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