src/apps_plugin.c

   1 // TODO
   2 //
   3 // 1. disable RESET_OR_OVERFLOW check in charts
   4
   5 #ifdef HAVE_CONFIG_H
   6 #include <config.h>
   7 #endif
   8 #include <stdio.h>
   9 #include <stdlib.h>
  10 #include <string.h>
  11 #include <time.h>
  12 #include <unistd.h>
  13 #include <sys/types.h>
  14 #include <sys/time.h>
  15 #include <sys/wait.h>
  16 #include <sys/stat.h>
  17
  18 #include <sys/resource.h>
  19 #include <sys/stat.h>
  20
  21 #include <errno.h>
  22 #include <stdarg.h>
  23 #include <locale.h>
  24 #include <ctype.h>
  25 #include <fcntl.h>
  26
  27 #include <malloc.h>
  28 #include <dirent.h>
  29 #include <arpa/inet.h>
  30
  31 #include <sys/types.h>
  32 #include <pwd.h>
  33 #include <grp.h>
  34
  35 #include "avl.h"
  36
  37 #include "common.h"
  38 #include "log.h"
  39 #include "procfile.h"
  40 #include "../config.h"
  41
  42 #ifdef NETDATA_INTERNAL_CHECKS
  43 #include <sys/prctl.h>
  44 #endif
  45
  46 #define MAX_COMPARE_NAME 100
  47 #define MAX_NAME 100
  48 #define MAX_CMDLINE 1024
  49
  50 long processors = 1;
  51 long pid_max = 32768;
  52 int debug = 0;
  53
  54 int update_every = 1;
  55 unsigned long long file_counter = 0;
  56 int proc_pid_cmdline_is_needed = 0;
  57
  58 char *host_prefix = "";
  59 char *config_dir = CONFIG_DIR;
  60
  61 #ifdef NETDATA_INTERNAL_CHECKS
  62 // ----------------------------------------------------------------------------
  63 // memory debugger
  64 // do not use in production systems - it mis-aligns allocated memory
  65
  66 struct allocations {
  67         size_t allocations;
  68         size_t allocated;
  69         size_t allocated_max;
  70 } allocations = { 0, 0, 0 };
  71
  72 #define MALLOC_MARK (uint32_t)(0x0BADCAFE)
  73 #define MALLOC_PREFIX (sizeof(uint32_t) * 2)
  74 #define MALLOC_SUFFIX (sizeof(uint32_t))
  75 #define MALLOC_OVERHEAD (MALLOC_PREFIX + MALLOC_SUFFIX)
  76
  77 void *mark_allocation(void *allocated_ptr, size_t size_without_overheads) {
  78         uint32_t *real_ptr = (uint32_t *)allocated_ptr;
  79         real_ptr[0] = MALLOC_MARK;
  80         real_ptr[1] = (uint32_t) size_without_overheads;
  81
  82         uint32_t *end_ptr = (uint32_t *)(allocated_ptr + MALLOC_PREFIX + size_without_overheads);
  83         end_ptr[0] = MALLOC_MARK;
  84
  85         // fprintf(stderr, "MEMORY_POINTER: Allocated at %p, returning %p.\n", allocated_ptr, (void *)(allocated_ptr + MALLOC_PREFIX));
  86
  87         return allocated_ptr + MALLOC_PREFIX;
  88 }
  89
  90 void *check_allocation(const char *file, int line, const char *function, void *marked_ptr, size_t *size_without_overheads_ptr) {
  91         uint32_t *real_ptr = (uint32_t *)(marked_ptr - MALLOC_PREFIX);
  92
  93         // fprintf(stderr, "MEMORY_POINTER: Checking pointer at %p, real %p for %s/%u@%s.\n", marked_ptr, (void *)(marked_ptr - MALLOC_PREFIX), function, line, file);
  94
  95         if(real_ptr[0] != MALLOC_MARK) fatal("MEMORY: prefix MARK is not valid for %s/%u@%s.", function, line, file);
  96
  97         size_t size = real_ptr[1];
  98
  99         uint32_t *end_ptr = (uint32_t *)(marked_ptr + size);
 100         if(end_ptr[0] != MALLOC_MARK) fatal("MEMORY: suffix MARK of allocation with size %zu is not valid for %s/%u@%s.", size, function, line, file);
 101
 102         if(size_without_overheads_ptr) *size_without_overheads_ptr = size;
 103
 104         return real_ptr;
 105 }
 106
 107 void *malloc_debug(const char *file, int line, const char *function, size_t size) {
 108         void *ptr = malloc(size + MALLOC_OVERHEAD);
 109         if(!ptr) fatal("MEMORY: Cannot allocate %zu bytes for %s/%u@%s.", size, function, line, file);
 110
 111         allocations.allocated += size;
 112         allocations.allocations++;
 113
 114         debug(D_MEMORY, "MEMORY: Allocated %zu bytes for %s/%u@%s."
 115                 " Status: allocated %zu in %zu allocs."
 116                 , size
 117                 , function, line, file
 118                 , allocations.allocated
 119                 , allocations.allocations
 120         );
 121
 122         if(allocations.allocated > allocations.allocated_max) {
 123                 debug(D_MEMORY, "MEMORY: total allocation peak increased from %zu to %zu", allocations.allocated_max, allocations.allocated);
 124                 allocations.allocated_max = allocations.allocated;
 125         }
 126
 127         size_t csize;
 128         check_allocation(file, line, function, mark_allocation(ptr, size), &csize);
 129         if(size != csize) {
 130                 fatal("Invalid size.");
 131         }
 132
 133         return mark_allocation(ptr, size);
 134 }
 135
 136 void *calloc_debug(const char *file, int line, const char *function, size_t nmemb, size_t size) {
 137         void *ptr = malloc_debug(file, line, function, (nmemb * size));
 138         bzero(ptr, nmemb * size);
 139         return ptr;
 140 }
 141
 142 void free_debug(const char *file, int line, const char *function, void *ptr) {
 143         size_t size;
 144         void *real_ptr = check_allocation(file, line, function, ptr, &size);
 145
 146         bzero(real_ptr, size + MALLOC_OVERHEAD);
 147
 148         free(real_ptr);
 149         allocations.allocated -= size;
 150         allocations.allocations--;
 151
 152         debug(D_MEMORY, "MEMORY: freed %zu bytes for %s/%u@%s."
 153                 " Status: allocated %zu in %zu allocs."
 154                 , size
 155                 , function, line, file
 156                 , allocations.allocated
 157                 , allocations.allocations
 158         );
 159 }
 160
 161 void *realloc_debug(const char *file, int line, const char *function, void *ptr, size_t size) {
 162         if(!ptr) return malloc_debug(file, line, function, size);
 163         if(!size) { free_debug(file, line, function, ptr); return NULL; }
 164
 165         size_t old_size;
 166         void *real_ptr = check_allocation(file, line, function, ptr, &old_size);
 167
 168         void *new_ptr = realloc(real_ptr, size + MALLOC_OVERHEAD);
 169         if(!new_ptr) fatal("MEMORY: Cannot allocate %zu bytes for %s/%u@%s.", size, function, line, file);
 170
 171         allocations.allocated += size;
 172         allocations.allocated -= old_size;
 173
 174         debug(D_MEMORY, "MEMORY: Re-allocated from %zu to %zu bytes for %s/%u@%s."
 175                 " Status: allocated %z in %zu allocs."
 176                 , old_size, size
 177                 , function, line, file
 178                 , allocations.allocated
 179                 , allocations.allocations
 180         );
 181
 182         if(allocations.allocated > allocations.allocated_max) {
 183                 debug(D_MEMORY, "MEMORY: total allocation peak increased from %zu to %zu", allocations.allocated_max, allocations.allocated);
 184                 allocations.allocated_max = allocations.allocated;
 185         }
 186
 187         return mark_allocation(new_ptr, size);
 188 }
 189
 190 char *strdup_debug(const char *file, int line, const char *function, const char *ptr) {
 191         size_t size = 0;
 192         const char *s = ptr;
 193
 194         while(*s++) size++;
 195         size++;
 196
 197         char *p = malloc_debug(file, line, function, size);
 198         if(!p) fatal("Cannot allocate %zu bytes.", size);
 199
 200         memcpy(p, ptr, size);
 201         return p;
 202 }
 203
 204 #define malloc(size) malloc_debug(__FILE__, __LINE__, __FUNCTION__, (size))
 205 #define calloc(nmemb, size) calloc_debug(__FILE__, __LINE__, __FUNCTION__, (nmemb), (size))
 206 #define realloc(ptr, size) realloc_debug(__FILE__, __LINE__, __FUNCTION__, (ptr), (size))
 207 #define free(ptr) free_debug(__FILE__, __LINE__, __FUNCTION__, (ptr))
 208
 209 #ifdef strdup
 210 #undef strdup
 211 #endif
 212 #define strdup(ptr) strdup_debug(__FILE__, __LINE__, __FUNCTION__, (ptr))
 213
 214 #endif /* NETDATA_INTERNAL_CHECKS */
 215
 216 // ----------------------------------------------------------------------------
 217 // system functions
 218 // to retrieve settings of the system
 219
 220 long get_system_cpus(void) {
 221         procfile *ff = NULL;
 222
 223         int processors = 0;
 224
 225         char filename[FILENAME_MAX + 1];
 226         snprintfz(filename, FILENAME_MAX, "%s/proc/stat", host_prefix);
 227
 228         ff = procfile_open(filename, NULL, PROCFILE_FLAG_DEFAULT);
 229         if(!ff) return 1;
 230
 231         ff = procfile_readall(ff);
 232         if(!ff) {
 233                 procfile_close(ff);
 234                 return 1;
 235         }
 236
 237         unsigned int i;
 238         for(i = 0; i < procfile_lines(ff); i++) {
 239                 if(!procfile_linewords(ff, i)) continue;
 240
 241                 if(strncmp(procfile_lineword(ff, i, 0), "cpu", 3) == 0) processors++;
 242         }
 243         processors--;
 244         if(processors < 1) processors = 1;
 245
 246         procfile_close(ff);
 247         return processors;
 248 }
 249
 250 long get_system_pid_max(void) {
 251         procfile *ff = NULL;
 252         long mpid = 32768;
 253
 254         char filename[FILENAME_MAX + 1];
 255         snprintfz(filename, FILENAME_MAX, "%s/proc/sys/kernel/pid_max", host_prefix);
 256         ff = procfile_open(filename, NULL, PROCFILE_FLAG_DEFAULT);
 257         if(!ff) return mpid;
 258
 259         ff = procfile_readall(ff);
 260         if(!ff) {
 261                 procfile_close(ff);
 262                 return mpid;
 263         }
 264
 265         mpid = atol(procfile_lineword(ff, 0, 0));
 266         if(!mpid) mpid = 32768;
 267
 268         procfile_close(ff);
 269         return mpid;
 270 }
 271
 272 // ----------------------------------------------------------------------------
 273 // target
 274 // target is the structure that process data are aggregated
 275
 276 struct target {
 277         char compare[MAX_COMPARE_NAME + 1];
 278         uint32_t comparehash;
 279         size_t comparelen;
 280
 281         char id[MAX_NAME + 1];
 282         uint32_t idhash;
 283
 284         char name[MAX_NAME + 1];
 285
 286         uid_t uid;
 287         gid_t gid;
 288
 289         unsigned long long minflt;
 290         unsigned long long cminflt;
 291         unsigned long long majflt;
 292         unsigned long long cmajflt;
 293         unsigned long long utime;
 294         unsigned long long stime;
 295         unsigned long long cutime;
 296         unsigned long long cstime;
 297         unsigned long long num_threads;
 298         unsigned long long rss;
 299
 300         unsigned long long fix_minflt;
 301         unsigned long long fix_cminflt;
 302         unsigned long long fix_majflt;
 303         unsigned long long fix_cmajflt;
 304         unsigned long long fix_utime;
 305         unsigned long long fix_stime;
 306         unsigned long long fix_cutime;
 307         unsigned long long fix_cstime;
 308
 309         unsigned long long statm_size;
 310         unsigned long long statm_resident;
 311         unsigned long long statm_share;
 312         unsigned long long statm_text;
 313         unsigned long long statm_lib;
 314         unsigned long long statm_data;
 315         unsigned long long statm_dirty;
 316
 317         unsigned long long io_logical_bytes_read;
 318         unsigned long long io_logical_bytes_written;
 319         unsigned long long io_read_calls;
 320         unsigned long long io_write_calls;
 321         unsigned long long io_storage_bytes_read;
 322         unsigned long long io_storage_bytes_written;
 323         unsigned long long io_cancelled_write_bytes;
 324
 325         unsigned long long fix_io_logical_bytes_read;
 326         unsigned long long fix_io_logical_bytes_written;
 327         unsigned long long fix_io_read_calls;
 328         unsigned long long fix_io_write_calls;
 329         unsigned long long fix_io_storage_bytes_read;
 330         unsigned long long fix_io_storage_bytes_written;
 331         unsigned long long fix_io_cancelled_write_bytes;
 332
 333         int *fds;
 334         unsigned long long openfiles;
 335         unsigned long long openpipes;
 336         unsigned long long opensockets;
 337         unsigned long long openinotifies;
 338         unsigned long long openeventfds;
 339         unsigned long long opentimerfds;
 340         unsigned long long opensignalfds;
 341         unsigned long long openeventpolls;
 342         unsigned long long openother;
 343
 344         unsigned long processes;        // how many processes have been merged to this
 345         int exposed;                            // if set, we have sent this to netdata
 346         int hidden;                                     // if set, we set the hidden flag on the dimension
 347         int debug;
 348         int ends_with;
 349         int starts_with;            // if set, the compare string matches only the
 350                                                                 // beginning of the command
 351
 352         struct target *target;          // the one that will be reported to netdata
 353         struct target *next;
 354 };
 355
 356
 357 // ----------------------------------------------------------------------------
 358 // apps_groups.conf
 359 // aggregate all processes in groups, to have a limited number of dimensions
 360
 361 struct target *apps_groups_root_target = NULL;
 362 struct target *apps_groups_default_target = NULL;
 363 long apps_groups_targets = 0;
 364
 365 struct target *users_root_target = NULL;
 366 struct target *groups_root_target = NULL;
 367
 368 struct target *get_users_target(uid_t uid)
 369 {
 370         struct target *w;
 371         for(w = users_root_target ; w ; w = w->next)
 372                 if(w->uid == uid) return w;
 373
 374         w = calloc(sizeof(struct target), 1);
 375         if(unlikely(!w)) {
 376                 error("Cannot allocate %lu bytes of memory", (unsigned long)sizeof(struct target));
 377                 return NULL;
 378         }
 379
 380         snprintfz(w->compare, MAX_COMPARE_NAME, "%d", uid);
 381         w->comparehash = simple_hash(w->compare);
 382         w->comparelen = strlen(w->compare);
 383
 384         snprintfz(w->id, MAX_NAME, "%d", uid);
 385         w->idhash = simple_hash(w->id);
 386
 387         struct passwd *pw = getpwuid(uid);
 388         if(!pw)
 389                 snprintfz(w->name, MAX_NAME, "%d", uid);
 390         else
 391                 snprintfz(w->name, MAX_NAME, "%s", pw->pw_name);
 392
 393         netdata_fix_chart_name(w->name);
 394
 395         w->uid = uid;
 396
 397         w->next = users_root_target;
 398         users_root_target = w;
 399
 400         if(unlikely(debug))
 401                 fprintf(stderr, "apps.plugin: added uid %d ('%s') target\n", w->uid, w->name);
 402
 403         return w;
 404 }
 405
 406 struct target *get_groups_target(gid_t gid)
 407 {
 408         struct target *w;
 409         for(w = groups_root_target ; w ; w = w->next)
 410                 if(w->gid == gid) return w;
 411
 412         w = calloc(sizeof(struct target), 1);
 413         if(unlikely(!w)) {
 414                 error("Cannot allocate %lu bytes of memory", (unsigned long)sizeof(struct target));
 415                 return NULL;
 416         }
 417
 418         snprintfz(w->compare, MAX_COMPARE_NAME, "%d", gid);
 419         w->comparehash = simple_hash(w->compare);
 420         w->comparelen = strlen(w->compare);
 421
 422         snprintfz(w->id, MAX_NAME, "%d", gid);
 423         w->idhash = simple_hash(w->id);
 424
 425         struct group *gr = getgrgid(gid);
 426         if(!gr)
 427                 snprintfz(w->name, MAX_NAME, "%d", gid);
 428         else
 429                 snprintfz(w->name, MAX_NAME, "%s", gr->gr_name);
 430
 431         netdata_fix_chart_name(w->name);
 432
 433         w->gid = gid;
 434
 435         w->next = groups_root_target;
 436         groups_root_target = w;
 437
 438         if(unlikely(debug))
 439                 fprintf(stderr, "apps.plugin: added gid %d ('%s') target\n", w->gid, w->name);
 440
 441         return w;
 442 }
 443
 444 // find or create a new target
 445 // there are targets that are just aggregated to other target (the second argument)
 446 struct target *get_apps_groups_target(const char *id, struct target *target)
 447 {
 448         int tdebug = 0, thidden = 0, ends_with = 0;
 449         const char *nid = id;
 450
 451         while(nid[0] == '-' || nid[0] == '+' || nid[0] == '*') {
 452                 if(nid[0] == '-') thidden = 1;
 453                 if(nid[0] == '+') tdebug = 1;
 454                 if(nid[0] == '*') ends_with = 1;
 455                 nid++;
 456         }
 457         uint32_t hash = simple_hash(id);
 458
 459         struct target *w;
 460         for(w = apps_groups_root_target ; w ; w = w->next) {
 461                 if(w->idhash == hash && strncmp(nid, w->id, MAX_NAME) == 0)
 462                         return w;
 463         }
 464
 465         w = calloc(sizeof(struct target), 1);
 466         if(unlikely(!w)) {
 467                 error("Cannot allocate %lu bytes of memory", (unsigned long)sizeof(struct target));
 468                 return NULL;
 469         }
 470
 471         strncpyz(w->id, nid, MAX_NAME);
 472         w->idhash = simple_hash(w->id);
 473
 474         strncpyz(w->name, nid, MAX_NAME);
 475
 476         strncpyz(w->compare, nid, MAX_COMPARE_NAME);
 477         int len = strlen(w->compare);
 478         if(w->compare[len - 1] == '*') {
 479                 w->compare[len - 1] = '\0';
 480                 w->starts_with = 1;
 481         }
 482         w->ends_with = ends_with;
 483
 484         if(w->starts_with && w->ends_with)
 485                 proc_pid_cmdline_is_needed = 1;
 486
 487         w->comparehash = simple_hash(w->compare);
 488         w->comparelen = strlen(w->compare);
 489
 490         w->hidden = thidden;
 491         w->debug = tdebug;
 492         w->target = target;
 493
 494         w->next = apps_groups_root_target;
 495         apps_groups_root_target = w;
 496
 497         if(unlikely(debug))
 498                 fprintf(stderr, "apps.plugin: ADDING TARGET ID '%s', process name '%s' (%s), aggregated on target '%s', options: %s %s\n"
 499                         , w->id
 500                                 , w->compare, (w->starts_with && w->ends_with)?"substring":((w->starts_with)?"prefix":((w->ends_with)?"suffix":"exact"))
 501                                 , w->target?w->target->id:w->id
 502                                 , (w->hidden)?"hidden":"-"
 503                                 , (w->debug)?"debug":"-"
 504                 );
 505
 506         return w;
 507 }
 508
 509 // read the apps_groups.conf file
 510 int read_apps_groups_conf(const char *name)
 511 {
 512         char filename[FILENAME_MAX + 1];
 513
 514         snprintfz(filename, FILENAME_MAX, "%s/apps_%s.conf", config_dir, name);
 515
 516         if(unlikely(debug))
 517                 fprintf(stderr, "apps.plugin: process groups file: '%s'\n", filename);
 518
 519         // ----------------------------------------
 520
 521         procfile *ff = procfile_open(filename, " :\t", PROCFILE_FLAG_DEFAULT);
 522         if(!ff) return 1;
 523
 524         procfile_set_quotes(ff, "'\"");
 525
 526         ff = procfile_readall(ff);
 527         if(!ff) {
 528                 procfile_close(ff);
 529                 return 1;
 530         }
 531
 532         unsigned long line, lines = procfile_lines(ff);
 533
 534         for(line = 0; line < lines ;line++) {
 535                 unsigned long word, words = procfile_linewords(ff, line);
 536                 struct target *w = NULL;
 537
 538                 char *t = procfile_lineword(ff, line, 0);
 539                 if(!t || !*t) continue;
 540
 541                 for(word = 0; word < words ;word++) {
 542                         char *s = procfile_lineword(ff, line, word);
 543                         if(!s || !*s) continue;
 544                         if(*s == '#') break;
 545
 546                         if(t == s) continue;
 547
 548                         struct target *n = get_apps_groups_target(s, w);
 549                         if(!n) {
 550                                 error("Cannot create target '%s' (line %d, word %d)", s, line, word);
 551                                 continue;
 552                         }
 553
 554                         if(!w) w = n;
 555                 }
 556
 557                 if(w) {
 558                         int tdebug = 0, thidden = 0;
 559
 560                         while(t[0] == '-' || t[0] == '+') {
 561                                 if(t[0] == '-') thidden = 1;
 562                                 if(t[0] == '+') tdebug = 1;
 563                                 t++;
 564                         }
 565
 566                         strncpyz(w->name, t, MAX_NAME);
 567                         w->hidden = thidden;
 568                         w->debug = tdebug;
 569
 570                         if(unlikely(debug))
 571                                 fprintf(stderr, "apps.plugin: AGGREGATION TARGET NAME '%s' on ID '%s', process name '%s' (%s), aggregated on target '%s', options: %s %s\n"
 572                                                 , w->name
 573                                                 , w->id
 574                                                 , w->compare, (w->starts_with && w->ends_with)?"substring":((w->starts_with)?"prefix":((w->ends_with)?"suffix":"exact"))
 575                                                 , w->target?w->target->id:w->id
 576                                                 , (w->hidden)?"hidden":"-"
 577                                                 , (w->debug)?"debug":"-"
 578                                 );
 579                 }
 580         }
 581
 582         procfile_close(ff);
 583
 584         apps_groups_default_target = get_apps_groups_target("p+!o@w#e$i^r&7*5(-i)l-o_", NULL); // match nothing
 585         if(!apps_groups_default_target)
 586                 error("Cannot create default target");
 587         else
 588                 strncpyz(apps_groups_default_target->name, "other", MAX_NAME);
 589
 590         return 0;
 591 }
 592
 593
 594 // ----------------------------------------------------------------------------
 595 // data to store for each pid
 596 // see: man proc
 597
 598 struct pid_stat {
 599         int32_t pid;
 600         char comm[MAX_COMPARE_NAME + 1];
 601         char cmdline[MAX_CMDLINE + 1];
 602
 603         // char state;
 604         int32_t ppid;
 605         // int32_t pgrp;
 606         // int32_t session;
 607         // int32_t tty_nr;
 608         // int32_t tpgid;
 609         // uint64_t flags;
 610         unsigned long long minflt;
 611         unsigned long long cminflt;
 612         unsigned long long majflt;
 613         unsigned long long cmajflt;
 614         unsigned long long utime;
 615         unsigned long long stime;
 616         unsigned long long cutime;
 617         unsigned long long cstime;
 618         // int64_t priority;
 619         // int64_t nice;
 620         int32_t num_threads;
 621         // int64_t itrealvalue;
 622         // unsigned long long starttime;
 623         // unsigned long long vsize;
 624         unsigned long long rss;
 625         // unsigned long long rsslim;
 626         // unsigned long long starcode;
 627         // unsigned long long endcode;
 628         // unsigned long long startstack;
 629         // unsigned long long kstkesp;
 630         // unsigned long long kstkeip;
 631         // uint64_t signal;
 632         // uint64_t blocked;
 633         // uint64_t sigignore;
 634         // uint64_t sigcatch;
 635         // uint64_t wchan;
 636         // uint64_t nswap;
 637         // uint64_t cnswap;
 638         // int32_t exit_signal;
 639         // int32_t processor;
 640         // uint32_t rt_priority;
 641         // uint32_t policy;
 642         // unsigned long long delayacct_blkio_ticks;
 643         // uint64_t guest_time;
 644         // int64_t cguest_time;
 645
 646         uid_t uid;
 647         gid_t gid;
 648
 649         unsigned long long statm_size;
 650         unsigned long long statm_resident;
 651         unsigned long long statm_share;
 652         unsigned long long statm_text;
 653         unsigned long long statm_lib;
 654         unsigned long long statm_data;
 655         unsigned long long statm_dirty;
 656
 657         unsigned long long io_logical_bytes_read;
 658         unsigned long long io_logical_bytes_written;
 659         unsigned long long io_read_calls;
 660         unsigned long long io_write_calls;
 661         unsigned long long io_storage_bytes_read;
 662         unsigned long long io_storage_bytes_written;
 663         unsigned long long io_cancelled_write_bytes;
 664
 665         // we need the last values
 666         // for all incremental counters
 667         // so that when a process switches users/groups
 668         // we will subtract these values from the old
 669         // target
 670         unsigned long long last_minflt;
 671         unsigned long long last_cminflt;
 672         unsigned long long last_majflt;
 673         unsigned long long last_cmajflt;
 674         unsigned long long last_utime;
 675         unsigned long long last_stime;
 676         unsigned long long last_cutime;
 677         unsigned long long last_cstime;
 678
 679         unsigned long long last_io_logical_bytes_read;
 680         unsigned long long last_io_logical_bytes_written;
 681         unsigned long long last_io_read_calls;
 682         unsigned long long last_io_write_calls;
 683         unsigned long long last_io_storage_bytes_read;
 684         unsigned long long last_io_storage_bytes_written;
 685         unsigned long long last_io_cancelled_write_bytes;
 686
 687 #ifdef AGGREGATE_CHILDREN_TO_PARENTS
 688         unsigned long long old_utime;
 689         unsigned long long old_stime;
 690         unsigned long long old_minflt;
 691         unsigned long long old_majflt;
 692
 693         unsigned long long old_cutime;
 694         unsigned long long old_cstime;
 695         unsigned long long old_cminflt;
 696         unsigned long long old_cmajflt;
 697
 698         unsigned long long fix_cutime;
 699         unsigned long long fix_cstime;
 700         unsigned long long fix_cminflt;
 701         unsigned long long fix_cmajflt;
 702
 703         unsigned long long diff_cutime;
 704         unsigned long long diff_cstime;
 705         unsigned long long diff_cminflt;
 706         unsigned long long diff_cmajflt;
 707 #endif /* AGGREGATE_CHILDREN_TO_PARENTS */
 708
 709         int *fds;                                               // array of fds it uses
 710         int fds_size;                                   // the size of the fds array
 711
 712         int children_count;                             // number of processes directly referencing this
 713         int updated;                                    // 1 when update
 714         int merged;                                             // 1 when it has been merged to its parent
 715         int new_entry;
 716
 717         struct target *target;                  // app_groups.conf targets
 718         struct target *user_target;             // uid based targets
 719         struct target *group_target;    // gid based targets
 720
 721         struct pid_stat *parent;
 722         struct pid_stat *prev;
 723         struct pid_stat *next;
 724
 725 } *root_of_pids = NULL, **all_pids;
 726
 727 long all_pids_count = 0;
 728
 729 struct pid_stat *get_pid_entry(pid_t pid)
 730 {
 731         if(all_pids[pid]) {
 732                 all_pids[pid]->new_entry = 0;
 733                 return all_pids[pid];
 734         }
 735
 736         all_pids[pid] = calloc(sizeof(struct pid_stat), 1);
 737         if(!all_pids[pid]) {
 738                 error("Cannot allocate %lu bytes of memory", (unsigned long)sizeof(struct pid_stat));
 739                 return NULL;
 740         }
 741
 742         all_pids[pid]->fds = calloc(sizeof(int), 100);
 743         if(!all_pids[pid]->fds)
 744                 error("Cannot allocate %ld bytes of memory", (unsigned long)(sizeof(int) * 100));
 745         else all_pids[pid]->fds_size = 100;
 746
 747         if(root_of_pids) root_of_pids->prev = all_pids[pid];
 748         all_pids[pid]->next = root_of_pids;
 749         root_of_pids = all_pids[pid];
 750
 751         all_pids[pid]->pid = pid;
 752         all_pids[pid]->new_entry = 1;
 753
 754         return all_pids[pid];
 755 }
 756
 757 void del_pid_entry(pid_t pid)
 758 {
 759         if(!all_pids[pid]) return;
 760
 761         if(debug) fprintf(stderr, "apps.plugin: process %d %s exited, deleting it.\n", pid, all_pids[pid]->comm);
 762
 763         if(root_of_pids == all_pids[pid]) root_of_pids = all_pids[pid]->next;
 764         if(all_pids[pid]->next) all_pids[pid]->next->prev = all_pids[pid]->prev;
 765         if(all_pids[pid]->prev) all_pids[pid]->prev->next = all_pids[pid]->next;
 766
 767         if(all_pids[pid]->fds) free(all_pids[pid]->fds);
 768         free(all_pids[pid]);
 769         all_pids[pid] = NULL;
 770 }
 771
 772
 773 // ----------------------------------------------------------------------------
 774 // update pids from proc
 775
 776 int read_proc_pid_cmdline(struct pid_stat *p) {
 777         char filename[FILENAME_MAX + 1];
 778         snprintfz(filename, FILENAME_MAX, "%s/proc/%d/cmdline", host_prefix, p->pid);
 779
 780         int fd = open(filename, O_RDONLY, 0666);
 781         if(unlikely(fd == -1)) return 1;
 782
 783         int i, bytes = read(fd, p->cmdline, MAX_CMDLINE);
 784         close(fd);
 785
 786         if(bytes <= 0) {
 787                 // copy the command to the command line
 788                 strncpyz(p->cmdline, p->comm, MAX_CMDLINE);
 789                 return 0;
 790         }
 791
 792         p->cmdline[bytes] = '\0';
 793         for(i = 0; i < bytes ; i++)
 794                 if(!p->cmdline[i]) p->cmdline[i] = ' ';
 795
 796         if(unlikely(debug))
 797                 fprintf(stderr, "Read file '%s' contents: %s\n", filename, p->cmdline);
 798
 799         return 0;
 800 }
 801
 802 int read_proc_pid_ownership(struct pid_stat *p) {
 803         char filename[FILENAME_MAX + 1];
 804
 805         snprintfz(filename, FILENAME_MAX, "%s/proc/%d", host_prefix, p->pid);
 806
 807         // ----------------------------------------
 808         // read uid and gid
 809
 810         struct stat st;
 811         if(stat(filename, &st) != 0)
 812                 return 1;
 813
 814         p->uid = st.st_uid;
 815         p->gid = st.st_gid;
 816
 817         return 0;
 818 }
 819
 820 int read_proc_pid_stat(struct pid_stat *p) {
 821         static procfile *ff = NULL;
 822
 823         char filename[FILENAME_MAX + 1];
 824
 825         snprintfz(filename, FILENAME_MAX, "%s/proc/%d/stat", host_prefix, p->pid);
 826
 827         // ----------------------------------------
 828
 829         int set_quotes = (!ff)?1:0;
 830
 831         ff = procfile_reopen(ff, filename, NULL, PROCFILE_FLAG_NO_ERROR_ON_FILE_IO);
 832         if(!ff) return 1;
 833
 834         // if(set_quotes) procfile_set_quotes(ff, "()");
 835         if(set_quotes) procfile_set_open_close(ff, "(", ")");
 836
 837         ff = procfile_readall(ff);
 838         if(!ff) {
 839                 // procfile_close(ff);
 840                 return 1;
 841         }
 842
 843         file_counter++;
 844
 845         // parse the process name
 846         unsigned int i = 0;
 847         strncpyz(p->comm, procfile_lineword(ff, 0, 1), MAX_COMPARE_NAME);
 848
 849         // p->pid                       = atol(procfile_lineword(ff, 0, 0+i));
 850         // comm is at 1
 851         // p->state                     = *(procfile_lineword(ff, 0, 2+i));
 852         p->ppid                         = (int32_t) atol(procfile_lineword(ff, 0, 3 + i));
 853         // p->pgrp                      = atol(procfile_lineword(ff, 0, 4+i));
 854         // p->session           = atol(procfile_lineword(ff, 0, 5+i));
 855         // p->tty_nr            = atol(procfile_lineword(ff, 0, 6+i));
 856         // p->tpgid                     = atol(procfile_lineword(ff, 0, 7+i));
 857         // p->flags                     = strtoull(procfile_lineword(ff, 0, 8+i), NULL, 10);
 858         p->minflt                       = strtoull(procfile_lineword(ff, 0, 9+i), NULL, 10);
 859         p->cminflt                      = strtoull(procfile_lineword(ff, 0, 10+i), NULL, 10);
 860         p->majflt                       = strtoull(procfile_lineword(ff, 0, 11+i), NULL, 10);
 861         p->cmajflt                      = strtoull(procfile_lineword(ff, 0, 12+i), NULL, 10);
 862         p->utime                        = strtoull(procfile_lineword(ff, 0, 13+i), NULL, 10);
 863         p->stime                        = strtoull(procfile_lineword(ff, 0, 14+i), NULL, 10);
 864         p->cutime                       = strtoull(procfile_lineword(ff, 0, 15+i), NULL, 10);
 865         p->cstime                       = strtoull(procfile_lineword(ff, 0, 16+i), NULL, 10);
 866         // p->priority          = strtoull(procfile_lineword(ff, 0, 17+i), NULL, 10);
 867         // p->nice                      = strtoull(procfile_lineword(ff, 0, 18+i), NULL, 10);
 868         p->num_threads          = (int32_t) atol(procfile_lineword(ff, 0, 19 + i));
 869         // p->itrealvalue       = strtoull(procfile_lineword(ff, 0, 20+i), NULL, 10);
 870         // p->starttime         = strtoull(procfile_lineword(ff, 0, 21+i), NULL, 10);
 871         // p->vsize                     = strtoull(procfile_lineword(ff, 0, 22+i), NULL, 10);
 872         p->rss                          = strtoull(procfile_lineword(ff, 0, 23+i), NULL, 10);
 873         // p->rsslim            = strtoull(procfile_lineword(ff, 0, 24+i), NULL, 10);
 874         // p->starcode          = strtoull(procfile_lineword(ff, 0, 25+i), NULL, 10);
 875         // p->endcode           = strtoull(procfile_lineword(ff, 0, 26+i), NULL, 10);
 876         // p->startstack        = strtoull(procfile_lineword(ff, 0, 27+i), NULL, 10);
 877         // p->kstkesp           = strtoull(procfile_lineword(ff, 0, 28+i), NULL, 10);
 878         // p->kstkeip           = strtoull(procfile_lineword(ff, 0, 29+i), NULL, 10);
 879         // p->signal            = strtoull(procfile_lineword(ff, 0, 30+i), NULL, 10);
 880         // p->blocked           = strtoull(procfile_lineword(ff, 0, 31+i), NULL, 10);
 881         // p->sigignore         = strtoull(procfile_lineword(ff, 0, 32+i), NULL, 10);
 882         // p->sigcatch          = strtoull(procfile_lineword(ff, 0, 33+i), NULL, 10);
 883         // p->wchan                     = strtoull(procfile_lineword(ff, 0, 34+i), NULL, 10);
 884         // p->nswap                     = strtoull(procfile_lineword(ff, 0, 35+i), NULL, 10);
 885         // p->cnswap            = strtoull(procfile_lineword(ff, 0, 36+i), NULL, 10);
 886         // p->exit_signal       = atol(procfile_lineword(ff, 0, 37+i));
 887         // p->processor         = atol(procfile_lineword(ff, 0, 38+i));
 888         // p->rt_priority       = strtoul(procfile_lineword(ff, 0, 39+i), NULL, 10);
 889         // p->policy            = strtoul(procfile_lineword(ff, 0, 40+i), NULL, 10);
 890         // p->delayacct_blkio_ticks             = strtoull(procfile_lineword(ff, 0, 41+i), NULL, 10);
 891         // p->guest_time        = strtoull(procfile_lineword(ff, 0, 42+i), NULL, 10);
 892         // p->cguest_time       = strtoull(procfile_lineword(ff, 0, 43), NULL, 10);
 893
 894         if(debug || (p->target && p->target->debug))
 895                 fprintf(stderr, "apps.plugin: READ PROC/PID/STAT: %s/proc/%d/stat, process: '%s' VALUES: utime=%llu, stime=%llu, cutime=%llu, cstime=%llu, minflt=%llu, majflt=%llu, cminflt=%llu, cmajflt=%llu, threads=%d\n", host_prefix, p->pid, p->comm, p->utime, p->stime, p->cutime, p->cstime, p->minflt, p->majflt, p->cminflt, p->cmajflt, p->num_threads);
 896
 897         // procfile_close(ff);
 898         return 0;
 899 }
 900
 901 int read_proc_pid_statm(struct pid_stat *p) {
 902         static procfile *ff = NULL;
 903
 904         char filename[FILENAME_MAX + 1];
 905
 906         snprintfz(filename, FILENAME_MAX, "%s/proc/%d/statm", host_prefix, p->pid);
 907
 908         ff = procfile_reopen(ff, filename, NULL, PROCFILE_FLAG_NO_ERROR_ON_FILE_IO);
 909         if(!ff) return 1;
 910
 911         ff = procfile_readall(ff);
 912         if(!ff) {
 913                 // procfile_close(ff);
 914                 return 1;
 915         }
 916
 917         file_counter++;
 918
 919         p->statm_size                   = strtoull(procfile_lineword(ff, 0, 0), NULL, 10);
 920         p->statm_resident               = strtoull(procfile_lineword(ff, 0, 1), NULL, 10);
 921         p->statm_share                  = strtoull(procfile_lineword(ff, 0, 2), NULL, 10);
 922         p->statm_text                   = strtoull(procfile_lineword(ff, 0, 3), NULL, 10);
 923         p->statm_lib                    = strtoull(procfile_lineword(ff, 0, 4), NULL, 10);
 924         p->statm_data                   = strtoull(procfile_lineword(ff, 0, 5), NULL, 10);
 925         p->statm_dirty                  = strtoull(procfile_lineword(ff, 0, 6), NULL, 10);
 926
 927         // procfile_close(ff);
 928         return 0;
 929 }
 930
 931 int read_proc_pid_io(struct pid_stat *p) {
 932         static procfile *ff = NULL;
 933
 934         char filename[FILENAME_MAX + 1];
 935
 936         snprintfz(filename, FILENAME_MAX, "%s/proc/%d/io", host_prefix, p->pid);
 937
 938         ff = procfile_reopen(ff, filename, NULL, PROCFILE_FLAG_NO_ERROR_ON_FILE_IO);
 939         if(!ff) return 1;
 940
 941         ff = procfile_readall(ff);
 942         if(!ff) {
 943                 // procfile_close(ff);
 944                 return 1;
 945         }
 946
 947         file_counter++;
 948
 949         p->io_logical_bytes_read                = strtoull(procfile_lineword(ff, 0, 1), NULL, 10);
 950         p->io_logical_bytes_written     = strtoull(procfile_lineword(ff, 1, 1), NULL, 10);
 951         p->io_read_calls                                = strtoull(procfile_lineword(ff, 2, 1), NULL, 10);
 952         p->io_write_calls                               = strtoull(procfile_lineword(ff, 3, 1), NULL, 10);
 953         p->io_storage_bytes_read                = strtoull(procfile_lineword(ff, 4, 1), NULL, 10);
 954         p->io_storage_bytes_written     = strtoull(procfile_lineword(ff, 5, 1), NULL, 10);
 955         p->io_cancelled_write_bytes             = strtoull(procfile_lineword(ff, 6, 1), NULL, 10);
 956
 957         // procfile_close(ff);
 958         return 0;
 959 }
 960
 961
 962 // ----------------------------------------------------------------------------
 963 // file descriptor
 964 // this is used to keep a global list of all open files of the system
 965 // it is needed in order to calculate the unique files processes have open
 966
 967 #define FILE_DESCRIPTORS_INCREASE_STEP 100
 968
 969 struct file_descriptor {
 970         avl avl;
 971 #ifdef NETDATA_INTERNAL_CHECKS
 972         uint32_t magic;
 973 #endif /* NETDATA_INTERNAL_CHECKS */
 974         uint32_t hash;
 975         const char *name;
 976         int type;
 977         int count;
 978         int pos;
 979 } *all_files = NULL;
 980
 981 int all_files_len = 0;
 982 int all_files_size = 0;
 983
 984 int file_descriptor_compare(void* a, void* b) {
 985 #ifdef NETDATA_INTERNAL_CHECKS
 986         if(((struct file_descriptor *)a)->magic != 0x0BADCAFE || ((struct file_descriptor *)b)->magic != 0x0BADCAFE)
 987                 error("Corrupted index data detected. Please report this.");
 988 #endif /* NETDATA_INTERNAL_CHECKS */
 989
 990         if(((struct file_descriptor *)a)->hash < ((struct file_descriptor *)b)->hash)
 991                 return -1;
 992
 993         else if(((struct file_descriptor *)a)->hash > ((struct file_descriptor *)b)->hash)
 994                 return 1;
 995
 996         else
 997                 return strcmp(((struct file_descriptor *)a)->name, ((struct file_descriptor *)b)->name);
 998 }
 999
1000 int file_descriptor_iterator(avl *a) { if(a) {}; return 0; }
1001
1002 avl_tree all_files_index = {
1003                 NULL,
1004                 file_descriptor_compare
1005 };
1006
1007 static struct file_descriptor *file_descriptor_find(const char *name, uint32_t hash) {
1008         struct file_descriptor *result = NULL, tmp;
1009         tmp.hash = (hash)?hash:simple_hash(name);
1010         tmp.name = name;
1011         tmp.count = 0;
1012         tmp.pos = 0;
1013 #ifdef NETDATA_INTERNAL_CHECKS
1014         tmp.magic = 0x0BADCAFE;
1015 #endif /* NETDATA_INTERNAL_CHECKS */
1016
1017         avl_search(&all_files_index, (avl *) &tmp, file_descriptor_iterator, (avl **) &result);
1018         return result;
1019 }
1020
1021 #define file_descriptor_add(fd) avl_insert(&all_files_index, (avl *)(fd))
1022 #define file_descriptor_remove(fd) avl_remove(&all_files_index, (avl *)(fd))
1023
1024 #define FILETYPE_OTHER 0
1025 #define FILETYPE_FILE 1
1026 #define FILETYPE_PIPE 2
1027 #define FILETYPE_SOCKET 3
1028 #define FILETYPE_INOTIFY 4
1029 #define FILETYPE_EVENTFD 5
1030 #define FILETYPE_EVENTPOLL 6
1031 #define FILETYPE_TIMERFD 7
1032 #define FILETYPE_SIGNALFD 8
1033
1034 void file_descriptor_not_used(int id)
1035 {
1036         if(id > 0 && id < all_files_size) {
1037
1038 #ifdef NETDATA_INTERNAL_CHECKS
1039                 if(all_files[id].magic != 0x0BADCAFE) {
1040                         error("Ignoring request to remove empty file id %d.", id);
1041                         return;
1042                 }
1043 #endif /* NETDATA_INTERNAL_CHECKS */
1044
1045                 if(debug) fprintf(stderr, "apps.plugin: decreasing slot %d (count = %d).\n", id, all_files[id].count);
1046
1047                 if(all_files[id].count > 0) {
1048                         all_files[id].count--;
1049
1050                         if(!all_files[id].count) {
1051                                 if(debug) fprintf(stderr, "apps.plugin:   >> slot %d is empty.\n", id);
1052                                 file_descriptor_remove(&all_files[id]);
1053 #ifdef NETDATA_INTERNAL_CHECKS
1054                                 all_files[id].magic = 0x00000000;
1055 #endif /* NETDATA_INTERNAL_CHECKS */
1056                                 all_files_len--;
1057                         }
1058                 }
1059                 else
1060                         error("Request to decrease counter of fd %d (%s), while the use counter is 0", id, all_files[id].name);
1061         }
1062         else    error("Request to decrease counter of fd %d, which is outside the array size (1 to %d)", id, all_files_size);
1063 }
1064
1065 int file_descriptor_find_or_add(const char *name)
1066 {
1067         static int last_pos = 0;
1068         uint32_t hash = simple_hash(name);
1069
1070         if(debug) fprintf(stderr, "apps.plugin: adding or finding name '%s' with hash %u\n", name, hash);
1071
1072         struct file_descriptor *fd = file_descriptor_find(name, hash);
1073         if(fd) {
1074                 // found
1075                 if(debug) fprintf(stderr, "apps.plugin:   >> found on slot %d\n", fd->pos);
1076                 fd->count++;
1077                 return fd->pos;
1078         }
1079         // not found
1080
1081         // check we have enough memory to add it
1082         if(!all_files || all_files_len == all_files_size) {
1083                 void *old = all_files;
1084                 int i;
1085
1086                 // there is no empty slot
1087                 if(debug) fprintf(stderr, "apps.plugin: extending fd array to %d entries\n", all_files_size + FILE_DESCRIPTORS_INCREASE_STEP);
1088                 all_files = realloc(all_files, (all_files_size + FILE_DESCRIPTORS_INCREASE_STEP) * sizeof(struct file_descriptor));
1089
1090                 // if the address changed, we have to rebuild the index
1091                 // since all pointers are now invalid
1092                 if(old && old != (void *)all_files) {
1093                         if(debug) fprintf(stderr, "apps.plugin:   >> re-indexing.\n");
1094                         all_files_index.root = NULL;
1095                         for(i = 0; i < all_files_size; i++) {
1096                                 if(!all_files[i].count) continue;
1097                                 file_descriptor_add(&all_files[i]);
1098                         }
1099                         if(debug) fprintf(stderr, "apps.plugin:   >> re-indexing done.\n");
1100                 }
1101
1102                 for(i = all_files_size; i < (all_files_size + FILE_DESCRIPTORS_INCREASE_STEP); i++) {
1103                         all_files[i].count = 0;
1104                         all_files[i].name = NULL;
1105 #ifdef NETDATA_INTERNAL_CHECKS
1106                         all_files[i].magic = 0x00000000;
1107 #endif /* NETDATA_INTERNAL_CHECKS */
1108                         all_files[i].pos = i;
1109                 }
1110
1111                 if(!all_files_size) all_files_len = 1;
1112                 all_files_size += FILE_DESCRIPTORS_INCREASE_STEP;
1113         }
1114
1115         if(debug) fprintf(stderr, "apps.plugin:   >> searching for empty slot.\n");
1116
1117         // search for an empty slot
1118         int i, c;
1119         for(i = 0, c = last_pos ; i < all_files_size ; i++, c++) {
1120                 if(c >= all_files_size) c = 0;
1121                 if(c == 0) continue;
1122
1123                 if(!all_files[c].count) {
1124                         if(debug) fprintf(stderr, "apps.plugin:   >> Examining slot %d.\n", c);
1125
1126 #ifdef NETDATA_INTERNAL_CHECKS
1127                         if(all_files[c].magic == 0x0BADCAFE && all_files[c].name && file_descriptor_find(all_files[c].name, all_files[c].hash))
1128                                 error("fd on position %d is not cleared properly. It still has %s in it.\n", c, all_files[c].name);
1129 #endif /* NETDATA_INTERNAL_CHECKS */
1130
1131                         if(debug) fprintf(stderr, "apps.plugin:   >> %s fd position %d for %s (last name: %s)\n", all_files[c].name?"re-using":"using", c, name, all_files[c].name);
1132                         if(all_files[c].name) free((void *)all_files[c].name);
1133                         all_files[c].name = NULL;
1134                         last_pos = c;
1135                         break;
1136                 }
1137         }
1138         if(i == all_files_size) {
1139                 fatal("We should find an empty slot, but there isn't any");
1140                 exit(1);
1141         }
1142         if(debug) fprintf(stderr, "apps.plugin:   >> updating slot %d.\n", c);
1143
1144         all_files_len++;
1145
1146         // else we have an empty slot in 'c'
1147
1148         int type;
1149         if(name[0] == '/') type = FILETYPE_FILE;
1150         else if(strncmp(name, "pipe:", 5) == 0) type = FILETYPE_PIPE;
1151         else if(strncmp(name, "socket:", 7) == 0) type = FILETYPE_SOCKET;
1152         else if(strcmp(name, "anon_inode:inotify") == 0 || strcmp(name, "inotify") == 0) type = FILETYPE_INOTIFY;
1153         else if(strcmp(name, "anon_inode:[eventfd]") == 0) type = FILETYPE_EVENTFD;
1154         else if(strcmp(name, "anon_inode:[eventpoll]") == 0) type = FILETYPE_EVENTPOLL;
1155         else if(strcmp(name, "anon_inode:[timerfd]") == 0) type = FILETYPE_TIMERFD;
1156         else if(strcmp(name, "anon_inode:[signalfd]") == 0) type = FILETYPE_SIGNALFD;
1157         else if(strncmp(name, "anon_inode:", 11) == 0) {
1158                 if(debug) fprintf(stderr, "apps.plugin: FIXME: unknown anonymous inode: %s\n", name);
1159                 type = FILETYPE_OTHER;
1160         }
1161         else {
1162                 if(debug) fprintf(stderr, "apps.plugin: FIXME: cannot understand linkname: %s\n", name);
1163                 type = FILETYPE_OTHER;
1164         }
1165
1166         all_files[c].name = strdup(name);
1167         all_files[c].hash = hash;
1168         all_files[c].type = type;
1169         all_files[c].pos  = c;
1170         all_files[c].count = 1;
1171 #ifdef NETDATA_INTERNAL_CHECKS
1172         all_files[c].magic = 0x0BADCAFE;
1173 #endif /* NETDATA_INTERNAL_CHECKS */
1174         file_descriptor_add(&all_files[c]);
1175
1176         if(debug) fprintf(stderr, "apps.plugin: using fd position %d (name: %s)\n", c, all_files[c].name);
1177
1178         return c;
1179 }
1180
1181 int read_pid_file_descriptors(struct pid_stat *p) {
1182         char dirname[FILENAME_MAX+1];
1183
1184         snprintfz(dirname, FILENAME_MAX, "%s/proc/%d/fd", host_prefix, p->pid);
1185         DIR *fds = opendir(dirname);
1186         if(fds) {
1187                 int c;
1188                 struct dirent *de;
1189                 char fdname[FILENAME_MAX + 1];
1190                 char linkname[FILENAME_MAX + 1];
1191
1192                 // make the array negative
1193                 for(c = 0 ; c < p->fds_size ; c++)
1194                         p->fds[c] = -p->fds[c];
1195
1196                 while((de = readdir(fds))) {
1197                         if(strcmp(de->d_name, ".") == 0 || strcmp(de->d_name, "..") == 0)
1198                                 continue;
1199
1200                         // check if the fds array is small
1201                         int fdid = atoi(de->d_name);
1202                         if(fdid < 0) continue;
1203                         if(fdid >= p->fds_size) {
1204                                 // it is small, extend it
1205                                 if(debug) fprintf(stderr, "apps.plugin: extending fd memory slots for %s from %d to %d\n", p->comm, p->fds_size, fdid + 100);
1206                                 p->fds = realloc(p->fds, (fdid + 100) * sizeof(int));
1207                                 if(!p->fds) {
1208                                         fatal("Cannot re-allocate fds for %s", p->comm);
1209                                         break;
1210                                 }
1211
1212                                 // and initialize it
1213                                 for(c = p->fds_size ; c < (fdid + 100) ; c++) p->fds[c] = 0;
1214                                 p->fds_size = fdid + 100;
1215                         }
1216
1217                         if(p->fds[fdid] == 0) {
1218                                 // we don't know this fd, get it
1219
1220                                 sprintf(fdname, "%s/proc/%d/fd/%s", host_prefix, p->pid, de->d_name);
1221                                 ssize_t l = readlink(fdname, linkname, FILENAME_MAX);
1222                                 if(l == -1) {
1223                                         if(debug || (p->target && p->target->debug)) {
1224                                                 if(debug || (p->target && p->target->debug))
1225                                                         error("Cannot read link %s", fdname);
1226                                         }
1227                                         continue;
1228                                 }
1229                                 linkname[l] = '\0';
1230                                 file_counter++;
1231
1232                                 // if another process already has this, we will get
1233                                 // the same id
1234                                 p->fds[fdid] = file_descriptor_find_or_add(linkname);
1235                         }
1236
1237                         // else make it positive again, we need it
1238                         // of course, the actual file may have changed, but we don't care so much
1239                         // FIXME: we could compare the inode as returned by readdir direct structure
1240                         else p->fds[fdid] = -p->fds[fdid];
1241                 }
1242                 closedir(fds);
1243
1244                 // remove all the negative file descriptors
1245                 for(c = 0 ; c < p->fds_size ; c++) if(p->fds[c] < 0) {
1246                         file_descriptor_not_used(-p->fds[c]);
1247                         p->fds[c] = 0;
1248                 }
1249         }
1250         else return 1;
1251
1252         return 0;
1253 }
1254
1255 // ----------------------------------------------------------------------------
1256
1257 // 1. read all files in /proc
1258 // 2. for each numeric directory:
1259 //    i.   read /proc/pid/stat
1260 //    ii.  read /proc/pid/statm
1261 //    iii. read /proc/pid/io (requires root access)
1262 //    iii. read the entries in directory /proc/pid/fd (requires root access)
1263 //         for each entry:
1264 //         a. find or create a struct file_descriptor
1265 //         b. cleanup any old/unused file_descriptors
1266
1267 // after all these, some pids may be linked to targets, while others may not
1268
1269 // in case of errors, only 1 every 1000 errors is printed
1270 // to avoid filling up all disk space
1271 // if debug is enabled, all errors are printed
1272
1273 int collect_data_for_all_processes_from_proc(void)
1274 {
1275         char dirname[FILENAME_MAX + 1];
1276
1277         snprintfz(dirname, FILENAME_MAX, "%s/proc", host_prefix);
1278         DIR *dir = opendir(dirname);
1279         if(!dir) return 0;
1280
1281         struct dirent *file = NULL;
1282         struct pid_stat *p = NULL;
1283
1284         // mark them all as un-updated
1285         all_pids_count = 0;
1286         for(p = root_of_pids; p ; p = p->next) {
1287                 all_pids_count++;
1288                 p->parent = NULL;
1289                 p->updated = 0;
1290                 p->children_count = 0;
1291                 p->merged = 0;
1292                 p->new_entry = 0;
1293
1294         p->last_minflt  = p->minflt;
1295         p->last_cminflt  = p->cminflt;
1296         p->last_majflt  = p->majflt;
1297         p->last_cmajflt  = p->cmajflt;
1298         p->last_utime  = p->utime;
1299         p->last_stime  = p->stime;
1300         p->last_cutime  = p->cutime;
1301         p->last_cstime  = p->cstime;
1302
1303         p->last_io_logical_bytes_read  = p->io_logical_bytes_read;
1304         p->last_io_logical_bytes_written  = p->io_logical_bytes_written;
1305         p->last_io_read_calls  = p->io_read_calls;
1306         p->last_io_write_calls  = p->io_write_calls;
1307         p->last_io_storage_bytes_read  = p->io_storage_bytes_read;
1308         p->last_io_storage_bytes_written  = p->io_storage_bytes_written;
1309         p->last_io_cancelled_write_bytes  = p->io_cancelled_write_bytes;
1310         }
1311
1312         while((file = readdir(dir))) {
1313                 char *endptr = file->d_name;
1314                 pid_t pid = (pid_t) strtoul(file->d_name, &endptr, 10);
1315
1316                 // make sure we read a valid number
1317                 if(unlikely(pid <= 0 || pid > pid_max || endptr == file->d_name || *endptr != '\0'))
1318                         continue;
1319
1320                 p = get_pid_entry(pid);
1321                 if(unlikely(!p)) continue;
1322
1323
1324                 // --------------------------------------------------------------------
1325                 // /proc/<pid>/stat
1326
1327                 if(unlikely(read_proc_pid_stat(p))) {
1328                                 error("Cannot process %s/proc/%d/stat", host_prefix, pid);
1329
1330                         // there is no reason to proceed if we cannot get its status
1331                         continue;
1332                 }
1333
1334                 // check its parent pid
1335                 if(unlikely(p->ppid < 0 || p->ppid > pid_max)) {
1336                                 error("Pid %d states invalid parent pid %d. Using 0.", pid, p->ppid);
1337
1338                         p->ppid = 0;
1339                 }
1340
1341                 // --------------------------------------------------------------------
1342                 // /proc/<pid>/cmdline
1343
1344                 if(proc_pid_cmdline_is_needed) {
1345                         if(unlikely(read_proc_pid_cmdline(p))) {
1346                                         error("Cannot process %s/proc/%d/cmdline", host_prefix, pid);
1347                         }
1348                 }
1349
1350                 // --------------------------------------------------------------------
1351                 // /proc/<pid>/statm
1352
1353                 if(unlikely(read_proc_pid_statm(p))) {
1354                                 error("Cannot process %s/proc/%d/statm", host_prefix, pid);
1355
1356                         // there is no reason to proceed if we cannot get its memory status
1357                         continue;
1358                 }
1359
1360
1361                 // --------------------------------------------------------------------
1362                 // /proc/<pid>/io
1363
1364                 if(unlikely(read_proc_pid_io(p))) {
1365                                 error("Cannot process %s/proc/%d/io", host_prefix, pid);
1366
1367                         // on systems without /proc/X/io
1368                         // allow proceeding without I/O information
1369                         // continue;
1370                 }
1371
1372                 // --------------------------------------------------------------------
1373                 // <pid> ownership
1374
1375                 if(unlikely(read_proc_pid_ownership(p))) {
1376                                 error("Cannot stat %s/proc/%d", host_prefix, pid);
1377                 }
1378
1379                 // --------------------------------------------------------------------
1380                 // link it
1381
1382                 // check if it is target
1383                 // we do this only once, the first time this pid is loaded
1384                 if(unlikely(p->new_entry)) {
1385                         if(debug) fprintf(stderr, "apps.plugin: \tJust added %s\n", p->comm);
1386                         uint32_t hash = simple_hash(p->comm);
1387                         size_t pclen = strlen(p->comm);
1388
1389                         struct target *w;
1390                         for(w = apps_groups_root_target; w ; w = w->next) {
1391                                 // if(debug || (p->target && p->target->debug)) fprintf(stderr, "apps.plugin: \t\tcomparing '%s' with '%s'\n", w->compare, p->comm);
1392
1393                                 // find it - 4 cases:
1394                                 // 1. the target is not a pattern
1395                                 // 2. the target has the prefix
1396                                 // 3. the target has the suffix
1397                                 // 4. the target is something inside cmdline
1398                                 if(     (!w->starts_with && !w->ends_with && w->comparehash == hash && !strcmp(w->compare, p->comm))
1399                                        || (w->starts_with && !w->ends_with && !strncmp(w->compare, p->comm, w->comparelen))
1400                                        || (!w->starts_with && w->ends_with && pclen >= w->comparelen && !strcmp(w->compare, &p->comm[pclen - w->comparelen]))
1401                                        || (proc_pid_cmdline_is_needed && w->starts_with && w->ends_with && strstr(p->cmdline, w->compare))
1402                                                 ) {
1403                                         if(w->target) p->target = w->target;
1404                                         else p->target = w;
1405
1406                                         if(debug || (p->target && p->target->debug))
1407                                                 fprintf(stderr, "apps.plugin: \t\t%s linked to target %s\n", p->comm, p->target->name);
1408                                 }
1409                         }
1410                 }
1411
1412                 // --------------------------------------------------------------------
1413                 // /proc/<pid>/fd
1414
1415                 if(unlikely(read_pid_file_descriptors(p))) {
1416                                 error("Cannot process entries in %s/proc/%d/fd", host_prefix, pid);
1417                 }
1418
1419                 // --------------------------------------------------------------------
1420                 // done!
1421
1422                 // mark it as updated
1423                 p->updated = 1;
1424         }
1425
1426         closedir(dir);
1427
1428         return 1;
1429 }
1430
1431
1432 // ----------------------------------------------------------------------------
1433
1434 #ifdef AGGREGATE_CHILDREN_TO_PARENTS
1435 // print a tree view of all processes
1436 int debug_childrens_aggregations(pid_t pid, int level) {
1437         struct pid_stat *p = NULL;
1438         char b[level+3];
1439         int i, ret = 0;
1440
1441         for(i = 0; i < level; i++) b[i] = '\t';
1442         b[level] = '|';
1443         b[level+1] = '-';
1444         b[level+2] = '\0';
1445
1446         for(p = root_of_pids; p ; p = p->next) {
1447                 if(p->ppid == pid) {
1448                         ret += debug_childrens_aggregations(p->pid, level+1);
1449                 }
1450         }
1451
1452         p = all_pids[pid];
1453         if(p) {
1454                 if(!p->updated) ret += 1;
1455                 if(ret) fprintf(stderr, "%s %s %d [%s, %s] c=%d u=%llu+%llu, s=%llu+%llu, cu=%llu+%llu, cs=%llu+%llu, n=%llu+%llu, j=%llu+%llu, cn=%llu+%llu, cj=%llu+%llu\n"
1456                         , b, p->comm, p->pid, p->updated?"OK":"KILLED", p->target->name, p->children_count
1457                         , p->utime, p->utime - p->old_utime
1458                         , p->stime, p->stime - p->old_stime
1459                         , p->cutime, p->cutime - p->old_cutime
1460                         , p->cstime, p->cstime - p->old_cstime
1461                         , p->minflt, p->minflt - p->old_minflt
1462                         , p->majflt, p->majflt - p->old_majflt
1463                         , p->cminflt, p->cminflt - p->old_cminflt
1464                         , p->cmajflt, p->cmajflt - p->old_cmajflt
1465                         );
1466         }
1467
1468         return ret;
1469 }
1470 #endif /* AGGREGATE_CHILDREN_TO_PARENTS */
1471
1472
1473
1474 // ----------------------------------------------------------------------------
1475 // update statistics on the targets
1476
1477 // 1. link all childs to their parents
1478 // 2. go from bottom to top, marking as merged all childs to their parents
1479 //    this step links all parents without a target to the child target, if any
1480 // 3. link all top level processes (the ones not merged) to the default target
1481 // 4. go from top to bottom, linking all childs without a target, to their parent target
1482 //    after this step, all processes have a target
1483 // [5. for each killed pid (updated = 0), remove its usage from its target]
1484 // 6. zero all apps_groups_targets
1485 // 7. concentrate all values on the apps_groups_targets
1486 // 8. remove all killed processes
1487 // 9. find the unique file count for each target
1488 // check: update_apps_groups_statistics()
1489
1490 void link_all_processes_to_their_parents(void) {
1491         struct pid_stat *p = NULL;
1492
1493         // link all children to their parents
1494         // and update children count on parents
1495         for(p = root_of_pids; p ; p = p->next) {
1496                 // for each process found running
1497
1498                 if(p->ppid > 0
1499                                 && p->ppid <= pid_max
1500                                 && all_pids[p->ppid]
1501                         ) {
1502                         // for valid processes
1503
1504                         if(debug || (p->target && p->target->debug))
1505                                 fprintf(stderr, "apps.plugin: \tparent of %d (%s) is %d (%s)\n", p->pid, p->comm, p->ppid, all_pids[p->ppid]->comm);
1506
1507                         p->parent = all_pids[p->ppid];
1508                         p->parent->children_count++;
1509                 }
1510                 else if(p->ppid != 0)
1511                         error("pid %d %s states parent %d, but the later does not exist.", p->pid, p->comm, p->ppid);
1512         }
1513 }
1514
1515 #ifdef AGGREGATE_CHILDREN_TO_PARENTS
1516 void aggregate_children_to_parents(void) {
1517         struct pid_stat *p = NULL;
1518
1519         // for each killed process, remove its values from the parents
1520         // sums (we had already added them in a previous loop)
1521         for(p = root_of_pids; p ; p = p->next) {
1522                 if(p->updated) continue;
1523
1524                 if(debug) fprintf(stderr, "apps.plugin: UNMERGING %d %s\n", p->pid, p->comm);
1525
1526                 unsigned long long diff_utime = p->utime + p->cutime + p->fix_cutime;
1527                 unsigned long long diff_stime = p->stime + p->cstime + p->fix_cstime;
1528                 unsigned long long diff_minflt = p->minflt + p->cminflt + p->fix_cminflt;
1529                 unsigned long long diff_majflt = p->majflt + p->cmajflt + p->fix_cmajflt;
1530
1531                 struct pid_stat *t = p;
1532                 while((t = t->parent)) {
1533                         if(!t->updated) continue;
1534
1535                         unsigned long long x;
1536                         if(diff_utime && t->diff_cutime) {
1537                                 x = (t->diff_cutime < diff_utime)?t->diff_cutime:diff_utime;
1538                                 diff_utime -= x;
1539                                 t->diff_cutime -= x;
1540                                 t->fix_cutime += x;
1541                                 if(debug) fprintf(stderr, "apps.plugin: \t cutime %llu from %d %s %s\n", x, t->pid, t->comm, t->target->name);
1542                         }
1543                         if(diff_stime && t->diff_cstime) {
1544                                 x = (t->diff_cstime < diff_stime)?t->diff_cstime:diff_stime;
1545                                 diff_stime -= x;
1546                                 t->diff_cstime -= x;
1547                                 t->fix_cstime += x;
1548                                 if(debug) fprintf(stderr, "apps.plugin: \t cstime %llu from %d %s %s\n", x, t->pid, t->comm, t->target->name);
1549                         }
1550                         if(diff_minflt && t->diff_cminflt) {
1551                                 x = (t->diff_cminflt < diff_minflt)?t->diff_cminflt:diff_minflt;
1552                                 diff_minflt -= x;
1553                                 t->diff_cminflt -= x;
1554                                 t->fix_cminflt += x;
1555                                 if(debug) fprintf(stderr, "apps.plugin: \t cminflt %llu from %d %s %s\n", x, t->pid, t->comm, t->target->name);
1556                         }
1557                         if(diff_majflt && t->diff_cmajflt) {
1558                                 x = (t->diff_cmajflt < diff_majflt)?t->diff_cmajflt:diff_majflt;
1559                                 diff_majflt -= x;
1560                                 t->diff_cmajflt -= x;
1561                                 t->fix_cmajflt += x;
1562                                 if(debug) fprintf(stderr, "apps.plugin: \t cmajflt %llu from %d %s %s\n", x, t->pid, t->comm, t->target->name);
1563                         }
1564                 }
1565
1566                 if(diff_utime) error("Cannot fix up utime %llu", diff_utime);
1567                 if(diff_stime) error("Cannot fix up stime %llu", diff_stime);
1568                 if(diff_minflt) error("Cannot fix up minflt %llu", diff_minflt);
1569                 if(diff_majflt) error("Cannot fix up majflt %llu", diff_majflt);
1570         }
1571 }
1572 #endif /* AGGREGATE_CHILDREN_TO_PARENTS */
1573
1574 void cleanup_non_existing_pids(void) {
1575         int c;
1576         struct pid_stat *p = NULL;
1577
1578         for(p = root_of_pids; p ;) {
1579                 if(!p->updated) {
1580 //                      fprintf(stderr, "\tEXITED %d %s [parent %d %s, target %s] utime=%llu, stime=%llu, cutime=%llu, cstime=%llu, minflt=%llu, majflt=%llu, cminflt=%llu, cmajflt=%llu\n", p->pid, p->comm, p->parent->pid, p->parent->comm, p->target->name,  p->utime, p->stime, p->cutime, p->cstime, p->minflt, p->majflt, p->cminflt, p->cmajflt);
1581
1582                         for(c = 0 ; c < p->fds_size ; c++) if(p->fds[c] > 0) {
1583                                 file_descriptor_not_used(p->fds[c]);
1584                                 p->fds[c] = 0;
1585                         }
1586
1587                         pid_t r = p->pid;
1588                         p = p->next;
1589                         del_pid_entry(r);
1590                 }
1591                 else p = p->next;
1592         }
1593 }
1594
1595 void apply_apps_groups_targets_inheritance(void) {
1596         struct pid_stat *p = NULL;
1597
1598         // children that do not have a target
1599         // inherit their target from their parent
1600         int found = 1;
1601         while(found) {
1602                 found = 0;
1603                 for(p = root_of_pids; p ; p = p->next) {
1604                         // if this process does not have a target
1605                         // and it has a parent
1606                         // and its parent has a target
1607                         // then, set the parent's target to this process
1608                         if(unlikely(!p->target && p->parent && p->parent->target)) {
1609                                 p->target = p->parent->target;
1610                                 found++;
1611
1612                                 if(debug || (p->target && p->target->debug))
1613                                         fprintf(stderr, "apps.plugin: \t\tTARGET INHERITANCE: %s is inherited by %d (%s) from its parent %d (%s).\n", p->target->name, p->pid, p->comm, p->parent->pid, p->parent->comm);
1614                         }
1615                 }
1616         }
1617
1618
1619         // find all the procs with 0 childs and merge them to their parents
1620         // repeat, until nothing more can be done.
1621         found = 1;
1622         while(found) {
1623                 found = 0;
1624                 for(p = root_of_pids; p ; p = p->next) {
1625                         // if this process does not have any children
1626                         // and is not already merged
1627                         // and has a parent
1628                         // and its parent has children
1629                         // and the target of this process and its parent is the same, or the parent does not have a target
1630                         // and its parent is not init
1631                         // then, mark them as merged.
1632                         if(unlikely(
1633                                         !p->children_count
1634                                         && !p->merged
1635                                         && p->parent
1636                                         && p->parent->children_count
1637                                         && (p->target == p->parent->target || !p->parent->target)
1638                                         && p->ppid != 1
1639                                 )) {
1640                                 p->parent->children_count--;
1641                                 p->merged = 1;
1642
1643                                 // the parent inherits the child's target, if it does not have a target itself
1644                                 if(unlikely(p->target && !p->parent->target)) {
1645                                         p->parent->target = p->target;
1646
1647                                         if(debug || (p->target && p->target->debug))
1648                                                 fprintf(stderr, "apps.plugin: \t\tTARGET INHERITANCE: %s is inherited by %d (%s) from its child %d (%s).\n", p->target->name, p->parent->pid, p->parent->comm, p->pid, p->comm);
1649                                 }
1650
1651                                 found++;
1652                         }
1653                 }
1654
1655                 if(debug)
1656                         fprintf(stderr, "apps.plugin: merged %d processes\n", found);
1657         }
1658
1659         // init goes always to default target
1660         if(all_pids[1])
1661                 all_pids[1]->target = apps_groups_default_target;
1662
1663         // give a default target on all top level processes
1664         for(p = root_of_pids; p ; p = p->next) {
1665                 // if the process is not merged itself
1666                 // then is is a top level process
1667                 if(!p->merged && !p->target)
1668                         p->target = apps_groups_default_target;
1669
1670 #ifdef AGGREGATE_CHILDREN_TO_PARENTS
1671                 // by the way, update the diffs
1672                 // will be used later for subtracting killed process times
1673                 p->diff_cutime = p->utime - p->cutime;
1674                 p->diff_cstime = p->stime - p->cstime;
1675                 p->diff_cminflt = p->minflt - p->cminflt;
1676                 p->diff_cmajflt = p->majflt - p->cmajflt;
1677 #endif /* AGGREGATE_CHILDREN_TO_PARENTS */
1678         }
1679
1680         // give a target to all merged child processes
1681         found = 1;
1682         while(found) {
1683                 found = 0;
1684                 for(p = root_of_pids; p ; p = p->next) {
1685                         if(unlikely(!p->target && p->merged && p->parent && p->parent->target)) {
1686                                 p->target = p->parent->target;
1687                                 found++;
1688
1689                                 if(debug || (p->target && p->target->debug))
1690                                         fprintf(stderr, "apps.plugin: \t\tTARGET INHERITANCE: %s is inherited by %d (%s) from its parent %d (%s) at phase 2.\n", p->target->name, p->pid, p->comm, p->parent->pid, p->parent->comm);
1691                         }
1692                 }
1693         }
1694 }
1695
1696 long zero_all_targets(struct target *root) {
1697         struct target *w;
1698         long count = 0;
1699
1700         for (w = root; w ; w = w->next) {
1701                 count++;
1702
1703                 if(w->fds) free(w->fds);
1704                 w->fds = NULL;
1705
1706                 w->minflt = 0;
1707                 w->majflt = 0;
1708                 w->utime = 0;
1709                 w->stime = 0;
1710                 w->cminflt = 0;
1711                 w->cmajflt = 0;
1712                 w->cutime = 0;
1713                 w->cstime = 0;
1714                 w->num_threads = 0;
1715                 w->rss = 0;
1716                 w->processes = 0;
1717
1718                 w->statm_size = 0;
1719                 w->statm_resident = 0;
1720                 w->statm_share = 0;
1721                 w->statm_text = 0;
1722                 w->statm_lib = 0;
1723                 w->statm_data = 0;
1724                 w->statm_dirty = 0;
1725
1726                 w->io_logical_bytes_read = 0;
1727                 w->io_logical_bytes_written = 0;
1728                 w->io_read_calls = 0;
1729                 w->io_write_calls = 0;
1730                 w->io_storage_bytes_read = 0;
1731                 w->io_storage_bytes_written = 0;
1732                 w->io_cancelled_write_bytes = 0;
1733         }
1734
1735         return count;
1736 }
1737
1738 void aggregate_pid_on_target(struct target *w, struct pid_stat *p, struct target *o) {
1739         if(unlikely(!w->fds)) {
1740                 w->fds = calloc(sizeof(int), (size_t) all_files_size);
1741                 if(unlikely(!w->fds))
1742                         error("Cannot allocate memory for fds in %s", w->name);
1743         }
1744
1745         if(likely(p->updated)) {
1746                 w->cutime += p->cutime; // - p->fix_cutime;
1747                 w->cstime += p->cstime; // - p->fix_cstime;
1748                 w->cminflt += p->cminflt; // - p->fix_cminflt;
1749                 w->cmajflt += p->cmajflt; // - p->fix_cmajflt;
1750
1751                 w->utime += p->utime; //+ (p->pid != 1)?(p->cutime - p->fix_cutime):0;
1752                 w->stime += p->stime; //+ (p->pid != 1)?(p->cstime - p->fix_cstime):0;
1753                 w->minflt += p->minflt; //+ (p->pid != 1)?(p->cminflt - p->fix_cminflt):0;
1754                 w->majflt += p->majflt; //+ (p->pid != 1)?(p->cmajflt - p->fix_cmajflt):0;
1755
1756                 //if(p->num_threads < 0)
1757                 //      error("Negative threads number for pid '%s' (%d): %d", p->comm, p->pid, p->num_threads);
1758
1759                 //if(p->num_threads > 10000)
1760                 //      error("Excessive threads number for pid '%s' (%d): %d", p->comm, p->pid, p->num_threads);
1761
1762                 w->num_threads += p->num_threads;
1763                 w->rss += p->rss;
1764
1765                 w->statm_size += p->statm_size;
1766                 w->statm_resident += p->statm_resident;
1767                 w->statm_share += p->statm_share;
1768                 w->statm_text += p->statm_text;
1769                 w->statm_lib += p->statm_lib;
1770                 w->statm_data += p->statm_data;
1771                 w->statm_dirty += p->statm_dirty;
1772
1773                 w->io_logical_bytes_read += p->io_logical_bytes_read;
1774                 w->io_logical_bytes_written += p->io_logical_bytes_written;
1775                 w->io_read_calls += p->io_read_calls;
1776                 w->io_write_calls += p->io_write_calls;
1777                 w->io_storage_bytes_read += p->io_storage_bytes_read;
1778                 w->io_storage_bytes_written += p->io_storage_bytes_written;
1779                 w->io_cancelled_write_bytes += p->io_cancelled_write_bytes;
1780
1781                 w->processes++;
1782
1783                 if(likely(w->fds)) {
1784                         int c;
1785                         for(c = 0; c < p->fds_size ;c++) {
1786                                 if(p->fds[c] == 0) continue;
1787
1788                                 if(likely(p->fds[c] < all_files_size)) {
1789                                         if(w->fds) w->fds[p->fds[c]]++;
1790                                 }
1791                                 else
1792                                         error("Invalid fd number %d", p->fds[c]);
1793                         }
1794                 }
1795
1796                 if(unlikely(debug || w->debug))
1797                         fprintf(stderr, "apps.plugin: \tAgregating %s pid %d on %s utime=%llu, stime=%llu, cutime=%llu, cstime=%llu, minflt=%llu, majflt=%llu, cminflt=%llu, cmajflt=%llu\n", p->comm, p->pid, w->name, p->utime, p->stime, p->cutime, p->cstime, p->minflt, p->majflt, p->cminflt, p->cmajflt);
1798
1799 /*              if(p->utime - p->old_utime > 100) fprintf(stderr, "BIG CHANGE: %d %s utime increased by %llu from %llu to %llu\n", p->pid, p->comm, p->utime - p->old_utime, p->old_utime, p->utime);
1800                 if(p->cutime - p->old_cutime > 100) fprintf(stderr, "BIG CHANGE: %d %s cutime increased by %llu from %llu to %llu\n", p->pid, p->comm, p->cutime - p->old_cutime, p->old_cutime, p->cutime);
1801                 if(p->stime - p->old_stime > 100) fprintf(stderr, "BIG CHANGE: %d %s stime increased by %llu from %llu to %llu\n", p->pid, p->comm, p->stime - p->old_stime, p->old_stime, p->stime);
1802                 if(p->cstime - p->old_cstime > 100) fprintf(stderr, "BIG CHANGE: %d %s cstime increased by %llu from %llu to %llu\n", p->pid, p->comm, p->cstime - p->old_cstime, p->old_cstime, p->cstime);
1803                 if(p->minflt - p->old_minflt > 5000) fprintf(stderr, "BIG CHANGE: %d %s minflt increased by %llu from %llu to %llu\n", p->pid, p->comm, p->minflt - p->old_minflt, p->old_minflt, p->minflt);
1804                 if(p->majflt - p->old_majflt > 5000) fprintf(stderr, "BIG CHANGE: %d %s majflt increased by %llu from %llu to %llu\n", p->pid, p->comm, p->majflt - p->old_majflt, p->old_majflt, p->majflt);
1805                 if(p->cminflt - p->old_cminflt > 15000) fprintf(stderr, "BIG CHANGE: %d %s cminflt increased by %llu from %llu to %llu\n", p->pid, p->comm, p->cminflt - p->old_cminflt, p->old_cminflt, p->cminflt);
1806                 if(p->cmajflt - p->old_cmajflt > 15000) fprintf(stderr, "BIG CHANGE: %d %s cmajflt increased by %llu from %llu to %llu\n", p->pid, p->comm, p->cmajflt - p->old_cmajflt, p->old_cmajflt, p->cmajflt);
1807 */
1808 #ifdef AGGREGATE_CHILDREN_TO_PARENTS
1809                 p->old_utime = p->utime;
1810                 p->old_cutime = p->cutime;
1811                 p->old_stime = p->stime;
1812                 p->old_cstime = p->cstime;
1813                 p->old_minflt = p->minflt;
1814                 p->old_majflt = p->majflt;
1815                 p->old_cminflt = p->cminflt;
1816                 p->old_cmajflt = p->cmajflt;
1817 #endif /* AGGREGATE_CHILDREN_TO_PARENTS */
1818
1819                 if(o) {
1820                         // since the process switched target
1821                         // for all incremental values
1822                         // we have to subtract its OLD values from the new target
1823                         // and add its OLD values to the old target
1824
1825                         // IMPORTANT
1826                         // We add/subtract the last/OLD values we added to the target
1827
1828                         w->fix_cutime -= p->last_cutime;
1829                         w->fix_cstime -= p->last_cstime;
1830                         w->fix_cminflt -= p->last_cminflt;
1831                         w->fix_cmajflt -= p->last_cmajflt;
1832
1833                         w->fix_utime -= p->last_utime;
1834                         w->fix_stime -= p->last_stime;
1835                         w->fix_minflt -= p->last_minflt;
1836                         w->fix_majflt -= p->last_majflt;
1837
1838
1839                         w->fix_io_logical_bytes_read -= p->last_io_logical_bytes_read;
1840                         w->fix_io_logical_bytes_written -= p->last_io_logical_bytes_written;
1841                         w->fix_io_read_calls -= p->last_io_read_calls;
1842                         w->fix_io_write_calls -= p->last_io_write_calls;
1843                         w->fix_io_storage_bytes_read -= p->last_io_storage_bytes_read;
1844                         w->fix_io_storage_bytes_written -= p->last_io_storage_bytes_written;
1845                         w->fix_io_cancelled_write_bytes -= p->last_io_cancelled_write_bytes;
1846
1847                         // ---
1848
1849                         o->fix_cutime += p->last_cutime;
1850                         o->fix_cstime += p->last_cstime;
1851                         o->fix_cminflt += p->last_cminflt;
1852                         o->fix_cmajflt += p->last_cmajflt;
1853
1854                         o->fix_utime += p->last_utime;
1855                         o->fix_stime += p->last_stime;
1856                         o->fix_minflt += p->last_minflt;
1857                         o->fix_majflt += p->last_majflt;
1858
1859                         o->fix_io_logical_bytes_read += p->last_io_logical_bytes_read;
1860                         o->fix_io_logical_bytes_written += p->last_io_logical_bytes_written;
1861                         o->fix_io_read_calls += p->last_io_read_calls;
1862                         o->fix_io_write_calls += p->last_io_write_calls;
1863                         o->fix_io_storage_bytes_read += p->last_io_storage_bytes_read;
1864                         o->fix_io_storage_bytes_written += p->last_io_storage_bytes_written;
1865                         o->fix_io_cancelled_write_bytes += p->last_io_cancelled_write_bytes;
1866                 }
1867         }
1868         else {
1869                 // if(o) fprintf(stderr, "apps.plugin: \t\tpid %d (%s) is not updated by OLD target %s (%s) is present.\n", p->pid, p->comm, o->id, o->name);
1870
1871                 // since the process has exited, the user
1872                 // will see a drop in our charts, because the incremental
1873                 // values of this process will not be there
1874
1875                 // add them to the fix_* values and they will be added to
1876                 // the reported values, so that the report goes steady
1877                 w->fix_minflt += p->minflt;
1878                 w->fix_majflt += p->majflt;
1879                 w->fix_utime += p->utime;
1880                 w->fix_stime += p->stime;
1881                 w->fix_cminflt += p->cminflt;
1882                 w->fix_cmajflt += p->cmajflt;
1883                 w->fix_cutime += p->cutime;
1884                 w->fix_cstime += p->cstime;
1885
1886                 w->fix_io_logical_bytes_read += p->io_logical_bytes_read;
1887                 w->fix_io_logical_bytes_written += p->io_logical_bytes_written;
1888                 w->fix_io_read_calls += p->io_read_calls;
1889                 w->fix_io_write_calls += p->io_write_calls;
1890                 w->fix_io_storage_bytes_read += p->io_storage_bytes_read;
1891                 w->fix_io_storage_bytes_written += p->io_storage_bytes_written;
1892                 w->fix_io_cancelled_write_bytes += p->io_cancelled_write_bytes;
1893         }
1894
1895 }
1896
1897 void count_targets_fds(struct target *root) {
1898         int c;
1899         struct target *w;
1900
1901         for (w = root; w ; w = w->next) {
1902                 if(!w->fds) continue;
1903
1904                 w->openfiles = 0;
1905                 w->openpipes = 0;
1906                 w->opensockets = 0;
1907                 w->openinotifies = 0;
1908                 w->openeventfds = 0;
1909                 w->opentimerfds = 0;
1910                 w->opensignalfds = 0;
1911                 w->openeventpolls = 0;
1912                 w->openother = 0;
1913
1914                 for(c = 1; c < all_files_size ;c++) {
1915                         if(w->fds[c] > 0)
1916                                 switch(all_files[c].type) {
1917                                 case FILETYPE_FILE:
1918                                         w->openfiles++;
1919                                         break;
1920
1921                                 case FILETYPE_PIPE:
1922                                         w->openpipes++;
1923                                         break;
1924
1925                                 case FILETYPE_SOCKET:
1926                                         w->opensockets++;
1927                                         break;
1928
1929                                 case FILETYPE_INOTIFY:
1930                                         w->openinotifies++;
1931                                         break;
1932
1933                                 case FILETYPE_EVENTFD:
1934                                         w->openeventfds++;
1935                                         break;
1936
1937                                 case FILETYPE_TIMERFD:
1938                                         w->opentimerfds++;
1939                                         break;
1940
1941                                 case FILETYPE_SIGNALFD:
1942                                         w->opensignalfds++;
1943                                         break;
1944
1945                                 case FILETYPE_EVENTPOLL:
1946                                         w->openeventpolls++;
1947                                         break;
1948
1949                                 default:
1950                                         w->openother++;
1951                         }
1952                 }
1953
1954                 free(w->fds);
1955                 w->fds = NULL;
1956         }
1957 }
1958
1959 void calculate_netdata_statistics(void)
1960 {
1961         link_all_processes_to_their_parents();
1962         apply_apps_groups_targets_inheritance();
1963
1964 #ifdef AGGREGATE_CHILDREN_TO_PARENTS
1965         aggregate_children_to_parents();
1966 #endif /* AGGREGATE_CHILDREN_TO_PARENTS */
1967
1968         zero_all_targets(users_root_target);
1969         zero_all_targets(groups_root_target);
1970         apps_groups_targets = zero_all_targets(apps_groups_root_target);
1971
1972 #ifdef AGGREGATE_CHILDREN_TO_PARENTS
1973         if(debug)
1974                 debug_childrens_aggregations(0, 1);
1975 #endif /* AGGREGATE_CHILDREN_TO_PARENTS */
1976
1977         // this has to be done, before the cleanup
1978         struct pid_stat *p = NULL;
1979         struct target *w = NULL, *o = NULL;
1980
1981         // concentrate everything on the apps_groups_targets
1982         for(p = root_of_pids; p ; p = p->next) {
1983
1984                 // --------------------------------------------------------------------
1985                 // apps_groups targets
1986                 if(likely(p->target))
1987                         aggregate_pid_on_target(p->target, p, NULL);
1988                 else
1989                         error("pid %d %s was left without a target!", p->pid, p->comm);
1990
1991
1992                 // --------------------------------------------------------------------
1993                 // user targets
1994                 o = p->user_target;
1995                 if(likely(p->user_target && p->user_target->uid == p->uid))
1996                         w = p->user_target;
1997                 else {
1998                         if(unlikely(debug && p->user_target))
1999                                         fprintf(stderr, "apps.plugin: \t\tpid %d (%s) switched user from %d (%s) to %d.\n", p->pid, p->comm, p->user_target->uid, p->user_target->name, p->uid);
2000
2001                         w = p->user_target = get_users_target(p->uid);
2002                 }
2003
2004                 if(likely(w))
2005                         aggregate_pid_on_target(w, p, o);
2006                 else
2007                         error("pid %d %s was left without a user target!", p->pid, p->comm);
2008
2009
2010                 // --------------------------------------------------------------------
2011                 // group targets
2012                 o = p->group_target;
2013                 if(likely(p->group_target && p->group_target->gid == p->gid))
2014                         w = p->group_target;
2015                 else {
2016                         if(unlikely(debug && p->group_target))
2017                                         fprintf(stderr, "apps.plugin: \t\tpid %d (%s) switched group from %d (%s) to %d.\n", p->pid, p->comm, p->group_target->gid, p->group_target->name, p->gid);
2018
2019                         w = p->group_target = get_groups_target(p->gid);
2020                 }
2021
2022                 if(likely(w))
2023                         aggregate_pid_on_target(w, p, o);
2024                 else
2025                         error("pid %d %s was left without a group target!", p->pid, p->comm);
2026
2027         }
2028
2029         count_targets_fds(apps_groups_root_target);
2030         count_targets_fds(users_root_target);
2031         count_targets_fds(groups_root_target);
2032
2033         cleanup_non_existing_pids();
2034 }
2035
2036 // ----------------------------------------------------------------------------
2037 // update chart dimensions
2038
2039 unsigned long long send_resource_usage_to_netdata() {
2040         static struct timeval last = { 0, 0 };
2041         static struct rusage me_last;
2042
2043         struct timeval now;
2044         struct rusage me;
2045
2046         unsigned long long usec;
2047         unsigned long long cpuuser;
2048         unsigned long long cpusyst;
2049
2050         if(!last.tv_sec) {
2051                 gettimeofday(&last, NULL);
2052                 getrusage(RUSAGE_SELF, &me_last);
2053
2054                 // the first time, give a zero to allow
2055                 // netdata calibrate to the current time
2056                 // usec = update_every * 1000000ULL;
2057                 usec = 0ULL;
2058                 cpuuser = 0;
2059                 cpusyst = 0;
2060         }
2061         else {
2062                 gettimeofday(&now, NULL);
2063                 getrusage(RUSAGE_SELF, &me);
2064
2065                 usec = usecdiff(&now, &last);
2066                 cpuuser = me.ru_utime.tv_sec * 1000000ULL + me.ru_utime.tv_usec;
2067                 cpusyst = me.ru_stime.tv_sec * 1000000ULL + me.ru_stime.tv_usec;
2068
2069                 bcopy(&now, &last, sizeof(struct timeval));
2070                 bcopy(&me, &me_last, sizeof(struct rusage));
2071         }
2072
2073         fprintf(stdout, "BEGIN netdata.apps_cpu %llu\n", usec);
2074         fprintf(stdout, "SET user = %llu\n", cpuuser);
2075         fprintf(stdout, "SET system = %llu\n", cpusyst);
2076         fprintf(stdout, "END\n");
2077
2078         fprintf(stdout, "BEGIN netdata.apps_files %llu\n", usec);
2079         fprintf(stdout, "SET files = %llu\n", file_counter);
2080         fprintf(stdout, "SET pids = %ld\n", all_pids_count);
2081         fprintf(stdout, "SET fds = %d\n", all_files_len);
2082         fprintf(stdout, "SET targets = %ld\n", apps_groups_targets);
2083         fprintf(stdout, "END\n");
2084
2085         return usec;
2086 }
2087
2088 void send_collected_data_to_netdata(struct target *root, const char *type, unsigned long long usec)
2089 {
2090         struct target *w;
2091
2092         fprintf(stdout, "BEGIN %s.cpu %llu\n", type, usec);
2093         for (w = root; w ; w = w->next) {
2094                 if(w->target || (!w->processes && !w->exposed)) continue;
2095
2096                 fprintf(stdout, "SET %s = %llu\n", w->name, w->utime + w->stime + w->fix_utime + w->fix_stime);
2097         }
2098         fprintf(stdout, "END\n");
2099
2100         fprintf(stdout, "BEGIN %s.cpu_user %llu\n", type, usec);
2101         for (w = root; w ; w = w->next) {
2102                 if(w->target || (!w->processes && !w->exposed)) continue;
2103
2104                 fprintf(stdout, "SET %s = %llu\n", w->name, w->utime + w->fix_utime);
2105         }
2106         fprintf(stdout, "END\n");
2107
2108         fprintf(stdout, "BEGIN %s.cpu_system %llu\n", type, usec);
2109         for (w = root; w ; w = w->next) {
2110                 if(w->target || (!w->processes && !w->exposed)) continue;
2111
2112                 fprintf(stdout, "SET %s = %llu\n", w->name, w->stime + w->fix_stime);
2113         }
2114         fprintf(stdout, "END\n");
2115
2116         fprintf(stdout, "BEGIN %s.threads %llu\n", type, usec);
2117         for (w = root; w ; w = w->next) {
2118                 if(w->target || (!w->processes && !w->exposed)) continue;
2119
2120                 fprintf(stdout, "SET %s = %llu\n", w->name, w->num_threads);
2121         }
2122         fprintf(stdout, "END\n");
2123
2124         fprintf(stdout, "BEGIN %s.processes %llu\n", type, usec);
2125         for (w = root; w ; w = w->next) {
2126                 if(w->target || (!w->processes && !w->exposed)) continue;
2127
2128                 fprintf(stdout, "SET %s = %lu\n", w->name, w->processes);
2129         }
2130         fprintf(stdout, "END\n");
2131
2132         fprintf(stdout, "BEGIN %s.mem %llu\n", type, usec);
2133         for (w = root; w ; w = w->next) {
2134                 if(w->target || (!w->processes && !w->exposed)) continue;
2135
2136                 fprintf(stdout, "SET %s = %lld\n", w->name, (long long)w->statm_resident - (long long)w->statm_share);
2137         }
2138         fprintf(stdout, "END\n");
2139
2140         fprintf(stdout, "BEGIN %s.minor_faults %llu\n", type, usec);
2141         for (w = root; w ; w = w->next) {
2142                 if(w->target || (!w->processes && !w->exposed)) continue;
2143
2144                 fprintf(stdout, "SET %s = %llu\n", w->name, w->minflt + w->fix_minflt);
2145         }
2146         fprintf(stdout, "END\n");
2147
2148         fprintf(stdout, "BEGIN %s.major_faults %llu\n", type, usec);
2149         for (w = root; w ; w = w->next) {
2150                 if(w->target || (!w->processes && !w->exposed)) continue;
2151
2152                 fprintf(stdout, "SET %s = %llu\n", w->name, w->majflt + w->fix_majflt);
2153         }
2154         fprintf(stdout, "END\n");
2155
2156         fprintf(stdout, "BEGIN %s.lreads %llu\n", type, usec);
2157         for (w = root; w ; w = w->next) {
2158                 if(w->target || (!w->processes && !w->exposed)) continue;
2159
2160                 fprintf(stdout, "SET %s = %llu\n", w->name, w->io_logical_bytes_read + w->fix_io_logical_bytes_read);
2161         }
2162         fprintf(stdout, "END\n");
2163
2164         fprintf(stdout, "BEGIN %s.lwrites %llu\n", type, usec);
2165         for (w = root; w ; w = w->next) {
2166                 if(w->target || (!w->processes && !w->exposed)) continue;
2167
2168                 fprintf(stdout, "SET %s = %llu\n", w->name, w->io_logical_bytes_written + w->fix_io_logical_bytes_written);
2169         }
2170         fprintf(stdout, "END\n");
2171
2172         fprintf(stdout, "BEGIN %s.preads %llu\n", type, usec);
2173         for (w = root; w ; w = w->next) {
2174                 if(w->target || (!w->processes && !w->exposed)) continue;
2175
2176                 fprintf(stdout, "SET %s = %llu\n", w->name, w->io_storage_bytes_read + w->fix_io_storage_bytes_read);
2177         }
2178         fprintf(stdout, "END\n");
2179
2180         fprintf(stdout, "BEGIN %s.pwrites %llu\n", type, usec);
2181         for (w = root; w ; w = w->next) {
2182                 if(w->target || (!w->processes && !w->exposed)) continue;
2183
2184                 fprintf(stdout, "SET %s = %llu\n", w->name, w->io_storage_bytes_written + w->fix_io_storage_bytes_written);
2185         }
2186         fprintf(stdout, "END\n");
2187
2188         fprintf(stdout, "BEGIN %s.files %llu\n", type, usec);
2189         for (w = root; w ; w = w->next) {
2190                 if(w->target || (!w->processes && !w->exposed)) continue;
2191
2192                 fprintf(stdout, "SET %s = %llu\n", w->name, w->openfiles);
2193         }
2194         fprintf(stdout, "END\n");
2195
2196         fprintf(stdout, "BEGIN %s.sockets %llu\n", type, usec);
2197         for (w = root; w ; w = w->next) {
2198                 if(w->target || (!w->processes && !w->exposed)) continue;
2199
2200                 fprintf(stdout, "SET %s = %llu\n", w->name, w->opensockets);
2201         }
2202         fprintf(stdout, "END\n");
2203
2204         fprintf(stdout, "BEGIN %s.pipes %llu\n", type, usec);
2205         for (w = root; w ; w = w->next) {
2206                 if(w->target || (!w->processes && !w->exposed)) continue;
2207
2208                 fprintf(stdout, "SET %s = %llu\n", w->name, w->openpipes);
2209         }
2210         fprintf(stdout, "END\n");
2211
2212         fflush(stdout);
2213 }
2214
2215
2216 // ----------------------------------------------------------------------------
2217 // generate the charts
2218
2219 void send_charts_updates_to_netdata(struct target *root, const char *type, const char *title)
2220 {
2221         struct target *w;
2222         int newly_added = 0;
2223
2224         for(w = root ; w ; w = w->next)
2225                 if(!w->exposed && w->processes) {
2226                         newly_added++;
2227                         w->exposed = 1;
2228                         if(debug || w->debug) fprintf(stderr, "apps.plugin: %s just added - regenerating charts.\n", w->name);
2229                 }
2230
2231         // nothing more to show
2232         if(!newly_added) return;
2233
2234         // we have something new to show
2235         // update the charts
2236         fprintf(stdout, "CHART %s.cpu '' '%s CPU Time (%ld%% = %ld core%s)' 'cpu time %%' cpu %s.cpu stacked 20001 %d\n", type, title, (processors * 100), processors, (processors>1)?"s":"", type, update_every);
2237         for (w = root; w ; w = w->next) {
2238                 if(w->target || (!w->processes && !w->exposed)) continue;
2239
2240                 fprintf(stdout, "DIMENSION %s '' incremental 100 %u %s\n", w->name, hz, w->hidden ? "hidden,noreset" : "noreset");
2241         }
2242
2243         fprintf(stdout, "CHART %s.mem '' '%s Dedicated Memory (w/o shared)' 'MB' mem %s.mem stacked 20003 %d\n", type, title, type, update_every);
2244         for (w = root; w ; w = w->next) {
2245                 if(w->target || (!w->processes && !w->exposed)) continue;
2246
2247                 fprintf(stdout, "DIMENSION %s '' absolute %ld %ld noreset\n", w->name, sysconf(_SC_PAGESIZE), 1024L*1024L);
2248         }
2249
2250         fprintf(stdout, "CHART %s.threads '' '%s Threads' 'threads' processes %s.threads stacked 20005 %d\n", type, title, type, update_every);
2251         for (w = root; w ; w = w->next) {
2252                 if(w->target || (!w->processes && !w->exposed)) continue;
2253
2254                 fprintf(stdout, "DIMENSION %s '' absolute 1 1 noreset\n", w->name);
2255         }
2256
2257         fprintf(stdout, "CHART %s.processes '' '%s Processes' 'processes' processes %s.processes stacked 20004 %d\n", type, title, type, update_every);
2258         for (w = root; w ; w = w->next) {
2259                 if(w->target || (!w->processes && !w->exposed)) continue;
2260
2261                 fprintf(stdout, "DIMENSION %s '' absolute 1 1 noreset\n", w->name);
2262         }
2263
2264         fprintf(stdout, "CHART %s.cpu_user '' '%s CPU User Time (%ld%% = %ld core%s)' 'cpu time %%' cpu %s.cpu_user stacked 20020 %d\n", type, title, (processors * 100), processors, (processors>1)?"s":"", type, update_every);
2265         for (w = root; w ; w = w->next) {
2266                 if(w->target || (!w->processes && !w->exposed)) continue;
2267
2268                 fprintf(stdout, "DIMENSION %s '' incremental 100 %ld noreset\n", w->name, hz * processors);
2269         }
2270
2271         fprintf(stdout, "CHART %s.cpu_system '' '%s CPU System Time (%ld%% = %ld core%s)' 'cpu time %%' cpu %s.cpu_system stacked 20021 %d\n", type, title, (processors * 100), processors, (processors>1)?"s":"", type, update_every);
2272         for (w = root; w ; w = w->next) {
2273                 if(w->target || (!w->processes && !w->exposed)) continue;
2274
2275                 fprintf(stdout, "DIMENSION %s '' incremental 100 %ld noreset\n", w->name, hz * processors);
2276         }
2277
2278         fprintf(stdout, "CHART %s.major_faults '' '%s Major Page Faults (swap read)' 'page faults/s' swap %s.major_faults stacked 20010 %d\n", type, title, type, update_every);
2279         for (w = root; w ; w = w->next) {
2280                 if(w->target || (!w->processes && !w->exposed)) continue;
2281
2282                 fprintf(stdout, "DIMENSION %s '' incremental 1 1 noreset\n", w->name);
2283         }
2284
2285         fprintf(stdout, "CHART %s.minor_faults '' '%s Minor Page Faults' 'page faults/s' mem %s.minor_faults stacked 20011 %d\n", type, title, type, update_every);
2286         for (w = root; w ; w = w->next) {
2287                 if(w->target || (!w->processes && !w->exposed)) continue;
2288
2289                 fprintf(stdout, "DIMENSION %s '' incremental 1 1 noreset\n", w->name);
2290         }
2291
2292         fprintf(stdout, "CHART %s.lreads '' '%s Disk Logical Reads' 'kilobytes/s' disk %s.lreads stacked 20042 %d\n", type, title, type, update_every);
2293         for (w = root; w ; w = w->next) {
2294                 if(w->target || (!w->processes && !w->exposed)) continue;
2295
2296                 fprintf(stdout, "DIMENSION %s '' incremental 1 %d noreset\n", w->name, 1024);
2297         }
2298
2299         fprintf(stdout, "CHART %s.lwrites '' '%s I/O Logical Writes' 'kilobytes/s' disk %s.lwrites stacked 20042 %d\n", type, title, type, update_every);
2300         for (w = root; w ; w = w->next) {
2301                 if(w->target || (!w->processes && !w->exposed)) continue;
2302
2303                 fprintf(stdout, "DIMENSION %s '' incremental 1 %d noreset\n", w->name, 1024);
2304         }
2305
2306         fprintf(stdout, "CHART %s.preads '' '%s Disk Reads' 'kilobytes/s' disk %s.preads stacked 20002 %d\n", type, title, type, update_every);
2307         for (w = root; w ; w = w->next) {
2308                 if(w->target || (!w->processes && !w->exposed)) continue;
2309
2310                 fprintf(stdout, "DIMENSION %s '' incremental 1 %d noreset\n", w->name, 1024);
2311         }
2312
2313         fprintf(stdout, "CHART %s.pwrites '' '%s Disk Writes' 'kilobytes/s' disk %s.pwrites stacked 20002 %d\n", type, title, type, update_every);
2314         for (w = root; w ; w = w->next) {
2315                 if(w->target || (!w->processes && !w->exposed)) continue;
2316
2317                 fprintf(stdout, "DIMENSION %s '' incremental 1 %d noreset\n", w->name, 1024);
2318         }
2319
2320         fprintf(stdout, "CHART %s.files '' '%s Open Files' 'open files' disk %s.files stacked 20050 %d\n", type, title, type, update_every);
2321         for (w = root; w ; w = w->next) {
2322                 if(w->target || (!w->processes && !w->exposed)) continue;
2323
2324                 fprintf(stdout, "DIMENSION %s '' absolute 1 1 noreset\n", w->name);
2325         }
2326
2327         fprintf(stdout, "CHART %s.sockets '' '%s Open Sockets' 'open sockets' net %s.sockets stacked 20051 %d\n", type, title, type, update_every);
2328         for (w = root; w ; w = w->next) {
2329                 if(w->target || (!w->processes && !w->exposed)) continue;
2330
2331                 fprintf(stdout, "DIMENSION %s '' absolute 1 1 noreset\n", w->name);
2332         }
2333
2334         fprintf(stdout, "CHART %s.pipes '' '%s Pipes' 'open pipes' processes %s.pipes stacked 20053 %d\n", type, title, type, update_every);
2335         for (w = root; w ; w = w->next) {
2336                 if(w->target || (!w->processes && !w->exposed)) continue;
2337
2338                 fprintf(stdout, "DIMENSION %s '' absolute 1 1 noreset\n", w->name);
2339         }
2340 }
2341
2342
2343 // ----------------------------------------------------------------------------
2344 // parse command line arguments
2345
2346 void parse_args(int argc, char **argv)
2347 {
2348         int i, freq = 0;
2349         char *name = NULL;
2350
2351         for(i = 1; i < argc; i++) {
2352                 if(!freq) {
2353                         int n = atoi(argv[i]);
2354                         if(n > 0) {
2355                                 freq = n;
2356                                 continue;
2357                         }
2358                 }
2359
2360                 if(strcmp("debug", argv[i]) == 0) {
2361                         debug = 1;
2362                         debug_flags = 0xffffffff;
2363                         continue;
2364                 }
2365
2366                 if(!name) {
2367                         name = argv[i];
2368                         continue;
2369                 }
2370
2371                 error("Cannot understand option %s", argv[i]);
2372                 exit(1);
2373         }
2374
2375         if(freq > 0) update_every = freq;
2376         if(!name) name = "groups";
2377
2378         if(read_apps_groups_conf(name)) {
2379                 error("Cannot read process groups %s", name);
2380                 exit(1);
2381         }
2382 }
2383
2384 int main(int argc, char **argv)
2385 {
2386         // debug_flags = D_PROCFILE;
2387
2388         // set the name for logging
2389         program_name = "apps.plugin";
2390
2391         // disable syslog for apps.plugin
2392         error_log_syslog = 0;
2393
2394         // set errors flood protection to 100 logs per hour
2395         error_log_errors_per_period = 100;
2396         error_log_throttle_period = 3600;
2397
2398         host_prefix = getenv("NETDATA_HOST_PREFIX");
2399         if(host_prefix == NULL) {
2400                 info("NETDATA_HOST_PREFIX is not passed from netdata");
2401                 host_prefix = "";
2402         }
2403         else info("Found NETDATA_HOST_PREFIX='%s'", host_prefix);
2404
2405         config_dir = getenv("NETDATA_CONFIG_DIR");
2406         if(config_dir == NULL) {
2407                 info("NETDATA_CONFIG_DIR is not passed from netdata");
2408                 config_dir = CONFIG_DIR;
2409         }
2410         else info("Found NETDATA_CONFIG_DIR='%s'", config_dir);
2411
2412 #ifdef NETDATA_INTERNAL_CHECKS
2413         if(debug_flags != 0) {
2414                 struct rlimit rl = { RLIM_INFINITY, RLIM_INFINITY };
2415                 if(setrlimit(RLIMIT_CORE, &rl) != 0)
2416                         info("Cannot request unlimited core dumps for debugging... Proceeding anyway...");
2417                 prctl(PR_SET_DUMPABLE, 1, 0, 0, 0);
2418         }
2419 #endif /* NETDATA_INTERNAL_CHECKS */
2420
2421         info("starting...");
2422
2423         procfile_adaptive_initial_allocation = 1;
2424
2425         time_t started_t = time(NULL);
2426         time_t current_t;
2427         get_HZ();
2428         pid_max = get_system_pid_max();
2429         processors = get_system_cpus();
2430
2431         parse_args(argc, argv);
2432
2433         all_pids = calloc(sizeof(struct pid_stat *), (size_t) pid_max);
2434         if(!all_pids) {
2435                 error("Cannot allocate %lu bytes of memory.", sizeof(struct pid_stat *) * pid_max);
2436                 printf("DISABLE\n");
2437                 exit(1);
2438         }
2439
2440         fprintf(stdout, "CHART netdata.apps_cpu '' 'Apps Plugin CPU' 'milliseconds/s' apps.plugin netdata.apps_cpu stacked 140000 %1$d\n"
2441                         "DIMENSION user '' incremental 1 1000\n"
2442                         "DIMENSION system '' incremental 1 1000\n"
2443                         "CHART netdata.apps_files '' 'Apps Plugin Files' 'files/s' apps.plugin netdata.apps_files line 140001 %1$d\n"
2444                         "DIMENSION files '' incremental 1 1\n"
2445                   "DIMENSION pids '' absolute 1 1\n"
2446                         "DIMENSION fds '' absolute 1 1\n"
2447                         "DIMENSION targets '' absolute 1 1\n", update_every);
2448
2449 #ifndef PROFILING_MODE
2450         unsigned long long sunext = (time(NULL) - (time(NULL) % update_every) + update_every) * 1000000ULL;
2451         unsigned long long sunow;
2452 #endif /* PROFILING_MODE */
2453
2454         unsigned long long counter = 1;
2455         for(;1; counter++) {
2456 #ifndef PROFILING_MODE
2457                 // delay until it is our time to run
2458                 while((sunow = timems()) < sunext)
2459                         usleep((useconds_t)(sunext - sunow));
2460
2461                 // find the next time we need to run
2462                 while(timems() > sunext)
2463                         sunext += update_every * 1000000ULL;
2464 #endif /* PROFILING_MODE */
2465
2466                 if(!collect_data_for_all_processes_from_proc()) {
2467                         error("Cannot collect /proc data for running processes. Disabling apps.plugin...");
2468                         printf("DISABLE\n");
2469                         exit(1);
2470                 }
2471
2472                 calculate_netdata_statistics();
2473
2474                 unsigned long long dt = send_resource_usage_to_netdata();
2475
2476                 // this is smart enough to show only newly added apps, when needed
2477                 send_charts_updates_to_netdata(apps_groups_root_target, "apps", "Apps");
2478                 send_charts_updates_to_netdata(users_root_target, "users", "Users");
2479                 send_charts_updates_to_netdata(groups_root_target, "groups", "User Groups");
2480
2481                 send_collected_data_to_netdata(apps_groups_root_target, "apps", dt);
2482                 send_collected_data_to_netdata(users_root_target, "users", dt);
2483                 send_collected_data_to_netdata(groups_root_target, "groups", dt);
2484
2485                 if(debug) fprintf(stderr, "apps.plugin: done Loop No %llu\n", counter);
2486
2487                 current_t = time(NULL);
2488
2489 #ifndef PROFILING_MODE
2490                 // restart check (14400 seconds)
2491                 if(current_t - started_t > 14400) exit(0);
2492 #else
2493                 if(current_t - started_t > 10) exit(0);
2494 #endif /* PROFILING_MODE */
2495         }
2496 }