src/cgtop/cgtop.c

   1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
   2
   3 #include <errno.h>
   4 #include <getopt.h>
   5 #include <signal.h>
   6 #include <stdint.h>
   7 #include <stdlib.h>
   8 #include <unistd.h>
   9
  10 #include "sd-bus.h"
  11
  12 #include "alloc-util.h"
  13 #include "bus-error.h"
  14 #include "bus-util.h"
  15 #include "cgroup-show.h"
  16 #include "cgroup-util.h"
  17 #include "fd-util.h"
  18 #include "fileio.h"
  19 #include "hashmap.h"
  20 #include "main-func.h"
  21 #include "missing_sched.h"
  22 #include "parse-util.h"
  23 #include "path-util.h"
  24 #include "pretty-print.h"
  25 #include "process-util.h"
  26 #include "procfs-util.h"
  27 #include "sort-util.h"
  28 #include "stdio-util.h"
  29 #include "strv.h"
  30 #include "terminal-util.h"
  31 #include "unit-name.h"
  32 #include "virt.h"
  33
  34 typedef struct Group {
  35         char *path;
  36
  37         bool n_tasks_valid:1;
  38         bool cpu_valid:1;
  39         bool memory_valid:1;
  40         bool io_valid:1;
  41
  42         uint64_t n_tasks;
  43
  44         unsigned cpu_iteration;
  45         nsec_t cpu_usage;
  46         nsec_t cpu_timestamp;
  47         double cpu_fraction;
  48
  49         uint64_t memory;
  50
  51         unsigned io_iteration;
  52         uint64_t io_input, io_output;
  53         nsec_t io_timestamp;
  54         uint64_t io_input_bps, io_output_bps;
  55 } Group;
  56
  57 static unsigned arg_depth = 3;
  58 static unsigned arg_iterations = (unsigned) -1;
  59 static bool arg_batch = false;
  60 static bool arg_raw = false;
  61 static usec_t arg_delay = 1*USEC_PER_SEC;
  62 static char* arg_machine = NULL;
  63 static char* arg_root = NULL;
  64 static bool arg_recursive = true;
  65 static bool arg_recursive_unset = false;
  66
  67 static enum {
  68         COUNT_PIDS,
  69         COUNT_USERSPACE_PROCESSES,
  70         COUNT_ALL_PROCESSES,
  71 } arg_count = COUNT_PIDS;
  72
  73 static enum {
  74         ORDER_PATH,
  75         ORDER_TASKS,
  76         ORDER_CPU,
  77         ORDER_MEMORY,
  78         ORDER_IO,
  79 } arg_order = ORDER_CPU;
  80
  81 static enum {
  82         CPU_PERCENT,
  83         CPU_TIME,
  84 } arg_cpu_type = CPU_PERCENT;
  85
  86 static Group *group_free(Group *g) {
  87         if (!g)
  88                 return NULL;
  89
  90         free(g->path);
  91         return mfree(g);
  92 }
  93
  94
  95 static const char *maybe_format_timespan(char *buf, size_t l, usec_t t, usec_t accuracy) {
  96         if (arg_raw) {
  97                snprintf(buf, l, USEC_FMT, t);
  98                return buf;
  99         }
 100         return format_timespan(buf, l, t, accuracy);
 101 }
 102
 103 static const char *maybe_format_bytes(char *buf, size_t l, bool is_valid, uint64_t t) {
 104         if (!is_valid)
 105                 return "-";
 106         if (arg_raw) {
 107                 snprintf(buf, l, "%" PRIu64, t);
 108                 return buf;
 109         }
 110         return format_bytes(buf, l, t);
 111 }
 112
 113 static bool is_root_cgroup(const char *path) {
 114
 115         /* Returns true if the specified path belongs to the root cgroup. The root cgroup is special on cgroup v2 as it
 116          * carries only very few attributes in order not to export multiple truth about system state as most
 117          * information is available elsewhere in /proc anyway. We need to be able to deal with that, and need to get
 118          * our data from different sources in that case.
 119          *
 120          * There's one extra complication in all of this, though 😣: if the path to the cgroup indicates we are in the
 121          * root cgroup this might actually not be the case, because cgroup namespacing might be in effect
 122          * (CLONE_NEWCGROUP). Since there's no nice way to distinguish a real cgroup root from a fake namespaced one we
 123          * do an explicit container check here, under the assumption that CLONE_NEWCGROUP is generally used when
 124          * container managers are used too.
 125          *
 126          * Note that checking for a container environment is kinda ugly, since in theory people could use cgtop from
 127          * inside a container where cgroup namespacing is turned off to watch the host system. However, that's mostly a
 128          * theoretic usecase, and if people actually try all they'll lose is accounting for the top-level cgroup. Which
 129          * isn't too bad. */
 130
 131         if (detect_container() > 0)
 132                 return false;
 133
 134         return empty_or_root(path);
 135 }
 136
 137 static int process(
 138                 const char *controller,
 139                 const char *path,
 140                 Hashmap *a,
 141                 Hashmap *b,
 142                 unsigned iteration,
 143                 Group **ret) {
 144
 145         Group *g;
 146         int r, all_unified;
 147
 148         assert(controller);
 149         assert(path);
 150         assert(a);
 151
 152         all_unified = cg_all_unified();
 153         if (all_unified < 0)
 154                 return all_unified;
 155
 156         g = hashmap_get(a, path);
 157         if (!g) {
 158                 g = hashmap_get(b, path);
 159                 if (!g) {
 160                         g = new0(Group, 1);
 161                         if (!g)
 162                                 return -ENOMEM;
 163
 164                         g->path = strdup(path);
 165                         if (!g->path) {
 166                                 group_free(g);
 167                                 return -ENOMEM;
 168                         }
 169
 170                         r = hashmap_put(a, g->path, g);
 171                         if (r < 0) {
 172                                 group_free(g);
 173                                 return r;
 174                         }
 175                 } else {
 176                         r = hashmap_move_one(a, b, path);
 177                         if (r < 0)
 178                                 return r;
 179
 180                         g->cpu_valid = g->memory_valid = g->io_valid = g->n_tasks_valid = false;
 181                 }
 182         }
 183
 184         if (streq(controller, SYSTEMD_CGROUP_CONTROLLER) &&
 185             IN_SET(arg_count, COUNT_ALL_PROCESSES, COUNT_USERSPACE_PROCESSES)) {
 186                 _cleanup_fclose_ FILE *f = NULL;
 187                 pid_t pid;
 188
 189                 r = cg_enumerate_processes(controller, path, &f);
 190                 if (r == -ENOENT)
 191                         return 0;
 192                 if (r < 0)
 193                         return r;
 194
 195                 g->n_tasks = 0;
 196                 while (cg_read_pid(f, &pid) > 0) {
 197
 198                         if (arg_count == COUNT_USERSPACE_PROCESSES && is_kernel_thread(pid) > 0)
 199                                 continue;
 200
 201                         g->n_tasks++;
 202                 }
 203
 204                 if (g->n_tasks > 0)
 205                         g->n_tasks_valid = true;
 206
 207         } else if (streq(controller, "pids") && arg_count == COUNT_PIDS) {
 208
 209                 if (is_root_cgroup(path)) {
 210                         r = procfs_tasks_get_current(&g->n_tasks);
 211                         if (r < 0)
 212                                 return r;
 213                 } else {
 214                         _cleanup_free_ char *p = NULL, *v = NULL;
 215
 216                         r = cg_get_path(controller, path, "pids.current", &p);
 217                         if (r < 0)
 218                                 return r;
 219
 220                         r = read_one_line_file(p, &v);
 221                         if (r == -ENOENT)
 222                                 return 0;
 223                         if (r < 0)
 224                                 return r;
 225
 226                         r = safe_atou64(v, &g->n_tasks);
 227                         if (r < 0)
 228                                 return r;
 229                 }
 230
 231                 if (g->n_tasks > 0)
 232                         g->n_tasks_valid = true;
 233
 234         } else if (streq(controller, "memory")) {
 235
 236                 if (is_root_cgroup(path)) {
 237                         r = procfs_memory_get_used(&g->memory);
 238                         if (r < 0)
 239                                 return r;
 240                 } else {
 241                         _cleanup_free_ char *p = NULL, *v = NULL;
 242
 243                         if (all_unified)
 244                                 r = cg_get_path(controller, path, "memory.current", &p);
 245                         else
 246                                 r = cg_get_path(controller, path, "memory.usage_in_bytes", &p);
 247                         if (r < 0)
 248                                 return r;
 249
 250                         r = read_one_line_file(p, &v);
 251                         if (r == -ENOENT)
 252                                 return 0;
 253                         if (r < 0)
 254                                 return r;
 255
 256                         r = safe_atou64(v, &g->memory);
 257                         if (r < 0)
 258                                 return r;
 259                 }
 260
 261                 if (g->memory > 0)
 262                         g->memory_valid = true;
 263
 264         } else if ((streq(controller, "io") && all_unified) ||
 265                    (streq(controller, "blkio") && !all_unified)) {
 266                 _cleanup_fclose_ FILE *f = NULL;
 267                 _cleanup_free_ char *p = NULL;
 268                 uint64_t wr = 0, rd = 0;
 269                 nsec_t timestamp;
 270
 271                 r = cg_get_path(controller, path, all_unified ? "io.stat" : "blkio.io_service_bytes", &p);
 272                 if (r < 0)
 273                         return r;
 274
 275                 f = fopen(p, "re");
 276                 if (!f) {
 277                         if (errno == ENOENT)
 278                                 return 0;
 279                         return -errno;
 280                 }
 281
 282                 for (;;) {
 283                         _cleanup_free_ char *line = NULL;
 284                         uint64_t k, *q;
 285                         char *l;
 286
 287                         r = read_line(f, LONG_LINE_MAX, &line);
 288                         if (r < 0)
 289                                 return r;
 290                         if (r == 0)
 291                                 break;
 292
 293                         /* Trim and skip the device */
 294                         l = strstrip(line);
 295                         l += strcspn(l, WHITESPACE);
 296                         l += strspn(l, WHITESPACE);
 297
 298                         if (all_unified) {
 299                                 while (!isempty(l)) {
 300                                         if (sscanf(l, "rbytes=%" SCNu64, &k))
 301                                                 rd += k;
 302                                         else if (sscanf(l, "wbytes=%" SCNu64, &k))
 303                                                 wr += k;
 304
 305                                         l += strcspn(l, WHITESPACE);
 306                                         l += strspn(l, WHITESPACE);
 307                                 }
 308                         } else {
 309                                 if (first_word(l, "Read")) {
 310                                         l += 4;
 311                                         q = &rd;
 312                                 } else if (first_word(l, "Write")) {
 313                                         l += 5;
 314                                         q = &wr;
 315                                 } else
 316                                         continue;
 317
 318                                 l += strspn(l, WHITESPACE);
 319                                 r = safe_atou64(l, &k);
 320                                 if (r < 0)
 321                                         continue;
 322
 323                                 *q += k;
 324                         }
 325                 }
 326
 327                 timestamp = now_nsec(CLOCK_MONOTONIC);
 328
 329                 if (g->io_iteration == iteration - 1) {
 330                         uint64_t x, yr, yw;
 331
 332                         x = (uint64_t) (timestamp - g->io_timestamp);
 333                         if (x < 1)
 334                                 x = 1;
 335
 336                         if (rd > g->io_input)
 337                                 yr = rd - g->io_input;
 338                         else
 339                                 yr = 0;
 340
 341                         if (wr > g->io_output)
 342                                 yw = wr - g->io_output;
 343                         else
 344                                 yw = 0;
 345
 346                         if (yr > 0 || yw > 0) {
 347                                 g->io_input_bps = (yr * 1000000000ULL) / x;
 348                                 g->io_output_bps = (yw * 1000000000ULL) / x;
 349                                 g->io_valid = true;
 350                         }
 351                 }
 352
 353                 g->io_input = rd;
 354                 g->io_output = wr;
 355                 g->io_timestamp = timestamp;
 356                 g->io_iteration = iteration;
 357         } else if (STR_IN_SET(controller, "cpu", "cpuacct") || cpu_accounting_is_cheap()) {
 358                 _cleanup_free_ char *p = NULL, *v = NULL;
 359                 uint64_t new_usage;
 360                 nsec_t timestamp;
 361
 362                 if (is_root_cgroup(path)) {
 363                         r = procfs_cpu_get_usage(&new_usage);
 364                         if (r < 0)
 365                                 return r;
 366                 } else if (all_unified) {
 367                         _cleanup_free_ char *val = NULL;
 368
 369                         if (!streq(controller, "cpu"))
 370                                 return 0;
 371
 372                         r = cg_get_keyed_attribute("cpu", path, "cpu.stat", STRV_MAKE("usage_usec"), &val);
 373                         if (IN_SET(r, -ENOENT, -ENXIO))
 374                                 return 0;
 375                         if (r < 0)
 376                                 return r;
 377
 378                         r = safe_atou64(val, &new_usage);
 379                         if (r < 0)
 380                                 return r;
 381
 382                         new_usage *= NSEC_PER_USEC;
 383                 } else {
 384                         if (!streq(controller, "cpuacct"))
 385                                 return 0;
 386
 387                         r = cg_get_path(controller, path, "cpuacct.usage", &p);
 388                         if (r < 0)
 389                                 return r;
 390
 391                         r = read_one_line_file(p, &v);
 392                         if (r == -ENOENT)
 393                                 return 0;
 394                         if (r < 0)
 395                                 return r;
 396
 397                         r = safe_atou64(v, &new_usage);
 398                         if (r < 0)
 399                                 return r;
 400                 }
 401
 402                 timestamp = now_nsec(CLOCK_MONOTONIC);
 403
 404                 if (g->cpu_iteration == iteration - 1 &&
 405                     (nsec_t) new_usage > g->cpu_usage) {
 406
 407                         nsec_t x, y;
 408
 409                         x = timestamp - g->cpu_timestamp;
 410                         if (x < 1)
 411                                 x = 1;
 412
 413                         y = (nsec_t) new_usage - g->cpu_usage;
 414                         g->cpu_fraction = (double) y / (double) x;
 415                         g->cpu_valid = true;
 416                 }
 417
 418                 g->cpu_usage = (nsec_t) new_usage;
 419                 g->cpu_timestamp = timestamp;
 420                 g->cpu_iteration = iteration;
 421
 422         }
 423
 424         if (ret)
 425                 *ret = g;
 426
 427         return 0;
 428 }
 429
 430 static int refresh_one(
 431                 const char *controller,
 432                 const char *path,
 433                 Hashmap *a,
 434                 Hashmap *b,
 435                 unsigned iteration,
 436                 unsigned depth,
 437                 Group **ret) {
 438
 439         _cleanup_closedir_ DIR *d = NULL;
 440         Group *ours = NULL;
 441         int r;
 442
 443         assert(controller);
 444         assert(path);
 445         assert(a);
 446
 447         if (depth > arg_depth)
 448                 return 0;
 449
 450         r = process(controller, path, a, b, iteration, &ours);
 451         if (r < 0)
 452                 return r;
 453
 454         r = cg_enumerate_subgroups(controller, path, &d);
 455         if (r == -ENOENT)
 456                 return 0;
 457         if (r < 0)
 458                 return r;
 459
 460         for (;;) {
 461                 _cleanup_free_ char *fn = NULL, *p = NULL;
 462                 Group *child = NULL;
 463
 464                 r = cg_read_subgroup(d, &fn);
 465                 if (r < 0)
 466                         return r;
 467                 if (r == 0)
 468                         break;
 469
 470                 p = path_join(path, fn);
 471                 if (!p)
 472                         return -ENOMEM;
 473
 474                 path_simplify(p, false);
 475
 476                 r = refresh_one(controller, p, a, b, iteration, depth + 1, &child);
 477                 if (r < 0)
 478                         return r;
 479
 480                 if (arg_recursive &&
 481                     IN_SET(arg_count, COUNT_ALL_PROCESSES, COUNT_USERSPACE_PROCESSES) &&
 482                     child &&
 483                     child->n_tasks_valid &&
 484                     streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
 485
 486                         /* Recursively sum up processes */
 487
 488                         if (ours->n_tasks_valid)
 489                                 ours->n_tasks += child->n_tasks;
 490                         else {
 491                                 ours->n_tasks = child->n_tasks;
 492                                 ours->n_tasks_valid = true;
 493                         }
 494                 }
 495         }
 496
 497         if (ret)
 498                 *ret = ours;
 499
 500         return 1;
 501 }
 502
 503 static int refresh(const char *root, Hashmap *a, Hashmap *b, unsigned iteration) {
 504         const char *c;
 505         int r;
 506
 507         FOREACH_STRING(c, SYSTEMD_CGROUP_CONTROLLER, "cpu", "cpuacct", "memory", "io", "blkio", "pids") {
 508                 r = refresh_one(c, root, a, b, iteration, 0, NULL);
 509                 if (r < 0)
 510                         return r;
 511         }
 512
 513         return 0;
 514 }
 515
 516 static int group_compare(Group * const *a, Group * const *b) {
 517         const Group *x = *a, *y = *b;
 518         int r;
 519
 520         if (arg_order != ORDER_TASKS || arg_recursive) {
 521                 /* Let's make sure that the parent is always before
 522                  * the child. Except when ordering by tasks and
 523                  * recursive summing is off, since that is actually
 524                  * not accumulative for all children. */
 525
 526                 if (path_startswith(empty_to_root(y->path), empty_to_root(x->path)))
 527                         return -1;
 528                 if (path_startswith(empty_to_root(x->path), empty_to_root(y->path)))
 529                         return 1;
 530         }
 531
 532         switch (arg_order) {
 533
 534         case ORDER_PATH:
 535                 break;
 536
 537         case ORDER_CPU:
 538                 if (arg_cpu_type == CPU_PERCENT) {
 539                         if (x->cpu_valid && y->cpu_valid) {
 540                                 r = CMP(y->cpu_fraction, x->cpu_fraction);
 541                                 if (r != 0)
 542                                         return r;
 543                         } else if (x->cpu_valid)
 544                                 return -1;
 545                         else if (y->cpu_valid)
 546                                 return 1;
 547                 } else {
 548                         r = CMP(y->cpu_usage, x->cpu_usage);
 549                         if (r != 0)
 550                                 return r;
 551                 }
 552
 553                 break;
 554
 555         case ORDER_TASKS:
 556                 if (x->n_tasks_valid && y->n_tasks_valid) {
 557                         r = CMP(y->n_tasks, x->n_tasks);
 558                         if (r != 0)
 559                                 return r;
 560                 } else if (x->n_tasks_valid)
 561                         return -1;
 562                 else if (y->n_tasks_valid)
 563                         return 1;
 564
 565                 break;
 566
 567         case ORDER_MEMORY:
 568                 if (x->memory_valid && y->memory_valid) {
 569                         r = CMP(y->memory, x->memory);
 570                         if (r != 0)
 571                                 return r;
 572                 } else if (x->memory_valid)
 573                         return -1;
 574                 else if (y->memory_valid)
 575                         return 1;
 576
 577                 break;
 578
 579         case ORDER_IO:
 580                 if (x->io_valid && y->io_valid) {
 581                         r = CMP(y->io_input_bps + y->io_output_bps, x->io_input_bps + x->io_output_bps);
 582                         if (r != 0)
 583                                 return r;
 584                 } else if (x->io_valid)
 585                         return -1;
 586                 else if (y->io_valid)
 587                         return 1;
 588         }
 589
 590         return path_compare(x->path, y->path);
 591 }
 592
 593 static void display(Hashmap *a) {
 594         Group *g;
 595         Group **array;
 596         signed path_columns;
 597         unsigned rows, n = 0, j, maxtcpu = 0, maxtpath = 3; /* 3 for ellipsize() to work properly */
 598         char buffer[MAX4(21U, FORMAT_BYTES_MAX, FORMAT_TIMESPAN_MAX, DECIMAL_STR_MAX(usec_t))];
 599
 600         assert(a);
 601
 602         if (!terminal_is_dumb())
 603                 fputs(ANSI_HOME_CLEAR, stdout);
 604
 605         array = newa(Group*, hashmap_size(a));
 606
 607         HASHMAP_FOREACH(g, a)
 608                 if (g->n_tasks_valid || g->cpu_valid || g->memory_valid || g->io_valid)
 609                         array[n++] = g;
 610
 611         typesafe_qsort(array, n, group_compare);
 612
 613         /* Find the longest names in one run */
 614         for (j = 0; j < n; j++) {
 615                 unsigned cputlen, pathtlen;
 616
 617                 maybe_format_timespan(buffer, sizeof(buffer), (usec_t) (array[j]->cpu_usage / NSEC_PER_USEC), 0);
 618                 cputlen = strlen(buffer);
 619                 maxtcpu = MAX(maxtcpu, cputlen);
 620
 621                 pathtlen = strlen(array[j]->path);
 622                 maxtpath = MAX(maxtpath, pathtlen);
 623         }
 624
 625         if (arg_cpu_type == CPU_PERCENT)
 626                 xsprintf(buffer, "%6s", "%CPU");
 627         else
 628                 xsprintf(buffer, "%*s", maxtcpu, "CPU Time");
 629
 630         rows = lines();
 631         if (rows <= 10)
 632                 rows = 10;
 633
 634         if (on_tty()) {
 635                 const char *on, *off;
 636
 637                 path_columns = columns() - 36 - strlen(buffer);
 638                 if (path_columns < 10)
 639                         path_columns = 10;
 640
 641                 on = ansi_highlight_underline();
 642                 off = ansi_underline();
 643
 644                 printf("%s%s%-*s%s %s%7s%s %s%s%s %s%8s%s %s%8s%s %s%8s%s%s\n",
 645                        ansi_underline(),
 646                        arg_order == ORDER_PATH ? on : "", path_columns, "Control Group",
 647                        arg_order == ORDER_PATH ? off : "",
 648                        arg_order == ORDER_TASKS ? on : "", arg_count == COUNT_PIDS ? "Tasks" : arg_count == COUNT_USERSPACE_PROCESSES ? "Procs" : "Proc+",
 649                        arg_order == ORDER_TASKS ? off : "",
 650                        arg_order == ORDER_CPU ? on : "", buffer,
 651                        arg_order == ORDER_CPU ? off : "",
 652                        arg_order == ORDER_MEMORY ? on : "", "Memory",
 653                        arg_order == ORDER_MEMORY ? off : "",
 654                        arg_order == ORDER_IO ? on : "", "Input/s",
 655                        arg_order == ORDER_IO ? off : "",
 656                        arg_order == ORDER_IO ? on : "", "Output/s",
 657                        arg_order == ORDER_IO ? off : "",
 658                        ansi_normal());
 659         } else
 660                 path_columns = maxtpath;
 661
 662         for (j = 0; j < n; j++) {
 663                 _cleanup_free_ char *ellipsized = NULL;
 664                 const char *path;
 665
 666                 if (on_tty() && j + 6 > rows)
 667                         break;
 668
 669                 g = array[j];
 670
 671                 path = empty_to_root(g->path);
 672                 ellipsized = ellipsize(path, path_columns, 33);
 673                 printf("%-*s", path_columns, ellipsized ?: path);
 674
 675                 if (g->n_tasks_valid)
 676                         printf(" %7" PRIu64, g->n_tasks);
 677                 else
 678                         fputs("       -", stdout);
 679
 680                 if (arg_cpu_type == CPU_PERCENT) {
 681                         if (g->cpu_valid)
 682                                 printf(" %6.1f", g->cpu_fraction*100);
 683                         else
 684                                 fputs("      -", stdout);
 685                 } else
 686                         printf(" %*s", maxtcpu, maybe_format_timespan(buffer, sizeof(buffer), (usec_t) (g->cpu_usage / NSEC_PER_USEC), 0));
 687
 688                 printf(" %8s", maybe_format_bytes(buffer, sizeof(buffer), g->memory_valid, g->memory));
 689                 printf(" %8s", maybe_format_bytes(buffer, sizeof(buffer), g->io_valid, g->io_input_bps));
 690                 printf(" %8s", maybe_format_bytes(buffer, sizeof(buffer), g->io_valid, g->io_output_bps));
 691
 692                 putchar('\n');
 693         }
 694 }
 695
 696 static int help(void) {
 697         _cleanup_free_ char *link = NULL;
 698         int r;
 699
 700         r = terminal_urlify_man("systemd-cgtop", "1", &link);
 701         if (r < 0)
 702                 return log_oom();
 703
 704         printf("%s [OPTIONS...] [CGROUP]\n\n"
 705                "Show top control groups by their resource usage.\n\n"
 706                "  -h --help           Show this help\n"
 707                "     --version        Show package version\n"
 708                "  -p --order=path     Order by path\n"
 709                "  -t --order=tasks    Order by number of tasks/processes\n"
 710                "  -c --order=cpu      Order by CPU load (default)\n"
 711                "  -m --order=memory   Order by memory load\n"
 712                "  -i --order=io       Order by IO load\n"
 713                "  -r --raw            Provide raw (not human-readable) numbers\n"
 714                "     --cpu=percentage Show CPU usage as percentage (default)\n"
 715                "     --cpu=time       Show CPU usage as time\n"
 716                "  -P                  Count userspace processes instead of tasks (excl. kernel)\n"
 717                "  -k                  Count all processes instead of tasks (incl. kernel)\n"
 718                "     --recursive=BOOL Sum up process count recursively\n"
 719                "  -d --delay=DELAY    Delay between updates\n"
 720                "  -n --iterations=N   Run for N iterations before exiting\n"
 721                "  -1                  Shortcut for --iterations=1\n"
 722                "  -b --batch          Run in batch mode, accepting no input\n"
 723                "     --depth=DEPTH    Maximum traversal depth (default: %u)\n"
 724                "  -M --machine=       Show container\n"
 725                "\nSee the %s for details.\n"
 726                , program_invocation_short_name
 727                , arg_depth
 728                , link
 729         );
 730
 731         return 0;
 732 }
 733
 734 static int parse_argv(int argc, char *argv[]) {
 735         enum {
 736                 ARG_VERSION = 0x100,
 737                 ARG_DEPTH,
 738                 ARG_CPU_TYPE,
 739                 ARG_ORDER,
 740                 ARG_RECURSIVE,
 741         };
 742
 743         static const struct option options[] = {
 744                 { "help",         no_argument,       NULL, 'h'           },
 745                 { "version",      no_argument,       NULL, ARG_VERSION   },
 746                 { "delay",        required_argument, NULL, 'd'           },
 747                 { "iterations",   required_argument, NULL, 'n'           },
 748                 { "batch",        no_argument,       NULL, 'b'           },
 749                 { "raw",          no_argument,       NULL, 'r'           },
 750                 { "depth",        required_argument, NULL, ARG_DEPTH     },
 751                 { "cpu",          optional_argument, NULL, ARG_CPU_TYPE  },
 752                 { "order",        required_argument, NULL, ARG_ORDER     },
 753                 { "recursive",    required_argument, NULL, ARG_RECURSIVE },
 754                 { "machine",      required_argument, NULL, 'M'           },
 755                 {}
 756         };
 757
 758         int c, r;
 759
 760         assert(argc >= 1);
 761         assert(argv);
 762
 763         while ((c = getopt_long(argc, argv, "hptcmin:brd:kPM:1", options, NULL)) >= 0)
 764
 765                 switch (c) {
 766
 767                 case 'h':
 768                         return help();
 769
 770                 case ARG_VERSION:
 771                         return version();
 772
 773                 case ARG_CPU_TYPE:
 774                         if (optarg) {
 775                                 if (streq(optarg, "time"))
 776                                         arg_cpu_type = CPU_TIME;
 777                                 else if (streq(optarg, "percentage"))
 778                                         arg_cpu_type = CPU_PERCENT;
 779                                 else
 780                                         return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
 781                                                                "Unknown argument to --cpu=: %s",
 782                                                                optarg);
 783                         } else
 784                                 arg_cpu_type = CPU_TIME;
 785
 786                         break;
 787
 788                 case ARG_DEPTH:
 789                         r = safe_atou(optarg, &arg_depth);
 790                         if (r < 0)
 791                                 return log_error_errno(r, "Failed to parse depth parameter '%s': %m", optarg);
 792
 793                         break;
 794
 795                 case 'd':
 796                         r = parse_sec(optarg, &arg_delay);
 797                         if (r < 0)
 798                                 return log_error_errno(r, "Failed to parse delay parameter '%s': %m", optarg);
 799                         if (arg_delay <= 0)
 800                                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
 801                                                        "Invalid delay parameter '%s'",
 802                                                        optarg);
 803
 804                         break;
 805
 806                 case 'n':
 807                         r = safe_atou(optarg, &arg_iterations);
 808                         if (r < 0)
 809                                 return log_error_errno(r, "Failed to parse iterations parameter '%s': %m", optarg);
 810
 811                         break;
 812
 813                 case '1':
 814                         arg_iterations = 1;
 815                         break;
 816
 817                 case 'b':
 818                         arg_batch = true;
 819                         break;
 820
 821                 case 'r':
 822                         arg_raw = true;
 823                         break;
 824
 825                 case 'p':
 826                         arg_order = ORDER_PATH;
 827                         break;
 828
 829                 case 't':
 830                         arg_order = ORDER_TASKS;
 831                         break;
 832
 833                 case 'c':
 834                         arg_order = ORDER_CPU;
 835                         break;
 836
 837                 case 'm':
 838                         arg_order = ORDER_MEMORY;
 839                         break;
 840
 841                 case 'i':
 842                         arg_order = ORDER_IO;
 843                         break;
 844
 845                 case ARG_ORDER:
 846                         if (streq(optarg, "path"))
 847                                 arg_order = ORDER_PATH;
 848                         else if (streq(optarg, "tasks"))
 849                                 arg_order = ORDER_TASKS;
 850                         else if (streq(optarg, "cpu"))
 851                                 arg_order = ORDER_CPU;
 852                         else if (streq(optarg, "memory"))
 853                                 arg_order = ORDER_MEMORY;
 854                         else if (streq(optarg, "io"))
 855                                 arg_order = ORDER_IO;
 856                         else
 857                                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
 858                                                        "Invalid argument to --order=: %s",
 859                                                        optarg);
 860                         break;
 861
 862                 case 'k':
 863                         arg_count = COUNT_ALL_PROCESSES;
 864                         break;
 865
 866                 case 'P':
 867                         arg_count = COUNT_USERSPACE_PROCESSES;
 868                         break;
 869
 870                 case ARG_RECURSIVE:
 871                         r = parse_boolean(optarg);
 872                         if (r < 0)
 873                                 return log_error_errno(r, "Failed to parse --recursive= argument '%s': %m", optarg);
 874
 875                         arg_recursive = r;
 876                         arg_recursive_unset = r == 0;
 877                         break;
 878
 879                 case 'M':
 880                         arg_machine = optarg;
 881                         break;
 882
 883                 case '?':
 884                         return -EINVAL;
 885
 886                 default:
 887                         assert_not_reached("Unhandled option");
 888                 }
 889
 890         if (optind == argc - 1)
 891                 arg_root = argv[optind];
 892         else if (optind < argc)
 893                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
 894                                        "Too many arguments.");
 895
 896         return 1;
 897 }
 898
 899 static const char* counting_what(void) {
 900         if (arg_count == COUNT_PIDS)
 901                 return "tasks";
 902         else if (arg_count == COUNT_ALL_PROCESSES)
 903                 return "all processes (incl. kernel)";
 904         else
 905                 return "userspace processes (excl. kernel)";
 906 }
 907
 908 DEFINE_PRIVATE_HASH_OPS_WITH_VALUE_DESTRUCTOR(group_hash_ops, char, path_hash_func, path_compare, Group, group_free);
 909
 910 static int run(int argc, char *argv[]) {
 911         _cleanup_hashmap_free_ Hashmap *a = NULL, *b = NULL;
 912         unsigned iteration = 0;
 913         usec_t last_refresh = 0;
 914         bool quit = false, immediate_refresh = false;
 915         _cleanup_free_ char *root = NULL;
 916         CGroupMask mask;
 917         int r;
 918
 919         log_setup_cli();
 920
 921         r = parse_argv(argc, argv);
 922         if (r <= 0)
 923                 return r;
 924
 925         r = cg_mask_supported(&mask);
 926         if (r < 0)
 927                 return log_error_errno(r, "Failed to determine supported controllers: %m");
 928
 929         arg_count = (mask & CGROUP_MASK_PIDS) ? COUNT_PIDS : COUNT_USERSPACE_PROCESSES;
 930
 931         if (arg_recursive_unset && arg_count == COUNT_PIDS)
 932                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
 933                                        "Non-recursive counting is only supported when counting processes, not tasks. Use -P or -k.");
 934
 935         r = show_cgroup_get_path_and_warn(arg_machine, arg_root, &root);
 936         if (r < 0)
 937                 return log_error_errno(r, "Failed to get root control group path: %m");
 938         log_debug("CGroup path: %s", root);
 939
 940         a = hashmap_new(&group_hash_ops);
 941         b = hashmap_new(&group_hash_ops);
 942         if (!a || !b)
 943                 return log_oom();
 944
 945         signal(SIGWINCH, columns_lines_cache_reset);
 946
 947         if (arg_iterations == (unsigned) -1)
 948                 arg_iterations = on_tty() ? 0 : 1;
 949
 950         while (!quit) {
 951                 usec_t t;
 952                 char key;
 953                 char h[FORMAT_TIMESPAN_MAX];
 954
 955                 t = now(CLOCK_MONOTONIC);
 956
 957                 if (t >= last_refresh + arg_delay || immediate_refresh) {
 958
 959                         r = refresh(root, a, b, iteration++);
 960                         if (r < 0)
 961                                 return log_error_errno(r, "Failed to refresh: %m");
 962
 963                         hashmap_clear(b);
 964                         SWAP_TWO(a, b);
 965
 966                         last_refresh = t;
 967                         immediate_refresh = false;
 968                 }
 969
 970                 display(b);
 971
 972                 if (arg_iterations && iteration >= arg_iterations)
 973                         break;
 974
 975                 if (!on_tty()) /* non-TTY: Empty newline as delimiter between polls */
 976                         fputs("\n", stdout);
 977                 fflush(stdout);
 978
 979                 if (arg_batch)
 980                         (void) usleep(last_refresh + arg_delay - t);
 981                 else {
 982                         r = read_one_char(stdin, &key, last_refresh + arg_delay - t, NULL);
 983                         if (r == -ETIMEDOUT)
 984                                 continue;
 985                         if (r < 0)
 986                                 return log_error_errno(r, "Couldn't read key: %m");
 987                 }
 988
 989                 if (on_tty()) { /* TTY: Clear any user keystroke */
 990                         fputs("\r \r", stdout);
 991                         fflush(stdout);
 992                 }
 993
 994                 if (arg_batch)
 995                         continue;
 996
 997                 switch (key) {
 998
 999                 case ' ':
1000                         immediate_refresh = true;
1001                         break;
1002
1003                 case 'q':
1004                         quit = true;
1005                         break;
1006
1007                 case 'p':
1008                         arg_order = ORDER_PATH;
1009                         break;
1010
1011                 case 't':
1012                         arg_order = ORDER_TASKS;
1013                         break;
1014
1015                 case 'c':
1016                         arg_order = ORDER_CPU;
1017                         break;
1018
1019                 case 'm':
1020                         arg_order = ORDER_MEMORY;
1021                         break;
1022
1023                 case 'i':
1024                         arg_order = ORDER_IO;
1025                         break;
1026
1027                 case '%':
1028                         arg_cpu_type = arg_cpu_type == CPU_TIME ? CPU_PERCENT : CPU_TIME;
1029                         break;
1030
1031                 case 'k':
1032                         arg_count = arg_count != COUNT_ALL_PROCESSES ? COUNT_ALL_PROCESSES : COUNT_PIDS;
1033                         fprintf(stdout, "\nCounting: %s.", counting_what());
1034                         fflush(stdout);
1035                         sleep(1);
1036                         break;
1037
1038                 case 'P':
1039                         arg_count = arg_count != COUNT_USERSPACE_PROCESSES ? COUNT_USERSPACE_PROCESSES : COUNT_PIDS;
1040                         fprintf(stdout, "\nCounting: %s.", counting_what());
1041                         fflush(stdout);
1042                         sleep(1);
1043                         break;
1044
1045                 case 'r':
1046                         if (arg_count == COUNT_PIDS)
1047                                 fprintf(stdout, "\n\aCannot toggle recursive counting, not available in task counting mode.");
1048                         else {
1049                                 arg_recursive = !arg_recursive;
1050                                 fprintf(stdout, "\nRecursive process counting: %s", yes_no(arg_recursive));
1051                         }
1052                         fflush(stdout);
1053                         sleep(1);
1054                         break;
1055
1056                 case '+':
1057                         if (arg_delay < USEC_PER_SEC)
1058                                 arg_delay += USEC_PER_MSEC*250;
1059                         else
1060                                 arg_delay += USEC_PER_SEC;
1061
1062                         fprintf(stdout, "\nIncreased delay to %s.", format_timespan(h, sizeof(h), arg_delay, 0));
1063                         fflush(stdout);
1064                         sleep(1);
1065                         break;
1066
1067                 case '-':
1068                         if (arg_delay <= USEC_PER_MSEC*500)
1069                                 arg_delay = USEC_PER_MSEC*250;
1070                         else if (arg_delay < USEC_PER_MSEC*1250)
1071                                 arg_delay -= USEC_PER_MSEC*250;
1072                         else
1073                                 arg_delay -= USEC_PER_SEC;
1074
1075                         fprintf(stdout, "\nDecreased delay to %s.", format_timespan(h, sizeof(h), arg_delay, 0));
1076                         fflush(stdout);
1077                         sleep(1);
1078                         break;
1079
1080                 case '?':
1081                 case 'h':
1082
1083 #define ON ANSI_HIGHLIGHT
1084 #define OFF ANSI_NORMAL
1085
1086                         fprintf(stdout,
1087                                 "\t<" ON "p" OFF "> By path; <" ON "t" OFF "> By tasks/procs; <" ON "c" OFF "> By CPU; <" ON "m" OFF "> By memory; <" ON "i" OFF "> By I/O\n"
1088                                 "\t<" ON "+" OFF "> Inc. delay; <" ON "-" OFF "> Dec. delay; <" ON "%%" OFF "> Toggle time; <" ON "SPACE" OFF "> Refresh\n"
1089                                 "\t<" ON "P" OFF "> Toggle count userspace processes; <" ON "k" OFF "> Toggle count all processes\n"
1090                                 "\t<" ON "r" OFF "> Count processes recursively; <" ON "q" OFF "> Quit");
1091                         fflush(stdout);
1092                         sleep(3);
1093                         break;
1094
1095                 default:
1096                         if (key < ' ')
1097                                 fprintf(stdout, "\nUnknown key '\\x%x'. Ignoring.", key);
1098                         else
1099                                 fprintf(stdout, "\nUnknown key '%c'. Ignoring.", key);
1100                         fflush(stdout);
1101                         sleep(1);
1102                         break;
1103                 }
1104         }
1105
1106         return 0;
1107 }
1108
1109 DEFINE_MAIN_FUNCTION(run);