src/cgtop/cgtop.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2
   3 #include <alloca.h>
   4 #include <errno.h>
   5 #include <getopt.h>
   6 #include <signal.h>
   7 #include <stdint.h>
   8 #include <stdlib.h>
   9 #include <string.h>
  10 #include <unistd.h>
  11
  12 #include "sd-bus.h"
  13
  14 #include "alloc-util.h"
  15 #include "bus-error.h"
  16 #include "bus-util.h"
  17 #include "cgroup-show.h"
  18 #include "cgroup-util.h"
  19 #include "fd-util.h"
  20 #include "fileio.h"
  21 #include "hashmap.h"
  22 #include "main-func.h"
  23 #include "parse-util.h"
  24 #include "path-util.h"
  25 #include "process-util.h"
  26 #include "procfs-util.h"
  27 #include "stdio-util.h"
  28 #include "strv.h"
  29 #include "terminal-util.h"
  30 #include "unit-name.h"
  31 #include "util.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 void group_free(Group *g) {
  87         assert(g);
  88
  89         free(g->path);
  90         free(g);
  91 }
  92
  93 static void group_hashmap_clear(Hashmap *h) {
  94         hashmap_clear_with_destructor(h, group_free);
  95 }
  96
  97 static void group_hashmap_free(Hashmap *h) {
  98         group_hashmap_clear(h);
  99         hashmap_free(h);
 100 }
 101
 102 DEFINE_TRIVIAL_CLEANUP_FUNC(Hashmap*, group_hashmap_free);
 103
 104 static const char *maybe_format_bytes(char *buf, size_t l, bool is_valid, uint64_t t) {
 105         if (!is_valid)
 106                 return "-";
 107         if (arg_raw) {
 108                 snprintf(buf, l, "%" PRIu64, t);
 109                 return buf;
 110         }
 111         return format_bytes(buf, l, t);
 112 }
 113
 114 static bool is_root_cgroup(const char *path) {
 115
 116         /* Returns true if the specified path belongs to the root cgroup. The root cgroup is special on cgroupsv2 as it
 117          * carries only very few attributes in order not to export multiple truth about system state as most
 118          * information is available elsewhere in /proc anyway. We need to be able to deal with that, and need to get
 119          * our data from different sources in that case.
 120          *
 121          * There's one extra complication in all of this, though 😣: if the path to the cgroup indicates we are in the
 122          * root cgroup this might actually not be the case, because cgroup namespacing might be in effect
 123          * (CLONE_NEWCGROUP). Since there's no nice way to distuingish a real cgroup root from a fake namespaced one we
 124          * do an explicit container check here, under the assumption that CLONE_NEWCGROUP is generally used when
 125          * container managers are used too.
 126          *
 127          * Note that checking for a container environment is kinda ugly, since in theory people could use cgtop from
 128          * inside a container where cgroup namespacing is turned off to watch the host system. However, that's mostly a
 129          * theoretic usecase, and if people actually try all they'll lose is accounting for the top-level cgroup. Which
 130          * isn't too bad. */
 131
 132         if (detect_container() > 0)
 133                 return false;
 134
 135         return empty_or_root(path);
 136 }
 137
 138 static int process(
 139                 const char *controller,
 140                 const char *path,
 141                 Hashmap *a,
 142                 Hashmap *b,
 143                 unsigned iteration,
 144                 Group **ret) {
 145
 146         Group *g;
 147         int r, all_unified;
 148
 149         assert(controller);
 150         assert(path);
 151         assert(a);
 152
 153         all_unified = cg_all_unified();
 154         if (all_unified < 0)
 155                 return all_unified;
 156
 157         g = hashmap_get(a, path);
 158         if (!g) {
 159                 g = hashmap_get(b, path);
 160                 if (!g) {
 161                         g = new0(Group, 1);
 162                         if (!g)
 163                                 return -ENOMEM;
 164
 165                         g->path = strdup(path);
 166                         if (!g->path) {
 167                                 group_free(g);
 168                                 return -ENOMEM;
 169                         }
 170
 171                         r = hashmap_put(a, g->path, g);
 172                         if (r < 0) {
 173                                 group_free(g);
 174                                 return r;
 175                         }
 176                 } else {
 177                         r = hashmap_move_one(a, b, path);
 178                         if (r < 0)
 179                                 return r;
 180
 181                         g->cpu_valid = g->memory_valid = g->io_valid = g->n_tasks_valid = false;
 182                 }
 183         }
 184
 185         if (streq(controller, SYSTEMD_CGROUP_CONTROLLER) &&
 186             IN_SET(arg_count, COUNT_ALL_PROCESSES, COUNT_USERSPACE_PROCESSES)) {
 187                 _cleanup_fclose_ FILE *f = NULL;
 188                 pid_t pid;
 189
 190                 r = cg_enumerate_processes(controller, path, &f);
 191                 if (r == -ENOENT)
 192                         return 0;
 193                 if (r < 0)
 194                         return r;
 195
 196                 g->n_tasks = 0;
 197                 while (cg_read_pid(f, &pid) > 0) {
 198
 199                         if (arg_count == COUNT_USERSPACE_PROCESSES && is_kernel_thread(pid) > 0)
 200                                 continue;
 201
 202                         g->n_tasks++;
 203                 }
 204
 205                 if (g->n_tasks > 0)
 206                         g->n_tasks_valid = true;
 207
 208         } else if (streq(controller, "pids") && arg_count == COUNT_PIDS) {
 209
 210                 if (is_root_cgroup(path)) {
 211                         r = procfs_tasks_get_current(&g->n_tasks);
 212                         if (r < 0)
 213                                 return r;
 214                 } else {
 215                         _cleanup_free_ char *p = NULL, *v = NULL;
 216
 217                         r = cg_get_path(controller, path, "pids.current", &p);
 218                         if (r < 0)
 219                                 return r;
 220
 221                         r = read_one_line_file(p, &v);
 222                         if (r == -ENOENT)
 223                                 return 0;
 224                         if (r < 0)
 225                                 return r;
 226
 227                         r = safe_atou64(v, &g->n_tasks);
 228                         if (r < 0)
 229                                 return r;
 230                 }
 231
 232                 if (g->n_tasks > 0)
 233                         g->n_tasks_valid = true;
 234
 235         } else if (STR_IN_SET(controller, "cpu", "cpuacct") || cpu_accounting_is_cheap()) {
 236                 _cleanup_free_ char *p = NULL, *v = NULL;
 237                 uint64_t new_usage;
 238                 nsec_t timestamp;
 239
 240                 if (is_root_cgroup(path)) {
 241                         r = procfs_cpu_get_usage(&new_usage);
 242                         if (r < 0)
 243                                 return r;
 244                 } else if (all_unified) {
 245                         _cleanup_free_ char *val = NULL;
 246
 247                         if (!streq(controller, "cpu"))
 248                                 return 0;
 249
 250                         r = cg_get_keyed_attribute("cpu", path, "cpu.stat", STRV_MAKE("usage_usec"), &val);
 251                         if (IN_SET(r, -ENOENT, -ENXIO))
 252                                 return 0;
 253                         if (r < 0)
 254                                 return r;
 255
 256                         r = safe_atou64(val, &new_usage);
 257                         if (r < 0)
 258                                 return r;
 259
 260                         new_usage *= NSEC_PER_USEC;
 261                 } else {
 262                         if (!streq(controller, "cpuacct"))
 263                                 return 0;
 264
 265                         r = cg_get_path(controller, path, "cpuacct.usage", &p);
 266                         if (r < 0)
 267                                 return r;
 268
 269                         r = read_one_line_file(p, &v);
 270                         if (r == -ENOENT)
 271                                 return 0;
 272                         if (r < 0)
 273                                 return r;
 274
 275                         r = safe_atou64(v, &new_usage);
 276                         if (r < 0)
 277                                 return r;
 278                 }
 279
 280                 timestamp = now_nsec(CLOCK_MONOTONIC);
 281
 282                 if (g->cpu_iteration == iteration - 1 &&
 283                     (nsec_t) new_usage > g->cpu_usage) {
 284
 285                         nsec_t x, y;
 286
 287                         x = timestamp - g->cpu_timestamp;
 288                         if (x < 1)
 289                                 x = 1;
 290
 291                         y = (nsec_t) new_usage - g->cpu_usage;
 292                         g->cpu_fraction = (double) y / (double) x;
 293                         g->cpu_valid = true;
 294                 }
 295
 296                 g->cpu_usage = (nsec_t) new_usage;
 297                 g->cpu_timestamp = timestamp;
 298                 g->cpu_iteration = iteration;
 299
 300         } else if (streq(controller, "memory")) {
 301
 302                 if (is_root_cgroup(path)) {
 303                         r = procfs_memory_get_current(&g->memory);
 304                         if (r < 0)
 305                                 return r;
 306                 } else {
 307                         _cleanup_free_ char *p = NULL, *v = NULL;
 308
 309                         if (all_unified)
 310                                 r = cg_get_path(controller, path, "memory.current", &p);
 311                         else
 312                                 r = cg_get_path(controller, path, "memory.usage_in_bytes", &p);
 313                         if (r < 0)
 314                                 return r;
 315
 316                         r = read_one_line_file(p, &v);
 317                         if (r == -ENOENT)
 318                                 return 0;
 319                         if (r < 0)
 320                                 return r;
 321
 322                         r = safe_atou64(v, &g->memory);
 323                         if (r < 0)
 324                                 return r;
 325                 }
 326
 327                 if (g->memory > 0)
 328                         g->memory_valid = true;
 329
 330         } else if ((streq(controller, "io") && all_unified) ||
 331                    (streq(controller, "blkio") && !all_unified)) {
 332                 _cleanup_fclose_ FILE *f = NULL;
 333                 _cleanup_free_ char *p = NULL;
 334                 uint64_t wr = 0, rd = 0;
 335                 nsec_t timestamp;
 336
 337                 r = cg_get_path(controller, path, all_unified ? "io.stat" : "blkio.io_service_bytes", &p);
 338                 if (r < 0)
 339                         return r;
 340
 341                 f = fopen(p, "re");
 342                 if (!f) {
 343                         if (errno == ENOENT)
 344                                 return 0;
 345                         return -errno;
 346                 }
 347
 348                 for (;;) {
 349                         _cleanup_free_ char *line = NULL;
 350                         uint64_t k, *q;
 351                         char *l;
 352
 353                         r = read_line(f, LONG_LINE_MAX, &line);
 354                         if (r < 0)
 355                                 return r;
 356                         if (r == 0)
 357                                 break;
 358
 359                         /* Trim and skip the device */
 360                         l = strstrip(line);
 361                         l += strcspn(l, WHITESPACE);
 362                         l += strspn(l, WHITESPACE);
 363
 364                         if (all_unified) {
 365                                 while (!isempty(l)) {
 366                                         if (sscanf(l, "rbytes=%" SCNu64, &k))
 367                                                 rd += k;
 368                                         else if (sscanf(l, "wbytes=%" SCNu64, &k))
 369                                                 wr += k;
 370
 371                                         l += strcspn(l, WHITESPACE);
 372                                         l += strspn(l, WHITESPACE);
 373                                 }
 374                         } else {
 375                                 if (first_word(l, "Read")) {
 376                                         l += 4;
 377                                         q = &rd;
 378                                 } else if (first_word(l, "Write")) {
 379                                         l += 5;
 380                                         q = &wr;
 381                                 } else
 382                                         continue;
 383
 384                                 l += strspn(l, WHITESPACE);
 385                                 r = safe_atou64(l, &k);
 386                                 if (r < 0)
 387                                         continue;
 388
 389                                 *q += k;
 390                         }
 391                 }
 392
 393                 timestamp = now_nsec(CLOCK_MONOTONIC);
 394
 395                 if (g->io_iteration == iteration - 1) {
 396                         uint64_t x, yr, yw;
 397
 398                         x = (uint64_t) (timestamp - g->io_timestamp);
 399                         if (x < 1)
 400                                 x = 1;
 401
 402                         if (rd > g->io_input)
 403                                 yr = rd - g->io_input;
 404                         else
 405                                 yr = 0;
 406
 407                         if (wr > g->io_output)
 408                                 yw = wr - g->io_output;
 409                         else
 410                                 yw = 0;
 411
 412                         if (yr > 0 || yw > 0) {
 413                                 g->io_input_bps = (yr * 1000000000ULL) / x;
 414                                 g->io_output_bps = (yw * 1000000000ULL) / x;
 415                                 g->io_valid = true;
 416                         }
 417                 }
 418
 419                 g->io_input = rd;
 420                 g->io_output = wr;
 421                 g->io_timestamp = timestamp;
 422                 g->io_iteration = iteration;
 423         }
 424
 425         if (ret)
 426                 *ret = g;
 427
 428         return 0;
 429 }
 430
 431 static int refresh_one(
 432                 const char *controller,
 433                 const char *path,
 434                 Hashmap *a,
 435                 Hashmap *b,
 436                 unsigned iteration,
 437                 unsigned depth,
 438                 Group **ret) {
 439
 440         _cleanup_closedir_ DIR *d = NULL;
 441         Group *ours = NULL;
 442         int r;
 443
 444         assert(controller);
 445         assert(path);
 446         assert(a);
 447
 448         if (depth > arg_depth)
 449                 return 0;
 450
 451         r = process(controller, path, a, b, iteration, &ours);
 452         if (r < 0)
 453                 return r;
 454
 455         r = cg_enumerate_subgroups(controller, path, &d);
 456         if (r == -ENOENT)
 457                 return 0;
 458         if (r < 0)
 459                 return r;
 460
 461         for (;;) {
 462                 _cleanup_free_ char *fn = NULL, *p = NULL;
 463                 Group *child = NULL;
 464
 465                 r = cg_read_subgroup(d, &fn);
 466                 if (r < 0)
 467                         return r;
 468                 if (r == 0)
 469                         break;
 470
 471                 p = strjoin(path, "/", fn);
 472                 if (!p)
 473                         return -ENOMEM;
 474
 475                 path_simplify(p, false);
 476
 477                 r = refresh_one(controller, p, a, b, iteration, depth + 1, &child);
 478                 if (r < 0)
 479                         return r;
 480
 481                 if (arg_recursive &&
 482                     IN_SET(arg_count, COUNT_ALL_PROCESSES, COUNT_USERSPACE_PROCESSES) &&
 483                     child &&
 484                     child->n_tasks_valid &&
 485                     streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
 486
 487                         /* Recursively sum up processes */
 488
 489                         if (ours->n_tasks_valid)
 490                                 ours->n_tasks += child->n_tasks;
 491                         else {
 492                                 ours->n_tasks = child->n_tasks;
 493                                 ours->n_tasks_valid = true;
 494                         }
 495                 }
 496         }
 497
 498         if (ret)
 499                 *ret = ours;
 500
 501         return 1;
 502 }
 503
 504 static int refresh(const char *root, Hashmap *a, Hashmap *b, unsigned iteration) {
 505         const char *c;
 506         int r;
 507
 508         FOREACH_STRING(c, SYSTEMD_CGROUP_CONTROLLER, "cpu", "cpuacct", "memory", "io", "blkio", "pids") {
 509                 r = refresh_one(c, root, a, b, iteration, 0, NULL);
 510                 if (r < 0)
 511                         return r;
 512         }
 513
 514         return 0;
 515 }
 516
 517 static int group_compare(Group * const *a, Group * const *b) {
 518         const Group *x = *a, *y = *b;
 519         int r;
 520
 521         if (arg_order != ORDER_TASKS || arg_recursive) {
 522                 /* Let's make sure that the parent is always before
 523                  * the child. Except when ordering by tasks and
 524                  * recursive summing is off, since that is actually
 525                  * not accumulative for all children. */
 526
 527                 if (path_startswith(empty_to_root(y->path), empty_to_root(x->path)))
 528                         return -1;
 529                 if (path_startswith(empty_to_root(x->path), empty_to_root(y->path)))
 530                         return 1;
 531         }
 532
 533         switch (arg_order) {
 534
 535         case ORDER_PATH:
 536                 break;
 537
 538         case ORDER_CPU:
 539                 if (arg_cpu_type == CPU_PERCENT) {
 540                         if (x->cpu_valid && y->cpu_valid) {
 541                                 r = CMP(y->cpu_fraction, x->cpu_fraction);
 542                                 if (r != 0)
 543                                         return r;
 544                         } else if (x->cpu_valid)
 545                                 return -1;
 546                         else if (y->cpu_valid)
 547                                 return 1;
 548                 } else {
 549                         r = CMP(y->cpu_usage, x->cpu_usage);
 550                         if (r != 0)
 551                                 return r;
 552                 }
 553
 554                 break;
 555
 556         case ORDER_TASKS:
 557                 if (x->n_tasks_valid && y->n_tasks_valid) {
 558                         r = CMP(y->n_tasks, x->n_tasks);
 559                         if (r != 0)
 560                                 return r;
 561                 } else if (x->n_tasks_valid)
 562                         return -1;
 563                 else if (y->n_tasks_valid)
 564                         return 1;
 565
 566                 break;
 567
 568         case ORDER_MEMORY:
 569                 if (x->memory_valid && y->memory_valid) {
 570                         r = CMP(y->memory, x->memory);
 571                         if (r != 0)
 572                                 return r;
 573                 } else if (x->memory_valid)
 574                         return -1;
 575                 else if (y->memory_valid)
 576                         return 1;
 577
 578                 break;
 579
 580         case ORDER_IO:
 581                 if (x->io_valid && y->io_valid) {
 582                         r = CMP(y->io_input_bps + y->io_output_bps, x->io_input_bps + x->io_output_bps);
 583                         if (r != 0)
 584                                 return r;
 585                 } else if (x->io_valid)
 586                         return -1;
 587                 else if (y->io_valid)
 588                         return 1;
 589         }
 590
 591         return path_compare(x->path, y->path);
 592 }
 593
 594 static void display(Hashmap *a) {
 595         Iterator i;
 596         Group *g;
 597         Group **array;
 598         signed path_columns;
 599         unsigned rows, n = 0, j, maxtcpu = 0, maxtpath = 3; /* 3 for ellipsize() to work properly */
 600         char buffer[MAX3(21, FORMAT_BYTES_MAX, FORMAT_TIMESPAN_MAX)];
 601
 602         assert(a);
 603
 604         if (!terminal_is_dumb())
 605                 fputs(ANSI_HOME_CLEAR, stdout);
 606
 607         array = newa(Group*, hashmap_size(a));
 608
 609         HASHMAP_FOREACH(g, a, i)
 610                 if (g->n_tasks_valid || g->cpu_valid || g->memory_valid || g->io_valid)
 611                         array[n++] = g;
 612
 613         typesafe_qsort(array, n, group_compare);
 614
 615         /* Find the longest names in one run */
 616         for (j = 0; j < n; j++) {
 617                 unsigned cputlen, pathtlen;
 618
 619                 format_timespan(buffer, sizeof(buffer), (usec_t) (array[j]->cpu_usage / NSEC_PER_USEC), 0);
 620                 cputlen = strlen(buffer);
 621                 maxtcpu = MAX(maxtcpu, cputlen);
 622
 623                 pathtlen = strlen(array[j]->path);
 624                 maxtpath = MAX(maxtpath, pathtlen);
 625         }
 626
 627         if (arg_cpu_type == CPU_PERCENT)
 628                 xsprintf(buffer, "%6s", "%CPU");
 629         else
 630                 xsprintf(buffer, "%*s", maxtcpu, "CPU Time");
 631
 632         rows = lines();
 633         if (rows <= 10)
 634                 rows = 10;
 635
 636         if (on_tty()) {
 637                 const char *on, *off;
 638
 639                 path_columns = columns() - 36 - strlen(buffer);
 640                 if (path_columns < 10)
 641                         path_columns = 10;
 642
 643                 on = ansi_highlight_underline();
 644                 off = ansi_underline();
 645
 646                 printf("%s%s%-*s%s %s%7s%s %s%s%s %s%8s%s %s%8s%s %s%8s%s%s\n",
 647                        ansi_underline(),
 648                        arg_order == ORDER_PATH ? on : "", path_columns, "Control Group",
 649                        arg_order == ORDER_PATH ? off : "",
 650                        arg_order == ORDER_TASKS ? on : "", arg_count == COUNT_PIDS ? "Tasks" : arg_count == COUNT_USERSPACE_PROCESSES ? "Procs" : "Proc+",
 651                        arg_order == ORDER_TASKS ? off : "",
 652                        arg_order == ORDER_CPU ? on : "", buffer,
 653                        arg_order == ORDER_CPU ? off : "",
 654                        arg_order == ORDER_MEMORY ? on : "", "Memory",
 655                        arg_order == ORDER_MEMORY ? off : "",
 656                        arg_order == ORDER_IO ? on : "", "Input/s",
 657                        arg_order == ORDER_IO ? off : "",
 658                        arg_order == ORDER_IO ? on : "", "Output/s",
 659                        arg_order == ORDER_IO ? off : "",
 660                        ansi_normal());
 661         } else
 662                 path_columns = maxtpath;
 663
 664         for (j = 0; j < n; j++) {
 665                 _cleanup_free_ char *ellipsized = NULL;
 666                 const char *path;
 667
 668                 if (on_tty() && j + 6 > rows)
 669                         break;
 670
 671                 g = array[j];
 672
 673                 path = empty_to_root(g->path);
 674                 ellipsized = ellipsize(path, path_columns, 33);
 675                 printf("%-*s", path_columns, ellipsized ?: path);
 676
 677                 if (g->n_tasks_valid)
 678                         printf(" %7" PRIu64, g->n_tasks);
 679                 else
 680                         fputs("       -", stdout);
 681
 682                 if (arg_cpu_type == CPU_PERCENT) {
 683                         if (g->cpu_valid)
 684                                 printf(" %6.1f", g->cpu_fraction*100);
 685                         else
 686                                 fputs("      -", stdout);
 687                 } else
 688                         printf(" %*s", maxtcpu, format_timespan(buffer, sizeof(buffer), (usec_t) (g->cpu_usage / NSEC_PER_USEC), 0));
 689
 690                 printf(" %8s", maybe_format_bytes(buffer, sizeof(buffer), g->memory_valid, g->memory));
 691                 printf(" %8s", maybe_format_bytes(buffer, sizeof(buffer), g->io_valid, g->io_input_bps));
 692                 printf(" %8s", maybe_format_bytes(buffer, sizeof(buffer), g->io_valid, g->io_output_bps));
 693
 694                 putchar('\n');
 695         }
 696 }
 697
 698 static int help(void) {
 699         _cleanup_free_ char *link = NULL;
 700         int r;
 701
 702         r = terminal_urlify_man("systemd-cgtop", "1", &link);
 703         if (r < 0)
 704                 return log_oom();
 705
 706         printf("%s [OPTIONS...] [CGROUP]\n\n"
 707                "Show top control groups by their resource usage.\n\n"
 708                "  -h --help           Show this help\n"
 709                "     --version        Show package version\n"
 710                "  -p --order=path     Order by path\n"
 711                "  -t --order=tasks    Order by number of tasks/processes\n"
 712                "  -c --order=cpu      Order by CPU load (default)\n"
 713                "  -m --order=memory   Order by memory load\n"
 714                "  -i --order=io       Order by IO load\n"
 715                "  -r --raw            Provide raw (not human-readable) numbers\n"
 716                "     --cpu=percentage Show CPU usage as percentage (default)\n"
 717                "     --cpu=time       Show CPU usage as time\n"
 718                "  -P                  Count userspace processes instead of tasks (excl. kernel)\n"
 719                "  -k                  Count all processes instead of tasks (incl. kernel)\n"
 720                "     --recursive=BOOL Sum up process count recursively\n"
 721                "  -d --delay=DELAY    Delay between updates\n"
 722                "  -n --iterations=N   Run for N iterations before exiting\n"
 723                "  -1                  Shortcut for --iterations=1\n"
 724                "  -b --batch          Run in batch mode, accepting no input\n"
 725                "     --depth=DEPTH    Maximum traversal depth (default: %u)\n"
 726                "  -M --machine=       Show container\n"
 727                "\nSee the %s for details.\n"
 728                , program_invocation_short_name
 729                , arg_depth
 730                , link
 731         );
 732
 733         return 0;
 734 }
 735
 736 static int parse_argv(int argc, char *argv[]) {
 737         enum {
 738                 ARG_VERSION = 0x100,
 739                 ARG_DEPTH,
 740                 ARG_CPU_TYPE,
 741                 ARG_ORDER,
 742                 ARG_RECURSIVE,
 743         };
 744
 745         static const struct option options[] = {
 746                 { "help",         no_argument,       NULL, 'h'           },
 747                 { "version",      no_argument,       NULL, ARG_VERSION   },
 748                 { "delay",        required_argument, NULL, 'd'           },
 749                 { "iterations",   required_argument, NULL, 'n'           },
 750                 { "batch",        no_argument,       NULL, 'b'           },
 751                 { "raw",          no_argument,       NULL, 'r'           },
 752                 { "depth",        required_argument, NULL, ARG_DEPTH     },
 753                 { "cpu",          optional_argument, NULL, ARG_CPU_TYPE  },
 754                 { "order",        required_argument, NULL, ARG_ORDER     },
 755                 { "recursive",    required_argument, NULL, ARG_RECURSIVE },
 756                 { "machine",      required_argument, NULL, 'M'           },
 757                 {}
 758         };
 759
 760         int c, r;
 761
 762         assert(argc >= 1);
 763         assert(argv);
 764
 765         while ((c = getopt_long(argc, argv, "hptcmin:brd:kPM:1", options, NULL)) >= 0)
 766
 767                 switch (c) {
 768
 769                 case 'h':
 770                         return help();
 771
 772                 case ARG_VERSION:
 773                         return version();
 774
 775                 case ARG_CPU_TYPE:
 776                         if (optarg) {
 777                                 if (streq(optarg, "time"))
 778                                         arg_cpu_type = CPU_TIME;
 779                                 else if (streq(optarg, "percentage"))
 780                                         arg_cpu_type = CPU_PERCENT;
 781                                 else {
 782                                         log_error("Unknown argument to --cpu=: %s", optarg);
 783                                         return -EINVAL;
 784                                 }
 785                         } else
 786                                 arg_cpu_type = CPU_TIME;
 787
 788                         break;
 789
 790                 case ARG_DEPTH:
 791                         r = safe_atou(optarg, &arg_depth);
 792                         if (r < 0)
 793                                 return log_error_errno(r, "Failed to parse depth parameter '%s': %m", optarg);
 794
 795                         break;
 796
 797                 case 'd':
 798                         r = parse_sec(optarg, &arg_delay);
 799                         if (r < 0)
 800                                 return log_error_errno(r, "Failed to parse delay parameter '%s': %m", optarg);
 801                         if (arg_delay <= 0) {
 802                                 log_error("Invalid delay parameter '%s'", optarg);
 803                                 return -EINVAL;
 804                         }
 805
 806                         break;
 807
 808                 case 'n':
 809                         r = safe_atou(optarg, &arg_iterations);
 810                         if (r < 0)
 811                                 return log_error_errno(r, "Failed to parse iterations parameter '%s': %m", optarg);
 812
 813                         break;
 814
 815                 case '1':
 816                         arg_iterations = 1;
 817                         break;
 818
 819                 case 'b':
 820                         arg_batch = true;
 821                         break;
 822
 823                 case 'r':
 824                         arg_raw = true;
 825                         break;
 826
 827                 case 'p':
 828                         arg_order = ORDER_PATH;
 829                         break;
 830
 831                 case 't':
 832                         arg_order = ORDER_TASKS;
 833                         break;
 834
 835                 case 'c':
 836                         arg_order = ORDER_CPU;
 837                         break;
 838
 839                 case 'm':
 840                         arg_order = ORDER_MEMORY;
 841                         break;
 842
 843                 case 'i':
 844                         arg_order = ORDER_IO;
 845                         break;
 846
 847                 case ARG_ORDER:
 848                         if (streq(optarg, "path"))
 849                                 arg_order = ORDER_PATH;
 850                         else if (streq(optarg, "tasks"))
 851                                 arg_order = ORDER_TASKS;
 852                         else if (streq(optarg, "cpu"))
 853                                 arg_order = ORDER_CPU;
 854                         else if (streq(optarg, "memory"))
 855                                 arg_order = ORDER_MEMORY;
 856                         else if (streq(optarg, "io"))
 857                                 arg_order = ORDER_IO;
 858                         else {
 859                                 log_error("Invalid argument to --order=: %s", optarg);
 860                                 return -EINVAL;
 861                         }
 862                         break;
 863
 864                 case 'k':
 865                         arg_count = COUNT_ALL_PROCESSES;
 866                         break;
 867
 868                 case 'P':
 869                         arg_count = COUNT_USERSPACE_PROCESSES;
 870                         break;
 871
 872                 case ARG_RECURSIVE:
 873                         r = parse_boolean(optarg);
 874                         if (r < 0)
 875                                 return log_error_errno(r, "Failed to parse --recursive= argument '%s': %m", optarg);
 876
 877                         arg_recursive = r;
 878                         arg_recursive_unset = r == 0;
 879                         break;
 880
 881                 case 'M':
 882                         arg_machine = optarg;
 883                         break;
 884
 885                 case '?':
 886                         return -EINVAL;
 887
 888                 default:
 889                         assert_not_reached("Unhandled option");
 890                 }
 891
 892         if (optind == argc - 1)
 893                 arg_root = argv[optind];
 894         else if (optind < argc) {
 895                 log_error("Too many arguments.");
 896                 return -EINVAL;
 897         }
 898
 899         return 1;
 900 }
 901
 902 static const char* counting_what(void) {
 903         if (arg_count == COUNT_PIDS)
 904                 return "tasks";
 905         else if (arg_count == COUNT_ALL_PROCESSES)
 906                 return "all processes (incl. kernel)";
 907         else
 908                 return "userspace processes (excl. kernel)";
 909 }
 910
 911 static int run(int argc, char *argv[]) {
 912         _cleanup_(group_hashmap_freep) Hashmap *a = NULL, *b = NULL;
 913         unsigned iteration = 0;
 914         usec_t last_refresh = 0;
 915         bool quit = false, immediate_refresh = false;
 916         _cleanup_free_ char *root = NULL;
 917         CGroupMask mask;
 918         int r;
 919
 920         log_parse_environment();
 921         log_open();
 922
 923         r = parse_argv(argc, argv);
 924         if (r <= 0)
 925                 return r;
 926
 927         r = cg_mask_supported(&mask);
 928         if (r < 0)
 929                 return log_error_errno(r, "Failed to determine supported controllers: %m");
 930
 931         arg_count = (mask & CGROUP_MASK_PIDS) ? COUNT_PIDS : COUNT_USERSPACE_PROCESSES;
 932
 933         if (arg_recursive_unset && arg_count == COUNT_PIDS) {
 934                 log_error("Non-recursive counting is only supported when counting processes, not tasks. Use -P or -k.");
 935                 return -EINVAL;
 936         }
 937
 938         r = show_cgroup_get_path_and_warn(arg_machine, arg_root, &root);
 939         if (r < 0)
 940                 return log_error_errno(r, "Failed to get root control group path: %m");
 941         log_debug("Cgroup path: %s", root);
 942
 943         a = hashmap_new(&path_hash_ops);
 944         b = hashmap_new(&path_hash_ops);
 945         if (!a || !b)
 946                 return log_oom();
 947
 948         signal(SIGWINCH, columns_lines_cache_reset);
 949
 950         if (arg_iterations == (unsigned) -1)
 951                 arg_iterations = on_tty() ? 0 : 1;
 952
 953         while (!quit) {
 954                 Hashmap *c;
 955                 usec_t t;
 956                 char key;
 957                 char h[FORMAT_TIMESPAN_MAX];
 958
 959                 t = now(CLOCK_MONOTONIC);
 960
 961                 if (t >= last_refresh + arg_delay || immediate_refresh) {
 962
 963                         r = refresh(root, a, b, iteration++);
 964                         if (r < 0)
 965                                 return log_error_errno(r, "Failed to refresh: %m");
 966
 967                         group_hashmap_clear(b);
 968
 969                         c = a;
 970                         a = b;
 971                         b = c;
 972
 973                         last_refresh = t;
 974                         immediate_refresh = false;
 975                 }
 976
 977                 display(b);
 978
 979                 if (arg_iterations && iteration >= arg_iterations)
 980                         break;
 981
 982                 if (!on_tty()) /* non-TTY: Empty newline as delimiter between polls */
 983                         fputs("\n", stdout);
 984                 fflush(stdout);
 985
 986                 if (arg_batch)
 987                         (void) usleep(last_refresh + arg_delay - t);
 988                 else {
 989                         r = read_one_char(stdin, &key, last_refresh + arg_delay - t, NULL);
 990                         if (r == -ETIMEDOUT)
 991                                 continue;
 992                         if (r < 0)
 993                                 return log_error_errno(r, "Couldn't read key: %m");
 994                 }
 995
 996                 if (on_tty()) { /* TTY: Clear any user keystroke */
 997                         fputs("\r \r", stdout);
 998                         fflush(stdout);
 999                 }
1000
1001                 if (arg_batch)
1002                         continue;
1003
1004                 switch (key) {
1005
1006                 case ' ':
1007                         immediate_refresh = true;
1008                         break;
1009
1010                 case 'q':
1011                         quit = true;
1012                         break;
1013
1014                 case 'p':
1015                         arg_order = ORDER_PATH;
1016                         break;
1017
1018                 case 't':
1019                         arg_order = ORDER_TASKS;
1020                         break;
1021
1022                 case 'c':
1023                         arg_order = ORDER_CPU;
1024                         break;
1025
1026                 case 'm':
1027                         arg_order = ORDER_MEMORY;
1028                         break;
1029
1030                 case 'i':
1031                         arg_order = ORDER_IO;
1032                         break;
1033
1034                 case '%':
1035                         arg_cpu_type = arg_cpu_type == CPU_TIME ? CPU_PERCENT : CPU_TIME;
1036                         break;
1037
1038                 case 'k':
1039                         arg_count = arg_count != COUNT_ALL_PROCESSES ? COUNT_ALL_PROCESSES : COUNT_PIDS;
1040                         fprintf(stdout, "\nCounting: %s.", counting_what());
1041                         fflush(stdout);
1042                         sleep(1);
1043                         break;
1044
1045                 case 'P':
1046                         arg_count = arg_count != COUNT_USERSPACE_PROCESSES ? COUNT_USERSPACE_PROCESSES : COUNT_PIDS;
1047                         fprintf(stdout, "\nCounting: %s.", counting_what());
1048                         fflush(stdout);
1049                         sleep(1);
1050                         break;
1051
1052                 case 'r':
1053                         if (arg_count == COUNT_PIDS)
1054                                 fprintf(stdout, "\n\aCannot toggle recursive counting, not available in task counting mode.");
1055                         else {
1056                                 arg_recursive = !arg_recursive;
1057                                 fprintf(stdout, "\nRecursive process counting: %s", yes_no(arg_recursive));
1058                         }
1059                         fflush(stdout);
1060                         sleep(1);
1061                         break;
1062
1063                 case '+':
1064                         if (arg_delay < USEC_PER_SEC)
1065                                 arg_delay += USEC_PER_MSEC*250;
1066                         else
1067                                 arg_delay += USEC_PER_SEC;
1068
1069                         fprintf(stdout, "\nIncreased delay to %s.", format_timespan(h, sizeof(h), arg_delay, 0));
1070                         fflush(stdout);
1071                         sleep(1);
1072                         break;
1073
1074                 case '-':
1075                         if (arg_delay <= USEC_PER_MSEC*500)
1076                                 arg_delay = USEC_PER_MSEC*250;
1077                         else if (arg_delay < USEC_PER_MSEC*1250)
1078                                 arg_delay -= USEC_PER_MSEC*250;
1079                         else
1080                                 arg_delay -= USEC_PER_SEC;
1081
1082                         fprintf(stdout, "\nDecreased delay to %s.", format_timespan(h, sizeof(h), arg_delay, 0));
1083                         fflush(stdout);
1084                         sleep(1);
1085                         break;
1086
1087                 case '?':
1088                 case 'h':
1089
1090 #define ON ANSI_HIGHLIGHT
1091 #define OFF ANSI_NORMAL
1092
1093                         fprintf(stdout,
1094                                 "\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"
1095                                 "\t<" ON "+" OFF "> Inc. delay; <" ON "-" OFF "> Dec. delay; <" ON "%%" OFF "> Toggle time; <" ON "SPACE" OFF "> Refresh\n"
1096                                 "\t<" ON "P" OFF "> Toggle count userspace processes; <" ON "k" OFF "> Toggle count all processes\n"
1097                                 "\t<" ON "r" OFF "> Count processes recursively; <" ON "q" OFF "> Quit");
1098                         fflush(stdout);
1099                         sleep(3);
1100                         break;
1101
1102                 default:
1103                         if (key < ' ')
1104                                 fprintf(stdout, "\nUnknown key '\\x%x'. Ignoring.", key);
1105                         else
1106                                 fprintf(stdout, "\nUnknown key '%c'. Ignoring.", key);
1107                         fflush(stdout);
1108                         sleep(1);
1109                         break;
1110                 }
1111         }
1112
1113         return 0;
1114 }
1115
1116 DEFINE_MAIN_FUNCTION(run);