src/apps_plugin.c

   1 // TODO
   2 //
   3 // 1. support -ch option to set global /proc and /sys prefix
   4 // 2. disable RESET_OR_OVERFLOW check in charts
   5
   6 #ifdef HAVE_CONFIG_H
   7 #include <config.h>
   8 #endif
   9 #include <stdio.h>
  10 #include <stdlib.h>
  11 #include <string.h>
  12 #include <time.h>
  13 #include <unistd.h>
  14 #include <sys/types.h>
  15 #include <sys/time.h>
  16 #include <sys/wait.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 "common.h"
  32 #include "log.h"
  33 #include "avl.h"
  34 #include "procfile.h"
  35
  36 #define MAX_COMPARE_NAME 15
  37 #define MAX_NAME 100
  38
  39 unsigned long long Hertz = 1;
  40
  41 long processors = 1;
  42 long pid_max = 32768;
  43 int debug = 0;
  44
  45 int update_every = 1;
  46 unsigned long long file_counter = 0;
  47
  48 #define PROC_BUFFER 4096
  49
  50 // ----------------------------------------------------------------------------
  51 // memory debugger
  52
  53 struct allocations {
  54         size_t allocations;
  55         size_t allocated;
  56         size_t allocated_max;
  57 } allocations = { 0, 0, 0 };
  58
  59 #define MALLOC_MARK (uint32_t)(0x0BADCAFE)
  60 #define MALLOC_PREFIX (sizeof(uint32_t) * 2)
  61 #define MALLOC_SUFFIX (sizeof(uint32_t))
  62 #define MALLOC_OVERHEAD (MALLOC_PREFIX + MALLOC_SUFFIX)
  63
  64 void *mark_allocation(void *allocated_ptr, size_t size_without_overheads) {
  65         uint32_t *real_ptr = (uint32_t *)allocated_ptr;
  66         real_ptr[0] = MALLOC_MARK;
  67         real_ptr[1] = size_without_overheads;
  68
  69         uint32_t *end_ptr = (uint32_t *)(allocated_ptr + MALLOC_PREFIX + size_without_overheads);
  70         end_ptr[0] = MALLOC_MARK;
  71
  72         // fprintf(stderr, "MEMORY_POINTER: Allocated at %p, returning %p.\n", allocated_ptr, (void *)(allocated_ptr + MALLOC_PREFIX));
  73
  74         return (void *)(allocated_ptr + MALLOC_PREFIX);
  75 }
  76
  77 void *check_allocation(const char *file, int line, const char *function, void *marked_ptr, size_t *size_without_overheads_ptr) {
  78         uint32_t *real_ptr = (uint32_t *)(marked_ptr - MALLOC_PREFIX);
  79
  80         // 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);
  81
  82         if(real_ptr[0] != MALLOC_MARK) fatal("MEMORY: prefix MARK is not valid for %s/%u@%s.", function, line, file);
  83
  84         size_t size = real_ptr[1];
  85
  86         uint32_t *end_ptr = (uint32_t *)(marked_ptr + size);
  87         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);
  88
  89         if(size_without_overheads_ptr) *size_without_overheads_ptr = size;
  90
  91         return real_ptr;
  92 }
  93
  94 void *malloc_debug(const char *file, int line, const char *function, size_t size) {
  95         void *ptr = malloc(size + MALLOC_OVERHEAD);
  96         if(!ptr) fatal("MEMORY: Cannot allocate %zu bytes for %s/%u@%s.", size, function, line, file);
  97
  98         allocations.allocated += size;
  99         allocations.allocations++;
 100
 101         debug(D_MEMORY, "MEMORY: Allocated %zu bytes for %s/%u@%s."
 102                 " Status: allocated %zu in %zu allocs."
 103                 , size
 104                 , function, line, file
 105                 , allocations.allocated
 106                 , allocations.allocations
 107         );
 108
 109         if(allocations.allocated > allocations.allocated_max) {
 110                 debug(D_MEMORY, "MEMORY: total allocation peak increased from %zu to %zu", allocations.allocated_max, allocations.allocated);
 111                 allocations.allocated_max = allocations.allocated;
 112         }
 113
 114         size_t csize;
 115         check_allocation(file, line, function, mark_allocation(ptr, size), &csize);
 116         if(size != csize) {
 117                 fatal("Invalid size.");
 118         }
 119
 120         return mark_allocation(ptr, size);
 121 }
 122
 123 void *calloc_debug(const char *file, int line, const char *function, size_t nmemb, size_t size) {
 124         void *ptr = malloc_debug(file, line, function, (nmemb * size));
 125         bzero(ptr, nmemb * size);
 126         return ptr;
 127 }
 128
 129 void free_debug(const char *file, int line, const char *function, void *ptr) {
 130         size_t size;
 131         void *real_ptr = check_allocation(file, line, function, ptr, &size);
 132
 133         bzero(real_ptr, size + MALLOC_OVERHEAD);
 134
 135         free(real_ptr);
 136         allocations.allocated -= size;
 137         allocations.allocations--;
 138
 139         debug(D_MEMORY, "MEMORY: freed %zu bytes for %s/%u@%s."
 140                 " Status: allocated %zu in %zu allocs."
 141                 , size
 142                 , function, line, file
 143                 , allocations.allocated
 144                 , allocations.allocations
 145         );
 146 }
 147
 148 void *realloc_debug(const char *file, int line, const char *function, void *ptr, size_t size) {
 149         if(!ptr) return malloc_debug(file, line, function, size);
 150         if(!size) { free_debug(file, line, function, ptr); return NULL; }
 151
 152         size_t old_size;
 153         void *real_ptr = check_allocation(file, line, function, ptr, &old_size);
 154
 155         void *new_ptr = realloc(real_ptr, size + MALLOC_OVERHEAD);
 156         if(!new_ptr) fatal("MEMORY: Cannot allocate %zu bytes for %s/%u@%s.", size, function, line, file);
 157
 158         allocations.allocated += size;
 159         allocations.allocated -= old_size;
 160
 161         debug(D_MEMORY, "MEMORY: Re-allocated from %zu to %zu bytes for %s/%u@%s."
 162                 " Status: allocated %z in %zu allocs."
 163                 , old_size, size
 164                 , function, line, file
 165                 , allocations.allocated
 166                 , allocations.allocations
 167         );
 168
 169         if(allocations.allocated > allocations.allocated_max) {
 170                 debug(D_MEMORY, "MEMORY: total allocation peak increased from %zu to %zu", allocations.allocated_max, allocations.allocated);
 171                 allocations.allocated_max = allocations.allocated;
 172         }
 173
 174         return mark_allocation(new_ptr, size);
 175 }
 176
 177 char *strdup_debug(const char *file, int line, const char *function, const char *ptr) {
 178         size_t size = 0;
 179         const char *s = ptr;
 180
 181         while(*s++) size++;
 182         size++;
 183
 184         char *p = malloc_debug(file, line, function, size);
 185         if(!p) fatal("Cannot allocate %zu bytes.", size);
 186
 187         memcpy(p, ptr, size);
 188         return p;
 189 }
 190
 191 #define malloc(size) malloc_debug(__FILE__, __LINE__, __FUNCTION__, (size))
 192 #define calloc(nmemb, size) calloc_debug(__FILE__, __LINE__, __FUNCTION__, (nmemb), (size))
 193 #define realloc(ptr, size) realloc_debug(__FILE__, __LINE__, __FUNCTION__, (ptr), (size))
 194 #define free(ptr) free_debug(__FILE__, __LINE__, __FUNCTION__, (ptr))
 195
 196 #ifdef strdup
 197 #undef strdup
 198 #endif
 199 #define strdup(ptr) strdup_debug(__FILE__, __LINE__, __FUNCTION__, (ptr))
 200
 201 // ----------------------------------------------------------------------------
 202 // helper functions
 203
 204 procfile *ff = NULL;
 205
 206 long get_processors(void) {
 207         int processors = 0;
 208
 209         ff = procfile_reopen(ff, "/proc/stat", "", PROCFILE_FLAG_DEFAULT);
 210         if(!ff) return 1;
 211
 212         ff = procfile_readall(ff);
 213         if(!ff) {
 214                 // procfile_close(ff);
 215                 return 1;
 216         }
 217
 218         unsigned int i;
 219         for(i = 0; i < procfile_lines(ff); i++) {
 220                 if(!procfile_linewords(ff, i)) continue;
 221
 222                 if(strncmp(procfile_lineword(ff, i, 0), "cpu", 3) == 0) processors++;
 223         }
 224         processors--;
 225         if(processors < 1) processors = 1;
 226
 227         // procfile_close(ff);
 228         return processors;
 229 }
 230
 231 long get_pid_max(void) {
 232         long mpid = 32768;
 233
 234         ff = procfile_reopen(ff, "/proc/sys/kernel/pid_max", "", PROCFILE_FLAG_DEFAULT);
 235         if(!ff) return mpid;
 236
 237         ff = procfile_readall(ff);
 238         if(!ff) {
 239                 // procfile_close(ff);
 240                 return mpid;
 241         }
 242
 243         mpid = atol(procfile_lineword(ff, 0, 0));
 244         if(!mpid) mpid = 32768;
 245
 246         // procfile_close(ff);
 247         return mpid;
 248 }
 249
 250 unsigned long long get_hertz(void)
 251 {
 252         unsigned long long hz = 1;
 253
 254 #ifdef _SC_CLK_TCK
 255         if((hz = sysconf(_SC_CLK_TCK)) > 0) {
 256                 return hz;
 257         }
 258 #endif
 259
 260 #ifdef HZ
 261         hz = (unsigned long long)HZ;    /* <asm/param.h> */
 262 #else
 263         /* If 32-bit or big-endian (not Alpha or ia64), assume HZ is 100. */
 264         hz = (sizeof(long)==sizeof(int) || htons(999)==999) ? 100UL : 1024UL;
 265 #endif
 266
 267         error("Unknown HZ value. Assuming %llu.", hz);
 268         return hz;
 269 }
 270
 271
 272 // ----------------------------------------------------------------------------
 273 // target
 274 // target is the point to aggregate a process tree values
 275
 276 struct target {
 277         char compare[MAX_COMPARE_NAME + 1];
 278         char id[MAX_NAME + 1];
 279         char name[MAX_NAME + 1];
 280
 281         unsigned long long minflt;
 282         unsigned long long cminflt;
 283         unsigned long long majflt;
 284         unsigned long long cmajflt;
 285         unsigned long long utime;
 286         unsigned long long stime;
 287         unsigned long long cutime;
 288         unsigned long long cstime;
 289         unsigned long long num_threads;
 290         unsigned long long rss;
 291
 292         unsigned long long fix_minflt;
 293         unsigned long long fix_cminflt;
 294         unsigned long long fix_majflt;
 295         unsigned long long fix_cmajflt;
 296         unsigned long long fix_utime;
 297         unsigned long long fix_stime;
 298         unsigned long long fix_cutime;
 299         unsigned long long fix_cstime;
 300
 301         unsigned long long statm_size;
 302         unsigned long long statm_resident;
 303         unsigned long long statm_share;
 304         unsigned long long statm_text;
 305         unsigned long long statm_lib;
 306         unsigned long long statm_data;
 307         unsigned long long statm_dirty;
 308
 309         unsigned long long io_logical_bytes_read;
 310         unsigned long long io_logical_bytes_written;
 311         unsigned long long io_read_calls;
 312         unsigned long long io_write_calls;
 313         unsigned long long io_storage_bytes_read;
 314         unsigned long long io_storage_bytes_written;
 315         unsigned long long io_cancelled_write_bytes;
 316
 317         unsigned long long fix_io_logical_bytes_read;
 318         unsigned long long fix_io_logical_bytes_written;
 319         unsigned long long fix_io_read_calls;
 320         unsigned long long fix_io_write_calls;
 321         unsigned long long fix_io_storage_bytes_read;
 322         unsigned long long fix_io_storage_bytes_written;
 323         unsigned long long fix_io_cancelled_write_bytes;
 324
 325         int *fds;
 326         unsigned long long openfiles;
 327         unsigned long long openpipes;
 328         unsigned long long opensockets;
 329         unsigned long long openinotifies;
 330         unsigned long long openeventfds;
 331         unsigned long long opentimerfds;
 332         unsigned long long opensignalfds;
 333         unsigned long long openeventpolls;
 334         unsigned long long openother;
 335
 336         unsigned long processes;        // how many processes have been merged to this
 337         int exposed;                            // if set, we have sent this to netdata
 338         int hidden;                                     // if set, we set the hidden flag on the dimension
 339         int debug;
 340
 341         struct target *target;  // the one that will be reported to netdata
 342         struct target *next;
 343 } *target_root = NULL, *default_target = NULL;
 344
 345 long targets = 0;
 346
 347 // find or create a target
 348 // there are targets that are just agregated to other target (the second argument)
 349 struct target *get_target(const char *id, struct target *target)
 350 {
 351         const char *nid = id;
 352         if(nid[0] == '-') nid++;
 353
 354         struct target *w;
 355         for(w = target_root ; w ; w = w->next)
 356                 if(strncmp(nid, w->id, MAX_NAME) == 0) return w;
 357
 358         w = calloc(sizeof(struct target), 1);
 359         if(!w) {
 360                 error("Cannot allocate %lu bytes of memory", (unsigned long)sizeof(struct target));
 361                 return NULL;
 362         }
 363
 364         strncpy(w->id, nid, MAX_NAME);
 365         strncpy(w->name, nid, MAX_NAME);
 366         strncpy(w->compare, nid, MAX_COMPARE_NAME);
 367         if(id[0] == '-') w->hidden = 1;
 368
 369         w->target = target;
 370
 371         w->next = target_root;
 372         target_root = w;
 373
 374         if(debug) fprintf(stderr, "apps.plugin: adding hook for process '%s', compare '%s' on target '%s'\n", w->id, w->compare, w->target?w->target->id:"");
 375
 376         return w;
 377 }
 378
 379 // read the process groups file
 380 int read_process_groups(const char *name)
 381 {
 382         char buffer[4096+1];
 383         char filename[FILENAME_MAX + 1];
 384
 385         snprintf(filename, FILENAME_MAX, "%s/apps_%s.conf", CONFIG_DIR, name);
 386
 387         if(debug) fprintf(stderr, "apps.plugin: process groups file: '%s'\n", filename);
 388         FILE *fp = fopen(filename, "r");
 389         if(!fp) {
 390                 error("Cannot open file '%s' (%s)", filename, strerror(errno));
 391                 return 1;
 392         }
 393
 394         long line = 0;
 395         while(fgets(buffer, 4096, fp) != NULL) {
 396                 int whidden = 0, wdebug = 0;
 397                 line++;
 398
 399                 // if(debug) fprintf(stderr, "apps.plugin: \tread %s\n", buffer);
 400
 401                 char *s = buffer, *t, *p;
 402                 s = trim(s);
 403                 if(!s || !*s || *s == '#') continue;
 404
 405                 if(debug) fprintf(stderr, "apps.plugin: \tread %s\n", s);
 406
 407                 // the target name
 408                 t = strsep(&s, ":");
 409                 if(t) t = trim(t);
 410                 if(!t || !*t) continue;
 411
 412                 while(t[0]) {
 413                         int stop = 1;
 414
 415                         switch(t[0]) {
 416                                 case '-':
 417                                         stop = 0;
 418                                         whidden = 1;
 419                                         t++;
 420                                         break;
 421
 422                                 case '+':
 423                                         stop = 0;
 424                                         wdebug = 1;
 425                                         t++;
 426                                         break;
 427                         }
 428
 429                         if(stop) break;
 430                 }
 431
 432                 if(debug) fprintf(stderr, "apps.plugin: \t\ttarget %s\n", t);
 433
 434                 struct target *w = NULL;
 435                 long count = 0;
 436
 437                 // the process names
 438                 while((p = strsep(&s, " "))) {
 439                         p = trim(p);
 440                         if(!p || !*p) continue;
 441
 442                         struct target *n = get_target(p, w);
 443                         n->hidden = whidden;
 444                         n->debug = wdebug;
 445                         if(!w) w = n;
 446
 447                         count++;
 448                 }
 449
 450                 if(w) strncpy(w->name, t, MAX_NAME);
 451                 if(!count) error("The line %ld on file '%s', for group '%s' does not state any process names.", line, filename, t);
 452         }
 453         fclose(fp);
 454
 455         default_target = get_target("+p!o@w#e$i^r&7*5(-i)l-o_", NULL); // match nothing
 456         strncpy(default_target->name, "other", MAX_NAME);
 457
 458         return 0;
 459 }
 460
 461
 462 // ----------------------------------------------------------------------------
 463 // data to store for each pid
 464 // see: man proc
 465
 466 struct pid_stat {
 467         int32_t pid;
 468         char comm[MAX_COMPARE_NAME + 1];
 469         // char state;
 470         int32_t ppid;
 471         // int32_t pgrp;
 472         // int32_t session;
 473         // int32_t tty_nr;
 474         // int32_t tpgid;
 475         // uint64_t flags;
 476         unsigned long long minflt;
 477         unsigned long long cminflt;
 478         unsigned long long majflt;
 479         unsigned long long cmajflt;
 480         unsigned long long utime;
 481         unsigned long long stime;
 482         unsigned long long cutime;
 483         unsigned long long cstime;
 484         // int64_t priority;
 485         // int64_t nice;
 486         int32_t num_threads;
 487         // int64_t itrealvalue;
 488         // unsigned long long starttime;
 489         // unsigned long long vsize;
 490         unsigned long long rss;
 491         // unsigned long long rsslim;
 492         // unsigned long long starcode;
 493         // unsigned long long endcode;
 494         // unsigned long long startstack;
 495         // unsigned long long kstkesp;
 496         // unsigned long long kstkeip;
 497         // uint64_t signal;
 498         // uint64_t blocked;
 499         // uint64_t sigignore;
 500         // uint64_t sigcatch;
 501         // uint64_t wchan;
 502         // uint64_t nswap;
 503         // uint64_t cnswap;
 504         // int32_t exit_signal;
 505         // int32_t processor;
 506         // uint32_t rt_priority;
 507         // uint32_t policy;
 508         // unsigned long long delayacct_blkio_ticks;
 509         // uint64_t guest_time;
 510         // int64_t cguest_time;
 511
 512         unsigned long long statm_size;
 513         unsigned long long statm_resident;
 514         unsigned long long statm_share;
 515         unsigned long long statm_text;
 516         unsigned long long statm_lib;
 517         unsigned long long statm_data;
 518         unsigned long long statm_dirty;
 519
 520         unsigned long long io_logical_bytes_read;
 521         unsigned long long io_logical_bytes_written;
 522         unsigned long long io_read_calls;
 523         unsigned long long io_write_calls;
 524         unsigned long long io_storage_bytes_read;
 525         unsigned long long io_storage_bytes_written;
 526         unsigned long long io_cancelled_write_bytes;
 527
 528 #ifdef INCLUDE_CHILDS
 529         unsigned long long old_utime;
 530         unsigned long long old_stime;
 531         unsigned long long old_minflt;
 532         unsigned long long old_majflt;
 533
 534         unsigned long long old_cutime;
 535         unsigned long long old_cstime;
 536         unsigned long long old_cminflt;
 537         unsigned long long old_cmajflt;
 538
 539         unsigned long long fix_cutime;
 540         unsigned long long fix_cstime;
 541         unsigned long long fix_cminflt;
 542         unsigned long long fix_cmajflt;
 543
 544         unsigned long long diff_cutime;
 545         unsigned long long diff_cstime;
 546         unsigned long long diff_cminflt;
 547         unsigned long long diff_cmajflt;
 548 #endif
 549
 550         int *fds;                                       // array of fds it uses
 551         int fds_size;                           // the size of the fds array
 552
 553         int childs;                                     // number of processes directly referencing this
 554         int updated;                            // 1 when update
 555         int merged;                                     // 1 when it has been merged to its parent
 556         int new_entry;
 557         struct target *target;
 558         struct pid_stat *parent;
 559         struct pid_stat *prev;
 560         struct pid_stat *next;
 561 } *root_of_pids = NULL, **all_pids;
 562
 563 long all_pids_count = 0;
 564
 565 struct pid_stat *get_pid_entry(pid_t pid)
 566 {
 567         if(all_pids[pid]) {
 568                 all_pids[pid]->new_entry = 0;
 569                 return all_pids[pid];
 570         }
 571
 572         all_pids[pid] = calloc(sizeof(struct pid_stat), 1);
 573         if(!all_pids[pid]) {
 574                 error("Cannot allocate %lu bytes of memory", (unsigned long)sizeof(struct pid_stat));
 575                 return NULL;
 576         }
 577
 578         all_pids[pid]->fds = calloc(sizeof(int), 100);
 579         if(!all_pids[pid]->fds)
 580                 error("Cannot allocate %ld bytes of memory", (unsigned long)(sizeof(int) * 100));
 581         else all_pids[pid]->fds_size = 100;
 582
 583         if(root_of_pids) root_of_pids->prev = all_pids[pid];
 584         all_pids[pid]->next = root_of_pids;
 585         root_of_pids = all_pids[pid];
 586
 587         all_pids[pid]->pid = pid;
 588         all_pids[pid]->new_entry = 1;
 589
 590         return all_pids[pid];
 591 }
 592
 593 void del_pid_entry(pid_t pid)
 594 {
 595         if(!all_pids[pid]) return;
 596
 597         if(debug) fprintf(stderr, "apps.plugin: process %d %s exited, deleting it.\n", pid, all_pids[pid]->comm);
 598
 599         if(root_of_pids == all_pids[pid]) root_of_pids = all_pids[pid]->next;
 600         if(all_pids[pid]->next) all_pids[pid]->next->prev = all_pids[pid]->prev;
 601         if(all_pids[pid]->prev) all_pids[pid]->prev->next = all_pids[pid]->next;
 602
 603         if(all_pids[pid]->fds) free(all_pids[pid]->fds);
 604         free(all_pids[pid]);
 605         all_pids[pid] = NULL;
 606 }
 607
 608
 609 // ----------------------------------------------------------------------------
 610 // update pids from proc
 611
 612 int read_proc_pid_stat(struct pid_stat *p) {
 613         char filename[FILENAME_MAX + 1];
 614
 615         snprintf(filename, FILENAME_MAX, "/proc/%d/stat", p->pid);
 616
 617         ff = procfile_reopen(ff, filename, NULL, PROCFILE_FLAG_NO_ERROR_ON_FILE_IO);
 618         if(!ff) return 1;
 619
 620         ff = procfile_readall(ff);
 621         if(!ff) {
 622                 // procfile_close(ff);
 623                 return 1;
 624         }
 625
 626         file_counter++;
 627
 628         p->comm[0] = '\0';
 629         p->comm[MAX_COMPARE_NAME] = '\0';
 630         size_t blen = 0;
 631
 632         char *s = procfile_lineword(ff, 0, 1);
 633         if(*s == '(') s++;
 634         size_t len = strlen(s);
 635         unsigned int i = 0;
 636         while(len && s[len - 1] != ')') {
 637                 if(blen < MAX_COMPARE_NAME) {
 638                         strncpy(&p->comm[blen], s, MAX_COMPARE_NAME - blen);
 639                         blen = strlen(p->comm);
 640                 }
 641
 642                 i++;
 643                 s = procfile_lineword(ff, 0, 1+i);
 644                 len = strlen(s);
 645         }
 646         if(len && s[len - 1] == ')') s[len - 1] = '\0';
 647         if(blen < MAX_COMPARE_NAME)
 648                 strncpy(&p->comm[blen], s, MAX_COMPARE_NAME - blen);
 649
 650         // p->pid                       = atol(procfile_lineword(ff, 0, 0+i));
 651         // comm is at 1
 652         // p->state                     = *(procfile_lineword(ff, 0, 2+i));
 653         p->ppid                         = atol(procfile_lineword(ff, 0, 3+i));
 654         // p->pgrp                      = atol(procfile_lineword(ff, 0, 4+i));
 655         // p->session           = atol(procfile_lineword(ff, 0, 5+i));
 656         // p->tty_nr            = atol(procfile_lineword(ff, 0, 6+i));
 657         // p->tpgid                     = atol(procfile_lineword(ff, 0, 7+i));
 658         // p->flags                     = strtoull(procfile_lineword(ff, 0, 8+i), NULL, 10);
 659         p->minflt                       = strtoull(procfile_lineword(ff, 0, 9+i), NULL, 10);
 660         p->cminflt                      = strtoull(procfile_lineword(ff, 0, 10+i), NULL, 10);
 661         p->majflt                       = strtoull(procfile_lineword(ff, 0, 11+i), NULL, 10);
 662         p->cmajflt                      = strtoull(procfile_lineword(ff, 0, 12+i), NULL, 10);
 663         p->utime                        = strtoull(procfile_lineword(ff, 0, 13+i), NULL, 10);
 664         p->stime                        = strtoull(procfile_lineword(ff, 0, 14+i), NULL, 10);
 665         p->cutime                       = strtoull(procfile_lineword(ff, 0, 15+i), NULL, 10);
 666         p->cstime                       = strtoull(procfile_lineword(ff, 0, 16+i), NULL, 10);
 667         // p->priority          = strtoull(procfile_lineword(ff, 0, 17+i), NULL, 10);
 668         // p->nice                      = strtoull(procfile_lineword(ff, 0, 18+i), NULL, 10);
 669         p->num_threads          = atol(procfile_lineword(ff, 0, 19+i));
 670         // p->itrealvalue       = strtoull(procfile_lineword(ff, 0, 20+i), NULL, 10);
 671         // p->starttime         = strtoull(procfile_lineword(ff, 0, 21+i), NULL, 10);
 672         // p->vsize                     = strtoull(procfile_lineword(ff, 0, 22+i), NULL, 10);
 673         p->rss                          = strtoull(procfile_lineword(ff, 0, 23+i), NULL, 10);
 674         // p->rsslim            = strtoull(procfile_lineword(ff, 0, 24+i), NULL, 10);
 675         // p->starcode          = strtoull(procfile_lineword(ff, 0, 25+i), NULL, 10);
 676         // p->endcode           = strtoull(procfile_lineword(ff, 0, 26+i), NULL, 10);
 677         // p->startstack        = strtoull(procfile_lineword(ff, 0, 27+i), NULL, 10);
 678         // p->kstkesp           = strtoull(procfile_lineword(ff, 0, 28+i), NULL, 10);
 679         // p->kstkeip           = strtoull(procfile_lineword(ff, 0, 29+i), NULL, 10);
 680         // p->signal            = strtoull(procfile_lineword(ff, 0, 30+i), NULL, 10);
 681         // p->blocked           = strtoull(procfile_lineword(ff, 0, 31+i), NULL, 10);
 682         // p->sigignore         = strtoull(procfile_lineword(ff, 0, 32+i), NULL, 10);
 683         // p->sigcatch          = strtoull(procfile_lineword(ff, 0, 33+i), NULL, 10);
 684         // p->wchan                     = strtoull(procfile_lineword(ff, 0, 34+i), NULL, 10);
 685         // p->nswap                     = strtoull(procfile_lineword(ff, 0, 35+i), NULL, 10);
 686         // p->cnswap            = strtoull(procfile_lineword(ff, 0, 36+i), NULL, 10);
 687         // p->exit_signal       = atol(procfile_lineword(ff, 0, 37+i));
 688         // p->processor         = atol(procfile_lineword(ff, 0, 38+i));
 689         // p->rt_priority       = strtoul(procfile_lineword(ff, 0, 39+i), NULL, 10);
 690         // p->policy            = strtoul(procfile_lineword(ff, 0, 40+i), NULL, 10);
 691         // p->delayacct_blkio_ticks             = strtoull(procfile_lineword(ff, 0, 41+i), NULL, 10);
 692         // p->guest_time        = strtoull(procfile_lineword(ff, 0, 42+i), NULL, 10);
 693         // p->cguest_time       = strtoull(procfile_lineword(ff, 0, 43), NULL, 10);
 694
 695         if(debug || (p->target && p->target->debug)) fprintf(stderr, "apps.plugin: VALUES: %s utime=%llu, stime=%llu, cutime=%llu, cstime=%llu, minflt=%llu, majflt=%llu, cminflt=%llu, cmajflt=%llu, threads=%d\n", p->comm, p->utime, p->stime, p->cutime, p->cstime, p->minflt, p->majflt, p->cminflt, p->cmajflt, p->num_threads);
 696
 697         // procfile_close(ff);
 698         return 0;
 699 }
 700
 701 int read_proc_pid_statm(struct pid_stat *p) {
 702         char filename[FILENAME_MAX + 1];
 703
 704         snprintf(filename, FILENAME_MAX, "/proc/%d/statm", p->pid);
 705
 706         ff = procfile_reopen(ff, filename, NULL, PROCFILE_FLAG_NO_ERROR_ON_FILE_IO);
 707         if(!ff) return 1;
 708
 709         ff = procfile_readall(ff);
 710         if(!ff) {
 711                 // procfile_close(ff);
 712                 return 1;
 713         }
 714
 715         file_counter++;
 716
 717         p->statm_size                   = strtoull(procfile_lineword(ff, 0, 0), NULL, 10);
 718         p->statm_resident               = strtoull(procfile_lineword(ff, 0, 1), NULL, 10);
 719         p->statm_share                  = strtoull(procfile_lineword(ff, 0, 2), NULL, 10);
 720         p->statm_text                   = strtoull(procfile_lineword(ff, 0, 3), NULL, 10);
 721         p->statm_lib                    = strtoull(procfile_lineword(ff, 0, 4), NULL, 10);
 722         p->statm_data                   = strtoull(procfile_lineword(ff, 0, 5), NULL, 10);
 723         p->statm_dirty                  = strtoull(procfile_lineword(ff, 0, 6), NULL, 10);
 724
 725         // procfile_close(ff);
 726         return 0;
 727 }
 728
 729 int read_proc_pid_io(struct pid_stat *p) {
 730         char filename[FILENAME_MAX + 1];
 731
 732         snprintf(filename, FILENAME_MAX, "/proc/%d/io", p->pid);
 733
 734         ff = procfile_reopen(ff, filename, NULL, PROCFILE_FLAG_NO_ERROR_ON_FILE_IO);
 735         if(!ff) return 1;
 736
 737         ff = procfile_readall(ff);
 738         if(!ff) {
 739                 // procfile_close(ff);
 740                 return 1;
 741         }
 742
 743         file_counter++;
 744
 745         p->io_logical_bytes_read                = strtoull(procfile_lineword(ff, 0, 1), NULL, 10);
 746         p->io_logical_bytes_written     = strtoull(procfile_lineword(ff, 1, 1), NULL, 10);
 747         p->io_read_calls                                = strtoull(procfile_lineword(ff, 2, 1), NULL, 10);
 748         p->io_write_calls                               = strtoull(procfile_lineword(ff, 3, 1), NULL, 10);
 749         p->io_storage_bytes_read                = strtoull(procfile_lineword(ff, 4, 1), NULL, 10);
 750         p->io_storage_bytes_written     = strtoull(procfile_lineword(ff, 5, 1), NULL, 10);
 751         p->io_cancelled_write_bytes             = strtoull(procfile_lineword(ff, 6, 1), NULL, 10);
 752
 753         // procfile_close(ff);
 754         return 0;
 755 }
 756
 757
 758 // ----------------------------------------------------------------------------
 759
 760 #ifdef INCLUDE_CHILDS
 761 // print a tree view of all processes
 762 int walk_down(pid_t pid, int level) {
 763         struct pid_stat *p = NULL;
 764         char b[level+3];
 765         int i, ret = 0;
 766
 767         for(i = 0; i < level; i++) b[i] = '\t';
 768         b[level] = '|';
 769         b[level+1] = '-';
 770         b[level+2] = '\0';
 771
 772         for(p = root_of_pids; p ; p = p->next) {
 773                 if(p->ppid == pid) {
 774                         ret += walk_down(p->pid, level+1);
 775                 }
 776         }
 777
 778         p = all_pids[pid];
 779         if(p) {
 780                 if(!p->updated) ret += 1;
 781                 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"
 782                         , b, p->comm, p->pid, p->updated?"OK":"KILLED", p->target->name, p->childs
 783                         , p->utime, p->utime - p->old_utime
 784                         , p->stime, p->stime - p->old_stime
 785                         , p->cutime, p->cutime - p->old_cutime
 786                         , p->cstime, p->cstime - p->old_cstime
 787                         , p->minflt, p->minflt - p->old_minflt
 788                         , p->majflt, p->majflt - p->old_majflt
 789                         , p->cminflt, p->cminflt - p->old_cminflt
 790                         , p->cmajflt, p->cmajflt - p->old_cmajflt
 791                         );
 792         }
 793
 794         return ret;
 795 }
 796 #endif
 797
 798
 799 // ----------------------------------------------------------------------------
 800 // file descriptor
 801 // this is used to keep a global list of all open files of the system
 802 // it is needed in order to figure out the unique files a process tree has open
 803
 804 #define FILE_DESCRIPTORS_INCREASE_STEP 100
 805
 806 struct file_descriptor {
 807         avl avl;
 808         uint32_t magic;
 809         uint32_t hash;
 810         const char *name;
 811         int type;
 812         long count;
 813         long pos;
 814 } *all_files = NULL;
 815
 816 int all_files_len = 0;
 817 int all_files_size = 0;
 818
 819 int file_descriptor_compare(void* a, void* b) {
 820         if(((struct file_descriptor *)a)->magic != 0x0BADCAFE || ((struct file_descriptor *)b)->magic != 0x0BADCAFE)
 821                 error("Corrupted index data detected. Please report this.");
 822
 823         if(((struct file_descriptor *)a)->hash < ((struct file_descriptor *)b)->hash)
 824                 return -1;
 825         else if(((struct file_descriptor *)a)->hash > ((struct file_descriptor *)b)->hash)
 826                 return 1;
 827         else return strcmp(((struct file_descriptor *)a)->name, ((struct file_descriptor *)b)->name);
 828 }
 829 int file_descriptor_iterator(avl *a) { if(a) {}; return 0; }
 830
 831 avl_tree all_files_index = {
 832                 NULL,
 833                 file_descriptor_compare,
 834                 PTHREAD_RWLOCK_INITIALIZER
 835 };
 836
 837 static struct file_descriptor *file_descriptor_find(const char *name, uint32_t hash) {
 838         struct file_descriptor *result = NULL, tmp;
 839         tmp.hash = (hash)?hash:simple_hash(name);
 840         tmp.name = name;
 841         tmp.count = 0;
 842         tmp.pos = 0;
 843         tmp.magic = 0x0BADCAFE;
 844
 845         avl_search(&all_files_index, (avl *)&tmp, file_descriptor_iterator, (avl **)&result);
 846         return result;
 847 }
 848
 849 #define file_descriptor_add(fd) avl_insert(&all_files_index, (avl *)(fd))
 850 #define file_descriptor_remove(fd) avl_remove(&all_files_index, (avl *)(fd))
 851
 852 #define FILETYPE_OTHER 0
 853 #define FILETYPE_FILE 1
 854 #define FILETYPE_PIPE 2
 855 #define FILETYPE_SOCKET 3
 856 #define FILETYPE_INOTIFY 4
 857 #define FILETYPE_EVENTFD 5
 858 #define FILETYPE_EVENTPOLL 6
 859 #define FILETYPE_TIMERFD 7
 860 #define FILETYPE_SIGNALFD 8
 861
 862 void file_descriptor_not_used(int id)
 863 {
 864         if(id > 0 && id < all_files_size) {
 865                 if(all_files[id].magic != 0x0BADCAFE) {
 866                         error("Ignoring request to remove empty file id %d.", id);
 867                         return;
 868                 }
 869
 870                 if(debug) fprintf(stderr, "apps.plugin: decreasing slot %d (count = %ld).\n", id, all_files[id].count);
 871
 872                 if(all_files[id].count > 0) {
 873                         all_files[id].count--;
 874
 875                         if(!all_files[id].count) {
 876                                 if(debug) fprintf(stderr, "apps.plugin:   >> slot %d is empty.\n", id);
 877                                 file_descriptor_remove(&all_files[id]);
 878                                 all_files[id].magic = 0x00000000;
 879                                 all_files_len--;
 880                         }
 881                 }
 882                 else
 883                         error("Request to decrease counter of fd %d (%s), while the use counter is 0", id, all_files[id].name);
 884         }
 885         else    error("Request to decrease counter of fd %d, which is outside the array size (1 to %d)", id, all_files_size);
 886 }
 887
 888 unsigned long file_descriptor_find_or_add(const char *name)
 889 {
 890         static int last_pos = 0;
 891         uint32_t hash = simple_hash(name);
 892
 893         if(debug) fprintf(stderr, "apps.plugin: adding or finding name '%s' with hash %u\n", name, hash);
 894
 895         struct file_descriptor *fd = file_descriptor_find(name, hash);
 896         if(fd) {
 897                 // found
 898                 if(debug) fprintf(stderr, "apps.plugin:   >> found on slot %ld\n", fd->pos);
 899                 fd->count++;
 900                 return fd->pos;
 901         }
 902         // not found
 903
 904         // check we have enough memory to add it
 905         if(!all_files || all_files_len == all_files_size) {
 906                 void *old = all_files;
 907                 int i;
 908
 909                 // there is no empty slot
 910                 if(debug) fprintf(stderr, "apps.plugin: extending fd array to %d entries\n", all_files_size + FILE_DESCRIPTORS_INCREASE_STEP);
 911                 all_files = realloc(all_files, (all_files_size + FILE_DESCRIPTORS_INCREASE_STEP) * sizeof(struct file_descriptor));
 912
 913                 // if the address changed, we have to rebuild the index
 914                 // since all pointers are now invalid
 915                 if(old && old != (void *)all_files) {
 916                         if(debug) fprintf(stderr, "apps.plugin:   >> re-indexing.\n");
 917                         all_files_index.root = NULL;
 918                         for(i = 0; i < all_files_size; i++) {
 919                                 if(!all_files[i].count) continue;
 920                                 file_descriptor_add(&all_files[i]);
 921                         }
 922                         if(debug) fprintf(stderr, "apps.plugin:   >> re-indexing done.\n");
 923                 }
 924
 925                 for(i = all_files_size; i < (all_files_size + FILE_DESCRIPTORS_INCREASE_STEP); i++) {
 926                         all_files[i].count = 0;
 927                         all_files[i].name = NULL;
 928                         all_files[i].magic = 0x00000000;
 929                         all_files[i].pos = i;
 930                 }
 931
 932                 if(!all_files_size) all_files_len = 1;
 933                 all_files_size += FILE_DESCRIPTORS_INCREASE_STEP;
 934         }
 935
 936         if(debug) fprintf(stderr, "apps.plugin:   >> searching for empty slot.\n");
 937
 938         // search for an empty slot
 939         int i, c;
 940         for(i = 0, c = last_pos ; i < all_files_size ; i++, c++) {
 941                 if(c >= all_files_size) c = 0;
 942                 if(c == 0) continue;
 943
 944                 if(!all_files[c].count) {
 945                         if(debug) fprintf(stderr, "apps.plugin:   >> Examining slot %d.\n", c);
 946
 947                         if(all_files[c].magic == 0x0BADCAFE && all_files[c].name && file_descriptor_find(all_files[c].name, all_files[c].hash))
 948                                 error("fd on position %d is not cleared properly. It still has %s in it.\n", c, all_files[c].name);
 949
 950                         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);
 951                         if(all_files[c].name) free((void *)all_files[c].name);
 952                         all_files[c].name = NULL;
 953                         last_pos = c;
 954                         break;
 955                 }
 956         }
 957         if(i == all_files_size) {
 958                 fatal("We should find an empty slot, but there isn't any");
 959                 exit(1);
 960         }
 961         if(debug) fprintf(stderr, "apps.plugin:   >> updating slot %d.\n", c);
 962
 963         all_files_len++;
 964
 965         // else we have an empty slot in 'c'
 966
 967         int type = FILETYPE_OTHER;
 968         if(name[0] == '/') type = FILETYPE_FILE;
 969         else if(strncmp(name, "pipe:", 5) == 0) type = FILETYPE_PIPE;
 970         else if(strncmp(name, "socket:", 7) == 0) type = FILETYPE_SOCKET;
 971         else if(strcmp(name, "anon_inode:inotify") == 0 || strcmp(name, "inotify") == 0) type = FILETYPE_INOTIFY;
 972         else if(strcmp(name, "anon_inode:[eventfd]") == 0) type = FILETYPE_EVENTFD;
 973         else if(strcmp(name, "anon_inode:[eventpoll]") == 0) type = FILETYPE_EVENTPOLL;
 974         else if(strcmp(name, "anon_inode:[timerfd]") == 0) type = FILETYPE_TIMERFD;
 975         else if(strcmp(name, "anon_inode:[signalfd]") == 0) type = FILETYPE_SIGNALFD;
 976         else if(strncmp(name, "anon_inode:", 11) == 0) {
 977                 if(debug) fprintf(stderr, "apps.plugin: FIXME: unknown anonymous inode: %s\n", name);
 978                 type = FILETYPE_OTHER;
 979         }
 980         else {
 981                 if(debug) fprintf(stderr, "apps.plugin: FIXME: cannot understand linkname: %s\n", name);
 982                 type = FILETYPE_OTHER;
 983         }
 984
 985         all_files[c].name = strdup(name);
 986         all_files[c].hash = hash;
 987         all_files[c].type = type;
 988         all_files[c].pos  = c;
 989         all_files[c].count = 1;
 990         all_files[c].magic = 0x0BADCAFE;
 991
 992         file_descriptor_add(&all_files[c]);
 993
 994         if(debug) fprintf(stderr, "apps.plugin: using fd position %d (name: %s)\n", c, all_files[c].name);
 995
 996         return c;
 997 }
 998
 999
1000 // 1. read all files in /proc
1001 // 2. for each numeric directory:
1002 //    i.   read /proc/pid/stat
1003 //    ii.  read /proc/pid/statm
1004 //    iii. read /proc/pid/io (requires root access)
1005 //    iii. read the entries in directory /proc/pid/fd (requires root access)
1006 //         for each entry:
1007 //         a. find or create a struct file_descriptor
1008 //         b. cleanup any old/unused file_descriptors
1009
1010 // after all these, some pids may be linked to targets, while others may not
1011
1012 // in case of errors, only 1 every 1000 errors is printed
1013 // to avoid filling up all disk space
1014 // if debug is enabled, all errors are printed
1015
1016 int update_from_proc(void)
1017 {
1018         static long count_errors = 0;
1019
1020         char filename[FILENAME_MAX+1];
1021         DIR *dir = opendir("/proc");
1022         if(!dir) return 0;
1023
1024         struct dirent *file = NULL;
1025         struct pid_stat *p = NULL;
1026
1027         // mark them all as un-updated
1028         all_pids_count = 0;
1029         for(p = root_of_pids; p ; p = p->next) {
1030                 all_pids_count++;
1031                 p->parent = NULL;
1032                 p->updated = 0;
1033                 p->childs = 0;
1034                 p->merged = 0;
1035                 p->new_entry = 0;
1036         }
1037
1038         while((file = readdir(dir))) {
1039                 char *endptr = file->d_name;
1040                 pid_t pid = strtoul(file->d_name, &endptr, 10);
1041                 if(pid <= 0 || pid > pid_max || endptr == file->d_name || *endptr != '\0') continue;
1042
1043                 p = get_pid_entry(pid);
1044                 if(!p) continue;
1045
1046                 // --------------------------------------------------------------------
1047                 // /proc/<pid>/stat
1048
1049                 if(read_proc_pid_stat(p)) {
1050                         if(!count_errors++ || debug || (p->target && p->target->debug))
1051                                 error("Cannot process /proc/%d/stat", pid);
1052
1053                         continue;
1054                 }
1055                 if(p->ppid < 0 || p->ppid > pid_max) p->ppid = 0;
1056
1057
1058                 // --------------------------------------------------------------------
1059                 // /proc/<pid>/statm
1060
1061                 if(read_proc_pid_statm(p)) {
1062                         if(!count_errors++ || debug || (p->target && p->target->debug))
1063                                 error("Cannot process /proc/%d/statm", pid);
1064
1065                         continue;
1066                 }
1067
1068
1069                 // --------------------------------------------------------------------
1070                 // /proc/<pid>/io
1071
1072                 if(read_proc_pid_io(p)) {
1073                         if(!count_errors++ || debug || (p->target && p->target->debug))
1074                                 error("Cannot process /proc/%d/io", pid);
1075
1076                         continue;
1077                 }
1078
1079                 // --------------------------------------------------------------------
1080                 // link it
1081
1082                 // check if it is target
1083                 // we do this only once, the first time this pid is loaded
1084                 if(p->new_entry) {
1085                         if(debug) fprintf(stderr, "apps.plugin: \tJust added %s\n", p->comm);
1086
1087                         struct target *w;
1088                         for(w = target_root; w ; w = w->next) {
1089                                 // if(debug || (p->target && p->target->debug)) fprintf(stderr, "apps.plugin: \t\tcomparing '%s' with '%s'\n", w->compare, p->comm);
1090
1091                                 if(strcmp(w->compare, p->comm) == 0) {
1092                                         if(w->target) p->target = w->target;
1093                                         else p->target = w;
1094
1095                                         if(debug || (p->target && p->target->debug)) fprintf(stderr, "apps.plugin: \t\t%s linked to target %s\n", p->comm, p->target->name);
1096                                 }
1097                         }
1098                 }
1099
1100                 // --------------------------------------------------------------------
1101                 // /proc/<pid>/fd
1102
1103                 snprintf(filename, FILENAME_MAX, "/proc/%s/fd", file->d_name);
1104                 DIR *fds = opendir(filename);
1105                 if(fds) {
1106                         int c;
1107                         struct dirent *de;
1108                         char fdname[FILENAME_MAX + 1];
1109                         char linkname[FILENAME_MAX + 1];
1110
1111                         // make the array negative
1112                         for(c = 0 ; c < p->fds_size ; c++) p->fds[c] = -p->fds[c];
1113
1114                         while((de = readdir(fds))) {
1115                                 if(strcmp(de->d_name, ".") == 0 || strcmp(de->d_name, "..") == 0) continue;
1116
1117                                 // check if the fds array is small
1118                                 int fdid = atol(de->d_name);
1119                                 if(fdid < 0) continue;
1120                                 if(fdid >= p->fds_size) {
1121                                         // it is small, extend it
1122                                         if(debug) fprintf(stderr, "apps.plugin: extending fd memory slots for %s from %d to %d\n", p->comm, p->fds_size, fdid + 100);
1123                                         p->fds = realloc(p->fds, (fdid + 100) * sizeof(int));
1124                                         if(!p->fds) {
1125                                                 error("Cannot re-allocate fds for %s", p->comm);
1126                                                 break;
1127                                         }
1128
1129                                         // and initialize it
1130                                         for(c = p->fds_size ; c < (fdid + 100) ; c++) p->fds[c] = 0;
1131                                         p->fds_size = fdid + 100;
1132                                 }
1133
1134                                 if(p->fds[fdid] == 0) {
1135                                         // we don't know this fd, get it
1136
1137                                         sprintf(fdname, "/proc/%s/fd/%s", file->d_name, de->d_name);
1138                                         int l = readlink(fdname, linkname, FILENAME_MAX);
1139                                         if(l == -1) {
1140                                                 if(debug || (p->target && p->target->debug)) {
1141                                                         if(!count_errors++ || debug || (p->target && p->target->debug))
1142                                                                 error("Cannot read link %s", fdname);
1143                                                 }
1144                                                 continue;
1145                                         }
1146                                         linkname[l] = '\0';
1147                                         file_counter++;
1148
1149                                         // if another process already has this, we will get
1150                                         // the same id
1151                                         p->fds[fdid] = file_descriptor_find_or_add(linkname);
1152                                 }
1153
1154                                 // else make it positive again, we need it
1155                                 // of course, the actual file may have changed, but we don't care so much
1156                                 // FIXME: we could compare the inode as returned by readdir direct structure
1157                                 else p->fds[fdid] = -p->fds[fdid];
1158                         }
1159                         closedir(fds);
1160
1161                         // remove all the negative file descriptors
1162                         for(c = 0 ; c < p->fds_size ; c++) if(p->fds[c] < 0) {
1163                                 file_descriptor_not_used(-p->fds[c]);
1164                                 p->fds[c] = 0;
1165                         }
1166                 }
1167
1168                 // --------------------------------------------------------------------
1169                 // done!
1170
1171                 // mark it as updated
1172                 p->updated = 1;
1173         }
1174         if(count_errors > 1000) {
1175                 error("%ld more errors encountered\n", count_errors - 1);
1176                 count_errors = 0;
1177         }
1178
1179         closedir(dir);
1180
1181         return 1;
1182 }
1183
1184
1185 // ----------------------------------------------------------------------------
1186 // update statistics on the targets
1187
1188 // 1. link all childs to their parents
1189 // 2. go from bottom to top, marking as merged all childs to their parents
1190 //    this step links all parents without a target to the child target, if any
1191 // 3. link all top level processes (the ones not merged) to the default target
1192 // 4. go from top to bottom, linking all childs without a target, to their parent target
1193 //    after this step, all processes have a target
1194 // [5. for each killed pid (updated = 0), remove its usage from its target]
1195 // 6. zero all targets
1196 // 7. concentrate all values on the targets
1197 // 8. remove all killed processes
1198 // 9. find the unique file count for each target
1199
1200 void update_statistics(void)
1201 {
1202         int c;
1203         struct pid_stat *p = NULL;
1204
1205         // link all parents and update childs count
1206         for(p = root_of_pids; p ; p = p->next) {
1207                 if(p->ppid > 0 && p->ppid <= pid_max && all_pids[p->ppid]) {
1208                         if(debug || (p->target && p->target->debug)) fprintf(stderr, "apps.plugin: \tparent of %d %s is %d %s\n", p->pid, p->comm, p->ppid, all_pids[p->ppid]->comm);
1209
1210                         p->parent = all_pids[p->ppid];
1211                         p->parent->childs++;
1212                 }
1213                 else if(p->ppid != 0) error("pid %d %s states parent %d, but the later does not exist.", p->pid, p->comm, p->ppid);
1214         }
1215
1216         // find all the procs with 0 childs and merge them to their parents
1217         // repeat, until nothing more can be done.
1218         int found = 1;
1219         while(found) {
1220                 found = 0;
1221                 for(p = root_of_pids; p ; p = p->next) {
1222                         // if this process does not have any childs, and
1223                         // is not already merged, and
1224                         // its parent has childs waiting to be merged, and
1225                         // the target of this process and its parent is the same, or the parent does not have a target, or this process does not have a parent
1226                         // and its parent is not init
1227                         // then... merge them!
1228                         if(!p->childs && !p->merged && p->parent && p->parent->childs && (p->target == p->parent->target || !p->parent->target || !p->target) && p->ppid != 1) {
1229                                 p->parent->childs--;
1230                                 p->merged = 1;
1231
1232                                 // the parent inherits the child's target, if it does not have a target itself
1233                                 if(p->target && !p->parent->target) {
1234                                         p->parent->target = p->target;
1235                                         if(debug || (p->target && p->target->debug)) fprintf(stderr, "apps.plugin: \t\ttarget %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);
1236                                 }
1237
1238                                 found++;
1239                         }
1240                 }
1241                 if(debug) fprintf(stderr, "apps.plugin: merged %d processes\n", found);
1242         }
1243
1244         // give a default target on all top level processes
1245         // init goes always to default target
1246         if(all_pids[1]) all_pids[1]->target = default_target;
1247
1248         for(p = root_of_pids; p ; p = p->next) {
1249                 // if the process is not merged itself
1250                 // then is is a top level process
1251                 if(!p->merged && !p->target) p->target = default_target;
1252
1253 #ifdef INCLUDE_CHILDS
1254                 // by the way, update the diffs
1255                 // will be used later for substracting killed process times
1256                 p->diff_cutime = p->utime - p->cutime;
1257                 p->diff_cstime = p->stime - p->cstime;
1258                 p->diff_cminflt = p->minflt - p->cminflt;
1259                 p->diff_cmajflt = p->majflt - p->cmajflt;
1260 #endif
1261         }
1262
1263         // give a target to all merged child processes
1264         found = 1;
1265         while(found) {
1266                 found = 0;
1267                 for(p = root_of_pids; p ; p = p->next) {
1268                         if(!p->target && p->merged && p->parent && p->parent->target) {
1269                                 p->target = p->parent->target;
1270                                 found++;
1271                         }
1272                 }
1273         }
1274
1275 #ifdef INCLUDE_CHILDS
1276         // for each killed process, remove its values from the parents
1277         // sums (we had already added them in a previous loop)
1278         for(p = root_of_pids; p ; p = p->next) {
1279                 if(p->updated) continue;
1280
1281                 if(debug) fprintf(stderr, "apps.plugin: UNMERGING %d %s\n", p->pid, p->comm);
1282
1283                 unsigned long long diff_utime = p->utime + p->cutime + p->fix_cutime;
1284                 unsigned long long diff_stime = p->stime + p->cstime + p->fix_cstime;
1285                 unsigned long long diff_minflt = p->minflt + p->cminflt + p->fix_cminflt;
1286                 unsigned long long diff_majflt = p->majflt + p->cmajflt + p->fix_cmajflt;
1287
1288                 struct pid_stat *t = p;
1289                 while((t = t->parent)) {
1290                         if(!t->updated) continue;
1291
1292                         unsigned long long x;
1293                         if(diff_utime && t->diff_cutime) {
1294                                 x = (t->diff_cutime < diff_utime)?t->diff_cutime:diff_utime;
1295                                 diff_utime -= x;
1296                                 t->diff_cutime -= x;
1297                                 t->fix_cutime += x;
1298                                 if(debug) fprintf(stderr, "apps.plugin: \t cutime %llu from %d %s %s\n", x, t->pid, t->comm, t->target->name);
1299                         }
1300                         if(diff_stime && t->diff_cstime) {
1301                                 x = (t->diff_cstime < diff_stime)?t->diff_cstime:diff_stime;
1302                                 diff_stime -= x;
1303                                 t->diff_cstime -= x;
1304                                 t->fix_cstime += x;
1305                                 if(debug) fprintf(stderr, "apps.plugin: \t cstime %llu from %d %s %s\n", x, t->pid, t->comm, t->target->name);
1306                         }
1307                         if(diff_minflt && t->diff_cminflt) {
1308                                 x = (t->diff_cminflt < diff_minflt)?t->diff_cminflt:diff_minflt;
1309                                 diff_minflt -= x;
1310                                 t->diff_cminflt -= x;
1311                                 t->fix_cminflt += x;
1312                                 if(debug) fprintf(stderr, "apps.plugin: \t cminflt %llu from %d %s %s\n", x, t->pid, t->comm, t->target->name);
1313                         }
1314                         if(diff_majflt && t->diff_cmajflt) {
1315                                 x = (t->diff_cmajflt < diff_majflt)?t->diff_cmajflt:diff_majflt;
1316                                 diff_majflt -= x;
1317                                 t->diff_cmajflt -= x;
1318                                 t->fix_cmajflt += x;
1319                                 if(debug) fprintf(stderr, "apps.plugin: \t cmajflt %llu from %d %s %s\n", x, t->pid, t->comm, t->target->name);
1320                         }
1321                 }
1322
1323                 if(diff_utime) error("Cannot fix up utime %llu", diff_utime);
1324                 if(diff_stime) error("Cannot fix up stime %llu", diff_stime);
1325                 if(diff_minflt) error("Cannot fix up minflt %llu", diff_minflt);
1326                 if(diff_majflt) error("Cannot fix up majflt %llu", diff_majflt);
1327         }
1328 #endif
1329
1330         // zero all the targets
1331         targets = 0;
1332         struct target *w;
1333         for (w = target_root; w ; w = w->next) {
1334                 targets++;
1335
1336                 w->fds = calloc(sizeof(int), all_files_size);
1337                 if(!w->fds)
1338                         error("Cannot allocate memory for fds in %s", w->name);
1339
1340                 w->minflt = 0;
1341                 w->majflt = 0;
1342                 w->utime = 0;
1343                 w->stime = 0;
1344                 w->cminflt = 0;
1345                 w->cmajflt = 0;
1346                 w->cutime = 0;
1347                 w->cstime = 0;
1348                 w->num_threads = 0;
1349                 w->rss = 0;
1350                 w->processes = 0;
1351
1352                 w->statm_size = 0;
1353                 w->statm_resident = 0;
1354                 w->statm_share = 0;
1355                 w->statm_text = 0;
1356                 w->statm_lib = 0;
1357                 w->statm_data = 0;
1358                 w->statm_dirty = 0;
1359
1360                 w->io_logical_bytes_read = 0;
1361                 w->io_logical_bytes_written = 0;
1362                 w->io_read_calls = 0;
1363                 w->io_write_calls = 0;
1364                 w->io_storage_bytes_read = 0;
1365                 w->io_storage_bytes_written = 0;
1366                 w->io_cancelled_write_bytes = 0;
1367         }
1368
1369 #ifdef INCLUDE_CHILDS
1370         if(debug) walk_down(0, 1);
1371 #endif
1372
1373         // concentrate everything on the targets
1374         for(p = root_of_pids; p ; p = p->next) {
1375                 if(!p->target) {
1376                         error("pid %d %s was left without a target!", p->pid, p->comm);
1377                         continue;
1378                 }
1379
1380                 if(p->updated) {
1381                         p->target->cutime += p->cutime; // - p->fix_cutime;
1382                         p->target->cstime += p->cstime; // - p->fix_cstime;
1383                         p->target->cminflt += p->cminflt; // - p->fix_cminflt;
1384                         p->target->cmajflt += p->cmajflt; // - p->fix_cmajflt;
1385
1386                         p->target->utime += p->utime; //+ (p->pid != 1)?(p->cutime - p->fix_cutime):0;
1387                         p->target->stime += p->stime; //+ (p->pid != 1)?(p->cstime - p->fix_cstime):0;
1388                         p->target->minflt += p->minflt; //+ (p->pid != 1)?(p->cminflt - p->fix_cminflt):0;
1389                         p->target->majflt += p->majflt; //+ (p->pid != 1)?(p->cmajflt - p->fix_cmajflt):0;
1390
1391                         //if(p->num_threads < 0)
1392                         //      error("Negative threads number for pid '%s' (%d): %d", p->comm, p->pid, p->num_threads);
1393
1394                         //if(p->num_threads > 10000)
1395                         //      error("Excessive threads number for pid '%s' (%d): %d", p->comm, p->pid, p->num_threads);
1396
1397                         p->target->num_threads += p->num_threads;
1398                         p->target->rss += p->rss;
1399
1400                         p->target->statm_size += p->statm_size;
1401                         p->target->statm_resident += p->statm_resident;
1402                         p->target->statm_share += p->statm_share;
1403                         p->target->statm_text += p->statm_text;
1404                         p->target->statm_lib += p->statm_lib;
1405                         p->target->statm_data += p->statm_data;
1406                         p->target->statm_dirty += p->statm_dirty;
1407
1408                         p->target->io_logical_bytes_read += p->io_logical_bytes_read;
1409                         p->target->io_logical_bytes_written += p->io_logical_bytes_written;
1410                         p->target->io_read_calls += p->io_read_calls;
1411                         p->target->io_write_calls += p->io_write_calls;
1412                         p->target->io_storage_bytes_read += p->io_storage_bytes_read;
1413                         p->target->io_storage_bytes_written += p->io_storage_bytes_written;
1414                         p->target->io_cancelled_write_bytes += p->io_cancelled_write_bytes;
1415
1416                         p->target->processes++;
1417
1418                         for(c = 0; c < p->fds_size ;c++) {
1419                                 if(p->fds[c] == 0) continue;
1420                                 if(p->fds[c] < all_files_size) {
1421                                         if(p->target->fds) p->target->fds[p->fds[c]]++;
1422                                 }
1423                                 else
1424                                         error("Invalid fd number %d", p->fds[c]);
1425                         }
1426
1427                         if(debug || p->target->debug) 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, p->target->name, p->utime, p->stime, p->cutime, p->cstime, p->minflt, p->majflt, p->cminflt, p->cmajflt);
1428
1429 /*                      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);
1430                         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);
1431                         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);
1432                         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);
1433                         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);
1434                         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);
1435                         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);
1436                         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);
1437 */
1438 #ifdef INCLUDE_CHILDS
1439                         p->old_utime = p->utime;
1440                         p->old_cutime = p->cutime;
1441                         p->old_stime = p->stime;
1442                         p->old_cstime = p->cstime;
1443                         p->old_minflt = p->minflt;
1444                         p->old_majflt = p->majflt;
1445                         p->old_cminflt = p->cminflt;
1446                         p->old_cmajflt = p->cmajflt;
1447 #endif
1448                 }
1449                 else {
1450                         // since the process has exited, the user
1451                         // will see a drop in our charts, because the incremental
1452                         // values of this process will not be there
1453
1454                         // add them to the fix_* values and they will be added to
1455                         // the reported values, so that the report goes steady
1456                         p->target->fix_minflt += p->minflt;
1457                         p->target->fix_majflt += p->majflt;
1458                         p->target->fix_utime += p->utime;
1459                         p->target->fix_stime += p->stime;
1460                         p->target->fix_cminflt += p->cminflt;
1461                         p->target->fix_cmajflt += p->cmajflt;
1462                         p->target->fix_cutime += p->cutime;
1463                         p->target->fix_cstime += p->cstime;
1464
1465                         p->target->fix_io_logical_bytes_read += p->io_logical_bytes_read;
1466                         p->target->fix_io_logical_bytes_written += p->io_logical_bytes_written;
1467                         p->target->fix_io_read_calls += p->io_read_calls;
1468                         p->target->fix_io_write_calls += p->io_write_calls;
1469                         p->target->fix_io_storage_bytes_read += p->io_storage_bytes_read;
1470                         p->target->fix_io_storage_bytes_written += p->io_storage_bytes_written;
1471                         p->target->fix_io_cancelled_write_bytes += p->io_cancelled_write_bytes;
1472                 }
1473         }
1474
1475 //      fprintf(stderr, "\n");
1476         // cleanup all un-updated processed (exited, killed, etc)
1477         for(p = root_of_pids; p ;) {
1478                 if(!p->updated) {
1479 //                      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);
1480
1481                         for(c = 0 ; c < p->fds_size ; c++) if(p->fds[c] > 0) {
1482                                 file_descriptor_not_used(p->fds[c]);
1483                                 p->fds[c] = 0;
1484                         }
1485
1486                         pid_t r = p->pid;
1487                         p = p->next;
1488                         del_pid_entry(r);
1489                 }
1490                 else p = p->next;
1491         }
1492
1493         for (w = target_root; w ; w = w->next) {
1494                 w->openfiles = 0;
1495                 w->openpipes = 0;
1496                 w->opensockets = 0;
1497                 w->openinotifies = 0;
1498                 w->openeventfds = 0;
1499                 w->opentimerfds = 0;
1500                 w->opensignalfds = 0;
1501                 w->openeventpolls = 0;
1502                 w->openother = 0;
1503
1504                 for(c = 1; c < all_files_size ;c++) {
1505                         if(w->fds && w->fds[c] > 0) switch(all_files[c].type) {
1506                                 case FILETYPE_FILE:
1507                                         w->openfiles++;
1508                                         break;
1509
1510                                 case FILETYPE_PIPE:
1511                                         w->openpipes++;
1512                                         break;
1513
1514                                 case FILETYPE_SOCKET:
1515                                         w->opensockets++;
1516                                         break;
1517
1518                                 case FILETYPE_INOTIFY:
1519                                         w->openinotifies++;
1520                                         break;
1521
1522                                 case FILETYPE_EVENTFD:
1523                                         w->openeventfds++;
1524                                         break;
1525
1526                                 case FILETYPE_TIMERFD:
1527                                         w->opentimerfds++;
1528                                         break;
1529
1530                                 case FILETYPE_SIGNALFD:
1531                                         w->opensignalfds++;
1532                                         break;
1533
1534                                 case FILETYPE_EVENTPOLL:
1535                                         w->openeventpolls++;
1536                                         break;
1537
1538                                 default:
1539                                         w->openother++;
1540                         }
1541                 }
1542
1543                 free(w->fds);
1544                 w->fds = NULL;
1545         }
1546 }
1547
1548 // ----------------------------------------------------------------------------
1549 // update chart dimensions
1550
1551 void show_dimensions(void)
1552 {
1553         static struct timeval last = { 0, 0 };
1554         static struct rusage me_last;
1555
1556         struct target *w;
1557         struct timeval now;
1558         struct rusage me;
1559
1560         unsigned long long usec;
1561         unsigned long long cpuuser;
1562         unsigned long long cpusyst;
1563
1564         if(!last.tv_sec) {
1565                 gettimeofday(&last, NULL);
1566                 getrusage(RUSAGE_SELF, &me_last);
1567
1568                 // the first time, give a zero to allow
1569                 // netdata calibrate to the current time
1570                 // usec = update_every * 1000000ULL;
1571                 usec = 0ULL;
1572                 cpuuser = 0;
1573                 cpusyst = 0;
1574         }
1575         else {
1576                 gettimeofday(&now, NULL);
1577                 getrusage(RUSAGE_SELF, &me);
1578
1579                 usec = usecdiff(&now, &last);
1580                 cpuuser = me.ru_utime.tv_sec * 1000000ULL + me.ru_utime.tv_usec;
1581                 cpusyst = me.ru_stime.tv_sec * 1000000ULL + me.ru_stime.tv_usec;
1582
1583                 bcopy(&now, &last, sizeof(struct timeval));
1584                 bcopy(&me, &me_last, sizeof(struct rusage));
1585         }
1586
1587         fprintf(stdout, "BEGIN apps.cpu %llu\n", usec);
1588         for (w = target_root; w ; w = w->next) {
1589                 if(w->target || (!w->processes && !w->exposed)) continue;
1590
1591                 fprintf(stdout, "SET %s = %llu\n", w->name, w->utime + w->stime + w->fix_utime + w->fix_stime);
1592         }
1593         fprintf(stdout, "END\n");
1594
1595         fprintf(stdout, "BEGIN apps.cpu_user %llu\n", usec);
1596         for (w = target_root; w ; w = w->next) {
1597                 if(w->target || (!w->processes && !w->exposed)) continue;
1598
1599                 fprintf(stdout, "SET %s = %llu\n", w->name, w->utime + w->fix_utime);
1600         }
1601         fprintf(stdout, "END\n");
1602
1603         fprintf(stdout, "BEGIN apps.cpu_system %llu\n", usec);
1604         for (w = target_root; w ; w = w->next) {
1605                 if(w->target || (!w->processes && !w->exposed)) continue;
1606
1607                 fprintf(stdout, "SET %s = %llu\n", w->name, w->stime + w->fix_stime);
1608         }
1609         fprintf(stdout, "END\n");
1610
1611         fprintf(stdout, "BEGIN apps.threads %llu\n", usec);
1612         for (w = target_root; w ; w = w->next) {
1613                 if(w->target || (!w->processes && !w->exposed)) continue;
1614
1615                 fprintf(stdout, "SET %s = %llu\n", w->name, w->num_threads);
1616         }
1617         fprintf(stdout, "END\n");
1618
1619         fprintf(stdout, "BEGIN apps.processes %llu\n", usec);
1620         for (w = target_root; w ; w = w->next) {
1621                 if(w->target || (!w->processes && !w->exposed)) continue;
1622
1623                 fprintf(stdout, "SET %s = %lu\n", w->name, w->processes);
1624         }
1625         fprintf(stdout, "END\n");
1626
1627         fprintf(stdout, "BEGIN apps.mem %llu\n", usec);
1628         for (w = target_root; w ; w = w->next) {
1629                 if(w->target || (!w->processes && !w->exposed)) continue;
1630
1631                 fprintf(stdout, "SET %s = %lld\n", w->name, (long long)w->statm_resident - (long long)w->statm_share);
1632         }
1633         fprintf(stdout, "END\n");
1634
1635         fprintf(stdout, "BEGIN apps.minor_faults %llu\n", usec);
1636         for (w = target_root; w ; w = w->next) {
1637                 if(w->target || (!w->processes && !w->exposed)) continue;
1638
1639                 fprintf(stdout, "SET %s = %llu\n", w->name, w->minflt + w->fix_minflt);
1640         }
1641         fprintf(stdout, "END\n");
1642
1643         fprintf(stdout, "BEGIN apps.major_faults %llu\n", usec);
1644         for (w = target_root; w ; w = w->next) {
1645                 if(w->target || (!w->processes && !w->exposed)) continue;
1646
1647                 fprintf(stdout, "SET %s = %llu\n", w->name, w->majflt + w->fix_majflt);
1648         }
1649         fprintf(stdout, "END\n");
1650
1651         fprintf(stdout, "BEGIN apps.lreads %llu\n", usec);
1652         for (w = target_root; w ; w = w->next) {
1653                 if(w->target || (!w->processes && !w->exposed)) continue;
1654
1655                 fprintf(stdout, "SET %s = %llu\n", w->name, w->io_logical_bytes_read);
1656         }
1657         fprintf(stdout, "END\n");
1658
1659         fprintf(stdout, "BEGIN apps.lwrites %llu\n", usec);
1660         for (w = target_root; w ; w = w->next) {
1661                 if(w->target || (!w->processes && !w->exposed)) continue;
1662
1663                 fprintf(stdout, "SET %s = %llu\n", w->name, w->io_logical_bytes_written);
1664         }
1665         fprintf(stdout, "END\n");
1666
1667         fprintf(stdout, "BEGIN apps.preads %llu\n", usec);
1668         for (w = target_root; w ; w = w->next) {
1669                 if(w->target || (!w->processes && !w->exposed)) continue;
1670
1671                 fprintf(stdout, "SET %s = %llu\n", w->name, w->io_storage_bytes_read);
1672         }
1673         fprintf(stdout, "END\n");
1674
1675         fprintf(stdout, "BEGIN apps.pwrites %llu\n", usec);
1676         for (w = target_root; w ; w = w->next) {
1677                 if(w->target || (!w->processes && !w->exposed)) continue;
1678
1679                 fprintf(stdout, "SET %s = %llu\n", w->name, w->io_storage_bytes_written);
1680         }
1681         fprintf(stdout, "END\n");
1682
1683         fprintf(stdout, "BEGIN apps.files %llu\n", usec);
1684         for (w = target_root; w ; w = w->next) {
1685                 if(w->target || (!w->processes && !w->exposed)) continue;
1686
1687                 fprintf(stdout, "SET %s = %llu\n", w->name, w->openfiles);
1688         }
1689         fprintf(stdout, "END\n");
1690
1691         fprintf(stdout, "BEGIN apps.sockets %llu\n", usec);
1692         for (w = target_root; w ; w = w->next) {
1693                 if(w->target || (!w->processes && !w->exposed)) continue;
1694
1695                 fprintf(stdout, "SET %s = %llu\n", w->name, w->opensockets);
1696         }
1697         fprintf(stdout, "END\n");
1698
1699         fprintf(stdout, "BEGIN apps.pipes %llu\n", usec);
1700         for (w = target_root; w ; w = w->next) {
1701                 if(w->target || (!w->processes && !w->exposed)) continue;
1702
1703                 fprintf(stdout, "SET %s = %llu\n", w->name, w->openpipes);
1704         }
1705         fprintf(stdout, "END\n");
1706
1707         fprintf(stdout, "BEGIN netdata.apps_cpu %llu\n", usec);
1708         fprintf(stdout, "SET user = %llu\n", cpuuser);
1709         fprintf(stdout, "SET system = %llu\n", cpusyst);
1710         fprintf(stdout, "END\n");
1711
1712         fprintf(stdout, "BEGIN netdata.apps_files %llu\n", usec);
1713         fprintf(stdout, "SET files = %llu\n", file_counter);
1714         fprintf(stdout, "SET pids = %ld\n", all_pids_count);
1715         fprintf(stdout, "SET fds = %d\n", all_files_len);
1716         fprintf(stdout, "SET targets = %ld\n", targets);
1717         fprintf(stdout, "END\n");
1718
1719         fflush(stdout);
1720 }
1721
1722
1723 // ----------------------------------------------------------------------------
1724 // generate the charts
1725
1726 void show_charts(void)
1727 {
1728         struct target *w;
1729         int newly_added = 0;
1730
1731         for(w = target_root ; w ; w = w->next)
1732                 if(!w->exposed && w->processes) {
1733                         newly_added++;
1734                         w->exposed = 1;
1735                         if(debug || w->debug) fprintf(stderr, "apps.plugin: %s just added - regenerating charts.\n", w->name);
1736                 }
1737
1738         // nothing more to show
1739         if(!newly_added) return;
1740
1741         // we have something new to show
1742         // update the charts
1743         fprintf(stdout, "CHART apps.cpu '' 'Apps CPU Time (%ld%% = %ld core%s)' 'cpu time %%' apps apps stacked 20001 %d\n", (processors * 100), processors, (processors>1)?"s":"", update_every);
1744         for (w = target_root; w ; w = w->next) {
1745                 if(w->target || (!w->processes && !w->exposed)) continue;
1746
1747                 fprintf(stdout, "DIMENSION %s '' incremental 100 %llu %s\n", w->name, (unsigned long long)(Hertz), w->hidden?"hidden,noreset":"noreset");
1748         }
1749
1750         fprintf(stdout, "CHART apps.mem '' 'Apps Dedicated Memory (w/o shared)' 'MB' apps apps stacked 20003 %d\n", update_every);
1751         for (w = target_root; w ; w = w->next) {
1752                 if(w->target || (!w->processes && !w->exposed)) continue;
1753
1754                 fprintf(stdout, "DIMENSION %s '' absolute %ld %ld noreset\n", w->name, sysconf(_SC_PAGESIZE), 1024L*1024L);
1755         }
1756
1757         fprintf(stdout, "CHART apps.threads '' 'Apps Threads' 'threads' apps apps stacked 20005 %d\n", update_every);
1758         for (w = target_root; w ; w = w->next) {
1759                 if(w->target || (!w->processes && !w->exposed)) continue;
1760
1761                 fprintf(stdout, "DIMENSION %s '' absolute 1 1 noreset\n", w->name);
1762         }
1763
1764         fprintf(stdout, "CHART apps.processes '' 'Apps Processes' 'processes' apps apps stacked 20004 %d\n", update_every);
1765         for (w = target_root; w ; w = w->next) {
1766                 if(w->target || (!w->processes && !w->exposed)) continue;
1767
1768                 fprintf(stdout, "DIMENSION %s '' absolute 1 1 noreset\n", w->name);
1769         }
1770
1771         fprintf(stdout, "CHART apps.cpu_user '' 'Apps CPU User Time (%ld%% = %ld core%s)' 'cpu time %%' apps none stacked 20020 %d\n", (processors * 100), processors, (processors>1)?"s":"", update_every);
1772         for (w = target_root; w ; w = w->next) {
1773                 if(w->target || (!w->processes && !w->exposed)) continue;
1774
1775                 fprintf(stdout, "DIMENSION %s '' incremental 100 %llu noreset\n", w->name, Hertz * processors);
1776         }
1777
1778         fprintf(stdout, "CHART apps.cpu_system '' 'Apps CPU System Time (%ld%% = %ld core%s)' 'cpu time %%' apps none stacked 20021 %d\n", (processors * 100), processors, (processors>1)?"s":"", update_every);
1779         for (w = target_root; w ; w = w->next) {
1780                 if(w->target || (!w->processes && !w->exposed)) continue;
1781
1782                 fprintf(stdout, "DIMENSION %s '' incremental 100 %llu noreset\n", w->name, Hertz * processors);
1783         }
1784
1785         fprintf(stdout, "CHART apps.major_faults '' 'Apps Major Page Faults (swaps in)' 'page faults/s' apps apps stacked 20010 %d\n", update_every);
1786         for (w = target_root; w ; w = w->next) {
1787                 if(w->target || (!w->processes && !w->exposed)) continue;
1788
1789                 fprintf(stdout, "DIMENSION %s '' incremental 1 1 noreset\n", w->name);
1790         }
1791
1792         fprintf(stdout, "CHART apps.minor_faults '' 'Apps Minor Page Faults' 'page faults/s' apps none stacked 20011 %d\n", update_every);
1793         for (w = target_root; w ; w = w->next) {
1794                 if(w->target || (!w->processes && !w->exposed)) continue;
1795
1796                 fprintf(stdout, "DIMENSION %s '' incremental 1 1 noreset\n", w->name);
1797         }
1798
1799         fprintf(stdout, "CHART apps.lreads '' 'Apps Disk Logical Reads' 'kilobytes/s' apps none stacked 20042 %d\n", update_every);
1800         for (w = target_root; w ; w = w->next) {
1801                 if(w->target || (!w->processes && !w->exposed)) continue;
1802
1803                 fprintf(stdout, "DIMENSION %s '' incremental 1 %d noreset\n", w->name, 1024);
1804         }
1805
1806         fprintf(stdout, "CHART apps.lwrites '' 'Apps I/O Logical Writes' 'kilobytes/s' apps none stacked 20042 %d\n", update_every);
1807         for (w = target_root; w ; w = w->next) {
1808                 if(w->target || (!w->processes && !w->exposed)) continue;
1809
1810                 fprintf(stdout, "DIMENSION %s '' incremental 1 %d noreset\n", w->name, 1024);
1811         }
1812
1813         fprintf(stdout, "CHART apps.preads '' 'Apps Disk Reads' 'kilobytes/s' apps apps stacked 20002 %d\n", update_every);
1814         for (w = target_root; w ; w = w->next) {
1815                 if(w->target || (!w->processes && !w->exposed)) continue;
1816
1817                 fprintf(stdout, "DIMENSION %s '' incremental 1 %d noreset\n", w->name, 1024);
1818         }
1819
1820         fprintf(stdout, "CHART apps.pwrites '' 'Apps Disk Writes' 'kilobytes/s' apps apps stacked 20002 %d\n", update_every);
1821         for (w = target_root; w ; w = w->next) {
1822                 if(w->target || (!w->processes && !w->exposed)) continue;
1823
1824                 fprintf(stdout, "DIMENSION %s '' incremental 1 %d noreset\n", w->name, 1024);
1825         }
1826
1827         fprintf(stdout, "CHART apps.files '' 'Apps Open Files' 'open files' apps apps stacked 20050 %d\n", update_every);
1828         for (w = target_root; w ; w = w->next) {
1829                 if(w->target || (!w->processes && !w->exposed)) continue;
1830
1831                 fprintf(stdout, "DIMENSION %s '' absolute 1 1 noreset\n", w->name);
1832         }
1833
1834         fprintf(stdout, "CHART apps.sockets '' 'Apps Open Sockets' 'open sockets' apps apps stacked 20051 %d\n", update_every);
1835         for (w = target_root; w ; w = w->next) {
1836                 if(w->target || (!w->processes && !w->exposed)) continue;
1837
1838                 fprintf(stdout, "DIMENSION %s '' absolute 1 1 noreset\n", w->name);
1839         }
1840
1841         fprintf(stdout, "CHART apps.pipes '' 'Apps Pipes' 'open pipes' apps none stacked 20053 %d\n", update_every);
1842         for (w = target_root; w ; w = w->next) {
1843                 if(w->target || (!w->processes && !w->exposed)) continue;
1844
1845                 fprintf(stdout, "DIMENSION %s '' absolute 1 1 noreset\n", w->name);
1846         }
1847
1848         fprintf(stdout, "CHART netdata.apps_cpu '' 'Apps Plugin CPU' 'milliseconds/s' netdata netdata stacked 10000 %d\n", update_every);
1849         fprintf(stdout, "DIMENSION user '' incremental 1 %d\n", 1000);
1850         fprintf(stdout, "DIMENSION system '' incremental 1 %d\n", 1000);
1851
1852         fprintf(stdout, "CHART netdata.apps_files '' 'Apps Plugin Files' 'files/s' netdata netdata line 10001 %d\n", update_every);
1853         fprintf(stdout, "DIMENSION files '' incremental 1 1\n");
1854         fprintf(stdout, "DIMENSION pids '' absolute 1 1\n");
1855         fprintf(stdout, "DIMENSION fds '' absolute 1 1\n");
1856         fprintf(stdout, "DIMENSION targets '' absolute 1 1\n");
1857
1858         fflush(stdout);
1859 }
1860
1861
1862 // ----------------------------------------------------------------------------
1863 // parse command line arguments
1864
1865 void parse_args(int argc, char **argv)
1866 {
1867         int i, freq = 0;
1868         char *name = NULL;
1869
1870         for(i = 1; i < argc; i++) {
1871                 if(!freq) {
1872                         int n = atoi(argv[i]);
1873                         if(n > 0) {
1874                                 freq = n;
1875                                 continue;
1876                         }
1877                 }
1878
1879                 if(strcmp("debug", argv[i]) == 0) {
1880                         debug = 1;
1881                         debug_flags = 0xffffffff;
1882                         continue;
1883                 }
1884
1885                 if(!name) {
1886                         name = argv[i];
1887                         continue;
1888                 }
1889
1890                 error("Cannot understand option %s", argv[i]);
1891                 exit(1);
1892         }
1893
1894         if(freq > 0) update_every = freq;
1895         if(!name) name = "groups";
1896
1897         if(read_process_groups(name)) {
1898                 error("Cannot read process groups %s", name);
1899                 exit(1);
1900         }
1901 }
1902
1903 int main(int argc, char **argv)
1904 {
1905         // debug_flags = D_PROCFILE;
1906
1907         // set the name for logging
1908         program_name = "apps.plugin";
1909
1910         info("starting...");
1911
1912         procfile_adaptive_initial_allocation = 1;
1913
1914         unsigned long started_t = time(NULL), current_t;
1915         Hertz = get_hertz();
1916         pid_max = get_pid_max();
1917         processors = get_processors();
1918
1919         parse_args(argc, argv);
1920
1921         all_pids = calloc(sizeof(struct pid_stat *), pid_max);
1922         if(!all_pids) {
1923                 error("Cannot allocate %lu bytes of memory.", sizeof(struct pid_stat *) * pid_max);
1924                 printf("DISABLE\n");
1925                 exit(1);
1926         }
1927
1928         unsigned long long counter = 1;
1929         unsigned long long usec = 0, susec = 0;
1930         struct timeval last, now;
1931         gettimeofday(&last, NULL);
1932
1933         for(;1; counter++) {
1934                 if(!update_from_proc()) {
1935                         error("Cannot allocate %lu bytes of memory.", sizeof(struct pid_stat *) * pid_max);
1936                         printf("DISABLE\n");
1937                         exit(1);
1938                 }
1939
1940                 update_statistics();
1941                 show_charts();          // this is smart enough to show only newly added apps, when needed
1942                 show_dimensions();
1943
1944                 if(debug) fprintf(stderr, "apps.plugin: done Loop No %llu\n", counter);
1945                 fflush(NULL);
1946
1947                 gettimeofday(&now, NULL);
1948                 usec = usecdiff(&now, &last) - susec;
1949                 if(debug) fprintf(stderr, "apps.plugin: last loop took %llu usec (worked for %llu, sleeped for %llu).\n", usec + susec, usec, susec);
1950
1951                 // if the last loop took less than half the time
1952                 // wait the rest of the time
1953                 if(usec < (update_every * 1000000ULL / 2)) susec = (update_every * 1000000ULL) - usec;
1954                 else susec = update_every * 1000000ULL / 2;
1955
1956                 usleep(susec);
1957                 bcopy(&now, &last, sizeof(struct timeval));
1958
1959                 current_t = time(NULL);
1960                 if(current_t - started_t > 3600) exit(0);
1961         }
1962 }