1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
4 This file is part of systemd.
6 Copyright 2011 Lennart Poettering
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.
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.
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/>.
26 #include <sys/statvfs.h>
31 #include <attr/xattr.h>
34 #include "journal-def.h"
35 #include "journal-file.h"
36 #include "journal-authenticate.h"
41 #define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
42 #define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
44 #define COMPRESSION_SIZE_THRESHOLD (512ULL)
46 /* This is the minimum journal file size */
47 #define JOURNAL_FILE_SIZE_MIN (4ULL*1024ULL*1024ULL) /* 4 MiB */
49 /* These are the lower and upper bounds if we deduce the max_use value
50 * from the file system size */
51 #define DEFAULT_MAX_USE_LOWER (1ULL*1024ULL*1024ULL) /* 1 MiB */
52 #define DEFAULT_MAX_USE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
54 /* This is the upper bound if we deduce max_size from max_use */
55 #define DEFAULT_MAX_SIZE_UPPER (128ULL*1024ULL*1024ULL) /* 128 MiB */
57 /* This is the upper bound if we deduce the keep_free value from the
59 #define DEFAULT_KEEP_FREE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
61 /* This is the keep_free value when we can't determine the system
63 #define DEFAULT_KEEP_FREE (1024ULL*1024ULL) /* 1 MB */
65 /* n_data was the first entry we added after the initial file format design */
66 #define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
68 /* How many entries to keep in the entry array chain cache at max */
69 #define CHAIN_CACHE_MAX 20
71 /* How much to increase the journal file size at once each time we allocate something new. */
72 #define FILE_SIZE_INCREASE (8ULL*1024ULL*1024ULL) /* 8MB */
74 static int journal_file_set_online(JournalFile *f) {
80 if (!(f->fd >= 0 && f->header))
83 switch(f->header->state) {
88 f->header->state = STATE_ONLINE;
97 int journal_file_set_offline(JournalFile *f) {
103 if (!(f->fd >= 0 && f->header))
106 if (f->header->state != STATE_ONLINE)
111 f->header->state = STATE_OFFLINE;
118 void journal_file_close(JournalFile *f) {
122 /* Write the final tag */
123 if (f->seal && f->writable)
124 journal_file_append_tag(f);
127 /* Sync everything to disk, before we mark the file offline */
128 if (f->mmap && f->fd >= 0)
129 mmap_cache_close_fd(f->mmap, f->fd);
131 journal_file_set_offline(f);
134 munmap(f->header, PAGE_ALIGN(sizeof(Header)));
137 close_nointr_nofail(f->fd);
142 mmap_cache_unref(f->mmap);
144 hashmap_free_free(f->chain_cache);
147 free(f->compress_buffer);
152 munmap(f->fss_file, PAGE_ALIGN(f->fss_file_size));
153 else if (f->fsprg_state)
154 free(f->fsprg_state);
159 gcry_md_close(f->hmac);
165 static int journal_file_init_header(JournalFile *f, JournalFile *template) {
173 memcpy(h.signature, HEADER_SIGNATURE, 8);
174 h.header_size = htole64(ALIGN64(sizeof(h)));
176 h.incompatible_flags =
177 htole32(f->compress ? HEADER_INCOMPATIBLE_COMPRESSED : 0);
180 htole32(f->seal ? HEADER_COMPATIBLE_SEALED : 0);
182 r = sd_id128_randomize(&h.file_id);
187 h.seqnum_id = template->header->seqnum_id;
188 h.tail_entry_seqnum = template->header->tail_entry_seqnum;
190 h.seqnum_id = h.file_id;
192 k = pwrite(f->fd, &h, sizeof(h), 0);
202 static int journal_file_refresh_header(JournalFile *f) {
208 r = sd_id128_get_machine(&f->header->machine_id);
212 r = sd_id128_get_boot(&boot_id);
216 if (sd_id128_equal(boot_id, f->header->boot_id))
217 f->tail_entry_monotonic_valid = true;
219 f->header->boot_id = boot_id;
221 journal_file_set_online(f);
223 /* Sync the online state to disk */
229 static int journal_file_verify_header(JournalFile *f) {
232 if (memcmp(f->header->signature, HEADER_SIGNATURE, 8))
235 /* In both read and write mode we refuse to open files with
236 * incompatible flags we don't know */
238 if ((le32toh(f->header->incompatible_flags) & ~HEADER_INCOMPATIBLE_COMPRESSED) != 0)
239 return -EPROTONOSUPPORT;
241 if (f->header->incompatible_flags != 0)
242 return -EPROTONOSUPPORT;
245 /* When open for writing we refuse to open files with
246 * compatible flags, too */
249 if ((le32toh(f->header->compatible_flags) & ~HEADER_COMPATIBLE_SEALED) != 0)
250 return -EPROTONOSUPPORT;
252 if (f->header->compatible_flags != 0)
253 return -EPROTONOSUPPORT;
257 if (f->header->state >= _STATE_MAX)
260 /* The first addition was n_data, so check that we are at least this large */
261 if (le64toh(f->header->header_size) < HEADER_SIZE_MIN)
264 if (JOURNAL_HEADER_SEALED(f->header) && !JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
267 if ((le64toh(f->header->header_size) + le64toh(f->header->arena_size)) > (uint64_t) f->last_stat.st_size)
270 if (le64toh(f->header->tail_object_offset) > (le64toh(f->header->header_size) + le64toh(f->header->arena_size)))
273 if (!VALID64(le64toh(f->header->data_hash_table_offset)) ||
274 !VALID64(le64toh(f->header->field_hash_table_offset)) ||
275 !VALID64(le64toh(f->header->tail_object_offset)) ||
276 !VALID64(le64toh(f->header->entry_array_offset)))
279 if (le64toh(f->header->data_hash_table_offset) < le64toh(f->header->header_size) ||
280 le64toh(f->header->field_hash_table_offset) < le64toh(f->header->header_size) ||
281 le64toh(f->header->tail_object_offset) < le64toh(f->header->header_size) ||
282 le64toh(f->header->entry_array_offset) < le64toh(f->header->header_size))
287 sd_id128_t machine_id;
290 r = sd_id128_get_machine(&machine_id);
294 if (!sd_id128_equal(machine_id, f->header->machine_id))
297 state = f->header->state;
299 if (state == STATE_ONLINE) {
300 log_debug("Journal file %s is already online. Assuming unclean closing.", f->path);
302 } else if (state == STATE_ARCHIVED)
304 else if (state != STATE_OFFLINE) {
305 log_debug("Journal file %s has unknown state %u.", f->path, state);
310 f->compress = JOURNAL_HEADER_COMPRESSED(f->header);
312 f->seal = JOURNAL_HEADER_SEALED(f->header);
317 static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size) {
318 uint64_t old_size, new_size;
323 /* We assume that this file is not sparse, and we know that
324 * for sure, since we always call posix_fallocate()
328 le64toh(f->header->header_size) +
329 le64toh(f->header->arena_size);
331 new_size = PAGE_ALIGN(offset + size);
332 if (new_size < le64toh(f->header->header_size))
333 new_size = le64toh(f->header->header_size);
335 if (new_size <= old_size)
338 if (f->metrics.max_size > 0 && new_size > f->metrics.max_size)
341 if (new_size > f->metrics.min_size && f->metrics.keep_free > 0) {
344 if (fstatvfs(f->fd, &svfs) >= 0) {
347 available = svfs.f_bfree * svfs.f_bsize;
349 if (available >= f->metrics.keep_free)
350 available -= f->metrics.keep_free;
354 if (new_size - old_size > available)
359 /* Increase by larger blocks at once */
360 new_size = ((new_size+FILE_SIZE_INCREASE-1) / FILE_SIZE_INCREASE) * FILE_SIZE_INCREASE;
361 if (f->metrics.max_size > 0 && new_size > f->metrics.max_size)
362 new_size = f->metrics.max_size;
364 /* Note that the glibc fallocate() fallback is very
365 inefficient, hence we try to minimize the allocation area
367 r = posix_fallocate(f->fd, old_size, new_size - old_size);
371 if (fstat(f->fd, &f->last_stat) < 0)
374 f->header->arena_size = htole64(new_size - le64toh(f->header->header_size));
379 static int journal_file_move_to(JournalFile *f, int context, bool keep_always, uint64_t offset, uint64_t size, void **ret) {
386 /* Avoid SIGBUS on invalid accesses */
387 if (offset + size > (uint64_t) f->last_stat.st_size) {
388 /* Hmm, out of range? Let's refresh the fstat() data
389 * first, before we trust that check. */
391 if (fstat(f->fd, &f->last_stat) < 0 ||
392 offset + size > (uint64_t) f->last_stat.st_size)
393 return -EADDRNOTAVAIL;
396 return mmap_cache_get(f->mmap, f->fd, f->prot, context, keep_always, offset, size, &f->last_stat, ret);
399 static uint64_t minimum_header_size(Object *o) {
401 static const uint64_t table[] = {
402 [OBJECT_DATA] = sizeof(DataObject),
403 [OBJECT_FIELD] = sizeof(FieldObject),
404 [OBJECT_ENTRY] = sizeof(EntryObject),
405 [OBJECT_DATA_HASH_TABLE] = sizeof(HashTableObject),
406 [OBJECT_FIELD_HASH_TABLE] = sizeof(HashTableObject),
407 [OBJECT_ENTRY_ARRAY] = sizeof(EntryArrayObject),
408 [OBJECT_TAG] = sizeof(TagObject),
411 if (o->object.type >= ELEMENTSOF(table) || table[o->object.type] <= 0)
412 return sizeof(ObjectHeader);
414 return table[o->object.type];
417 int journal_file_move_to_object(JournalFile *f, int type, uint64_t offset, Object **ret) {
426 /* Objects may only be located at multiple of 64 bit */
427 if (!VALID64(offset))
431 r = journal_file_move_to(f, type_to_context(type), false, offset, sizeof(ObjectHeader), &t);
436 s = le64toh(o->object.size);
438 if (s < sizeof(ObjectHeader))
441 if (o->object.type <= OBJECT_UNUSED)
444 if (s < minimum_header_size(o))
447 if (type > 0 && o->object.type != type)
450 if (s > sizeof(ObjectHeader)) {
451 r = journal_file_move_to(f, o->object.type, false, offset, s, &t);
462 static uint64_t journal_file_entry_seqnum(JournalFile *f, uint64_t *seqnum) {
467 r = le64toh(f->header->tail_entry_seqnum) + 1;
470 /* If an external seqnum counter was passed, we update
471 * both the local and the external one, and set it to
472 * the maximum of both */
480 f->header->tail_entry_seqnum = htole64(r);
482 if (f->header->head_entry_seqnum == 0)
483 f->header->head_entry_seqnum = htole64(r);
488 int journal_file_append_object(JournalFile *f, int type, uint64_t size, Object **ret, uint64_t *offset) {
495 assert(type > 0 && type < _OBJECT_TYPE_MAX);
496 assert(size >= sizeof(ObjectHeader));
500 r = journal_file_set_online(f);
504 p = le64toh(f->header->tail_object_offset);
506 p = le64toh(f->header->header_size);
508 r = journal_file_move_to_object(f, -1, p, &tail);
512 p += ALIGN64(le64toh(tail->object.size));
515 r = journal_file_allocate(f, p, size);
519 r = journal_file_move_to(f, type, false, p, size, &t);
526 o->object.type = type;
527 o->object.size = htole64(size);
529 f->header->tail_object_offset = htole64(p);
530 f->header->n_objects = htole64(le64toh(f->header->n_objects) + 1);
538 static int journal_file_setup_data_hash_table(JournalFile *f) {
545 /* We estimate that we need 1 hash table entry per 768 of
546 journal file and we want to make sure we never get beyond
547 75% fill level. Calculate the hash table size for the
548 maximum file size based on these metrics. */
550 s = (f->metrics.max_size * 4 / 768 / 3) * sizeof(HashItem);
551 if (s < DEFAULT_DATA_HASH_TABLE_SIZE)
552 s = DEFAULT_DATA_HASH_TABLE_SIZE;
554 log_debug("Reserving %"PRIu64" entries in hash table.", s / sizeof(HashItem));
556 r = journal_file_append_object(f,
557 OBJECT_DATA_HASH_TABLE,
558 offsetof(Object, hash_table.items) + s,
563 memzero(o->hash_table.items, s);
565 f->header->data_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
566 f->header->data_hash_table_size = htole64(s);
571 static int journal_file_setup_field_hash_table(JournalFile *f) {
578 /* We use a fixed size hash table for the fields as this
579 * number should grow very slowly only */
581 s = DEFAULT_FIELD_HASH_TABLE_SIZE;
582 r = journal_file_append_object(f,
583 OBJECT_FIELD_HASH_TABLE,
584 offsetof(Object, hash_table.items) + s,
589 memzero(o->hash_table.items, s);
591 f->header->field_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
592 f->header->field_hash_table_size = htole64(s);
597 static int journal_file_map_data_hash_table(JournalFile *f) {
604 p = le64toh(f->header->data_hash_table_offset);
605 s = le64toh(f->header->data_hash_table_size);
607 r = journal_file_move_to(f,
608 OBJECT_DATA_HASH_TABLE,
615 f->data_hash_table = t;
619 static int journal_file_map_field_hash_table(JournalFile *f) {
626 p = le64toh(f->header->field_hash_table_offset);
627 s = le64toh(f->header->field_hash_table_size);
629 r = journal_file_move_to(f,
630 OBJECT_FIELD_HASH_TABLE,
637 f->field_hash_table = t;
641 static int journal_file_link_field(
654 if (o->object.type != OBJECT_FIELD)
657 /* This might alter the window we are looking at */
659 o->field.next_hash_offset = o->field.head_data_offset = 0;
661 h = hash % (le64toh(f->header->field_hash_table_size) / sizeof(HashItem));
662 p = le64toh(f->field_hash_table[h].tail_hash_offset);
664 f->field_hash_table[h].head_hash_offset = htole64(offset);
666 r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
670 o->field.next_hash_offset = htole64(offset);
673 f->field_hash_table[h].tail_hash_offset = htole64(offset);
675 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
676 f->header->n_fields = htole64(le64toh(f->header->n_fields) + 1);
681 static int journal_file_link_data(
694 if (o->object.type != OBJECT_DATA)
697 /* This might alter the window we are looking at */
699 o->data.next_hash_offset = o->data.next_field_offset = 0;
700 o->data.entry_offset = o->data.entry_array_offset = 0;
701 o->data.n_entries = 0;
703 h = hash % (le64toh(f->header->data_hash_table_size) / sizeof(HashItem));
704 p = le64toh(f->data_hash_table[h].tail_hash_offset);
706 /* Only entry in the hash table is easy */
707 f->data_hash_table[h].head_hash_offset = htole64(offset);
709 /* Move back to the previous data object, to patch in
712 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
716 o->data.next_hash_offset = htole64(offset);
719 f->data_hash_table[h].tail_hash_offset = htole64(offset);
721 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
722 f->header->n_data = htole64(le64toh(f->header->n_data) + 1);
727 int journal_file_find_field_object_with_hash(
729 const void *field, uint64_t size, uint64_t hash,
730 Object **ret, uint64_t *offset) {
732 uint64_t p, osize, h;
736 assert(field && size > 0);
738 osize = offsetof(Object, field.payload) + size;
740 if (f->header->field_hash_table_size == 0)
743 h = hash % (le64toh(f->header->field_hash_table_size) / sizeof(HashItem));
744 p = le64toh(f->field_hash_table[h].head_hash_offset);
749 r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
753 if (le64toh(o->field.hash) == hash &&
754 le64toh(o->object.size) == osize &&
755 memcmp(o->field.payload, field, size) == 0) {
765 p = le64toh(o->field.next_hash_offset);
771 int journal_file_find_field_object(
773 const void *field, uint64_t size,
774 Object **ret, uint64_t *offset) {
779 assert(field && size > 0);
781 hash = hash64(field, size);
783 return journal_file_find_field_object_with_hash(f,
788 int journal_file_find_data_object_with_hash(
790 const void *data, uint64_t size, uint64_t hash,
791 Object **ret, uint64_t *offset) {
793 uint64_t p, osize, h;
797 assert(data || size == 0);
799 osize = offsetof(Object, data.payload) + size;
801 if (f->header->data_hash_table_size == 0)
804 h = hash % (le64toh(f->header->data_hash_table_size) / sizeof(HashItem));
805 p = le64toh(f->data_hash_table[h].head_hash_offset);
810 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
814 if (le64toh(o->data.hash) != hash)
817 if (o->object.flags & OBJECT_COMPRESSED) {
821 l = le64toh(o->object.size);
822 if (l <= offsetof(Object, data.payload))
825 l -= offsetof(Object, data.payload);
827 if (!uncompress_blob(o->data.payload, l, &f->compress_buffer, &f->compress_buffer_size, &rsize, 0))
831 memcmp(f->compress_buffer, data, size) == 0) {
842 return -EPROTONOSUPPORT;
845 } else if (le64toh(o->object.size) == osize &&
846 memcmp(o->data.payload, data, size) == 0) {
858 p = le64toh(o->data.next_hash_offset);
864 int journal_file_find_data_object(
866 const void *data, uint64_t size,
867 Object **ret, uint64_t *offset) {
872 assert(data || size == 0);
874 hash = hash64(data, size);
876 return journal_file_find_data_object_with_hash(f,
881 static int journal_file_append_field(
883 const void *field, uint64_t size,
884 Object **ret, uint64_t *offset) {
892 assert(field && size > 0);
894 hash = hash64(field, size);
896 r = journal_file_find_field_object_with_hash(f, field, size, hash, &o, &p);
910 osize = offsetof(Object, field.payload) + size;
911 r = journal_file_append_object(f, OBJECT_FIELD, osize, &o, &p);
915 o->field.hash = htole64(hash);
916 memcpy(o->field.payload, field, size);
918 r = journal_file_link_field(f, o, p, hash);
922 /* The linking might have altered the window, so let's
923 * refresh our pointer */
924 r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
929 r = journal_file_hmac_put_object(f, OBJECT_FIELD, o, p);
943 static int journal_file_append_data(
945 const void *data, uint64_t size,
946 Object **ret, uint64_t *offset) {
952 bool compressed = false;
956 assert(data || size == 0);
958 hash = hash64(data, size);
960 r = journal_file_find_data_object_with_hash(f, data, size, hash, &o, &p);
974 osize = offsetof(Object, data.payload) + size;
975 r = journal_file_append_object(f, OBJECT_DATA, osize, &o, &p);
979 o->data.hash = htole64(hash);
983 size >= COMPRESSION_SIZE_THRESHOLD) {
986 compressed = compress_blob(data, size, o->data.payload, &rsize);
989 o->object.size = htole64(offsetof(Object, data.payload) + rsize);
990 o->object.flags |= OBJECT_COMPRESSED;
992 log_debug("Compressed data object %"PRIu64" -> %"PRIu64, size, rsize);
997 if (!compressed && size > 0)
998 memcpy(o->data.payload, data, size);
1000 r = journal_file_link_data(f, o, p, hash);
1004 /* The linking might have altered the window, so let's
1005 * refresh our pointer */
1006 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1013 eq = memchr(data, '=', size);
1014 if (eq && eq > data) {
1018 /* Create field object ... */
1019 r = journal_file_append_field(f, data, (uint8_t*) eq - (uint8_t*) data, &fo, &fp);
1023 /* ... and link it in. */
1024 o->data.next_field_offset = fo->field.head_data_offset;
1025 fo->field.head_data_offset = le64toh(p);
1029 r = journal_file_hmac_put_object(f, OBJECT_DATA, o, p);
1043 uint64_t journal_file_entry_n_items(Object *o) {
1046 if (o->object.type != OBJECT_ENTRY)
1049 return (le64toh(o->object.size) - offsetof(Object, entry.items)) / sizeof(EntryItem);
1052 uint64_t journal_file_entry_array_n_items(Object *o) {
1055 if (o->object.type != OBJECT_ENTRY_ARRAY)
1058 return (le64toh(o->object.size) - offsetof(Object, entry_array.items)) / sizeof(uint64_t);
1061 uint64_t journal_file_hash_table_n_items(Object *o) {
1064 if (o->object.type != OBJECT_DATA_HASH_TABLE &&
1065 o->object.type != OBJECT_FIELD_HASH_TABLE)
1068 return (le64toh(o->object.size) - offsetof(Object, hash_table.items)) / sizeof(HashItem);
1071 static int link_entry_into_array(JournalFile *f,
1076 uint64_t n = 0, ap = 0, q, i, a, hidx;
1084 a = le64toh(*first);
1085 i = hidx = le64toh(*idx);
1088 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
1092 n = journal_file_entry_array_n_items(o);
1094 o->entry_array.items[i] = htole64(p);
1095 *idx = htole64(hidx + 1);
1101 a = le64toh(o->entry_array.next_entry_array_offset);
1112 r = journal_file_append_object(f, OBJECT_ENTRY_ARRAY,
1113 offsetof(Object, entry_array.items) + n * sizeof(uint64_t),
1119 r = journal_file_hmac_put_object(f, OBJECT_ENTRY_ARRAY, o, q);
1124 o->entry_array.items[i] = htole64(p);
1127 *first = htole64(q);
1129 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, ap, &o);
1133 o->entry_array.next_entry_array_offset = htole64(q);
1136 if (JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
1137 f->header->n_entry_arrays = htole64(le64toh(f->header->n_entry_arrays) + 1);
1139 *idx = htole64(hidx + 1);
1144 static int link_entry_into_array_plus_one(JournalFile *f,
1159 *extra = htole64(p);
1163 i = htole64(le64toh(*idx) - 1);
1164 r = link_entry_into_array(f, first, &i, p);
1169 *idx = htole64(le64toh(*idx) + 1);
1173 static int journal_file_link_entry_item(JournalFile *f, Object *o, uint64_t offset, uint64_t i) {
1180 p = le64toh(o->entry.items[i].object_offset);
1184 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1188 return link_entry_into_array_plus_one(f,
1189 &o->data.entry_offset,
1190 &o->data.entry_array_offset,
1195 static int journal_file_link_entry(JournalFile *f, Object *o, uint64_t offset) {
1203 if (o->object.type != OBJECT_ENTRY)
1206 __sync_synchronize();
1208 /* Link up the entry itself */
1209 r = link_entry_into_array(f,
1210 &f->header->entry_array_offset,
1211 &f->header->n_entries,
1216 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
1218 if (f->header->head_entry_realtime == 0)
1219 f->header->head_entry_realtime = o->entry.realtime;
1221 f->header->tail_entry_realtime = o->entry.realtime;
1222 f->header->tail_entry_monotonic = o->entry.monotonic;
1224 f->tail_entry_monotonic_valid = true;
1226 /* Link up the items */
1227 n = journal_file_entry_n_items(o);
1228 for (i = 0; i < n; i++) {
1229 r = journal_file_link_entry_item(f, o, offset, i);
1237 static int journal_file_append_entry_internal(
1239 const dual_timestamp *ts,
1241 const EntryItem items[], unsigned n_items,
1243 Object **ret, uint64_t *offset) {
1250 assert(items || n_items == 0);
1253 osize = offsetof(Object, entry.items) + (n_items * sizeof(EntryItem));
1255 r = journal_file_append_object(f, OBJECT_ENTRY, osize, &o, &np);
1259 o->entry.seqnum = htole64(journal_file_entry_seqnum(f, seqnum));
1260 memcpy(o->entry.items, items, n_items * sizeof(EntryItem));
1261 o->entry.realtime = htole64(ts->realtime);
1262 o->entry.monotonic = htole64(ts->monotonic);
1263 o->entry.xor_hash = htole64(xor_hash);
1264 o->entry.boot_id = f->header->boot_id;
1267 r = journal_file_hmac_put_object(f, OBJECT_ENTRY, o, np);
1272 r = journal_file_link_entry(f, o, np);
1285 void journal_file_post_change(JournalFile *f) {
1288 /* inotify() does not receive IN_MODIFY events from file
1289 * accesses done via mmap(). After each access we hence
1290 * trigger IN_MODIFY by truncating the journal file to its
1291 * current size which triggers IN_MODIFY. */
1293 __sync_synchronize();
1295 if (ftruncate(f->fd, f->last_stat.st_size) < 0)
1296 log_error("Failed to truncate file to its own size: %m");
1299 static int entry_item_cmp(const void *_a, const void *_b) {
1300 const EntryItem *a = _a, *b = _b;
1302 if (le64toh(a->object_offset) < le64toh(b->object_offset))
1304 if (le64toh(a->object_offset) > le64toh(b->object_offset))
1309 int journal_file_append_entry(JournalFile *f, const dual_timestamp *ts, const struct iovec iovec[], unsigned n_iovec, uint64_t *seqnum, Object **ret, uint64_t *offset) {
1313 uint64_t xor_hash = 0;
1314 struct dual_timestamp _ts;
1317 assert(iovec || n_iovec == 0);
1320 dual_timestamp_get(&_ts);
1324 if (f->tail_entry_monotonic_valid &&
1325 ts->monotonic < le64toh(f->header->tail_entry_monotonic))
1329 r = journal_file_maybe_append_tag(f, ts->realtime);
1334 /* alloca() can't take 0, hence let's allocate at least one */
1335 items = alloca(sizeof(EntryItem) * MAX(1u, n_iovec));
1337 for (i = 0; i < n_iovec; i++) {
1341 r = journal_file_append_data(f, iovec[i].iov_base, iovec[i].iov_len, &o, &p);
1345 xor_hash ^= le64toh(o->data.hash);
1346 items[i].object_offset = htole64(p);
1347 items[i].hash = o->data.hash;
1350 /* Order by the position on disk, in order to improve seek
1351 * times for rotating media. */
1352 qsort_safe(items, n_iovec, sizeof(EntryItem), entry_item_cmp);
1354 r = journal_file_append_entry_internal(f, ts, xor_hash, items, n_iovec, seqnum, ret, offset);
1356 journal_file_post_change(f);
1361 typedef struct ChainCacheItem {
1362 uint64_t first; /* the array at the begin of the chain */
1363 uint64_t array; /* the cached array */
1364 uint64_t begin; /* the first item in the cached array */
1365 uint64_t total; /* the total number of items in all arrays before this one in the chain */
1366 uint64_t last_index; /* the last index we looked at, to optimize locality when bisecting */
1369 static void chain_cache_put(
1376 uint64_t last_index) {
1379 /* If the chain item to cache for this chain is the
1380 * first one it's not worth caching anything */
1384 if (hashmap_size(h) >= CHAIN_CACHE_MAX)
1385 ci = hashmap_steal_first(h);
1387 ci = new(ChainCacheItem, 1);
1394 if (hashmap_put(h, &ci->first, ci) < 0) {
1399 assert(ci->first == first);
1404 ci->last_index = last_index;
1407 static int generic_array_get(
1411 Object **ret, uint64_t *offset) {
1414 uint64_t p = 0, a, t = 0;
1422 /* Try the chain cache first */
1423 ci = hashmap_get(f->chain_cache, &first);
1424 if (ci && i > ci->total) {
1433 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
1437 k = journal_file_entry_array_n_items(o);
1439 p = le64toh(o->entry_array.items[i]);
1445 a = le64toh(o->entry_array.next_entry_array_offset);
1451 /* Let's cache this item for the next invocation */
1452 chain_cache_put(f->chain_cache, ci, first, a, le64toh(o->entry_array.items[0]), t, i);
1454 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1467 static int generic_array_get_plus_one(
1472 Object **ret, uint64_t *offset) {
1481 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
1494 return generic_array_get(f, first, i-1, ret, offset);
1503 static int generic_array_bisect(
1508 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
1509 direction_t direction,
1514 uint64_t a, p, t = 0, i = 0, last_p = 0, last_index = (uint64_t) -1;
1515 bool subtract_one = false;
1516 Object *o, *array = NULL;
1521 assert(test_object);
1523 /* Start with the first array in the chain */
1526 ci = hashmap_get(f->chain_cache, &first);
1527 if (ci && n > ci->total) {
1528 /* Ah, we have iterated this bisection array chain
1529 * previously! Let's see if we can skip ahead in the
1530 * chain, as far as the last time. But we can't jump
1531 * backwards in the chain, so let's check that
1534 r = test_object(f, ci->begin, needle);
1538 if (r == TEST_LEFT) {
1539 /* OK, what we are looking for is right of the
1540 * begin of this EntryArray, so let's jump
1541 * straight to previously cached array in the
1547 last_index = ci->last_index;
1552 uint64_t left, right, k, lp;
1554 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &array);
1558 k = journal_file_entry_array_n_items(array);
1564 lp = p = le64toh(array->entry_array.items[i]);
1568 r = test_object(f, p, needle);
1572 if (r == TEST_FOUND)
1573 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
1575 if (r == TEST_RIGHT) {
1579 if (last_index != (uint64_t) -1) {
1580 assert(last_index <= right);
1582 /* If we cached the last index we
1583 * looked at, let's try to not to jump
1584 * too wildly around and see if we can
1585 * limit the range to look at early to
1586 * the immediate neighbors of the last
1587 * index we looked at. */
1589 if (last_index > 0) {
1590 uint64_t x = last_index - 1;
1592 p = le64toh(array->entry_array.items[x]);
1596 r = test_object(f, p, needle);
1600 if (r == TEST_FOUND)
1601 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
1603 if (r == TEST_RIGHT)
1609 if (last_index < right) {
1610 uint64_t y = last_index + 1;
1612 p = le64toh(array->entry_array.items[y]);
1616 r = test_object(f, p, needle);
1620 if (r == TEST_FOUND)
1621 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
1623 if (r == TEST_RIGHT)
1631 if (left == right) {
1632 if (direction == DIRECTION_UP)
1633 subtract_one = true;
1639 assert(left < right);
1640 i = (left + right) / 2;
1642 p = le64toh(array->entry_array.items[i]);
1646 r = test_object(f, p, needle);
1650 if (r == TEST_FOUND)
1651 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
1653 if (r == TEST_RIGHT)
1661 if (direction == DIRECTION_UP) {
1663 subtract_one = true;
1674 last_index = (uint64_t) -1;
1675 a = le64toh(array->entry_array.next_entry_array_offset);
1681 if (subtract_one && t == 0 && i == 0)
1684 /* Let's cache this item for the next invocation */
1685 chain_cache_put(f->chain_cache, ci, first, a, le64toh(array->entry_array.items[0]), t, subtract_one ? (i > 0 ? i-1 : (uint64_t) -1) : i);
1687 if (subtract_one && i == 0)
1689 else if (subtract_one)
1690 p = le64toh(array->entry_array.items[i-1]);
1692 p = le64toh(array->entry_array.items[i]);
1694 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1705 *idx = t + i + (subtract_one ? -1 : 0);
1711 static int generic_array_bisect_plus_one(
1717 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
1718 direction_t direction,
1724 bool step_back = false;
1728 assert(test_object);
1733 /* This bisects the array in object 'first', but first checks
1735 r = test_object(f, extra, needle);
1739 if (r == TEST_FOUND)
1740 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
1742 /* if we are looking with DIRECTION_UP then we need to first
1743 see if in the actual array there is a matching entry, and
1744 return the last one of that. But if there isn't any we need
1745 to return this one. Hence remember this, and return it
1748 step_back = direction == DIRECTION_UP;
1750 if (r == TEST_RIGHT) {
1751 if (direction == DIRECTION_DOWN)
1757 r = generic_array_bisect(f, first, n-1, needle, test_object, direction, ret, offset, idx);
1759 if (r == 0 && step_back)
1768 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
1784 _pure_ static int test_object_offset(JournalFile *f, uint64_t p, uint64_t needle) {
1790 else if (p < needle)
1796 int journal_file_move_to_entry_by_offset(
1799 direction_t direction,
1803 return generic_array_bisect(f,
1804 le64toh(f->header->entry_array_offset),
1805 le64toh(f->header->n_entries),
1813 static int test_object_seqnum(JournalFile *f, uint64_t p, uint64_t needle) {
1820 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1824 if (le64toh(o->entry.seqnum) == needle)
1826 else if (le64toh(o->entry.seqnum) < needle)
1832 int journal_file_move_to_entry_by_seqnum(
1835 direction_t direction,
1839 return generic_array_bisect(f,
1840 le64toh(f->header->entry_array_offset),
1841 le64toh(f->header->n_entries),
1848 static int test_object_realtime(JournalFile *f, uint64_t p, uint64_t needle) {
1855 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1859 if (le64toh(o->entry.realtime) == needle)
1861 else if (le64toh(o->entry.realtime) < needle)
1867 int journal_file_move_to_entry_by_realtime(
1870 direction_t direction,
1874 return generic_array_bisect(f,
1875 le64toh(f->header->entry_array_offset),
1876 le64toh(f->header->n_entries),
1878 test_object_realtime,
1883 static int test_object_monotonic(JournalFile *f, uint64_t p, uint64_t needle) {
1890 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1894 if (le64toh(o->entry.monotonic) == needle)
1896 else if (le64toh(o->entry.monotonic) < needle)
1902 static inline int find_data_object_by_boot_id(
1907 char t[sizeof("_BOOT_ID=")-1 + 32 + 1] = "_BOOT_ID=";
1909 sd_id128_to_string(boot_id, t + 9);
1910 return journal_file_find_data_object(f, t, sizeof(t) - 1, o, b);
1913 int journal_file_move_to_entry_by_monotonic(
1917 direction_t direction,
1926 r = find_data_object_by_boot_id(f, boot_id, &o, NULL);
1932 return generic_array_bisect_plus_one(f,
1933 le64toh(o->data.entry_offset),
1934 le64toh(o->data.entry_array_offset),
1935 le64toh(o->data.n_entries),
1937 test_object_monotonic,
1942 int journal_file_next_entry(
1944 Object *o, uint64_t p,
1945 direction_t direction,
1946 Object **ret, uint64_t *offset) {
1952 assert(p > 0 || !o);
1954 n = le64toh(f->header->n_entries);
1959 i = direction == DIRECTION_DOWN ? 0 : n - 1;
1961 if (o->object.type != OBJECT_ENTRY)
1964 r = generic_array_bisect(f,
1965 le64toh(f->header->entry_array_offset),
1966 le64toh(f->header->n_entries),
1975 if (direction == DIRECTION_DOWN) {
1988 /* And jump to it */
1989 return generic_array_get(f,
1990 le64toh(f->header->entry_array_offset),
1995 int journal_file_skip_entry(
1997 Object *o, uint64_t p,
1999 Object **ret, uint64_t *offset) {
2008 if (o->object.type != OBJECT_ENTRY)
2011 r = generic_array_bisect(f,
2012 le64toh(f->header->entry_array_offset),
2013 le64toh(f->header->n_entries),
2022 /* Calculate new index */
2024 if ((uint64_t) -skip >= i)
2027 i = i - (uint64_t) -skip;
2029 i += (uint64_t) skip;
2031 n = le64toh(f->header->n_entries);
2038 return generic_array_get(f,
2039 le64toh(f->header->entry_array_offset),
2044 int journal_file_next_entry_for_data(
2046 Object *o, uint64_t p,
2047 uint64_t data_offset,
2048 direction_t direction,
2049 Object **ret, uint64_t *offset) {
2056 assert(p > 0 || !o);
2058 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2062 n = le64toh(d->data.n_entries);
2067 i = direction == DIRECTION_DOWN ? 0 : n - 1;
2069 if (o->object.type != OBJECT_ENTRY)
2072 r = generic_array_bisect_plus_one(f,
2073 le64toh(d->data.entry_offset),
2074 le64toh(d->data.entry_array_offset),
2075 le64toh(d->data.n_entries),
2085 if (direction == DIRECTION_DOWN) {
2099 return generic_array_get_plus_one(f,
2100 le64toh(d->data.entry_offset),
2101 le64toh(d->data.entry_array_offset),
2106 int journal_file_move_to_entry_by_offset_for_data(
2108 uint64_t data_offset,
2110 direction_t direction,
2111 Object **ret, uint64_t *offset) {
2118 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2122 return generic_array_bisect_plus_one(f,
2123 le64toh(d->data.entry_offset),
2124 le64toh(d->data.entry_array_offset),
2125 le64toh(d->data.n_entries),
2132 int journal_file_move_to_entry_by_monotonic_for_data(
2134 uint64_t data_offset,
2137 direction_t direction,
2138 Object **ret, uint64_t *offset) {
2146 /* First, seek by time */
2147 r = find_data_object_by_boot_id(f, boot_id, &o, &b);
2153 r = generic_array_bisect_plus_one(f,
2154 le64toh(o->data.entry_offset),
2155 le64toh(o->data.entry_array_offset),
2156 le64toh(o->data.n_entries),
2158 test_object_monotonic,
2164 /* And now, continue seeking until we find an entry that
2165 * exists in both bisection arrays */
2171 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2175 r = generic_array_bisect_plus_one(f,
2176 le64toh(d->data.entry_offset),
2177 le64toh(d->data.entry_array_offset),
2178 le64toh(d->data.n_entries),
2186 r = journal_file_move_to_object(f, OBJECT_DATA, b, &o);
2190 r = generic_array_bisect_plus_one(f,
2191 le64toh(o->data.entry_offset),
2192 le64toh(o->data.entry_array_offset),
2193 le64toh(o->data.n_entries),
2215 int journal_file_move_to_entry_by_seqnum_for_data(
2217 uint64_t data_offset,
2219 direction_t direction,
2220 Object **ret, uint64_t *offset) {
2227 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2231 return generic_array_bisect_plus_one(f,
2232 le64toh(d->data.entry_offset),
2233 le64toh(d->data.entry_array_offset),
2234 le64toh(d->data.n_entries),
2241 int journal_file_move_to_entry_by_realtime_for_data(
2243 uint64_t data_offset,
2245 direction_t direction,
2246 Object **ret, uint64_t *offset) {
2253 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2257 return generic_array_bisect_plus_one(f,
2258 le64toh(d->data.entry_offset),
2259 le64toh(d->data.entry_array_offset),
2260 le64toh(d->data.n_entries),
2262 test_object_realtime,
2267 void journal_file_dump(JournalFile *f) {
2274 journal_file_print_header(f);
2276 p = le64toh(f->header->header_size);
2278 r = journal_file_move_to_object(f, -1, p, &o);
2282 switch (o->object.type) {
2285 printf("Type: OBJECT_UNUSED\n");
2289 printf("Type: OBJECT_DATA\n");
2293 printf("Type: OBJECT_FIELD\n");
2297 printf("Type: OBJECT_ENTRY seqnum=%"PRIu64" monotonic=%"PRIu64" realtime=%"PRIu64"\n",
2298 le64toh(o->entry.seqnum),
2299 le64toh(o->entry.monotonic),
2300 le64toh(o->entry.realtime));
2303 case OBJECT_FIELD_HASH_TABLE:
2304 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
2307 case OBJECT_DATA_HASH_TABLE:
2308 printf("Type: OBJECT_DATA_HASH_TABLE\n");
2311 case OBJECT_ENTRY_ARRAY:
2312 printf("Type: OBJECT_ENTRY_ARRAY\n");
2316 printf("Type: OBJECT_TAG seqnum=%"PRIu64" epoch=%"PRIu64"\n",
2317 le64toh(o->tag.seqnum),
2318 le64toh(o->tag.epoch));
2322 printf("Type: unknown (%u)\n", o->object.type);
2326 if (o->object.flags & OBJECT_COMPRESSED)
2327 printf("Flags: COMPRESSED\n");
2329 if (p == le64toh(f->header->tail_object_offset))
2332 p = p + ALIGN64(le64toh(o->object.size));
2337 log_error("File corrupt");
2340 static const char* format_timestamp_safe(char *buf, size_t l, usec_t t) {
2343 x = format_timestamp(buf, l, t);
2349 void journal_file_print_header(JournalFile *f) {
2350 char a[33], b[33], c[33], d[33];
2351 char x[FORMAT_TIMESTAMP_MAX], y[FORMAT_TIMESTAMP_MAX], z[FORMAT_TIMESTAMP_MAX];
2353 char bytes[FORMAT_BYTES_MAX];
2357 printf("File Path: %s\n"
2361 "Sequential Number ID: %s\n"
2363 "Compatible Flags:%s%s\n"
2364 "Incompatible Flags:%s%s\n"
2365 "Header size: %"PRIu64"\n"
2366 "Arena size: %"PRIu64"\n"
2367 "Data Hash Table Size: %"PRIu64"\n"
2368 "Field Hash Table Size: %"PRIu64"\n"
2369 "Rotate Suggested: %s\n"
2370 "Head Sequential Number: %"PRIu64"\n"
2371 "Tail Sequential Number: %"PRIu64"\n"
2372 "Head Realtime Timestamp: %s\n"
2373 "Tail Realtime Timestamp: %s\n"
2374 "Tail Monotonic Timestamp: %s\n"
2375 "Objects: %"PRIu64"\n"
2376 "Entry Objects: %"PRIu64"\n",
2378 sd_id128_to_string(f->header->file_id, a),
2379 sd_id128_to_string(f->header->machine_id, b),
2380 sd_id128_to_string(f->header->boot_id, c),
2381 sd_id128_to_string(f->header->seqnum_id, d),
2382 f->header->state == STATE_OFFLINE ? "OFFLINE" :
2383 f->header->state == STATE_ONLINE ? "ONLINE" :
2384 f->header->state == STATE_ARCHIVED ? "ARCHIVED" : "UNKNOWN",
2385 JOURNAL_HEADER_SEALED(f->header) ? " SEALED" : "",
2386 (le32toh(f->header->compatible_flags) & ~HEADER_COMPATIBLE_SEALED) ? " ???" : "",
2387 JOURNAL_HEADER_COMPRESSED(f->header) ? " COMPRESSED" : "",
2388 (le32toh(f->header->incompatible_flags) & ~HEADER_INCOMPATIBLE_COMPRESSED) ? " ???" : "",
2389 le64toh(f->header->header_size),
2390 le64toh(f->header->arena_size),
2391 le64toh(f->header->data_hash_table_size) / sizeof(HashItem),
2392 le64toh(f->header->field_hash_table_size) / sizeof(HashItem),
2393 yes_no(journal_file_rotate_suggested(f, 0)),
2394 le64toh(f->header->head_entry_seqnum),
2395 le64toh(f->header->tail_entry_seqnum),
2396 format_timestamp_safe(x, sizeof(x), le64toh(f->header->head_entry_realtime)),
2397 format_timestamp_safe(y, sizeof(y), le64toh(f->header->tail_entry_realtime)),
2398 format_timespan(z, sizeof(z), le64toh(f->header->tail_entry_monotonic), USEC_PER_MSEC),
2399 le64toh(f->header->n_objects),
2400 le64toh(f->header->n_entries));
2402 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
2403 printf("Data Objects: %"PRIu64"\n"
2404 "Data Hash Table Fill: %.1f%%\n",
2405 le64toh(f->header->n_data),
2406 100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem))));
2408 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
2409 printf("Field Objects: %"PRIu64"\n"
2410 "Field Hash Table Fill: %.1f%%\n",
2411 le64toh(f->header->n_fields),
2412 100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem))));
2414 if (JOURNAL_HEADER_CONTAINS(f->header, n_tags))
2415 printf("Tag Objects: %"PRIu64"\n",
2416 le64toh(f->header->n_tags));
2417 if (JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
2418 printf("Entry Array Objects: %"PRIu64"\n",
2419 le64toh(f->header->n_entry_arrays));
2421 if (fstat(f->fd, &st) >= 0)
2422 printf("Disk usage: %s\n", format_bytes(bytes, sizeof(bytes), (off_t) st.st_blocks * 512ULL));
2425 int journal_file_open(
2431 JournalMetrics *metrics,
2432 MMapCache *mmap_cache,
2433 JournalFile *template,
2434 JournalFile **ret) {
2438 bool newly_created = false;
2443 if ((flags & O_ACCMODE) != O_RDONLY &&
2444 (flags & O_ACCMODE) != O_RDWR)
2447 if (!endswith(fname, ".journal") &&
2448 !endswith(fname, ".journal~"))
2451 f = new0(JournalFile, 1);
2459 f->prot = prot_from_flags(flags);
2460 f->writable = (flags & O_ACCMODE) != O_RDONLY;
2462 f->compress = compress;
2469 f->mmap = mmap_cache_ref(mmap_cache);
2471 f->mmap = mmap_cache_new();
2478 f->path = strdup(fname);
2484 f->chain_cache = hashmap_new(uint64_hash_func, uint64_compare_func);
2485 if (!f->chain_cache) {
2490 f->fd = open(f->path, f->flags|O_CLOEXEC, f->mode);
2496 if (fstat(f->fd, &f->last_stat) < 0) {
2501 if (f->last_stat.st_size == 0 && f->writable) {
2505 /* Let's attach the creation time to the journal file,
2506 * so that the vacuuming code knows the age of this
2507 * file even if the file might end up corrupted one
2508 * day... Ideally we'd just use the creation time many
2509 * file systems maintain for each file, but there is
2510 * currently no usable API to query this, hence let's
2511 * emulate this via extended attributes. If extended
2512 * attributes are not supported we'll just skip this,
2513 * and rely solely on mtime/atime/ctime of the file.*/
2515 crtime = htole64((uint64_t) now(CLOCK_REALTIME));
2516 fsetxattr(f->fd, "user.crtime_usec", &crtime, sizeof(crtime), XATTR_CREATE);
2520 /* Try to load the FSPRG state, and if we can't, then
2521 * just don't do sealing */
2523 r = journal_file_fss_load(f);
2529 r = journal_file_init_header(f, template);
2533 if (fstat(f->fd, &f->last_stat) < 0) {
2538 newly_created = true;
2541 if (f->last_stat.st_size < (off_t) HEADER_SIZE_MIN) {
2546 f->header = mmap(NULL, PAGE_ALIGN(sizeof(Header)), prot_from_flags(flags), MAP_SHARED, f->fd, 0);
2547 if (f->header == MAP_FAILED) {
2553 if (!newly_created) {
2554 r = journal_file_verify_header(f);
2560 if (!newly_created && f->writable) {
2561 r = journal_file_fss_load(f);
2569 journal_default_metrics(metrics, f->fd);
2570 f->metrics = *metrics;
2571 } else if (template)
2572 f->metrics = template->metrics;
2574 r = journal_file_refresh_header(f);
2580 r = journal_file_hmac_setup(f);
2585 if (newly_created) {
2586 r = journal_file_setup_field_hash_table(f);
2590 r = journal_file_setup_data_hash_table(f);
2595 r = journal_file_append_first_tag(f);
2601 r = journal_file_map_field_hash_table(f);
2605 r = journal_file_map_data_hash_table(f);
2613 journal_file_close(f);
2618 int journal_file_rotate(JournalFile **f, bool compress, bool seal) {
2619 _cleanup_free_ char *p = NULL;
2621 JournalFile *old_file, *new_file = NULL;
2629 if (!old_file->writable)
2632 if (!endswith(old_file->path, ".journal"))
2635 l = strlen(old_file->path);
2636 r = asprintf(&p, "%.*s@" SD_ID128_FORMAT_STR "-%016"PRIx64"-%016"PRIx64".journal",
2637 (int) l - 8, old_file->path,
2638 SD_ID128_FORMAT_VAL(old_file->header->seqnum_id),
2639 le64toh((*f)->header->head_entry_seqnum),
2640 le64toh((*f)->header->head_entry_realtime));
2644 r = rename(old_file->path, p);
2648 old_file->header->state = STATE_ARCHIVED;
2650 r = journal_file_open(old_file->path, old_file->flags, old_file->mode, compress, seal, NULL, old_file->mmap, old_file, &new_file);
2651 journal_file_close(old_file);
2657 int journal_file_open_reliably(
2663 JournalMetrics *metrics,
2664 MMapCache *mmap_cache,
2665 JournalFile *template,
2666 JournalFile **ret) {
2670 _cleanup_free_ char *p = NULL;
2672 r = journal_file_open(fname, flags, mode, compress, seal,
2673 metrics, mmap_cache, template, ret);
2674 if (r != -EBADMSG && /* corrupted */
2675 r != -ENODATA && /* truncated */
2676 r != -EHOSTDOWN && /* other machine */
2677 r != -EPROTONOSUPPORT && /* incompatible feature */
2678 r != -EBUSY && /* unclean shutdown */
2679 r != -ESHUTDOWN /* already archived */)
2682 if ((flags & O_ACCMODE) == O_RDONLY)
2685 if (!(flags & O_CREAT))
2688 if (!endswith(fname, ".journal"))
2691 /* The file is corrupted. Rotate it away and try it again (but only once) */
2694 if (asprintf(&p, "%.*s@%016llx-%016" PRIx64 ".journal~",
2696 (unsigned long long) now(CLOCK_REALTIME),
2700 r = rename(fname, p);
2704 log_warning("File %s corrupted or uncleanly shut down, renaming and replacing.", fname);
2706 return journal_file_open(fname, flags, mode, compress, seal,
2707 metrics, mmap_cache, template, ret);
2710 int journal_file_copy_entry(JournalFile *from, JournalFile *to, Object *o, uint64_t p, uint64_t *seqnum, Object **ret, uint64_t *offset) {
2712 uint64_t q, xor_hash = 0;
2725 ts.monotonic = le64toh(o->entry.monotonic);
2726 ts.realtime = le64toh(o->entry.realtime);
2728 n = journal_file_entry_n_items(o);
2729 /* alloca() can't take 0, hence let's allocate at least one */
2730 items = alloca(sizeof(EntryItem) * MAX(1u, n));
2732 for (i = 0; i < n; i++) {
2739 q = le64toh(o->entry.items[i].object_offset);
2740 le_hash = o->entry.items[i].hash;
2742 r = journal_file_move_to_object(from, OBJECT_DATA, q, &o);
2746 if (le_hash != o->data.hash)
2749 l = le64toh(o->object.size) - offsetof(Object, data.payload);
2752 /* We hit the limit on 32bit machines */
2753 if ((uint64_t) t != l)
2756 if (o->object.flags & OBJECT_COMPRESSED) {
2760 if (!uncompress_blob(o->data.payload, l, &from->compress_buffer, &from->compress_buffer_size, &rsize, 0))
2763 data = from->compress_buffer;
2766 return -EPROTONOSUPPORT;
2769 data = o->data.payload;
2771 r = journal_file_append_data(to, data, l, &u, &h);
2775 xor_hash ^= le64toh(u->data.hash);
2776 items[i].object_offset = htole64(h);
2777 items[i].hash = u->data.hash;
2779 r = journal_file_move_to_object(from, OBJECT_ENTRY, p, &o);
2784 return journal_file_append_entry_internal(to, &ts, xor_hash, items, n, seqnum, ret, offset);
2787 void journal_default_metrics(JournalMetrics *m, int fd) {
2788 uint64_t fs_size = 0;
2790 char a[FORMAT_BYTES_MAX], b[FORMAT_BYTES_MAX], c[FORMAT_BYTES_MAX], d[FORMAT_BYTES_MAX];
2795 if (fstatvfs(fd, &ss) >= 0)
2796 fs_size = ss.f_frsize * ss.f_blocks;
2798 if (m->max_use == (uint64_t) -1) {
2801 m->max_use = PAGE_ALIGN(fs_size / 10); /* 10% of file system size */
2803 if (m->max_use > DEFAULT_MAX_USE_UPPER)
2804 m->max_use = DEFAULT_MAX_USE_UPPER;
2806 if (m->max_use < DEFAULT_MAX_USE_LOWER)
2807 m->max_use = DEFAULT_MAX_USE_LOWER;
2809 m->max_use = DEFAULT_MAX_USE_LOWER;
2811 m->max_use = PAGE_ALIGN(m->max_use);
2813 if (m->max_use < JOURNAL_FILE_SIZE_MIN*2)
2814 m->max_use = JOURNAL_FILE_SIZE_MIN*2;
2817 if (m->max_size == (uint64_t) -1) {
2818 m->max_size = PAGE_ALIGN(m->max_use / 8); /* 8 chunks */
2820 if (m->max_size > DEFAULT_MAX_SIZE_UPPER)
2821 m->max_size = DEFAULT_MAX_SIZE_UPPER;
2823 m->max_size = PAGE_ALIGN(m->max_size);
2825 if (m->max_size < JOURNAL_FILE_SIZE_MIN)
2826 m->max_size = JOURNAL_FILE_SIZE_MIN;
2828 if (m->max_size*2 > m->max_use)
2829 m->max_use = m->max_size*2;
2831 if (m->min_size == (uint64_t) -1)
2832 m->min_size = JOURNAL_FILE_SIZE_MIN;
2834 m->min_size = PAGE_ALIGN(m->min_size);
2836 if (m->min_size < JOURNAL_FILE_SIZE_MIN)
2837 m->min_size = JOURNAL_FILE_SIZE_MIN;
2839 if (m->min_size > m->max_size)
2840 m->max_size = m->min_size;
2843 if (m->keep_free == (uint64_t) -1) {
2846 m->keep_free = PAGE_ALIGN(fs_size * 3 / 20); /* 15% of file system size */
2848 if (m->keep_free > DEFAULT_KEEP_FREE_UPPER)
2849 m->keep_free = DEFAULT_KEEP_FREE_UPPER;
2852 m->keep_free = DEFAULT_KEEP_FREE;
2855 log_debug("Fixed max_use=%s max_size=%s min_size=%s keep_free=%s",
2856 format_bytes(a, sizeof(a), m->max_use),
2857 format_bytes(b, sizeof(b), m->max_size),
2858 format_bytes(c, sizeof(c), m->min_size),
2859 format_bytes(d, sizeof(d), m->keep_free));
2862 int journal_file_get_cutoff_realtime_usec(JournalFile *f, usec_t *from, usec_t *to) {
2867 if (f->header->head_entry_realtime == 0)
2870 *from = le64toh(f->header->head_entry_realtime);
2874 if (f->header->tail_entry_realtime == 0)
2877 *to = le64toh(f->header->tail_entry_realtime);
2883 int journal_file_get_cutoff_monotonic_usec(JournalFile *f, sd_id128_t boot_id, usec_t *from, usec_t *to) {
2891 r = find_data_object_by_boot_id(f, boot_id, &o, &p);
2895 if (le64toh(o->data.n_entries) <= 0)
2899 r = journal_file_move_to_object(f, OBJECT_ENTRY, le64toh(o->data.entry_offset), &o);
2903 *from = le64toh(o->entry.monotonic);
2907 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
2911 r = generic_array_get_plus_one(f,
2912 le64toh(o->data.entry_offset),
2913 le64toh(o->data.entry_array_offset),
2914 le64toh(o->data.n_entries)-1,
2919 *to = le64toh(o->entry.monotonic);
2925 bool journal_file_rotate_suggested(JournalFile *f, usec_t max_file_usec) {
2928 /* If we gained new header fields we gained new features,
2929 * hence suggest a rotation */
2930 if (le64toh(f->header->header_size) < sizeof(Header)) {
2931 log_debug("%s uses an outdated header, suggesting rotation.", f->path);
2935 /* Let's check if the hash tables grew over a certain fill
2936 * level (75%, borrowing this value from Java's hash table
2937 * implementation), and if so suggest a rotation. To calculate
2938 * the fill level we need the n_data field, which only exists
2939 * in newer versions. */
2941 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
2942 if (le64toh(f->header->n_data) * 4ULL > (le64toh(f->header->data_hash_table_size) / sizeof(HashItem)) * 3ULL) {
2943 log_debug("Data hash table of %s has a fill level at %.1f (%"PRIu64" of %"PRIu64" items, %llu file size, %"PRIu64" bytes per hash table item), suggesting rotation.",
2945 100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem))),
2946 le64toh(f->header->n_data),
2947 le64toh(f->header->data_hash_table_size) / sizeof(HashItem),
2948 (unsigned long long) f->last_stat.st_size,
2949 f->last_stat.st_size / le64toh(f->header->n_data));
2953 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
2954 if (le64toh(f->header->n_fields) * 4ULL > (le64toh(f->header->field_hash_table_size) / sizeof(HashItem)) * 3ULL) {
2955 log_debug("Field hash table of %s has a fill level at %.1f (%"PRIu64" of %"PRIu64" items), suggesting rotation.",
2957 100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem))),
2958 le64toh(f->header->n_fields),
2959 le64toh(f->header->field_hash_table_size) / sizeof(HashItem));
2963 /* Are the data objects properly indexed by field objects? */
2964 if (JOURNAL_HEADER_CONTAINS(f->header, n_data) &&
2965 JOURNAL_HEADER_CONTAINS(f->header, n_fields) &&
2966 le64toh(f->header->n_data) > 0 &&
2967 le64toh(f->header->n_fields) == 0)
2970 if (max_file_usec > 0) {
2973 h = le64toh(f->header->head_entry_realtime);
2974 t = now(CLOCK_REALTIME);
2976 if (h > 0 && t > h + max_file_usec)