src/journal/journal-file.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 General Public License as published by
  10   the Free Software Foundation; either version 2 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   General Public License for more details.
  17
  18   You should have received a copy of the GNU General Public License
  19   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  20 ***/
  21
  22 #include <sys/mman.h>
  23 #include <errno.h>
  24 #include <sys/uio.h>
  25 #include <unistd.h>
  26 #include <sys/statvfs.h>
  27 #include <fcntl.h>
  28 #include <stddef.h>
  29
  30 #include "journal-def.h"
  31 #include "journal-file.h"
  32 #include "lookup3.h"
  33
  34 #define DEFAULT_ARENA_MAX_SIZE (16ULL*1024ULL*1024ULL*1024ULL)
  35 #define DEFAULT_ARENA_MIN_SIZE (256ULL*1024ULL)
  36 #define DEFAULT_ARENA_KEEP_FREE (1ULL*1024ULL*1024ULL)
  37
  38 #define DEFAULT_MAX_USE (16ULL*1024ULL*1024ULL*16ULL)
  39
  40 #define DEFAULT_HASH_TABLE_SIZE (2047ULL*16ULL)
  41 #define DEFAULT_BISECT_TABLE_SIZE ((DEFAULT_ARENA_MAX_SIZE/(64ULL*1024ULL))*8ULL)
  42
  43 #define DEFAULT_WINDOW_SIZE (128ULL*1024ULL*1024ULL)
  44
  45 static const char signature[] = { 'L', 'P', 'K', 'S', 'H', 'H', 'R', 'H' };
  46
  47 #define ALIGN64(x) (((x) + 7ULL) & ~7ULL)
  48
  49 void journal_file_close(JournalFile *f) {
  50         assert(f);
  51
  52         if (f->header) {
  53                 if (f->writable && f->header->state == htole32(STATE_ONLINE))
  54                         f->header->state = htole32(STATE_OFFLINE);
  55
  56                 munmap(f->header, PAGE_ALIGN(sizeof(Header)));
  57         }
  58
  59         if (f->hash_table_window)
  60                 munmap(f->hash_table_window, f->hash_table_window_size);
  61
  62         if (f->bisect_table_window)
  63                 munmap(f->bisect_table_window, f->bisect_table_window_size);
  64
  65         if (f->window)
  66                 munmap(f->window, f->window_size);
  67
  68         if (f->fd >= 0)
  69                 close_nointr_nofail(f->fd);
  70
  71         free(f->path);
  72         free(f);
  73 }
  74
  75 static int journal_file_init_header(JournalFile *f, JournalFile *template) {
  76         Header h;
  77         ssize_t k;
  78         int r;
  79
  80         assert(f);
  81
  82         zero(h);
  83         memcpy(h.signature, signature, 8);
  84         h.arena_offset = htole64(ALIGN64(sizeof(h)));
  85         h.arena_max_size = htole64(DEFAULT_ARENA_MAX_SIZE);
  86         h.arena_min_size = htole64(DEFAULT_ARENA_MIN_SIZE);
  87         h.arena_keep_free = htole64(DEFAULT_ARENA_KEEP_FREE);
  88
  89         r = sd_id128_randomize(&h.file_id);
  90         if (r < 0)
  91                 return r;
  92
  93         if (template) {
  94                 h.seqnum_id = template->header->seqnum_id;
  95                 h.seqnum = template->header->seqnum;
  96         } else
  97                 h.seqnum_id = h.file_id;
  98
  99         k = pwrite(f->fd, &h, sizeof(h), 0);
 100         if (k < 0)
 101                 return -errno;
 102
 103         if (k != sizeof(h))
 104                 return -EIO;
 105
 106         return 0;
 107 }
 108
 109 static int journal_file_refresh_header(JournalFile *f) {
 110         int r;
 111
 112         assert(f);
 113
 114         r = sd_id128_get_machine(&f->header->machine_id);
 115         if (r < 0)
 116                 return r;
 117
 118         r = sd_id128_get_boot(&f->header->boot_id);
 119         if (r < 0)
 120                 return r;
 121
 122         f->header->state = htole32(STATE_ONLINE);
 123         return 0;
 124 }
 125
 126 static int journal_file_verify_header(JournalFile *f) {
 127         assert(f);
 128
 129         if (memcmp(f->header, signature, 8))
 130                 return -EBADMSG;
 131
 132         if (f->header->incompatible_flags != 0)
 133                 return -EPROTONOSUPPORT;
 134
 135         if ((uint64_t) f->last_stat.st_size < (le64toh(f->header->arena_offset) + le64toh(f->header->arena_size)))
 136                 return -ENODATA;
 137
 138         if (f->writable) {
 139                 uint32_t state;
 140                 sd_id128_t machine_id;
 141                 int r;
 142
 143                 r = sd_id128_get_machine(&machine_id);
 144                 if (r < 0)
 145                         return r;
 146
 147                 if (!sd_id128_equal(machine_id, f->header->machine_id))
 148                         return -EHOSTDOWN;
 149
 150                 state = le32toh(f->header->state);
 151
 152                 if (state == STATE_ONLINE)
 153                         log_debug("Journal file %s is already online. Assuming unclean closing. Ignoring.", f->path);
 154                 else if (state == STATE_ARCHIVED)
 155                         return -ESHUTDOWN;
 156                 else if (state != STATE_OFFLINE)
 157                         log_debug("Journal file %s has unknown state %u. Ignoring.", f->path, state);
 158         }
 159
 160         return 0;
 161 }
 162
 163 static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size) {
 164         uint64_t asize;
 165         uint64_t old_size, new_size;
 166
 167         assert(f);
 168
 169         if (offset < le64toh(f->header->arena_offset))
 170                 return -EINVAL;
 171
 172         new_size = PAGE_ALIGN(offset + size);
 173
 174         /* We assume that this file is not sparse, and we know that
 175          * for sure, since we always call posix_fallocate()
 176          * ourselves */
 177
 178         old_size =
 179                 le64toh(f->header->arena_offset) +
 180                 le64toh(f->header->arena_size);
 181
 182         if (old_size >= new_size)
 183                 return 0;
 184
 185         asize = new_size - le64toh(f->header->arena_offset);
 186
 187         if (asize > le64toh(f->header->arena_min_size)) {
 188                 struct statvfs svfs;
 189
 190                 if (fstatvfs(f->fd, &svfs) >= 0) {
 191                         uint64_t available;
 192
 193                         available = svfs.f_bfree * svfs.f_bsize;
 194
 195                         if (available >= f->header->arena_keep_free)
 196                                 available -= f->header->arena_keep_free;
 197                         else
 198                                 available = 0;
 199
 200                         if (new_size - old_size > available)
 201                                 return -E2BIG;
 202                 }
 203         }
 204
 205         if (asize > le64toh(f->header->arena_max_size))
 206                 return -E2BIG;
 207
 208         if (posix_fallocate(f->fd, old_size, new_size - old_size) < 0)
 209                 return -errno;
 210
 211         if (fstat(f->fd, &f->last_stat) < 0)
 212                 return -errno;
 213
 214         f->header->arena_size = htole64(asize);
 215
 216         return 0;
 217 }
 218
 219 static int journal_file_map(
 220                 JournalFile *f,
 221                 uint64_t offset,
 222                 uint64_t size,
 223                 void **_window,
 224                 uint64_t *_woffset,
 225                 uint64_t *_wsize,
 226                 void **ret) {
 227
 228         uint64_t woffset, wsize;
 229         void *window;
 230
 231         assert(f);
 232         assert(size > 0);
 233         assert(ret);
 234
 235         woffset = offset & ~((uint64_t) page_size() - 1ULL);
 236         wsize = size + (offset - woffset);
 237         wsize = PAGE_ALIGN(wsize);
 238
 239         window = mmap(NULL, wsize, f->prot, MAP_SHARED, f->fd, woffset);
 240         if (window == MAP_FAILED)
 241                 return -errno;
 242
 243         if (_window)
 244                 *_window = window;
 245
 246         if (_woffset)
 247                 *_woffset = woffset;
 248
 249         if (_wsize)
 250                 *_wsize = wsize;
 251
 252         *ret = (uint8_t*) window + (offset - woffset);
 253
 254         return 0;
 255 }
 256
 257 static int journal_file_move_to(JournalFile *f, uint64_t offset, uint64_t size, void **ret) {
 258         void *p;
 259         uint64_t delta;
 260         int r;
 261
 262         assert(f);
 263         assert(ret);
 264
 265         if (_likely_(f->window &&
 266                      f->window_offset <= offset &&
 267                      f->window_offset+f->window_size >= offset + size)) {
 268
 269                 *ret = (uint8_t*) f->window + (offset - f->window_offset);
 270                 return 0;
 271         }
 272
 273         if (f->window) {
 274                 if (munmap(f->window, f->window_size) < 0)
 275                         return -errno;
 276
 277                 f->window = NULL;
 278                 f->window_size = f->window_offset = 0;
 279         }
 280
 281         if (size < DEFAULT_WINDOW_SIZE) {
 282                 /* If the default window size is larger then what was
 283                  * asked for extend the mapping a bit in the hope to
 284                  * minimize needed remappings later on. We add half
 285                  * the window space before and half behind the
 286                  * requested mapping */
 287
 288                 delta = PAGE_ALIGN((DEFAULT_WINDOW_SIZE - size) / 2);
 289
 290                 if (offset < delta)
 291                         delta = offset;
 292
 293                 offset -= delta;
 294                 size += (DEFAULT_WINDOW_SIZE - delta);
 295         } else
 296                 delta = 0;
 297
 298         r = journal_file_map(f,
 299                              offset, size,
 300                              &f->window, &f->window_offset, &f->window_size,
 301                              & p);
 302
 303         if (r < 0)
 304                 return r;
 305
 306         *ret = (uint8_t*) p + delta;
 307         return 0;
 308 }
 309
 310 static bool verify_hash(Object *o) {
 311         uint64_t t;
 312
 313         assert(o);
 314
 315         t = le64toh(o->object.type);
 316         if (t == OBJECT_DATA) {
 317                 uint64_t s, h1, h2;
 318
 319                 s = le64toh(o->object.size);
 320
 321                 h1 = le64toh(o->data.hash);
 322                 h2 = hash64(o->data.payload, s - offsetof(Object, data.payload));
 323
 324                 return h1 == h2;
 325         }
 326
 327         return true;
 328 }
 329
 330 int journal_file_move_to_object(JournalFile *f, uint64_t offset, int type, Object **ret) {
 331         int r;
 332         void *t;
 333         Object *o;
 334         uint64_t s;
 335
 336         assert(f);
 337         assert(ret);
 338
 339         r = journal_file_move_to(f, offset, sizeof(ObjectHeader), &t);
 340         if (r < 0)
 341                 return r;
 342
 343         o = (Object*) t;
 344         s = le64toh(o->object.size);
 345
 346         if (s < sizeof(ObjectHeader))
 347                 return -EBADMSG;
 348
 349         if (type >= 0 && le64toh(o->object.type) != type)
 350                 return -EBADMSG;
 351
 352         if (s > sizeof(ObjectHeader)) {
 353                 r = journal_file_move_to(f, offset, s, &t);
 354                 if (r < 0)
 355                         return r;
 356
 357                 o = (Object*) t;
 358         }
 359
 360         if (!verify_hash(o))
 361                 return -EBADMSG;
 362
 363         *ret = o;
 364         return 0;
 365 }
 366
 367 static uint64_t journal_file_seqnum(JournalFile *f) {
 368         uint64_t r;
 369
 370         assert(f);
 371
 372         r = le64toh(f->header->seqnum) + 1;
 373         f->header->seqnum = htole64(r);
 374
 375         return r;
 376 }
 377
 378 static int journal_file_append_object(JournalFile *f, uint64_t size, Object **ret, uint64_t *offset) {
 379         int r;
 380         uint64_t p;
 381         Object *tail, *o;
 382         void *t;
 383
 384         assert(f);
 385         assert(size >= sizeof(ObjectHeader));
 386         assert(offset);
 387         assert(ret);
 388
 389         p = le64toh(f->header->tail_object_offset);
 390
 391         if (p == 0)
 392                 p = le64toh(f->header->arena_offset);
 393         else {
 394                 r = journal_file_move_to_object(f, p, -1, &tail);
 395                 if (r < 0)
 396                         return r;
 397
 398                 p += ALIGN64(le64toh(tail->object.size));
 399         }
 400
 401         r = journal_file_allocate(f, p, size);
 402         if (r < 0)
 403                 return r;
 404
 405         r = journal_file_move_to(f, p, size, &t);
 406         if (r < 0)
 407                 return r;
 408
 409         o = (Object*) t;
 410
 411         zero(o->object);
 412         o->object.type = htole64(OBJECT_UNUSED);
 413         zero(o->object.reserved);
 414         o->object.size = htole64(size);
 415
 416         f->header->tail_object_offset = htole64(p);
 417         if (f->header->head_object_offset == 0)
 418                 f->header->head_object_offset = htole64(p);
 419
 420         f->header->n_objects = htole64(le64toh(f->header->n_objects) + 1);
 421
 422         *ret = o;
 423         *offset = p;
 424
 425         return 0;
 426 }
 427
 428 static int journal_file_setup_hash_table(JournalFile *f) {
 429         uint64_t s, p;
 430         Object *o;
 431         int r;
 432
 433         assert(f);
 434
 435         s = DEFAULT_HASH_TABLE_SIZE;
 436         r = journal_file_append_object(f, offsetof(Object, hash_table.table) + s, &o, &p);
 437         if (r < 0)
 438                 return r;
 439
 440         o->object.type = htole64(OBJECT_HASH_TABLE);
 441         memset(o->hash_table.table, 0, s);
 442
 443         f->header->hash_table_offset = htole64(p + offsetof(Object, hash_table.table));
 444         f->header->hash_table_size = htole64(s);
 445
 446         return 0;
 447 }
 448
 449 static int journal_file_setup_bisect_table(JournalFile *f) {
 450         uint64_t s, p;
 451         Object *o;
 452         int r;
 453
 454         assert(f);
 455
 456         s = DEFAULT_BISECT_TABLE_SIZE;
 457         r = journal_file_append_object(f, offsetof(Object, bisect_table.table) + s, &o, &p);
 458         if (r < 0)
 459                 return r;
 460
 461         o->object.type = htole64(OBJECT_BISECT_TABLE);
 462         memset(o->bisect_table.table, 0, s);
 463
 464         f->header->bisect_table_offset = htole64(p + offsetof(Object, bisect_table.table));
 465         f->header->bisect_table_size = htole64(s);
 466
 467         return 0;
 468 }
 469
 470 static int journal_file_map_hash_table(JournalFile *f) {
 471         uint64_t s, p;
 472         void *t;
 473         int r;
 474
 475         assert(f);
 476
 477         p = le64toh(f->header->hash_table_offset);
 478         s = le64toh(f->header->hash_table_size);
 479
 480         r = journal_file_map(f,
 481                              p, s,
 482                              &f->hash_table_window, NULL, &f->hash_table_window_size,
 483                              &t);
 484         if (r < 0)
 485                 return r;
 486
 487         f->hash_table = t;
 488         return 0;
 489 }
 490
 491 static int journal_file_map_bisect_table(JournalFile *f) {
 492         uint64_t s, p;
 493         void *t;
 494         int r;
 495
 496         assert(f);
 497
 498         p = le64toh(f->header->bisect_table_offset);
 499         s = le64toh(f->header->bisect_table_size);
 500
 501         r = journal_file_map(f,
 502                              p, s,
 503                              &f->bisect_table_window, NULL, &f->bisect_table_window_size,
 504                              &t);
 505
 506         if (r < 0)
 507                 return r;
 508
 509         f->bisect_table = t;
 510         return 0;
 511 }
 512
 513 static int journal_file_link_data(JournalFile *f, Object *o, uint64_t offset, uint64_t hash_index) {
 514         uint64_t p;
 515         int r;
 516
 517         assert(f);
 518         assert(o);
 519         assert(offset > 0);
 520         assert(o->object.type == htole64(OBJECT_DATA));
 521
 522         o->data.head_entry_offset = o->data.tail_entry_offset = 0;
 523         o->data.next_hash_offset = 0;
 524
 525         p = le64toh(f->hash_table[hash_index].tail_hash_offset);
 526         if (p == 0) {
 527                 /* Only entry in the hash table is easy */
 528
 529                 o->data.prev_hash_offset = 0;
 530                 f->hash_table[hash_index].head_hash_offset = htole64(offset);
 531         } else {
 532                 o->data.prev_hash_offset = htole64(p);
 533
 534                 /* Temporarily move back to the previous data object,
 535                  * to patch in pointer */
 536
 537                 r = journal_file_move_to_object(f, p, OBJECT_DATA, &o);
 538                 if (r < 0)
 539                         return r;
 540
 541                 o->data.next_hash_offset = offset;
 542
 543                 r = journal_file_move_to_object(f, offset, OBJECT_DATA, &o);
 544                 if (r < 0)
 545                         return r;
 546         }
 547
 548         f->hash_table[hash_index].tail_hash_offset = htole64(offset);
 549
 550         return 0;
 551 }
 552
 553 static int journal_file_append_data(JournalFile *f, const void *data, uint64_t size, Object **ret, uint64_t *offset) {
 554         uint64_t hash, h, p, np;
 555         uint64_t osize;
 556         Object *o;
 557         int r;
 558
 559         assert(f);
 560         assert(data || size == 0);
 561
 562         osize = offsetof(Object, data.payload) + size;
 563
 564         hash = hash64(data, size);
 565         h = hash % (le64toh(f->header->hash_table_size) / sizeof(HashItem));
 566         p = le64toh(f->hash_table[h].head_hash_offset);
 567
 568         while (p != 0) {
 569                 /* Look for this data object in the hash table */
 570
 571                 r = journal_file_move_to_object(f, p, OBJECT_DATA, &o);
 572                 if (r < 0)
 573                         return r;
 574
 575                 if (le64toh(o->object.size) == osize &&
 576                     memcmp(o->data.payload, data, size) == 0) {
 577
 578                         if (le64toh(o->data.hash) != hash)
 579                                 return -EBADMSG;
 580
 581                         if (ret)
 582                                 *ret = o;
 583
 584                         if (offset)
 585                                 *offset = p;
 586
 587                         return 0;
 588                 }
 589
 590                 p = le64toh(o->data.next_hash_offset);
 591         }
 592
 593         r = journal_file_append_object(f, osize, &o, &np);
 594         if (r < 0)
 595                 return r;
 596
 597         o->object.type = htole64(OBJECT_DATA);
 598         o->data.hash = htole64(hash);
 599         memcpy(o->data.payload, data, size);
 600
 601         r = journal_file_link_data(f, o, np, h);
 602         if (r < 0)
 603                 return r;
 604
 605         if (ret)
 606                 *ret = o;
 607
 608         if (offset)
 609                 *offset = np;
 610
 611         return 0;
 612 }
 613
 614 uint64_t journal_file_entry_n_items(Object *o) {
 615         assert(o);
 616         assert(o->object.type == htole64(OBJECT_ENTRY));
 617
 618         return (le64toh(o->object.size) - offsetof(Object, entry.items)) / sizeof(EntryItem);
 619 }
 620
 621 static int journal_file_link_entry_item(JournalFile *f, Object *o, uint64_t offset, uint64_t i) {
 622         uint64_t p, q;
 623         int r;
 624         assert(f);
 625         assert(o);
 626         assert(offset > 0);
 627
 628         p = le64toh(o->entry.items[i].object_offset);
 629         if (p == 0)
 630                 return -EINVAL;
 631
 632         o->entry.items[i].next_entry_offset = 0;
 633
 634         /* Move to the data object */
 635         r = journal_file_move_to_object(f, p, OBJECT_DATA, &o);
 636         if (r < 0)
 637                 return r;
 638
 639         q = le64toh(o->data.tail_entry_offset);
 640         o->data.tail_entry_offset = htole64(offset);
 641
 642         if (q == 0)
 643                 o->data.head_entry_offset = htole64(offset);
 644         else {
 645                 uint64_t n, j;
 646
 647                 /* Move to previous entry */
 648                 r = journal_file_move_to_object(f, q, OBJECT_ENTRY, &o);
 649                 if (r < 0)
 650                         return r;
 651
 652                 n = journal_file_entry_n_items(o);
 653                 for (j = 0; j < n; j++)
 654                         if (le64toh(o->entry.items[j].object_offset) == p)
 655                                 break;
 656
 657                 if (j >= n)
 658                         return -EBADMSG;
 659
 660                 o->entry.items[j].next_entry_offset = offset;
 661         }
 662
 663         /* Move back to original entry */
 664         r = journal_file_move_to_object(f, offset, OBJECT_ENTRY, &o);
 665         if (r < 0)
 666                 return r;
 667
 668         o->entry.items[i].prev_entry_offset = q;
 669         return 0;
 670 }
 671
 672 static int journal_file_link_entry(JournalFile *f, Object *o, uint64_t offset) {
 673         uint64_t p, i, n, k, a, b;
 674         int r;
 675
 676         assert(f);
 677         assert(o);
 678         assert(offset > 0);
 679         assert(o->object.type == htole64(OBJECT_ENTRY));
 680
 681         /* Link up the entry itself */
 682         p = le64toh(f->header->tail_entry_offset);
 683
 684         o->entry.prev_entry_offset = f->header->tail_entry_offset;
 685         o->entry.next_entry_offset = 0;
 686
 687         if (p == 0) {
 688                 f->header->head_entry_offset = htole64(offset);
 689                 f->header->head_entry_realtime = o->entry.realtime;
 690         } else {
 691                 /* Temporarily move back to the previous entry, to
 692                  * patch in pointer */
 693
 694                 r = journal_file_move_to_object(f, p, OBJECT_ENTRY, &o);
 695                 if (r < 0)
 696                         return r;
 697
 698                 o->entry.next_entry_offset = htole64(offset);
 699
 700                 r = journal_file_move_to_object(f, offset, OBJECT_ENTRY, &o);
 701                 if (r < 0)
 702                         return r;
 703         }
 704
 705         f->header->tail_entry_offset = htole64(offset);
 706         f->header->tail_entry_realtime = o->entry.realtime;
 707
 708         /* Link up the items */
 709         n = journal_file_entry_n_items(o);
 710         for (i = 0; i < n; i++) {
 711                 r = journal_file_link_entry_item(f, o, offset, i);
 712                 if (r < 0)
 713                         return r;
 714         }
 715
 716         /* Link up the entry in the bisect table */
 717         n = le64toh(f->header->bisect_table_size) / sizeof(uint64_t);
 718         k = le64toh(f->header->arena_max_size) / n;
 719
 720         a = (le64toh(f->header->last_bisect_offset) + k - 1) / k;
 721         b = offset / k;
 722
 723         for (; a <= b; a++)
 724                 f->bisect_table[a] = htole64(offset);
 725
 726         f->header->last_bisect_offset = htole64(offset + le64toh(o->object.size));
 727
 728         return 0;
 729 }
 730
 731 static int journal_file_append_entry_internal(
 732                 JournalFile *f,
 733                 const dual_timestamp *ts,
 734                 uint64_t xor_hash,
 735                 const EntryItem items[], unsigned n_items,
 736                 Object **ret, uint64_t *offset) {
 737         uint64_t np;
 738         uint64_t osize;
 739         Object *o;
 740         int r;
 741
 742         assert(f);
 743         assert(items || n_items == 0);
 744
 745         osize = offsetof(Object, entry.items) + (n_items * sizeof(EntryItem));
 746
 747         r = journal_file_append_object(f, osize, &o, &np);
 748         if (r < 0)
 749                 return r;
 750
 751         o->object.type = htole64(OBJECT_ENTRY);
 752         o->entry.seqnum = htole64(journal_file_seqnum(f));
 753         memcpy(o->entry.items, items, n_items * sizeof(EntryItem));
 754         o->entry.realtime = ts ? htole64(ts->realtime) : 0;
 755         o->entry.monotonic = ts ? htole64(ts->monotonic) : 0;
 756         o->entry.xor_hash = htole64(xor_hash);
 757         o->entry.boot_id = f->header->boot_id;
 758
 759         r = journal_file_link_entry(f, o, np);
 760         if (r < 0)
 761                 return r;
 762
 763         if (ret)
 764                 *ret = o;
 765
 766         if (offset)
 767                 *offset = np;
 768
 769         return 0;
 770 }
 771
 772 int journal_file_append_entry(JournalFile *f, const dual_timestamp *ts, const struct iovec iovec[], unsigned n_iovec, Object **ret, uint64_t *offset) {
 773         unsigned i;
 774         EntryItem *items;
 775         int r;
 776         uint64_t xor_hash = 0;
 777
 778         assert(f);
 779         assert(iovec || n_iovec == 0);
 780
 781         items = new(EntryItem, n_iovec);
 782         if (!items)
 783                 return -ENOMEM;
 784
 785         for (i = 0; i < n_iovec; i++) {
 786                 uint64_t p;
 787                 Object *o;
 788
 789                 r = journal_file_append_data(f, iovec[i].iov_base, iovec[i].iov_len, &o, &p);
 790                 if (r < 0)
 791                         goto finish;
 792
 793                 xor_hash ^= le64toh(o->data.hash);
 794                 items[i].object_offset = htole64(p);
 795         }
 796
 797         r = journal_file_append_entry_internal(f, ts, xor_hash, items, n_iovec, ret, offset);
 798
 799 finish:
 800         free(items);
 801
 802         return r;
 803 }
 804
 805 int journal_file_move_to_entry(JournalFile *f, uint64_t seqnum, Object **ret, uint64_t *offset) {
 806         Object *o;
 807         uint64_t lower, upper, p, n, k;
 808         int r;
 809
 810         assert(f);
 811
 812         n = le64toh(f->header->bisect_table_size) / sizeof(uint64_t);
 813         k = le64toh(f->header->arena_max_size) / n;
 814
 815         lower = 0;
 816         upper = le64toh(f->header->last_bisect_offset)/k+1;
 817
 818         while (lower < upper) {
 819                 k = (upper + lower) / 2;
 820                 p = le64toh(f->bisect_table[k]);
 821
 822                 if (p == 0) {
 823                         upper = k;
 824                         continue;
 825                 }
 826
 827                 r = journal_file_move_to_object(f, p, OBJECT_ENTRY, &o);
 828                 if (r < 0)
 829                         return r;
 830
 831                 if (o->entry.seqnum == seqnum) {
 832                         if (ret)
 833                                 *ret = o;
 834
 835                         if (offset)
 836                                 *offset = p;
 837
 838                         return 1;
 839                 } else if (seqnum < o->entry.seqnum)
 840                         upper = k;
 841                 else if (seqnum > o->entry.seqnum)
 842                         lower = k+1;
 843         }
 844
 845         assert(lower == upper);
 846
 847         if (lower <= 0)
 848                 return 0;
 849
 850         /* The object we are looking for is between
 851          * bisect_table[lower-1] and bisect_table[lower] */
 852
 853         p = le64toh(f->bisect_table[lower-1]);
 854
 855         for (;;) {
 856                 r = journal_file_move_to_object(f, p, OBJECT_ENTRY, &o);
 857                 if (r < 0)
 858                         return r;
 859
 860                 if (o->entry.seqnum == seqnum) {
 861                         if (ret)
 862                                 *ret = o;
 863
 864                         if (offset)
 865                                 *offset = p;
 866
 867                         return 1;
 868
 869                 } if (seqnum < o->entry.seqnum)
 870                         return 0;
 871
 872                 if (o->entry.next_entry_offset == 0)
 873                         return 0;
 874
 875                 p = le64toh(o->entry.next_entry_offset);
 876         }
 877
 878         return 0;
 879 }
 880
 881 int journal_file_next_entry(JournalFile *f, Object *o, Object **ret, uint64_t *offset) {
 882         uint64_t np;
 883         int r;
 884
 885         assert(f);
 886
 887         if (!o)
 888                 np = le64toh(f->header->head_entry_offset);
 889         else {
 890                 if (le64toh(o->object.type) != OBJECT_ENTRY)
 891                         return -EINVAL;
 892
 893                 np = le64toh(o->entry.next_entry_offset);
 894         }
 895
 896         if (np == 0)
 897                 return 0;
 898
 899         r = journal_file_move_to_object(f, np, OBJECT_ENTRY, &o);
 900         if (r < 0)
 901                 return r;
 902
 903         if (ret)
 904                 *ret = o;
 905
 906         if (offset)
 907                 *offset = np;
 908
 909         return 1;
 910 }
 911
 912 int journal_file_prev_entry(JournalFile *f, Object *o, Object **ret, uint64_t *offset) {
 913         uint64_t np;
 914         int r;
 915
 916         assert(f);
 917
 918         if (!o)
 919                 np = le64toh(f->header->tail_entry_offset);
 920         else {
 921                 if (le64toh(o->object.type) != OBJECT_ENTRY)
 922                         return -EINVAL;
 923
 924                 np = le64toh(o->entry.prev_entry_offset);
 925         }
 926
 927         if (np == 0)
 928                 return 0;
 929
 930         r = journal_file_move_to_object(f, np, OBJECT_ENTRY, &o);
 931         if (r < 0)
 932                 return r;
 933
 934         if (ret)
 935                 *ret = o;
 936
 937         if (offset)
 938                 *offset = np;
 939
 940         return 1;
 941 }
 942
 943 int journal_file_find_first_entry(JournalFile *f, const void *data, uint64_t size, Object **ret, uint64_t *offset) {
 944         uint64_t p, osize, hash, h;
 945         int r;
 946
 947         assert(f);
 948         assert(data || size == 0);
 949
 950         osize = offsetof(Object, data.payload) + size;
 951
 952         hash = hash64(data, size);
 953         h = hash % (le64toh(f->header->hash_table_size) / sizeof(HashItem));
 954         p = le64toh(f->hash_table[h].head_hash_offset);
 955
 956         while (p != 0) {
 957                 Object *o;
 958
 959                 r = journal_file_move_to_object(f, p, OBJECT_DATA, &o);
 960                 if (r < 0)
 961                         return r;
 962
 963                 if (le64toh(o->object.size) == osize &&
 964                     memcmp(o->data.payload, data, size) == 0) {
 965
 966                         if (le64toh(o->data.hash) != hash)
 967                                 return -EBADMSG;
 968
 969                         if (o->data.head_entry_offset == 0)
 970                                 return 0;
 971
 972                         p = le64toh(o->data.head_entry_offset);
 973                         r = journal_file_move_to_object(f, p, OBJECT_ENTRY, &o);
 974                         if (r < 0)
 975                                 return r;
 976
 977                         if (ret)
 978                                 *ret = o;
 979
 980                         if (offset)
 981                                 *offset = p;
 982
 983                         return 1;
 984                 }
 985
 986                 p = le64toh(o->data.next_hash_offset);
 987         }
 988
 989         return 0;
 990 }
 991
 992 int journal_file_find_last_entry(JournalFile *f, const void *data, uint64_t size, Object **ret, uint64_t *offset) {
 993         uint64_t p, osize, hash, h;
 994         int r;
 995
 996         assert(f);
 997         assert(data || size == 0);
 998
 999         osize = offsetof(Object, data.payload) + size;
1000
1001         hash = hash64(data, size);
1002         h = hash % (le64toh(f->header->hash_table_size) / sizeof(HashItem));
1003         p = le64toh(f->hash_table[h].tail_hash_offset);
1004
1005         while (p != 0) {
1006                 Object *o;
1007
1008                 r = journal_file_move_to_object(f, p, OBJECT_DATA, &o);
1009                 if (r < 0)
1010                         return r;
1011
1012                 if (le64toh(o->object.size) == osize &&
1013                     memcmp(o->data.payload, data, size) == 0) {
1014
1015                         if (le64toh(o->data.hash) != hash)
1016                                 return -EBADMSG;
1017
1018                         if (o->data.tail_entry_offset == 0)
1019                                 return 0;
1020
1021                         p = le64toh(o->data.tail_entry_offset);
1022                         r = journal_file_move_to_object(f, p, OBJECT_ENTRY, &o);
1023                         if (r < 0)
1024                                 return r;
1025
1026                         if (ret)
1027                                 *ret = o;
1028
1029                         if (offset)
1030                                 *offset = p;
1031
1032                         return 1;
1033                 }
1034
1035                 p = le64toh(o->data.prev_hash_offset);
1036         }
1037
1038         return 0;
1039 }
1040
1041 void journal_file_dump(JournalFile *f) {
1042         char a[33], b[33], c[33];
1043         Object *o;
1044         int r;
1045         uint64_t p;
1046
1047         assert(f);
1048
1049         printf("File ID: %s\n"
1050                "Machine ID: %s\n"
1051                "Boot ID: %s\n"
1052                "Arena size: %llu\n",
1053                sd_id128_to_string(f->header->file_id, a),
1054                sd_id128_to_string(f->header->machine_id, b),
1055                sd_id128_to_string(f->header->boot_id, c),
1056                (unsigned long long) le64toh(f->header->arena_size));
1057
1058         p = le64toh(f->header->head_object_offset);
1059         while (p != 0) {
1060                 r = journal_file_move_to_object(f, p, -1, &o);
1061                 if (r < 0)
1062                         goto fail;
1063
1064                 switch (o->object.type) {
1065
1066                 case OBJECT_UNUSED:
1067                         printf("Type: OBJECT_UNUSED\n");
1068                         break;
1069
1070                 case OBJECT_DATA:
1071                         printf("Type: OBJECT_DATA\n");
1072                         break;
1073
1074                 case OBJECT_ENTRY:
1075                         printf("Type: OBJECT_ENTRY %llu\n", (unsigned long long) le64toh(o->entry.seqnum));
1076                         break;
1077
1078                 case OBJECT_HASH_TABLE:
1079                         printf("Type: OBJECT_HASH_TABLE\n");
1080                         break;
1081
1082                 case OBJECT_BISECT_TABLE:
1083                         printf("Type: OBJECT_BISECT_TABLE\n");
1084                         break;
1085                 }
1086
1087                 if (p == le64toh(f->header->tail_object_offset))
1088                         p = 0;
1089                 else
1090                         p = p + ALIGN64(le64toh(o->object.size));
1091         }
1092
1093         return;
1094 fail:
1095         log_error("File corrupt");
1096 }
1097
1098 int journal_file_open(
1099                 const char *fname,
1100                 int flags,
1101                 mode_t mode,
1102                 JournalFile *template,
1103                 JournalFile **ret) {
1104
1105         JournalFile *f;
1106         int r;
1107         bool newly_created = false;
1108
1109         assert(fname);
1110
1111         if ((flags & O_ACCMODE) != O_RDONLY &&
1112             (flags & O_ACCMODE) != O_RDWR)
1113                 return -EINVAL;
1114
1115         f = new0(JournalFile, 1);
1116         if (!f)
1117                 return -ENOMEM;
1118
1119         f->fd = -1;
1120         f->flags = flags;
1121         f->mode = mode;
1122         f->writable = (flags & O_ACCMODE) != O_RDONLY;
1123         f->prot = prot_from_flags(flags);
1124
1125         f->path = strdup(fname);
1126         if (!f->path) {
1127                 r = -ENOMEM;
1128                 goto fail;
1129         }
1130
1131         f->fd = open(f->path, f->flags|O_CLOEXEC, f->mode);
1132         if (f->fd < 0) {
1133                 r = -errno;
1134                 goto fail;
1135         }
1136
1137         if (fstat(f->fd, &f->last_stat) < 0) {
1138                 r = -errno;
1139                 goto fail;
1140         }
1141
1142         if (f->last_stat.st_size == 0 && f->writable) {
1143                 newly_created = true;
1144
1145                 r = journal_file_init_header(f, template);
1146                 if (r < 0)
1147                         goto fail;
1148
1149                 if (fstat(f->fd, &f->last_stat) < 0) {
1150                         r = -errno;
1151                         goto fail;
1152                 }
1153         }
1154
1155         if (f->last_stat.st_size < (off_t) sizeof(Header)) {
1156                 r = -EIO;
1157                 goto fail;
1158         }
1159
1160         f->header = mmap(NULL, PAGE_ALIGN(sizeof(Header)), prot_from_flags(flags), MAP_SHARED, f->fd, 0);
1161         if (f->header == MAP_FAILED) {
1162                 f->header = NULL;
1163                 r = -errno;
1164                 goto fail;
1165         }
1166
1167         if (!newly_created) {
1168                 r = journal_file_verify_header(f);
1169                 if (r < 0)
1170                         goto fail;
1171         }
1172
1173         if (f->writable) {
1174                 r = journal_file_refresh_header(f);
1175                 if (r < 0)
1176                         goto fail;
1177         }
1178
1179         if (newly_created) {
1180
1181                 r = journal_file_setup_hash_table(f);
1182                 if (r < 0)
1183                         goto fail;
1184
1185                 r = journal_file_setup_bisect_table(f);
1186                 if (r < 0)
1187                         goto fail;
1188         }
1189
1190         r = journal_file_map_hash_table(f);
1191         if (r < 0)
1192                 goto fail;
1193
1194         r = journal_file_map_bisect_table(f);
1195         if (r < 0)
1196                 goto fail;
1197
1198         if (ret)
1199                 *ret = f;
1200
1201         return 0;
1202
1203 fail:
1204         journal_file_close(f);
1205
1206         return r;
1207 }
1208
1209 int journal_file_rotate(JournalFile **f) {
1210         char *p;
1211         size_t l;
1212         JournalFile *old_file, *new_file = NULL;
1213         int r;
1214
1215         assert(f);
1216         assert(*f);
1217
1218         old_file = *f;
1219
1220         if (!old_file->writable)
1221                 return -EINVAL;
1222
1223         if (!endswith(old_file->path, ".journal"))
1224                 return -EINVAL;
1225
1226         l = strlen(old_file->path);
1227
1228         p = new(char, l + 1 + 16 + 1 + 32 + 1 + 16 + 1);
1229         if (!p)
1230                 return -ENOMEM;
1231
1232         memcpy(p, old_file->path, l - 8);
1233         p[l-8] = '@';
1234         sd_id128_to_string(old_file->header->seqnum_id, p + l - 8 + 1);
1235         snprintf(p + l - 8 + 1 + 32, 1 + 16 + 1 + 16 + 8 + 1,
1236                  "-%016llx-%016llx.journal",
1237                  (unsigned long long) le64toh((*f)->header->seqnum),
1238                  (unsigned long long) le64toh((*f)->header->tail_entry_realtime));
1239
1240         r = rename(old_file->path, p);
1241         free(p);
1242
1243         if (r < 0)
1244                 return -errno;
1245
1246         old_file->header->state = le32toh(STATE_ARCHIVED);
1247
1248         r = journal_file_open(old_file->path, old_file->flags, old_file->mode, old_file, &new_file);
1249         journal_file_close(old_file);
1250
1251         *f = new_file;
1252         return r;
1253 }
1254
1255 struct vacuum_info {
1256         off_t usage;
1257         char *filename;
1258
1259         uint64_t realtime;
1260         sd_id128_t seqnum_id;
1261         uint64_t seqnum;
1262 };
1263
1264 static int vacuum_compare(const void *_a, const void *_b) {
1265         const struct vacuum_info *a, *b;
1266
1267         a = _a;
1268         b = _b;
1269
1270         if (sd_id128_equal(a->seqnum_id, b->seqnum_id)) {
1271                 if (a->seqnum < b->seqnum)
1272                         return -1;
1273                 else if (a->seqnum > b->seqnum)
1274                         return 1;
1275                 else
1276                         return 0;
1277         }
1278
1279         if (a->realtime < b->realtime)
1280                 return -1;
1281         else if (a->realtime > b->realtime)
1282                 return 1;
1283         else
1284                 return memcmp(&a->seqnum_id, &b->seqnum_id, 16);
1285 }
1286
1287 int journal_directory_vacuum(const char *directory, uint64_t max_use, uint64_t min_free) {
1288         DIR *d;
1289         int r = 0;
1290         struct vacuum_info *list = NULL;
1291         unsigned n_list = 0, n_allocated = 0, i;
1292         uint64_t sum = 0;
1293
1294         assert(directory);
1295
1296         if (max_use <= 0)
1297                 max_use = DEFAULT_MAX_USE;
1298
1299         d = opendir(directory);
1300         if (!d)
1301                 return -errno;
1302
1303         for (;;) {
1304                 int k;
1305                 struct dirent buf, *de;
1306                 size_t q;
1307                 struct stat st;
1308                 char *p;
1309                 unsigned long long seqnum, realtime;
1310                 sd_id128_t seqnum_id;
1311
1312                 k = readdir_r(d, &buf, &de);
1313                 if (k != 0) {
1314                         r = -k;
1315                         goto finish;
1316                 }
1317
1318                 if (!de)
1319                         break;
1320
1321                 if (!dirent_is_file_with_suffix(de, ".journal"))
1322                         continue;
1323
1324                 q = strlen(de->d_name);
1325
1326                 if (q < 1 + 32 + 1 + 16 + 1 + 16 + 8)
1327                         continue;
1328
1329                 if (de->d_name[q-8-16-1] != '-' ||
1330                     de->d_name[q-8-16-1-16-1] != '-' ||
1331                     de->d_name[q-8-16-1-16-1-32-1] != '@')
1332                         continue;
1333
1334                 if (fstatat(dirfd(d), de->d_name, &st, AT_SYMLINK_NOFOLLOW) < 0)
1335                         continue;
1336
1337                 if (!S_ISREG(st.st_mode))
1338                         continue;
1339
1340                 p = strdup(de->d_name);
1341                 if (!p) {
1342                         r = -ENOMEM;
1343                         goto finish;
1344                 }
1345
1346                 de->d_name[q-8-16-1-16-1] = 0;
1347                 if (sd_id128_from_string(de->d_name + q-8-16-1-16-1-32, &seqnum_id) < 0) {
1348                         free(p);
1349                         continue;
1350                 }
1351
1352                 if (sscanf(de->d_name + q-8-16-1-16, "%16llx-%16llx.journal", &seqnum, &realtime) != 2) {
1353                         free(p);
1354                         continue;
1355                 }
1356
1357                 if (n_list >= n_allocated) {
1358                         struct vacuum_info *j;
1359
1360                         n_allocated = MAX(n_allocated * 2U, 8U);
1361                         j = realloc(list, n_allocated * sizeof(struct vacuum_info));
1362                         if (!j) {
1363                                 free(p);
1364                                 r = -ENOMEM;
1365                                 goto finish;
1366                         }
1367
1368                         list = j;
1369                 }
1370
1371                 list[n_list].filename = p;
1372                 list[n_list].usage = (uint64_t) st.st_blksize * (uint64_t) st.st_blocks;
1373                 list[n_list].seqnum = seqnum;
1374                 list[n_list].realtime = realtime;
1375                 list[n_list].seqnum_id = seqnum_id;
1376
1377                 sum += list[n_list].usage;
1378
1379                 n_list ++;
1380         }
1381
1382         qsort(list, n_list, sizeof(struct vacuum_info), vacuum_compare);
1383
1384         for(i = 0; i < n_list; i++) {
1385                 struct statvfs ss;
1386
1387                 if (fstatvfs(dirfd(d), &ss) < 0) {
1388                         r = -errno;
1389                         goto finish;
1390                 }
1391
1392                 if (sum <= max_use &&
1393                     (uint64_t) ss.f_bavail * (uint64_t) ss.f_bsize >= min_free)
1394                         break;
1395
1396                 if (unlinkat(dirfd(d), list[i].filename, 0) >= 0) {
1397                         log_debug("Deleted archived journal %s/%s.", directory, list[i].filename);
1398                         sum -= list[i].usage;
1399                 } else if (errno != ENOENT)
1400                         log_warning("Failed to delete %s/%s: %m", directory, list[i].filename);
1401         }
1402
1403 finish:
1404         for (i = 0; i < n_list; i++)
1405                 free(list[i].filename);
1406
1407         free(list);
1408
1409         return r;
1410 }