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);
299 if (offset + size > (uint64_t) f->last_stat.st_size) {
300 /* Hmm, out of range? Let's refresh the fstat() data
301 * first, before we trust that check. */
303 if (fstat(f->fd, &f->last_stat) < 0 ||
304 offset + size > (uint64_t) f->last_stat.st_size)
305 return -EADDRNOTAVAIL;
310 if (_likely_(w->ptr &&
311 w->offset <= offset &&
312 w->offset + w->size >= offset + size)) {
314 *ret = (uint8_t*) w->ptr + (offset - w->offset);
319 if (munmap(w->ptr, w->size) < 0)
323 w->size = w->offset = 0;
326 if (size < DEFAULT_WINDOW_SIZE) {
327 /* If the default window size is larger then what was
328 * asked for extend the mapping a bit in the hope to
329 * minimize needed remappings later on. We add half
330 * the window space before and half behind the
331 * requested mapping */
333 delta = (DEFAULT_WINDOW_SIZE - size) / 2;
339 size = DEFAULT_WINDOW_SIZE;
343 if (offset + size > (uint64_t) f->last_stat.st_size)
344 size = (uint64_t) f->last_stat.st_size - offset;
347 return -EADDRNOTAVAIL;
349 r = journal_file_map(f,
351 &w->ptr, &w->offset, &w->size,
357 *ret = (uint8_t*) p + delta;
361 static bool verify_hash(Object *o) {
366 if (o->object.type == OBJECT_DATA && !(o->object.flags & OBJECT_COMPRESSED)) {
367 h1 = le64toh(o->data.hash);
368 h2 = hash64(o->data.payload, le64toh(o->object.size) - offsetof(Object, data.payload));
369 } else if (o->object.type == OBJECT_FIELD) {
370 h1 = le64toh(o->field.hash);
371 h2 = hash64(o->field.payload, le64toh(o->object.size) - offsetof(Object, field.payload));
378 int journal_file_move_to_object(JournalFile *f, int type, uint64_t offset, Object **ret) {
386 assert(type < _OBJECT_TYPE_MAX);
388 r = journal_file_move_to(f, type >= 0 ? type : WINDOW_UNKNOWN, offset, sizeof(ObjectHeader), &t);
393 s = le64toh(o->object.size);
395 if (s < sizeof(ObjectHeader))
398 if (type >= 0 && o->object.type != type)
401 if (s > sizeof(ObjectHeader)) {
402 r = journal_file_move_to(f, o->object.type, offset, s, &t);
416 static uint64_t journal_file_seqnum(JournalFile *f, uint64_t *seqnum) {
421 r = le64toh(f->header->seqnum) + 1;
424 /* If an external seqnum counter was passed, we update
425 * both the local and the external one, and set it to
426 * the maximum of both */
434 f->header->seqnum = htole64(r);
436 if (f->header->first_seqnum == 0)
437 f->header->first_seqnum = htole64(r);
442 static int journal_file_append_object(JournalFile *f, int type, uint64_t size, Object **ret, uint64_t *offset) {
449 assert(size >= sizeof(ObjectHeader));
453 p = le64toh(f->header->tail_object_offset);
455 p = le64toh(f->header->arena_offset);
457 r = journal_file_move_to_object(f, -1, p, &tail);
461 p += ALIGN64(le64toh(tail->object.size));
464 r = journal_file_allocate(f, p, size);
468 r = journal_file_move_to(f, type, p, size, &t);
475 o->object.type = type;
476 o->object.size = htole64(size);
478 f->header->tail_object_offset = htole64(p);
479 f->header->n_objects = htole64(le64toh(f->header->n_objects) + 1);
487 static int journal_file_setup_data_hash_table(JournalFile *f) {
494 s = DEFAULT_DATA_HASH_TABLE_SIZE;
495 r = journal_file_append_object(f,
496 OBJECT_DATA_HASH_TABLE,
497 offsetof(Object, hash_table.items) + s,
502 memset(o->hash_table.items, 0, s);
504 f->header->data_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
505 f->header->data_hash_table_size = htole64(s);
510 static int journal_file_setup_field_hash_table(JournalFile *f) {
517 s = DEFAULT_FIELD_HASH_TABLE_SIZE;
518 r = journal_file_append_object(f,
519 OBJECT_FIELD_HASH_TABLE,
520 offsetof(Object, hash_table.items) + s,
525 memset(o->hash_table.items, 0, s);
527 f->header->field_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
528 f->header->field_hash_table_size = htole64(s);
533 static int journal_file_map_data_hash_table(JournalFile *f) {
540 p = le64toh(f->header->data_hash_table_offset);
541 s = le64toh(f->header->data_hash_table_size);
543 r = journal_file_move_to(f,
544 WINDOW_DATA_HASH_TABLE,
550 f->data_hash_table = t;
554 static int journal_file_map_field_hash_table(JournalFile *f) {
561 p = le64toh(f->header->field_hash_table_offset);
562 s = le64toh(f->header->field_hash_table_size);
564 r = journal_file_move_to(f,
565 WINDOW_FIELD_HASH_TABLE,
571 f->field_hash_table = t;
575 static int journal_file_link_data(JournalFile *f, Object *o, uint64_t offset, uint64_t hash) {
582 assert(o->object.type == OBJECT_DATA);
584 o->data.next_hash_offset = o->data.next_field_offset = 0;
585 o->data.entry_offset = o->data.entry_array_offset = 0;
586 o->data.n_entries = 0;
588 h = hash % (le64toh(f->header->data_hash_table_size) / sizeof(HashItem));
589 p = le64toh(f->data_hash_table[h].head_hash_offset);
591 /* Only entry in the hash table is easy */
592 f->data_hash_table[h].head_hash_offset = htole64(offset);
594 /* Temporarily move back to the previous data object,
595 * to patch in pointer */
597 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
601 o->data.next_hash_offset = htole64(offset);
603 r = journal_file_move_to_object(f, OBJECT_DATA, offset, &o);
608 f->data_hash_table[h].tail_hash_offset = htole64(offset);
613 int journal_file_find_data_object_with_hash(
615 const void *data, uint64_t size, uint64_t hash,
616 Object **ret, uint64_t *offset) {
617 uint64_t p, osize, h;
621 assert(data || size == 0);
623 osize = offsetof(Object, data.payload) + size;
625 if (f->header->data_hash_table_size == 0)
628 h = hash % (le64toh(f->header->data_hash_table_size) / sizeof(HashItem));
629 p = le64toh(f->data_hash_table[h].head_hash_offset);
634 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
638 if (le64toh(o->data.hash) != hash)
641 if (o->object.flags & OBJECT_COMPRESSED) {
646 l = le64toh(o->object.size);
647 if (l <= offsetof(Object, data.payload))
650 l -= offsetof(Object, data.payload);
652 if (!uncompress_blob(o->data.payload, l, &f->compress_buffer, &f->compress_buffer_size, &rsize))
655 if ((uint64_t) rsize == size &&
656 memcmp(f->compress_buffer, data, size) == 0) {
667 return -EPROTONOSUPPORT;
670 } else if (le64toh(o->object.size) == osize &&
671 memcmp(o->data.payload, data, size) == 0) {
683 p = le64toh(o->data.next_hash_offset);
689 int journal_file_find_data_object(
691 const void *data, uint64_t size,
692 Object **ret, uint64_t *offset) {
697 assert(data || size == 0);
699 hash = hash64(data, size);
701 return journal_file_find_data_object_with_hash(f,
706 static int journal_file_append_data(JournalFile *f, const void *data, uint64_t size, Object **ret, uint64_t *offset) {
711 bool compressed = false;
714 assert(data || size == 0);
716 hash = hash64(data, size);
718 r = journal_file_find_data_object_with_hash(f, data, size, hash, &o, &p);
732 osize = offsetof(Object, data.payload) + size;
733 r = journal_file_append_object(f, OBJECT_DATA, osize, &o, &p);
737 o->data.hash = htole64(hash);
741 size >= COMPRESSION_SIZE_THRESHOLD) {
744 compressed = compress_blob(data, size, o->data.payload, &rsize);
747 o->object.size = htole64(offsetof(Object, data.payload) + rsize);
748 o->object.flags |= OBJECT_COMPRESSED;
750 f->header->incompatible_flags = htole32(le32toh(f->header->incompatible_flags) | HEADER_INCOMPATIBLE_COMPRESSED);
752 log_debug("Compressed data object %lu -> %lu", (unsigned long) size, (unsigned long) rsize);
758 memcpy(o->data.payload, data, size);
760 r = journal_file_link_data(f, o, p, hash);
773 uint64_t journal_file_entry_n_items(Object *o) {
775 assert(o->object.type == htole64(OBJECT_ENTRY));
777 return (le64toh(o->object.size) - offsetof(Object, entry.items)) / sizeof(EntryItem);
780 static uint64_t journal_file_entry_array_n_items(Object *o) {
782 assert(o->object.type == htole64(OBJECT_ENTRY_ARRAY));
784 return (le64toh(o->object.size) - offsetof(Object, entry_array.items)) / sizeof(uint64_t);
787 static int link_entry_into_array(JournalFile *f,
792 uint64_t n = 0, ap = 0, q, i, a, hidx;
801 i = hidx = le64toh(*idx);
804 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
808 n = journal_file_entry_array_n_items(o);
810 o->entry_array.items[i] = htole64(p);
811 *idx = htole64(hidx + 1);
817 a = le64toh(o->entry_array.next_entry_array_offset);
828 r = journal_file_append_object(f, OBJECT_ENTRY_ARRAY,
829 offsetof(Object, entry_array.items) + n * sizeof(uint64_t),
834 o->entry_array.items[i] = htole64(p);
839 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, ap, &o);
843 o->entry_array.next_entry_array_offset = htole64(q);
846 *idx = htole64(hidx + 1);
851 static int link_entry_into_array_plus_one(JournalFile *f,
870 i = le64toh(*idx) - 1;
871 r = link_entry_into_array(f, first, &i, p);
876 *idx = htole64(le64toh(*idx) + 1);
880 static int journal_file_link_entry_item(JournalFile *f, Object *o, uint64_t offset, uint64_t i) {
887 p = le64toh(o->entry.items[i].object_offset);
891 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
895 return link_entry_into_array_plus_one(f,
896 &o->data.entry_offset,
897 &o->data.entry_array_offset,
902 static int journal_file_link_entry(JournalFile *f, Object *o, uint64_t offset) {
909 assert(o->object.type == OBJECT_ENTRY);
911 __sync_synchronize();
913 /* Link up the entry itself */
914 r = link_entry_into_array(f,
915 &f->header->entry_array_offset,
916 &f->header->n_entries,
921 /* log_debug("=> %s seqnr=%lu n_entries=%lu", f->path, (unsigned long) o->entry.seqnum, (unsigned long) f->header->n_entries); */
923 if (f->header->head_entry_realtime == 0)
924 f->header->head_entry_realtime = o->entry.realtime;
926 f->header->tail_entry_realtime = o->entry.realtime;
927 f->header->tail_entry_monotonic = o->entry.monotonic;
929 f->tail_entry_monotonic_valid = true;
931 /* Link up the items */
932 n = journal_file_entry_n_items(o);
933 for (i = 0; i < n; i++) {
934 r = journal_file_link_entry_item(f, o, offset, i);
942 static int journal_file_append_entry_internal(
944 const dual_timestamp *ts,
946 const EntryItem items[], unsigned n_items,
948 Object **ret, uint64_t *offset) {
955 assert(items || n_items == 0);
958 osize = offsetof(Object, entry.items) + (n_items * sizeof(EntryItem));
960 r = journal_file_append_object(f, OBJECT_ENTRY, osize, &o, &np);
964 o->entry.seqnum = htole64(journal_file_seqnum(f, seqnum));
965 memcpy(o->entry.items, items, n_items * sizeof(EntryItem));
966 o->entry.realtime = htole64(ts->realtime);
967 o->entry.monotonic = htole64(ts->monotonic);
968 o->entry.xor_hash = htole64(xor_hash);
969 o->entry.boot_id = f->header->boot_id;
971 r = journal_file_link_entry(f, o, np);
984 void journal_file_post_change(JournalFile *f) {
987 /* inotify() does not receive IN_MODIFY events from file
988 * accesses done via mmap(). After each access we hence
989 * trigger IN_MODIFY by truncating the journal file to its
990 * current size which triggers IN_MODIFY. */
992 __sync_synchronize();
994 if (ftruncate(f->fd, f->last_stat.st_size) < 0)
995 log_error("Failed to to truncate file to its own size: %m");
998 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) {
1002 uint64_t xor_hash = 0;
1003 struct dual_timestamp _ts;
1006 assert(iovec || n_iovec == 0);
1012 dual_timestamp_get(&_ts);
1016 if (f->tail_entry_monotonic_valid &&
1017 ts->monotonic < le64toh(f->header->tail_entry_monotonic))
1020 items = alloca(sizeof(EntryItem) * n_iovec);
1022 for (i = 0; i < n_iovec; i++) {
1026 r = journal_file_append_data(f, iovec[i].iov_base, iovec[i].iov_len, &o, &p);
1030 xor_hash ^= le64toh(o->data.hash);
1031 items[i].object_offset = htole64(p);
1032 items[i].hash = o->data.hash;
1035 r = journal_file_append_entry_internal(f, ts, xor_hash, items, n_iovec, seqnum, ret, offset);
1037 journal_file_post_change(f);
1042 static int generic_array_get(JournalFile *f,
1045 Object **ret, uint64_t *offset) {
1057 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
1061 n = journal_file_entry_array_n_items(o);
1063 p = le64toh(o->entry_array.items[i]);
1068 a = le64toh(o->entry_array.next_entry_array_offset);
1071 if (a <= 0 || p <= 0)
1074 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1087 static int generic_array_get_plus_one(JournalFile *f,
1091 Object **ret, uint64_t *offset) {
1100 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
1113 return generic_array_get(f, first, i-1, ret, offset);
1122 static int generic_array_bisect(JournalFile *f,
1126 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
1127 direction_t direction,
1132 uint64_t a, p, t = 0, i = 0, last_p = 0;
1133 bool subtract_one = false;
1134 Object *o, *array = NULL;
1138 assert(test_object);
1142 uint64_t left, right, k, lp;
1144 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &array);
1148 k = journal_file_entry_array_n_items(array);
1154 lp = p = le64toh(array->entry_array.items[i]);
1158 r = test_object(f, p, needle);
1162 if (r == TEST_FOUND)
1163 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
1165 if (r == TEST_RIGHT) {
1169 if (left == right) {
1170 if (direction == DIRECTION_UP)
1171 subtract_one = true;
1177 assert(left < right);
1179 i = (left + right) / 2;
1180 p = le64toh(array->entry_array.items[i]);
1184 r = test_object(f, p, needle);
1188 if (r == TEST_FOUND)
1189 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
1191 if (r == TEST_RIGHT)
1205 a = le64toh(array->entry_array.next_entry_array_offset);
1211 if (subtract_one && t == 0 && i == 0)
1214 if (subtract_one && i == 0)
1216 else if (subtract_one)
1217 p = le64toh(array->entry_array.items[i-1]);
1219 p = le64toh(array->entry_array.items[i]);
1221 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1232 *idx = t + i - (subtract_one ? 1 : 0);
1237 static int generic_array_bisect_plus_one(JournalFile *f,
1242 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
1243 direction_t direction,
1251 assert(test_object);
1256 /* This bisects the array in object 'first', but first checks
1258 r = test_object(f, extra, needle);
1261 else if (r == TEST_FOUND) {
1264 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
1278 } else if (r == TEST_RIGHT)
1281 r = generic_array_bisect(f, first, n-1, needle, test_object, direction, ret, offset, idx);
1289 static int test_object_seqnum(JournalFile *f, uint64_t p, uint64_t needle) {
1296 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1300 if (le64toh(o->entry.seqnum) == needle)
1302 else if (le64toh(o->entry.seqnum) < needle)
1308 int journal_file_move_to_entry_by_seqnum(
1311 direction_t direction,
1315 return generic_array_bisect(f,
1316 le64toh(f->header->entry_array_offset),
1317 le64toh(f->header->n_entries),
1324 static int test_object_realtime(JournalFile *f, uint64_t p, uint64_t needle) {
1331 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1335 if (le64toh(o->entry.realtime) == needle)
1337 else if (le64toh(o->entry.realtime) < needle)
1343 int journal_file_move_to_entry_by_realtime(
1346 direction_t direction,
1350 return generic_array_bisect(f,
1351 le64toh(f->header->entry_array_offset),
1352 le64toh(f->header->n_entries),
1354 test_object_realtime,
1359 static int test_object_monotonic(JournalFile *f, uint64_t p, uint64_t needle) {
1366 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1370 if (le64toh(o->entry.monotonic) == needle)
1372 else if (le64toh(o->entry.monotonic) < needle)
1378 int journal_file_move_to_entry_by_monotonic(
1382 direction_t direction,
1386 char t[8+32+1] = "_BOOT_ID=";
1390 sd_id128_to_string(boot_id, t + 8);
1392 r = journal_file_find_data_object(f, t, strlen(t), &o, NULL);
1398 return generic_array_bisect_plus_one(f,
1399 le64toh(o->data.entry_offset),
1400 le64toh(o->data.entry_array_offset),
1401 le64toh(o->data.n_entries),
1403 test_object_monotonic,
1408 static int test_object_offset(JournalFile *f, uint64_t p, uint64_t needle) {
1414 else if (p < needle)
1420 int journal_file_next_entry(
1422 Object *o, uint64_t p,
1423 direction_t direction,
1424 Object **ret, uint64_t *offset) {
1430 assert(p > 0 || !o);
1432 n = le64toh(f->header->n_entries);
1437 i = direction == DIRECTION_DOWN ? 0 : n - 1;
1439 if (o->object.type != OBJECT_ENTRY)
1442 r = generic_array_bisect(f,
1443 le64toh(f->header->entry_array_offset),
1444 le64toh(f->header->n_entries),
1453 if (direction == DIRECTION_DOWN) {
1466 /* And jump to it */
1467 return generic_array_get(f,
1468 le64toh(f->header->entry_array_offset),
1473 int journal_file_skip_entry(
1475 Object *o, uint64_t p,
1477 Object **ret, uint64_t *offset) {
1486 if (o->object.type != OBJECT_ENTRY)
1489 r = generic_array_bisect(f,
1490 le64toh(f->header->entry_array_offset),
1491 le64toh(f->header->n_entries),
1500 /* Calculate new index */
1502 if ((uint64_t) -skip >= i)
1505 i = i - (uint64_t) -skip;
1507 i += (uint64_t) skip;
1509 n = le64toh(f->header->n_entries);
1516 return generic_array_get(f,
1517 le64toh(f->header->entry_array_offset),
1522 int journal_file_next_entry_for_data(
1524 Object *o, uint64_t p,
1525 uint64_t data_offset,
1526 direction_t direction,
1527 Object **ret, uint64_t *offset) {
1534 assert(p > 0 || !o);
1536 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
1540 n = le64toh(d->data.n_entries);
1545 i = direction == DIRECTION_DOWN ? 0 : n - 1;
1547 if (o->object.type != OBJECT_ENTRY)
1550 r = generic_array_bisect_plus_one(f,
1551 le64toh(d->data.entry_offset),
1552 le64toh(d->data.entry_array_offset),
1553 le64toh(d->data.n_entries),
1563 if (direction == DIRECTION_DOWN) {
1577 return generic_array_get_plus_one(f,
1578 le64toh(d->data.entry_offset),
1579 le64toh(d->data.entry_array_offset),
1584 int journal_file_move_to_entry_by_seqnum_for_data(
1586 uint64_t data_offset,
1588 direction_t direction,
1589 Object **ret, uint64_t *offset) {
1594 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
1598 return generic_array_bisect_plus_one(f,
1599 le64toh(d->data.entry_offset),
1600 le64toh(d->data.entry_array_offset),
1601 le64toh(d->data.n_entries),
1608 int journal_file_move_to_entry_by_realtime_for_data(
1610 uint64_t data_offset,
1612 direction_t direction,
1613 Object **ret, uint64_t *offset) {
1618 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
1622 return generic_array_bisect_plus_one(f,
1623 le64toh(d->data.entry_offset),
1624 le64toh(d->data.entry_array_offset),
1625 le64toh(d->data.n_entries),
1627 test_object_realtime,
1632 void journal_file_dump(JournalFile *f) {
1633 char a[33], b[33], c[33];
1640 printf("File Path: %s\n"
1644 "Arena size: %llu\n"
1648 sd_id128_to_string(f->header->file_id, a),
1649 sd_id128_to_string(f->header->machine_id, b),
1650 sd_id128_to_string(f->header->boot_id, c),
1651 (unsigned long long) le64toh(f->header->arena_size),
1652 (unsigned long) le64toh(f->header->n_objects),
1653 (unsigned long) le64toh(f->header->n_entries));
1655 p = le64toh(f->header->arena_offset);
1657 r = journal_file_move_to_object(f, -1, p, &o);
1661 switch (o->object.type) {
1664 printf("Type: OBJECT_UNUSED\n");
1668 printf("Type: OBJECT_DATA\n");
1672 printf("Type: OBJECT_ENTRY %llu %llu %llu\n",
1673 (unsigned long long) le64toh(o->entry.seqnum),
1674 (unsigned long long) le64toh(o->entry.monotonic),
1675 (unsigned long long) le64toh(o->entry.realtime));
1678 case OBJECT_FIELD_HASH_TABLE:
1679 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
1682 case OBJECT_DATA_HASH_TABLE:
1683 printf("Type: OBJECT_DATA_HASH_TABLE\n");
1686 case OBJECT_ENTRY_ARRAY:
1687 printf("Type: OBJECT_ENTRY_ARRAY\n");
1691 if (o->object.flags & OBJECT_COMPRESSED)
1692 printf("Flags: COMPRESSED\n");
1694 if (p == le64toh(f->header->tail_object_offset))
1697 p = p + ALIGN64(le64toh(o->object.size));
1702 log_error("File corrupt");
1705 int journal_file_open(
1709 JournalFile *template,
1710 JournalFile **ret) {
1714 bool newly_created = false;
1718 if ((flags & O_ACCMODE) != O_RDONLY &&
1719 (flags & O_ACCMODE) != O_RDWR)
1722 f = new0(JournalFile, 1);
1729 f->writable = (flags & O_ACCMODE) != O_RDONLY;
1730 f->prot = prot_from_flags(flags);
1732 f->path = strdup(fname);
1738 f->fd = open(f->path, f->flags|O_CLOEXEC, f->mode);
1744 if (fstat(f->fd, &f->last_stat) < 0) {
1749 if (f->last_stat.st_size == 0 && f->writable) {
1750 newly_created = true;
1752 r = journal_file_init_header(f, template);
1756 if (fstat(f->fd, &f->last_stat) < 0) {
1762 if (f->last_stat.st_size < (off_t) sizeof(Header)) {
1767 f->header = mmap(NULL, PAGE_ALIGN(sizeof(Header)), prot_from_flags(flags), MAP_SHARED, f->fd, 0);
1768 if (f->header == MAP_FAILED) {
1774 if (!newly_created) {
1775 r = journal_file_verify_header(f);
1781 r = journal_file_refresh_header(f);
1786 if (newly_created) {
1788 r = journal_file_setup_field_hash_table(f);
1792 r = journal_file_setup_data_hash_table(f);
1797 r = journal_file_map_field_hash_table(f);
1801 r = journal_file_map_data_hash_table(f);
1811 journal_file_close(f);
1816 int journal_file_rotate(JournalFile **f) {
1819 JournalFile *old_file, *new_file = NULL;
1827 if (!old_file->writable)
1830 if (!endswith(old_file->path, ".journal"))
1833 l = strlen(old_file->path);
1835 p = new(char, l + 1 + 16 + 1 + 32 + 1 + 16 + 1);
1839 memcpy(p, old_file->path, l - 8);
1841 sd_id128_to_string(old_file->header->seqnum_id, p + l - 8 + 1);
1842 snprintf(p + l - 8 + 1 + 32, 1 + 16 + 1 + 16 + 8 + 1,
1843 "-%016llx-%016llx.journal",
1844 (unsigned long long) le64toh((*f)->header->seqnum),
1845 (unsigned long long) le64toh((*f)->header->tail_entry_realtime));
1847 r = rename(old_file->path, p);
1853 old_file->header->state = le32toh(STATE_ARCHIVED);
1855 r = journal_file_open(old_file->path, old_file->flags, old_file->mode, old_file, &new_file);
1856 journal_file_close(old_file);
1862 struct vacuum_info {
1867 sd_id128_t seqnum_id;
1871 static int vacuum_compare(const void *_a, const void *_b) {
1872 const struct vacuum_info *a, *b;
1877 if (sd_id128_equal(a->seqnum_id, b->seqnum_id)) {
1878 if (a->seqnum < b->seqnum)
1880 else if (a->seqnum > b->seqnum)
1886 if (a->realtime < b->realtime)
1888 else if (a->realtime > b->realtime)
1891 return memcmp(&a->seqnum_id, &b->seqnum_id, 16);
1894 int journal_directory_vacuum(const char *directory, uint64_t max_use, uint64_t min_free) {
1897 struct vacuum_info *list = NULL;
1898 unsigned n_list = 0, n_allocated = 0, i;
1906 d = opendir(directory);
1912 struct dirent buf, *de;
1916 unsigned long long seqnum, realtime;
1917 sd_id128_t seqnum_id;
1919 k = readdir_r(d, &buf, &de);
1928 if (!dirent_is_file_with_suffix(de, ".journal"))
1931 q = strlen(de->d_name);
1933 if (q < 1 + 32 + 1 + 16 + 1 + 16 + 8)
1936 if (de->d_name[q-8-16-1] != '-' ||
1937 de->d_name[q-8-16-1-16-1] != '-' ||
1938 de->d_name[q-8-16-1-16-1-32-1] != '@')
1941 if (fstatat(dirfd(d), de->d_name, &st, AT_SYMLINK_NOFOLLOW) < 0)
1944 if (!S_ISREG(st.st_mode))
1947 p = strdup(de->d_name);
1953 de->d_name[q-8-16-1-16-1] = 0;
1954 if (sd_id128_from_string(de->d_name + q-8-16-1-16-1-32, &seqnum_id) < 0) {
1959 if (sscanf(de->d_name + q-8-16-1-16, "%16llx-%16llx.journal", &seqnum, &realtime) != 2) {
1964 if (n_list >= n_allocated) {
1965 struct vacuum_info *j;
1967 n_allocated = MAX(n_allocated * 2U, 8U);
1968 j = realloc(list, n_allocated * sizeof(struct vacuum_info));
1978 list[n_list].filename = p;
1979 list[n_list].usage = (uint64_t) st.st_blksize * (uint64_t) st.st_blocks;
1980 list[n_list].seqnum = seqnum;
1981 list[n_list].realtime = realtime;
1982 list[n_list].seqnum_id = seqnum_id;
1984 sum += list[n_list].usage;
1989 qsort(list, n_list, sizeof(struct vacuum_info), vacuum_compare);
1991 for(i = 0; i < n_list; i++) {
1994 if (fstatvfs(dirfd(d), &ss) < 0) {
1999 if (sum <= max_use &&
2000 (uint64_t) ss.f_bavail * (uint64_t) ss.f_bsize >= min_free)
2003 if (unlinkat(dirfd(d), list[i].filename, 0) >= 0) {
2004 log_debug("Deleted archived journal %s/%s.", directory, list[i].filename);
2005 sum -= list[i].usage;
2006 } else if (errno != ENOENT)
2007 log_warning("Failed to delete %s/%s: %m", directory, list[i].filename);
2011 for (i = 0; i < n_list; i++)
2012 free(list[i].filename);
2022 int journal_file_copy_entry(JournalFile *from, JournalFile *to, Object *o, uint64_t p, uint64_t *seqnum, Object **ret, uint64_t *offset) {
2024 uint64_t q, xor_hash = 0;
2037 ts.monotonic = le64toh(o->entry.monotonic);
2038 ts.realtime = le64toh(o->entry.realtime);
2040 if (to->tail_entry_monotonic_valid &&
2041 ts.monotonic < le64toh(to->header->tail_entry_monotonic))
2044 if (ts.realtime < le64toh(to->header->tail_entry_realtime))
2047 n = journal_file_entry_n_items(o);
2048 items = alloca(sizeof(EntryItem) * n);
2050 for (i = 0; i < n; i++) {
2051 uint64_t le_hash, l, h;
2056 q = le64toh(o->entry.items[i].object_offset);
2057 le_hash = o->entry.items[i].hash;
2059 r = journal_file_move_to_object(from, OBJECT_DATA, q, &o);
2063 if (le_hash != o->data.hash)
2066 l = le64toh(o->object.size) - offsetof(Object, data.payload);
2069 /* We hit the limit on 32bit machines */
2070 if ((uint64_t) t != l)
2073 if (o->object.flags & OBJECT_COMPRESSED) {
2077 if (!uncompress_blob(o->data.payload, l, &from->compress_buffer, &from->compress_buffer_size, &rsize))
2080 data = from->compress_buffer;
2083 return -EPROTONOSUPPORT;
2086 data = o->data.payload;
2088 r = journal_file_append_data(to, data, l, &u, &h);
2092 xor_hash ^= le64toh(u->data.hash);
2093 items[i].object_offset = htole64(h);
2094 items[i].hash = u->data.hash;
2096 r = journal_file_move_to_object(from, OBJECT_ENTRY, p, &o);
2101 return journal_file_append_entry_internal(to, &ts, xor_hash, items, n, seqnum, ret, offset);
2104 void journal_default_metrics(JournalMetrics *m, int fd) {
2105 uint64_t fs_size = 0;
2107 char a[64], b[64], c[64], d[64];
2112 if (fstatvfs(fd, &ss) >= 0)
2113 fs_size = ss.f_frsize * ss.f_blocks;
2115 if (m->max_use == (uint64_t) -1) {
2118 m->max_use = PAGE_ALIGN(fs_size / 10); /* 10% of file system size */
2120 if (m->max_use > DEFAULT_MAX_USE_UPPER)
2121 m->max_use = DEFAULT_MAX_USE_UPPER;
2123 if (m->max_use < DEFAULT_MAX_USE_LOWER)
2124 m->max_use = DEFAULT_MAX_USE_LOWER;
2126 m->max_use = DEFAULT_MAX_USE_LOWER;
2128 m->max_use = PAGE_ALIGN(m->max_use);
2130 if (m->max_use < JOURNAL_FILE_SIZE_MIN*2)
2131 m->max_use = JOURNAL_FILE_SIZE_MIN*2;
2134 if (m->max_size == (uint64_t) -1) {
2135 m->max_size = PAGE_ALIGN(m->max_use / 8); /* 8 chunks */
2137 if (m->max_size > DEFAULT_MAX_SIZE_UPPER)
2138 m->max_size = DEFAULT_MAX_SIZE_UPPER;
2140 m->max_size = PAGE_ALIGN(m->max_size);
2142 if (m->max_size < JOURNAL_FILE_SIZE_MIN)
2143 m->max_size = JOURNAL_FILE_SIZE_MIN;
2145 if (m->max_size*2 > m->max_use)
2146 m->max_use = m->max_size*2;
2148 if (m->min_size == (uint64_t) -1)
2149 m->min_size = JOURNAL_FILE_SIZE_MIN;
2151 m->min_size = PAGE_ALIGN(m->min_size);
2153 if (m->min_size < JOURNAL_FILE_SIZE_MIN)
2154 m->min_size = JOURNAL_FILE_SIZE_MIN;
2156 if (m->min_size > m->max_size)
2157 m->max_size = m->min_size;
2160 if (m->keep_free == (uint64_t) -1) {
2163 m->keep_free = PAGE_ALIGN(fs_size / 20); /* 5% of file system size */
2165 if (m->keep_free > DEFAULT_KEEP_FREE_UPPER)
2166 m->keep_free = DEFAULT_KEEP_FREE_UPPER;
2169 m->keep_free = DEFAULT_KEEP_FREE;
2172 log_debug("Fixed max_use=%s max_size=%s min_size=%s keep_free=%s",
2173 format_bytes(a, sizeof(a), m->max_use),
2174 format_bytes(b, sizeof(b), m->max_size),
2175 format_bytes(c, sizeof(c), m->min_size),
2176 format_bytes(d, sizeof(d), m->keep_free));