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 /* This is the minimum journal file size */
43 #define JOURNAL_FILE_SIZE_MIN (64ULL*1024ULL)
45 /* These are the lower and upper bounds if we deduce the max_use value
46 * from the file system size */
47 #define DEFAULT_MAX_USE_LOWER (1ULL*1024ULL*1024ULL) /* 1 MiB */
48 #define DEFAULT_MAX_USE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
50 /* This is the upper bound if we deduce max_size from max_use */
51 #define DEFAULT_MAX_SIZE_UPPER (16ULL*1024ULL*1024ULL) /* 16 MiB */
53 /* This is the upper bound if we deduce the keep_free value from the
55 #define DEFAULT_KEEP_FREE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
57 /* This is the keep_free value when we can't determine the system
59 #define DEFAULT_KEEP_FREE (1024ULL*1024ULL) /* 1 MB */
61 static const char signature[] = { 'L', 'P', 'K', 'S', 'H', 'H', 'R', 'H' };
63 #define ALIGN64(x) (((x) + 7ULL) & ~7ULL)
65 void journal_file_close(JournalFile *f) {
70 if (f->header && f->writable)
71 f->header->state = STATE_OFFLINE;
74 for (t = 0; t < _WINDOW_MAX; t++)
75 if (f->windows[t].ptr)
76 munmap(f->windows[t].ptr, f->windows[t].size);
79 close_nointr_nofail(f->fd);
84 free(f->compress_buffer);
90 static int journal_file_init_header(JournalFile *f, JournalFile *template) {
98 memcpy(h.signature, signature, 8);
99 h.arena_offset = htole64(ALIGN64(sizeof(h)));
101 r = sd_id128_randomize(&h.file_id);
106 h.seqnum_id = template->header->seqnum_id;
107 h.seqnum = template->header->seqnum;
109 h.seqnum_id = h.file_id;
111 k = pwrite(f->fd, &h, sizeof(h), 0);
121 static int journal_file_refresh_header(JournalFile *f) {
127 r = sd_id128_get_machine(&f->header->machine_id);
131 r = sd_id128_get_boot(&boot_id);
135 if (sd_id128_equal(boot_id, f->header->boot_id))
136 f->tail_entry_monotonic_valid = true;
138 f->header->boot_id = boot_id;
140 f->header->state = STATE_ONLINE;
142 __sync_synchronize();
147 static int journal_file_verify_header(JournalFile *f) {
150 if (memcmp(f->header, signature, 8))
154 if ((le64toh(f->header->incompatible_flags) & ~HEADER_INCOMPATIBLE_COMPRESSED) != 0)
155 return -EPROTONOSUPPORT;
157 if (f->header->incompatible_flags != 0)
158 return -EPROTONOSUPPORT;
161 if ((uint64_t) f->last_stat.st_size < (le64toh(f->header->arena_offset) + le64toh(f->header->arena_size)))
166 sd_id128_t machine_id;
169 r = sd_id128_get_machine(&machine_id);
173 if (!sd_id128_equal(machine_id, f->header->machine_id))
176 state = f->header->state;
178 if (state == STATE_ONLINE)
179 log_debug("Journal file %s is already online. Assuming unclean closing. Ignoring.", f->path);
180 else if (state == STATE_ARCHIVED)
182 else if (state != STATE_OFFLINE)
183 log_debug("Journal file %s has unknown state %u. Ignoring.", f->path, state);
189 static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size) {
190 uint64_t old_size, new_size;
194 /* We assume that this file is not sparse, and we know that
195 * for sure, since we always call posix_fallocate()
199 le64toh(f->header->arena_offset) +
200 le64toh(f->header->arena_size);
202 new_size = PAGE_ALIGN(offset + size);
203 if (new_size < le64toh(f->header->arena_offset))
204 new_size = le64toh(f->header->arena_offset);
206 if (new_size <= old_size)
209 if (f->metrics.max_size > 0 &&
210 new_size > f->metrics.max_size)
213 if (new_size > f->metrics.min_size &&
214 f->metrics.keep_free > 0) {
217 if (fstatvfs(f->fd, &svfs) >= 0) {
220 available = svfs.f_bfree * svfs.f_bsize;
222 if (available >= f->metrics.keep_free)
223 available -= f->metrics.keep_free;
227 if (new_size - old_size > available)
232 /* Note that the glibc fallocate() fallback is very
233 inefficient, hence we try to minimize the allocation area
235 if (posix_fallocate(f->fd, old_size, new_size - old_size) < 0)
238 if (fstat(f->fd, &f->last_stat) < 0)
241 f->header->arena_size = new_size - htole64(f->header->arena_offset);
246 static int journal_file_map(
255 uint64_t woffset, wsize;
262 woffset = offset & ~((uint64_t) page_size() - 1ULL);
263 wsize = size + (offset - woffset);
264 wsize = PAGE_ALIGN(wsize);
266 /* Avoid SIGBUS on invalid accesses */
267 if (woffset + wsize > (uint64_t) PAGE_ALIGN(f->last_stat.st_size))
268 return -EADDRNOTAVAIL;
270 window = mmap(NULL, wsize, f->prot, MAP_SHARED, f->fd, woffset);
271 if (window == MAP_FAILED)
283 *ret = (uint8_t*) window + (offset - woffset);
288 static int journal_file_move_to(JournalFile *f, int wt, uint64_t offset, uint64_t size, void **ret) {
297 assert(wt < _WINDOW_MAX);
301 if (_likely_(w->ptr &&
302 w->offset <= offset &&
303 w->offset + w->size >= offset + size)) {
305 *ret = (uint8_t*) w->ptr + (offset - w->offset);
310 if (munmap(w->ptr, w->size) < 0)
314 w->size = w->offset = 0;
317 if (size < DEFAULT_WINDOW_SIZE) {
318 /* If the default window size is larger then what was
319 * asked for extend the mapping a bit in the hope to
320 * minimize needed remappings later on. We add half
321 * the window space before and half behind the
322 * requested mapping */
324 delta = PAGE_ALIGN((DEFAULT_WINDOW_SIZE - size) / 2);
330 size += (DEFAULT_WINDOW_SIZE - delta);
334 if (offset > (uint64_t) f->last_stat.st_size)
335 return -EADDRNOTAVAIL;
337 if (offset + size > (uint64_t) f->last_stat.st_size)
338 size = PAGE_ALIGN((uint64_t) f->last_stat.st_size - offset);
341 return -EADDRNOTAVAIL;
343 r = journal_file_map(f,
345 &w->ptr, &w->offset, &w->size,
351 *ret = (uint8_t*) p + delta;
355 static bool verify_hash(Object *o) {
360 if (o->object.type == OBJECT_DATA && !(o->object.flags & OBJECT_COMPRESSED)) {
361 h1 = le64toh(o->data.hash);
362 h2 = hash64(o->data.payload, le64toh(o->object.size) - offsetof(Object, data.payload));
363 } else if (o->object.type == OBJECT_FIELD) {
364 h1 = le64toh(o->field.hash);
365 h2 = hash64(o->field.payload, le64toh(o->object.size) - offsetof(Object, field.payload));
372 int journal_file_move_to_object(JournalFile *f, int type, uint64_t offset, Object **ret) {
380 assert(type < _OBJECT_TYPE_MAX);
382 r = journal_file_move_to(f, type >= 0 ? type : WINDOW_UNKNOWN, offset, sizeof(ObjectHeader), &t);
387 s = le64toh(o->object.size);
389 if (s < sizeof(ObjectHeader))
392 if (type >= 0 && o->object.type != type)
395 if (s > sizeof(ObjectHeader)) {
396 r = journal_file_move_to(f, o->object.type, offset, s, &t);
410 static uint64_t journal_file_seqnum(JournalFile *f, uint64_t *seqnum) {
415 r = le64toh(f->header->seqnum) + 1;
418 /* If an external seqnum counter was passed, we update
419 * both the local and the external one, and set it to
420 * the maximum of both */
428 f->header->seqnum = htole64(r);
430 if (f->header->first_seqnum == 0)
431 f->header->first_seqnum = htole64(r);
436 static int journal_file_append_object(JournalFile *f, int type, uint64_t size, Object **ret, uint64_t *offset) {
443 assert(size >= sizeof(ObjectHeader));
447 p = le64toh(f->header->tail_object_offset);
449 p = le64toh(f->header->arena_offset);
451 r = journal_file_move_to_object(f, -1, p, &tail);
455 p += ALIGN64(le64toh(tail->object.size));
458 r = journal_file_allocate(f, p, size);
462 r = journal_file_move_to(f, type, p, size, &t);
469 o->object.type = type;
470 o->object.size = htole64(size);
472 f->header->tail_object_offset = htole64(p);
473 f->header->n_objects = htole64(le64toh(f->header->n_objects) + 1);
481 static int journal_file_setup_data_hash_table(JournalFile *f) {
488 s = DEFAULT_DATA_HASH_TABLE_SIZE;
489 r = journal_file_append_object(f,
490 OBJECT_DATA_HASH_TABLE,
491 offsetof(Object, hash_table.items) + s,
496 memset(o->hash_table.items, 0, s);
498 f->header->data_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
499 f->header->data_hash_table_size = htole64(s);
504 static int journal_file_setup_field_hash_table(JournalFile *f) {
511 s = DEFAULT_FIELD_HASH_TABLE_SIZE;
512 r = journal_file_append_object(f,
513 OBJECT_FIELD_HASH_TABLE,
514 offsetof(Object, hash_table.items) + s,
519 memset(o->hash_table.items, 0, s);
521 f->header->field_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
522 f->header->field_hash_table_size = htole64(s);
527 static int journal_file_map_data_hash_table(JournalFile *f) {
534 p = le64toh(f->header->data_hash_table_offset);
535 s = le64toh(f->header->data_hash_table_size);
537 r = journal_file_move_to(f,
538 WINDOW_DATA_HASH_TABLE,
544 f->data_hash_table = t;
548 static int journal_file_map_field_hash_table(JournalFile *f) {
555 p = le64toh(f->header->field_hash_table_offset);
556 s = le64toh(f->header->field_hash_table_size);
558 r = journal_file_move_to(f,
559 WINDOW_FIELD_HASH_TABLE,
565 f->field_hash_table = t;
569 static int journal_file_link_data(JournalFile *f, Object *o, uint64_t offset, uint64_t hash) {
576 assert(o->object.type == OBJECT_DATA);
578 o->data.next_hash_offset = o->data.next_field_offset = 0;
579 o->data.entry_offset = o->data.entry_array_offset = 0;
580 o->data.n_entries = 0;
582 h = hash % (le64toh(f->header->data_hash_table_size) / sizeof(HashItem));
583 p = le64toh(f->data_hash_table[h].head_hash_offset);
585 /* Only entry in the hash table is easy */
586 f->data_hash_table[h].head_hash_offset = htole64(offset);
588 /* Temporarily move back to the previous data object,
589 * to patch in pointer */
591 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
595 o->data.next_hash_offset = htole64(offset);
597 r = journal_file_move_to_object(f, OBJECT_DATA, offset, &o);
602 f->data_hash_table[h].tail_hash_offset = htole64(offset);
607 int journal_file_find_data_object_with_hash(
609 const void *data, uint64_t size, uint64_t hash,
610 Object **ret, uint64_t *offset) {
611 uint64_t p, osize, h;
615 assert(data || size == 0);
617 osize = offsetof(Object, data.payload) + size;
619 if (f->header->data_hash_table_size == 0)
622 h = hash % (le64toh(f->header->data_hash_table_size) / sizeof(HashItem));
623 p = le64toh(f->data_hash_table[h].head_hash_offset);
628 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
632 if (le64toh(o->data.hash) != hash)
635 if (o->object.flags & OBJECT_COMPRESSED) {
639 l = le64toh(o->object.size);
640 if (l <= offsetof(Object, data.payload))
643 l -= offsetof(Object, data.payload);
645 if (!uncompress_blob(o->data.payload, l, &f->compress_buffer, &f->compress_buffer_size, &rsize))
649 memcmp(f->compress_buffer, data, size) == 0) {
660 return -EPROTONOSUPPORT;
663 } else if (le64toh(o->object.size) == osize &&
664 memcmp(o->data.payload, data, size) == 0) {
676 p = le64toh(o->data.next_hash_offset);
682 int journal_file_find_data_object(
684 const void *data, uint64_t size,
685 Object **ret, uint64_t *offset) {
690 assert(data || size == 0);
692 hash = hash64(data, size);
694 return journal_file_find_data_object_with_hash(f,
699 static int journal_file_append_data(JournalFile *f, const void *data, uint64_t size, Object **ret, uint64_t *offset) {
704 bool compressed = false;
707 assert(data || size == 0);
709 hash = hash64(data, size);
711 r = journal_file_find_data_object_with_hash(f, data, size, hash, &o, &p);
725 osize = offsetof(Object, data.payload) + size;
726 r = journal_file_append_object(f, OBJECT_DATA, osize, &o, &p);
730 o->data.hash = htole64(hash);
734 size >= COMPRESSION_SIZE_THRESHOLD) {
737 compressed = compress_blob(data, size, o->data.payload, &rsize);
740 o->object.size = htole64(offsetof(Object, data.payload) + rsize);
741 o->object.flags |= OBJECT_COMPRESSED;
743 f->header->incompatible_flags = htole32(le32toh(f->header->incompatible_flags) | HEADER_INCOMPATIBLE_COMPRESSED);
745 log_debug("Compressed data object %lu -> %lu", (unsigned long) size, (unsigned long) rsize);
751 memcpy(o->data.payload, data, size);
753 r = journal_file_link_data(f, o, p, hash);
766 uint64_t journal_file_entry_n_items(Object *o) {
768 assert(o->object.type == htole64(OBJECT_ENTRY));
770 return (le64toh(o->object.size) - offsetof(Object, entry.items)) / sizeof(EntryItem);
773 static uint64_t journal_file_entry_array_n_items(Object *o) {
775 assert(o->object.type == htole64(OBJECT_ENTRY_ARRAY));
777 return (le64toh(o->object.size) - offsetof(Object, entry_array.items)) / sizeof(uint64_t);
780 static int link_entry_into_array(JournalFile *f,
785 uint64_t n = 0, ap = 0, q, i, a, hidx;
794 i = hidx = le64toh(*idx);
797 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
801 n = journal_file_entry_array_n_items(o);
803 o->entry_array.items[i] = htole64(p);
804 *idx = htole64(hidx + 1);
810 a = le64toh(o->entry_array.next_entry_array_offset);
821 r = journal_file_append_object(f, OBJECT_ENTRY_ARRAY,
822 offsetof(Object, entry_array.items) + n * sizeof(uint64_t),
827 o->entry_array.items[i] = htole64(p);
832 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, ap, &o);
836 o->entry_array.next_entry_array_offset = htole64(q);
839 *idx = htole64(hidx + 1);
844 static int link_entry_into_array_plus_one(JournalFile *f,
863 i = le64toh(*idx) - 1;
864 r = link_entry_into_array(f, first, &i, p);
869 *idx = htole64(le64toh(*idx) + 1);
873 static int journal_file_link_entry_item(JournalFile *f, Object *o, uint64_t offset, uint64_t i) {
880 p = le64toh(o->entry.items[i].object_offset);
884 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
888 return link_entry_into_array_plus_one(f,
889 &o->data.entry_offset,
890 &o->data.entry_array_offset,
895 static int journal_file_link_entry(JournalFile *f, Object *o, uint64_t offset) {
902 assert(o->object.type == OBJECT_ENTRY);
904 __sync_synchronize();
906 /* Link up the entry itself */
907 r = link_entry_into_array(f,
908 &f->header->entry_array_offset,
909 &f->header->n_entries,
914 log_error("=> %s seqnr=%lu n_entries=%lu", f->path, (unsigned long) o->entry.seqnum, (unsigned long) f->header->n_entries);
916 if (f->header->head_entry_realtime == 0)
917 f->header->head_entry_realtime = o->entry.realtime;
919 f->header->tail_entry_realtime = o->entry.realtime;
920 f->header->tail_entry_monotonic = o->entry.monotonic;
922 f->tail_entry_monotonic_valid = true;
924 /* Link up the items */
925 n = journal_file_entry_n_items(o);
926 for (i = 0; i < n; i++) {
927 r = journal_file_link_entry_item(f, o, offset, i);
935 static int journal_file_append_entry_internal(
937 const dual_timestamp *ts,
939 const EntryItem items[], unsigned n_items,
941 Object **ret, uint64_t *offset) {
948 assert(items || n_items == 0);
951 osize = offsetof(Object, entry.items) + (n_items * sizeof(EntryItem));
953 r = journal_file_append_object(f, OBJECT_ENTRY, osize, &o, &np);
957 o->entry.seqnum = htole64(journal_file_seqnum(f, seqnum));
958 memcpy(o->entry.items, items, n_items * sizeof(EntryItem));
959 o->entry.realtime = htole64(ts->realtime);
960 o->entry.monotonic = htole64(ts->monotonic);
961 o->entry.xor_hash = htole64(xor_hash);
962 o->entry.boot_id = f->header->boot_id;
964 r = journal_file_link_entry(f, o, np);
977 void journal_file_post_change(JournalFile *f) {
980 /* inotify() does not receive IN_MODIFY events from file
981 * accesses done via mmap(). After each access we hence
982 * trigger IN_MODIFY by truncating the journal file to its
983 * current size which triggers IN_MODIFY. */
985 __sync_synchronize();
987 if (ftruncate(f->fd, f->last_stat.st_size) < 0)
988 log_error("Failed to to truncate file to its own size: %m");
991 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) {
995 uint64_t xor_hash = 0;
996 struct dual_timestamp _ts;
999 assert(iovec || n_iovec == 0);
1005 dual_timestamp_get(&_ts);
1009 if (f->tail_entry_monotonic_valid &&
1010 ts->monotonic < le64toh(f->header->tail_entry_monotonic))
1013 if (ts->realtime < le64toh(f->header->tail_entry_realtime))
1016 items = alloca(sizeof(EntryItem) * n_iovec);
1018 for (i = 0; i < n_iovec; i++) {
1022 r = journal_file_append_data(f, iovec[i].iov_base, iovec[i].iov_len, &o, &p);
1026 xor_hash ^= le64toh(o->data.hash);
1027 items[i].object_offset = htole64(p);
1028 items[i].hash = o->data.hash;
1031 r = journal_file_append_entry_internal(f, ts, xor_hash, items, n_iovec, seqnum, ret, offset);
1033 journal_file_post_change(f);
1038 static int generic_array_get(JournalFile *f,
1041 Object **ret, uint64_t *offset) {
1053 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
1057 n = journal_file_entry_array_n_items(o);
1059 p = le64toh(o->entry_array.items[i]);
1064 a = le64toh(o->entry_array.next_entry_array_offset);
1067 if (a <= 0 || p <= 0)
1070 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1083 static int generic_array_get_plus_one(JournalFile *f,
1087 Object **ret, uint64_t *offset) {
1096 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
1109 return generic_array_get(f, first, i-1, ret, offset);
1118 static int generic_array_bisect(JournalFile *f,
1122 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
1123 direction_t direction,
1128 uint64_t a, p, t = 0, i = 0, last_p = 0;
1129 bool subtract_one = false;
1130 Object *o, *array = NULL;
1134 assert(test_object);
1138 uint64_t left, right, k, lp;
1140 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &array);
1144 k = journal_file_entry_array_n_items(array);
1150 lp = p = le64toh(array->entry_array.items[i]);
1154 r = test_object(f, p, needle);
1158 if (r == TEST_FOUND)
1159 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
1161 if (r == TEST_RIGHT) {
1165 if (left == right) {
1166 if (direction == DIRECTION_UP)
1167 subtract_one = true;
1173 assert(left < right);
1175 i = (left + right) / 2;
1176 p = le64toh(array->entry_array.items[i]);
1180 r = test_object(f, p, needle);
1184 if (r == TEST_FOUND)
1185 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
1187 if (r == TEST_RIGHT)
1201 a = le64toh(array->entry_array.next_entry_array_offset);
1207 if (subtract_one && t == 0 && i == 0)
1210 if (subtract_one && i == 0)
1212 else if (subtract_one)
1213 p = le64toh(array->entry_array.items[i-1]);
1215 p = le64toh(array->entry_array.items[i]);
1217 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1228 *idx = t + i - (subtract_one ? 1 : 0);
1233 static int generic_array_bisect_plus_one(JournalFile *f,
1238 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
1239 direction_t direction,
1247 assert(test_object);
1252 /* This bisects the array in object 'first', but first checks
1254 r = test_object(f, extra, needle);
1257 else if (r == TEST_FOUND) {
1260 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
1274 } else if (r == TEST_RIGHT)
1277 r = generic_array_bisect(f, first, n-1, needle, test_object, direction, ret, offset, idx);
1285 static int test_object_seqnum(JournalFile *f, uint64_t p, uint64_t needle) {
1292 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1296 if (le64toh(o->entry.seqnum) == needle)
1298 else if (le64toh(o->entry.seqnum) < needle)
1304 int journal_file_move_to_entry_by_seqnum(
1307 direction_t direction,
1311 return generic_array_bisect(f,
1312 le64toh(f->header->entry_array_offset),
1313 le64toh(f->header->n_entries),
1320 static int test_object_realtime(JournalFile *f, uint64_t p, uint64_t needle) {
1327 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1331 if (le64toh(o->entry.realtime) == needle)
1333 else if (le64toh(o->entry.realtime) < needle)
1339 int journal_file_move_to_entry_by_realtime(
1342 direction_t direction,
1346 return generic_array_bisect(f,
1347 le64toh(f->header->entry_array_offset),
1348 le64toh(f->header->n_entries),
1350 test_object_realtime,
1355 static int test_object_monotonic(JournalFile *f, uint64_t p, uint64_t needle) {
1362 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1366 if (le64toh(o->entry.monotonic) == needle)
1368 else if (le64toh(o->entry.monotonic) < needle)
1374 int journal_file_move_to_entry_by_monotonic(
1378 direction_t direction,
1382 char t[8+32+1] = "_BOOT_ID=";
1386 sd_id128_to_string(boot_id, t + 8);
1388 r = journal_file_find_data_object(f, t, strlen(t), &o, NULL);
1394 return generic_array_bisect_plus_one(f,
1395 le64toh(o->data.entry_offset),
1396 le64toh(o->data.entry_array_offset),
1397 le64toh(o->data.n_entries),
1399 test_object_monotonic,
1404 static int test_object_offset(JournalFile *f, uint64_t p, uint64_t needle) {
1410 else if (p < needle)
1416 int journal_file_next_entry(
1418 Object *o, uint64_t p,
1419 direction_t direction,
1420 Object **ret, uint64_t *offset) {
1426 assert(p > 0 || !o);
1428 n = le64toh(f->header->n_entries);
1433 i = direction == DIRECTION_DOWN ? 0 : n - 1;
1435 if (o->object.type != OBJECT_ENTRY)
1438 r = generic_array_bisect(f,
1439 le64toh(f->header->entry_array_offset),
1440 le64toh(f->header->n_entries),
1449 if (direction == DIRECTION_DOWN) {
1462 /* And jump to it */
1463 return generic_array_get(f,
1464 le64toh(f->header->entry_array_offset),
1469 int journal_file_skip_entry(
1471 Object *o, uint64_t p,
1473 Object **ret, uint64_t *offset) {
1482 if (o->object.type != OBJECT_ENTRY)
1485 r = generic_array_bisect(f,
1486 le64toh(f->header->entry_array_offset),
1487 le64toh(f->header->n_entries),
1496 /* Calculate new index */
1498 if ((uint64_t) -skip >= i)
1501 i = i - (uint64_t) -skip;
1503 i += (uint64_t) skip;
1505 n = le64toh(f->header->n_entries);
1512 return generic_array_get(f,
1513 le64toh(f->header->entry_array_offset),
1518 int journal_file_next_entry_for_data(
1520 Object *o, uint64_t p,
1521 uint64_t data_offset,
1522 direction_t direction,
1523 Object **ret, uint64_t *offset) {
1530 assert(p > 0 || !o);
1532 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
1536 n = le64toh(d->data.n_entries);
1541 i = direction == DIRECTION_DOWN ? 0 : n - 1;
1543 if (o->object.type != OBJECT_ENTRY)
1546 r = generic_array_bisect_plus_one(f,
1547 le64toh(d->data.entry_offset),
1548 le64toh(d->data.entry_array_offset),
1549 le64toh(d->data.n_entries),
1559 if (direction == DIRECTION_DOWN) {
1573 return generic_array_get_plus_one(f,
1574 le64toh(d->data.entry_offset),
1575 le64toh(d->data.entry_array_offset),
1580 int journal_file_move_to_entry_by_seqnum_for_data(
1582 uint64_t data_offset,
1584 direction_t direction,
1585 Object **ret, uint64_t *offset) {
1590 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
1594 return generic_array_bisect_plus_one(f,
1595 le64toh(d->data.entry_offset),
1596 le64toh(d->data.entry_array_offset),
1597 le64toh(d->data.n_entries),
1604 int journal_file_move_to_entry_by_realtime_for_data(
1606 uint64_t data_offset,
1608 direction_t direction,
1609 Object **ret, uint64_t *offset) {
1614 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
1618 return generic_array_bisect_plus_one(f,
1619 le64toh(d->data.entry_offset),
1620 le64toh(d->data.entry_array_offset),
1621 le64toh(d->data.n_entries),
1623 test_object_realtime,
1628 void journal_file_dump(JournalFile *f) {
1629 char a[33], b[33], c[33];
1636 printf("File Path: %s\n"
1640 "Arena size: %llu\n"
1644 sd_id128_to_string(f->header->file_id, a),
1645 sd_id128_to_string(f->header->machine_id, b),
1646 sd_id128_to_string(f->header->boot_id, c),
1647 (unsigned long long) le64toh(f->header->arena_size),
1648 (unsigned long) le64toh(f->header->n_objects),
1649 (unsigned long) le64toh(f->header->n_entries));
1651 p = le64toh(f->header->arena_offset);
1653 r = journal_file_move_to_object(f, -1, p, &o);
1657 switch (o->object.type) {
1660 printf("Type: OBJECT_UNUSED\n");
1664 printf("Type: OBJECT_DATA\n");
1668 printf("Type: OBJECT_ENTRY %llu %llu %llu\n",
1669 (unsigned long long) le64toh(o->entry.seqnum),
1670 (unsigned long long) le64toh(o->entry.monotonic),
1671 (unsigned long long) le64toh(o->entry.realtime));
1674 case OBJECT_FIELD_HASH_TABLE:
1675 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
1678 case OBJECT_DATA_HASH_TABLE:
1679 printf("Type: OBJECT_DATA_HASH_TABLE\n");
1682 case OBJECT_ENTRY_ARRAY:
1683 printf("Type: OBJECT_ENTRY_ARRAY\n");
1687 if (o->object.flags & OBJECT_COMPRESSED)
1688 printf("Flags: COMPRESSED\n");
1690 if (p == le64toh(f->header->tail_object_offset))
1693 p = p + ALIGN64(le64toh(o->object.size));
1698 log_error("File corrupt");
1701 int journal_file_open(
1705 JournalFile *template,
1706 JournalFile **ret) {
1710 bool newly_created = false;
1714 if ((flags & O_ACCMODE) != O_RDONLY &&
1715 (flags & O_ACCMODE) != O_RDWR)
1718 f = new0(JournalFile, 1);
1725 f->writable = (flags & O_ACCMODE) != O_RDONLY;
1726 f->prot = prot_from_flags(flags);
1728 f->path = strdup(fname);
1734 f->fd = open(f->path, f->flags|O_CLOEXEC, f->mode);
1740 if (fstat(f->fd, &f->last_stat) < 0) {
1745 if (f->last_stat.st_size == 0 && f->writable) {
1746 newly_created = true;
1748 r = journal_file_init_header(f, template);
1752 if (fstat(f->fd, &f->last_stat) < 0) {
1758 if (f->last_stat.st_size < (off_t) sizeof(Header)) {
1763 f->header = mmap(NULL, PAGE_ALIGN(sizeof(Header)), prot_from_flags(flags), MAP_SHARED, f->fd, 0);
1764 if (f->header == MAP_FAILED) {
1770 if (!newly_created) {
1771 r = journal_file_verify_header(f);
1777 r = journal_file_refresh_header(f);
1782 if (newly_created) {
1784 r = journal_file_setup_field_hash_table(f);
1788 r = journal_file_setup_data_hash_table(f);
1793 r = journal_file_map_field_hash_table(f);
1797 r = journal_file_map_data_hash_table(f);
1807 journal_file_close(f);
1812 int journal_file_rotate(JournalFile **f) {
1815 JournalFile *old_file, *new_file = NULL;
1823 if (!old_file->writable)
1826 if (!endswith(old_file->path, ".journal"))
1829 l = strlen(old_file->path);
1831 p = new(char, l + 1 + 16 + 1 + 32 + 1 + 16 + 1);
1835 memcpy(p, old_file->path, l - 8);
1837 sd_id128_to_string(old_file->header->seqnum_id, p + l - 8 + 1);
1838 snprintf(p + l - 8 + 1 + 32, 1 + 16 + 1 + 16 + 8 + 1,
1839 "-%016llx-%016llx.journal",
1840 (unsigned long long) le64toh((*f)->header->seqnum),
1841 (unsigned long long) le64toh((*f)->header->tail_entry_realtime));
1843 r = rename(old_file->path, p);
1849 old_file->header->state = le32toh(STATE_ARCHIVED);
1851 r = journal_file_open(old_file->path, old_file->flags, old_file->mode, old_file, &new_file);
1852 journal_file_close(old_file);
1858 struct vacuum_info {
1863 sd_id128_t seqnum_id;
1867 static int vacuum_compare(const void *_a, const void *_b) {
1868 const struct vacuum_info *a, *b;
1873 if (sd_id128_equal(a->seqnum_id, b->seqnum_id)) {
1874 if (a->seqnum < b->seqnum)
1876 else if (a->seqnum > b->seqnum)
1882 if (a->realtime < b->realtime)
1884 else if (a->realtime > b->realtime)
1887 return memcmp(&a->seqnum_id, &b->seqnum_id, 16);
1890 int journal_directory_vacuum(const char *directory, uint64_t max_use, uint64_t min_free) {
1893 struct vacuum_info *list = NULL;
1894 unsigned n_list = 0, n_allocated = 0, i;
1902 d = opendir(directory);
1908 struct dirent buf, *de;
1912 unsigned long long seqnum, realtime;
1913 sd_id128_t seqnum_id;
1915 k = readdir_r(d, &buf, &de);
1924 if (!dirent_is_file_with_suffix(de, ".journal"))
1927 q = strlen(de->d_name);
1929 if (q < 1 + 32 + 1 + 16 + 1 + 16 + 8)
1932 if (de->d_name[q-8-16-1] != '-' ||
1933 de->d_name[q-8-16-1-16-1] != '-' ||
1934 de->d_name[q-8-16-1-16-1-32-1] != '@')
1937 if (fstatat(dirfd(d), de->d_name, &st, AT_SYMLINK_NOFOLLOW) < 0)
1940 if (!S_ISREG(st.st_mode))
1943 p = strdup(de->d_name);
1949 de->d_name[q-8-16-1-16-1] = 0;
1950 if (sd_id128_from_string(de->d_name + q-8-16-1-16-1-32, &seqnum_id) < 0) {
1955 if (sscanf(de->d_name + q-8-16-1-16, "%16llx-%16llx.journal", &seqnum, &realtime) != 2) {
1960 if (n_list >= n_allocated) {
1961 struct vacuum_info *j;
1963 n_allocated = MAX(n_allocated * 2U, 8U);
1964 j = realloc(list, n_allocated * sizeof(struct vacuum_info));
1974 list[n_list].filename = p;
1975 list[n_list].usage = (uint64_t) st.st_blksize * (uint64_t) st.st_blocks;
1976 list[n_list].seqnum = seqnum;
1977 list[n_list].realtime = realtime;
1978 list[n_list].seqnum_id = seqnum_id;
1980 sum += list[n_list].usage;
1985 qsort(list, n_list, sizeof(struct vacuum_info), vacuum_compare);
1987 for(i = 0; i < n_list; i++) {
1990 if (fstatvfs(dirfd(d), &ss) < 0) {
1995 if (sum <= max_use &&
1996 (uint64_t) ss.f_bavail * (uint64_t) ss.f_bsize >= min_free)
1999 if (unlinkat(dirfd(d), list[i].filename, 0) >= 0) {
2000 log_debug("Deleted archived journal %s/%s.", directory, list[i].filename);
2001 sum -= list[i].usage;
2002 } else if (errno != ENOENT)
2003 log_warning("Failed to delete %s/%s: %m", directory, list[i].filename);
2007 for (i = 0; i < n_list; i++)
2008 free(list[i].filename);
2018 int journal_file_copy_entry(JournalFile *from, JournalFile *to, Object *o, uint64_t p, uint64_t *seqnum, Object **ret, uint64_t *offset) {
2020 uint64_t q, xor_hash = 0;
2033 ts.monotonic = le64toh(o->entry.monotonic);
2034 ts.realtime = le64toh(o->entry.realtime);
2036 if (to->tail_entry_monotonic_valid &&
2037 ts.monotonic < le64toh(to->header->tail_entry_monotonic))
2040 if (ts.realtime < le64toh(to->header->tail_entry_realtime))
2043 n = journal_file_entry_n_items(o);
2044 items = alloca(sizeof(EntryItem) * n);
2046 for (i = 0; i < n; i++) {
2047 uint64_t le_hash, l, h;
2052 q = le64toh(o->entry.items[i].object_offset);
2053 le_hash = o->entry.items[i].hash;
2055 r = journal_file_move_to_object(from, OBJECT_DATA, q, &o);
2059 if (le_hash != o->data.hash)
2062 l = le64toh(o->object.size) - offsetof(Object, data.payload);
2065 /* We hit the limit on 32bit machines */
2066 if ((uint64_t) t != l)
2069 if (o->object.flags & OBJECT_COMPRESSED) {
2073 if (!uncompress_blob(o->data.payload, l, &from->compress_buffer, &from->compress_buffer_size, &rsize))
2076 data = from->compress_buffer;
2079 return -EPROTONOSUPPORT;
2082 data = o->data.payload;
2084 r = journal_file_append_data(to, data, l, &u, &h);
2088 xor_hash ^= le64toh(u->data.hash);
2089 items[i].object_offset = htole64(h);
2090 items[i].hash = u->data.hash;
2092 r = journal_file_move_to_object(from, OBJECT_ENTRY, p, &o);
2097 return journal_file_append_entry_internal(to, &ts, xor_hash, items, n, seqnum, ret, offset);
2100 void journal_default_metrics(JournalMetrics *m, int fd) {
2101 uint64_t fs_size = 0;
2103 char a[64], b[64], c[64], d[64];
2108 if (fstatvfs(fd, &ss) >= 0)
2109 fs_size = ss.f_frsize * ss.f_blocks;
2111 if (m->max_use == (uint64_t) -1) {
2114 m->max_use = PAGE_ALIGN(fs_size / 10); /* 10% of file system size */
2116 if (m->max_use > DEFAULT_MAX_USE_UPPER)
2117 m->max_use = DEFAULT_MAX_USE_UPPER;
2119 if (m->max_use < DEFAULT_MAX_USE_LOWER)
2120 m->max_use = DEFAULT_MAX_USE_LOWER;
2122 m->max_use = DEFAULT_MAX_USE_LOWER;
2124 m->max_use = PAGE_ALIGN(m->max_use);
2126 if (m->max_use < JOURNAL_FILE_SIZE_MIN*2)
2127 m->max_use = JOURNAL_FILE_SIZE_MIN*2;
2130 if (m->max_size == (uint64_t) -1) {
2131 m->max_size = PAGE_ALIGN(m->max_use / 8); /* 8 chunks */
2133 if (m->max_size > DEFAULT_MAX_SIZE_UPPER)
2134 m->max_size = DEFAULT_MAX_SIZE_UPPER;
2136 m->max_size = PAGE_ALIGN(m->max_size);
2138 if (m->max_size < JOURNAL_FILE_SIZE_MIN)
2139 m->max_size = JOURNAL_FILE_SIZE_MIN;
2141 if (m->max_size*2 > m->max_use)
2142 m->max_use = m->max_size*2;
2144 if (m->min_size == (uint64_t) -1)
2145 m->min_size = JOURNAL_FILE_SIZE_MIN;
2147 m->min_size = PAGE_ALIGN(m->min_size);
2149 if (m->min_size < JOURNAL_FILE_SIZE_MIN)
2150 m->min_size = JOURNAL_FILE_SIZE_MIN;
2152 if (m->min_size > m->max_size)
2153 m->max_size = m->min_size;
2156 if (m->keep_free == (uint64_t) -1) {
2159 m->keep_free = PAGE_ALIGN(fs_size / 20); /* 5% of file system size */
2161 if (m->keep_free > DEFAULT_KEEP_FREE_UPPER)
2162 m->keep_free = DEFAULT_KEEP_FREE_UPPER;
2165 m->keep_free = DEFAULT_KEEP_FREE;
2168 log_debug("Fixed max_use=%s max_size=%s min_size=%s keep_free=%s",
2169 format_bytes(a, sizeof(a), m->max_use),
2170 format_bytes(b, sizeof(b), m->max_size),
2171 format_bytes(c, sizeof(c), m->min_size),
2172 format_bytes(d, sizeof(d), m->keep_free));