src/journal/sd-journal.c

   1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
   2
   3 /***
   4   This file is part of systemd.
   5
   6   Copyright 2011 Lennart Poettering
   7
   8   systemd is free software; you can redistribute it and/or modify it
   9   under the terms of the GNU Lesser General Public License as published by
  10   the Free Software Foundation; either version 2.1 of the License, or
  11   (at your option) any later version.
  12
  13   systemd is distributed in the hope that it will be useful, but
  14   WITHOUT ANY WARRANTY; without even the implied warranty of
  15   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  16   Lesser General Public License for more details.
  17
  18   You should have received a copy of the GNU Lesser General Public License
  19   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  20 ***/
  21
  22 #include <errno.h>
  23 #include <fcntl.h>
  24 #include <stddef.h>
  25 #include <unistd.h>
  26 #include <sys/inotify.h>
  27 #include <sys/poll.h>
  28 #include <sys/vfs.h>
  29 #include <linux/magic.h>
  30
  31 #include "sd-journal.h"
  32 #include "journal-def.h"
  33 #include "journal-file.h"
  34 #include "hashmap.h"
  35 #include "list.h"
  36 #include "strv.h"
  37 #include "path-util.h"
  38 #include "lookup3.h"
  39 #include "compress.h"
  40 #include "journal-internal.h"
  41 #include "missing.h"
  42 #include "catalog.h"
  43 #include "replace-var.h"
  44 #include "fileio.h"
  45
  46 #define JOURNAL_FILES_MAX 1024
  47
  48 #define JOURNAL_FILES_RECHECK_USEC (2 * USEC_PER_SEC)
  49
  50 #define REPLACE_VAR_MAX 256
  51
  52 #define DEFAULT_DATA_THRESHOLD (64*1024)
  53
  54 static void remove_file_real(sd_journal *j, JournalFile *f);
  55
  56 static bool journal_pid_changed(sd_journal *j) {
  57         assert(j);
  58
  59         /* We don't support people creating a journal object and
  60          * keeping it around over a fork(). Let's complain. */
  61
  62         return j->original_pid != getpid();
  63 }
  64
  65 /* We return an error here only if we didn't manage to
  66    memorize the real error. */
  67 static int set_put_error(sd_journal *j, int r) {
  68         int k;
  69
  70         if (r >= 0)
  71                 return r;
  72
  73         k = set_ensure_allocated(&j->errors, trivial_hash_func, trivial_compare_func);
  74         if (k < 0)
  75                 return k;
  76
  77         return set_put(j->errors, INT_TO_PTR(r));
  78 }
  79
  80 static void detach_location(sd_journal *j) {
  81         Iterator i;
  82         JournalFile *f;
  83
  84         assert(j);
  85
  86         j->current_file = NULL;
  87         j->current_field = 0;
  88
  89         HASHMAP_FOREACH(f, j->files, i)
  90                 f->current_offset = 0;
  91 }
  92
  93 static void reset_location(sd_journal *j) {
  94         assert(j);
  95
  96         detach_location(j);
  97         zero(j->current_location);
  98 }
  99
 100 static void init_location(Location *l, LocationType type, JournalFile *f, Object *o) {
 101         assert(l);
 102         assert(type == LOCATION_DISCRETE || type == LOCATION_SEEK);
 103         assert(f);
 104         assert(o->object.type == OBJECT_ENTRY);
 105
 106         l->type = type;
 107         l->seqnum = le64toh(o->entry.seqnum);
 108         l->seqnum_id = f->header->seqnum_id;
 109         l->realtime = le64toh(o->entry.realtime);
 110         l->monotonic = le64toh(o->entry.monotonic);
 111         l->boot_id = o->entry.boot_id;
 112         l->xor_hash = le64toh(o->entry.xor_hash);
 113
 114         l->seqnum_set = l->realtime_set = l->monotonic_set = l->xor_hash_set = true;
 115 }
 116
 117 static void set_location(sd_journal *j, LocationType type, JournalFile *f, Object *o,
 118                          direction_t direction, uint64_t offset) {
 119         assert(j);
 120         assert(type == LOCATION_DISCRETE || type == LOCATION_SEEK);
 121         assert(f);
 122         assert(o);
 123
 124         init_location(&j->current_location, type, f, o);
 125
 126         j->current_file = f;
 127         j->current_field = 0;
 128
 129         f->last_direction = direction;
 130         f->current_offset = offset;
 131 }
 132
 133 static int match_is_valid(const void *data, size_t size) {
 134         const char *b, *p;
 135
 136         assert(data);
 137
 138         if (size < 2)
 139                 return false;
 140
 141         if (startswith(data, "__"))
 142                 return false;
 143
 144         b = data;
 145         for (p = b; p < b + size; p++) {
 146
 147                 if (*p == '=')
 148                         return p > b;
 149
 150                 if (*p == '_')
 151                         continue;
 152
 153                 if (*p >= 'A' && *p <= 'Z')
 154                         continue;
 155
 156                 if (*p >= '0' && *p <= '9')
 157                         continue;
 158
 159                 return false;
 160         }
 161
 162         return false;
 163 }
 164
 165 static bool same_field(const void *_a, size_t s, const void *_b, size_t t) {
 166         const uint8_t *a = _a, *b = _b;
 167         size_t j;
 168
 169         for (j = 0; j < s && j < t; j++) {
 170
 171                 if (a[j] != b[j])
 172                         return false;
 173
 174                 if (a[j] == '=')
 175                         return true;
 176         }
 177
 178         assert_not_reached("\"=\" not found");
 179 }
 180
 181 static Match *match_new(Match *p, MatchType t) {
 182         Match *m;
 183
 184         m = new0(Match, 1);
 185         if (!m)
 186                 return NULL;
 187
 188         m->type = t;
 189
 190         if (p) {
 191                 m->parent = p;
 192                 LIST_PREPEND(matches, p->matches, m);
 193         }
 194
 195         return m;
 196 }
 197
 198 static void match_free(Match *m) {
 199         assert(m);
 200
 201         while (m->matches)
 202                 match_free(m->matches);
 203
 204         if (m->parent)
 205                 LIST_REMOVE(matches, m->parent->matches, m);
 206
 207         free(m->data);
 208         free(m);
 209 }
 210
 211 static void match_free_if_empty(Match *m) {
 212         if (!m || m->matches)
 213                 return;
 214
 215         match_free(m);
 216 }
 217
 218 _public_ int sd_journal_add_match(sd_journal *j, const void *data, size_t size) {
 219         Match *l3, *l4, *add_here = NULL, *m;
 220         le64_t le_hash;
 221
 222         assert_return(j, -EINVAL);
 223         assert_return(!journal_pid_changed(j), -ECHILD);
 224         assert_return(data, -EINVAL);
 225
 226         if (size == 0)
 227                 size = strlen(data);
 228
 229         assert_return(match_is_valid(data, size), -EINVAL);
 230
 231         /* level 0: AND term
 232          * level 1: OR terms
 233          * level 2: AND terms
 234          * level 3: OR terms
 235          * level 4: concrete matches */
 236
 237         if (!j->level0) {
 238                 j->level0 = match_new(NULL, MATCH_AND_TERM);
 239                 if (!j->level0)
 240                         return -ENOMEM;
 241         }
 242
 243         if (!j->level1) {
 244                 j->level1 = match_new(j->level0, MATCH_OR_TERM);
 245                 if (!j->level1)
 246                         return -ENOMEM;
 247         }
 248
 249         if (!j->level2) {
 250                 j->level2 = match_new(j->level1, MATCH_AND_TERM);
 251                 if (!j->level2)
 252                         return -ENOMEM;
 253         }
 254
 255         assert(j->level0->type == MATCH_AND_TERM);
 256         assert(j->level1->type == MATCH_OR_TERM);
 257         assert(j->level2->type == MATCH_AND_TERM);
 258
 259         le_hash = htole64(hash64(data, size));
 260
 261         LIST_FOREACH(matches, l3, j->level2->matches) {
 262                 assert(l3->type == MATCH_OR_TERM);
 263
 264                 LIST_FOREACH(matches, l4, l3->matches) {
 265                         assert(l4->type == MATCH_DISCRETE);
 266
 267                         /* Exactly the same match already? Then ignore
 268                          * this addition */
 269                         if (l4->le_hash == le_hash &&
 270                             l4->size == size &&
 271                             memcmp(l4->data, data, size) == 0)
 272                                 return 0;
 273
 274                         /* Same field? Then let's add this to this OR term */
 275                         if (same_field(data, size, l4->data, l4->size)) {
 276                                 add_here = l3;
 277                                 break;
 278                         }
 279                 }
 280
 281                 if (add_here)
 282                         break;
 283         }
 284
 285         if (!add_here) {
 286                 add_here = match_new(j->level2, MATCH_OR_TERM);
 287                 if (!add_here)
 288                         goto fail;
 289         }
 290
 291         m = match_new(add_here, MATCH_DISCRETE);
 292         if (!m)
 293                 goto fail;
 294
 295         m->le_hash = le_hash;
 296         m->size = size;
 297         m->data = memdup(data, size);
 298         if (!m->data)
 299                 goto fail;
 300
 301         detach_location(j);
 302
 303         return 0;
 304
 305 fail:
 306         match_free_if_empty(add_here);
 307         match_free_if_empty(j->level2);
 308         match_free_if_empty(j->level1);
 309         match_free_if_empty(j->level0);
 310
 311         return -ENOMEM;
 312 }
 313
 314 _public_ int sd_journal_add_conjunction(sd_journal *j) {
 315         assert_return(j, -EINVAL);
 316         assert_return(!journal_pid_changed(j), -ECHILD);
 317
 318         if (!j->level0)
 319                 return 0;
 320
 321         if (!j->level1)
 322                 return 0;
 323
 324         if (!j->level1->matches)
 325                 return 0;
 326
 327         j->level1 = NULL;
 328         j->level2 = NULL;
 329
 330         return 0;
 331 }
 332
 333 _public_ int sd_journal_add_disjunction(sd_journal *j) {
 334         assert_return(j, -EINVAL);
 335         assert_return(!journal_pid_changed(j), -ECHILD);
 336
 337         if (!j->level0)
 338                 return 0;
 339
 340         if (!j->level1)
 341                 return 0;
 342
 343         if (!j->level2)
 344                 return 0;
 345
 346         if (!j->level2->matches)
 347                 return 0;
 348
 349         j->level2 = NULL;
 350         return 0;
 351 }
 352
 353 static char *match_make_string(Match *m) {
 354         char *p, *r;
 355         Match *i;
 356         bool enclose = false;
 357
 358         if (!m)
 359                 return strdup("none");
 360
 361         if (m->type == MATCH_DISCRETE)
 362                 return strndup(m->data, m->size);
 363
 364         p = NULL;
 365         LIST_FOREACH(matches, i, m->matches) {
 366                 char *t, *k;
 367
 368                 t = match_make_string(i);
 369                 if (!t) {
 370                         free(p);
 371                         return NULL;
 372                 }
 373
 374                 if (p) {
 375                         k = strjoin(p, m->type == MATCH_OR_TERM ? " OR " : " AND ", t, NULL);
 376                         free(p);
 377                         free(t);
 378
 379                         if (!k)
 380                                 return NULL;
 381
 382                         p = k;
 383
 384                         enclose = true;
 385                 } else
 386                         p = t;
 387         }
 388
 389         if (enclose) {
 390                 r = strjoin("(", p, ")", NULL);
 391                 free(p);
 392                 return r;
 393         }
 394
 395         return p;
 396 }
 397
 398 char *journal_make_match_string(sd_journal *j) {
 399         assert(j);
 400
 401         return match_make_string(j->level0);
 402 }
 403
 404 _public_ void sd_journal_flush_matches(sd_journal *j) {
 405         if (!j)
 406                 return;
 407
 408         if (j->level0)
 409                 match_free(j->level0);
 410
 411         j->level0 = j->level1 = j->level2 = NULL;
 412
 413         detach_location(j);
 414 }
 415
 416 static int compare_entry_order(JournalFile *af, Object *_ao,
 417                                JournalFile *bf, uint64_t bp) {
 418
 419         uint64_t a, b;
 420         Object *ao, *bo;
 421         int r;
 422
 423         assert(af);
 424         assert(bf);
 425         assert(_ao);
 426
 427         /* The mmap cache might invalidate the object from the first
 428          * file if we look at the one from the second file. Hence
 429          * temporarily copy the header of the first one, and look at
 430          * that only. */
 431         ao = alloca(offsetof(EntryObject, items));
 432         memcpy(ao, _ao, offsetof(EntryObject, items));
 433
 434         r = journal_file_move_to_object(bf, OBJECT_ENTRY, bp, &bo);
 435         if (r < 0)
 436                 return strcmp(af->path, bf->path);
 437
 438         /* We operate on two different files here, hence we can access
 439          * two objects at the same time, which we normally can't.
 440          *
 441          * If contents and timestamps match, these entries are
 442          * identical, even if the seqnum does not match */
 443
 444         if (sd_id128_equal(ao->entry.boot_id, bo->entry.boot_id) &&
 445             ao->entry.monotonic == bo->entry.monotonic &&
 446             ao->entry.realtime == bo->entry.realtime &&
 447             ao->entry.xor_hash == bo->entry.xor_hash)
 448                 return 0;
 449
 450         if (sd_id128_equal(af->header->seqnum_id, bf->header->seqnum_id)) {
 451
 452                 /* If this is from the same seqnum source, compare
 453                  * seqnums */
 454                 a = le64toh(ao->entry.seqnum);
 455                 b = le64toh(bo->entry.seqnum);
 456
 457                 if (a < b)
 458                         return -1;
 459                 if (a > b)
 460                         return 1;
 461
 462                 /* Wow! This is weird, different data but the same
 463                  * seqnums? Something is borked, but let's make the
 464                  * best of it and compare by time. */
 465         }
 466
 467         if (sd_id128_equal(ao->entry.boot_id, bo->entry.boot_id)) {
 468
 469                 /* If the boot id matches, compare monotonic time */
 470                 a = le64toh(ao->entry.monotonic);
 471                 b = le64toh(bo->entry.monotonic);
 472
 473                 if (a < b)
 474                         return -1;
 475                 if (a > b)
 476                         return 1;
 477         }
 478
 479         /* Otherwise, compare UTC time */
 480         a = le64toh(ao->entry.realtime);
 481         b = le64toh(bo->entry.realtime);
 482
 483         if (a < b)
 484                 return -1;
 485         if (a > b)
 486                 return 1;
 487
 488         /* Finally, compare by contents */
 489         a = le64toh(ao->entry.xor_hash);
 490         b = le64toh(bo->entry.xor_hash);
 491
 492         if (a < b)
 493                 return -1;
 494         if (a > b)
 495                 return 1;
 496
 497         return 0;
 498 }
 499
 500 _pure_ static int compare_with_location(JournalFile *af, Object *ao, Location *l) {
 501         uint64_t a;
 502
 503         assert(af);
 504         assert(ao);
 505         assert(l);
 506         assert(l->type == LOCATION_DISCRETE || l->type == LOCATION_SEEK);
 507
 508         if (l->monotonic_set &&
 509             sd_id128_equal(ao->entry.boot_id, l->boot_id) &&
 510             l->realtime_set &&
 511             le64toh(ao->entry.realtime) == l->realtime &&
 512             l->xor_hash_set &&
 513             le64toh(ao->entry.xor_hash) == l->xor_hash)
 514                 return 0;
 515
 516         if (l->seqnum_set &&
 517             sd_id128_equal(af->header->seqnum_id, l->seqnum_id)) {
 518
 519                 a = le64toh(ao->entry.seqnum);
 520
 521                 if (a < l->seqnum)
 522                         return -1;
 523                 if (a > l->seqnum)
 524                         return 1;
 525         }
 526
 527         if (l->monotonic_set &&
 528             sd_id128_equal(ao->entry.boot_id, l->boot_id)) {
 529
 530                 a = le64toh(ao->entry.monotonic);
 531
 532                 if (a < l->monotonic)
 533                         return -1;
 534                 if (a > l->monotonic)
 535                         return 1;
 536         }
 537
 538         if (l->realtime_set) {
 539
 540                 a = le64toh(ao->entry.realtime);
 541
 542                 if (a < l->realtime)
 543                         return -1;
 544                 if (a > l->realtime)
 545                         return 1;
 546         }
 547
 548         if (l->xor_hash_set) {
 549                 a = le64toh(ao->entry.xor_hash);
 550
 551                 if (a < l->xor_hash)
 552                         return -1;
 553                 if (a > l->xor_hash)
 554                         return 1;
 555         }
 556
 557         return 0;
 558 }
 559
 560 static int next_for_match(
 561                 sd_journal *j,
 562                 Match *m,
 563                 JournalFile *f,
 564                 uint64_t after_offset,
 565                 direction_t direction,
 566                 Object **ret,
 567                 uint64_t *offset) {
 568
 569         int r;
 570         uint64_t np = 0;
 571         Object *n;
 572
 573         assert(j);
 574         assert(m);
 575         assert(f);
 576
 577         if (m->type == MATCH_DISCRETE) {
 578                 uint64_t dp;
 579
 580                 r = journal_file_find_data_object_with_hash(f, m->data, m->size, le64toh(m->le_hash), NULL, &dp);
 581                 if (r <= 0)
 582                         return r;
 583
 584                 return journal_file_move_to_entry_by_offset_for_data(f, dp, after_offset, direction, ret, offset);
 585
 586         } else if (m->type == MATCH_OR_TERM) {
 587                 Match *i;
 588
 589                 /* Find the earliest match beyond after_offset */
 590
 591                 LIST_FOREACH(matches, i, m->matches) {
 592                         uint64_t cp;
 593
 594                         r = next_for_match(j, i, f, after_offset, direction, NULL, &cp);
 595                         if (r < 0)
 596                                 return r;
 597                         else if (r > 0) {
 598                                 if (np == 0 || (direction == DIRECTION_DOWN ? cp < np : cp > np))
 599                                         np = cp;
 600                         }
 601                 }
 602
 603                 if (np == 0)
 604                         return 0;
 605
 606         } else if (m->type == MATCH_AND_TERM) {
 607                 Match *i, *last_moved;
 608
 609                 /* Always jump to the next matching entry and repeat
 610                  * this until we find an offset that matches for all
 611                  * matches. */
 612
 613                 if (!m->matches)
 614                         return 0;
 615
 616                 r = next_for_match(j, m->matches, f, after_offset, direction, NULL, &np);
 617                 if (r <= 0)
 618                         return r;
 619
 620                 assert(direction == DIRECTION_DOWN ? np >= after_offset : np <= after_offset);
 621                 last_moved = m->matches;
 622
 623                 LIST_LOOP_BUT_ONE(matches, i, m->matches, last_moved) {
 624                         uint64_t cp;
 625
 626                         r = next_for_match(j, i, f, np, direction, NULL, &cp);
 627                         if (r <= 0)
 628                                 return r;
 629
 630                         assert(direction == DIRECTION_DOWN ? cp >= np : cp <= np);
 631                         if (direction == DIRECTION_DOWN ? cp > np : cp < np) {
 632                                 np = cp;
 633                                 last_moved = i;
 634                         }
 635                 }
 636         }
 637
 638         assert(np > 0);
 639
 640         r = journal_file_move_to_object(f, OBJECT_ENTRY, np, &n);
 641         if (r < 0)
 642                 return r;
 643
 644         if (ret)
 645                 *ret = n;
 646         if (offset)
 647                 *offset = np;
 648
 649         return 1;
 650 }
 651
 652 static int find_location_for_match(
 653                 sd_journal *j,
 654                 Match *m,
 655                 JournalFile *f,
 656                 direction_t direction,
 657                 Object **ret,
 658                 uint64_t *offset) {
 659
 660         int r;
 661
 662         assert(j);
 663         assert(m);
 664         assert(f);
 665
 666         if (m->type == MATCH_DISCRETE) {
 667                 uint64_t dp;
 668
 669                 r = journal_file_find_data_object_with_hash(f, m->data, m->size, le64toh(m->le_hash), NULL, &dp);
 670                 if (r <= 0)
 671                         return r;
 672
 673                 /* FIXME: missing: find by monotonic */
 674
 675                 if (j->current_location.type == LOCATION_HEAD)
 676                         return journal_file_next_entry_for_data(f, NULL, 0, dp, DIRECTION_DOWN, ret, offset);
 677                 if (j->current_location.type == LOCATION_TAIL)
 678                         return journal_file_next_entry_for_data(f, NULL, 0, dp, DIRECTION_UP, ret, offset);
 679                 if (j->current_location.seqnum_set && sd_id128_equal(j->current_location.seqnum_id, f->header->seqnum_id))
 680                         return journal_file_move_to_entry_by_seqnum_for_data(f, dp, j->current_location.seqnum, direction, ret, offset);
 681                 if (j->current_location.monotonic_set) {
 682                         r = journal_file_move_to_entry_by_monotonic_for_data(f, dp, j->current_location.boot_id, j->current_location.monotonic, direction, ret, offset);
 683                         if (r != -ENOENT)
 684                                 return r;
 685                 }
 686                 if (j->current_location.realtime_set)
 687                         return journal_file_move_to_entry_by_realtime_for_data(f, dp, j->current_location.realtime, direction, ret, offset);
 688
 689                 return journal_file_next_entry_for_data(f, NULL, 0, dp, direction, ret, offset);
 690
 691         } else if (m->type == MATCH_OR_TERM) {
 692                 uint64_t np = 0;
 693                 Object *n;
 694                 Match *i;
 695
 696                 /* Find the earliest match */
 697
 698                 LIST_FOREACH(matches, i, m->matches) {
 699                         uint64_t cp;
 700
 701                         r = find_location_for_match(j, i, f, direction, NULL, &cp);
 702                         if (r < 0)
 703                                 return r;
 704                         else if (r > 0) {
 705                                 if (np == 0 || (direction == DIRECTION_DOWN ? np > cp : np < cp))
 706                                         np = cp;
 707                         }
 708                 }
 709
 710                 if (np == 0)
 711                         return 0;
 712
 713                 r = journal_file_move_to_object(f, OBJECT_ENTRY, np, &n);
 714                 if (r < 0)
 715                         return r;
 716
 717                 if (ret)
 718                         *ret = n;
 719                 if (offset)
 720                         *offset = np;
 721
 722                 return 1;
 723
 724         } else {
 725                 Match *i;
 726                 uint64_t np = 0;
 727
 728                 assert(m->type == MATCH_AND_TERM);
 729
 730                 /* First jump to the last match, and then find the
 731                  * next one where all matches match */
 732
 733                 if (!m->matches)
 734                         return 0;
 735
 736                 LIST_FOREACH(matches, i, m->matches) {
 737                         uint64_t cp;
 738
 739                         r = find_location_for_match(j, i, f, direction, NULL, &cp);
 740                         if (r <= 0)
 741                                 return r;
 742
 743                         if (np == 0 || (direction == DIRECTION_DOWN ? cp > np : cp < np))
 744                                 np = cp;
 745                 }
 746
 747                 return next_for_match(j, m, f, np, direction, ret, offset);
 748         }
 749 }
 750
 751 static int find_location_with_matches(
 752                 sd_journal *j,
 753                 JournalFile *f,
 754                 direction_t direction,
 755                 Object **ret,
 756                 uint64_t *offset) {
 757
 758         int r;
 759
 760         assert(j);
 761         assert(f);
 762         assert(ret);
 763         assert(offset);
 764
 765         if (!j->level0) {
 766                 /* No matches is simple */
 767
 768                 if (j->current_location.type == LOCATION_HEAD)
 769                         return journal_file_next_entry(f, NULL, 0, DIRECTION_DOWN, ret, offset);
 770                 if (j->current_location.type == LOCATION_TAIL)
 771                         return journal_file_next_entry(f, NULL, 0, DIRECTION_UP, ret, offset);
 772                 if (j->current_location.seqnum_set && sd_id128_equal(j->current_location.seqnum_id, f->header->seqnum_id))
 773                         return journal_file_move_to_entry_by_seqnum(f, j->current_location.seqnum, direction, ret, offset);
 774                 if (j->current_location.monotonic_set) {
 775                         r = journal_file_move_to_entry_by_monotonic(f, j->current_location.boot_id, j->current_location.monotonic, direction, ret, offset);
 776                         if (r != -ENOENT)
 777                                 return r;
 778                 }
 779                 if (j->current_location.realtime_set)
 780                         return journal_file_move_to_entry_by_realtime(f, j->current_location.realtime, direction, ret, offset);
 781
 782                 return journal_file_next_entry(f, NULL, 0, direction, ret, offset);
 783         } else
 784                 return find_location_for_match(j, j->level0, f, direction, ret, offset);
 785 }
 786
 787 static int next_with_matches(
 788                 sd_journal *j,
 789                 JournalFile *f,
 790                 direction_t direction,
 791                 Object **ret,
 792                 uint64_t *offset) {
 793
 794         Object *c;
 795         uint64_t cp;
 796
 797         assert(j);
 798         assert(f);
 799         assert(ret);
 800         assert(offset);
 801
 802         c = *ret;
 803         cp = *offset;
 804
 805         /* No matches is easy. We simple advance the file
 806          * pointer by one. */
 807         if (!j->level0)
 808                 return journal_file_next_entry(f, c, cp, direction, ret, offset);
 809
 810         /* If we have a match then we look for the next matching entry
 811          * with an offset at least one step larger */
 812         return next_for_match(j, j->level0, f, direction == DIRECTION_DOWN ? cp+1 : cp-1, direction, ret, offset);
 813 }
 814
 815 static int next_beyond_location(sd_journal *j, JournalFile *f, direction_t direction, Object **ret, uint64_t *offset) {
 816         Object *c;
 817         uint64_t cp;
 818         int r;
 819
 820         assert(j);
 821         assert(f);
 822
 823         if (f->last_direction == direction && f->current_offset > 0) {
 824                 cp = f->current_offset;
 825
 826                 r = journal_file_move_to_object(f, OBJECT_ENTRY, cp, &c);
 827                 if (r < 0)
 828                         return r;
 829
 830                 r = next_with_matches(j, f, direction, &c, &cp);
 831                 if (r <= 0)
 832                         return r;
 833         } else {
 834                 r = find_location_with_matches(j, f, direction, &c, &cp);
 835                 if (r <= 0)
 836                         return r;
 837         }
 838
 839         /* OK, we found the spot, now let's advance until an entry
 840          * that is actually different from what we were previously
 841          * looking at. This is necessary to handle entries which exist
 842          * in two (or more) journal files, and which shall all be
 843          * suppressed but one. */
 844
 845         for (;;) {
 846                 bool found;
 847
 848                 if (j->current_location.type == LOCATION_DISCRETE) {
 849                         int k;
 850
 851                         k = compare_with_location(f, c, &j->current_location);
 852                         if (direction == DIRECTION_DOWN)
 853                                 found = k > 0;
 854                         else
 855                                 found = k < 0;
 856                 } else
 857                         found = true;
 858
 859                 if (found) {
 860                         if (ret)
 861                                 *ret = c;
 862                         if (offset)
 863                                 *offset = cp;
 864                         return 1;
 865                 }
 866
 867                 r = next_with_matches(j, f, direction, &c, &cp);
 868                 if (r <= 0)
 869                         return r;
 870         }
 871 }
 872
 873 static int real_journal_next(sd_journal *j, direction_t direction) {
 874         JournalFile *f, *new_file = NULL;
 875         uint64_t new_offset = 0;
 876         uint64_t p = 0;
 877         Iterator i;
 878         Object *o;
 879         int r;
 880
 881         assert_return(j, -EINVAL);
 882         assert_return(!journal_pid_changed(j), -ECHILD);
 883
 884         HASHMAP_FOREACH(f, j->files, i) {
 885                 bool found;
 886
 887                 r = next_beyond_location(j, f, direction, &o, &p);
 888                 if (r < 0) {
 889                         log_debug("Can't iterate through %s, ignoring: %s", f->path, strerror(-r));
 890                         remove_file_real(j, f);
 891                         continue;
 892                 } else if (r == 0)
 893                         continue;
 894
 895                 if (!new_file)
 896                         found = true;
 897                 else {
 898                         int k;
 899
 900                         k = compare_entry_order(f, o, new_file, new_offset);
 901
 902                         found = direction == DIRECTION_DOWN ? k < 0 : k > 0;
 903                 }
 904
 905                 if (found) {
 906                         new_file = f;
 907                         new_offset = p;
 908                 }
 909         }
 910
 911         if (!new_file)
 912                 return 0;
 913
 914         r = journal_file_move_to_object(new_file, OBJECT_ENTRY, new_offset, &o);
 915         if (r < 0)
 916                 return r;
 917
 918         set_location(j, LOCATION_DISCRETE, new_file, o, direction, new_offset);
 919
 920         return 1;
 921 }
 922
 923 _public_ int sd_journal_next(sd_journal *j) {
 924         return real_journal_next(j, DIRECTION_DOWN);
 925 }
 926
 927 _public_ int sd_journal_previous(sd_journal *j) {
 928         return real_journal_next(j, DIRECTION_UP);
 929 }
 930
 931 static int real_journal_next_skip(sd_journal *j, direction_t direction, uint64_t skip) {
 932         int c = 0, r;
 933
 934         assert_return(j, -EINVAL);
 935         assert_return(!journal_pid_changed(j), -ECHILD);
 936
 937         if (skip == 0) {
 938                 /* If this is not a discrete skip, then at least
 939                  * resolve the current location */
 940                 if (j->current_location.type != LOCATION_DISCRETE)
 941                         return real_journal_next(j, direction);
 942
 943                 return 0;
 944         }
 945
 946         do {
 947                 r = real_journal_next(j, direction);
 948                 if (r < 0)
 949                         return r;
 950
 951                 if (r == 0)
 952                         return c;
 953
 954                 skip--;
 955                 c++;
 956         } while (skip > 0);
 957
 958         return c;
 959 }
 960
 961 _public_ int sd_journal_next_skip(sd_journal *j, uint64_t skip) {
 962         return real_journal_next_skip(j, DIRECTION_DOWN, skip);
 963 }
 964
 965 _public_ int sd_journal_previous_skip(sd_journal *j, uint64_t skip) {
 966         return real_journal_next_skip(j, DIRECTION_UP, skip);
 967 }
 968
 969 _public_ int sd_journal_get_cursor(sd_journal *j, char **cursor) {
 970         Object *o;
 971         int r;
 972         char bid[33], sid[33];
 973
 974         assert_return(j, -EINVAL);
 975         assert_return(!journal_pid_changed(j), -ECHILD);
 976         assert_return(cursor, -EINVAL);
 977
 978         if (!j->current_file || j->current_file->current_offset <= 0)
 979                 return -EADDRNOTAVAIL;
 980
 981         r = journal_file_move_to_object(j->current_file, OBJECT_ENTRY, j->current_file->current_offset, &o);
 982         if (r < 0)
 983                 return r;
 984
 985         sd_id128_to_string(j->current_file->header->seqnum_id, sid);
 986         sd_id128_to_string(o->entry.boot_id, bid);
 987
 988         if (asprintf(cursor,
 989                      "s=%s;i=%"PRIx64";b=%s;m=%"PRIx64";t=%"PRIx64";x=%"PRIx64,
 990                      sid, le64toh(o->entry.seqnum),
 991                      bid, le64toh(o->entry.monotonic),
 992                      le64toh(o->entry.realtime),
 993                      le64toh(o->entry.xor_hash)) < 0)
 994                 return -ENOMEM;
 995
 996         return 0;
 997 }
 998
 999 _public_ int sd_journal_seek_cursor(sd_journal *j, const char *cursor) {
1000         char *w, *state;
1001         size_t l;
1002         unsigned long long seqnum, monotonic, realtime, xor_hash;
1003         bool
1004                 seqnum_id_set = false,
1005                 seqnum_set = false,
1006                 boot_id_set = false,
1007                 monotonic_set = false,
1008                 realtime_set = false,
1009                 xor_hash_set = false;
1010         sd_id128_t seqnum_id, boot_id;
1011
1012         assert_return(j, -EINVAL);
1013         assert_return(!journal_pid_changed(j), -ECHILD);
1014         assert_return(!isempty(cursor), -EINVAL);
1015
1016         FOREACH_WORD_SEPARATOR(w, l, cursor, ";", state) {
1017                 char *item;
1018                 int k = 0;
1019
1020                 if (l < 2 || w[1] != '=')
1021                         return -EINVAL;
1022
1023                 item = strndup(w, l);
1024                 if (!item)
1025                         return -ENOMEM;
1026
1027                 switch (w[0]) {
1028
1029                 case 's':
1030                         seqnum_id_set = true;
1031                         k = sd_id128_from_string(item+2, &seqnum_id);
1032                         break;
1033
1034                 case 'i':
1035                         seqnum_set = true;
1036                         if (sscanf(item+2, "%llx", &seqnum) != 1)
1037                                 k = -EINVAL;
1038                         break;
1039
1040                 case 'b':
1041                         boot_id_set = true;
1042                         k = sd_id128_from_string(item+2, &boot_id);
1043                         break;
1044
1045                 case 'm':
1046                         monotonic_set = true;
1047                         if (sscanf(item+2, "%llx", &monotonic) != 1)
1048                                 k = -EINVAL;
1049                         break;
1050
1051                 case 't':
1052                         realtime_set = true;
1053                         if (sscanf(item+2, "%llx", &realtime) != 1)
1054                                 k = -EINVAL;
1055                         break;
1056
1057                 case 'x':
1058                         xor_hash_set = true;
1059                         if (sscanf(item+2, "%llx", &xor_hash) != 1)
1060                                 k = -EINVAL;
1061                         break;
1062                 }
1063
1064                 free(item);
1065
1066                 if (k < 0)
1067                         return k;
1068         }
1069
1070         if ((!seqnum_set || !seqnum_id_set) &&
1071             (!monotonic_set || !boot_id_set) &&
1072             !realtime_set)
1073                 return -EINVAL;
1074
1075         reset_location(j);
1076
1077         j->current_location.type = LOCATION_SEEK;
1078
1079         if (realtime_set) {
1080                 j->current_location.realtime = (uint64_t) realtime;
1081                 j->current_location.realtime_set = true;
1082         }
1083
1084         if (seqnum_set && seqnum_id_set) {
1085                 j->current_location.seqnum = (uint64_t) seqnum;
1086                 j->current_location.seqnum_id = seqnum_id;
1087                 j->current_location.seqnum_set = true;
1088         }
1089
1090         if (monotonic_set && boot_id_set) {
1091                 j->current_location.monotonic = (uint64_t) monotonic;
1092                 j->current_location.boot_id = boot_id;
1093                 j->current_location.monotonic_set = true;
1094         }
1095
1096         if (xor_hash_set) {
1097                 j->current_location.xor_hash = (uint64_t) xor_hash;
1098                 j->current_location.xor_hash_set = true;
1099         }
1100
1101         return 0;
1102 }
1103
1104 _public_ int sd_journal_test_cursor(sd_journal *j, const char *cursor) {
1105         int r;
1106         char *w, *state;
1107         size_t l;
1108         Object *o;
1109
1110         assert_return(j, -EINVAL);
1111         assert_return(!journal_pid_changed(j), -ECHILD);
1112         assert_return(!isempty(cursor), -EINVAL);
1113
1114         if (!j->current_file || j->current_file->current_offset <= 0)
1115                 return -EADDRNOTAVAIL;
1116
1117         r = journal_file_move_to_object(j->current_file, OBJECT_ENTRY, j->current_file->current_offset, &o);
1118         if (r < 0)
1119                 return r;
1120
1121         FOREACH_WORD_SEPARATOR(w, l, cursor, ";", state) {
1122                 _cleanup_free_ char *item = NULL;
1123                 sd_id128_t id;
1124                 unsigned long long ll;
1125                 int k = 0;
1126
1127                 if (l < 2 || w[1] != '=')
1128                         return -EINVAL;
1129
1130                 item = strndup(w, l);
1131                 if (!item)
1132                         return -ENOMEM;
1133
1134                 switch (w[0]) {
1135
1136                 case 's':
1137                         k = sd_id128_from_string(item+2, &id);
1138                         if (k < 0)
1139                                 return k;
1140                         if (!sd_id128_equal(id, j->current_file->header->seqnum_id))
1141                                 return 0;
1142                         break;
1143
1144                 case 'i':
1145                         if (sscanf(item+2, "%llx", &ll) != 1)
1146                                 return -EINVAL;
1147                         if (ll != le64toh(o->entry.seqnum))
1148                                 return 0;
1149                         break;
1150
1151                 case 'b':
1152                         k = sd_id128_from_string(item+2, &id);
1153                         if (k < 0)
1154                                 return k;
1155                         if (!sd_id128_equal(id, o->entry.boot_id))
1156                                 return 0;
1157                         break;
1158
1159                 case 'm':
1160                         if (sscanf(item+2, "%llx", &ll) != 1)
1161                                 return -EINVAL;
1162                         if (ll != le64toh(o->entry.monotonic))
1163                                 return 0;
1164                         break;
1165
1166                 case 't':
1167                         if (sscanf(item+2, "%llx", &ll) != 1)
1168                                 return -EINVAL;
1169                         if (ll != le64toh(o->entry.realtime))
1170                                 return 0;
1171                         break;
1172
1173                 case 'x':
1174                         if (sscanf(item+2, "%llx", &ll) != 1)
1175                                 return -EINVAL;
1176                         if (ll != le64toh(o->entry.xor_hash))
1177                                 return 0;
1178                         break;
1179                 }
1180         }
1181
1182         return 1;
1183 }
1184
1185
1186 _public_ int sd_journal_seek_monotonic_usec(sd_journal *j, sd_id128_t boot_id, uint64_t usec) {
1187         assert_return(j, -EINVAL);
1188         assert_return(!journal_pid_changed(j), -ECHILD);
1189
1190         reset_location(j);
1191         j->current_location.type = LOCATION_SEEK;
1192         j->current_location.boot_id = boot_id;
1193         j->current_location.monotonic = usec;
1194         j->current_location.monotonic_set = true;
1195
1196         return 0;
1197 }
1198
1199 _public_ int sd_journal_seek_realtime_usec(sd_journal *j, uint64_t usec) {
1200         assert_return(j, -EINVAL);
1201         assert_return(!journal_pid_changed(j), -ECHILD);
1202
1203         reset_location(j);
1204         j->current_location.type = LOCATION_SEEK;
1205         j->current_location.realtime = usec;
1206         j->current_location.realtime_set = true;
1207
1208         return 0;
1209 }
1210
1211 _public_ int sd_journal_seek_head(sd_journal *j) {
1212         assert_return(j, -EINVAL);
1213         assert_return(!journal_pid_changed(j), -ECHILD);
1214
1215         reset_location(j);
1216         j->current_location.type = LOCATION_HEAD;
1217
1218         return 0;
1219 }
1220
1221 _public_ int sd_journal_seek_tail(sd_journal *j) {
1222         assert_return(j, -EINVAL);
1223         assert_return(!journal_pid_changed(j), -ECHILD);
1224
1225         reset_location(j);
1226         j->current_location.type = LOCATION_TAIL;
1227
1228         return 0;
1229 }
1230
1231 static void check_network(sd_journal *j, int fd) {
1232         struct statfs sfs;
1233
1234         assert(j);
1235
1236         if (j->on_network)
1237                 return;
1238
1239         if (fstatfs(fd, &sfs) < 0)
1240                 return;
1241
1242         j->on_network =
1243                 F_TYPE_EQUAL(sfs.f_type, CIFS_MAGIC_NUMBER) ||
1244                 F_TYPE_EQUAL(sfs.f_type, CODA_SUPER_MAGIC) ||
1245                 F_TYPE_EQUAL(sfs.f_type, NCP_SUPER_MAGIC) ||
1246                 F_TYPE_EQUAL(sfs.f_type, NFS_SUPER_MAGIC) ||
1247                 F_TYPE_EQUAL(sfs.f_type, SMB_SUPER_MAGIC);
1248 }
1249
1250 static bool file_has_type_prefix(const char *prefix, const char *filename) {
1251         const char *full, *tilded, *atted;
1252
1253         full = strappenda(prefix, ".journal");
1254         tilded = strappenda(full, "~");
1255         atted = strappenda(prefix, "@");
1256
1257         return streq(filename, full) ||
1258                streq(filename, tilded) ||
1259                startswith(filename, atted);
1260 }
1261
1262 static bool file_type_wanted(int flags, const char *filename) {
1263         if (!endswith(filename, ".journal") && !endswith(filename, ".journal~"))
1264                 return false;
1265
1266         /* no flags set → every type is OK */
1267         if (!(flags & (SD_JOURNAL_SYSTEM | SD_JOURNAL_CURRENT_USER)))
1268                 return true;
1269
1270         if (flags & SD_JOURNAL_SYSTEM && file_has_type_prefix("system", filename))
1271                 return true;
1272
1273         if (flags & SD_JOURNAL_CURRENT_USER) {
1274                 char prefix[5 + DECIMAL_STR_MAX(uid_t) + 1];
1275
1276                 assert_se(snprintf(prefix, sizeof(prefix), "user-%lu", (unsigned long) getuid())
1277                           < (int) sizeof(prefix));
1278
1279                 if (file_has_type_prefix(prefix, filename))
1280                         return true;
1281         }
1282
1283         return false;
1284 }
1285
1286 static int add_any_file(sd_journal *j, const char *path) {
1287         JournalFile *f = NULL;
1288         int r;
1289
1290         assert(j);
1291         assert(path);
1292
1293         if (hashmap_get(j->files, path))
1294                 return 0;
1295
1296         if (hashmap_size(j->files) >= JOURNAL_FILES_MAX) {
1297                 log_warning("Too many open journal files, not adding %s.", path);
1298                 return set_put_error(j, -ETOOMANYREFS);
1299         }
1300
1301         r = journal_file_open(path, O_RDONLY, 0, false, false, NULL, j->mmap, NULL, &f);
1302         if (r < 0)
1303                 return r;
1304
1305         /* journal_file_dump(f); */
1306
1307         r = hashmap_put(j->files, f->path, f);
1308         if (r < 0) {
1309                 journal_file_close(f);
1310                 return r;
1311         }
1312
1313         log_debug("File %s added.", f->path);
1314
1315         check_network(j, f->fd);
1316
1317         j->current_invalidate_counter ++;
1318
1319         return 0;
1320 }
1321
1322 static int add_file(sd_journal *j, const char *prefix, const char *filename) {
1323         _cleanup_free_ char *path = NULL;
1324         int r;
1325
1326         assert(j);
1327         assert(prefix);
1328         assert(filename);
1329
1330         if (j->no_new_files ||
1331             !file_type_wanted(j->flags, filename))
1332                 return 0;
1333
1334         path = strjoin(prefix, "/", filename, NULL);
1335         if (!path)
1336                 return -ENOMEM;
1337
1338         r = add_any_file(j, path);
1339         if (r == -ENOENT)
1340                 return 0;
1341         return 0;
1342 }
1343
1344 static int remove_file(sd_journal *j, const char *prefix, const char *filename) {
1345         _cleanup_free_ char *path;
1346         JournalFile *f;
1347
1348         assert(j);
1349         assert(prefix);
1350         assert(filename);
1351
1352         path = strjoin(prefix, "/", filename, NULL);
1353         if (!path)
1354                 return -ENOMEM;
1355
1356         f = hashmap_get(j->files, path);
1357         if (!f)
1358                 return 0;
1359
1360         remove_file_real(j, f);
1361         return 0;
1362 }
1363
1364 static void remove_file_real(sd_journal *j, JournalFile *f) {
1365         assert(j);
1366         assert(f);
1367
1368         hashmap_remove(j->files, f->path);
1369
1370         log_debug("File %s removed.", f->path);
1371
1372         if (j->current_file == f) {
1373                 j->current_file = NULL;
1374                 j->current_field = 0;
1375         }
1376
1377         if (j->unique_file == f) {
1378                 j->unique_file = NULL;
1379                 j->unique_offset = 0;
1380         }
1381
1382         journal_file_close(f);
1383
1384         j->current_invalidate_counter ++;
1385 }
1386
1387 static int add_directory(sd_journal *j, const char *prefix, const char *dirname) {
1388         _cleanup_free_ char *path = NULL;
1389         int r;
1390         _cleanup_closedir_ DIR *d = NULL;
1391         sd_id128_t id, mid;
1392         Directory *m;
1393
1394         assert(j);
1395         assert(prefix);
1396         assert(dirname);
1397
1398         log_debug("Considering %s/%s.", prefix, dirname);
1399
1400         if ((j->flags & SD_JOURNAL_LOCAL_ONLY) &&
1401             (sd_id128_from_string(dirname, &id) < 0 ||
1402              sd_id128_get_machine(&mid) < 0 ||
1403              !(sd_id128_equal(id, mid) || path_startswith(prefix, "/run"))))
1404             return 0;
1405
1406         path = strjoin(prefix, "/", dirname, NULL);
1407         if (!path)
1408                 return -ENOMEM;
1409
1410         d = opendir(path);
1411         if (!d) {
1412                 log_debug("Failed to open %s: %m", path);
1413                 if (errno == ENOENT)
1414                         return 0;
1415                 return -errno;
1416         }
1417
1418         m = hashmap_get(j->directories_by_path, path);
1419         if (!m) {
1420                 m = new0(Directory, 1);
1421                 if (!m)
1422                         return -ENOMEM;
1423
1424                 m->is_root = false;
1425                 m->path = path;
1426
1427                 if (hashmap_put(j->directories_by_path, m->path, m) < 0) {
1428                         free(m);
1429                         return -ENOMEM;
1430                 }
1431
1432                 path = NULL; /* avoid freeing in cleanup */
1433                 j->current_invalidate_counter ++;
1434
1435                 log_debug("Directory %s added.", m->path);
1436
1437         } else if (m->is_root)
1438                 return 0;
1439
1440         if (m->wd <= 0 && j->inotify_fd >= 0) {
1441
1442                 m->wd = inotify_add_watch(j->inotify_fd, m->path,
1443                                           IN_CREATE|IN_MOVED_TO|IN_MODIFY|IN_ATTRIB|IN_DELETE|
1444                                           IN_DELETE_SELF|IN_MOVE_SELF|IN_UNMOUNT|IN_MOVED_FROM|
1445                                           IN_ONLYDIR);
1446
1447                 if (m->wd > 0 && hashmap_put(j->directories_by_wd, INT_TO_PTR(m->wd), m) < 0)
1448                         inotify_rm_watch(j->inotify_fd, m->wd);
1449         }
1450
1451         for (;;) {
1452                 struct dirent *de;
1453
1454                 errno = 0;
1455                 de = readdir(d);
1456                 if (!de && errno != 0) {
1457                         r = -errno;
1458                         log_debug("Failed to read directory %s: %s",
1459                                   m->path, strerror(errno));
1460                         return r;
1461                 }
1462                 if (!de)
1463                         break;
1464
1465                 if (dirent_is_file_with_suffix(de, ".journal") ||
1466                     dirent_is_file_with_suffix(de, ".journal~")) {
1467                         r = add_file(j, m->path, de->d_name);
1468                         if (r < 0) {
1469                                 log_debug("Failed to add file %s/%s: %s",
1470                                           m->path, de->d_name, strerror(-r));
1471                                 r = set_put_error(j, r);
1472                                 if (r < 0)
1473                                         return r;
1474                         }
1475                 }
1476         }
1477
1478         check_network(j, dirfd(d));
1479
1480         return 0;
1481 }
1482
1483 static int add_root_directory(sd_journal *j, const char *p) {
1484         _cleanup_closedir_ DIR *d = NULL;
1485         Directory *m;
1486         int r;
1487
1488         assert(j);
1489         assert(p);
1490
1491         if ((j->flags & SD_JOURNAL_RUNTIME_ONLY) &&
1492             !path_startswith(p, "/run"))
1493                 return -EINVAL;
1494
1495         if (j->prefix)
1496                 p = strappenda(j->prefix, p);
1497
1498         d = opendir(p);
1499         if (!d)
1500                 return -errno;
1501
1502         m = hashmap_get(j->directories_by_path, p);
1503         if (!m) {
1504                 m = new0(Directory, 1);
1505                 if (!m)
1506                         return -ENOMEM;
1507
1508                 m->is_root = true;
1509                 m->path = strdup(p);
1510                 if (!m->path) {
1511                         free(m);
1512                         return -ENOMEM;
1513                 }
1514
1515                 if (hashmap_put(j->directories_by_path, m->path, m) < 0) {
1516                         free(m->path);
1517                         free(m);
1518                         return -ENOMEM;
1519                 }
1520
1521                 j->current_invalidate_counter ++;
1522
1523                 log_debug("Root directory %s added.", m->path);
1524
1525         } else if (!m->is_root)
1526                 return 0;
1527
1528         if (m->wd <= 0 && j->inotify_fd >= 0) {
1529
1530                 m->wd = inotify_add_watch(j->inotify_fd, m->path,
1531                                           IN_CREATE|IN_MOVED_TO|IN_MODIFY|IN_ATTRIB|IN_DELETE|
1532                                           IN_ONLYDIR);
1533
1534                 if (m->wd > 0 && hashmap_put(j->directories_by_wd, INT_TO_PTR(m->wd), m) < 0)
1535                         inotify_rm_watch(j->inotify_fd, m->wd);
1536         }
1537
1538         if (j->no_new_files)
1539                 return 0;
1540
1541         for (;;) {
1542                 struct dirent *de;
1543                 sd_id128_t id;
1544
1545                 errno = 0;
1546                 de = readdir(d);
1547                 if (!de && errno != 0) {
1548                         r = -errno;
1549                         log_debug("Failed to read directory %s: %s",
1550                                   m->path, strerror(errno));
1551                         return r;
1552                 }
1553                 if (!de)
1554                         break;
1555
1556                 if (dirent_is_file_with_suffix(de, ".journal") ||
1557                     dirent_is_file_with_suffix(de, ".journal~")) {
1558                         r = add_file(j, m->path, de->d_name);
1559                         if (r < 0) {
1560                                 log_debug("Failed to add file %s/%s: %s",
1561                                           m->path, de->d_name, strerror(-r));
1562                                 r = set_put_error(j, r);
1563                                 if (r < 0)
1564                                         return r;
1565                         }
1566                 } else if ((de->d_type == DT_DIR || de->d_type == DT_LNK || de->d_type == DT_UNKNOWN) &&
1567                            sd_id128_from_string(de->d_name, &id) >= 0) {
1568
1569                         r = add_directory(j, m->path, de->d_name);
1570                         if (r < 0)
1571                                 log_debug("Failed to add directory %s/%s: %s", m->path, de->d_name, strerror(-r));
1572                 }
1573         }
1574
1575         check_network(j, dirfd(d));
1576
1577         return 0;
1578 }
1579
1580 static int remove_directory(sd_journal *j, Directory *d) {
1581         assert(j);
1582
1583         if (d->wd > 0) {
1584                 hashmap_remove(j->directories_by_wd, INT_TO_PTR(d->wd));
1585
1586                 if (j->inotify_fd >= 0)
1587                         inotify_rm_watch(j->inotify_fd, d->wd);
1588         }
1589
1590         hashmap_remove(j->directories_by_path, d->path);
1591
1592         if (d->is_root)
1593                 log_debug("Root directory %s removed.", d->path);
1594         else
1595                 log_debug("Directory %s removed.", d->path);
1596
1597         free(d->path);
1598         free(d);
1599
1600         return 0;
1601 }
1602
1603 static int add_search_paths(sd_journal *j) {
1604         int r;
1605         const char search_paths[] =
1606                 "/run/log/journal\0"
1607                 "/var/log/journal\0";
1608         const char *p;
1609
1610         assert(j);
1611
1612         /* We ignore most errors here, since the idea is to only open
1613          * what's actually accessible, and ignore the rest. */
1614
1615         NULSTR_FOREACH(p, search_paths) {
1616                 r = add_root_directory(j, p);
1617                 if (r < 0 && r != -ENOENT) {
1618                         r = set_put_error(j, r);
1619                         if (r < 0)
1620                                 return r;
1621                 }
1622         }
1623
1624         return 0;
1625 }
1626
1627 static int add_current_paths(sd_journal *j) {
1628         Iterator i;
1629         JournalFile *f;
1630
1631         assert(j);
1632         assert(j->no_new_files);
1633
1634         /* Simply adds all directories for files we have open as
1635          * "root" directories. We don't expect errors here, so we
1636          * treat them as fatal. */
1637
1638         HASHMAP_FOREACH(f, j->files, i) {
1639                 _cleanup_free_ char *dir;
1640                 int r;
1641
1642                 dir = dirname_malloc(f->path);
1643                 if (!dir)
1644                         return -ENOMEM;
1645
1646                 r = add_root_directory(j, dir);
1647                 if (r < 0) {
1648                         set_put_error(j, r);
1649                         return r;
1650                 }
1651         }
1652
1653         return 0;
1654 }
1655
1656
1657 static int allocate_inotify(sd_journal *j) {
1658         assert(j);
1659
1660         if (j->inotify_fd < 0) {
1661                 j->inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
1662                 if (j->inotify_fd < 0)
1663                         return -errno;
1664         }
1665
1666         if (!j->directories_by_wd) {
1667                 j->directories_by_wd = hashmap_new(trivial_hash_func, trivial_compare_func);
1668                 if (!j->directories_by_wd)
1669                         return -ENOMEM;
1670         }
1671
1672         return 0;
1673 }
1674
1675 static sd_journal *journal_new(int flags, const char *path) {
1676         sd_journal *j;
1677
1678         j = new0(sd_journal, 1);
1679         if (!j)
1680                 return NULL;
1681
1682         j->original_pid = getpid();
1683         j->inotify_fd = -1;
1684         j->flags = flags;
1685         j->data_threshold = DEFAULT_DATA_THRESHOLD;
1686
1687         if (path) {
1688                 j->path = strdup(path);
1689                 if (!j->path)
1690                         goto fail;
1691         }
1692
1693         j->files = hashmap_new(string_hash_func, string_compare_func);
1694         j->directories_by_path = hashmap_new(string_hash_func, string_compare_func);
1695         j->mmap = mmap_cache_new();
1696         if (!j->files || !j->directories_by_path || !j->mmap)
1697                 goto fail;
1698
1699         return j;
1700
1701 fail:
1702         sd_journal_close(j);
1703         return NULL;
1704 }
1705
1706 _public_ int sd_journal_open(sd_journal **ret, int flags) {
1707         sd_journal *j;
1708         int r;
1709
1710         assert_return(ret, -EINVAL);
1711         assert_return((flags & ~(SD_JOURNAL_LOCAL_ONLY|SD_JOURNAL_RUNTIME_ONLY|SD_JOURNAL_SYSTEM|SD_JOURNAL_CURRENT_USER)) == 0, -EINVAL);
1712
1713         j = journal_new(flags, NULL);
1714         if (!j)
1715                 return -ENOMEM;
1716
1717         r = add_search_paths(j);
1718         if (r < 0)
1719                 goto fail;
1720
1721         *ret = j;
1722         return 0;
1723
1724 fail:
1725         sd_journal_close(j);
1726
1727         return r;
1728 }
1729
1730 _public_ int sd_journal_open_container(sd_journal **ret, const char *machine, int flags) {
1731         _cleanup_free_ char *root = NULL, *class = NULL;
1732         sd_journal *j;
1733         char *p;
1734         int r;
1735
1736         assert_return(machine, -EINVAL);
1737         assert_return(ret, -EINVAL);
1738         assert_return((flags & ~(SD_JOURNAL_LOCAL_ONLY|SD_JOURNAL_SYSTEM)) == 0, -EINVAL);
1739         assert_return(filename_is_safe(machine), -EINVAL);
1740
1741         p = strappenda("/run/systemd/machines/", machine);
1742         r = parse_env_file(p, NEWLINE, "ROOT", &root, "CLASS", &class, NULL);
1743         if (r == -ENOENT)
1744                 return -EHOSTDOWN;
1745         if (r < 0)
1746                 return r;
1747         if (!root)
1748                 return -ENODATA;
1749
1750         if (!streq_ptr(class, "container"))
1751                 return -EIO;
1752
1753         j = journal_new(flags, NULL);
1754         if (!j)
1755                 return -ENOMEM;
1756
1757         j->prefix = root;
1758         root = NULL;
1759
1760         r = add_search_paths(j);
1761         if (r < 0)
1762                 goto fail;
1763
1764         *ret = j;
1765         return 0;
1766
1767 fail:
1768         sd_journal_close(j);
1769         return r;
1770 }
1771
1772 _public_ int sd_journal_open_directory(sd_journal **ret, const char *path, int flags) {
1773         sd_journal *j;
1774         int r;
1775
1776         assert_return(ret, -EINVAL);
1777         assert_return(path, -EINVAL);
1778         assert_return(flags == 0, -EINVAL);
1779
1780         j = journal_new(flags, path);
1781         if (!j)
1782                 return -ENOMEM;
1783
1784         r = add_root_directory(j, path);
1785         if (r < 0) {
1786                 set_put_error(j, r);
1787                 goto fail;
1788         }
1789
1790         *ret = j;
1791         return 0;
1792
1793 fail:
1794         sd_journal_close(j);
1795
1796         return r;
1797 }
1798
1799 _public_ int sd_journal_open_files(sd_journal **ret, const char **paths, int flags) {
1800         sd_journal *j;
1801         const char **path;
1802         int r;
1803
1804         assert_return(ret, -EINVAL);
1805         assert_return(flags == 0, -EINVAL);
1806
1807         j = journal_new(flags, NULL);
1808         if (!j)
1809                 return -ENOMEM;
1810
1811         STRV_FOREACH(path, paths) {
1812                 r = add_any_file(j, *path);
1813                 if (r < 0) {
1814                         log_error("Failed to open %s: %s", *path, strerror(-r));
1815                         goto fail;
1816                 }
1817         }
1818
1819         j->no_new_files = true;
1820
1821         *ret = j;
1822         return 0;
1823
1824 fail:
1825         sd_journal_close(j);
1826
1827         return r;
1828 }
1829
1830 _public_ void sd_journal_close(sd_journal *j) {
1831         Directory *d;
1832         JournalFile *f;
1833
1834         if (!j)
1835                 return;
1836
1837         sd_journal_flush_matches(j);
1838
1839         while ((f = hashmap_steal_first(j->files)))
1840                 journal_file_close(f);
1841
1842         hashmap_free(j->files);
1843
1844         while ((d = hashmap_first(j->directories_by_path)))
1845                 remove_directory(j, d);
1846
1847         while ((d = hashmap_first(j->directories_by_wd)))
1848                 remove_directory(j, d);
1849
1850         hashmap_free(j->directories_by_path);
1851         hashmap_free(j->directories_by_wd);
1852
1853         if (j->inotify_fd >= 0)
1854                 close_nointr_nofail(j->inotify_fd);
1855
1856         if (j->mmap) {
1857                 log_debug("mmap cache statistics: %u hit, %u miss", mmap_cache_get_hit(j->mmap), mmap_cache_get_missed(j->mmap));
1858                 mmap_cache_unref(j->mmap);
1859         }
1860
1861         free(j->path);
1862         free(j->prefix);
1863         free(j->unique_field);
1864         set_free(j->errors);
1865         free(j);
1866 }
1867
1868 _public_ int sd_journal_get_realtime_usec(sd_journal *j, uint64_t *ret) {
1869         Object *o;
1870         JournalFile *f;
1871         int r;
1872
1873         assert_return(j, -EINVAL);
1874         assert_return(!journal_pid_changed(j), -ECHILD);
1875         assert_return(ret, -EINVAL);
1876
1877         f = j->current_file;
1878         if (!f)
1879                 return -EADDRNOTAVAIL;
1880
1881         if (f->current_offset <= 0)
1882                 return -EADDRNOTAVAIL;
1883
1884         r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &o);
1885         if (r < 0)
1886                 return r;
1887
1888         *ret = le64toh(o->entry.realtime);
1889         return 0;
1890 }
1891
1892 _public_ int sd_journal_get_monotonic_usec(sd_journal *j, uint64_t *ret, sd_id128_t *ret_boot_id) {
1893         Object *o;
1894         JournalFile *f;
1895         int r;
1896         sd_id128_t id;
1897
1898         assert_return(j, -EINVAL);
1899         assert_return(!journal_pid_changed(j), -ECHILD);
1900
1901         f = j->current_file;
1902         if (!f)
1903                 return -EADDRNOTAVAIL;
1904
1905         if (f->current_offset <= 0)
1906                 return -EADDRNOTAVAIL;
1907
1908         r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &o);
1909         if (r < 0)
1910                 return r;
1911
1912         if (ret_boot_id)
1913                 *ret_boot_id = o->entry.boot_id;
1914         else {
1915                 r = sd_id128_get_boot(&id);
1916                 if (r < 0)
1917                         return r;
1918
1919                 if (!sd_id128_equal(id, o->entry.boot_id))
1920                         return -ESTALE;
1921         }
1922
1923         if (ret)
1924                 *ret = le64toh(o->entry.monotonic);
1925
1926         return 0;
1927 }
1928
1929 static bool field_is_valid(const char *field) {
1930         const char *p;
1931
1932         assert(field);
1933
1934         if (isempty(field))
1935                 return false;
1936
1937         if (startswith(field, "__"))
1938                 return false;
1939
1940         for (p = field; *p; p++) {
1941
1942                 if (*p == '_')
1943                         continue;
1944
1945                 if (*p >= 'A' && *p <= 'Z')
1946                         continue;
1947
1948                 if (*p >= '0' && *p <= '9')
1949                         continue;
1950
1951                 return false;
1952         }
1953
1954         return true;
1955 }
1956
1957 _public_ int sd_journal_get_data(sd_journal *j, const char *field, const void **data, size_t *size) {
1958         JournalFile *f;
1959         uint64_t i, n;
1960         size_t field_length;
1961         int r;
1962         Object *o;
1963
1964         assert_return(j, -EINVAL);
1965         assert_return(!journal_pid_changed(j), -ECHILD);
1966         assert_return(field, -EINVAL);
1967         assert_return(data, -EINVAL);
1968         assert_return(size, -EINVAL);
1969         assert_return(field_is_valid(field), -EINVAL);
1970
1971         f = j->current_file;
1972         if (!f)
1973                 return -EADDRNOTAVAIL;
1974
1975         if (f->current_offset <= 0)
1976                 return -EADDRNOTAVAIL;
1977
1978         r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &o);
1979         if (r < 0)
1980                 return r;
1981
1982         field_length = strlen(field);
1983
1984         n = journal_file_entry_n_items(o);
1985         for (i = 0; i < n; i++) {
1986                 uint64_t p, l;
1987                 le64_t le_hash;
1988                 size_t t;
1989
1990                 p = le64toh(o->entry.items[i].object_offset);
1991                 le_hash = o->entry.items[i].hash;
1992                 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1993                 if (r < 0)
1994                         return r;
1995
1996                 if (le_hash != o->data.hash)
1997                         return -EBADMSG;
1998
1999                 l = le64toh(o->object.size) - offsetof(Object, data.payload);
2000
2001                 if (o->object.flags & OBJECT_COMPRESSED) {
2002
2003 #ifdef HAVE_XZ
2004                         if (uncompress_startswith(o->data.payload, l,
2005                                                   &f->compress_buffer, &f->compress_buffer_size,
2006                                                   field, field_length, '=')) {
2007
2008                                 uint64_t rsize;
2009
2010                                 if (!uncompress_blob(o->data.payload, l,
2011                                                      &f->compress_buffer, &f->compress_buffer_size, &rsize,
2012                                                      j->data_threshold))
2013                                         return -EBADMSG;
2014
2015                                 *data = f->compress_buffer;
2016                                 *size = (size_t) rsize;
2017
2018                                 return 0;
2019                         }
2020 #else
2021                         return -EPROTONOSUPPORT;
2022 #endif
2023
2024                 } else if (l >= field_length+1 &&
2025                            memcmp(o->data.payload, field, field_length) == 0 &&
2026                            o->data.payload[field_length] == '=') {
2027
2028                         t = (size_t) l;
2029
2030                         if ((uint64_t) t != l)
2031                                 return -E2BIG;
2032
2033                         *data = o->data.payload;
2034                         *size = t;
2035
2036                         return 0;
2037                 }
2038
2039                 r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &o);
2040                 if (r < 0)
2041                         return r;
2042         }
2043
2044         return -ENOENT;
2045 }
2046
2047 static int return_data(sd_journal *j, JournalFile *f, Object *o, const void **data, size_t *size) {
2048         size_t t;
2049         uint64_t l;
2050
2051         l = le64toh(o->object.size) - offsetof(Object, data.payload);
2052         t = (size_t) l;
2053
2054         /* We can't read objects larger than 4G on a 32bit machine */
2055         if ((uint64_t) t != l)
2056                 return -E2BIG;
2057
2058         if (o->object.flags & OBJECT_COMPRESSED) {
2059 #ifdef HAVE_XZ
2060                 uint64_t rsize;
2061
2062                 if (!uncompress_blob(o->data.payload, l, &f->compress_buffer, &f->compress_buffer_size, &rsize, j->data_threshold))
2063                         return -EBADMSG;
2064
2065                 *data = f->compress_buffer;
2066                 *size = (size_t) rsize;
2067 #else
2068                 return -EPROTONOSUPPORT;
2069 #endif
2070         } else {
2071                 *data = o->data.payload;
2072                 *size = t;
2073         }
2074
2075         return 0;
2076 }
2077
2078 _public_ int sd_journal_enumerate_data(sd_journal *j, const void **data, size_t *size) {
2079         JournalFile *f;
2080         uint64_t p, n;
2081         le64_t le_hash;
2082         int r;
2083         Object *o;
2084
2085         assert_return(j, -EINVAL);
2086         assert_return(!journal_pid_changed(j), -ECHILD);
2087         assert_return(data, -EINVAL);
2088         assert_return(size, -EINVAL);
2089
2090         f = j->current_file;
2091         if (!f)
2092                 return -EADDRNOTAVAIL;
2093
2094         if (f->current_offset <= 0)
2095                 return -EADDRNOTAVAIL;
2096
2097         r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &o);
2098         if (r < 0)
2099                 return r;
2100
2101         n = journal_file_entry_n_items(o);
2102         if (j->current_field >= n)
2103                 return 0;
2104
2105         p = le64toh(o->entry.items[j->current_field].object_offset);
2106         le_hash = o->entry.items[j->current_field].hash;
2107         r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
2108         if (r < 0)
2109                 return r;
2110
2111         if (le_hash != o->data.hash)
2112                 return -EBADMSG;
2113
2114         r = return_data(j, f, o, data, size);
2115         if (r < 0)
2116                 return r;
2117
2118         j->current_field ++;
2119
2120         return 1;
2121 }
2122
2123 _public_ void sd_journal_restart_data(sd_journal *j) {
2124         if (!j)
2125                 return;
2126
2127         j->current_field = 0;
2128 }
2129
2130 _public_ int sd_journal_get_fd(sd_journal *j) {
2131         int r;
2132
2133         assert_return(j, -EINVAL);
2134         assert_return(!journal_pid_changed(j), -ECHILD);
2135
2136         if (j->inotify_fd >= 0)
2137                 return j->inotify_fd;
2138
2139         r = allocate_inotify(j);
2140         if (r < 0)
2141                 return r;
2142
2143         /* Iterate through all dirs again, to add them to the
2144          * inotify */
2145         if (j->no_new_files)
2146                 r = add_current_paths(j);
2147         else if (j->path)
2148                 r = add_root_directory(j, j->path);
2149         else
2150                 r = add_search_paths(j);
2151         if (r < 0)
2152                 return r;
2153
2154         return j->inotify_fd;
2155 }
2156
2157 _public_ int sd_journal_get_events(sd_journal *j) {
2158         int fd;
2159
2160         assert_return(j, -EINVAL);
2161         assert_return(!journal_pid_changed(j), -ECHILD);
2162
2163         fd = sd_journal_get_fd(j);
2164         if (fd < 0)
2165                 return fd;
2166
2167         return POLLIN;
2168 }
2169
2170 _public_ int sd_journal_get_timeout(sd_journal *j, uint64_t *timeout_usec) {
2171         int fd;
2172
2173         assert_return(j, -EINVAL);
2174         assert_return(!journal_pid_changed(j), -ECHILD);
2175         assert_return(timeout_usec, -EINVAL);
2176
2177         fd = sd_journal_get_fd(j);
2178         if (fd < 0)
2179                 return fd;
2180
2181         if (!j->on_network) {
2182                 *timeout_usec = (uint64_t) -1;
2183                 return 0;
2184         }
2185
2186         /* If we are on the network we need to regularly check for
2187          * changes manually */
2188
2189         *timeout_usec = j->last_process_usec + JOURNAL_FILES_RECHECK_USEC;
2190         return 1;
2191 }
2192
2193 static void process_inotify_event(sd_journal *j, struct inotify_event *e) {
2194         Directory *d;
2195         int r;
2196
2197         assert(j);
2198         assert(e);
2199
2200         /* Is this a subdirectory we watch? */
2201         d = hashmap_get(j->directories_by_wd, INT_TO_PTR(e->wd));
2202         if (d) {
2203                 sd_id128_t id;
2204
2205                 if (!(e->mask & IN_ISDIR) && e->len > 0 &&
2206                     (endswith(e->name, ".journal") ||
2207                      endswith(e->name, ".journal~"))) {
2208
2209                         /* Event for a journal file */
2210
2211                         if (e->mask & (IN_CREATE|IN_MOVED_TO|IN_MODIFY|IN_ATTRIB)) {
2212                                 r = add_file(j, d->path, e->name);
2213                                 if (r < 0) {
2214                                         log_debug("Failed to add file %s/%s: %s",
2215                                                   d->path, e->name, strerror(-r));
2216                                         set_put_error(j, r);
2217                                 }
2218
2219                         } else if (e->mask & (IN_DELETE|IN_MOVED_FROM|IN_UNMOUNT)) {
2220
2221                                 r = remove_file(j, d->path, e->name);
2222                                 if (r < 0)
2223                                         log_debug("Failed to remove file %s/%s: %s", d->path, e->name, strerror(-r));
2224                         }
2225
2226                 } else if (!d->is_root && e->len == 0) {
2227
2228                         /* Event for a subdirectory */
2229
2230                         if (e->mask & (IN_DELETE_SELF|IN_MOVE_SELF|IN_UNMOUNT)) {
2231                                 r = remove_directory(j, d);
2232                                 if (r < 0)
2233                                         log_debug("Failed to remove directory %s: %s", d->path, strerror(-r));
2234                         }
2235
2236
2237                 } else if (d->is_root && (e->mask & IN_ISDIR) && e->len > 0 && sd_id128_from_string(e->name, &id) >= 0) {
2238
2239                         /* Event for root directory */
2240
2241                         if (e->mask & (IN_CREATE|IN_MOVED_TO|IN_MODIFY|IN_ATTRIB)) {
2242                                 r = add_directory(j, d->path, e->name);
2243                                 if (r < 0)
2244                                         log_debug("Failed to add directory %s/%s: %s", d->path, e->name, strerror(-r));
2245                         }
2246                 }
2247
2248                 return;
2249         }
2250
2251         if (e->mask & IN_IGNORED)
2252                 return;
2253
2254         log_warning("Unknown inotify event.");
2255 }
2256
2257 static int determine_change(sd_journal *j) {
2258         bool b;
2259
2260         assert(j);
2261
2262         b = j->current_invalidate_counter != j->last_invalidate_counter;
2263         j->last_invalidate_counter = j->current_invalidate_counter;
2264
2265         return b ? SD_JOURNAL_INVALIDATE : SD_JOURNAL_APPEND;
2266 }
2267
2268 _public_ int sd_journal_process(sd_journal *j) {
2269         uint8_t buffer[sizeof(struct inotify_event) + FILENAME_MAX] _alignas_(struct inotify_event);
2270         bool got_something = false;
2271
2272         assert_return(j, -EINVAL);
2273         assert_return(!journal_pid_changed(j), -ECHILD);
2274
2275         j->last_process_usec = now(CLOCK_MONOTONIC);
2276
2277         for (;;) {
2278                 struct inotify_event *e;
2279                 ssize_t l;
2280
2281                 l = read(j->inotify_fd, buffer, sizeof(buffer));
2282                 if (l < 0) {
2283                         if (errno == EAGAIN || errno == EINTR)
2284                                 return got_something ? determine_change(j) : SD_JOURNAL_NOP;
2285
2286                         return -errno;
2287                 }
2288
2289                 got_something = true;
2290
2291                 e = (struct inotify_event*) buffer;
2292                 while (l > 0) {
2293                         size_t step;
2294
2295                         process_inotify_event(j, e);
2296
2297                         step = sizeof(struct inotify_event) + e->len;
2298                         assert(step <= (size_t) l);
2299
2300                         e = (struct inotify_event*) ((uint8_t*) e + step);
2301                         l -= step;
2302                 }
2303         }
2304 }
2305
2306 _public_ int sd_journal_wait(sd_journal *j, uint64_t timeout_usec) {
2307         int r;
2308         uint64_t t;
2309
2310         assert_return(j, -EINVAL);
2311         assert_return(!journal_pid_changed(j), -ECHILD);
2312
2313         if (j->inotify_fd < 0) {
2314
2315                 /* This is the first invocation, hence create the
2316                  * inotify watch */
2317                 r = sd_journal_get_fd(j);
2318                 if (r < 0)
2319                         return r;
2320
2321                 /* The journal might have changed since the context
2322                  * object was created and we weren't watching before,
2323                  * hence don't wait for anything, and return
2324                  * immediately. */
2325                 return determine_change(j);
2326         }
2327
2328         r = sd_journal_get_timeout(j, &t);
2329         if (r < 0)
2330                 return r;
2331
2332         if (t != (uint64_t) -1) {
2333                 usec_t n;
2334
2335                 n = now(CLOCK_MONOTONIC);
2336                 t = t > n ? t - n : 0;
2337
2338                 if (timeout_usec == (uint64_t) -1 || timeout_usec > t)
2339                         timeout_usec = t;
2340         }
2341
2342         do {
2343                 r = fd_wait_for_event(j->inotify_fd, POLLIN, timeout_usec);
2344         } while (r == -EINTR);
2345
2346         if (r < 0)
2347                 return r;
2348
2349         return sd_journal_process(j);
2350 }
2351
2352 _public_ int sd_journal_get_cutoff_realtime_usec(sd_journal *j, uint64_t *from, uint64_t *to) {
2353         Iterator i;
2354         JournalFile *f;
2355         bool first = true;
2356         uint64_t fmin = 0, tmax = 0;
2357         int r;
2358
2359         assert_return(j, -EINVAL);
2360         assert_return(!journal_pid_changed(j), -ECHILD);
2361         assert_return(from || to, -EINVAL);
2362         assert_return(from != to, -EINVAL);
2363
2364         HASHMAP_FOREACH(f, j->files, i) {
2365                 usec_t fr, t;
2366
2367                 r = journal_file_get_cutoff_realtime_usec(f, &fr, &t);
2368                 if (r == -ENOENT)
2369                         continue;
2370                 if (r < 0)
2371                         return r;
2372                 if (r == 0)
2373                         continue;
2374
2375                 if (first) {
2376                         fmin = fr;
2377                         tmax = t;
2378                         first = false;
2379                 } else {
2380                         fmin = MIN(fr, fmin);
2381                         tmax = MAX(t, tmax);
2382                 }
2383         }
2384
2385         if (from)
2386                 *from = fmin;
2387         if (to)
2388                 *to = tmax;
2389
2390         return first ? 0 : 1;
2391 }
2392
2393 _public_ int sd_journal_get_cutoff_monotonic_usec(sd_journal *j, sd_id128_t boot_id, uint64_t *from, uint64_t *to) {
2394         Iterator i;
2395         JournalFile *f;
2396         bool first = true;
2397         int r;
2398
2399         assert_return(j, -EINVAL);
2400         assert_return(!journal_pid_changed(j), -ECHILD);
2401         assert_return(from || to, -EINVAL);
2402         assert_return(from != to, -EINVAL);
2403
2404         HASHMAP_FOREACH(f, j->files, i) {
2405                 usec_t fr, t;
2406
2407                 r = journal_file_get_cutoff_monotonic_usec(f, boot_id, &fr, &t);
2408                 if (r == -ENOENT)
2409                         continue;
2410                 if (r < 0)
2411                         return r;
2412                 if (r == 0)
2413                         continue;
2414
2415                 if (first) {
2416                         if (from)
2417                                 *from = fr;
2418                         if (to)
2419                                 *to = t;
2420                         first = false;
2421                 } else {
2422                         if (from)
2423                                 *from = MIN(fr, *from);
2424                         if (to)
2425                                 *to = MAX(t, *to);
2426                 }
2427         }
2428
2429         return first ? 0 : 1;
2430 }
2431
2432 void journal_print_header(sd_journal *j) {
2433         Iterator i;
2434         JournalFile *f;
2435         bool newline = false;
2436
2437         assert(j);
2438
2439         HASHMAP_FOREACH(f, j->files, i) {
2440                 if (newline)
2441                         putchar('\n');
2442                 else
2443                         newline = true;
2444
2445                 journal_file_print_header(f);
2446         }
2447 }
2448
2449 _public_ int sd_journal_get_usage(sd_journal *j, uint64_t *bytes) {
2450         Iterator i;
2451         JournalFile *f;
2452         uint64_t sum = 0;
2453
2454         assert_return(j, -EINVAL);
2455         assert_return(!journal_pid_changed(j), -ECHILD);
2456         assert_return(bytes, -EINVAL);
2457
2458         HASHMAP_FOREACH(f, j->files, i) {
2459                 struct stat st;
2460
2461                 if (fstat(f->fd, &st) < 0)
2462                         return -errno;
2463
2464                 sum += (uint64_t) st.st_blocks * 512ULL;
2465         }
2466
2467         *bytes = sum;
2468         return 0;
2469 }
2470
2471 _public_ int sd_journal_query_unique(sd_journal *j, const char *field) {
2472         char *f;
2473
2474         assert_return(j, -EINVAL);
2475         assert_return(!journal_pid_changed(j), -ECHILD);
2476         assert_return(!isempty(field), -EINVAL);
2477         assert_return(field_is_valid(field), -EINVAL);
2478
2479         f = strdup(field);
2480         if (!f)
2481                 return -ENOMEM;
2482
2483         free(j->unique_field);
2484         j->unique_field = f;
2485         j->unique_file = NULL;
2486         j->unique_offset = 0;
2487
2488         return 0;
2489 }
2490
2491 _public_ int sd_journal_enumerate_unique(sd_journal *j, const void **data, size_t *l) {
2492         size_t k;
2493
2494         assert_return(j, -EINVAL);
2495         assert_return(!journal_pid_changed(j), -ECHILD);
2496         assert_return(data, -EINVAL);
2497         assert_return(l, -EINVAL);
2498         assert_return(j->unique_field, -EINVAL);
2499
2500         k = strlen(j->unique_field);
2501
2502         if (!j->unique_file) {
2503                 j->unique_file = hashmap_first(j->files);
2504                 if (!j->unique_file)
2505                         return 0;
2506                 j->unique_offset = 0;
2507         }
2508
2509         for (;;) {
2510                 JournalFile *of;
2511                 Iterator i;
2512                 Object *o;
2513                 const void *odata;
2514                 size_t ol;
2515                 bool found;
2516                 int r;
2517
2518                 /* Proceed to next data object in the field's linked list */
2519                 if (j->unique_offset == 0) {
2520                         r = journal_file_find_field_object(j->unique_file, j->unique_field, k, &o, NULL);
2521                         if (r < 0)
2522                                 return r;
2523
2524                         j->unique_offset = r > 0 ? le64toh(o->field.head_data_offset) : 0;
2525                 } else {
2526                         r = journal_file_move_to_object(j->unique_file, OBJECT_DATA, j->unique_offset, &o);
2527                         if (r < 0)
2528                                 return r;
2529
2530                         j->unique_offset = le64toh(o->data.next_field_offset);
2531                 }
2532
2533                 /* We reached the end of the list? Then start again, with the next file */
2534                 if (j->unique_offset == 0) {
2535                         JournalFile *n;
2536
2537                         n = hashmap_next(j->files, j->unique_file->path);
2538                         if (!n)
2539                                 return 0;
2540
2541                         j->unique_file = n;
2542                         continue;
2543                 }
2544
2545                 /* We do not use the type context here, but 0 instead,
2546                  * so that we can look at this data object at the same
2547                  * time as one on another file */
2548                 r = journal_file_move_to_object(j->unique_file, 0, j->unique_offset, &o);
2549                 if (r < 0)
2550                         return r;
2551
2552                 /* Let's do the type check by hand, since we used 0 context above. */
2553                 if (o->object.type != OBJECT_DATA) {
2554                         log_error("%s:offset " OFSfmt ": object has type %d, expected %d",
2555                                   j->unique_file->path, j->unique_offset,
2556                                   o->object.type, OBJECT_DATA);
2557                         return -EBADMSG;
2558                 }
2559
2560                 r = journal_file_object_keep(j->unique_file, o, j->unique_offset);
2561                 if (r < 0)
2562                         return r;
2563
2564                 r = return_data(j, j->unique_file, o, &odata, &ol);
2565                 if (r < 0)
2566                         return r;
2567
2568                 /* OK, now let's see if we already returned this data
2569                  * object by checking if it exists in the earlier
2570                  * traversed files. */
2571                 found = false;
2572                 HASHMAP_FOREACH(of, j->files, i) {
2573                         Object *oo;
2574                         uint64_t op;
2575
2576                         if (of == j->unique_file)
2577                                 break;
2578
2579                         /* Skip this file it didn't have any fields
2580                          * indexed */
2581                         if (JOURNAL_HEADER_CONTAINS(of->header, n_fields) &&
2582                             le64toh(of->header->n_fields) <= 0)
2583                                 continue;
2584
2585                         r = journal_file_find_data_object_with_hash(of, odata, ol, le64toh(o->data.hash), &oo, &op);
2586                         if (r < 0)
2587                                 return r;
2588
2589                         if (r > 0)
2590                                 found = true;
2591                 }
2592
2593                 if (found)
2594                         continue;
2595
2596                 r = journal_file_object_release(j->unique_file, o, j->unique_offset);
2597                 if (r < 0)
2598                         return r;
2599
2600                 r = return_data(j, j->unique_file, o, data, l);
2601                 if (r < 0)
2602                         return r;
2603
2604                 return 1;
2605         }
2606 }
2607
2608 _public_ void sd_journal_restart_unique(sd_journal *j) {
2609         if (!j)
2610                 return;
2611
2612         j->unique_file = NULL;
2613         j->unique_offset = 0;
2614 }
2615
2616 _public_ int sd_journal_reliable_fd(sd_journal *j) {
2617         assert_return(j, -EINVAL);
2618         assert_return(!journal_pid_changed(j), -ECHILD);
2619
2620         return !j->on_network;
2621 }
2622
2623 static char *lookup_field(const char *field, void *userdata) {
2624         sd_journal *j = userdata;
2625         const void *data;
2626         size_t size, d;
2627         int r;
2628
2629         assert(field);
2630         assert(j);
2631
2632         r = sd_journal_get_data(j, field, &data, &size);
2633         if (r < 0 ||
2634             size > REPLACE_VAR_MAX)
2635                 return strdup(field);
2636
2637         d = strlen(field) + 1;
2638
2639         return strndup((const char*) data + d, size - d);
2640 }
2641
2642 _public_ int sd_journal_get_catalog(sd_journal *j, char **ret) {
2643         const void *data;
2644         size_t size;
2645         sd_id128_t id;
2646         _cleanup_free_ char *text = NULL, *cid = NULL;
2647         char *t;
2648         int r;
2649
2650         assert_return(j, -EINVAL);
2651         assert_return(!journal_pid_changed(j), -ECHILD);
2652         assert_return(ret, -EINVAL);
2653
2654         r = sd_journal_get_data(j, "MESSAGE_ID", &data, &size);
2655         if (r < 0)
2656                 return r;
2657
2658         cid = strndup((const char*) data + 11, size - 11);
2659         if (!cid)
2660                 return -ENOMEM;
2661
2662         r = sd_id128_from_string(cid, &id);
2663         if (r < 0)
2664                 return r;
2665
2666         r = catalog_get(CATALOG_DATABASE, id, &text);
2667         if (r < 0)
2668                 return r;
2669
2670         t = replace_var(text, lookup_field, j);
2671         if (!t)
2672                 return -ENOMEM;
2673
2674         *ret = t;
2675         return 0;
2676 }
2677
2678 _public_ int sd_journal_get_catalog_for_message_id(sd_id128_t id, char **ret) {
2679         assert_return(ret, -EINVAL);
2680
2681         return catalog_get(CATALOG_DATABASE, id, ret);
2682 }
2683
2684 _public_ int sd_journal_set_data_threshold(sd_journal *j, size_t sz) {
2685         assert_return(j, -EINVAL);
2686         assert_return(!journal_pid_changed(j), -ECHILD);
2687
2688         j->data_threshold = sz;
2689         return 0;
2690 }
2691
2692 _public_ int sd_journal_get_data_threshold(sd_journal *j, size_t *sz) {
2693         assert_return(j, -EINVAL);
2694         assert_return(!journal_pid_changed(j), -ECHILD);
2695         assert_return(sz, -EINVAL);
2696
2697         *sz = j->data_threshold;
2698         return 0;
2699 }