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 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.
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.
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/>.
26 #include <sys/statvfs.h>
30 #include "journal-def.h"
31 #include "journal-file.h"
35 #define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*16ULL)
36 #define DEFAULT_FIELD_HASH_TABLE_SIZE (2047ULL*16ULL)
38 #define DEFAULT_WINDOW_SIZE (128ULL*1024ULL*1024ULL)
40 #define COMPRESSION_SIZE_THRESHOLD (64ULL)
42 static const char signature[] = { 'L', 'P', 'K', 'S', 'H', 'H', 'R', 'H' };
44 #define ALIGN64(x) (((x) + 7ULL) & ~7ULL)
46 void journal_file_close(JournalFile *f) {
51 if (f->header && f->writable)
52 f->header->state = STATE_OFFLINE;
55 for (t = 0; t < _WINDOW_MAX; t++)
56 if (f->windows[t].ptr)
57 munmap(f->windows[t].ptr, f->windows[t].size);
60 close_nointr_nofail(f->fd);
65 free(f->compress_buffer);
71 static int journal_file_init_header(JournalFile *f, JournalFile *template) {
79 memcpy(h.signature, signature, 8);
80 h.arena_offset = htole64(ALIGN64(sizeof(h)));
82 r = sd_id128_randomize(&h.file_id);
87 h.seqnum_id = template->header->seqnum_id;
88 h.seqnum = template->header->seqnum;
90 h.seqnum_id = h.file_id;
92 k = pwrite(f->fd, &h, sizeof(h), 0);
102 static int journal_file_refresh_header(JournalFile *f) {
108 r = sd_id128_get_machine(&f->header->machine_id);
112 r = sd_id128_get_boot(&boot_id);
116 if (sd_id128_equal(boot_id, f->header->boot_id))
117 f->tail_entry_monotonic_valid = true;
119 f->header->boot_id = boot_id;
121 f->header->state = STATE_ONLINE;
125 static int journal_file_verify_header(JournalFile *f) {
128 if (memcmp(f->header, signature, 8))
132 if ((le64toh(f->header->incompatible_flags) & ~HEADER_INCOMPATIBLE_COMPRESSED) != 0)
133 return -EPROTONOSUPPORT;
135 if (f->header->incompatible_flags != 0)
136 return -EPROTONOSUPPORT;
139 if ((uint64_t) f->last_stat.st_size < (le64toh(f->header->arena_offset) + le64toh(f->header->arena_size)))
144 sd_id128_t machine_id;
147 r = sd_id128_get_machine(&machine_id);
151 if (!sd_id128_equal(machine_id, f->header->machine_id))
154 state = f->header->state;
156 if (state == STATE_ONLINE)
157 log_debug("Journal file %s is already online. Assuming unclean closing. Ignoring.", f->path);
158 else if (state == STATE_ARCHIVED)
160 else if (state != STATE_OFFLINE)
161 log_debug("Journal file %s has unknown state %u. Ignoring.", f->path, state);
167 static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size) {
168 uint64_t old_size, new_size;
172 /* We assume that this file is not sparse, and we know that
173 * for sure, since we always call posix_fallocate()
177 le64toh(f->header->arena_offset) +
178 le64toh(f->header->arena_size);
180 new_size = PAGE_ALIGN(offset + size);
181 if (new_size < le64toh(f->header->arena_offset))
182 new_size = le64toh(f->header->arena_offset);
184 if (new_size <= old_size)
187 if (f->metrics.max_size > 0 &&
188 new_size > f->metrics.max_size)
191 if (new_size > f->metrics.min_size &&
192 f->metrics.keep_free > 0) {
195 if (fstatvfs(f->fd, &svfs) >= 0) {
198 available = svfs.f_bfree * svfs.f_bsize;
200 if (available >= f->metrics.keep_free)
201 available -= f->metrics.keep_free;
205 if (new_size - old_size > available)
210 /* Note that the glibc fallocate() fallback is very
211 inefficient, hence we try to minimize the allocation area
213 if (posix_fallocate(f->fd, old_size, new_size - old_size) < 0)
216 if (fstat(f->fd, &f->last_stat) < 0)
219 f->header->arena_size = new_size - htole64(f->header->arena_offset);
224 static int journal_file_map(
233 uint64_t woffset, wsize;
240 woffset = offset & ~((uint64_t) page_size() - 1ULL);
241 wsize = size + (offset - woffset);
242 wsize = PAGE_ALIGN(wsize);
244 window = mmap(NULL, wsize, f->prot, MAP_SHARED, f->fd, woffset);
245 if (window == MAP_FAILED)
257 *ret = (uint8_t*) window + (offset - woffset);
262 static int journal_file_move_to(JournalFile *f, int wt, uint64_t offset, uint64_t size, void **ret) {
271 assert(wt < _WINDOW_MAX);
275 if (_likely_(w->ptr &&
276 w->offset <= offset &&
277 w->offset + w->size >= offset + size)) {
279 *ret = (uint8_t*) w->ptr + (offset - w->offset);
284 if (munmap(w->ptr, w->size) < 0)
288 w->size = w->offset = 0;
291 if (size < DEFAULT_WINDOW_SIZE) {
292 /* If the default window size is larger then what was
293 * asked for extend the mapping a bit in the hope to
294 * minimize needed remappings later on. We add half
295 * the window space before and half behind the
296 * requested mapping */
298 delta = PAGE_ALIGN((DEFAULT_WINDOW_SIZE - size) / 2);
304 size += (DEFAULT_WINDOW_SIZE - delta);
308 r = journal_file_map(f,
310 &w->ptr, &w->offset, &w->size,
316 *ret = (uint8_t*) p + delta;
320 static bool verify_hash(Object *o) {
325 if (o->object.type == OBJECT_DATA && !(o->object.flags & OBJECT_COMPRESSED)) {
326 h1 = le64toh(o->data.hash);
327 h2 = hash64(o->data.payload, le64toh(o->object.size) - offsetof(Object, data.payload));
328 } else if (o->object.type == OBJECT_FIELD) {
329 h1 = le64toh(o->field.hash);
330 h2 = hash64(o->field.payload, le64toh(o->object.size) - offsetof(Object, field.payload));
337 int journal_file_move_to_object(JournalFile *f, int type, uint64_t offset, Object **ret) {
345 assert(type < _OBJECT_TYPE_MAX);
347 r = journal_file_move_to(f, type >= 0 ? type : WINDOW_UNKNOWN, offset, sizeof(ObjectHeader), &t);
352 s = le64toh(o->object.size);
354 if (s < sizeof(ObjectHeader))
357 if (type >= 0 && o->object.type != type)
360 if (s > sizeof(ObjectHeader)) {
361 r = journal_file_move_to(f, o->object.type, offset, s, &t);
375 static uint64_t journal_file_seqnum(JournalFile *f, uint64_t *seqnum) {
380 r = le64toh(f->header->seqnum) + 1;
383 /* If an external seqnum counter was passed, we update
384 * both the local and the external one, and set it to
385 * the maximum of both */
393 f->header->seqnum = htole64(r);
395 if (f->header->first_seqnum == 0)
396 f->header->first_seqnum = htole64(r);
401 static int journal_file_append_object(JournalFile *f, int type, uint64_t size, Object **ret, uint64_t *offset) {
408 assert(size >= sizeof(ObjectHeader));
412 p = le64toh(f->header->tail_object_offset);
414 p = le64toh(f->header->arena_offset);
416 r = journal_file_move_to_object(f, -1, p, &tail);
420 p += ALIGN64(le64toh(tail->object.size));
423 r = journal_file_allocate(f, p, size);
427 r = journal_file_move_to(f, type, p, size, &t);
434 o->object.type = type;
435 o->object.size = htole64(size);
437 f->header->tail_object_offset = htole64(p);
438 f->header->n_objects = htole64(le64toh(f->header->n_objects) + 1);
446 static int journal_file_setup_data_hash_table(JournalFile *f) {
453 s = DEFAULT_DATA_HASH_TABLE_SIZE;
454 r = journal_file_append_object(f,
455 OBJECT_DATA_HASH_TABLE,
456 offsetof(Object, hash_table.items) + s,
461 memset(o->hash_table.items, 0, s);
463 f->header->data_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
464 f->header->data_hash_table_size = htole64(s);
469 static int journal_file_setup_field_hash_table(JournalFile *f) {
476 s = DEFAULT_FIELD_HASH_TABLE_SIZE;
477 r = journal_file_append_object(f,
478 OBJECT_FIELD_HASH_TABLE,
479 offsetof(Object, hash_table.items) + s,
484 memset(o->hash_table.items, 0, s);
486 f->header->field_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
487 f->header->field_hash_table_size = htole64(s);
492 static int journal_file_map_data_hash_table(JournalFile *f) {
499 p = le64toh(f->header->data_hash_table_offset);
500 s = le64toh(f->header->data_hash_table_size);
502 r = journal_file_move_to(f,
503 WINDOW_DATA_HASH_TABLE,
509 f->data_hash_table = t;
513 static int journal_file_map_field_hash_table(JournalFile *f) {
520 p = le64toh(f->header->field_hash_table_offset);
521 s = le64toh(f->header->field_hash_table_size);
523 r = journal_file_move_to(f,
524 WINDOW_FIELD_HASH_TABLE,
530 f->field_hash_table = t;
534 static int journal_file_link_data(JournalFile *f, Object *o, uint64_t offset, uint64_t hash) {
541 assert(o->object.type == OBJECT_DATA);
543 o->data.next_hash_offset = o->data.next_field_offset = 0;
544 o->data.entry_offset = o->data.entry_array_offset = 0;
545 o->data.n_entries = 0;
547 h = hash % (le64toh(f->header->data_hash_table_size) / sizeof(HashItem));
548 p = le64toh(f->data_hash_table[h].head_hash_offset);
550 /* Only entry in the hash table is easy */
551 f->data_hash_table[h].head_hash_offset = htole64(offset);
553 /* Temporarily move back to the previous data object,
554 * to patch in pointer */
556 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
560 o->data.next_hash_offset = htole64(offset);
562 r = journal_file_move_to_object(f, OBJECT_DATA, offset, &o);
567 f->data_hash_table[h].tail_hash_offset = htole64(offset);
572 int journal_file_find_data_object_with_hash(
574 const void *data, uint64_t size, uint64_t hash,
575 Object **ret, uint64_t *offset) {
576 uint64_t p, osize, h;
580 assert(data || size == 0);
582 osize = offsetof(Object, data.payload) + size;
584 if (f->header->data_hash_table_size == 0)
587 h = hash % (le64toh(f->header->data_hash_table_size) / sizeof(HashItem));
588 p = le64toh(f->data_hash_table[h].head_hash_offset);
593 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
597 if (le64toh(o->data.hash) != hash)
600 if (o->object.flags & OBJECT_COMPRESSED) {
604 l = le64toh(o->object.size);
605 if (l <= offsetof(Object, data.payload))
608 l -= offsetof(Object, data.payload);
610 if (!uncompress_blob(o->data.payload, l, &f->compress_buffer, &f->compress_buffer_size, &rsize))
614 memcmp(f->compress_buffer, data, size) == 0) {
625 return -EPROTONOSUPPORT;
628 } else if (le64toh(o->object.size) == osize &&
629 memcmp(o->data.payload, data, size) == 0) {
640 p = le64toh(o->data.next_hash_offset);
646 int journal_file_find_data_object(
648 const void *data, uint64_t size,
649 Object **ret, uint64_t *offset) {
654 assert(data || size == 0);
656 hash = hash64(data, size);
658 return journal_file_find_data_object_with_hash(f,
663 static int journal_file_append_data(JournalFile *f, const void *data, uint64_t size, Object **ret, uint64_t *offset) {
668 bool compressed = false;
671 assert(data || size == 0);
673 hash = hash64(data, size);
675 r = journal_file_find_data_object_with_hash(f, data, size, hash, &o, &p);
689 osize = offsetof(Object, data.payload) + size;
690 r = journal_file_append_object(f, OBJECT_DATA, osize, &o, &p);
694 o->data.hash = htole64(hash);
698 size >= COMPRESSION_SIZE_THRESHOLD) {
701 compressed = compress_blob(data, size, o->data.payload, &rsize);
704 o->object.size = htole64(offsetof(Object, data.payload) + rsize);
705 o->object.flags |= OBJECT_COMPRESSED;
707 f->header->incompatible_flags = htole32(le32toh(f->header->incompatible_flags) | HEADER_INCOMPATIBLE_COMPRESSED);
709 log_debug("Compressed data object %lu -> %lu", (unsigned long) size, (unsigned long) rsize);
715 memcpy(o->data.payload, data, size);
717 r = journal_file_link_data(f, o, p, hash);
730 uint64_t journal_file_entry_n_items(Object *o) {
732 assert(o->object.type == htole64(OBJECT_ENTRY));
734 return (le64toh(o->object.size) - offsetof(Object, entry.items)) / sizeof(EntryItem);
737 static uint64_t journal_file_entry_array_n_items(Object *o) {
739 assert(o->object.type == htole64(OBJECT_ENTRY_ARRAY));
741 return (le64toh(o->object.size) - offsetof(Object, entry_array.items)) / sizeof(uint64_t);
744 static int link_entry_into_array(JournalFile *f,
749 uint64_t n = 0, ap = 0, q, i, a, hidx;
758 i = hidx = le64toh(*idx);
761 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
765 n = journal_file_entry_array_n_items(o);
767 o->entry_array.items[i] = htole64(p);
768 *idx = htole64(hidx + 1);
774 a = le64toh(o->entry_array.next_entry_array_offset);
785 r = journal_file_append_object(f, OBJECT_ENTRY_ARRAY,
786 offsetof(Object, entry_array.items) + n * sizeof(uint64_t),
791 o->entry_array.items[i] = htole64(p);
796 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, ap, &o);
800 o->entry_array.next_entry_array_offset = htole64(q);
803 *idx = htole64(hidx + 1);
808 static int link_entry_into_array_plus_one(JournalFile *f,
827 i = le64toh(*idx) - 1;
828 r = link_entry_into_array(f, first, &i, p);
833 *idx = htole64(le64toh(*idx) + 1);
837 static int journal_file_link_entry_item(JournalFile *f, Object *o, uint64_t offset, uint64_t i) {
844 p = le64toh(o->entry.items[i].object_offset);
848 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
852 return link_entry_into_array_plus_one(f,
853 &o->data.entry_offset,
854 &o->data.entry_array_offset,
859 static int journal_file_link_entry(JournalFile *f, Object *o, uint64_t offset) {
866 assert(o->object.type == OBJECT_ENTRY);
868 /* Link up the entry itself */
869 r = link_entry_into_array(f,
870 &f->header->entry_array_offset,
871 &f->header->n_entries,
876 log_error("=> %s seqnr=%lu n_entries=%lu", f->path, (unsigned long) o->entry.seqnum, (unsigned long) f->header->n_entries);
878 if (f->header->head_entry_realtime == 0)
879 f->header->head_entry_realtime = o->entry.realtime;
881 f->header->tail_entry_realtime = o->entry.realtime;
882 f->header->tail_entry_monotonic = o->entry.monotonic;
884 f->tail_entry_monotonic_valid = true;
886 /* Link up the items */
887 n = journal_file_entry_n_items(o);
888 for (i = 0; i < n; i++) {
889 r = journal_file_link_entry_item(f, o, offset, i);
897 static int journal_file_append_entry_internal(
899 const dual_timestamp *ts,
901 const EntryItem items[], unsigned n_items,
903 Object **ret, uint64_t *offset) {
910 assert(items || n_items == 0);
913 osize = offsetof(Object, entry.items) + (n_items * sizeof(EntryItem));
915 r = journal_file_append_object(f, OBJECT_ENTRY, osize, &o, &np);
919 o->entry.seqnum = htole64(journal_file_seqnum(f, seqnum));
920 memcpy(o->entry.items, items, n_items * sizeof(EntryItem));
921 o->entry.realtime = htole64(ts->realtime);
922 o->entry.monotonic = htole64(ts->monotonic);
923 o->entry.xor_hash = htole64(xor_hash);
924 o->entry.boot_id = f->header->boot_id;
926 r = journal_file_link_entry(f, o, np);
939 static void journal_file_post_change(JournalFile *f) {
942 /* inotify() does not receive IN_MODIFY events from file
943 * accesses done via mmap(). After each access we hence
944 * trigger IN_MODIFY by truncating the journal file to its
945 * current size which triggers IN_MODIFY. */
947 __sync_synchronize();
949 if (ftruncate(f->fd, f->last_stat.st_size) < 0)
950 log_error("Failed to to truncate file to its own size: %m");
953 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) {
957 uint64_t xor_hash = 0;
958 struct dual_timestamp _ts;
961 assert(iovec || n_iovec == 0);
967 dual_timestamp_get(&_ts);
971 if (f->tail_entry_monotonic_valid &&
972 ts->monotonic < le64toh(f->header->tail_entry_monotonic))
975 if (ts->realtime < le64toh(f->header->tail_entry_realtime))
978 items = new(EntryItem, n_iovec);
982 for (i = 0; i < n_iovec; i++) {
986 r = journal_file_append_data(f, iovec[i].iov_base, iovec[i].iov_len, &o, &p);
990 xor_hash ^= le64toh(o->data.hash);
991 items[i].object_offset = htole64(p);
992 items[i].hash = o->data.hash;
995 r = journal_file_append_entry_internal(f, ts, xor_hash, items, n_iovec, seqnum, ret, offset);
997 journal_file_post_change(f);
1005 static int generic_array_get(JournalFile *f,
1008 Object **ret, uint64_t *offset) {
1020 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
1024 n = journal_file_entry_array_n_items(o);
1026 p = le64toh(o->entry_array.items[i]);
1031 a = le64toh(o->entry_array.next_entry_array_offset);
1034 if (a <= 0 || p <= 0)
1037 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1050 static int generic_array_get_plus_one(JournalFile *f,
1054 Object **ret, uint64_t *offset) {
1063 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
1076 return generic_array_get(f, first, i-1, ret, offset);
1085 static int generic_array_bisect(JournalFile *f,
1089 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
1090 direction_t direction,
1095 uint64_t a, p, t = 0, i = 0, last_p = 0;
1096 bool subtract_one = false;
1097 Object *o, *array = NULL;
1101 assert(test_object);
1105 uint64_t left, right, k, lp;
1107 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &array);
1111 k = journal_file_entry_array_n_items(array);
1117 lp = p = le64toh(array->entry_array.items[i]);
1121 r = test_object(f, p, needle);
1125 if (r == TEST_FOUND)
1126 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
1128 if (r == TEST_RIGHT) {
1132 if (left == right) {
1133 if (direction == DIRECTION_UP)
1134 subtract_one = true;
1140 assert(left < right);
1142 i = (left + right) / 2;
1143 p = le64toh(array->entry_array.items[i]);
1147 r = test_object(f, p, needle);
1151 if (r == TEST_FOUND)
1152 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
1154 if (r == TEST_RIGHT)
1168 a = le64toh(array->entry_array.next_entry_array_offset);
1174 if (subtract_one && t == 0 && i == 0)
1177 if (subtract_one && i == 0)
1179 else if (subtract_one)
1180 p = le64toh(array->entry_array.items[i-1]);
1182 p = le64toh(array->entry_array.items[i]);
1184 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1195 *idx = t + i - (subtract_one ? 1 : 0);
1200 static int generic_array_bisect_plus_one(JournalFile *f,
1205 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
1206 direction_t direction,
1214 assert(test_object);
1219 /* This bisects the array in object 'first', but first checks
1222 r = test_object(f, extra, needle);
1225 else if (r == TEST_FOUND) {
1228 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
1237 } else if (r == TEST_RIGHT)
1240 r = generic_array_bisect(f, first, n-1, needle, test_object, direction, ret, offset, idx);
1248 static int test_object_seqnum(JournalFile *f, uint64_t p, uint64_t needle) {
1255 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1259 if (le64toh(o->entry.seqnum) == needle)
1261 else if (le64toh(o->entry.seqnum) < needle)
1267 int journal_file_move_to_entry_by_seqnum(
1270 direction_t direction,
1274 return generic_array_bisect(f,
1275 le64toh(f->header->entry_array_offset),
1276 le64toh(f->header->n_entries),
1283 static int test_object_realtime(JournalFile *f, uint64_t p, uint64_t needle) {
1290 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1294 if (le64toh(o->entry.realtime) == needle)
1296 else if (le64toh(o->entry.realtime) < needle)
1302 int journal_file_move_to_entry_by_realtime(
1305 direction_t direction,
1309 return generic_array_bisect(f,
1310 le64toh(f->header->entry_array_offset),
1311 le64toh(f->header->n_entries),
1313 test_object_realtime,
1318 static int test_object_monotonic(JournalFile *f, uint64_t p, uint64_t needle) {
1325 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1329 if (le64toh(o->entry.monotonic) == needle)
1331 else if (le64toh(o->entry.monotonic) < needle)
1337 int journal_file_move_to_entry_by_monotonic(
1341 direction_t direction,
1345 char t[8+32+1] = "_BOOT_ID=";
1349 sd_id128_to_string(boot_id, t + 8);
1351 r = journal_file_find_data_object(f, t, strlen(t), &o, NULL);
1357 return generic_array_bisect_plus_one(f,
1358 le64toh(o->data.entry_offset),
1359 le64toh(o->data.entry_array_offset),
1360 le64toh(o->data.n_entries),
1362 test_object_monotonic,
1367 static int test_object_offset(JournalFile *f, uint64_t p, uint64_t needle) {
1373 else if (p < needle)
1379 int journal_file_next_entry(
1381 Object *o, uint64_t p,
1382 direction_t direction,
1383 Object **ret, uint64_t *offset) {
1389 assert(p > 0 || !o);
1391 n = le64toh(f->header->n_entries);
1396 i = direction == DIRECTION_DOWN ? 0 : n - 1;
1398 if (o->object.type != OBJECT_ENTRY)
1401 r = generic_array_bisect(f,
1402 le64toh(f->header->entry_array_offset),
1403 le64toh(f->header->n_entries),
1412 if (direction == DIRECTION_DOWN) {
1425 /* And jump to it */
1426 return generic_array_get(f,
1427 le64toh(f->header->entry_array_offset),
1432 int journal_file_skip_entry(
1434 Object *o, uint64_t p,
1436 Object **ret, uint64_t *offset) {
1445 if (o->object.type != OBJECT_ENTRY)
1448 r = generic_array_bisect(f,
1449 le64toh(f->header->entry_array_offset),
1450 le64toh(f->header->n_entries),
1459 /* Calculate new index */
1461 if ((uint64_t) -skip >= i)
1464 i = i - (uint64_t) -skip;
1466 i += (uint64_t) skip;
1468 n = le64toh(f->header->n_entries);
1475 return generic_array_get(f,
1476 le64toh(f->header->entry_array_offset),
1481 int journal_file_next_entry_for_data(
1483 Object *o, uint64_t p,
1484 uint64_t data_offset,
1485 direction_t direction,
1486 Object **ret, uint64_t *offset) {
1493 assert(p > 0 || !o);
1495 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
1499 n = le64toh(d->data.n_entries);
1504 i = direction == DIRECTION_DOWN ? 0 : n - 1;
1506 if (o->object.type != OBJECT_ENTRY)
1509 r = generic_array_bisect_plus_one(f,
1510 le64toh(d->data.entry_offset),
1511 le64toh(d->data.entry_array_offset),
1512 le64toh(d->data.n_entries),
1522 if (direction == DIRECTION_DOWN) {
1536 return generic_array_get_plus_one(f,
1537 le64toh(d->data.entry_offset),
1538 le64toh(d->data.entry_array_offset),
1543 int journal_file_move_to_entry_by_seqnum_for_data(
1545 uint64_t data_offset,
1547 direction_t direction,
1548 Object **ret, uint64_t *offset) {
1553 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
1557 return generic_array_bisect_plus_one(f,
1558 le64toh(d->data.entry_offset),
1559 le64toh(d->data.entry_array_offset),
1560 le64toh(d->data.n_entries),
1567 int journal_file_move_to_entry_by_realtime_for_data(
1569 uint64_t data_offset,
1571 direction_t direction,
1572 Object **ret, uint64_t *offset) {
1577 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
1581 return generic_array_bisect_plus_one(f,
1582 le64toh(d->data.entry_offset),
1583 le64toh(d->data.entry_array_offset),
1584 le64toh(d->data.n_entries),
1586 test_object_realtime,
1591 void journal_file_dump(JournalFile *f) {
1592 char a[33], b[33], c[33];
1599 printf("File Path: %s\n"
1603 "Arena size: %llu\n"
1607 sd_id128_to_string(f->header->file_id, a),
1608 sd_id128_to_string(f->header->machine_id, b),
1609 sd_id128_to_string(f->header->boot_id, c),
1610 (unsigned long long) le64toh(f->header->arena_size),
1611 (unsigned long) le64toh(f->header->n_objects),
1612 (unsigned long) le64toh(f->header->n_entries));
1614 p = le64toh(f->header->arena_offset);
1616 r = journal_file_move_to_object(f, -1, p, &o);
1620 switch (o->object.type) {
1623 printf("Type: OBJECT_UNUSED\n");
1627 printf("Type: OBJECT_DATA\n");
1631 printf("Type: OBJECT_ENTRY %llu %llu %llu\n",
1632 (unsigned long long) le64toh(o->entry.seqnum),
1633 (unsigned long long) le64toh(o->entry.monotonic),
1634 (unsigned long long) le64toh(o->entry.realtime));
1637 case OBJECT_FIELD_HASH_TABLE:
1638 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
1641 case OBJECT_DATA_HASH_TABLE:
1642 printf("Type: OBJECT_DATA_HASH_TABLE\n");
1645 case OBJECT_ENTRY_ARRAY:
1646 printf("Type: OBJECT_ENTRY_ARRAY\n");
1650 if (o->object.flags & OBJECT_COMPRESSED)
1651 printf("Flags: COMPRESSED\n");
1653 if (p == le64toh(f->header->tail_object_offset))
1656 p = p + ALIGN64(le64toh(o->object.size));
1661 log_error("File corrupt");
1664 int journal_file_open(
1668 JournalFile *template,
1669 JournalFile **ret) {
1673 bool newly_created = false;
1677 if ((flags & O_ACCMODE) != O_RDONLY &&
1678 (flags & O_ACCMODE) != O_RDWR)
1681 f = new0(JournalFile, 1);
1688 f->writable = (flags & O_ACCMODE) != O_RDONLY;
1689 f->prot = prot_from_flags(flags);
1691 f->metrics.max_size = DEFAULT_MAX_SIZE;
1692 f->metrics.min_size = DEFAULT_MIN_SIZE;
1693 f->metrics.keep_free = DEFAULT_KEEP_FREE;
1695 f->path = strdup(fname);
1701 f->fd = open(f->path, f->flags|O_CLOEXEC, f->mode);
1707 if (fstat(f->fd, &f->last_stat) < 0) {
1712 if (f->last_stat.st_size == 0 && f->writable) {
1713 newly_created = true;
1715 r = journal_file_init_header(f, template);
1719 if (fstat(f->fd, &f->last_stat) < 0) {
1725 if (f->last_stat.st_size < (off_t) sizeof(Header)) {
1730 f->header = mmap(NULL, PAGE_ALIGN(sizeof(Header)), prot_from_flags(flags), MAP_SHARED, f->fd, 0);
1731 if (f->header == MAP_FAILED) {
1737 if (!newly_created) {
1738 r = journal_file_verify_header(f);
1744 r = journal_file_refresh_header(f);
1749 if (newly_created) {
1751 r = journal_file_setup_field_hash_table(f);
1755 r = journal_file_setup_data_hash_table(f);
1760 r = journal_file_map_field_hash_table(f);
1764 r = journal_file_map_data_hash_table(f);
1774 journal_file_close(f);
1779 int journal_file_rotate(JournalFile **f) {
1782 JournalFile *old_file, *new_file = NULL;
1790 if (!old_file->writable)
1793 if (!endswith(old_file->path, ".journal"))
1796 l = strlen(old_file->path);
1798 p = new(char, l + 1 + 16 + 1 + 32 + 1 + 16 + 1);
1802 memcpy(p, old_file->path, l - 8);
1804 sd_id128_to_string(old_file->header->seqnum_id, p + l - 8 + 1);
1805 snprintf(p + l - 8 + 1 + 32, 1 + 16 + 1 + 16 + 8 + 1,
1806 "-%016llx-%016llx.journal",
1807 (unsigned long long) le64toh((*f)->header->seqnum),
1808 (unsigned long long) le64toh((*f)->header->tail_entry_realtime));
1810 r = rename(old_file->path, p);
1816 old_file->header->state = le32toh(STATE_ARCHIVED);
1818 r = journal_file_open(old_file->path, old_file->flags, old_file->mode, old_file, &new_file);
1819 journal_file_close(old_file);
1825 struct vacuum_info {
1830 sd_id128_t seqnum_id;
1834 static int vacuum_compare(const void *_a, const void *_b) {
1835 const struct vacuum_info *a, *b;
1840 if (sd_id128_equal(a->seqnum_id, b->seqnum_id)) {
1841 if (a->seqnum < b->seqnum)
1843 else if (a->seqnum > b->seqnum)
1849 if (a->realtime < b->realtime)
1851 else if (a->realtime > b->realtime)
1854 return memcmp(&a->seqnum_id, &b->seqnum_id, 16);
1857 int journal_directory_vacuum(const char *directory, uint64_t max_use, uint64_t min_free) {
1860 struct vacuum_info *list = NULL;
1861 unsigned n_list = 0, n_allocated = 0, i;
1867 max_use = DEFAULT_MAX_USE;
1869 d = opendir(directory);
1875 struct dirent buf, *de;
1879 unsigned long long seqnum, realtime;
1880 sd_id128_t seqnum_id;
1882 k = readdir_r(d, &buf, &de);
1891 if (!dirent_is_file_with_suffix(de, ".journal"))
1894 q = strlen(de->d_name);
1896 if (q < 1 + 32 + 1 + 16 + 1 + 16 + 8)
1899 if (de->d_name[q-8-16-1] != '-' ||
1900 de->d_name[q-8-16-1-16-1] != '-' ||
1901 de->d_name[q-8-16-1-16-1-32-1] != '@')
1904 if (fstatat(dirfd(d), de->d_name, &st, AT_SYMLINK_NOFOLLOW) < 0)
1907 if (!S_ISREG(st.st_mode))
1910 p = strdup(de->d_name);
1916 de->d_name[q-8-16-1-16-1] = 0;
1917 if (sd_id128_from_string(de->d_name + q-8-16-1-16-1-32, &seqnum_id) < 0) {
1922 if (sscanf(de->d_name + q-8-16-1-16, "%16llx-%16llx.journal", &seqnum, &realtime) != 2) {
1927 if (n_list >= n_allocated) {
1928 struct vacuum_info *j;
1930 n_allocated = MAX(n_allocated * 2U, 8U);
1931 j = realloc(list, n_allocated * sizeof(struct vacuum_info));
1941 list[n_list].filename = p;
1942 list[n_list].usage = (uint64_t) st.st_blksize * (uint64_t) st.st_blocks;
1943 list[n_list].seqnum = seqnum;
1944 list[n_list].realtime = realtime;
1945 list[n_list].seqnum_id = seqnum_id;
1947 sum += list[n_list].usage;
1952 qsort(list, n_list, sizeof(struct vacuum_info), vacuum_compare);
1954 for(i = 0; i < n_list; i++) {
1957 if (fstatvfs(dirfd(d), &ss) < 0) {
1962 if (sum <= max_use &&
1963 (uint64_t) ss.f_bavail * (uint64_t) ss.f_bsize >= min_free)
1966 if (unlinkat(dirfd(d), list[i].filename, 0) >= 0) {
1967 log_debug("Deleted archived journal %s/%s.", directory, list[i].filename);
1968 sum -= list[i].usage;
1969 } else if (errno != ENOENT)
1970 log_warning("Failed to delete %s/%s: %m", directory, list[i].filename);
1974 for (i = 0; i < n_list; i++)
1975 free(list[i].filename);