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