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                 _cleanup_close_ int dfd = -1;
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 = take_fdopendir(&dfd);
1735                 if (!d) {
1736                         r = -errno;
1737                         goto fail;
1738                 }
1739
1740                 rewinddir(d);
1741         }
1742
1743         m = hashmap_get(j->directories_by_path, p);
1744         if (!m) {
1745                 m = new0(Directory, 1);
1746                 if (!m) {
1747                         r = -ENOMEM;
1748                         goto fail;
1749                 }
1750
1751                 m->is_root = true;
1752
1753                 m->path = strdup(p);
1754                 if (!m->path) {
1755                         free(m);
1756                         r = -ENOMEM;
1757                         goto fail;
1758                 }
1759
1760                 if (hashmap_put(j->directories_by_path, m->path, m) < 0) {
1761                         free(m->path);
1762                         free(m);
1763                         r = -ENOMEM;
1764                         goto fail;
1765                 }
1766
1767                 j->current_invalidate_counter++;
1768
1769                 log_debug("Root directory %s added.", m->path);
1770
1771         } else if (!m->is_root)
1772                 return 0;
1773
1774         directory_watch(j, m, dirfd(d),
1775                         IN_CREATE|IN_MOVED_TO|IN_MODIFY|IN_ATTRIB|IN_DELETE|
1776                         IN_ONLYDIR);
1777
1778         if (!j->no_new_files)
1779                 directory_enumerate(j, m, d);
1780
1781         check_network(j, dirfd(d));
1782
1783         return 0;
1784
1785 fail:
1786         k = journal_put_error(j, r, p);
1787         if (k < 0)
1788                 return k;
1789
1790         return r;
1791 }
1792
1793 static void remove_directory(sd_journal *j, Directory *d) {
1794         assert(j);
1795
1796         if (d->wd > 0) {
1797                 hashmap_remove(j->directories_by_wd, INT_TO_PTR(d->wd));
1798
1799                 if (j->inotify_fd >= 0)
1800                         (void) inotify_rm_watch(j->inotify_fd, d->wd);
1801         }
1802
1803         hashmap_remove(j->directories_by_path, d->path);
1804
1805         if (d->is_root)
1806                 log_debug("Root directory %s removed.", d->path);
1807         else
1808                 log_debug("Directory %s removed.", d->path);
1809
1810         free(d->path);
1811         free(d);
1812 }
1813
1814 static int add_search_paths(sd_journal *j) {
1815
1816         static const char search_paths[] =
1817                 "/run/log/journal\0"
1818                 "/var/log/journal\0";
1819         const char *p;
1820
1821         assert(j);
1822
1823         /* We ignore most errors here, since the idea is to only open
1824          * what's actually accessible, and ignore the rest. */
1825
1826         NULSTR_FOREACH(p, search_paths)
1827                 (void) add_root_directory(j, p, true);
1828
1829         if (!(j->flags & SD_JOURNAL_LOCAL_ONLY))
1830                 (void) add_root_directory(j, "/var/log/journal/remote", true);
1831
1832         return 0;
1833 }
1834
1835 static int add_current_paths(sd_journal *j) {
1836         Iterator i;
1837         JournalFile *f;
1838
1839         assert(j);
1840         assert(j->no_new_files);
1841
1842         /* Simply adds all directories for files we have open as directories. We don't expect errors here, so we
1843          * treat them as fatal. */
1844
1845         ORDERED_HASHMAP_FOREACH(f, j->files, i) {
1846                 _cleanup_free_ char *dir;
1847                 int r;
1848
1849                 dir = dirname_malloc(f->path);
1850                 if (!dir)
1851                         return -ENOMEM;
1852
1853                 r = add_directory(j, dir, NULL);
1854                 if (r < 0)
1855                         return r;
1856         }
1857
1858         return 0;
1859 }
1860
1861 static int allocate_inotify(sd_journal *j) {
1862         assert(j);
1863
1864         if (j->inotify_fd < 0) {
1865                 j->inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
1866                 if (j->inotify_fd < 0)
1867                         return -errno;
1868         }
1869
1870         return hashmap_ensure_allocated(&j->directories_by_wd, NULL);
1871 }
1872
1873 static sd_journal *journal_new(int flags, const char *path, const char *namespace) {
1874         _cleanup_(sd_journal_closep) sd_journal *j = NULL;
1875
1876         j = new0(sd_journal, 1);
1877         if (!j)
1878                 return NULL;
1879
1880         j->original_pid = getpid_cached();
1881         j->toplevel_fd = -1;
1882         j->inotify_fd = -1;
1883         j->flags = flags;
1884         j->data_threshold = DEFAULT_DATA_THRESHOLD;
1885
1886         if (path) {
1887                 char *t;
1888
1889                 t = strdup(path);
1890                 if (!t)
1891                         return NULL;
1892
1893                 if (flags & SD_JOURNAL_OS_ROOT)
1894                         j->prefix = t;
1895                 else
1896                         j->path = t;
1897         }
1898
1899         if (namespace) {
1900                 j->namespace = strdup(namespace);
1901                 if (!j->namespace)
1902                         return NULL;
1903         }
1904
1905         j->files = ordered_hashmap_new(&path_hash_ops);
1906         if (!j->files)
1907                 return NULL;
1908
1909         j->files_cache = ordered_hashmap_iterated_cache_new(j->files);
1910         j->directories_by_path = hashmap_new(&path_hash_ops);
1911         j->mmap = mmap_cache_new();
1912         if (!j->files_cache || !j->directories_by_path || !j->mmap)
1913                 return NULL;
1914
1915         return TAKE_PTR(j);
1916 }
1917
1918 #define OPEN_ALLOWED_FLAGS                              \
1919         (SD_JOURNAL_LOCAL_ONLY |                        \
1920          SD_JOURNAL_RUNTIME_ONLY |                      \
1921          SD_JOURNAL_SYSTEM |                            \
1922          SD_JOURNAL_CURRENT_USER |                      \
1923          SD_JOURNAL_ALL_NAMESPACES |                    \
1924          SD_JOURNAL_INCLUDE_DEFAULT_NAMESPACE)
1925
1926 _public_ int sd_journal_open_namespace(sd_journal **ret, const char *namespace, int flags) {
1927         _cleanup_(sd_journal_closep) sd_journal *j = NULL;
1928         int r;
1929
1930         assert_return(ret, -EINVAL);
1931         assert_return((flags & ~OPEN_ALLOWED_FLAGS) == 0, -EINVAL);
1932
1933         j = journal_new(flags, NULL, namespace);
1934         if (!j)
1935                 return -ENOMEM;
1936
1937         r = add_search_paths(j);
1938         if (r < 0)
1939                 return r;
1940
1941         *ret = TAKE_PTR(j);
1942         return 0;
1943 }
1944
1945 _public_ int sd_journal_open(sd_journal **ret, int flags) {
1946         return sd_journal_open_namespace(ret, NULL, flags);
1947 }
1948
1949 #define OPEN_CONTAINER_ALLOWED_FLAGS                    \
1950         (SD_JOURNAL_LOCAL_ONLY | SD_JOURNAL_SYSTEM)
1951
1952 _public_ int sd_journal_open_container(sd_journal **ret, const char *machine, int flags) {
1953         _cleanup_free_ char *root = NULL, *class = NULL;
1954         _cleanup_(sd_journal_closep) sd_journal *j = NULL;
1955         char *p;
1956         int r;
1957
1958         /* This is deprecated, people should use machined's OpenMachineRootDirectory() call instead in
1959          * combination with sd_journal_open_directory_fd(). */
1960
1961         assert_return(machine, -EINVAL);
1962         assert_return(ret, -EINVAL);
1963         assert_return((flags & ~OPEN_CONTAINER_ALLOWED_FLAGS) == 0, -EINVAL);
1964         assert_return(machine_name_is_valid(machine), -EINVAL);
1965
1966         p = strjoina("/run/systemd/machines/", machine);
1967         r = parse_env_file(NULL, p,
1968                            "ROOT", &root,
1969                            "CLASS", &class);
1970         if (r == -ENOENT)
1971                 return -EHOSTDOWN;
1972         if (r < 0)
1973                 return r;
1974         if (!root)
1975                 return -ENODATA;
1976
1977         if (!streq_ptr(class, "container"))
1978                 return -EIO;
1979
1980         j = journal_new(flags, root, NULL);
1981         if (!j)
1982                 return -ENOMEM;
1983
1984         r = add_search_paths(j);
1985         if (r < 0)
1986                 return r;
1987
1988         *ret = TAKE_PTR(j);
1989         return 0;
1990 }
1991
1992 #define OPEN_DIRECTORY_ALLOWED_FLAGS                    \
1993         (SD_JOURNAL_OS_ROOT |                           \
1994          SD_JOURNAL_SYSTEM | SD_JOURNAL_CURRENT_USER )
1995
1996 _public_ int sd_journal_open_directory(sd_journal **ret, const char *path, int flags) {
1997         _cleanup_(sd_journal_closep) sd_journal *j = NULL;
1998         int r;
1999
2000         assert_return(ret, -EINVAL);
2001         assert_return(path, -EINVAL);
2002         assert_return((flags & ~OPEN_DIRECTORY_ALLOWED_FLAGS) == 0, -EINVAL);
2003
2004         j = journal_new(flags, path, NULL);
2005         if (!j)
2006                 return -ENOMEM;
2007
2008         if (flags & SD_JOURNAL_OS_ROOT)
2009                 r = add_search_paths(j);
2010         else
2011                 r = add_root_directory(j, path, false);
2012         if (r < 0)
2013                 return r;
2014
2015         *ret = TAKE_PTR(j);
2016         return 0;
2017 }
2018
2019 _public_ int sd_journal_open_files(sd_journal **ret, const char **paths, int flags) {
2020         _cleanup_(sd_journal_closep) sd_journal *j = NULL;
2021         const char **path;
2022         int r;
2023
2024         assert_return(ret, -EINVAL);
2025         assert_return(flags == 0, -EINVAL);
2026
2027         j = journal_new(flags, NULL, NULL);
2028         if (!j)
2029                 return -ENOMEM;
2030
2031         STRV_FOREACH(path, paths) {
2032                 r = add_any_file(j, -1, *path);
2033                 if (r < 0)
2034                         return r;
2035         }
2036
2037         j->no_new_files = true;
2038
2039         *ret = TAKE_PTR(j);
2040         return 0;
2041 }
2042
2043 #define OPEN_DIRECTORY_FD_ALLOWED_FLAGS         \
2044         (SD_JOURNAL_OS_ROOT |                           \
2045          SD_JOURNAL_SYSTEM | SD_JOURNAL_CURRENT_USER )
2046
2047 _public_ int sd_journal_open_directory_fd(sd_journal **ret, int fd, int flags) {
2048         _cleanup_(sd_journal_closep) sd_journal *j = NULL;
2049         struct stat st;
2050         int r;
2051
2052         assert_return(ret, -EINVAL);
2053         assert_return(fd >= 0, -EBADF);
2054         assert_return((flags & ~OPEN_DIRECTORY_FD_ALLOWED_FLAGS) == 0, -EINVAL);
2055
2056         if (fstat(fd, &st) < 0)
2057                 return -errno;
2058
2059         if (!S_ISDIR(st.st_mode))
2060                 return -EBADFD;
2061
2062         j = journal_new(flags, NULL, NULL);
2063         if (!j)
2064                 return -ENOMEM;
2065
2066         j->toplevel_fd = fd;
2067
2068         if (flags & SD_JOURNAL_OS_ROOT)
2069                 r = add_search_paths(j);
2070         else
2071                 r = add_root_directory(j, NULL, false);
2072         if (r < 0)
2073                 return r;
2074
2075         *ret = TAKE_PTR(j);
2076         return 0;
2077 }
2078
2079 _public_ int sd_journal_open_files_fd(sd_journal **ret, int fds[], unsigned n_fds, int flags) {
2080         Iterator iterator;
2081         JournalFile *f;
2082         _cleanup_(sd_journal_closep) sd_journal *j = NULL;
2083         unsigned i;
2084         int r;
2085
2086         assert_return(ret, -EINVAL);
2087         assert_return(n_fds > 0, -EBADF);
2088         assert_return(flags == 0, -EINVAL);
2089
2090         j = journal_new(flags, NULL, NULL);
2091         if (!j)
2092                 return -ENOMEM;
2093
2094         for (i = 0; i < n_fds; i++) {
2095                 struct stat st;
2096
2097                 if (fds[i] < 0) {
2098                         r = -EBADF;
2099                         goto fail;
2100                 }
2101
2102                 if (fstat(fds[i], &st) < 0) {
2103                         r = -errno;
2104                         goto fail;
2105                 }
2106
2107                 r = stat_verify_regular(&st);
2108                 if (r < 0)
2109                         goto fail;
2110
2111                 r = add_any_file(j, fds[i], NULL);
2112                 if (r < 0)
2113                         goto fail;
2114         }
2115
2116         j->no_new_files = true;
2117         j->no_inotify = true;
2118
2119         *ret = TAKE_PTR(j);
2120         return 0;
2121
2122 fail:
2123         /* If we fail, make sure we don't take possession of the files we managed to make use of successfully, and they
2124          * remain open */
2125         ORDERED_HASHMAP_FOREACH(f, j->files, iterator)
2126                 f->close_fd = false;
2127
2128         return r;
2129 }
2130
2131 _public_ void sd_journal_close(sd_journal *j) {
2132         Directory *d;
2133
2134         if (!j)
2135                 return;
2136
2137         sd_journal_flush_matches(j);
2138
2139         ordered_hashmap_free_with_destructor(j->files, journal_file_close);
2140         iterated_cache_free(j->files_cache);
2141
2142         while ((d = hashmap_first(j->directories_by_path)))
2143                 remove_directory(j, d);
2144
2145         while ((d = hashmap_first(j->directories_by_wd)))
2146                 remove_directory(j, d);
2147
2148         hashmap_free(j->directories_by_path);
2149         hashmap_free(j->directories_by_wd);
2150
2151         safe_close(j->inotify_fd);
2152
2153         if (j->mmap) {
2154                 log_debug("mmap cache statistics: %u hit, %u miss", mmap_cache_get_hit(j->mmap), mmap_cache_get_missed(j->mmap));
2155                 mmap_cache_unref(j->mmap);
2156         }
2157
2158         hashmap_free_free(j->errors);
2159
2160         free(j->path);
2161         free(j->prefix);
2162         free(j->namespace);
2163         free(j->unique_field);
2164         free(j->fields_buffer);
2165         free(j);
2166 }
2167
2168 _public_ int sd_journal_get_realtime_usec(sd_journal *j, uint64_t *ret) {
2169         Object *o;
2170         JournalFile *f;
2171         int r;
2172
2173         assert_return(j, -EINVAL);
2174         assert_return(!journal_pid_changed(j), -ECHILD);
2175         assert_return(ret, -EINVAL);
2176
2177         f = j->current_file;
2178         if (!f)
2179                 return -EADDRNOTAVAIL;
2180
2181         if (f->current_offset <= 0)
2182                 return -EADDRNOTAVAIL;
2183
2184         r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &o);
2185         if (r < 0)
2186                 return r;
2187
2188         *ret = le64toh(o->entry.realtime);
2189         return 0;
2190 }
2191
2192 _public_ int sd_journal_get_monotonic_usec(sd_journal *j, uint64_t *ret, sd_id128_t *ret_boot_id) {
2193         Object *o;
2194         JournalFile *f;
2195         int r;
2196         sd_id128_t id;
2197
2198         assert_return(j, -EINVAL);
2199         assert_return(!journal_pid_changed(j), -ECHILD);
2200
2201         f = j->current_file;
2202         if (!f)
2203                 return -EADDRNOTAVAIL;
2204
2205         if (f->current_offset <= 0)
2206                 return -EADDRNOTAVAIL;
2207
2208         r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &o);
2209         if (r < 0)
2210                 return r;
2211
2212         if (ret_boot_id)
2213                 *ret_boot_id = o->entry.boot_id;
2214         else {
2215                 r = sd_id128_get_boot(&id);
2216                 if (r < 0)
2217                         return r;
2218
2219                 if (!sd_id128_equal(id, o->entry.boot_id))
2220                         return -ESTALE;
2221         }
2222
2223         if (ret)
2224                 *ret = le64toh(o->entry.monotonic);
2225
2226         return 0;
2227 }
2228
2229 static bool field_is_valid(const char *field) {
2230         const char *p;
2231
2232         assert(field);
2233
2234         if (isempty(field))
2235                 return false;
2236
2237         if (startswith(field, "__"))
2238                 return false;
2239
2240         for (p = field; *p; p++) {
2241
2242                 if (*p == '_')
2243                         continue;
2244
2245                 if (*p >= 'A' && *p <= 'Z')
2246                         continue;
2247
2248                 if (*p >= '0' && *p <= '9')
2249                         continue;
2250
2251                 return false;
2252         }
2253
2254         return true;
2255 }
2256
2257 _public_ int sd_journal_get_data(sd_journal *j, const char *field, const void **data, size_t *size) {
2258         JournalFile *f;
2259         uint64_t i, n;
2260         size_t field_length;
2261         int r;
2262         Object *o;
2263
2264         assert_return(j, -EINVAL);
2265         assert_return(!journal_pid_changed(j), -ECHILD);
2266         assert_return(field, -EINVAL);
2267         assert_return(data, -EINVAL);
2268         assert_return(size, -EINVAL);
2269         assert_return(field_is_valid(field), -EINVAL);
2270
2271         f = j->current_file;
2272         if (!f)
2273                 return -EADDRNOTAVAIL;
2274
2275         if (f->current_offset <= 0)
2276                 return -EADDRNOTAVAIL;
2277
2278         r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &o);
2279         if (r < 0)
2280                 return r;
2281
2282         field_length = strlen(field);
2283
2284         n = journal_file_entry_n_items(o);
2285         for (i = 0; i < n; i++) {
2286                 uint64_t p, l;
2287                 le64_t le_hash;
2288                 size_t t;
2289                 int compression;
2290
2291                 p = le64toh(o->entry.items[i].object_offset);
2292                 le_hash = o->entry.items[i].hash;
2293                 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
2294                 if (r < 0)
2295                         return r;
2296
2297                 if (le_hash != o->data.hash)
2298                         return -EBADMSG;
2299
2300                 l = le64toh(o->object.size) - offsetof(Object, data.payload);
2301
2302                 compression = o->object.flags & OBJECT_COMPRESSION_MASK;
2303                 if (compression) {
2304 #if HAVE_XZ || HAVE_LZ4
2305                         r = decompress_startswith(compression,
2306                                                   o->data.payload, l,
2307                                                   &f->compress_buffer, &f->compress_buffer_size,
2308                                                   field, field_length, '=');
2309                         if (r < 0)
2310                                 log_debug_errno(r, "Cannot decompress %s object of length %"PRIu64" at offset "OFSfmt": %m",
2311                                                 object_compressed_to_string(compression), l, p);
2312                         else if (r > 0) {
2313
2314                                 size_t rsize;
2315
2316                                 r = decompress_blob(compression,
2317                                                     o->data.payload, l,
2318                                                     &f->compress_buffer, &f->compress_buffer_size, &rsize,
2319                                                     j->data_threshold);
2320                                 if (r < 0)
2321                                         return r;
2322
2323                                 *data = f->compress_buffer;
2324                                 *size = (size_t) rsize;
2325
2326                                 return 0;
2327                         }
2328 #else
2329                         return -EPROTONOSUPPORT;
2330 #endif
2331                 } else if (l >= field_length+1 &&
2332                            memcmp(o->data.payload, field, field_length) == 0 &&
2333                            o->data.payload[field_length] == '=') {
2334
2335                         t = (size_t) l;
2336
2337                         if ((uint64_t) t != l)
2338                                 return -E2BIG;
2339
2340                         *data = o->data.payload;
2341                         *size = t;
2342
2343                         return 0;
2344                 }
2345
2346                 r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &o);
2347                 if (r < 0)
2348                         return r;
2349         }
2350
2351         return -ENOENT;
2352 }
2353
2354 static int return_data(sd_journal *j, JournalFile *f, Object *o, const void **data, size_t *size) {
2355         size_t t;
2356         uint64_t l;
2357         int compression;
2358
2359         l = le64toh(o->object.size) - offsetof(Object, data.payload);
2360         t = (size_t) l;
2361
2362         /* We can't read objects larger than 4G on a 32bit machine */
2363         if ((uint64_t) t != l)
2364                 return -E2BIG;
2365
2366         compression = o->object.flags & OBJECT_COMPRESSION_MASK;
2367         if (compression) {
2368 #if HAVE_XZ || HAVE_LZ4
2369                 size_t rsize;
2370                 int r;
2371
2372                 r = decompress_blob(compression,
2373                                     o->data.payload, l, &f->compress_buffer,
2374                                     &f->compress_buffer_size, &rsize, j->data_threshold);
2375                 if (r < 0)
2376                         return r;
2377
2378                 *data = f->compress_buffer;
2379                 *size = (size_t) rsize;
2380 #else
2381                 return -EPROTONOSUPPORT;
2382 #endif
2383         } else {
2384                 *data = o->data.payload;
2385                 *size = t;
2386         }
2387
2388         return 0;
2389 }
2390
2391 _public_ int sd_journal_enumerate_data(sd_journal *j, const void **data, size_t *size) {
2392         JournalFile *f;
2393         uint64_t p, n;
2394         le64_t le_hash;
2395         int r;
2396         Object *o;
2397
2398         assert_return(j, -EINVAL);
2399         assert_return(!journal_pid_changed(j), -ECHILD);
2400         assert_return(data, -EINVAL);
2401         assert_return(size, -EINVAL);
2402
2403         f = j->current_file;
2404         if (!f)
2405                 return -EADDRNOTAVAIL;
2406
2407         if (f->current_offset <= 0)
2408                 return -EADDRNOTAVAIL;
2409
2410         r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &o);
2411         if (r < 0)
2412                 return r;
2413
2414         n = journal_file_entry_n_items(o);
2415         if (j->current_field >= n)
2416                 return 0;
2417
2418         p = le64toh(o->entry.items[j->current_field].object_offset);
2419         le_hash = o->entry.items[j->current_field].hash;
2420         r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
2421         if (r < 0)
2422                 return r;
2423
2424         if (le_hash != o->data.hash)
2425                 return -EBADMSG;
2426
2427         r = return_data(j, f, o, data, size);
2428         if (r < 0)
2429                 return r;
2430
2431         j->current_field++;
2432
2433         return 1;
2434 }
2435
2436 _public_ void sd_journal_restart_data(sd_journal *j) {
2437         if (!j)
2438                 return;
2439
2440         j->current_field = 0;
2441 }
2442
2443 static int reiterate_all_paths(sd_journal *j) {
2444         assert(j);
2445
2446         if (j->no_new_files)
2447                 return add_current_paths(j);
2448
2449         if (j->flags & SD_JOURNAL_OS_ROOT)
2450                 return add_search_paths(j);
2451
2452         if (j->toplevel_fd >= 0)
2453                 return add_root_directory(j, NULL, false);
2454
2455         if (j->path)
2456                 return add_root_directory(j, j->path, true);
2457
2458         return add_search_paths(j);
2459 }
2460
2461 _public_ int sd_journal_get_fd(sd_journal *j) {
2462         int r;
2463
2464         assert_return(j, -EINVAL);
2465         assert_return(!journal_pid_changed(j), -ECHILD);
2466
2467         if (j->no_inotify)
2468                 return -EMEDIUMTYPE;
2469
2470         if (j->inotify_fd >= 0)
2471                 return j->inotify_fd;
2472
2473         r = allocate_inotify(j);
2474         if (r < 0)
2475                 return r;
2476
2477         log_debug("Reiterating files to get inotify watches established.");
2478
2479         /* Iterate through all dirs again, to add them to the inotify */
2480         r = reiterate_all_paths(j);
2481         if (r < 0)
2482                 return r;
2483
2484         return j->inotify_fd;
2485 }
2486
2487 _public_ int sd_journal_get_events(sd_journal *j) {
2488         int fd;
2489
2490         assert_return(j, -EINVAL);
2491         assert_return(!journal_pid_changed(j), -ECHILD);
2492
2493         fd = sd_journal_get_fd(j);
2494         if (fd < 0)
2495                 return fd;
2496
2497         return POLLIN;
2498 }
2499
2500 _public_ int sd_journal_get_timeout(sd_journal *j, uint64_t *timeout_usec) {
2501         int fd;
2502
2503         assert_return(j, -EINVAL);
2504         assert_return(!journal_pid_changed(j), -ECHILD);
2505         assert_return(timeout_usec, -EINVAL);
2506
2507         fd = sd_journal_get_fd(j);
2508         if (fd < 0)
2509                 return fd;
2510
2511         if (!j->on_network) {
2512                 *timeout_usec = (uint64_t) -1;
2513                 return 0;
2514         }
2515
2516         /* If we are on the network we need to regularly check for
2517          * changes manually */
2518
2519         *timeout_usec = j->last_process_usec + JOURNAL_FILES_RECHECK_USEC;
2520         return 1;
2521 }
2522
2523 static void process_q_overflow(sd_journal *j) {
2524         JournalFile *f;
2525         Directory *m;
2526         Iterator i;
2527
2528         assert(j);
2529
2530         /* When the inotify queue overruns we need to enumerate and re-validate all journal files to bring our list
2531          * back in sync with what's on disk. For this we pick a new generation counter value. It'll be assigned to all
2532          * journal files we encounter. All journal files and all directories that don't carry it after reenumeration
2533          * are subject for unloading. */
2534
2535         log_debug("Inotify queue overrun, reiterating everything.");
2536
2537         j->generation++;
2538         (void) reiterate_all_paths(j);
2539
2540         ORDERED_HASHMAP_FOREACH(f, j->files, i) {
2541
2542                 if (f->last_seen_generation == j->generation)
2543                         continue;
2544
2545                 log_debug("File '%s' hasn't been seen in this enumeration, removing.", f->path);
2546                 remove_file_real(j, f);
2547         }
2548
2549         HASHMAP_FOREACH(m, j->directories_by_path, i) {
2550
2551                 if (m->last_seen_generation == j->generation)
2552                         continue;
2553
2554                 if (m->is_root) /* Never GC root directories */
2555                         continue;
2556
2557                 log_debug("Directory '%s' hasn't been seen in this enumeration, removing.", f->path);
2558                 remove_directory(j, m);
2559         }
2560
2561         log_debug("Reiteration complete.");
2562 }
2563
2564 static void process_inotify_event(sd_journal *j, const struct inotify_event *e) {
2565         Directory *d;
2566
2567         assert(j);
2568         assert(e);
2569
2570         if (e->mask & IN_Q_OVERFLOW) {
2571                 process_q_overflow(j);
2572                 return;
2573         }
2574
2575         /* Is this a subdirectory we watch? */
2576         d = hashmap_get(j->directories_by_wd, INT_TO_PTR(e->wd));
2577         if (d) {
2578                 if (!(e->mask & IN_ISDIR) && e->len > 0 &&
2579                     (endswith(e->name, ".journal") ||
2580                      endswith(e->name, ".journal~"))) {
2581
2582                         /* Event for a journal file */
2583
2584                         if (e->mask & (IN_CREATE|IN_MOVED_TO|IN_MODIFY|IN_ATTRIB))
2585                                 (void) add_file_by_name(j, d->path, e->name);
2586                         else if (e->mask & (IN_DELETE|IN_MOVED_FROM|IN_UNMOUNT))
2587                                 remove_file_by_name(j, d->path, e->name);
2588
2589                 } else if (!d->is_root && e->len == 0) {
2590
2591                         /* Event for a subdirectory */
2592
2593                         if (e->mask & (IN_DELETE_SELF|IN_MOVE_SELF|IN_UNMOUNT))
2594                                 remove_directory(j, d);
2595
2596                 } else if (d->is_root && (e->mask & IN_ISDIR) && e->len > 0 && id128_is_valid(e->name)) {
2597
2598                         /* Event for root directory */
2599
2600                         if (e->mask & (IN_CREATE|IN_MOVED_TO|IN_MODIFY|IN_ATTRIB))
2601                                 (void) add_directory(j, d->path, e->name);
2602                 }
2603
2604                 return;
2605         }
2606
2607         if (e->mask & IN_IGNORED)
2608                 return;
2609
2610         log_debug("Unexpected inotify event.");
2611 }
2612
2613 static int determine_change(sd_journal *j) {
2614         bool b;
2615
2616         assert(j);
2617
2618         b = j->current_invalidate_counter != j->last_invalidate_counter;
2619         j->last_invalidate_counter = j->current_invalidate_counter;
2620
2621         return b ? SD_JOURNAL_INVALIDATE : SD_JOURNAL_APPEND;
2622 }
2623
2624 _public_ int sd_journal_process(sd_journal *j) {
2625         bool got_something = false;
2626
2627         assert_return(j, -EINVAL);
2628         assert_return(!journal_pid_changed(j), -ECHILD);
2629
2630         if (j->inotify_fd < 0) /* We have no inotify fd yet? Then there's noting to process. */
2631                 return 0;
2632
2633         j->last_process_usec = now(CLOCK_MONOTONIC);
2634         j->last_invalidate_counter = j->current_invalidate_counter;
2635
2636         for (;;) {
2637                 union inotify_event_buffer buffer;
2638                 struct inotify_event *e;
2639                 ssize_t l;
2640
2641                 l = read(j->inotify_fd, &buffer, sizeof(buffer));
2642                 if (l < 0) {
2643                         if (IN_SET(errno, EAGAIN, EINTR))
2644                                 return got_something ? determine_change(j) : SD_JOURNAL_NOP;
2645
2646                         return -errno;
2647                 }
2648
2649                 got_something = true;
2650
2651                 FOREACH_INOTIFY_EVENT(e, buffer, l)
2652                         process_inotify_event(j, e);
2653         }
2654 }
2655
2656 _public_ int sd_journal_wait(sd_journal *j, uint64_t timeout_usec) {
2657         int r;
2658         uint64_t t;
2659
2660         assert_return(j, -EINVAL);
2661         assert_return(!journal_pid_changed(j), -ECHILD);
2662
2663         if (j->inotify_fd < 0) {
2664                 Iterator i;
2665                 JournalFile *f;
2666
2667                 /* This is the first invocation, hence create the
2668                  * inotify watch */
2669                 r = sd_journal_get_fd(j);
2670                 if (r < 0)
2671                         return r;
2672
2673                 /* Server might have done some vacuuming while we weren't watching.
2674                    Get rid of the deleted files now so they don't stay around indefinitely. */
2675                 ORDERED_HASHMAP_FOREACH(f, j->files, i) {
2676                         r = journal_file_fstat(f);
2677                         if (r == -EIDRM)
2678                                 remove_file_real(j, f);
2679                         else if (r < 0) {
2680                                 log_debug_errno(r,"Failed to fstat() journal file '%s' : %m", f->path);
2681                                 continue;
2682                         }
2683                 }
2684
2685                 /* The journal might have changed since the context
2686                  * object was created and we weren't watching before,
2687                  * hence don't wait for anything, and return
2688                  * immediately. */
2689                 return determine_change(j);
2690         }
2691
2692         r = sd_journal_get_timeout(j, &t);
2693         if (r < 0)
2694                 return r;
2695
2696         if (t != (uint64_t) -1) {
2697                 usec_t n;
2698
2699                 n = now(CLOCK_MONOTONIC);
2700                 t = t > n ? t - n : 0;
2701
2702                 if (timeout_usec == (uint64_t) -1 || timeout_usec > t)
2703                         timeout_usec = t;
2704         }
2705
2706         do {
2707                 r = fd_wait_for_event(j->inotify_fd, POLLIN, timeout_usec);
2708         } while (r == -EINTR);
2709
2710         if (r < 0)
2711                 return r;
2712
2713         return sd_journal_process(j);
2714 }
2715
2716 _public_ int sd_journal_get_cutoff_realtime_usec(sd_journal *j, uint64_t *from, uint64_t *to) {
2717         Iterator i;
2718         JournalFile *f;
2719         bool first = true;
2720         uint64_t fmin = 0, tmax = 0;
2721         int r;
2722
2723         assert_return(j, -EINVAL);
2724         assert_return(!journal_pid_changed(j), -ECHILD);
2725         assert_return(from || to, -EINVAL);
2726         assert_return(from != to, -EINVAL);
2727
2728         ORDERED_HASHMAP_FOREACH(f, j->files, i) {
2729                 usec_t fr, t;
2730
2731                 r = journal_file_get_cutoff_realtime_usec(f, &fr, &t);
2732                 if (r == -ENOENT)
2733                         continue;
2734                 if (r < 0)
2735                         return r;
2736                 if (r == 0)
2737                         continue;
2738
2739                 if (first) {
2740                         fmin = fr;
2741                         tmax = t;
2742                         first = false;
2743                 } else {
2744                         fmin = MIN(fr, fmin);
2745                         tmax = MAX(t, tmax);
2746                 }
2747         }
2748
2749         if (from)
2750                 *from = fmin;
2751         if (to)
2752                 *to = tmax;
2753
2754         return first ? 0 : 1;
2755 }
2756
2757 _public_ int sd_journal_get_cutoff_monotonic_usec(sd_journal *j, sd_id128_t boot_id, uint64_t *from, uint64_t *to) {
2758         Iterator i;
2759         JournalFile *f;
2760         bool found = false;
2761         int r;
2762
2763         assert_return(j, -EINVAL);
2764         assert_return(!journal_pid_changed(j), -ECHILD);
2765         assert_return(from || to, -EINVAL);
2766         assert_return(from != to, -EINVAL);
2767
2768         ORDERED_HASHMAP_FOREACH(f, j->files, i) {
2769                 usec_t fr, t;
2770
2771                 r = journal_file_get_cutoff_monotonic_usec(f, boot_id, &fr, &t);
2772                 if (r == -ENOENT)
2773                         continue;
2774                 if (r < 0)
2775                         return r;
2776                 if (r == 0)
2777                         continue;
2778
2779                 if (found) {
2780                         if (from)
2781                                 *from = MIN(fr, *from);
2782                         if (to)
2783                                 *to = MAX(t, *to);
2784                 } else {
2785                         if (from)
2786                                 *from = fr;
2787                         if (to)
2788                                 *to = t;
2789                         found = true;
2790                 }
2791         }
2792
2793         return found;
2794 }
2795
2796 void journal_print_header(sd_journal *j) {
2797         Iterator i;
2798         JournalFile *f;
2799         bool newline = false;
2800
2801         assert(j);
2802
2803         ORDERED_HASHMAP_FOREACH(f, j->files, i) {
2804                 if (newline)
2805                         putchar('\n');
2806                 else
2807                         newline = true;
2808
2809                 journal_file_print_header(f);
2810         }
2811 }
2812
2813 _public_ int sd_journal_get_usage(sd_journal *j, uint64_t *bytes) {
2814         Iterator i;
2815         JournalFile *f;
2816         uint64_t sum = 0;
2817
2818         assert_return(j, -EINVAL);
2819         assert_return(!journal_pid_changed(j), -ECHILD);
2820         assert_return(bytes, -EINVAL);
2821
2822         ORDERED_HASHMAP_FOREACH(f, j->files, i) {
2823                 struct stat st;
2824
2825                 if (fstat(f->fd, &st) < 0)
2826                         return -errno;
2827
2828                 sum += (uint64_t) st.st_blocks * 512ULL;
2829         }
2830
2831         *bytes = sum;
2832         return 0;
2833 }
2834
2835 _public_ int sd_journal_query_unique(sd_journal *j, const char *field) {
2836         char *f;
2837
2838         assert_return(j, -EINVAL);
2839         assert_return(!journal_pid_changed(j), -ECHILD);
2840         assert_return(!isempty(field), -EINVAL);
2841         assert_return(field_is_valid(field), -EINVAL);
2842
2843         f = strdup(field);
2844         if (!f)
2845                 return -ENOMEM;
2846
2847         free(j->unique_field);
2848         j->unique_field = f;
2849         j->unique_file = NULL;
2850         j->unique_offset = 0;
2851         j->unique_file_lost = false;
2852
2853         return 0;
2854 }
2855
2856 _public_ int sd_journal_enumerate_unique(sd_journal *j, const void **data, size_t *l) {
2857         size_t k;
2858
2859         assert_return(j, -EINVAL);
2860         assert_return(!journal_pid_changed(j), -ECHILD);
2861         assert_return(data, -EINVAL);
2862         assert_return(l, -EINVAL);
2863         assert_return(j->unique_field, -EINVAL);
2864
2865         k = strlen(j->unique_field);
2866
2867         if (!j->unique_file) {
2868                 if (j->unique_file_lost)
2869                         return 0;
2870
2871                 j->unique_file = ordered_hashmap_first(j->files);
2872                 if (!j->unique_file)
2873                         return 0;
2874
2875                 j->unique_offset = 0;
2876         }
2877
2878         for (;;) {
2879                 JournalFile *of;
2880                 Iterator i;
2881                 Object *o;
2882                 const void *odata;
2883                 size_t ol;
2884                 bool found;
2885                 int r;
2886
2887                 /* Proceed to next data object in the field's linked list */
2888                 if (j->unique_offset == 0) {
2889                         r = journal_file_find_field_object(j->unique_file, j->unique_field, k, &o, NULL);
2890                         if (r < 0)
2891                                 return r;
2892
2893                         j->unique_offset = r > 0 ? le64toh(o->field.head_data_offset) : 0;
2894                 } else {
2895                         r = journal_file_move_to_object(j->unique_file, OBJECT_DATA, j->unique_offset, &o);
2896                         if (r < 0)
2897                                 return r;
2898
2899                         j->unique_offset = le64toh(o->data.next_field_offset);
2900                 }
2901
2902                 /* We reached the end of the list? Then start again, with the next file */
2903                 if (j->unique_offset == 0) {
2904                         j->unique_file = ordered_hashmap_next(j->files, j->unique_file->path);
2905                         if (!j->unique_file)
2906                                 return 0;
2907
2908                         continue;
2909                 }
2910
2911                 /* We do not use OBJECT_DATA context here, but OBJECT_UNUSED
2912                  * instead, so that we can look at this data object at the same
2913                  * time as one on another file */
2914                 r = journal_file_move_to_object(j->unique_file, OBJECT_UNUSED, j->unique_offset, &o);
2915                 if (r < 0)
2916                         return r;
2917
2918                 /* Let's do the type check by hand, since we used 0 context above. */
2919                 if (o->object.type != OBJECT_DATA)
2920                         return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
2921                                                "%s:offset " OFSfmt ": object has type %d, expected %d",
2922                                                j->unique_file->path,
2923                                                j->unique_offset,
2924                                                o->object.type, OBJECT_DATA);
2925
2926                 r = return_data(j, j->unique_file, o, &odata, &ol);
2927                 if (r < 0)
2928                         return r;
2929
2930                 /* Check if we have at least the field name and "=". */
2931                 if (ol <= k)
2932                         return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
2933                                                "%s:offset " OFSfmt ": object has size %zu, expected at least %zu",
2934                                                j->unique_file->path,
2935                                                j->unique_offset, ol, k + 1);
2936
2937                 if (memcmp(odata, j->unique_field, k) || ((const char*) odata)[k] != '=')
2938                         return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
2939                                                "%s:offset " OFSfmt ": object does not start with \"%s=\"",
2940                                                j->unique_file->path,
2941                                                j->unique_offset,
2942                                                j->unique_field);
2943
2944                 /* OK, now let's see if we already returned this data
2945                  * object by checking if it exists in the earlier
2946                  * traversed files. */
2947                 found = false;
2948                 ORDERED_HASHMAP_FOREACH(of, j->files, i) {
2949                         if (of == j->unique_file)
2950                                 break;
2951
2952                         /* Skip this file it didn't have any fields indexed */
2953                         if (JOURNAL_HEADER_CONTAINS(of->header, n_fields) && le64toh(of->header->n_fields) <= 0)
2954                                 continue;
2955
2956                         r = journal_file_find_data_object_with_hash(of, odata, ol, le64toh(o->data.hash), NULL, NULL);
2957                         if (r < 0)
2958                                 return r;
2959                         if (r > 0) {
2960                                 found = true;
2961                                 break;
2962                         }
2963                 }
2964
2965                 if (found)
2966                         continue;
2967
2968                 r = return_data(j, j->unique_file, o, data, l);
2969                 if (r < 0)
2970                         return r;
2971
2972                 return 1;
2973         }
2974 }
2975
2976 _public_ void sd_journal_restart_unique(sd_journal *j) {
2977         if (!j)
2978                 return;
2979
2980         j->unique_file = NULL;
2981         j->unique_offset = 0;
2982         j->unique_file_lost = false;
2983 }
2984
2985 _public_ int sd_journal_enumerate_fields(sd_journal *j, const char **field) {
2986         int r;
2987
2988         assert_return(j, -EINVAL);
2989         assert_return(!journal_pid_changed(j), -ECHILD);
2990         assert_return(field, -EINVAL);
2991
2992         if (!j->fields_file) {
2993                 if (j->fields_file_lost)
2994                         return 0;
2995
2996                 j->fields_file = ordered_hashmap_first(j->files);
2997                 if (!j->fields_file)
2998                         return 0;
2999
3000                 j->fields_hash_table_index = 0;
3001                 j->fields_offset = 0;
3002         }
3003
3004         for (;;) {
3005                 JournalFile *f, *of;
3006                 Iterator i;
3007                 uint64_t m;
3008                 Object *o;
3009                 size_t sz;
3010                 bool found;
3011
3012                 f = j->fields_file;
3013
3014                 if (j->fields_offset == 0) {
3015                         bool eof = false;
3016
3017                         /* We are not yet positioned at any field. Let's pick the first one */
3018                         r = journal_file_map_field_hash_table(f);
3019                         if (r < 0)
3020                                 return r;
3021
3022                         m = le64toh(f->header->field_hash_table_size) / sizeof(HashItem);
3023                         for (;;) {
3024                                 if (j->fields_hash_table_index >= m) {
3025                                         /* Reached the end of the hash table, go to the next file. */
3026                                         eof = true;
3027                                         break;
3028                                 }
3029
3030                                 j->fields_offset = le64toh(f->field_hash_table[j->fields_hash_table_index].head_hash_offset);
3031
3032                                 if (j->fields_offset != 0)
3033                                         break;
3034
3035                                 /* Empty hash table bucket, go to next one */
3036                                 j->fields_hash_table_index++;
3037                         }
3038
3039                         if (eof) {
3040                                 /* Proceed with next file */
3041                                 j->fields_file = ordered_hashmap_next(j->files, f->path);
3042                                 if (!j->fields_file) {
3043                                         *field = NULL;
3044                                         return 0;
3045                                 }
3046
3047                                 j->fields_offset = 0;
3048                                 j->fields_hash_table_index = 0;
3049                                 continue;
3050                         }
3051
3052                 } else {
3053                         /* We are already positioned at a field. If so, let's figure out the next field from it */
3054
3055                         r = journal_file_move_to_object(f, OBJECT_FIELD, j->fields_offset, &o);
3056                         if (r < 0)
3057                                 return r;
3058
3059                         j->fields_offset = le64toh(o->field.next_hash_offset);
3060                         if (j->fields_offset == 0) {
3061                                 /* Reached the end of the hash table chain */
3062                                 j->fields_hash_table_index++;
3063                                 continue;
3064                         }
3065                 }
3066
3067                 /* We use OBJECT_UNUSED here, so that the iterator below doesn't remove our mmap window */
3068                 r = journal_file_move_to_object(f, OBJECT_UNUSED, j->fields_offset, &o);
3069                 if (r < 0)
3070                         return r;
3071
3072                 /* Because we used OBJECT_UNUSED above, we need to do our type check manually */
3073                 if (o->object.type != OBJECT_FIELD)
3074                         return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
3075                                                "%s:offset " OFSfmt ": object has type %i, expected %i",
3076                                                f->path, j->fields_offset,
3077                                                o->object.type, OBJECT_FIELD);
3078
3079                 sz = le64toh(o->object.size) - offsetof(Object, field.payload);
3080
3081                 /* Let's see if we already returned this field name before. */
3082                 found = false;
3083                 ORDERED_HASHMAP_FOREACH(of, j->files, i) {
3084                         if (of == f)
3085                                 break;
3086
3087                         /* Skip this file it didn't have any fields indexed */
3088                         if (JOURNAL_HEADER_CONTAINS(of->header, n_fields) && le64toh(of->header->n_fields) <= 0)
3089                                 continue;
3090
3091                         r = journal_file_find_field_object_with_hash(of, o->field.payload, sz, le64toh(o->field.hash), NULL, NULL);
3092                         if (r < 0)
3093                                 return r;
3094                         if (r > 0) {
3095                                 found = true;
3096                                 break;
3097                         }
3098                 }
3099
3100                 if (found)
3101                         continue;
3102
3103                 /* Check if this is really a valid string containing no NUL byte */
3104                 if (memchr(o->field.payload, 0, sz))
3105                         return -EBADMSG;
3106
3107                 if (sz > j->data_threshold)
3108                         sz = j->data_threshold;
3109
3110                 if (!GREEDY_REALLOC(j->fields_buffer, j->fields_buffer_allocated, sz + 1))
3111                         return -ENOMEM;
3112
3113                 memcpy(j->fields_buffer, o->field.payload, sz);
3114                 j->fields_buffer[sz] = 0;
3115
3116                 if (!field_is_valid(j->fields_buffer))
3117                         return -EBADMSG;
3118
3119                 *field = j->fields_buffer;
3120                 return 1;
3121         }
3122 }
3123
3124 _public_ void sd_journal_restart_fields(sd_journal *j) {
3125         if (!j)
3126                 return;
3127
3128         j->fields_file = NULL;
3129         j->fields_hash_table_index = 0;
3130         j->fields_offset = 0;
3131         j->fields_file_lost = false;
3132 }
3133
3134 _public_ int sd_journal_reliable_fd(sd_journal *j) {
3135         assert_return(j, -EINVAL);
3136         assert_return(!journal_pid_changed(j), -ECHILD);
3137
3138         return !j->on_network;
3139 }
3140
3141 static char *lookup_field(const char *field, void *userdata) {
3142         sd_journal *j = userdata;
3143         const void *data;
3144         size_t size, d;
3145         int r;
3146
3147         assert(field);
3148         assert(j);
3149
3150         r = sd_journal_get_data(j, field, &data, &size);
3151         if (r < 0 ||
3152             size > REPLACE_VAR_MAX)
3153                 return strdup(field);
3154
3155         d = strlen(field) + 1;
3156
3157         return strndup((const char*) data + d, size - d);
3158 }
3159
3160 _public_ int sd_journal_get_catalog(sd_journal *j, char **ret) {
3161         const void *data;
3162         size_t size;
3163         sd_id128_t id;
3164         _cleanup_free_ char *text = NULL, *cid = NULL;
3165         char *t;
3166         int r;
3167
3168         assert_return(j, -EINVAL);
3169         assert_return(!journal_pid_changed(j), -ECHILD);
3170         assert_return(ret, -EINVAL);
3171
3172         r = sd_journal_get_data(j, "MESSAGE_ID", &data, &size);
3173         if (r < 0)
3174                 return r;
3175
3176         cid = strndup((const char*) data + 11, size - 11);
3177         if (!cid)
3178                 return -ENOMEM;
3179
3180         r = sd_id128_from_string(cid, &id);
3181         if (r < 0)
3182                 return r;
3183
3184         r = catalog_get(CATALOG_DATABASE, id, &text);
3185         if (r < 0)
3186                 return r;
3187
3188         t = replace_var(text, lookup_field, j);
3189         if (!t)
3190                 return -ENOMEM;
3191
3192         *ret = t;
3193         return 0;
3194 }
3195
3196 _public_ int sd_journal_get_catalog_for_message_id(sd_id128_t id, char **ret) {
3197         assert_return(ret, -EINVAL);
3198
3199         return catalog_get(CATALOG_DATABASE, id, ret);
3200 }
3201
3202 _public_ int sd_journal_set_data_threshold(sd_journal *j, size_t sz) {
3203         assert_return(j, -EINVAL);
3204         assert_return(!journal_pid_changed(j), -ECHILD);
3205
3206         j->data_threshold = sz;
3207         return 0;
3208 }
3209
3210 _public_ int sd_journal_get_data_threshold(sd_journal *j, size_t *sz) {
3211         assert_return(j, -EINVAL);
3212         assert_return(!journal_pid_changed(j), -ECHILD);
3213         assert_return(sz, -EINVAL);
3214
3215         *sz = j->data_threshold;
3216         return 0;
3217 }
3218
3219 _public_ int sd_journal_has_runtime_files(sd_journal *j) {
3220         assert_return(j, -EINVAL);
3221
3222         return j->has_runtime_files;
3223 }
3224
3225 _public_ int sd_journal_has_persistent_files(sd_journal *j) {
3226         assert_return(j, -EINVAL);
3227
3228         return j->has_persistent_files;
3229 }