903
 904         if (r < 0)
 905                 return r;
 906
 907         return 1;
 908 }
 909
 910 int cg_split_spec(const char *spec, char **controller, char **path) {
 911         const char *e;
 912         char *t = NULL, *u = NULL;
 913
 914         assert(spec);
 915
 916         if (*spec == '/') {
 917                 if (!path_is_safe(spec))
 918                         return -EINVAL;
 919
 920                 if (path) {
 921                         t = strdup(spec);
 922                         if (!t)
 923                                 return -ENOMEM;
 924
 925                         *path = t;
 926                 }
 927
 928                 if (controller)
 929                         *controller = NULL;
 930
 931                 return 0;
 932         }
 933
 934         e = strchr(spec, ':');
 935         if (!e) {
 936                 if (!filename_is_safe(spec))
 937                         return -EINVAL;
 938
 939                 if (controller) {
 940                         t = strdup(spec);
 941                         if (!t)
 942                                 return -ENOMEM;
 943
 944                         *controller = t;
 945                 }
 946
 947                 if (path)
 948                         *path = NULL;
 949
 950                 return 0;
 951         }
 952
 953         t = strndup(spec, e-spec);
 954         if (!t)
 955                 return -ENOMEM;
 956         if (!filename_is_safe(t)) {
 957                 free(t);
 958                 return -EINVAL;
 959         }
 960
 961         u = strdup(e+1);
 962         if (!u) {
 963                 free(t);
 964                 return -ENOMEM;
 965         }
 966         if (!path_is_safe(u)) {
 967                 free(t);
 968                 free(u);
 969                 return -EINVAL;
 970         }
 971
 972         if (controller)
 973                 *controller = t;
 974         else
 975                 free(t);
 976
 977         if (path)
 978                 *path = u;
 979         else
 980                 free(u);
 981
 982         return 0;
 983 }
 984
 985 int cg_join_spec(const char *controller, const char *path, char **spec) {
 986         char *s;
 987
 988         assert(path);
 989
 990         if (!controller)
 991                 controller = "systemd";
 992         else if (controller[0] == 0 ||
 993                  strchr(controller, ':') ||
 994                  strchr(controller, '/'))
 995                 return -EINVAL;
 996
 997         if (!path_is_absolute(path))
 998                 return -EINVAL;
 999
1000         controller = normalize_controller(controller);
1001
1002         s = strjoin(controller, ":", path, NULL);
1003         if (!s)
1004                 return -ENOMEM;
1005
1006         *spec = s;
1007         return 0;
1008 }
1009
1010 int cg_mangle_path(const char *path, char **result) {
1011         char *t, *c, *p;
1012         int r;
1013
1014         assert(path);
1015         assert(result);
1016
1017         /* First check if it already is a filesystem path */
1018         if (path_startswith(path, "/sys/fs/cgroup")) {
1019
1020                 t = strdup(path);
1021                 if (!t)
1022                         return -ENOMEM;
1023
1024                 *result = t;
1025                 return 0;
1026         }
1027
1028         /* Otherwise treat it as cg spec */
1029         r = cg_split_spec(path, &c, &p);
1030         if (r < 0)
1031                 return r;
1032
1033         r = cg_get_path(c ? c : SYSTEMD_CGROUP_CONTROLLER, p ? p : "/", NULL, result);
1034         free(c);
1035         free(p);
1036
1037         return r;
1038 }
1039
1040 int cg_get_system_path(char **path) {
1041         char *p;
1042         int r;
1043
1044         assert(path);
1045
1046         r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 1, &p);
1047         if (r < 0) {
1048                 p = strdup("/system");
1049                 if (!p)
1050                         return -ENOMEM;
1051         }
1052
1053         if (endswith(p, "/system"))
1054                 *path = p;
1055         else {
1056                 char *q;
1057
1058                 q = strappend(p, "/system");
1059                 free(p);
1060                 if (!q)
1061                         return -ENOMEM;
1062
1063                 *path = q;
1064         }
1065
1066         return 0;
1067 }
1068
1069 int cg_get_root_path(char **path) {
1070         char *root, *e;
1071         int r;
1072
1073         assert(path);
1074
1075         r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 1, &root);
1076         if (r < 0)
1077                 return r;
1078
1079         e = endswith(root, "/system");
1080         if (e == root)
1081                 e[1] = 0;
1082         else if (e)
1083                 *e = 0;
1084
1085         *path = root;
1086         return 0;
1087 }
1088
1089 int cg_get_user_path(char **path) {
1090         _cleanup_free_ char *root = NULL;
1091         char *p;
1092
1093         assert(path);
1094
1095         /* Figure out the place to put user cgroups below. We use the
1096          * same as PID 1 has but with the "/system" suffix replaced by
1097          * "/user" */
1098
1099         if (cg_get_root_path(&root) < 0 || streq(root, "/"))
1100                 p = strdup("/user");
1101         else
1102                 p = strappend(root, "/user");
1103
1104         if (!p)
1105                 return -ENOMEM;
1106
1107         *path = p;
1108         return 0;
1109 }
1110
1111 int cg_get_machine_path(char **path) {
1112         _cleanup_free_ char *root = NULL;
1113         char *p;
1114
1115         assert(path);
1116
1117         if (cg_get_root_path(&root) < 0 || streq(root, "/"))
1118                 p = strdup("/machine");
1119         else
1120                 p = strappend(root, "/machine");
1121
1122         if (!p)
1123                 return -ENOMEM;
1124
1125         *path = p;
1126         return 0;
1127 }
1128
1129 char **cg_shorten_controllers(char **controllers) {
1130         char **f, **t;
1131
1132         if (!controllers)
1133                 return controllers;
1134
1135         for (f = controllers, t = controllers; *f; f++) {
1136                 const char *p;
1137                 int r;
1138
1139                 p = normalize_controller(*f);
1140
1141                 if (streq(*f, "systemd")) {
1142                         free(*f);
1143                         continue;
1144                 }
1145
1146                 r = check_hierarchy(p);
1147                 if (r < 0) {
1148                         log_debug("Controller %s is not available, removing from controllers list.", *f);
1149                         free(*f);
1150                         continue;
1151                 }
1152
1153                 *(t++) = *f;
1154         }
1155
1156         *t = NULL;
1157         return strv_uniq(controllers);
1158 }
1159
1160 int cg_pid_get_path_shifted(pid_t pid, char **root, char **cgroup) {
1161         _cleanup_free_ char *cg_root = NULL;
1162         char *cg_process, *p;
1163         int r;
1164
1165         r = cg_get_root_path(&cg_root);
1166         if (r < 0)
1167                 return r;
1168
1169         r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cg_process);
1170         if (r < 0)
1171                 return r;
1172
1173         p = path_startswith(cg_process, cg_root);
1174         if (p)
1175                 p--;
1176         else
1177                 p = cg_process;
1178
1179         if (cgroup) {
1180                 char* c;
1181
1182                 c = strdup(p);
1183                 if (!c) {
1184                         free(cg_process);
1185                         return -ENOMEM;
1186                 }
1187
1188                 *cgroup = c;
1189         }
1190
1191         if (root) {
1192                 cg_process[p-cg_process] = 0;
1193                 *root = cg_process;
1194         } else
1195                 free(cg_process);
1196
1197         return 0;
1198 }
1199
1200 /* non-static only for testing purposes */
1201 int cg_path_decode_unit(const char *cgroup, char **unit){
1202         char *p, *e, *c, *s, *k;
1203
1204         assert(cgroup);
1205         assert(unit);
1206
1207         e = strchrnul(cgroup, '/');
1208         c = strndupa(cgroup, e - cgroup);
1209
1210         /* Could this be a valid unit name? */
1211         if (!unit_name_is_valid(c, true))
1212                 return -EINVAL;
1213
1214         if (!unit_name_is_template(c))
1215                 s = strdup(c);
1216         else {
1217                 if (*e != '/')
1218                         return -EINVAL;
1219
1220                 e += strspn(e, "/");
1221                 p = strchrnul(e, '/');
1222
1223                 /* Don't allow empty instance strings */
1224                 if (p == e)
1225                         return -EINVAL;
1226
1227                 k = strndupa(e, p - e);
1228
1229                 s = unit_name_replace_instance(c, k);
1230         }
1231
1232         if (!s)
1233                 return -ENOMEM;
1234
1235         *unit = s;
1236         return 0;
1237 }
1238
1239 int cg_path_get_unit(const char *path, char **unit) {
1240         const char *e;
1241
1242         assert(path);
1243         assert(unit);
1244
1245         e = path_startswith(path, "/system/");
1246         if (!e)
1247                 return -ENOENT;
1248
1249         return cg_path_decode_unit(e, unit);
1250 }
1251
1252 int cg_pid_get_unit(pid_t pid, char **unit) {
1253         char _cleanup_free_ *cgroup = NULL;
1254         int r;
1255
1256         assert(unit);
1257
1258         r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1259         if (r < 0)
1260                 return r;
1261
1262         return cg_path_get_unit(cgroup, unit);
1263 }
1264
1265 static const char *skip_label(const char *e) {
1266         assert(e);
1267
1268         e = strchr(e, '/');
1269         if (!e)
1270                 return NULL;
1271
1272         e += strspn(e, "/");
1273         return e;
1274 }
1275
1276 int cg_path_get_user_unit(const char *path, char **unit) {
1277         const char *e;
1278
1279         assert(path);
1280         assert(unit);
1281
1282         /* We always have to parse the path from the beginning as unit
1283          * cgroups might have arbitrary child cgroups and we shouldn't get
1284          * confused by those */
1285
1286         e = path_startswith(path, "/user/");
1287         if (!e)
1288                 return -ENOENT;
1289
1290         /* Skip the user name */
1291         e = skip_label(e);
1292         if (!e)
1293                 return -ENOENT;
1294
1295         /* Skip the session ID */
1296         e = skip_label(e);
1297         if (!e)
1298                 return -ENOENT;
1299
1300         /* Skip the systemd cgroup */
1301         e = skip_label(e);
1302         if (!e)
1303                 return -ENOENT;
1304
1305         return cg_path_decode_unit(e, unit);
1306 }
1307
1308 int cg_pid_get_user_unit(pid_t pid, char **unit) {
1309         char _cleanup_free_ *cgroup = NULL;
1310         int r;
1311
1312         assert(unit);
1313
1314         r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1315         if (r < 0)
1316                 return r;
1317
1318         return cg_path_get_user_unit(cgroup, unit);
1319 }
1320
1321 int cg_path_get_machine_name(const char *path, char **machine) {
1322         const char *e, *n;
1323         char *s;
1324
1325         assert(path);
1326         assert(machine);
1327
1328         e = path_startswith(path, "/machine/");
1329         if (!e)
1330                 return -ENOENT;
1331
1332         n = strchrnul(e, '/');
1333         if (e == n)
1334                 return -ENOENT;
1335
1336         s = strndup(e, n - e);
1337         if (!s)
1338                 return -ENOMEM;
1339
1340         *machine = s;
1341         return 0;
1342 }
1343
1344 int cg_pid_get_machine_name(pid_t pid, char **machine) {
1345         char _cleanup_free_ *cgroup = NULL;
1346         int r;
1347
1348         assert(machine);
1349
1350         r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1351         if (r < 0)
1352                 return r;
1353
1354         return cg_path_get_machine_name(cgroup, machine);
1355 }
1356
1357 int cg_path_get_session(const char *path, char **session) {
1358         const char *e, *n;
1359         char *s;
1360
1361         assert(path);
1362         assert(session);
1363
1364         e = path_startswith(path, "/user/");
1365         if (!e)
1366                 return -ENOENT;
1367
1368         /* Skip the user name */
1369         e = skip_label(e);
1370         if (!e)
1371                 return -ENOENT;
1372
1373         n = strchrnul(e, '/');
1374         if (e == n)
1375                 return -ENOENT;
1376
1377         if (n - e == 6 && memcmp(e, "shared", 6) == 0)
1378                 return -ENOENT;
1379
1380         s = strndup(e, n - e);
1381         if (!s)
1382                 return -ENOMEM;
1383
1384         *session = s;
1385         return 0;
1386 }
1387
1388 int cg_pid_get_session(pid_t pid, char **session) {
1389         char _cleanup_free_ *cgroup = NULL;
1390         int r;
1391
1392         assert(session);
1393
1394         r = cg_pid_get_path_shifted(pid, NULL, &cgroup);
1395         if (r < 0)
1396                 return r;
1397
1398         return cg_path_get_session(cgroup, session);
1399 }
1400
1401 int cg_controller_from_attr(const char *attr, char **controller) {
1402         const char *dot;
1403         char *c;
1404
1405         assert(attr);
1406         assert(controller);
1407
1408         if (!filename_is_safe(attr))
1409                 return -EINVAL;
1410
1411         dot = strchr(attr, '.');
1412         if (!dot) {
1413                 *controller = NULL;
1414                 return 0;
1415         }
1416
1417         c = strndup(attr, dot - attr);
1418         if (!c)
1419                 return -ENOMEM;
1420
1421         if (!filename_is_safe(c)) {
1422                 free(c);
1423                 return -EINVAL;
1424         }
1425
1426         *controller = c;
1427         return 1;
1428 }