src/journal/sd-journal.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2
   3 #include <errno.h>
   4 #include <fcntl.h>
   5 #include <inttypes.h>
   6 #include <linux/magic.h>
   7 #include <poll.h>
   8 #include <stddef.h>
   9 #include <sys/inotify.h>
  10 #include <sys/vfs.h>
  11 #include <unistd.h>
  12
  13 #include "sd-journal.h"
  14
  15 #include "alloc-util.h"
  16 #include "catalog.h"
  17 #include "compress.h"
  18 #include "dirent-util.h"
  19 #include "env-file.h"
  20 #include "escape.h"
  21 #include "fd-util.h"
  22 #include "fileio.h"
  23 #include "format-util.h"
  24 #include "fs-util.h"
  25 #include "hashmap.h"
  26 #include "hostname-util.h"
  27 #include "id128-util.h"
  28 #include "io-util.h"
  29 #include "journal-def.h"
  30 #include "journal-file.h"
  31 #include "journal-internal.h"
  32 #include "list.h"
  33 #include "lookup3.h"
  34 #include "nulstr-util.h"
  35 #include "path-util.h"
  36 #include "process-util.h"
  37 #include "replace-var.h"
  38 #include "stat-util.h"
  39 #include "stdio-util.h"
  40 #include "string-util.h"
  41 #include "strv.h"
  42 #include "syslog-util.h"
  43
  44 #define JOURNAL_FILES_MAX 7168
  45
  46 #define JOURNAL_FILES_RECHECK_USEC (2 * USEC_PER_SEC)
  47
  48 #define REPLACE_VAR_MAX 256
  49
  50 #define DEFAULT_DATA_THRESHOLD (64*1024)
  51
  52 static void remove_file_real(sd_journal *j, JournalFile *f);
  53
  54 static bool journal_pid_changed(sd_journal *j) {
  55         assert(j);
  56
  57         /* We don't support people creating a journal object and
  58          * keeping it around over a fork(). Let's complain. */
  59
  60         return j->original_pid != getpid_cached();
  61 }
  62
  63 static int journal_put_error(sd_journal *j, int r, const char *path) {
  64         char *copy;
  65         int k;
  66
  67         /* Memorize an error we encountered, and store which
  68          * file/directory it was generated from. Note that we store
  69          * only *one* path per error code, as the error code is the
  70          * key into the hashmap, and the path is the value. This means
  71          * we keep track only of all error kinds, but not of all error
  72          * locations. This has the benefit that the hashmap cannot
  73          * grow beyond bounds.
  74          *
  75          * We return an error here only if we didn't manage to
  76          * memorize the real error. */
  77
  78         if (r >= 0)
  79                 return r;
  80
  81         k = hashmap_ensure_allocated(&j->errors, NULL);
  82         if (k < 0)
  83                 return k;
  84
  85         if (path) {
  86                 copy = strdup(path);
  87                 if (!copy)
  88                         return -ENOMEM;
  89         } else
  90                 copy = NULL;
  91
  92         k = hashmap_put(j->errors, INT_TO_PTR(r), copy);
  93         if (k < 0) {
  94                 free(copy);
  95
  96                 if (k == -EEXIST)
  97                         return 0;
  98
  99                 return k;
 100         }
 101
 102         return 0;
 103 }
 104
 105 static void detach_location(sd_journal *j) {
 106         Iterator i;
 107         JournalFile *f;
 108
 109         assert(j);
 110
 111         j->current_file = NULL;
 112         j->current_field = 0;
 113
 114         ORDERED_HASHMAP_FOREACH(f, j->files, i)
 115                 journal_file_reset_location(f);
 116 }
 117
 118 static void init_location(Location *l, LocationType type, JournalFile *f, Object *o) {
 119         assert(l);
 120         assert(IN_SET(type, LOCATION_DISCRETE, LOCATION_SEEK));
 121         assert(f);
 122
 123         *l = (Location) {
 124                 .type = type,
 125                 .seqnum = le64toh(o->entry.seqnum),
 126                 .seqnum_id = f->header->seqnum_id,
 127                 .realtime = le64toh(o->entry.realtime),
 128                 .monotonic = le64toh(o->entry.monotonic),
 129                 .boot_id = o->entry.boot_id,
 130                 .xor_hash = le64toh(o->entry.xor_hash),
 131                 .seqnum_set = true,
 132                 .realtime_set = true,
 133                 .monotonic_set = true,
 134                 .xor_hash_set = true,
 135         };
 136 }
 137
 138 static void set_location(sd_journal *j, JournalFile *f, Object *o) {
 139         assert(j);
 140         assert(f);
 141         assert(o);
 142
 143         init_location(&j->current_location, LOCATION_DISCRETE, f, o);
 144
 145         j->current_file = f;
 146         j->current_field = 0;
 147
 148         /* Let f know its candidate entry was picked. */
 149         assert(f->location_type == LOCATION_SEEK);
 150         f->location_type = LOCATION_DISCRETE;
 151 }
 152
 153 static int match_is_valid(const void *data, size_t size) {
 154         const char *b, *p;
 155
 156         assert(data);
 157
 158         if (size < 2)
 159                 return false;
 160
 161         if (((char*) data)[0] == '_' && ((char*) data)[1] == '_')
 162                 return false;
 163
 164         b = data;
 165         for (p = b; p < b + size; p++) {
 166
 167                 if (*p == '=')
 168                         return p > b;
 169
 170                 if (*p == '_')
 171                         continue;
 172
 173                 if (*p >= 'A' && *p <= 'Z')
 174                         continue;
 175
 176                 if (*p >= '0' && *p <= '9')
 177                         continue;
 178
 179                 return false;
 180         }
 181
 182         return false;
 183 }
 184
 185 static bool same_field(const void *_a, size_t s, const void *_b, size_t t) {
 186         const uint8_t *a = _a, *b = _b;
 187         size_t j;
 188
 189         for (j = 0; j < s && j < t; j++) {
 190
 191                 if (a[j] != b[j])
 192                         return false;
 193
 194                 if (a[j] == '=')
 195                         return true;
 196         }
 197
 198         assert_not_reached("\"=\" not found");
 199 }
 200
 201 static Match *match_new(Match *p, MatchType t) {
 202         Match *m;
 203
 204         m = new(Match, 1);
 205         if (!m)
 206                 return NULL;
 207
 208         *m = (Match) {
 209                 .type = t,
 210                 .parent = p,
 211         };
 212
 213         if (p)
 214                 LIST_PREPEND(matches, p->matches, m);
 215
 216         return m;
 217 }
 218
 219 static void match_free(Match *m) {
 220         assert(m);
 221
 222         while (m->matches)
 223                 match_free(m->matches);
 224
 225         if (m->parent)
 226                 LIST_REMOVE(matches, m->parent->matches, m);
 227
 228         free(m->data);
 229         free(m);
 230 }
 231
 232 static void match_free_if_empty(Match *m) {
 233         if (!m || m->matches)
 234                 return;
 235
 236         match_free(m);
 237 }
 238
 239 _public_ int sd_journal_add_match(sd_journal *j, const void *data, size_t size) {
 240         Match *l3, *l4, *add_here = NULL, *m;
 241         le64_t le_hash;
 242
 243         assert_return(j, -EINVAL);
 244         assert_return(!journal_pid_changed(j), -ECHILD);
 245         assert_return(data, -EINVAL);
 246
 247         if (size == 0)
 248                 size = strlen(data);
 249
 250         assert_return(match_is_valid(data, size), -EINVAL);
 251
 252         /* level 0: AND term
 253          * level 1: OR terms
 254          * level 2: AND terms
 255          * level 3: OR terms
 256          * level 4: concrete matches */
 257
 258         if (!j->level0) {
 259                 j->level0 = match_new(NULL, MATCH_AND_TERM);
 260                 if (!j->level0)
 261                         return -ENOMEM;
 262         }
 263
 264         if (!j->level1) {
 265                 j->level1 = match_new(j->level0, MATCH_OR_TERM);
 266                 if (!j->level1)
 267                         return -ENOMEM;
 268         }
 269
 270         if (!j->level2) {
 271                 j->level2 = match_new(j->level1, MATCH_AND_TERM);
 272                 if (!j->level2)
 273                         return -ENOMEM;
 274         }
 275
 276         assert(j->level0->type == MATCH_AND_TERM);
 277         assert(j->level1->type == MATCH_OR_TERM);
 278         assert(j->level2->type == MATCH_AND_TERM);
 279
 280         le_hash = htole64(hash64(data, size));
 281
 282         LIST_FOREACH(matches, l3, j->level2->matches) {
 283                 assert(l3->type == MATCH_OR_TERM);
 284
 285                 LIST_FOREACH(matches, l4, l3->matches) {
 286                         assert(l4->type == MATCH_DISCRETE);
 287
 288                         /* Exactly the same match already? Then ignore
 289                          * this addition */
 290                         if (l4->le_hash == le_hash &&
 291                             l4->size == size &&
 292                             memcmp(l4->data, data, size) == 0)
 293                                 return 0;
 294
 295                         /* Same field? Then let's add this to this OR term */
 296                         if (same_field(data, size, l4->data, l4->size)) {
 297                                 add_here = l3;
 298                                 break;
 299                         }
 300                 }
 301
 302                 if (add_here)
 303                         break;
 304         }
 305
 306         if (!add_here) {
 307                 add_here = match_new(j->level2, MATCH_OR_TERM);
 308                 if (!add_here)
 309                         goto fail;
 310         }
 311
 312         m = match_new(add_here, MATCH_DISCRETE);
 313         if (!m)
 314                 goto fail;
 315
 316         m->le_hash = le_hash;
 317         m->size = size;
 318         m->data = memdup(data, size);
 319         if (!m->data)
 320                 goto fail;
 321
 322         detach_location(j);
 323
 324         return 0;
 325
 326 fail:
 327         match_free_if_empty(add_here);
 328         match_free_if_empty(j->level2);
 329         match_free_if_empty(j->level1);
 330         match_free_if_empty(j->level0);
 331
 332         return -ENOMEM;
 333 }
 334
 335 _public_ int sd_journal_add_conjunction(sd_journal *j) {
 336         assert_return(j, -EINVAL);
 337         assert_return(!journal_pid_changed(j), -ECHILD);
 338
 339         if (!j->level0)
 340                 return 0;
 341
 342         if (!j->level1)
 343                 return 0;
 344
 345         if (!j->level1->matches)
 346                 return 0;
 347
 348         j->level1 = NULL;
 349         j->level2 = NULL;
 350
 351         return 0;
 352 }
 353
 354 _public_ int sd_journal_add_disjunction(sd_journal *j) {
 355         assert_return(j, -EINVAL);
 356         assert_return(!journal_pid_changed(j), -ECHILD);
 357
 358         if (!j->level0)
 359                 return 0;
 360
 361         if (!j->level1)
 362                 return 0;
 363
 364         if (!j->level2)
 365                 return 0;
 366
 367         if (!j->level2->matches)
 368                 return 0;
 369
 370         j->level2 = NULL;
 371         return 0;
 372 }
 373
 374 static char *match_make_string(Match *m) {
 375         char *p = NULL, *r;
 376         Match *i;
 377         bool enclose = false;
 378
 379         if (!m)
 380                 return strdup("none");
 381
 382         if (m->type == MATCH_DISCRETE)
 383                 return cescape_length(m->data, m->size);
 384
 385         LIST_FOREACH(matches, i, m->matches) {
 386                 char *t, *k;
 387
 388                 t = match_make_string(i);
 389                 if (!t)
 390                         return mfree(p);
 391
 392                 if (p) {
 393                         k = strjoin(p, m->type == MATCH_OR_TERM ? " OR " : " AND ", t);
 394                         free(p);
 395                         free(t);
 396
 397                         if (!k)
 398                                 return NULL;
 399
 400                         p = k;
 401
 402                         enclose = true;
 403                 } else
 404                         p = t;
 405         }
 406
 407         if (enclose) {
 408                 r = strjoin("(", p, ")");
 409                 free(p);
 410                 return r;
 411         }
 412
 413         return p;
 414 }
 415
 416 char *journal_make_match_string(sd_journal *j) {
 417         assert(j);
 418
 419         return match_make_string(j->level0);
 420 }
 421
 422 _public_ void sd_journal_flush_matches(sd_journal *j) {
 423         if (!j)
 424                 return;
 425
 426         if (j->level0)
 427                 match_free(j->level0);
 428
 429         j->level0 = j->level1 = j->level2 = NULL;
 430
 431         detach_location(j);
 432 }
 433
 434 _pure_ static int compare_with_location(const JournalFile *f, const Location *l, const JournalFile *current_file) {
 435         int r;
 436
 437         assert(f);
 438         assert(l);
 439         assert(f->location_type == LOCATION_SEEK);
 440         assert(IN_SET(l->type, LOCATION_DISCRETE, LOCATION_SEEK));
 441
 442         if (l->monotonic_set &&
 443             sd_id128_equal(f->current_boot_id, l->boot_id) &&
 444             l->realtime_set &&
 445             f->current_realtime == l->realtime &&
 446             l->xor_hash_set &&
 447             f->current_xor_hash == l->xor_hash &&
 448             f != current_file)
 449                 return 0;
 450
 451         if (l->seqnum_set &&
 452             sd_id128_equal(f->header->seqnum_id, l->seqnum_id)) {
 453
 454                 r = CMP(f->current_seqnum, l->seqnum);
 455                 if (r != 0)
 456                         return r;
 457         }
 458
 459         if (l->monotonic_set &&
 460             sd_id128_equal(f->current_boot_id, l->boot_id)) {
 461
 462                 r = CMP(f->current_monotonic, l->monotonic);
 463                 if (r != 0)
 464                         return r;
 465         }
 466
 467         if (l->realtime_set) {
 468
 469                 r = CMP(f->current_realtime, l->realtime);
 470                 if (r != 0)
 471                         return r;
 472         }
 473
 474         if (l->xor_hash_set) {
 475
 476                 r = CMP(f->current_xor_hash, l->xor_hash);
 477                 if (r != 0)
 478                         return r;
 479         }
 480
 481         return 0;
 482 }
 483
 484 static int next_for_match(
 485                 sd_journal *j,
 486                 Match *m,
 487                 JournalFile *f,
 488                 uint64_t after_offset,
 489                 direction_t direction,
 490                 Object **ret,
 491                 uint64_t *offset) {
 492
 493         int r;
 494         uint64_t np = 0;
 495         Object *n;
 496
 497         assert(j);
 498         assert(m);
 499         assert(f);
 500
 501         if (m->type == MATCH_DISCRETE) {
 502                 uint64_t dp;
 503
 504                 r = journal_file_find_data_object_with_hash(f, m->data, m->size, le64toh(m->le_hash), NULL, &dp);
 505                 if (r <= 0)
 506                         return r;
 507
 508                 return journal_file_move_to_entry_by_offset_for_data(f, dp, after_offset, direction, ret, offset);
 509
 510         } else if (m->type == MATCH_OR_TERM) {
 511                 Match *i;
 512
 513                 /* Find the earliest match beyond after_offset */
 514
 515                 LIST_FOREACH(matches, i, m->matches) {
 516                         uint64_t cp;
 517
 518                         r = next_for_match(j, i, f, after_offset, direction, NULL, &cp);
 519                         if (r < 0)
 520                                 return r;
 521                         else if (r > 0) {
 522                                 if (np == 0 || (direction == DIRECTION_DOWN ? cp < np : cp > np))
 523                                         np = cp;
 524                         }
 525                 }
 526
 527                 if (np == 0)
 528                         return 0;
 529
 530         } else if (m->type == MATCH_AND_TERM) {
 531                 Match *i, *last_moved;
 532
 533                 /* Always jump to the next matching entry and repeat
 534                  * this until we find an offset that matches for all
 535                  * matches. */
 536
 537                 if (!m->matches)
 538                         return 0;
 539
 540                 r = next_for_match(j, m->matches, f, after_offset, direction, NULL, &np);
 541                 if (r <= 0)
 542                         return r;
 543
 544                 assert(direction == DIRECTION_DOWN ? np >= after_offset : np <= after_offset);
 545                 last_moved = m->matches;
 546
 547                 LIST_LOOP_BUT_ONE(matches, i, m->matches, last_moved) {
 548                         uint64_t cp;
 549
 550                         r = next_for_match(j, i, f, np, direction, NULL, &cp);
 551                         if (r <= 0)
 552                                 return r;
 553
 554                         assert(direction == DIRECTION_DOWN ? cp >= np : cp <= np);
 555                         if (direction == DIRECTION_DOWN ? cp > np : cp < np) {
 556                                 np = cp;
 557                                 last_moved = i;
 558                         }
 559                 }
 560         }
 561
 562         assert(np > 0);
 563
 564         r = journal_file_move_to_object(f, OBJECT_ENTRY, np, &n);
 565         if (r < 0)
 566                 return r;
 567
 568         if (ret)
 569                 *ret = n;
 570         if (offset)
 571                 *offset = np;
 572
 573         return 1;
 574 }
 575
 576 static int find_location_for_match(
 577                 sd_journal *j,
 578                 Match *m,
 579                 JournalFile *f,
 580                 direction_t direction,
 581                 Object **ret,
 582                 uint64_t *offset) {
 583
 584         int r;
 585
 586         assert(j);
 587         assert(m);
 588         assert(f);
 589
 590         if (m->type == MATCH_DISCRETE) {
 591                 uint64_t dp;
 592
 593                 r = journal_file_find_data_object_with_hash(f, m->data, m->size, le64toh(m->le_hash), NULL, &dp);
 594                 if (r <= 0)
 595                         return r;
 596
 597                 /* FIXME: missing: find by monotonic */
 598
 599                 if (j->current_location.type == LOCATION_HEAD)
 600                         return journal_file_next_entry_for_data(f, NULL, 0, dp, DIRECTION_DOWN, ret, offset);
 601                 if (j->current_location.type == LOCATION_TAIL)
 602                         return journal_file_next_entry_for_data(f, NULL, 0, dp, DIRECTION_UP, ret, offset);
 603                 if (j->current_location.seqnum_set && sd_id128_equal(j->current_location.seqnum_id, f->header->seqnum_id))
 604                         return journal_file_move_to_entry_by_seqnum_for_data(f, dp, j->current_location.seqnum, direction, ret, offset);
 605                 if (j->current_location.monotonic_set) {
 606                         r = journal_file_move_to_entry_by_monotonic_for_data(f, dp, j->current_location.boot_id, j->current_location.monotonic, direction, ret, offset);
 607                         if (r != -ENOENT)
 608                                 return r;
 609                 }
 610                 if (j->current_location.realtime_set)
 611                         return journal_file_move_to_entry_by_realtime_for_data(f, dp, j->current_location.realtime, direction, ret, offset);
 612
 613                 return journal_file_next_entry_for_data(f, NULL, 0, dp, direction, ret, offset);
 614
 615         } else if (m->type == MATCH_OR_TERM) {
 616                 uint64_t np = 0;
 617                 Object *n;
 618                 Match *i;
 619
 620                 /* Find the earliest match */
 621
 622                 LIST_FOREACH(matches, i, m->matches) {
 623                         uint64_t cp;
 624
 625                         r = find_location_for_match(j, i, f, direction, NULL, &cp);
 626                         if (r < 0)
 627                                 return r;
 628                         else if (r > 0) {
 629                                 if (np == 0 || (direction == DIRECTION_DOWN ? np > cp : np < cp))
 630                                         np = cp;
 631                         }
 632                 }
 633
 634                 if (np == 0)
 635                         return 0;
 636
 637                 r = journal_file_move_to_object(f, OBJECT_ENTRY, np, &n);
 638                 if (r < 0)
 639                         return r;
 640
 641                 if (ret)
 642                         *ret = n;
 643                 if (offset)
 644                         *offset = np;
 645
 646                 return 1;
 647
 648         } else {
 649                 Match *i;
 650                 uint64_t np = 0;
 651
 652                 assert(m->type == MATCH_AND_TERM);
 653
 654                 /* First jump to the last match, and then find the
 655                  * next one where all matches match */
 656
 657                 if (!m->matches)
 658                         return 0;
 659
 660                 LIST_FOREACH(matches, i, m->matches) {
 661                         uint64_t cp;
 662
 663                         r = find_location_for_match(j, i, f, direction, NULL, &cp);
 664                         if (r <= 0)
 665                                 return r;
 666
 667                         if (np == 0 || (direction == DIRECTION_DOWN ? cp > np : cp < np))
 668                                 np = cp;
 669                 }
 670
 671                 return next_for_match(j, m, f, np, direction, ret, offset);
 672         }
 673 }
 674
 675 static int find_location_with_matches(
 676                 sd_journal *j,
 677                 JournalFile *f,
 678                 direction_t direction,
 679                 Object **ret,
 680                 uint64_t *offset) {
 681
 682         int r;
 683
 684         assert(j);
 685         assert(f);
 686         assert(ret);
 687         assert(offset);
 688
 689         if (!j->level0) {
 690                 /* No matches is simple */
 691
 692                 if (j->current_location.type == LOCATION_HEAD)
 693                         return journal_file_next_entry(f, 0, DIRECTION_DOWN, ret, offset);
 694                 if (j->current_location.type == LOCATION_TAIL)
 695                         return journal_file_next_entry(f, 0, DIRECTION_UP, ret, offset);
 696                 if (j->current_location.seqnum_set && sd_id128_equal(j->current_location.seqnum_id, f->header->seqnum_id))
 697                         return journal_file_move_to_entry_by_seqnum(f, j->current_location.seqnum, direction, ret, offset);
 698                 if (j->current_location.monotonic_set) {
 699                         r = journal_file_move_to_entry_by_monotonic(f, j->current_location.boot_id, j->current_location.monotonic, direction, ret, offset);
 700                         if (r != -ENOENT)
 701                                 return r;
 702                 }
 703                 if (j->current_location.realtime_set)
 704                         return journal_file_move_to_entry_by_realtime(f, j->current_location.realtime, direction, ret, offset);
 705
 706                 return journal_file_next_entry(f, 0, direction, ret, offset);
 707         } else
 708                 return find_location_for_match(j, j->level0, f, direction, ret, offset);
 709 }
 710
 711 static int next_with_matches(
 712                 sd_journal *j,
 713                 JournalFile *f,
 714                 direction_t direction,
 715                 Object **ret,
 716                 uint64_t *offset) {
 717
 718         assert(j);
 719         assert(f);
 720         assert(ret);
 721         assert(offset);
 722
 723         /* No matches is easy. We simple advance the file
 724          * pointer by one. */
 725         if (!j->level0)
 726                 return journal_file_next_entry(f, f->current_offset, direction, ret, offset);
 727
 728         /* If we have a match then we look for the next matching entry
 729          * with an offset at least one step larger */
 730         return next_for_match(j, j->level0, f,
 731                               direction == DIRECTION_DOWN ? f->current_offset + 1
 732                                                           : f->current_offset - 1,
 733                               direction, ret, offset);
 734 }
 735
 736 static int next_beyond_location(sd_journal *j, JournalFile *f, direction_t direction) {
 737         Object *c;
 738         uint64_t cp, n_entries;
 739         int r;
 740
 741         assert(j);
 742         assert(f);
 743
 744         n_entries = le64toh(f->header->n_entries);
 745
 746         /* If we hit EOF before, we don't need to look into this file again
 747          * unless direction changed or new entries appeared. */
 748         if (f->last_direction == direction && f->location_type == LOCATION_TAIL &&
 749             n_entries == f->last_n_entries)
 750                 return 0;
 751
 752         f->last_n_entries = n_entries;
 753
 754         if (f->last_direction == direction && f->current_offset > 0) {
 755                 /* LOCATION_SEEK here means we did the work in a previous
 756                  * iteration and the current location already points to a
 757                  * candidate entry. */
 758                 if (f->location_type != LOCATION_SEEK) {
 759                         r = next_with_matches(j, f, direction, &c, &cp);
 760                         if (r <= 0)
 761                                 return r;
 762
 763                         journal_file_save_location(f, c, cp);
 764                 }
 765         } else {
 766                 f->last_direction = direction;
 767
 768                 r = find_location_with_matches(j, f, direction, &c, &cp);
 769                 if (r <= 0)
 770                         return r;
 771
 772                 journal_file_save_location(f, c, cp);
 773         }
 774
 775         /* OK, we found the spot, now let's advance until an entry
 776          * that is actually different from what we were previously
 777          * looking at. This is necessary to handle entries which exist
 778          * in two (or more) journal files, and which shall all be
 779          * suppressed but one. */
 780
 781         for (;;) {
 782                 bool found;
 783
 784                 if (j->current_location.type == LOCATION_DISCRETE) {
 785                         int k;
 786
 787                         k = compare_with_location(f, &j->current_location, j->current_file);
 788
 789                         found = direction == DIRECTION_DOWN ? k > 0 : k < 0;
 790                 } else
 791                         found = true;
 792
 793                 if (found)
 794                         return 1;
 795
 796                 r = next_with_matches(j, f, direction, &c, &cp);
 797                 if (r <= 0)
 798                         return r;
 799
 800                 journal_file_save_location(f, c, cp);
 801         }
 802 }
 803
 804 static int real_journal_next(sd_journal *j, direction_t direction) {
 805         JournalFile *new_file = NULL;
 806         unsigned i, n_files;
 807         const void **files;
 808         Object *o;
 809         int r;
 810
 811         assert_return(j, -EINVAL);
 812         assert_return(!journal_pid_changed(j), -ECHILD);
 813
 814         r = iterated_cache_get(j->files_cache, NULL, &files, &n_files);
 815         if (r < 0)
 816                 return r;
 817
 818         for (i = 0; i < n_files; i++) {
 819                 JournalFile *f = (JournalFile *)files[i];
 820                 bool found;
 821
 822                 r = next_beyond_location(j, f, direction);
 823                 if (r < 0) {
 824                         log_debug_errno(r, "Can't iterate through %s, ignoring: %m", f->path);
 825                         remove_file_real(j, f);
 826                         continue;
 827                 } else if (r == 0) {
 828                         f->location_type = LOCATION_TAIL;
 829                         continue;
 830                 }
 831
 832                 if (!new_file)
 833                         found = true;
 834                 else {
 835                         int k;
 836
 837                         k = journal_file_compare_locations(f, new_file);
 838
 839                         found = direction == DIRECTION_DOWN ? k < 0 : k > 0;
 840                 }
 841
 842                 if (found)
 843                         new_file = f;
 844         }
 845
 846         if (!new_file)
 847                 return 0;
 848
 849         r = journal_file_move_to_object(new_file, OBJECT_ENTRY, new_file->current_offset, &o);
 850         if (r < 0)
 851                 return r;
 852
 853         set_location(j, new_file, o);
 854
 855         return 1;
 856 }
 857
 858 _public_ int sd_journal_next(sd_journal *j) {
 859         return real_journal_next(j, DIRECTION_DOWN);
 860 }
 861
 862 _public_ int sd_journal_previous(sd_journal *j) {
 863         return real_journal_next(j, DIRECTION_UP);
 864 }
 865
 866 static int real_journal_next_skip(sd_journal *j, direction_t direction, uint64_t skip) {
 867         int c = 0, r;
 868
 869         assert_return(j, -EINVAL);
 870         assert_return(!journal_pid_changed(j), -ECHILD);
 871
 872         if (skip == 0) {
 873                 /* If this is not a discrete skip, then at least
 874                  * resolve the current location */
 875                 if (j->current_location.type != LOCATION_DISCRETE) {
 876                         r = real_journal_next(j, direction);
 877                         if (r < 0)
 878                                 return r;
 879                 }
 880
 881                 return 0;
 882         }
 883
 884         do {
 885                 r = real_journal_next(j, direction);
 886                 if (r < 0)
 887                         return r;
 888
 889                 if (r == 0)
 890                         return c;
 891
 892                 skip--;
 893                 c++;
 894         } while (skip > 0);
 895
 896         return c;
 897 }
 898
 899 _public_ int sd_journal_next_skip(sd_journal *j, uint64_t skip) {
 900         return real_journal_next_skip(j, DIRECTION_DOWN, skip);
 901 }
 902
 903 _public_ int sd_journal_previous_skip(sd_journal *j, uint64_t skip) {
 904         return real_journal_next_skip(j, DIRECTION_UP, skip);
 905 }
 906
 907 _public_ int sd_journal_get_cursor(sd_journal *j, char **cursor) {
 908         Object *o;
 909         int r;
 910         char bid[SD_ID128_STRING_MAX], sid[SD_ID128_STRING_MAX];
 911
 912         assert_return(j, -EINVAL);
 913         assert_return(!journal_pid_changed(j), -ECHILD);
 914         assert_return(cursor, -EINVAL);
 915
 916         if (!j->current_file || j->current_file->current_offset <= 0)
 917                 return -EADDRNOTAVAIL;
 918
 919         r = journal_file_move_to_object(j->current_file, OBJECT_ENTRY, j->current_file->current_offset, &o);
 920         if (r < 0)
 921                 return r;
 922
 923         sd_id128_to_string(j->current_file->header->seqnum_id, sid);
 924         sd_id128_to_string(o->entry.boot_id, bid);
 925
 926         if (asprintf(cursor,
 927                      "s=%s;i=%"PRIx64";b=%s;m=%"PRIx64";t=%"PRIx64";x=%"PRIx64,
 928                      sid, le64toh(o->entry.seqnum),
 929                      bid, le64toh(o->entry.monotonic),
 930                      le64toh(o->entry.realtime),
 931                      le64toh(o->entry.xor_hash)) < 0)
 932                 return -ENOMEM;
 933
 934         return 0;
 935 }
 936
 937 _public_ int sd_journal_seek_cursor(sd_journal *j, const char *cursor) {
 938         const char *word, *state;
 939         size_t l;
 940         unsigned long long seqnum, monotonic, realtime, xor_hash;
 941         bool
 942                 seqnum_id_set = false,
 943                 seqnum_set = false,
 944                 boot_id_set = false,
 945                 monotonic_set = false,
 946                 realtime_set = false,
 947                 xor_hash_set = false;
 948         sd_id128_t seqnum_id, boot_id;
 949
 950         assert_return(j, -EINVAL);
 951         assert_return(!journal_pid_changed(j), -ECHILD);
 952         assert_return(!isempty(cursor), -EINVAL);
 953
 954         FOREACH_WORD_SEPARATOR(word, l, cursor, ";", state) {
 955                 char *item;
 956                 int k = 0;
 957
 958                 if (l < 2 || word[1] != '=')
 959                         return -EINVAL;
 960
 961                 item = strndup(word, l);
 962                 if (!item)
 963                         return -ENOMEM;
 964
 965                 switch (word[0]) {
 966
 967                 case 's':
 968                         seqnum_id_set = true;
 969                         k = sd_id128_from_string(item+2, &seqnum_id);
 970                         break;
 971
 972                 case 'i':
 973                         seqnum_set = true;
 974                         if (sscanf(item+2, "%llx", &seqnum) != 1)
 975                                 k = -EINVAL;
 976                         break;
 977
 978                 case 'b':
 979                         boot_id_set = true;
 980                         k = sd_id128_from_string(item+2, &boot_id);
 981                         break;
 982
 983                 case 'm':
 984                         monotonic_set = true;
 985                         if (sscanf(item+2, "%llx", &monotonic) != 1)
 986                                 k = -EINVAL;
 987                         break;
 988
 989                 case 't':
 990                         realtime_set = true;
 991                         if (sscanf(item+2, "%llx", &realtime) != 1)
 992                                 k = -EINVAL;
 993                         break;
 994
 995                 case 'x':
 996                         xor_hash_set = true;
 997                         if (sscanf(item+2, "%llx", &xor_hash) != 1)
 998                                 k = -EINVAL;
 999                         break;
1000                 }
1001
1002                 free(item);
1003
1004                 if (k < 0)
1005                         return k;
1006         }
1007
1008         if ((!seqnum_set || !seqnum_id_set) &&
1009             (!monotonic_set || !boot_id_set) &&
1010             !realtime_set)
1011                 return -EINVAL;
1012
1013         detach_location(j);
1014         j->current_location = (Location) {
1015                 .type = LOCATION_SEEK,
1016         };
1017
1018         if (realtime_set) {
1019                 j->current_location.realtime = (uint64_t) realtime;
1020                 j->current_location.realtime_set = true;
1021         }
1022
1023         if (seqnum_set && seqnum_id_set) {
1024                 j->current_location.seqnum = (uint64_t) seqnum;
1025                 j->current_location.seqnum_id = seqnum_id;
1026                 j->current_location.seqnum_set = true;
1027         }
1028
1029         if (monotonic_set && boot_id_set) {
1030                 j->current_location.monotonic = (uint64_t) monotonic;
1031                 j->current_location.boot_id = boot_id;
1032                 j->current_location.monotonic_set = true;
1033         }
1034
1035         if (xor_hash_set) {
1036                 j->current_location.xor_hash = (uint64_t) xor_hash;
1037                 j->current_location.xor_hash_set = true;
1038         }
1039
1040         return 0;
1041 }
1042
1043 _public_ int sd_journal_test_cursor(sd_journal *j, const char *cursor) {
1044         int r;
1045         Object *o;
1046
1047         assert_return(j, -EINVAL);
1048         assert_return(!journal_pid_changed(j), -ECHILD);
1049         assert_return(!isempty(cursor), -EINVAL);
1050
1051         if (!j->current_file || j->current_file->current_offset <= 0)
1052                 return -EADDRNOTAVAIL;
1053
1054         r = journal_file_move_to_object(j->current_file, OBJECT_ENTRY, j->current_file->current_offset, &o);
1055         if (r < 0)
1056                 return r;
1057
1058         for (;;) {
1059                 _cleanup_free_ char *item = NULL;
1060                 unsigned long long ll;
1061                 sd_id128_t id;
1062                 int k = 0;
1063
1064                 r = extract_first_word(&cursor, &item, ";", EXTRACT_DONT_COALESCE_SEPARATORS);
1065                 if (r < 0)
1066                         return r;
1067
1068                 if (r == 0)
1069                         break;
1070
1071                 if (strlen(item) < 2 || item[1] != '=')
1072                         return -EINVAL;
1073
1074                 switch (item[0]) {
1075
1076                 case 's':
1077                         k = sd_id128_from_string(item+2, &id);
1078                         if (k < 0)
1079                                 return k;
1080                         if (!sd_id128_equal(id, j->current_file->header->seqnum_id))
1081                                 return 0;
1082                         break;
1083
1084                 case 'i':
1085                         if (sscanf(item+2, "%llx", &ll) != 1)
1086                                 return -EINVAL;
1087                         if (ll != le64toh(o->entry.seqnum))
1088                                 return 0;
1089                         break;
1090
1091                 case 'b':
1092                         k = sd_id128_from_string(item+2, &id);
1093                         if (k < 0)
1094                                 return k;
1095                         if (!sd_id128_equal(id, o->entry.boot_id))
1096                                 return 0;
1097                         break;
1098
1099                 case 'm':
1100                         if (sscanf(item+2, "%llx", &ll) != 1)
1101                                 return -EINVAL;
1102                         if (ll != le64toh(o->entry.monotonic))
1103                                 return 0;
1104                         break;
1105
1106                 case 't':
1107                         if (sscanf(item+2, "%llx", &ll) != 1)
1108                                 return -EINVAL;
1109                         if (ll != le64toh(o->entry.realtime))
1110                                 return 0;
1111                         break;
1112
1113                 case 'x':
1114                         if (sscanf(item+2, "%llx", &ll) != 1)
1115                                 return -EINVAL;
1116                         if (ll != le64toh(o->entry.xor_hash))
1117                                 return 0;
1118                         break;
1119                 }
1120         }
1121
1122         return 1;
1123 }
1124
1125 _public_ int sd_journal_seek_monotonic_usec(sd_journal *j, sd_id128_t boot_id, uint64_t usec) {
1126         assert_return(j, -EINVAL);
1127         assert_return(!journal_pid_changed(j), -ECHILD);
1128
1129         detach_location(j);
1130
1131         j->current_location = (Location) {
1132                 .type = LOCATION_SEEK,
1133                 .boot_id = boot_id,
1134                 .monotonic = usec,
1135                 .monotonic_set = true,
1136         };
1137
1138         return 0;
1139 }
1140
1141 _public_ int sd_journal_seek_realtime_usec(sd_journal *j, uint64_t usec) {
1142         assert_return(j, -EINVAL);
1143         assert_return(!journal_pid_changed(j), -ECHILD);
1144
1145         detach_location(j);
1146
1147         j->current_location = (Location) {
1148                 .type = LOCATION_SEEK,
1149                 .realtime = usec,
1150                 .realtime_set = true,
1151         };
1152
1153         return 0;
1154 }
1155
1156 _public_ int sd_journal_seek_head(sd_journal *j) {
1157         assert_return(j, -EINVAL);
1158         assert_return(!journal_pid_changed(j), -ECHILD);
1159
1160         detach_location(j);
1161
1162         j->current_location = (Location) {
1163                 .type = LOCATION_HEAD,
1164         };
1165
1166         return 0;
1167 }
1168
1169 _public_ int sd_journal_seek_tail(sd_journal *j) {
1170         assert_return(j, -EINVAL);
1171         assert_return(!journal_pid_changed(j), -ECHILD);
1172
1173         detach_location(j);
1174
1175         j->current_location = (Location) {
1176                 .type = LOCATION_TAIL,
1177         };
1178
1179         return 0;
1180 }
1181
1182 static void check_network(sd_journal *j, int fd) {
1183         assert(j);
1184
1185         if (j->on_network)
1186                 return;
1187
1188         j->on_network = fd_is_network_fs(fd);
1189 }
1190
1191 static bool file_has_type_prefix(const char *prefix, const char *filename) {
1192         const char *full, *tilded, *atted;
1193
1194         full = strjoina(prefix, ".journal");
1195         tilded = strjoina(full, "~");
1196         atted = strjoina(prefix, "@");
1197
1198         return STR_IN_SET(filename, full, tilded) ||
1199                startswith(filename, atted);
1200 }
1201
1202 static bool file_type_wanted(int flags, const char *filename) {
1203         assert(filename);
1204
1205         if (!endswith(filename, ".journal") && !endswith(filename, ".journal~"))
1206                 return false;
1207
1208         /* no flags set → every type is OK */
1209         if (!(flags & (SD_JOURNAL_SYSTEM | SD_JOURNAL_CURRENT_USER)))
1210                 return true;
1211
1212         if (flags & SD_JOURNAL_SYSTEM && file_has_type_prefix("system", filename))
1213                 return true;
1214
1215         if (flags & SD_JOURNAL_CURRENT_USER) {
1216                 char prefix[5 + DECIMAL_STR_MAX(uid_t) + 1];
1217
1218                 xsprintf(prefix, "user-"UID_FMT, getuid());
1219
1220                 if (file_has_type_prefix(prefix, filename))
1221                         return true;
1222         }
1223
1224         return false;
1225 }
1226
1227 static bool path_has_prefix(sd_journal *j, const char *path, const char *prefix) {
1228         assert(j);
1229         assert(path);
1230         assert(prefix);
1231
1232         if (j->toplevel_fd >= 0)
1233                 return false;
1234
1235         return path_startswith(path, prefix);
1236 }
1237
1238 static void track_file_disposition(sd_journal *j, JournalFile *f) {
1239         assert(j);
1240         assert(f);
1241
1242         if (!j->has_runtime_files && path_has_prefix(j, f->path, "/run"))
1243                 j->has_runtime_files = true;
1244         else if (!j->has_persistent_files && path_has_prefix(j, f->path, "/var"))
1245                 j->has_persistent_files = true;
1246 }
1247
1248 static const char *skip_slash(const char *p) {
1249
1250         if (!p)
1251                 return NULL;
1252
1253         while (*p == '/')
1254                 p++;
1255
1256         return p;
1257 }
1258
1259 static int add_any_file(
1260                 sd_journal *j,
1261                 int fd,
1262                 const char *path) {
1263
1264         bool close_fd = false;
1265         JournalFile *f;
1266         struct stat st;
1267         int r, k;
1268
1269         assert(j);
1270         assert(fd >= 0 || path);
1271
1272         if (fd < 0) {
1273                 if (j->toplevel_fd >= 0)
1274                         /* If there's a top-level fd defined make the path relative, explicitly, since otherwise
1275                          * openat() ignores the first argument. */
1276
1277                         fd = openat(j->toplevel_fd, skip_slash(path), O_RDONLY|O_CLOEXEC|O_NONBLOCK);
1278                 else
1279                         fd = open(path, O_RDONLY|O_CLOEXEC|O_NONBLOCK);
1280                 if (fd < 0) {
1281                         r = log_debug_errno(errno, "Failed to open journal file %s: %m", path);
1282                         goto finish;
1283                 }
1284
1285                 close_fd = true;
1286
1287                 r = fd_nonblock(fd, false);
1288                 if (r < 0) {
1289                         r = log_debug_errno(errno, "Failed to turn off O_NONBLOCK for %s: %m", path);
1290                         goto finish;
1291                 }
1292         }
1293
1294         if (fstat(fd, &st) < 0) {
1295                 r = log_debug_errno(errno, "Failed to fstat file '%s': %m", path);
1296                 goto finish;
1297         }
1298
1299         r = stat_verify_regular(&st);
1300         if (r < 0) {
1301                 log_debug_errno(r, "Refusing to open '%s', as it is not a regular file.", path);
1302                 goto finish;
1303         }
1304
1305         f = ordered_hashmap_get(j->files, path);
1306         if (f) {
1307                 if (f->last_stat.st_dev == st.st_dev &&
1308                     f->last_stat.st_ino == st.st_ino) {
1309
1310                         /* We already track this file, under the same path and with the same device/inode numbers, it's
1311                          * hence really the same. Mark this file as seen in this generation. This is used to GC old
1312                          * files in process_q_overflow() to detect journal files that are still there and discern them
1313                          * from those which are gone. */
1314
1315                         f->last_seen_generation = j->generation;
1316                         r = 0;
1317                         goto finish;
1318                 }
1319
1320                 /* So we tracked a file under this name, but it has a different inode/device. In that case, it got
1321                  * replaced (probably due to rotation?), let's drop it hence from our list. */
1322                 remove_file_real(j, f);
1323                 f = NULL;
1324         }
1325
1326         if (ordered_hashmap_size(j->files) >= JOURNAL_FILES_MAX) {
1327                 log_debug("Too many open journal files, not adding %s.", path);
1328                 r = -ETOOMANYREFS;
1329                 goto finish;
1330         }
1331
1332         r = journal_file_open(fd, path, O_RDONLY, 0, false, 0, false, NULL, j->mmap, NULL, NULL, &f);
1333         if (r < 0) {
1334                 log_debug_errno(r, "Failed to open journal file %s: %m", path);
1335                 goto finish;
1336         }
1337
1338         /* journal_file_dump(f); */
1339
1340         r = ordered_hashmap_put(j->files, f->path, f);
1341         if (r < 0) {
1342                 f->close_fd = false; /* make sure journal_file_close() doesn't close the caller's fd (or our own). We'll let the caller do that, or ourselves */
1343                 (void) journal_file_close(f);
1344                 goto finish;
1345         }
1346
1347         close_fd = false; /* the fd is now owned by the JournalFile object */
1348
1349         f->last_seen_generation = j->generation;
1350
1351         track_file_disposition(j, f);
1352         check_network(j, f->fd);
1353
1354         j->current_invalidate_counter++;
1355
1356         log_debug("File %s added.", f->path);
1357
1358         r = 0;
1359
1360 finish:
1361         if (close_fd)
1362                 safe_close(fd);
1363
1364         if (r < 0) {
1365                 k = journal_put_error(j, r, path);
1366                 if (k < 0)
1367                         return k;
1368         }
1369
1370         return r;
1371 }
1372
1373 static int add_file_by_name(
1374                 sd_journal *j,
1375                 const char *prefix,
1376                 const char *filename) {
1377
1378         const char *path;
1379
1380         assert(j);
1381         assert(prefix);
1382         assert(filename);
1383
1384         if (j->no_new_files)
1385                 return 0;
1386
1387         if (!file_type_wanted(j->flags, filename))
1388                 return 0;
1389
1390         path = prefix_roota(prefix, filename);
1391         return add_any_file(j, -1, path);
1392 }
1393
1394 static void remove_file_by_name(
1395                 sd_journal *j,
1396                 const char *prefix,
1397                 const char *filename) {
1398
1399         const char *path;
1400         JournalFile *f;
1401
1402         assert(j);
1403         assert(prefix);
1404         assert(filename);
1405
1406         path = prefix_roota(prefix, filename);
1407         f = ordered_hashmap_get(j->files, path);
1408         if (!f)
1409                 return;
1410
1411         remove_file_real(j, f);
1412 }
1413
1414 static void remove_file_real(sd_journal *j, JournalFile *f) {
1415         assert(j);
1416         assert(f);
1417
1418         (void) ordered_hashmap_remove(j->files, f->path);
1419
1420         log_debug("File %s removed.", f->path);
1421
1422         if (j->current_file == f) {
1423                 j->current_file = NULL;
1424                 j->current_field = 0;
1425         }
1426
1427         if (j->unique_file == f) {
1428                 /* Jump to the next unique_file or NULL if that one was last */
1429                 j->unique_file = ordered_hashmap_next(j->files, j->unique_file->path);
1430                 j->unique_offset = 0;
1431                 if (!j->unique_file)
1432                         j->unique_file_lost = true;
1433         }
1434
1435         if (j->fields_file == f) {
1436                 j->fields_file = ordered_hashmap_next(j->files, j->fields_file->path);
1437                 j->fields_offset = 0;
1438                 if (!j->fields_file)
1439                         j->fields_file_lost = true;
1440         }
1441
1442         (void) journal_file_close(f);
1443
1444         j->current_invalidate_counter++;
1445 }
1446
1447 static int dirname_is_machine_id(const char *fn) {
1448         sd_id128_t id, machine;
1449         const char *e;
1450         int r;
1451
1452         /* Returns true if the specified directory name matches the local machine ID */
1453
1454         r = sd_id128_get_machine(&machine);
1455         if (r < 0)
1456                 return r;
1457
1458         e = strchr(fn, '.');
1459         if (e) {
1460                 const char *k;
1461
1462                 /* Looks like it has a namespace suffix. Verify that. */
1463                 if (!log_namespace_name_valid(e + 1))
1464                         return false;
1465
1466                 k = strndupa(fn, e - fn);
1467                 r = sd_id128_from_string(k, &id);
1468         } else
1469                 r = sd_id128_from_string(fn, &id);
1470         if (r < 0)
1471                 return r;
1472
1473         return sd_id128_equal(id, machine);
1474 }
1475
1476 static int dirname_has_namespace(const char *fn, const char *namespace) {
1477         const char *e;
1478
1479         /* Returns true if the specified directory name matches the specified namespace */
1480
1481         e = strchr(fn, '.');
1482         if (e) {
1483                 const char *k;
1484
1485                 if (!namespace)
1486                         return false;
1487
1488                 if (!streq(e + 1, namespace))
1489                         return false;
1490
1491                 k = strndupa(fn, e - fn);
1492                 return id128_is_valid(k);
1493         }
1494
1495         if (namespace)
1496                 return false;
1497
1498         return id128_is_valid(fn);
1499 }
1500
1501 static bool dirent_is_journal_file(const struct dirent *de) {
1502         assert(de);
1503
1504         /* Returns true if the specified directory entry looks like a journal file we might be interested in */
1505
1506         if (!IN_SET(de->d_type, DT_REG, DT_LNK, DT_UNKNOWN))
1507                 return false;
1508
1509         return endswith(de->d_name, ".journal") ||
1510                 endswith(de->d_name, ".journal~");
1511 }
1512
1513 static bool dirent_is_journal_subdir(const struct dirent *de) {
1514         const char *e, *n;
1515         assert(de);
1516
1517         /* returns true if the specified directory entry looks like a directory that might contain journal
1518          * files we might be interested in, i.e. is either a 128bit ID or a 128bit ID suffixed by a
1519          * namespace. */
1520
1521         if (!IN_SET(de->d_type, DT_DIR, DT_LNK, DT_UNKNOWN))
1522                 return false;
1523
1524         e = strchr(de->d_name, '.');
1525         if (!e)
1526                 return id128_is_valid(de->d_name); /* No namespace */
1527
1528         n = strndupa(de->d_name, e - de->d_name);
1529         if (!id128_is_valid(n))
1530                 return false;
1531
1532         return log_namespace_name_valid(e + 1);
1533 }
1534
1535 static int directory_open(sd_journal *j, const char *path, DIR **ret) {
1536         DIR *d;
1537
1538         assert(j);
1539         assert(path);
1540         assert(ret);
1541
1542         if (j->toplevel_fd < 0)
1543                 d = opendir(path);
1544         else
1545                 /* Open the specified directory relative to the toplevel fd. Enforce that the path specified is
1546                  * relative, by dropping the initial slash */
1547                 d = xopendirat(j->toplevel_fd, skip_slash(path), 0);
1548         if (!d)
1549                 return -errno;
1550
1551         *ret = d;
1552         return 0;
1553 }
1554
1555 static int add_directory(sd_journal *j, const char *prefix, const char *dirname);
1556
1557 static void directory_enumerate(sd_journal *j, Directory *m, DIR *d) {
1558         struct dirent *de;
1559
1560         assert(j);
1561         assert(m);
1562         assert(d);
1563
1564         FOREACH_DIRENT_ALL(de, d, goto fail) {
1565
1566                 if (dirent_is_journal_file(de))
1567                         (void) add_file_by_name(j, m->path, de->d_name);
1568
1569                 if (m->is_root && dirent_is_journal_subdir(de))
1570                         (void) add_directory(j, m->path, de->d_name);
1571         }
1572
1573         return;
1574
1575 fail:
1576         log_debug_errno(errno, "Failed to enumerate directory %s, ignoring: %m", m->path);
1577 }
1578
1579 static void directory_watch(sd_journal *j, Directory *m, int fd, uint32_t mask) {
1580         int r;
1581
1582         assert(j);
1583         assert(m);
1584         assert(fd >= 0);
1585
1586         /* Watch this directory if that's enabled and if it not being watched yet. */
1587
1588         if (m->wd > 0) /* Already have a watch? */
1589                 return;
1590         if (j->inotify_fd < 0) /* Not watching at all? */
1591                 return;
1592
1593         m->wd = inotify_add_watch_fd(j->inotify_fd, fd, mask);
1594         if (m->wd < 0) {
1595                 log_debug_errno(errno, "Failed to watch journal directory '%s', ignoring: %m", m->path);
1596                 return;
1597         }
1598
1599         r = hashmap_put(j->directories_by_wd, INT_TO_PTR(m->wd), m);
1600         if (r == -EEXIST)
1601                 log_debug_errno(r, "Directory '%s' already being watched under a different path, ignoring: %m", m->path);
1602         if (r < 0) {
1603                 log_debug_errno(r, "Failed to add watch for journal directory '%s' to hashmap, ignoring: %m", m->path);
1604                 (void) inotify_rm_watch(j->inotify_fd, m->wd);
1605                 m->wd = -1;
1606         }
1607 }
1608
1609 static int add_directory(
1610                 sd_journal *j,
1611                 const char *prefix,
1612                 const char *dirname) {
1613
1614         _cleanup_free_ char *path = NULL;
1615         _cleanup_closedir_ DIR *d = NULL;
1616         Directory *m;
1617         int r, k;
1618
1619         assert(j);
1620         assert(prefix);
1621
1622         /* Adds a journal file directory to watch. If the directory is already tracked this updates the inotify watch
1623          * and reenumerates directory contents */
1624
1625         path = path_join(prefix, dirname);
1626         if (!path) {
1627                 r = -ENOMEM;
1628                 goto fail;
1629         }
1630
1631         log_debug("Considering directory '%s'.", path);
1632
1633         /* We consider everything local that is in a directory for the local machine ID, or that is stored in /run */
1634         if ((j->flags & SD_JOURNAL_LOCAL_ONLY) &&
1635             !((dirname && dirname_is_machine_id(dirname) > 0) || path_has_prefix(j, path, "/run")))
1636                 return 0;
1637
1638         if (dirname &&
1639             (!(FLAGS_SET(j->flags, SD_JOURNAL_ALL_NAMESPACES) ||
1640                dirname_has_namespace(dirname, j->namespace) > 0 ||
1641                (FLAGS_SET(j->flags, SD_JOURNAL_INCLUDE_DEFAULT_NAMESPACE) && dirname_has_namespace(dirname, NULL) > 0))))
1642                 return 0;
1643
1644         r = directory_open(j, path, &d);
1645         if (r < 0) {
1646                 log_debug_errno(r, "Failed to open directory '%s': %m", path);
1647                 goto fail;
1648         }
1649
1650         m = hashmap_get(j->directories_by_path, path);
1651         if (!m) {
1652                 m = new(Directory, 1);
1653                 if (!m) {
1654                         r = -ENOMEM;
1655                         goto fail;
1656                 }
1657
1658                 *m = (Directory) {
1659                         .is_root = false,
1660                         .path = path,
1661                 };
1662
1663                 if (hashmap_put(j->directories_by_path, m->path, m) < 0) {
1664                         free(m);
1665                         r = -ENOMEM;
1666                         goto fail;
1667                 }
1668
1669                 path = NULL; /* avoid freeing in cleanup */
1670                 j->current_invalidate_counter++;
1671
1672                 log_debug("Directory %s added.", m->path);
1673
1674         } else if (m->is_root)
1675                 return 0; /* Don't 'downgrade' from root directory */
1676
1677         m->last_seen_generation = j->generation;
1678
1679         directory_watch(j, m, dirfd(d),
1680                         IN_CREATE|IN_MOVED_TO|IN_MODIFY|IN_ATTRIB|IN_DELETE|
1681                         IN_DELETE_SELF|IN_MOVE_SELF|IN_UNMOUNT|IN_MOVED_FROM|
1682                         IN_ONLYDIR);
1683
1684         if (!j->no_new_files)
1685                 directory_enumerate(j, m, d);
1686
1687         check_network(j, dirfd(d));
1688
1689         return 0;
1690
1691 fail:
1692         k = journal_put_error(j, r, path ?: prefix);
1693         if (k < 0)
1694                 return k;
1695
1696         return r;
1697 }
1698
1699 static int add_root_directory(sd_journal *j, const char *p, bool missing_ok) {
1700
1701         _cleanup_closedir_ DIR *d = NULL;
1702         Directory *m;
1703         int r, k;
1704
1705         assert(j);
1706
1707         /* Adds a root directory to our set of directories to use. If the root directory is already in the set, we
1708          * update the inotify logic, and renumerate the directory entries. This call may hence be called to initially
1709          * populate the set, as well as to update it later. */
1710
1711         if (p) {
1712                 /* If there's a path specified, use it. */
1713
1714                 log_debug("Considering root directory '%s'.", p);
1715
1716                 if ((j->flags & SD_JOURNAL_RUNTIME_ONLY) &&
1717                     !path_has_prefix(j, p, "/run"))
1718                         return -EINVAL;
1719
1720                 if (j->prefix)
1721                         p = strjoina(j->prefix, p);
1722
1723                 r = directory_open(j, p, &d);
1724                 if (r == -ENOENT && missing_ok)
1725                         return 0;
1726                 if (r < 0) {
1727                         log_debug_errno(r, "Failed to open root directory %s: %m", p);
1728                         goto fail;
1729                 }
1730         } else {
1731                 _cleanup_close_ int dfd = -1;
1732
1733                 /* If there's no path specified, then we use the top-level fd itself. We duplicate the fd here, since
1734                  * opendir() will take possession of the fd, and close it, which we don't want. */
1735
1736                 p = "."; /* store this as "." in the directories hashmap */
1737
1738                 dfd = fcntl(j->toplevel_fd, F_DUPFD_CLOEXEC, 3);
1739                 if (dfd < 0) {
1740                         r = -errno;
1741                         goto fail;
1742                 }
1743
1744                 d = take_fdopendir(&dfd);
1745                 if (!d) {
1746                         r = -errno;
1747                         goto fail;
1748                 }
1749
1750                 rewinddir(d);
1751         }
1752
1753         m = hashmap_get(j->directories_by_path, p);
1754         if (!m) {
1755                 m = new0(Directory, 1);
1756                 if (!m) {
1757                         r = -ENOMEM;
1758                         goto fail;
1759                 }
1760
1761                 m->is_root = true;
1762
1763                 m->path = strdup(p);
1764                 if (!m->path) {
1765                         free(m);
1766                         r = -ENOMEM;
1767                         goto fail;
1768                 }
1769
1770                 if (hashmap_put(j->directories_by_path, m->path, m) < 0) {
1771                         free(m->path);
1772                         free(m);
1773                         r = -ENOMEM;
1774                         goto fail;
1775                 }
1776
1777                 j->current_invalidate_counter++;
1778
1779                 log_debug("Root directory %s added.", m->path);
1780
1781         } else if (!m->is_root)
1782                 return 0;
1783
1784         directory_watch(j, m, dirfd(d),
1785                         IN_CREATE|IN_MOVED_TO|IN_MODIFY|IN_ATTRIB|IN_DELETE|
1786                         IN_ONLYDIR);
1787
1788         if (!j->no_new_files)
1789                 directory_enumerate(j, m, d);
1790
1791         check_network(j, dirfd(d));
1792
1793         return 0;
1794
1795 fail:
1796         k = journal_put_error(j, r, p);
1797         if (k < 0)
1798                 return k;
1799
1800         return r;
1801 }
1802
1803 static void remove_directory(sd_journal *j, Directory *d) {
1804         assert(j);
1805
1806         if (d->wd > 0) {
1807                 hashmap_remove(j->directories_by_wd, INT_TO_PTR(d->wd));
1808
1809                 if (j->inotify_fd >= 0)
1810                         (void) inotify_rm_watch(j->inotify_fd, d->wd);
1811         }
1812
1813         hashmap_remove(j->directories_by_path, d->path);
1814
1815         if (d->is_root)
1816                 log_debug("Root directory %s removed.", d->path);
1817         else
1818                 log_debug("Directory %s removed.", d->path);
1819
1820         free(d->path);
1821         free(d);
1822 }
1823
1824 static int add_search_paths(sd_journal *j) {
1825
1826         static const char search_paths[] =
1827                 "/run/log/journal\0"
1828                 "/var/log/journal\0";
1829         const char *p;
1830
1831         assert(j);
1832
1833         /* We ignore most errors here, since the idea is to only open
1834          * what's actually accessible, and ignore the rest. */
1835
1836         NULSTR_FOREACH(p, search_paths)
1837                 (void) add_root_directory(j, p, true);
1838
1839         if (!(j->flags & SD_JOURNAL_LOCAL_ONLY))
1840                 (void) add_root_directory(j, "/var/log/journal/remote", true);
1841
1842         return 0;
1843 }
1844
1845 static int add_current_paths(sd_journal *j) {
1846         Iterator i;
1847         JournalFile *f;
1848
1849         assert(j);
1850         assert(j->no_new_files);
1851
1852         /* Simply adds all directories for files we have open as directories. We don't expect errors here, so we
1853          * treat them as fatal. */
1854
1855         ORDERED_HASHMAP_FOREACH(f, j->files, i) {
1856                 _cleanup_free_ char *dir;
1857                 int r;
1858
1859                 dir = dirname_malloc(f->path);
1860                 if (!dir)
1861                         return -ENOMEM;
1862
1863                 r = add_directory(j, dir, NULL);
1864                 if (r < 0)
1865                         return r;
1866         }
1867
1868         return 0;
1869 }
1870
1871 static int allocate_inotify(sd_journal *j) {
1872         assert(j);
1873
1874         if (j->inotify_fd < 0) {
1875                 j->inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
1876                 if (j->inotify_fd < 0)
1877                         return -errno;
1878         }
1879
1880         return hashmap_ensure_allocated(&j->directories_by_wd, NULL);
1881 }
1882
1883 static sd_journal *journal_new(int flags, const char *path, const char *namespace) {
1884         _cleanup_(sd_journal_closep) sd_journal *j = NULL;
1885
1886         j = new0(sd_journal, 1);
1887         if (!j)
1888                 return NULL;
1889
1890         j->original_pid = getpid_cached();
1891         j->toplevel_fd = -1;
1892         j->inotify_fd = -1;
1893         j->flags = flags;
1894         j->data_threshold = DEFAULT_DATA_THRESHOLD;
1895
1896         if (path) {
1897                 char *t;
1898
1899                 t = strdup(path);
1900                 if (!t)
1901                         return NULL;
1902
1903                 if (flags & SD_JOURNAL_OS_ROOT)
1904                         j->prefix = t;
1905                 else
1906                         j->path = t;
1907         }
1908
1909         if (namespace) {
1910                 j->namespace = strdup(namespace);
1911                 if (!j->namespace)
1912                         return NULL;
1913         }
1914
1915         j->files = ordered_hashmap_new(&path_hash_ops);
1916         if (!j->files)
1917                 return NULL;
1918
1919         j->files_cache = ordered_hashmap_iterated_cache_new(j->files);
1920         j->directories_by_path = hashmap_new(&path_hash_ops);
1921         j->mmap = mmap_cache_new();
1922         if (!j->files_cache || !j->directories_by_path || !j->mmap)
1923                 return NULL;
1924
1925         return TAKE_PTR(j);
1926 }
1927
1928 #define OPEN_ALLOWED_FLAGS                              \
1929         (SD_JOURNAL_LOCAL_ONLY |                        \
1930          SD_JOURNAL_RUNTIME_ONLY |                      \
1931          SD_JOURNAL_SYSTEM |                            \
1932          SD_JOURNAL_CURRENT_USER |                      \
1933          SD_JOURNAL_ALL_NAMESPACES |                    \
1934          SD_JOURNAL_INCLUDE_DEFAULT_NAMESPACE)
1935
1936 _public_ int sd_journal_open_namespace(sd_journal **ret, const char *namespace, int flags) {
1937         _cleanup_(sd_journal_closep) sd_journal *j = NULL;
1938         int r;
1939
1940         assert_return(ret, -EINVAL);
1941         assert_return((flags & ~OPEN_ALLOWED_FLAGS) == 0, -EINVAL);
1942
1943         j = journal_new(flags, NULL, namespace);
1944         if (!j)
1945                 return -ENOMEM;
1946
1947         r = add_search_paths(j);
1948         if (r < 0)
1949                 return r;
1950
1951         *ret = TAKE_PTR(j);
1952         return 0;
1953 }
1954
1955 _public_ int sd_journal_open(sd_journal **ret, int flags) {
1956         return sd_journal_open_namespace(ret, NULL, flags);
1957 }
1958
1959 #define OPEN_CONTAINER_ALLOWED_FLAGS                    \
1960         (SD_JOURNAL_LOCAL_ONLY | SD_JOURNAL_SYSTEM)
1961
1962 _public_ int sd_journal_open_container(sd_journal **ret, const char *machine, int flags) {
1963         _cleanup_free_ char *root = NULL, *class = NULL;
1964         _cleanup_(sd_journal_closep) sd_journal *j = NULL;
1965         char *p;
1966         int r;
1967
1968         /* This is deprecated, people should use machined's OpenMachineRootDirectory() call instead in
1969          * combination with sd_journal_open_directory_fd(). */
1970
1971         assert_return(machine, -EINVAL);
1972         assert_return(ret, -EINVAL);
1973         assert_return((flags & ~OPEN_CONTAINER_ALLOWED_FLAGS) == 0, -EINVAL);
1974         assert_return(machine_name_is_valid(machine), -EINVAL);
1975
1976         p = strjoina("/run/systemd/machines/", machine);
1977         r = parse_env_file(NULL, p,
1978                            "ROOT", &root,
1979                            "CLASS", &class);
1980         if (r == -ENOENT)
1981                 return -EHOSTDOWN;
1982         if (r < 0)
1983                 return r;
1984         if (!root)
1985                 return -ENODATA;
1986
1987         if (!streq_ptr(class, "container"))
1988                 return -EIO;
1989
1990         j = journal_new(flags, root, NULL);
1991         if (!j)
1992                 return -ENOMEM;
1993
1994         r = add_search_paths(j);
1995         if (r < 0)
1996                 return r;
1997
1998         *ret = TAKE_PTR(j);
1999         return 0;
2000 }
2001
2002 #define OPEN_DIRECTORY_ALLOWED_FLAGS                    \
2003         (SD_JOURNAL_OS_ROOT |                           \
2004          SD_JOURNAL_SYSTEM | SD_JOURNAL_CURRENT_USER )
2005
2006 _public_ int sd_journal_open_directory(sd_journal **ret, const char *path, int flags) {
2007         _cleanup_(sd_journal_closep) sd_journal *j = NULL;
2008         int r;
2009
2010         assert_return(ret, -EINVAL);
2011         assert_return(path, -EINVAL);
2012         assert_return((flags & ~OPEN_DIRECTORY_ALLOWED_FLAGS) == 0, -EINVAL);
2013
2014         j = journal_new(flags, path, NULL);
2015         if (!j)
2016                 return -ENOMEM;
2017
2018         if (flags & SD_JOURNAL_OS_ROOT)
2019                 r = add_search_paths(j);
2020         else
2021                 r = add_root_directory(j, path, false);
2022         if (r < 0)
2023                 return r;
2024
2025         *ret = TAKE_PTR(j);
2026         return 0;
2027 }
2028
2029 _public_ int sd_journal_open_files(sd_journal **ret, const char **paths, int flags) {
2030         _cleanup_(sd_journal_closep) sd_journal *j = NULL;
2031         const char **path;
2032         int r;
2033
2034         assert_return(ret, -EINVAL);
2035         assert_return(flags == 0, -EINVAL);
2036
2037         j = journal_new(flags, NULL, NULL);
2038         if (!j)
2039                 return -ENOMEM;
2040
2041         STRV_FOREACH(path, paths) {
2042                 r = add_any_file(j, -1, *path);
2043                 if (r < 0)
2044                         return r;
2045         }
2046
2047         j->no_new_files = true;
2048
2049         *ret = TAKE_PTR(j);
2050         return 0;
2051 }
2052
2053 #define OPEN_DIRECTORY_FD_ALLOWED_FLAGS         \
2054         (SD_JOURNAL_OS_ROOT |                           \
2055          SD_JOURNAL_SYSTEM | SD_JOURNAL_CURRENT_USER )
2056
2057 _public_ int sd_journal_open_directory_fd(sd_journal **ret, int fd, int flags) {
2058         _cleanup_(sd_journal_closep) sd_journal *j = NULL;
2059         struct stat st;
2060         int r;
2061
2062         assert_return(ret, -EINVAL);
2063         assert_return(fd >= 0, -EBADF);
2064         assert_return((flags & ~OPEN_DIRECTORY_FD_ALLOWED_FLAGS) == 0, -EINVAL);
2065
2066         if (fstat(fd, &st) < 0)
2067                 return -errno;
2068
2069         if (!S_ISDIR(st.st_mode))
2070                 return -EBADFD;
2071
2072         j = journal_new(flags, NULL, NULL);
2073         if (!j)
2074                 return -ENOMEM;
2075
2076         j->toplevel_fd = fd;
2077
2078         if (flags & SD_JOURNAL_OS_ROOT)
2079                 r = add_search_paths(j);
2080         else
2081                 r = add_root_directory(j, NULL, false);
2082         if (r < 0)
2083                 return r;
2084
2085         *ret = TAKE_PTR(j);
2086         return 0;
2087 }
2088
2089 _public_ int sd_journal_open_files_fd(sd_journal **ret, int fds[], unsigned n_fds, int flags) {
2090         Iterator iterator;
2091         JournalFile *f;
2092         _cleanup_(sd_journal_closep) sd_journal *j = NULL;
2093         unsigned i;
2094         int r;
2095
2096         assert_return(ret, -EINVAL);
2097         assert_return(n_fds > 0, -EBADF);
2098         assert_return(flags == 0, -EINVAL);
2099
2100         j = journal_new(flags, NULL, NULL);
2101         if (!j)
2102                 return -ENOMEM;
2103
2104         for (i = 0; i < n_fds; i++) {
2105                 struct stat st;
2106
2107                 if (fds[i] < 0) {
2108                         r = -EBADF;
2109                         goto fail;
2110                 }
2111
2112                 if (fstat(fds[i], &st) < 0) {
2113                         r = -errno;
2114                         goto fail;
2115                 }
2116
2117                 r = stat_verify_regular(&st);
2118                 if (r < 0)
2119                         goto fail;
2120
2121                 r = add_any_file(j, fds[i], NULL);
2122                 if (r < 0)
2123                         goto fail;
2124         }
2125
2126         j->no_new_files = true;
2127         j->no_inotify = true;
2128
2129         *ret = TAKE_PTR(j);
2130         return 0;
2131
2132 fail:
2133         /* If we fail, make sure we don't take possession of the files we managed to make use of successfully, and they
2134          * remain open */
2135         ORDERED_HASHMAP_FOREACH(f, j->files, iterator)
2136                 f->close_fd = false;
2137
2138         return r;
2139 }
2140
2141 _public_ void sd_journal_close(sd_journal *j) {
2142         Directory *d;
2143
2144         if (!j)
2145                 return;
2146
2147         sd_journal_flush_matches(j);
2148
2149         ordered_hashmap_free_with_destructor(j->files, journal_file_close);
2150         iterated_cache_free(j->files_cache);
2151
2152         while ((d = hashmap_first(j->directories_by_path)))
2153                 remove_directory(j, d);
2154
2155         while ((d = hashmap_first(j->directories_by_wd)))
2156                 remove_directory(j, d);
2157
2158         hashmap_free(j->directories_by_path);
2159         hashmap_free(j->directories_by_wd);
2160
2161         safe_close(j->inotify_fd);
2162
2163         if (j->mmap) {
2164                 log_debug("mmap cache statistics: %u hit, %u miss", mmap_cache_get_hit(j->mmap), mmap_cache_get_missed(j->mmap));
2165                 mmap_cache_unref(j->mmap);
2166         }
2167
2168         hashmap_free_free(j->errors);
2169
2170         free(j->path);
2171         free(j->prefix);
2172         free(j->namespace);
2173         free(j->unique_field);
2174         free(j->fields_buffer);
2175         free(j);
2176 }
2177
2178 _public_ int sd_journal_get_realtime_usec(sd_journal *j, uint64_t *ret) {
2179         Object *o;
2180         JournalFile *f;
2181         int r;
2182
2183         assert_return(j, -EINVAL);
2184         assert_return(!journal_pid_changed(j), -ECHILD);
2185         assert_return(ret, -EINVAL);
2186
2187         f = j->current_file;
2188         if (!f)
2189                 return -EADDRNOTAVAIL;
2190
2191         if (f->current_offset <= 0)
2192                 return -EADDRNOTAVAIL;
2193
2194         r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &o);
2195         if (r < 0)
2196                 return r;
2197
2198         *ret = le64toh(o->entry.realtime);
2199         return 0;
2200 }
2201
2202 _public_ int sd_journal_get_monotonic_usec(sd_journal *j, uint64_t *ret, sd_id128_t *ret_boot_id) {
2203         Object *o;
2204         JournalFile *f;
2205         int r;
2206         sd_id128_t id;
2207
2208         assert_return(j, -EINVAL);
2209         assert_return(!journal_pid_changed(j), -ECHILD);
2210
2211         f = j->current_file;
2212         if (!f)
2213                 return -EADDRNOTAVAIL;
2214
2215         if (f->current_offset <= 0)
2216                 return -EADDRNOTAVAIL;
2217
2218         r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &o);
2219         if (r < 0)
2220                 return r;
2221
2222         if (ret_boot_id)
2223                 *ret_boot_id = o->entry.boot_id;
2224         else {
2225                 r = sd_id128_get_boot(&id);
2226                 if (r < 0)
2227                         return r;
2228
2229                 if (!sd_id128_equal(id, o->entry.boot_id))
2230                         return -ESTALE;
2231         }
2232
2233         if (ret)
2234                 *ret = le64toh(o->entry.monotonic);
2235
2236         return 0;
2237 }
2238
2239 static bool field_is_valid(const char *field) {
2240         const char *p;
2241
2242         assert(field);
2243
2244         if (isempty(field))
2245                 return false;
2246
2247         if (startswith(field, "__"))
2248                 return false;
2249
2250         for (p = field; *p; p++) {
2251
2252                 if (*p == '_')
2253                         continue;
2254
2255                 if (*p >= 'A' && *p <= 'Z')
2256                         continue;
2257
2258                 if (*p >= '0' && *p <= '9')
2259                         continue;
2260
2261                 return false;
2262         }
2263
2264         return true;
2265 }
2266
2267 _public_ int sd_journal_get_data(sd_journal *j, const char *field, const void **data, size_t *size) {
2268         JournalFile *f;
2269         uint64_t i, n;
2270         size_t field_length;
2271         int r;
2272         Object *o;
2273
2274         assert_return(j, -EINVAL);
2275         assert_return(!journal_pid_changed(j), -ECHILD);
2276         assert_return(field, -EINVAL);
2277         assert_return(data, -EINVAL);
2278         assert_return(size, -EINVAL);
2279         assert_return(field_is_valid(field), -EINVAL);
2280
2281         f = j->current_file;
2282         if (!f)
2283                 return -EADDRNOTAVAIL;
2284
2285         if (f->current_offset <= 0)
2286                 return -EADDRNOTAVAIL;
2287
2288         r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &o);
2289         if (r < 0)
2290                 return r;
2291
2292         field_length = strlen(field);
2293
2294         n = journal_file_entry_n_items(o);
2295         for (i = 0; i < n; i++) {
2296                 uint64_t p, l;
2297                 le64_t le_hash;
2298                 size_t t;
2299                 int compression;
2300
2301                 p = le64toh(o->entry.items[i].object_offset);
2302                 le_hash = o->entry.items[i].hash;
2303                 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
2304                 if (r < 0)
2305                         return r;
2306
2307                 if (le_hash != o->data.hash)
2308                         return -EBADMSG;
2309
2310                 l = le64toh(o->object.size) - offsetof(Object, data.payload);
2311
2312                 compression = o->object.flags & OBJECT_COMPRESSION_MASK;
2313                 if (compression) {
2314 #if HAVE_XZ || HAVE_LZ4
2315                         r = decompress_startswith(compression,
2316                                                   o->data.payload, l,
2317                                                   &f->compress_buffer, &f->compress_buffer_size,
2318                                                   field, field_length, '=');
2319                         if (r < 0)
2320                                 log_debug_errno(r, "Cannot decompress %s object of length %"PRIu64" at offset "OFSfmt": %m",
2321                                                 object_compressed_to_string(compression), l, p);
2322                         else if (r > 0) {
2323
2324                                 size_t rsize;
2325
2326                                 r = decompress_blob(compression,
2327                                                     o->data.payload, l,
2328                                                     &f->compress_buffer, &f->compress_buffer_size, &rsize,
2329                                                     j->data_threshold);
2330                                 if (r < 0)
2331                                         return r;
2332
2333                                 *data = f->compress_buffer;
2334                                 *size = (size_t) rsize;
2335
2336                                 return 0;
2337                         }
2338 #else
2339                         return -EPROTONOSUPPORT;
2340 #endif
2341                 } else if (l >= field_length+1 &&
2342                            memcmp(o->data.payload, field, field_length) == 0 &&
2343                            o->data.payload[field_length] == '=') {
2344
2345                         t = (size_t) l;
2346
2347                         if ((uint64_t) t != l)
2348                                 return -E2BIG;
2349
2350                         *data = o->data.payload;
2351                         *size = t;
2352
2353                         return 0;
2354                 }
2355
2356                 r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &o);
2357                 if (r < 0)
2358                         return r;
2359         }
2360
2361         return -ENOENT;
2362 }
2363
2364 static int return_data(sd_journal *j, JournalFile *f, Object *o, const void **data, size_t *size) {
2365         size_t t;
2366         uint64_t l;
2367         int compression;
2368
2369         l = le64toh(READ_NOW(o->object.size));
2370         if (l < offsetof(Object, data.payload))
2371                 return -EBADMSG;
2372         l -= offsetof(Object, data.payload);
2373         t = (size_t) l;
2374
2375         /* We can't read objects larger than 4G on a 32bit machine */
2376         if ((uint64_t) t != l)
2377                 return -E2BIG;
2378
2379         compression = o->object.flags & OBJECT_COMPRESSION_MASK;
2380         if (compression) {
2381 #if HAVE_XZ || HAVE_LZ4
2382                 size_t rsize;
2383                 int r;
2384
2385                 r = decompress_blob(compression,
2386                                     o->data.payload, l, &f->compress_buffer,
2387                                     &f->compress_buffer_size, &rsize, j->data_threshold);
2388                 if (r < 0)
2389                         return r;
2390
2391                 *data = f->compress_buffer;
2392                 *size = (size_t) rsize;
2393 #else
2394                 return -EPROTONOSUPPORT;
2395 #endif
2396         } else {
2397                 *data = o->data.payload;
2398                 *size = t;
2399         }
2400
2401         return 0;
2402 }
2403
2404 _public_ int sd_journal_enumerate_data(sd_journal *j, const void **data, size_t *size) {
2405         JournalFile *f;
2406         uint64_t p, n;
2407         le64_t le_hash;
2408         int r;
2409         Object *o;
2410
2411         assert_return(j, -EINVAL);
2412         assert_return(!journal_pid_changed(j), -ECHILD);
2413         assert_return(data, -EINVAL);
2414         assert_return(size, -EINVAL);
2415
2416         f = j->current_file;
2417         if (!f)
2418                 return -EADDRNOTAVAIL;
2419
2420         if (f->current_offset <= 0)
2421                 return -EADDRNOTAVAIL;
2422
2423         r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &o);
2424         if (r < 0)
2425                 return r;
2426
2427         n = journal_file_entry_n_items(o);
2428         if (j->current_field >= n)
2429                 return 0;
2430
2431         p = le64toh(o->entry.items[j->current_field].object_offset);
2432         le_hash = o->entry.items[j->current_field].hash;
2433         r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
2434         if (r < 0)
2435                 return r;
2436
2437         if (le_hash != o->data.hash)
2438                 return -EBADMSG;
2439
2440         r = return_data(j, f, o, data, size);
2441         if (r < 0)
2442                 return r;
2443
2444         j->current_field++;
2445
2446         return 1;
2447 }
2448
2449 _public_ void sd_journal_restart_data(sd_journal *j) {
2450         if (!j)
2451                 return;
2452
2453         j->current_field = 0;
2454 }
2455
2456 static int reiterate_all_paths(sd_journal *j) {
2457         assert(j);
2458
2459         if (j->no_new_files)
2460                 return add_current_paths(j);
2461
2462         if (j->flags & SD_JOURNAL_OS_ROOT)
2463                 return add_search_paths(j);
2464
2465         if (j->toplevel_fd >= 0)
2466                 return add_root_directory(j, NULL, false);
2467
2468         if (j->path)
2469                 return add_root_directory(j, j->path, true);
2470
2471         return add_search_paths(j);
2472 }
2473
2474 _public_ int sd_journal_get_fd(sd_journal *j) {
2475         int r;
2476
2477         assert_return(j, -EINVAL);
2478         assert_return(!journal_pid_changed(j), -ECHILD);
2479
2480         if (j->no_inotify)
2481                 return -EMEDIUMTYPE;
2482
2483         if (j->inotify_fd >= 0)
2484                 return j->inotify_fd;
2485
2486         r = allocate_inotify(j);
2487         if (r < 0)
2488                 return r;
2489
2490         log_debug("Reiterating files to get inotify watches established.");
2491
2492         /* Iterate through all dirs again, to add them to the inotify */
2493         r = reiterate_all_paths(j);
2494         if (r < 0)
2495                 return r;
2496
2497         return j->inotify_fd;
2498 }
2499
2500 _public_ int sd_journal_get_events(sd_journal *j) {
2501         int fd;
2502
2503         assert_return(j, -EINVAL);
2504         assert_return(!journal_pid_changed(j), -ECHILD);
2505
2506         fd = sd_journal_get_fd(j);
2507         if (fd < 0)
2508                 return fd;
2509
2510         return POLLIN;
2511 }
2512
2513 _public_ int sd_journal_get_timeout(sd_journal *j, uint64_t *timeout_usec) {
2514         int fd;
2515
2516         assert_return(j, -EINVAL);
2517         assert_return(!journal_pid_changed(j), -ECHILD);
2518         assert_return(timeout_usec, -EINVAL);
2519
2520         fd = sd_journal_get_fd(j);
2521         if (fd < 0)
2522                 return fd;
2523
2524         if (!j->on_network) {
2525                 *timeout_usec = (uint64_t) -1;
2526                 return 0;
2527         }
2528
2529         /* If we are on the network we need to regularly check for
2530          * changes manually */
2531
2532         *timeout_usec = j->last_process_usec + JOURNAL_FILES_RECHECK_USEC;
2533         return 1;
2534 }
2535
2536 static void process_q_overflow(sd_journal *j) {
2537         JournalFile *f;
2538         Directory *m;
2539         Iterator i;
2540
2541         assert(j);
2542
2543         /* When the inotify queue overruns we need to enumerate and re-validate all journal files to bring our list
2544          * back in sync with what's on disk. For this we pick a new generation counter value. It'll be assigned to all
2545          * journal files we encounter. All journal files and all directories that don't carry it after reenumeration
2546          * are subject for unloading. */
2547
2548         log_debug("Inotify queue overrun, reiterating everything.");
2549
2550         j->generation++;
2551         (void) reiterate_all_paths(j);
2552
2553         ORDERED_HASHMAP_FOREACH(f, j->files, i) {
2554
2555                 if (f->last_seen_generation == j->generation)
2556                         continue;
2557
2558                 log_debug("File '%s' hasn't been seen in this enumeration, removing.", f->path);
2559                 remove_file_real(j, f);
2560         }
2561
2562         HASHMAP_FOREACH(m, j->directories_by_path, i) {
2563
2564                 if (m->last_seen_generation == j->generation)
2565                         continue;
2566
2567                 if (m->is_root) /* Never GC root directories */
2568                         continue;
2569
2570                 log_debug("Directory '%s' hasn't been seen in this enumeration, removing.", f->path);
2571                 remove_directory(j, m);
2572         }
2573
2574         log_debug("Reiteration complete.");
2575 }
2576
2577 static void process_inotify_event(sd_journal *j, const struct inotify_event *e) {
2578         Directory *d;
2579
2580         assert(j);
2581         assert(e);
2582
2583         if (e->mask & IN_Q_OVERFLOW) {
2584                 process_q_overflow(j);
2585                 return;
2586         }
2587
2588         /* Is this a subdirectory we watch? */
2589         d = hashmap_get(j->directories_by_wd, INT_TO_PTR(e->wd));
2590         if (d) {
2591                 if (!(e->mask & IN_ISDIR) && e->len > 0 &&
2592                     (endswith(e->name, ".journal") ||
2593                      endswith(e->name, ".journal~"))) {
2594
2595                         /* Event for a journal file */
2596
2597                         if (e->mask & (IN_CREATE|IN_MOVED_TO|IN_MODIFY|IN_ATTRIB))
2598                                 (void) add_file_by_name(j, d->path, e->name);
2599                         else if (e->mask & (IN_DELETE|IN_MOVED_FROM|IN_UNMOUNT))
2600                                 remove_file_by_name(j, d->path, e->name);
2601
2602                 } else if (!d->is_root && e->len == 0) {
2603
2604                         /* Event for a subdirectory */
2605
2606                         if (e->mask & (IN_DELETE_SELF|IN_MOVE_SELF|IN_UNMOUNT))
2607                                 remove_directory(j, d);
2608
2609                 } else if (d->is_root && (e->mask & IN_ISDIR) && e->len > 0 && id128_is_valid(e->name)) {
2610
2611                         /* Event for root directory */
2612
2613                         if (e->mask & (IN_CREATE|IN_MOVED_TO|IN_MODIFY|IN_ATTRIB))
2614                                 (void) add_directory(j, d->path, e->name);
2615                 }
2616
2617                 return;
2618         }
2619
2620         if (e->mask & IN_IGNORED)
2621                 return;
2622
2623         log_debug("Unexpected inotify event.");
2624 }
2625
2626 static int determine_change(sd_journal *j) {
2627         bool b;
2628
2629         assert(j);
2630
2631         b = j->current_invalidate_counter != j->last_invalidate_counter;
2632         j->last_invalidate_counter = j->current_invalidate_counter;
2633
2634         return b ? SD_JOURNAL_INVALIDATE : SD_JOURNAL_APPEND;
2635 }
2636
2637 _public_ int sd_journal_process(sd_journal *j) {
2638         bool got_something = false;
2639
2640         assert_return(j, -EINVAL);
2641         assert_return(!journal_pid_changed(j), -ECHILD);
2642
2643         if (j->inotify_fd < 0) /* We have no inotify fd yet? Then there's noting to process. */
2644                 return 0;
2645
2646         j->last_process_usec = now(CLOCK_MONOTONIC);
2647         j->last_invalidate_counter = j->current_invalidate_counter;
2648
2649         for (;;) {
2650                 union inotify_event_buffer buffer;
2651                 struct inotify_event *e;
2652                 ssize_t l;
2653
2654                 l = read(j->inotify_fd, &buffer, sizeof(buffer));
2655                 if (l < 0) {
2656                         if (IN_SET(errno, EAGAIN, EINTR))
2657                                 return got_something ? determine_change(j) : SD_JOURNAL_NOP;
2658
2659                         return -errno;
2660                 }
2661
2662                 got_something = true;
2663
2664                 FOREACH_INOTIFY_EVENT(e, buffer, l)
2665                         process_inotify_event(j, e);
2666         }
2667 }
2668
2669 _public_ int sd_journal_wait(sd_journal *j, uint64_t timeout_usec) {
2670         int r;
2671         uint64_t t;
2672
2673         assert_return(j, -EINVAL);
2674         assert_return(!journal_pid_changed(j), -ECHILD);
2675
2676         if (j->inotify_fd < 0) {
2677                 Iterator i;
2678                 JournalFile *f;
2679
2680                 /* This is the first invocation, hence create the
2681                  * inotify watch */
2682                 r = sd_journal_get_fd(j);
2683                 if (r < 0)
2684                         return r;
2685
2686                 /* Server might have done some vacuuming while we weren't watching.
2687                    Get rid of the deleted files now so they don't stay around indefinitely. */
2688                 ORDERED_HASHMAP_FOREACH(f, j->files, i) {
2689                         r = journal_file_fstat(f);
2690                         if (r == -EIDRM)
2691                                 remove_file_real(j, f);
2692                         else if (r < 0) {
2693                                 log_debug_errno(r,"Failed to fstat() journal file '%s' : %m", f->path);
2694                                 continue;
2695                         }
2696                 }
2697
2698                 /* The journal might have changed since the context
2699                  * object was created and we weren't watching before,
2700                  * hence don't wait for anything, and return
2701                  * immediately. */
2702                 return determine_change(j);
2703         }
2704
2705         r = sd_journal_get_timeout(j, &t);
2706         if (r < 0)
2707                 return r;
2708
2709         if (t != (uint64_t) -1) {
2710                 usec_t n;
2711
2712                 n = now(CLOCK_MONOTONIC);
2713                 t = t > n ? t - n : 0;
2714
2715                 if (timeout_usec == (uint64_t) -1 || timeout_usec > t)
2716                         timeout_usec = t;
2717         }
2718
2719         do {
2720                 r = fd_wait_for_event(j->inotify_fd, POLLIN, timeout_usec);
2721         } while (r == -EINTR);
2722
2723         if (r < 0)
2724                 return r;
2725
2726         return sd_journal_process(j);
2727 }
2728
2729 _public_ int sd_journal_get_cutoff_realtime_usec(sd_journal *j, uint64_t *from, uint64_t *to) {
2730         Iterator i;
2731         JournalFile *f;
2732         bool first = true;
2733         uint64_t fmin = 0, tmax = 0;
2734         int r;
2735
2736         assert_return(j, -EINVAL);
2737         assert_return(!journal_pid_changed(j), -ECHILD);
2738         assert_return(from || to, -EINVAL);
2739         assert_return(from != to, -EINVAL);
2740
2741         ORDERED_HASHMAP_FOREACH(f, j->files, i) {
2742                 usec_t fr, t;
2743
2744                 r = journal_file_get_cutoff_realtime_usec(f, &fr, &t);
2745                 if (r == -ENOENT)
2746                         continue;
2747                 if (r < 0)
2748                         return r;
2749                 if (r == 0)
2750                         continue;
2751
2752                 if (first) {
2753                         fmin = fr;
2754                         tmax = t;
2755                         first = false;
2756                 } else {
2757                         fmin = MIN(fr, fmin);
2758                         tmax = MAX(t, tmax);
2759                 }
2760         }
2761
2762         if (from)
2763                 *from = fmin;
2764         if (to)
2765                 *to = tmax;
2766
2767         return first ? 0 : 1;
2768 }
2769
2770 _public_ int sd_journal_get_cutoff_monotonic_usec(sd_journal *j, sd_id128_t boot_id, uint64_t *from, uint64_t *to) {
2771         Iterator i;
2772         JournalFile *f;
2773         bool found = false;
2774         int r;
2775
2776         assert_return(j, -EINVAL);
2777         assert_return(!journal_pid_changed(j), -ECHILD);
2778         assert_return(from || to, -EINVAL);
2779         assert_return(from != to, -EINVAL);
2780
2781         ORDERED_HASHMAP_FOREACH(f, j->files, i) {
2782                 usec_t fr, t;
2783
2784                 r = journal_file_get_cutoff_monotonic_usec(f, boot_id, &fr, &t);
2785                 if (r == -ENOENT)
2786                         continue;
2787                 if (r < 0)
2788                         return r;
2789                 if (r == 0)
2790                         continue;
2791
2792                 if (found) {
2793                         if (from)
2794                                 *from = MIN(fr, *from);
2795                         if (to)
2796                                 *to = MAX(t, *to);
2797                 } else {
2798                         if (from)
2799                                 *from = fr;
2800                         if (to)
2801                                 *to = t;
2802                         found = true;
2803                 }
2804         }
2805
2806         return found;
2807 }
2808
2809 void journal_print_header(sd_journal *j) {
2810         Iterator i;
2811         JournalFile *f;
2812         bool newline = false;
2813
2814         assert(j);
2815
2816         ORDERED_HASHMAP_FOREACH(f, j->files, i) {
2817                 if (newline)
2818                         putchar('\n');
2819                 else
2820                         newline = true;
2821
2822                 journal_file_print_header(f);
2823         }
2824 }
2825
2826 _public_ int sd_journal_get_usage(sd_journal *j, uint64_t *bytes) {
2827         Iterator i;
2828         JournalFile *f;
2829         uint64_t sum = 0;
2830
2831         assert_return(j, -EINVAL);
2832         assert_return(!journal_pid_changed(j), -ECHILD);
2833         assert_return(bytes, -EINVAL);
2834
2835         ORDERED_HASHMAP_FOREACH(f, j->files, i) {
2836                 struct stat st;
2837
2838                 if (fstat(f->fd, &st) < 0)
2839                         return -errno;
2840
2841                 sum += (uint64_t) st.st_blocks * 512ULL;
2842         }
2843
2844         *bytes = sum;
2845         return 0;
2846 }
2847
2848 _public_ int sd_journal_query_unique(sd_journal *j, const char *field) {
2849         char *f;
2850
2851         assert_return(j, -EINVAL);
2852         assert_return(!journal_pid_changed(j), -ECHILD);
2853         assert_return(!isempty(field), -EINVAL);
2854         assert_return(field_is_valid(field), -EINVAL);
2855
2856         f = strdup(field);
2857         if (!f)
2858                 return -ENOMEM;
2859
2860         free(j->unique_field);
2861         j->unique_field = f;
2862         j->unique_file = NULL;
2863         j->unique_offset = 0;
2864         j->unique_file_lost = false;
2865
2866         return 0;
2867 }
2868
2869 _public_ int sd_journal_enumerate_unique(sd_journal *j, const void **data, size_t *l) {
2870         size_t k;
2871
2872         assert_return(j, -EINVAL);
2873         assert_return(!journal_pid_changed(j), -ECHILD);
2874         assert_return(data, -EINVAL);
2875         assert_return(l, -EINVAL);
2876         assert_return(j->unique_field, -EINVAL);
2877
2878         k = strlen(j->unique_field);
2879
2880         if (!j->unique_file) {
2881                 if (j->unique_file_lost)
2882                         return 0;
2883
2884                 j->unique_file = ordered_hashmap_first(j->files);
2885                 if (!j->unique_file)
2886                         return 0;
2887
2888                 j->unique_offset = 0;
2889         }
2890
2891         for (;;) {
2892                 JournalFile *of;
2893                 Iterator i;
2894                 Object *o;
2895                 const void *odata;
2896                 size_t ol;
2897                 bool found;
2898                 int r;
2899
2900                 /* Proceed to next data object in the field's linked list */
2901                 if (j->unique_offset == 0) {
2902                         r = journal_file_find_field_object(j->unique_file, j->unique_field, k, &o, NULL);
2903                         if (r < 0)
2904                                 return r;
2905
2906                         j->unique_offset = r > 0 ? le64toh(o->field.head_data_offset) : 0;
2907                 } else {
2908                         r = journal_file_move_to_object(j->unique_file, OBJECT_DATA, j->unique_offset, &o);
2909                         if (r < 0)
2910                                 return r;
2911
2912                         j->unique_offset = le64toh(o->data.next_field_offset);
2913                 }
2914
2915                 /* We reached the end of the list? Then start again, with the next file */
2916                 if (j->unique_offset == 0) {
2917                         j->unique_file = ordered_hashmap_next(j->files, j->unique_file->path);
2918                         if (!j->unique_file)
2919                                 return 0;
2920
2921                         continue;
2922                 }
2923
2924                 /* We do not use OBJECT_DATA context here, but OBJECT_UNUSED
2925                  * instead, so that we can look at this data object at the same
2926                  * time as one on another file */
2927                 r = journal_file_move_to_object(j->unique_file, OBJECT_UNUSED, j->unique_offset, &o);
2928                 if (r < 0)
2929                         return r;
2930
2931                 /* Let's do the type check by hand, since we used 0 context above. */
2932                 if (o->object.type != OBJECT_DATA)
2933                         return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
2934                                                "%s:offset " OFSfmt ": object has type %d, expected %d",
2935                                                j->unique_file->path,
2936                                                j->unique_offset,
2937                                                o->object.type, OBJECT_DATA);
2938
2939                 r = return_data(j, j->unique_file, o, &odata, &ol);
2940                 if (r < 0)
2941                         return r;
2942
2943                 /* Check if we have at least the field name and "=". */
2944                 if (ol <= k)
2945                         return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
2946                                                "%s:offset " OFSfmt ": object has size %zu, expected at least %zu",
2947                                                j->unique_file->path,
2948                                                j->unique_offset, ol, k + 1);
2949
2950                 if (memcmp(odata, j->unique_field, k) || ((const char*) odata)[k] != '=')
2951                         return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
2952                                                "%s:offset " OFSfmt ": object does not start with \"%s=\"",
2953                                                j->unique_file->path,
2954                                                j->unique_offset,
2955                                                j->unique_field);
2956
2957                 /* OK, now let's see if we already returned this data
2958                  * object by checking if it exists in the earlier
2959                  * traversed files. */
2960                 found = false;
2961                 ORDERED_HASHMAP_FOREACH(of, j->files, i) {
2962                         if (of == j->unique_file)
2963                                 break;
2964
2965                         /* Skip this file it didn't have any fields indexed */
2966                         if (JOURNAL_HEADER_CONTAINS(of->header, n_fields) && le64toh(of->header->n_fields) <= 0)
2967                                 continue;
2968
2969                         r = journal_file_find_data_object_with_hash(of, odata, ol, le64toh(o->data.hash), NULL, NULL);
2970                         if (r < 0)
2971                                 return r;
2972                         if (r > 0) {
2973                                 found = true;
2974                                 break;
2975                         }
2976                 }
2977
2978                 if (found)
2979                         continue;
2980
2981                 r = return_data(j, j->unique_file, o, data, l);
2982                 if (r < 0)
2983                         return r;
2984
2985                 return 1;
2986         }
2987 }
2988
2989 _public_ void sd_journal_restart_unique(sd_journal *j) {
2990         if (!j)
2991                 return;
2992
2993         j->unique_file = NULL;
2994         j->unique_offset = 0;
2995         j->unique_file_lost = false;
2996 }
2997
2998 _public_ int sd_journal_enumerate_fields(sd_journal *j, const char **field) {
2999         int r;
3000
3001         assert_return(j, -EINVAL);
3002         assert_return(!journal_pid_changed(j), -ECHILD);
3003         assert_return(field, -EINVAL);
3004
3005         if (!j->fields_file) {
3006                 if (j->fields_file_lost)
3007                         return 0;
3008
3009                 j->fields_file = ordered_hashmap_first(j->files);
3010                 if (!j->fields_file)
3011                         return 0;
3012
3013                 j->fields_hash_table_index = 0;
3014                 j->fields_offset = 0;
3015         }
3016
3017         for (;;) {
3018                 JournalFile *f, *of;
3019                 Iterator i;
3020                 uint64_t m;
3021                 Object *o;
3022                 size_t sz;
3023                 bool found;
3024
3025                 f = j->fields_file;
3026
3027                 if (j->fields_offset == 0) {
3028                         bool eof = false;
3029
3030                         /* We are not yet positioned at any field. Let's pick the first one */
3031                         r = journal_file_map_field_hash_table(f);
3032                         if (r < 0)
3033                                 return r;
3034
3035                         m = le64toh(f->header->field_hash_table_size) / sizeof(HashItem);
3036                         for (;;) {
3037                                 if (j->fields_hash_table_index >= m) {
3038                                         /* Reached the end of the hash table, go to the next file. */
3039                                         eof = true;
3040                                         break;
3041                                 }
3042
3043                                 j->fields_offset = le64toh(f->field_hash_table[j->fields_hash_table_index].head_hash_offset);
3044
3045                                 if (j->fields_offset != 0)
3046                                         break;
3047
3048                                 /* Empty hash table bucket, go to next one */
3049                                 j->fields_hash_table_index++;
3050                         }
3051
3052                         if (eof) {
3053                                 /* Proceed with next file */
3054                                 j->fields_file = ordered_hashmap_next(j->files, f->path);
3055                                 if (!j->fields_file) {
3056                                         *field = NULL;
3057                                         return 0;
3058                                 }
3059
3060                                 j->fields_offset = 0;
3061                                 j->fields_hash_table_index = 0;
3062                                 continue;
3063                         }
3064
3065                 } else {
3066                         /* We are already positioned at a field. If so, let's figure out the next field from it */
3067
3068                         r = journal_file_move_to_object(f, OBJECT_FIELD, j->fields_offset, &o);
3069                         if (r < 0)
3070                                 return r;
3071
3072                         j->fields_offset = le64toh(o->field.next_hash_offset);
3073                         if (j->fields_offset == 0) {
3074                                 /* Reached the end of the hash table chain */
3075                                 j->fields_hash_table_index++;
3076                                 continue;
3077                         }
3078                 }
3079
3080                 /* We use OBJECT_UNUSED here, so that the iterator below doesn't remove our mmap window */
3081                 r = journal_file_move_to_object(f, OBJECT_UNUSED, j->fields_offset, &o);
3082                 if (r < 0)
3083                         return r;
3084
3085                 /* Because we used OBJECT_UNUSED above, we need to do our type check manually */
3086                 if (o->object.type != OBJECT_FIELD)
3087                         return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
3088                                                "%s:offset " OFSfmt ": object has type %i, expected %i",
3089                                                f->path, j->fields_offset,
3090                                                o->object.type, OBJECT_FIELD);
3091
3092                 sz = le64toh(o->object.size) - offsetof(Object, field.payload);
3093
3094                 /* Let's see if we already returned this field name before. */
3095                 found = false;
3096                 ORDERED_HASHMAP_FOREACH(of, j->files, i) {
3097                         if (of == f)
3098                                 break;
3099
3100                         /* Skip this file it didn't have any fields indexed */
3101                         if (JOURNAL_HEADER_CONTAINS(of->header, n_fields) && le64toh(of->header->n_fields) <= 0)
3102                                 continue;
3103
3104                         r = journal_file_find_field_object_with_hash(of, o->field.payload, sz, le64toh(o->field.hash), NULL, NULL);
3105                         if (r < 0)
3106                                 return r;
3107                         if (r > 0) {
3108                                 found = true;
3109                                 break;
3110                         }
3111                 }
3112
3113                 if (found)
3114                         continue;
3115
3116                 /* Check if this is really a valid string containing no NUL byte */
3117                 if (memchr(o->field.payload, 0, sz))
3118                         return -EBADMSG;
3119
3120                 if (sz > j->data_threshold)
3121                         sz = j->data_threshold;
3122
3123                 if (!GREEDY_REALLOC(j->fields_buffer, j->fields_buffer_allocated, sz + 1))
3124                         return -ENOMEM;
3125
3126                 memcpy(j->fields_buffer, o->field.payload, sz);
3127                 j->fields_buffer[sz] = 0;
3128
3129                 if (!field_is_valid(j->fields_buffer))
3130                         return -EBADMSG;
3131
3132                 *field = j->fields_buffer;
3133                 return 1;
3134         }
3135 }
3136
3137 _public_ void sd_journal_restart_fields(sd_journal *j) {
3138         if (!j)
3139                 return;
3140
3141         j->fields_file = NULL;
3142         j->fields_hash_table_index = 0;
3143         j->fields_offset = 0;
3144         j->fields_file_lost = false;
3145 }
3146
3147 _public_ int sd_journal_reliable_fd(sd_journal *j) {
3148         assert_return(j, -EINVAL);
3149         assert_return(!journal_pid_changed(j), -ECHILD);
3150
3151         return !j->on_network;
3152 }
3153
3154 static char *lookup_field(const char *field, void *userdata) {
3155         sd_journal *j = userdata;
3156         const void *data;
3157         size_t size, d;
3158         int r;
3159
3160         assert(field);
3161         assert(j);
3162
3163         r = sd_journal_get_data(j, field, &data, &size);
3164         if (r < 0 ||
3165             size > REPLACE_VAR_MAX)
3166                 return strdup(field);
3167
3168         d = strlen(field) + 1;
3169
3170         return strndup((const char*) data + d, size - d);
3171 }
3172
3173 _public_ int sd_journal_get_catalog(sd_journal *j, char **ret) {
3174         const void *data;
3175         size_t size;
3176         sd_id128_t id;
3177         _cleanup_free_ char *text = NULL, *cid = NULL;
3178         char *t;
3179         int r;
3180
3181         assert_return(j, -EINVAL);
3182         assert_return(!journal_pid_changed(j), -ECHILD);
3183         assert_return(ret, -EINVAL);
3184
3185         r = sd_journal_get_data(j, "MESSAGE_ID", &data, &size);
3186         if (r < 0)
3187                 return r;
3188
3189         cid = strndup((const char*) data + 11, size - 11);
3190         if (!cid)
3191                 return -ENOMEM;
3192
3193         r = sd_id128_from_string(cid, &id);
3194         if (r < 0)
3195                 return r;
3196
3197         r = catalog_get(CATALOG_DATABASE, id, &text);
3198         if (r < 0)
3199                 return r;
3200
3201         t = replace_var(text, lookup_field, j);
3202         if (!t)
3203                 return -ENOMEM;
3204
3205         *ret = t;
3206         return 0;
3207 }
3208
3209 _public_ int sd_journal_get_catalog_for_message_id(sd_id128_t id, char **ret) {
3210         assert_return(ret, -EINVAL);
3211
3212         return catalog_get(CATALOG_DATABASE, id, ret);
3213 }
3214
3215 _public_ int sd_journal_set_data_threshold(sd_journal *j, size_t sz) {
3216         assert_return(j, -EINVAL);
3217         assert_return(!journal_pid_changed(j), -ECHILD);
3218
3219         j->data_threshold = sz;
3220         return 0;
3221 }
3222
3223 _public_ int sd_journal_get_data_threshold(sd_journal *j, size_t *sz) {
3224         assert_return(j, -EINVAL);
3225         assert_return(!journal_pid_changed(j), -ECHILD);
3226         assert_return(sz, -EINVAL);
3227
3228         *sz = j->data_threshold;
3229         return 0;
3230 }
3231
3232 _public_ int sd_journal_has_runtime_files(sd_journal *j) {
3233         assert_return(j, -EINVAL);
3234
3235         return j->has_runtime_files;
3236 }
3237
3238 _public_ int sd_journal_has_persistent_files(sd_journal *j) {
3239         assert_return(j, -EINVAL);
3240
3241         return j->has_persistent_files;
3242 }