2 This file is part of systemd.
4 Copyright 2011 Lennart Poettering
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
26 #include <sys/statvfs.h>
30 #include "alloc-util.h"
31 #include "btrfs-util.h"
32 #include "chattr-util.h"
35 #include "journal-authenticate.h"
36 #include "journal-def.h"
37 #include "journal-file.h"
39 #include "parse-util.h"
40 #include "random-util.h"
43 #include "string-util.h"
44 #include "xattr-util.h"
46 #define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
47 #define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
49 #define COMPRESSION_SIZE_THRESHOLD (512ULL)
51 /* This is the minimum journal file size */
52 #define JOURNAL_FILE_SIZE_MIN (512ULL*1024ULL) /* 512 KiB */
54 /* These are the lower and upper bounds if we deduce the max_use value
55 * from the file system size */
56 #define DEFAULT_MAX_USE_LOWER (1ULL*1024ULL*1024ULL) /* 1 MiB */
57 #define DEFAULT_MAX_USE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
59 /* This is the default minimal use limit, how much we'll use even if keep_free suggests otherwise. */
60 #define DEFAULT_MIN_USE (1ULL*1024ULL*1024ULL) /* 1 MiB */
62 /* This is the upper bound if we deduce max_size from max_use */
63 #define DEFAULT_MAX_SIZE_UPPER (128ULL*1024ULL*1024ULL) /* 128 MiB */
65 /* This is the upper bound if we deduce the keep_free value from the
67 #define DEFAULT_KEEP_FREE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
69 /* This is the keep_free value when we can't determine the system
71 #define DEFAULT_KEEP_FREE (1024ULL*1024ULL) /* 1 MB */
73 /* This is the default maximum number of journal files to keep around. */
74 #define DEFAULT_N_MAX_FILES (100)
76 /* n_data was the first entry we added after the initial file format design */
77 #define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
79 /* How many entries to keep in the entry array chain cache at max */
80 #define CHAIN_CACHE_MAX 20
82 /* How much to increase the journal file size at once each time we allocate something new. */
83 #define FILE_SIZE_INCREASE (8ULL*1024ULL*1024ULL) /* 8MB */
85 /* Reread fstat() of the file for detecting deletions at least this often */
86 #define LAST_STAT_REFRESH_USEC (5*USEC_PER_SEC)
88 /* The mmap context to use for the header we pick as one above the last defined typed */
89 #define CONTEXT_HEADER _OBJECT_TYPE_MAX
91 /* This may be called from a separate thread to prevent blocking the caller for the duration of fsync().
92 * As a result we use atomic operations on f->offline_state for inter-thread communications with
93 * journal_file_set_offline() and journal_file_set_online(). */
94 static void journal_file_set_offline_internal(JournalFile
*f
) {
100 switch (f
->offline_state
) {
102 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_CANCEL
, OFFLINE_DONE
))
106 case OFFLINE_AGAIN_FROM_SYNCING
:
107 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_SYNCING
, OFFLINE_SYNCING
))
111 case OFFLINE_AGAIN_FROM_OFFLINING
:
112 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_OFFLINING
, OFFLINE_SYNCING
))
116 case OFFLINE_SYNCING
:
119 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_OFFLINING
))
122 f
->header
->state
= STATE_OFFLINE
;
126 case OFFLINE_OFFLINING
:
127 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_OFFLINING
, OFFLINE_DONE
))
135 log_debug("OFFLINE_JOINED unexpected offline state for journal_file_set_offline_internal()");
141 static void * journal_file_set_offline_thread(void *arg
) {
142 JournalFile
*f
= arg
;
144 journal_file_set_offline_internal(f
);
149 static int journal_file_set_offline_thread_join(JournalFile
*f
) {
154 if (f
->offline_state
== OFFLINE_JOINED
)
157 r
= pthread_join(f
->offline_thread
, NULL
);
161 f
->offline_state
= OFFLINE_JOINED
;
163 if (mmap_cache_got_sigbus(f
->mmap
, f
->fd
))
169 /* Trigger a restart if the offline thread is mid-flight in a restartable state. */
170 static bool journal_file_set_offline_try_restart(JournalFile
*f
) {
172 switch (f
->offline_state
) {
173 case OFFLINE_AGAIN_FROM_SYNCING
:
174 case OFFLINE_AGAIN_FROM_OFFLINING
:
178 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_CANCEL
, OFFLINE_AGAIN_FROM_SYNCING
))
182 case OFFLINE_SYNCING
:
183 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_AGAIN_FROM_SYNCING
))
187 case OFFLINE_OFFLINING
:
188 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_OFFLINING
, OFFLINE_AGAIN_FROM_OFFLINING
))
198 /* Sets a journal offline.
200 * If wait is false then an offline is dispatched in a separate thread for a
201 * subsequent journal_file_set_offline() or journal_file_set_online() of the
202 * same journal to synchronize with.
204 * If wait is true, then either an existing offline thread will be restarted
205 * and joined, or if none exists the offline is simply performed in this
206 * context without involving another thread.
208 int journal_file_set_offline(JournalFile
*f
, bool wait
) {
217 if (!(f
->fd
>= 0 && f
->header
))
220 if (f
->header
->state
!= STATE_ONLINE
)
223 /* Restart an in-flight offline thread and wait if needed, or join a lingering done one. */
224 restarted
= journal_file_set_offline_try_restart(f
);
225 if ((restarted
&& wait
) || !restarted
) {
226 r
= journal_file_set_offline_thread_join(f
);
234 /* Initiate a new offline. */
235 f
->offline_state
= OFFLINE_SYNCING
;
237 if (wait
) /* Without using a thread if waiting. */
238 journal_file_set_offline_internal(f
);
240 r
= pthread_create(&f
->offline_thread
, NULL
, journal_file_set_offline_thread
, f
);
242 f
->offline_state
= OFFLINE_JOINED
;
250 static int journal_file_set_online(JournalFile
*f
) {
258 if (!(f
->fd
>= 0 && f
->header
))
262 switch (f
->offline_state
) {
264 /* No offline thread, no need to wait. */
268 case OFFLINE_SYNCING
:
269 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_CANCEL
))
271 /* Canceled syncing prior to offlining, no need to wait. */
274 case OFFLINE_AGAIN_FROM_SYNCING
:
275 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_SYNCING
, OFFLINE_CANCEL
))
277 /* Canceled restart from syncing, no need to wait. */
280 case OFFLINE_AGAIN_FROM_OFFLINING
:
281 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_OFFLINING
, OFFLINE_CANCEL
))
283 /* Canceled restart from offlining, must wait for offlining to complete however. */
285 /* fall through to wait */
289 r
= journal_file_set_offline_thread_join(f
);
299 if (mmap_cache_got_sigbus(f
->mmap
, f
->fd
))
302 switch (f
->header
->state
) {
307 f
->header
->state
= STATE_ONLINE
;
316 bool journal_file_is_offlining(JournalFile
*f
) {
319 __sync_synchronize();
321 if (f
->offline_state
== OFFLINE_DONE
||
322 f
->offline_state
== OFFLINE_JOINED
)
328 JournalFile
* journal_file_close(JournalFile
*f
) {
332 /* Write the final tag */
333 if (f
->seal
&& f
->writable
)
334 journal_file_append_tag(f
);
337 if (f
->post_change_timer
) {
340 if (sd_event_source_get_enabled(f
->post_change_timer
, &enabled
) >= 0)
341 if (enabled
== SD_EVENT_ONESHOT
)
342 journal_file_post_change(f
);
344 (void) sd_event_source_set_enabled(f
->post_change_timer
, SD_EVENT_OFF
);
345 sd_event_source_unref(f
->post_change_timer
);
348 journal_file_set_offline(f
, true);
350 if (f
->mmap
&& f
->fd
>= 0)
351 mmap_cache_close_fd(f
->mmap
, f
->fd
);
353 if (f
->fd
>= 0 && f
->defrag_on_close
) {
355 /* Be friendly to btrfs: turn COW back on again now,
356 * and defragment the file. We won't write to the file
357 * ever again, hence remove all fragmentation, and
358 * reenable all the good bits COW usually provides
359 * (such as data checksumming). */
361 (void) chattr_fd(f
->fd
, 0, FS_NOCOW_FL
);
362 (void) btrfs_defrag_fd(f
->fd
);
368 mmap_cache_unref(f
->mmap
);
370 ordered_hashmap_free_free(f
->chain_cache
);
372 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
373 free(f
->compress_buffer
);
378 munmap(f
->fss_file
, PAGE_ALIGN(f
->fss_file_size
));
380 free(f
->fsprg_state
);
385 gcry_md_close(f
->hmac
);
392 void journal_file_close_set(Set
*s
) {
397 while ((f
= set_steal_first(s
)))
398 (void) journal_file_close(f
);
401 static int journal_file_init_header(JournalFile
*f
, JournalFile
*template) {
408 memcpy(h
.signature
, HEADER_SIGNATURE
, 8);
409 h
.header_size
= htole64(ALIGN64(sizeof(h
)));
411 h
.incompatible_flags
|= htole32(
412 f
->compress_xz
* HEADER_INCOMPATIBLE_COMPRESSED_XZ
|
413 f
->compress_lz4
* HEADER_INCOMPATIBLE_COMPRESSED_LZ4
);
415 h
.compatible_flags
= htole32(
416 f
->seal
* HEADER_COMPATIBLE_SEALED
);
418 r
= sd_id128_randomize(&h
.file_id
);
423 h
.seqnum_id
= template->header
->seqnum_id
;
424 h
.tail_entry_seqnum
= template->header
->tail_entry_seqnum
;
426 h
.seqnum_id
= h
.file_id
;
428 k
= pwrite(f
->fd
, &h
, sizeof(h
), 0);
438 static int journal_file_refresh_header(JournalFile
*f
) {
445 r
= sd_id128_get_machine(&f
->header
->machine_id
);
449 r
= sd_id128_get_boot(&boot_id
);
453 if (sd_id128_equal(boot_id
, f
->header
->boot_id
))
454 f
->tail_entry_monotonic_valid
= true;
456 f
->header
->boot_id
= boot_id
;
458 r
= journal_file_set_online(f
);
460 /* Sync the online state to disk */
466 static int journal_file_verify_header(JournalFile
*f
) {
472 if (memcmp(f
->header
->signature
, HEADER_SIGNATURE
, 8))
475 /* In both read and write mode we refuse to open files with
476 * incompatible flags we don't know */
477 flags
= le32toh(f
->header
->incompatible_flags
);
478 if (flags
& ~HEADER_INCOMPATIBLE_SUPPORTED
) {
479 if (flags
& ~HEADER_INCOMPATIBLE_ANY
)
480 log_debug("Journal file %s has unknown incompatible flags %"PRIx32
,
481 f
->path
, flags
& ~HEADER_INCOMPATIBLE_ANY
);
482 flags
= (flags
& HEADER_INCOMPATIBLE_ANY
) & ~HEADER_INCOMPATIBLE_SUPPORTED
;
484 log_debug("Journal file %s uses incompatible flags %"PRIx32
485 " disabled at compilation time.", f
->path
, flags
);
486 return -EPROTONOSUPPORT
;
489 /* When open for writing we refuse to open files with
490 * compatible flags, too */
491 flags
= le32toh(f
->header
->compatible_flags
);
492 if (f
->writable
&& (flags
& ~HEADER_COMPATIBLE_SUPPORTED
)) {
493 if (flags
& ~HEADER_COMPATIBLE_ANY
)
494 log_debug("Journal file %s has unknown compatible flags %"PRIx32
,
495 f
->path
, flags
& ~HEADER_COMPATIBLE_ANY
);
496 flags
= (flags
& HEADER_COMPATIBLE_ANY
) & ~HEADER_COMPATIBLE_SUPPORTED
;
498 log_debug("Journal file %s uses compatible flags %"PRIx32
499 " disabled at compilation time.", f
->path
, flags
);
500 return -EPROTONOSUPPORT
;
503 if (f
->header
->state
>= _STATE_MAX
)
506 /* The first addition was n_data, so check that we are at least this large */
507 if (le64toh(f
->header
->header_size
) < HEADER_SIZE_MIN
)
510 if (JOURNAL_HEADER_SEALED(f
->header
) && !JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
513 if ((le64toh(f
->header
->header_size
) + le64toh(f
->header
->arena_size
)) > (uint64_t) f
->last_stat
.st_size
)
516 if (le64toh(f
->header
->tail_object_offset
) > (le64toh(f
->header
->header_size
) + le64toh(f
->header
->arena_size
)))
519 if (!VALID64(le64toh(f
->header
->data_hash_table_offset
)) ||
520 !VALID64(le64toh(f
->header
->field_hash_table_offset
)) ||
521 !VALID64(le64toh(f
->header
->tail_object_offset
)) ||
522 !VALID64(le64toh(f
->header
->entry_array_offset
)))
527 sd_id128_t machine_id
;
530 r
= sd_id128_get_machine(&machine_id
);
534 if (!sd_id128_equal(machine_id
, f
->header
->machine_id
))
537 state
= f
->header
->state
;
539 if (state
== STATE_ONLINE
) {
540 log_debug("Journal file %s is already online. Assuming unclean closing.", f
->path
);
542 } else if (state
== STATE_ARCHIVED
)
544 else if (state
!= STATE_OFFLINE
) {
545 log_debug("Journal file %s has unknown state %i.", f
->path
, state
);
550 f
->compress_xz
= JOURNAL_HEADER_COMPRESSED_XZ(f
->header
);
551 f
->compress_lz4
= JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
);
553 f
->seal
= JOURNAL_HEADER_SEALED(f
->header
);
558 static int journal_file_fstat(JournalFile
*f
) {
562 if (fstat(f
->fd
, &f
->last_stat
) < 0)
565 f
->last_stat_usec
= now(CLOCK_MONOTONIC
);
567 /* Refuse appending to files that are already deleted */
568 if (f
->last_stat
.st_nlink
<= 0)
574 static int journal_file_allocate(JournalFile
*f
, uint64_t offset
, uint64_t size
) {
575 uint64_t old_size
, new_size
;
581 /* We assume that this file is not sparse, and we know that
582 * for sure, since we always call posix_fallocate()
585 if (mmap_cache_got_sigbus(f
->mmap
, f
->fd
))
589 le64toh(f
->header
->header_size
) +
590 le64toh(f
->header
->arena_size
);
592 new_size
= PAGE_ALIGN(offset
+ size
);
593 if (new_size
< le64toh(f
->header
->header_size
))
594 new_size
= le64toh(f
->header
->header_size
);
596 if (new_size
<= old_size
) {
598 /* We already pre-allocated enough space, but before
599 * we write to it, let's check with fstat() if the
600 * file got deleted, in order make sure we don't throw
601 * away the data immediately. Don't check fstat() for
602 * all writes though, but only once ever 10s. */
604 if (f
->last_stat_usec
+ LAST_STAT_REFRESH_USEC
> now(CLOCK_MONOTONIC
))
607 return journal_file_fstat(f
);
610 /* Allocate more space. */
612 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
615 if (new_size
> f
->metrics
.min_size
&& f
->metrics
.keep_free
> 0) {
618 if (fstatvfs(f
->fd
, &svfs
) >= 0) {
621 available
= LESS_BY((uint64_t) svfs
.f_bfree
* (uint64_t) svfs
.f_bsize
, f
->metrics
.keep_free
);
623 if (new_size
- old_size
> available
)
628 /* Increase by larger blocks at once */
629 new_size
= ((new_size
+FILE_SIZE_INCREASE
-1) / FILE_SIZE_INCREASE
) * FILE_SIZE_INCREASE
;
630 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
631 new_size
= f
->metrics
.max_size
;
633 /* Note that the glibc fallocate() fallback is very
634 inefficient, hence we try to minimize the allocation area
636 r
= posix_fallocate(f
->fd
, old_size
, new_size
- old_size
);
640 f
->header
->arena_size
= htole64(new_size
- le64toh(f
->header
->header_size
));
642 return journal_file_fstat(f
);
645 static unsigned type_to_context(ObjectType type
) {
646 /* One context for each type, plus one catch-all for the rest */
647 assert_cc(_OBJECT_TYPE_MAX
<= MMAP_CACHE_MAX_CONTEXTS
);
648 assert_cc(CONTEXT_HEADER
< MMAP_CACHE_MAX_CONTEXTS
);
649 return type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
? type
: 0;
652 static int journal_file_move_to(JournalFile
*f
, ObjectType type
, bool keep_always
, uint64_t offset
, uint64_t size
, void **ret
) {
661 /* Avoid SIGBUS on invalid accesses */
662 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
) {
663 /* Hmm, out of range? Let's refresh the fstat() data
664 * first, before we trust that check. */
666 r
= journal_file_fstat(f
);
670 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
)
671 return -EADDRNOTAVAIL
;
674 return mmap_cache_get(f
->mmap
, f
->fd
, f
->prot
, type_to_context(type
), keep_always
, offset
, size
, &f
->last_stat
, ret
);
677 static uint64_t minimum_header_size(Object
*o
) {
679 static const uint64_t table
[] = {
680 [OBJECT_DATA
] = sizeof(DataObject
),
681 [OBJECT_FIELD
] = sizeof(FieldObject
),
682 [OBJECT_ENTRY
] = sizeof(EntryObject
),
683 [OBJECT_DATA_HASH_TABLE
] = sizeof(HashTableObject
),
684 [OBJECT_FIELD_HASH_TABLE
] = sizeof(HashTableObject
),
685 [OBJECT_ENTRY_ARRAY
] = sizeof(EntryArrayObject
),
686 [OBJECT_TAG
] = sizeof(TagObject
),
689 if (o
->object
.type
>= ELEMENTSOF(table
) || table
[o
->object
.type
] <= 0)
690 return sizeof(ObjectHeader
);
692 return table
[o
->object
.type
];
695 int journal_file_move_to_object(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
**ret
) {
704 /* Objects may only be located at multiple of 64 bit */
705 if (!VALID64(offset
))
708 r
= journal_file_move_to(f
, type
, false, offset
, sizeof(ObjectHeader
), &t
);
713 s
= le64toh(o
->object
.size
);
715 if (s
< sizeof(ObjectHeader
))
718 if (o
->object
.type
<= OBJECT_UNUSED
)
721 if (s
< minimum_header_size(o
))
724 if (type
> OBJECT_UNUSED
&& o
->object
.type
!= type
)
727 if (s
> sizeof(ObjectHeader
)) {
728 r
= journal_file_move_to(f
, type
, false, offset
, s
, &t
);
739 static uint64_t journal_file_entry_seqnum(JournalFile
*f
, uint64_t *seqnum
) {
745 r
= le64toh(f
->header
->tail_entry_seqnum
) + 1;
748 /* If an external seqnum counter was passed, we update
749 * both the local and the external one, and set it to
750 * the maximum of both */
758 f
->header
->tail_entry_seqnum
= htole64(r
);
760 if (f
->header
->head_entry_seqnum
== 0)
761 f
->header
->head_entry_seqnum
= htole64(r
);
766 int journal_file_append_object(JournalFile
*f
, ObjectType type
, uint64_t size
, Object
**ret
, uint64_t *offset
) {
774 assert(type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
);
775 assert(size
>= sizeof(ObjectHeader
));
779 r
= journal_file_set_online(f
);
783 p
= le64toh(f
->header
->tail_object_offset
);
785 p
= le64toh(f
->header
->header_size
);
787 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &tail
);
791 p
+= ALIGN64(le64toh(tail
->object
.size
));
794 r
= journal_file_allocate(f
, p
, size
);
798 r
= journal_file_move_to(f
, type
, false, p
, size
, &t
);
805 o
->object
.type
= type
;
806 o
->object
.size
= htole64(size
);
808 f
->header
->tail_object_offset
= htole64(p
);
809 f
->header
->n_objects
= htole64(le64toh(f
->header
->n_objects
) + 1);
817 static int journal_file_setup_data_hash_table(JournalFile
*f
) {
825 /* We estimate that we need 1 hash table entry per 768 bytes
826 of journal file and we want to make sure we never get
827 beyond 75% fill level. Calculate the hash table size for
828 the maximum file size based on these metrics. */
830 s
= (f
->metrics
.max_size
* 4 / 768 / 3) * sizeof(HashItem
);
831 if (s
< DEFAULT_DATA_HASH_TABLE_SIZE
)
832 s
= DEFAULT_DATA_HASH_TABLE_SIZE
;
834 log_debug("Reserving %"PRIu64
" entries in hash table.", s
/ sizeof(HashItem
));
836 r
= journal_file_append_object(f
,
837 OBJECT_DATA_HASH_TABLE
,
838 offsetof(Object
, hash_table
.items
) + s
,
843 memzero(o
->hash_table
.items
, s
);
845 f
->header
->data_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
846 f
->header
->data_hash_table_size
= htole64(s
);
851 static int journal_file_setup_field_hash_table(JournalFile
*f
) {
859 /* We use a fixed size hash table for the fields as this
860 * number should grow very slowly only */
862 s
= DEFAULT_FIELD_HASH_TABLE_SIZE
;
863 r
= journal_file_append_object(f
,
864 OBJECT_FIELD_HASH_TABLE
,
865 offsetof(Object
, hash_table
.items
) + s
,
870 memzero(o
->hash_table
.items
, s
);
872 f
->header
->field_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
873 f
->header
->field_hash_table_size
= htole64(s
);
878 int journal_file_map_data_hash_table(JournalFile
*f
) {
886 if (f
->data_hash_table
)
889 p
= le64toh(f
->header
->data_hash_table_offset
);
890 s
= le64toh(f
->header
->data_hash_table_size
);
892 r
= journal_file_move_to(f
,
893 OBJECT_DATA_HASH_TABLE
,
900 f
->data_hash_table
= t
;
904 int journal_file_map_field_hash_table(JournalFile
*f
) {
912 if (f
->field_hash_table
)
915 p
= le64toh(f
->header
->field_hash_table_offset
);
916 s
= le64toh(f
->header
->field_hash_table_size
);
918 r
= journal_file_move_to(f
,
919 OBJECT_FIELD_HASH_TABLE
,
926 f
->field_hash_table
= t
;
930 static int journal_file_link_field(
941 assert(f
->field_hash_table
);
945 if (o
->object
.type
!= OBJECT_FIELD
)
948 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
952 /* This might alter the window we are looking at */
953 o
->field
.next_hash_offset
= o
->field
.head_data_offset
= 0;
956 p
= le64toh(f
->field_hash_table
[h
].tail_hash_offset
);
958 f
->field_hash_table
[h
].head_hash_offset
= htole64(offset
);
960 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
964 o
->field
.next_hash_offset
= htole64(offset
);
967 f
->field_hash_table
[h
].tail_hash_offset
= htole64(offset
);
969 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
970 f
->header
->n_fields
= htole64(le64toh(f
->header
->n_fields
) + 1);
975 static int journal_file_link_data(
986 assert(f
->data_hash_table
);
990 if (o
->object
.type
!= OBJECT_DATA
)
993 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
997 /* This might alter the window we are looking at */
998 o
->data
.next_hash_offset
= o
->data
.next_field_offset
= 0;
999 o
->data
.entry_offset
= o
->data
.entry_array_offset
= 0;
1000 o
->data
.n_entries
= 0;
1003 p
= le64toh(f
->data_hash_table
[h
].tail_hash_offset
);
1005 /* Only entry in the hash table is easy */
1006 f
->data_hash_table
[h
].head_hash_offset
= htole64(offset
);
1008 /* Move back to the previous data object, to patch in
1011 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1015 o
->data
.next_hash_offset
= htole64(offset
);
1018 f
->data_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1020 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
1021 f
->header
->n_data
= htole64(le64toh(f
->header
->n_data
) + 1);
1026 int journal_file_find_field_object_with_hash(
1028 const void *field
, uint64_t size
, uint64_t hash
,
1029 Object
**ret
, uint64_t *offset
) {
1031 uint64_t p
, osize
, h
, m
;
1036 assert(field
&& size
> 0);
1038 /* If the field hash table is empty, we can't find anything */
1039 if (le64toh(f
->header
->field_hash_table_size
) <= 0)
1042 /* Map the field hash table, if it isn't mapped yet. */
1043 r
= journal_file_map_field_hash_table(f
);
1047 osize
= offsetof(Object
, field
.payload
) + size
;
1049 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1054 p
= le64toh(f
->field_hash_table
[h
].head_hash_offset
);
1059 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1063 if (le64toh(o
->field
.hash
) == hash
&&
1064 le64toh(o
->object
.size
) == osize
&&
1065 memcmp(o
->field
.payload
, field
, size
) == 0) {
1075 p
= le64toh(o
->field
.next_hash_offset
);
1081 int journal_file_find_field_object(
1083 const void *field
, uint64_t size
,
1084 Object
**ret
, uint64_t *offset
) {
1089 assert(field
&& size
> 0);
1091 hash
= hash64(field
, size
);
1093 return journal_file_find_field_object_with_hash(f
,
1098 int journal_file_find_data_object_with_hash(
1100 const void *data
, uint64_t size
, uint64_t hash
,
1101 Object
**ret
, uint64_t *offset
) {
1103 uint64_t p
, osize
, h
, m
;
1108 assert(data
|| size
== 0);
1110 /* If there's no data hash table, then there's no entry. */
1111 if (le64toh(f
->header
->data_hash_table_size
) <= 0)
1114 /* Map the data hash table, if it isn't mapped yet. */
1115 r
= journal_file_map_data_hash_table(f
);
1119 osize
= offsetof(Object
, data
.payload
) + size
;
1121 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1126 p
= le64toh(f
->data_hash_table
[h
].head_hash_offset
);
1131 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1135 if (le64toh(o
->data
.hash
) != hash
)
1138 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
1139 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
1143 l
= le64toh(o
->object
.size
);
1144 if (l
<= offsetof(Object
, data
.payload
))
1147 l
-= offsetof(Object
, data
.payload
);
1149 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
1150 o
->data
.payload
, l
, &f
->compress_buffer
, &f
->compress_buffer_size
, &rsize
, 0);
1154 if (rsize
== size
&&
1155 memcmp(f
->compress_buffer
, data
, size
) == 0) {
1166 return -EPROTONOSUPPORT
;
1168 } else if (le64toh(o
->object
.size
) == osize
&&
1169 memcmp(o
->data
.payload
, data
, size
) == 0) {
1181 p
= le64toh(o
->data
.next_hash_offset
);
1187 int journal_file_find_data_object(
1189 const void *data
, uint64_t size
,
1190 Object
**ret
, uint64_t *offset
) {
1195 assert(data
|| size
== 0);
1197 hash
= hash64(data
, size
);
1199 return journal_file_find_data_object_with_hash(f
,
1204 static int journal_file_append_field(
1206 const void *field
, uint64_t size
,
1207 Object
**ret
, uint64_t *offset
) {
1215 assert(field
&& size
> 0);
1217 hash
= hash64(field
, size
);
1219 r
= journal_file_find_field_object_with_hash(f
, field
, size
, hash
, &o
, &p
);
1233 osize
= offsetof(Object
, field
.payload
) + size
;
1234 r
= journal_file_append_object(f
, OBJECT_FIELD
, osize
, &o
, &p
);
1238 o
->field
.hash
= htole64(hash
);
1239 memcpy(o
->field
.payload
, field
, size
);
1241 r
= journal_file_link_field(f
, o
, p
, hash
);
1245 /* The linking might have altered the window, so let's
1246 * refresh our pointer */
1247 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1252 r
= journal_file_hmac_put_object(f
, OBJECT_FIELD
, o
, p
);
1266 static int journal_file_append_data(
1268 const void *data
, uint64_t size
,
1269 Object
**ret
, uint64_t *offset
) {
1274 int r
, compression
= 0;
1278 assert(data
|| size
== 0);
1280 hash
= hash64(data
, size
);
1282 r
= journal_file_find_data_object_with_hash(f
, data
, size
, hash
, &o
, &p
);
1296 osize
= offsetof(Object
, data
.payload
) + size
;
1297 r
= journal_file_append_object(f
, OBJECT_DATA
, osize
, &o
, &p
);
1301 o
->data
.hash
= htole64(hash
);
1303 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
1304 if (JOURNAL_FILE_COMPRESS(f
) && size
>= COMPRESSION_SIZE_THRESHOLD
) {
1307 compression
= compress_blob(data
, size
, o
->data
.payload
, size
- 1, &rsize
);
1309 if (compression
>= 0) {
1310 o
->object
.size
= htole64(offsetof(Object
, data
.payload
) + rsize
);
1311 o
->object
.flags
|= compression
;
1313 log_debug("Compressed data object %"PRIu64
" -> %zu using %s",
1314 size
, rsize
, object_compressed_to_string(compression
));
1316 /* Compression didn't work, we don't really care why, let's continue without compression */
1321 if (compression
== 0)
1322 memcpy_safe(o
->data
.payload
, data
, size
);
1324 r
= journal_file_link_data(f
, o
, p
, hash
);
1328 /* The linking might have altered the window, so let's
1329 * refresh our pointer */
1330 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1337 eq
= memchr(data
, '=', size
);
1338 if (eq
&& eq
> data
) {
1342 /* Create field object ... */
1343 r
= journal_file_append_field(f
, data
, (uint8_t*) eq
- (uint8_t*) data
, &fo
, &fp
);
1347 /* ... and link it in. */
1348 o
->data
.next_field_offset
= fo
->field
.head_data_offset
;
1349 fo
->field
.head_data_offset
= le64toh(p
);
1353 r
= journal_file_hmac_put_object(f
, OBJECT_DATA
, o
, p
);
1367 uint64_t journal_file_entry_n_items(Object
*o
) {
1370 if (o
->object
.type
!= OBJECT_ENTRY
)
1373 return (le64toh(o
->object
.size
) - offsetof(Object
, entry
.items
)) / sizeof(EntryItem
);
1376 uint64_t journal_file_entry_array_n_items(Object
*o
) {
1379 if (o
->object
.type
!= OBJECT_ENTRY_ARRAY
)
1382 return (le64toh(o
->object
.size
) - offsetof(Object
, entry_array
.items
)) / sizeof(uint64_t);
1385 uint64_t journal_file_hash_table_n_items(Object
*o
) {
1388 if (o
->object
.type
!= OBJECT_DATA_HASH_TABLE
&&
1389 o
->object
.type
!= OBJECT_FIELD_HASH_TABLE
)
1392 return (le64toh(o
->object
.size
) - offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
);
1395 static int link_entry_into_array(JournalFile
*f
,
1400 uint64_t n
= 0, ap
= 0, q
, i
, a
, hidx
;
1409 a
= le64toh(*first
);
1410 i
= hidx
= le64toh(*idx
);
1413 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
1417 n
= journal_file_entry_array_n_items(o
);
1419 o
->entry_array
.items
[i
] = htole64(p
);
1420 *idx
= htole64(hidx
+ 1);
1426 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
1437 r
= journal_file_append_object(f
, OBJECT_ENTRY_ARRAY
,
1438 offsetof(Object
, entry_array
.items
) + n
* sizeof(uint64_t),
1444 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY_ARRAY
, o
, q
);
1449 o
->entry_array
.items
[i
] = htole64(p
);
1452 *first
= htole64(q
);
1454 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, ap
, &o
);
1458 o
->entry_array
.next_entry_array_offset
= htole64(q
);
1461 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
1462 f
->header
->n_entry_arrays
= htole64(le64toh(f
->header
->n_entry_arrays
) + 1);
1464 *idx
= htole64(hidx
+ 1);
1469 static int link_entry_into_array_plus_one(JournalFile
*f
,
1484 *extra
= htole64(p
);
1488 i
= htole64(le64toh(*idx
) - 1);
1489 r
= link_entry_into_array(f
, first
, &i
, p
);
1494 *idx
= htole64(le64toh(*idx
) + 1);
1498 static int journal_file_link_entry_item(JournalFile
*f
, Object
*o
, uint64_t offset
, uint64_t i
) {
1505 p
= le64toh(o
->entry
.items
[i
].object_offset
);
1509 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1513 return link_entry_into_array_plus_one(f
,
1514 &o
->data
.entry_offset
,
1515 &o
->data
.entry_array_offset
,
1520 static int journal_file_link_entry(JournalFile
*f
, Object
*o
, uint64_t offset
) {
1529 if (o
->object
.type
!= OBJECT_ENTRY
)
1532 __sync_synchronize();
1534 /* Link up the entry itself */
1535 r
= link_entry_into_array(f
,
1536 &f
->header
->entry_array_offset
,
1537 &f
->header
->n_entries
,
1542 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
1544 if (f
->header
->head_entry_realtime
== 0)
1545 f
->header
->head_entry_realtime
= o
->entry
.realtime
;
1547 f
->header
->tail_entry_realtime
= o
->entry
.realtime
;
1548 f
->header
->tail_entry_monotonic
= o
->entry
.monotonic
;
1550 f
->tail_entry_monotonic_valid
= true;
1552 /* Link up the items */
1553 n
= journal_file_entry_n_items(o
);
1554 for (i
= 0; i
< n
; i
++) {
1555 r
= journal_file_link_entry_item(f
, o
, offset
, i
);
1563 static int journal_file_append_entry_internal(
1565 const dual_timestamp
*ts
,
1567 const EntryItem items
[], unsigned n_items
,
1569 Object
**ret
, uint64_t *offset
) {
1577 assert(items
|| n_items
== 0);
1580 osize
= offsetof(Object
, entry
.items
) + (n_items
* sizeof(EntryItem
));
1582 r
= journal_file_append_object(f
, OBJECT_ENTRY
, osize
, &o
, &np
);
1586 o
->entry
.seqnum
= htole64(journal_file_entry_seqnum(f
, seqnum
));
1587 memcpy_safe(o
->entry
.items
, items
, n_items
* sizeof(EntryItem
));
1588 o
->entry
.realtime
= htole64(ts
->realtime
);
1589 o
->entry
.monotonic
= htole64(ts
->monotonic
);
1590 o
->entry
.xor_hash
= htole64(xor_hash
);
1591 o
->entry
.boot_id
= f
->header
->boot_id
;
1594 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY
, o
, np
);
1599 r
= journal_file_link_entry(f
, o
, np
);
1612 void journal_file_post_change(JournalFile
*f
) {
1615 /* inotify() does not receive IN_MODIFY events from file
1616 * accesses done via mmap(). After each access we hence
1617 * trigger IN_MODIFY by truncating the journal file to its
1618 * current size which triggers IN_MODIFY. */
1620 __sync_synchronize();
1622 if (ftruncate(f
->fd
, f
->last_stat
.st_size
) < 0)
1623 log_debug_errno(errno
, "Failed to truncate file to its own size: %m");
1626 static int post_change_thunk(sd_event_source
*timer
, uint64_t usec
, void *userdata
) {
1629 journal_file_post_change(userdata
);
1634 static void schedule_post_change(JournalFile
*f
) {
1635 sd_event_source
*timer
;
1640 assert(f
->post_change_timer
);
1642 timer
= f
->post_change_timer
;
1644 r
= sd_event_source_get_enabled(timer
, &enabled
);
1646 log_debug_errno(r
, "Failed to get ftruncate timer state: %m");
1650 if (enabled
== SD_EVENT_ONESHOT
)
1653 r
= sd_event_now(sd_event_source_get_event(timer
), CLOCK_MONOTONIC
, &now
);
1655 log_debug_errno(r
, "Failed to get clock's now for scheduling ftruncate: %m");
1659 r
= sd_event_source_set_time(timer
, now
+f
->post_change_timer_period
);
1661 log_debug_errno(r
, "Failed to set time for scheduling ftruncate: %m");
1665 r
= sd_event_source_set_enabled(timer
, SD_EVENT_ONESHOT
);
1667 log_debug_errno(r
, "Failed to enable scheduled ftruncate: %m");
1674 /* On failure, let's simply post the change immediately. */
1675 journal_file_post_change(f
);
1678 /* Enable coalesced change posting in a timer on the provided sd_event instance */
1679 int journal_file_enable_post_change_timer(JournalFile
*f
, sd_event
*e
, usec_t t
) {
1680 _cleanup_(sd_event_source_unrefp
) sd_event_source
*timer
= NULL
;
1684 assert_return(!f
->post_change_timer
, -EINVAL
);
1688 r
= sd_event_add_time(e
, &timer
, CLOCK_MONOTONIC
, 0, 0, post_change_thunk
, f
);
1692 r
= sd_event_source_set_enabled(timer
, SD_EVENT_OFF
);
1696 f
->post_change_timer
= timer
;
1698 f
->post_change_timer_period
= t
;
1703 static int entry_item_cmp(const void *_a
, const void *_b
) {
1704 const EntryItem
*a
= _a
, *b
= _b
;
1706 if (le64toh(a
->object_offset
) < le64toh(b
->object_offset
))
1708 if (le64toh(a
->object_offset
) > le64toh(b
->object_offset
))
1713 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
) {
1717 uint64_t xor_hash
= 0;
1718 struct dual_timestamp _ts
;
1722 assert(iovec
|| n_iovec
== 0);
1725 dual_timestamp_get(&_ts
);
1730 r
= journal_file_maybe_append_tag(f
, ts
->realtime
);
1735 /* alloca() can't take 0, hence let's allocate at least one */
1736 items
= alloca(sizeof(EntryItem
) * MAX(1u, n_iovec
));
1738 for (i
= 0; i
< n_iovec
; i
++) {
1742 r
= journal_file_append_data(f
, iovec
[i
].iov_base
, iovec
[i
].iov_len
, &o
, &p
);
1746 xor_hash
^= le64toh(o
->data
.hash
);
1747 items
[i
].object_offset
= htole64(p
);
1748 items
[i
].hash
= o
->data
.hash
;
1751 /* Order by the position on disk, in order to improve seek
1752 * times for rotating media. */
1753 qsort_safe(items
, n_iovec
, sizeof(EntryItem
), entry_item_cmp
);
1755 r
= journal_file_append_entry_internal(f
, ts
, xor_hash
, items
, n_iovec
, seqnum
, ret
, offset
);
1757 /* If the memory mapping triggered a SIGBUS then we return an
1758 * IO error and ignore the error code passed down to us, since
1759 * it is very likely just an effect of a nullified replacement
1762 if (mmap_cache_got_sigbus(f
->mmap
, f
->fd
))
1765 if (f
->post_change_timer
)
1766 schedule_post_change(f
);
1768 journal_file_post_change(f
);
1773 typedef struct ChainCacheItem
{
1774 uint64_t first
; /* the array at the beginning of the chain */
1775 uint64_t array
; /* the cached array */
1776 uint64_t begin
; /* the first item in the cached array */
1777 uint64_t total
; /* the total number of items in all arrays before this one in the chain */
1778 uint64_t last_index
; /* the last index we looked at, to optimize locality when bisecting */
1781 static void chain_cache_put(
1788 uint64_t last_index
) {
1791 /* If the chain item to cache for this chain is the
1792 * first one it's not worth caching anything */
1796 if (ordered_hashmap_size(h
) >= CHAIN_CACHE_MAX
) {
1797 ci
= ordered_hashmap_steal_first(h
);
1800 ci
= new(ChainCacheItem
, 1);
1807 if (ordered_hashmap_put(h
, &ci
->first
, ci
) < 0) {
1812 assert(ci
->first
== first
);
1817 ci
->last_index
= last_index
;
1820 static int generic_array_get(
1824 Object
**ret
, uint64_t *offset
) {
1827 uint64_t p
= 0, a
, t
= 0;
1835 /* Try the chain cache first */
1836 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
1837 if (ci
&& i
> ci
->total
) {
1846 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
1850 k
= journal_file_entry_array_n_items(o
);
1852 p
= le64toh(o
->entry_array
.items
[i
]);
1858 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
1864 /* Let's cache this item for the next invocation */
1865 chain_cache_put(f
->chain_cache
, ci
, first
, a
, le64toh(o
->entry_array
.items
[0]), t
, i
);
1867 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
1880 static int generic_array_get_plus_one(
1885 Object
**ret
, uint64_t *offset
) {
1894 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
1907 return generic_array_get(f
, first
, i
-1, ret
, offset
);
1916 static int generic_array_bisect(
1921 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
1922 direction_t direction
,
1927 uint64_t a
, p
, t
= 0, i
= 0, last_p
= 0, last_index
= (uint64_t) -1;
1928 bool subtract_one
= false;
1929 Object
*o
, *array
= NULL
;
1934 assert(test_object
);
1936 /* Start with the first array in the chain */
1939 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
1940 if (ci
&& n
> ci
->total
) {
1941 /* Ah, we have iterated this bisection array chain
1942 * previously! Let's see if we can skip ahead in the
1943 * chain, as far as the last time. But we can't jump
1944 * backwards in the chain, so let's check that
1947 r
= test_object(f
, ci
->begin
, needle
);
1951 if (r
== TEST_LEFT
) {
1952 /* OK, what we are looking for is right of the
1953 * begin of this EntryArray, so let's jump
1954 * straight to previously cached array in the
1960 last_index
= ci
->last_index
;
1965 uint64_t left
, right
, k
, lp
;
1967 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
1971 k
= journal_file_entry_array_n_items(array
);
1977 lp
= p
= le64toh(array
->entry_array
.items
[i
]);
1981 r
= test_object(f
, p
, needle
);
1985 if (r
== TEST_FOUND
)
1986 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
1988 if (r
== TEST_RIGHT
) {
1992 if (last_index
!= (uint64_t) -1) {
1993 assert(last_index
<= right
);
1995 /* If we cached the last index we
1996 * looked at, let's try to not to jump
1997 * too wildly around and see if we can
1998 * limit the range to look at early to
1999 * the immediate neighbors of the last
2000 * index we looked at. */
2002 if (last_index
> 0) {
2003 uint64_t x
= last_index
- 1;
2005 p
= le64toh(array
->entry_array
.items
[x
]);
2009 r
= test_object(f
, p
, needle
);
2013 if (r
== TEST_FOUND
)
2014 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2016 if (r
== TEST_RIGHT
)
2022 if (last_index
< right
) {
2023 uint64_t y
= last_index
+ 1;
2025 p
= le64toh(array
->entry_array
.items
[y
]);
2029 r
= test_object(f
, p
, needle
);
2033 if (r
== TEST_FOUND
)
2034 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2036 if (r
== TEST_RIGHT
)
2044 if (left
== right
) {
2045 if (direction
== DIRECTION_UP
)
2046 subtract_one
= true;
2052 assert(left
< right
);
2053 i
= (left
+ right
) / 2;
2055 p
= le64toh(array
->entry_array
.items
[i
]);
2059 r
= test_object(f
, p
, needle
);
2063 if (r
== TEST_FOUND
)
2064 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2066 if (r
== TEST_RIGHT
)
2074 if (direction
== DIRECTION_UP
) {
2076 subtract_one
= true;
2087 last_index
= (uint64_t) -1;
2088 a
= le64toh(array
->entry_array
.next_entry_array_offset
);
2094 if (subtract_one
&& t
== 0 && i
== 0)
2097 /* Let's cache this item for the next invocation */
2098 chain_cache_put(f
->chain_cache
, ci
, first
, a
, le64toh(array
->entry_array
.items
[0]), t
, subtract_one
? (i
> 0 ? i
-1 : (uint64_t) -1) : i
);
2100 if (subtract_one
&& i
== 0)
2102 else if (subtract_one
)
2103 p
= le64toh(array
->entry_array
.items
[i
-1]);
2105 p
= le64toh(array
->entry_array
.items
[i
]);
2107 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2118 *idx
= t
+ i
+ (subtract_one
? -1 : 0);
2123 static int generic_array_bisect_plus_one(
2129 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2130 direction_t direction
,
2136 bool step_back
= false;
2140 assert(test_object
);
2145 /* This bisects the array in object 'first', but first checks
2147 r
= test_object(f
, extra
, needle
);
2151 if (r
== TEST_FOUND
)
2152 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2154 /* if we are looking with DIRECTION_UP then we need to first
2155 see if in the actual array there is a matching entry, and
2156 return the last one of that. But if there isn't any we need
2157 to return this one. Hence remember this, and return it
2160 step_back
= direction
== DIRECTION_UP
;
2162 if (r
== TEST_RIGHT
) {
2163 if (direction
== DIRECTION_DOWN
)
2169 r
= generic_array_bisect(f
, first
, n
-1, needle
, test_object
, direction
, ret
, offset
, idx
);
2171 if (r
== 0 && step_back
)
2180 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2196 _pure_
static int test_object_offset(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2202 else if (p
< needle
)
2208 static int test_object_seqnum(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2215 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2219 if (le64toh(o
->entry
.seqnum
) == needle
)
2221 else if (le64toh(o
->entry
.seqnum
) < needle
)
2227 int journal_file_move_to_entry_by_seqnum(
2230 direction_t direction
,
2236 return generic_array_bisect(f
,
2237 le64toh(f
->header
->entry_array_offset
),
2238 le64toh(f
->header
->n_entries
),
2245 static int test_object_realtime(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2252 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2256 if (le64toh(o
->entry
.realtime
) == needle
)
2258 else if (le64toh(o
->entry
.realtime
) < needle
)
2264 int journal_file_move_to_entry_by_realtime(
2267 direction_t direction
,
2273 return generic_array_bisect(f
,
2274 le64toh(f
->header
->entry_array_offset
),
2275 le64toh(f
->header
->n_entries
),
2277 test_object_realtime
,
2282 static int test_object_monotonic(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2289 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2293 if (le64toh(o
->entry
.monotonic
) == needle
)
2295 else if (le64toh(o
->entry
.monotonic
) < needle
)
2301 static int find_data_object_by_boot_id(
2307 char t
[sizeof("_BOOT_ID=")-1 + 32 + 1] = "_BOOT_ID=";
2309 sd_id128_to_string(boot_id
, t
+ 9);
2310 return journal_file_find_data_object(f
, t
, sizeof(t
) - 1, o
, b
);
2313 int journal_file_move_to_entry_by_monotonic(
2317 direction_t direction
,
2326 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, NULL
);
2332 return generic_array_bisect_plus_one(f
,
2333 le64toh(o
->data
.entry_offset
),
2334 le64toh(o
->data
.entry_array_offset
),
2335 le64toh(o
->data
.n_entries
),
2337 test_object_monotonic
,
2342 void journal_file_reset_location(JournalFile
*f
) {
2343 f
->location_type
= LOCATION_HEAD
;
2344 f
->current_offset
= 0;
2345 f
->current_seqnum
= 0;
2346 f
->current_realtime
= 0;
2347 f
->current_monotonic
= 0;
2348 zero(f
->current_boot_id
);
2349 f
->current_xor_hash
= 0;
2352 void journal_file_save_location(JournalFile
*f
, Object
*o
, uint64_t offset
) {
2353 f
->location_type
= LOCATION_SEEK
;
2354 f
->current_offset
= offset
;
2355 f
->current_seqnum
= le64toh(o
->entry
.seqnum
);
2356 f
->current_realtime
= le64toh(o
->entry
.realtime
);
2357 f
->current_monotonic
= le64toh(o
->entry
.monotonic
);
2358 f
->current_boot_id
= o
->entry
.boot_id
;
2359 f
->current_xor_hash
= le64toh(o
->entry
.xor_hash
);
2362 int journal_file_compare_locations(JournalFile
*af
, JournalFile
*bf
) {
2367 assert(af
->location_type
== LOCATION_SEEK
);
2368 assert(bf
->location_type
== LOCATION_SEEK
);
2370 /* If contents and timestamps match, these entries are
2371 * identical, even if the seqnum does not match */
2372 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
) &&
2373 af
->current_monotonic
== bf
->current_monotonic
&&
2374 af
->current_realtime
== bf
->current_realtime
&&
2375 af
->current_xor_hash
== bf
->current_xor_hash
)
2378 if (sd_id128_equal(af
->header
->seqnum_id
, bf
->header
->seqnum_id
)) {
2380 /* If this is from the same seqnum source, compare
2382 if (af
->current_seqnum
< bf
->current_seqnum
)
2384 if (af
->current_seqnum
> bf
->current_seqnum
)
2387 /* Wow! This is weird, different data but the same
2388 * seqnums? Something is borked, but let's make the
2389 * best of it and compare by time. */
2392 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
)) {
2394 /* If the boot id matches, compare monotonic time */
2395 if (af
->current_monotonic
< bf
->current_monotonic
)
2397 if (af
->current_monotonic
> bf
->current_monotonic
)
2401 /* Otherwise, compare UTC time */
2402 if (af
->current_realtime
< bf
->current_realtime
)
2404 if (af
->current_realtime
> bf
->current_realtime
)
2407 /* Finally, compare by contents */
2408 if (af
->current_xor_hash
< bf
->current_xor_hash
)
2410 if (af
->current_xor_hash
> bf
->current_xor_hash
)
2416 int journal_file_next_entry(
2419 direction_t direction
,
2420 Object
**ret
, uint64_t *offset
) {
2428 n
= le64toh(f
->header
->n_entries
);
2433 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2435 r
= generic_array_bisect(f
,
2436 le64toh(f
->header
->entry_array_offset
),
2437 le64toh(f
->header
->n_entries
),
2446 if (direction
== DIRECTION_DOWN
) {
2459 /* And jump to it */
2460 r
= generic_array_get(f
,
2461 le64toh(f
->header
->entry_array_offset
),
2468 (direction
== DIRECTION_DOWN
? ofs
<= p
: ofs
>= p
)) {
2469 log_debug("%s: entry array corrupted at entry %"PRIu64
,
2480 int journal_file_next_entry_for_data(
2482 Object
*o
, uint64_t p
,
2483 uint64_t data_offset
,
2484 direction_t direction
,
2485 Object
**ret
, uint64_t *offset
) {
2492 assert(p
> 0 || !o
);
2494 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2498 n
= le64toh(d
->data
.n_entries
);
2503 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2505 if (o
->object
.type
!= OBJECT_ENTRY
)
2508 r
= generic_array_bisect_plus_one(f
,
2509 le64toh(d
->data
.entry_offset
),
2510 le64toh(d
->data
.entry_array_offset
),
2511 le64toh(d
->data
.n_entries
),
2521 if (direction
== DIRECTION_DOWN
) {
2535 return generic_array_get_plus_one(f
,
2536 le64toh(d
->data
.entry_offset
),
2537 le64toh(d
->data
.entry_array_offset
),
2542 int journal_file_move_to_entry_by_offset_for_data(
2544 uint64_t data_offset
,
2546 direction_t direction
,
2547 Object
**ret
, uint64_t *offset
) {
2554 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2558 return generic_array_bisect_plus_one(f
,
2559 le64toh(d
->data
.entry_offset
),
2560 le64toh(d
->data
.entry_array_offset
),
2561 le64toh(d
->data
.n_entries
),
2568 int journal_file_move_to_entry_by_monotonic_for_data(
2570 uint64_t data_offset
,
2573 direction_t direction
,
2574 Object
**ret
, uint64_t *offset
) {
2582 /* First, seek by time */
2583 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &b
);
2589 r
= generic_array_bisect_plus_one(f
,
2590 le64toh(o
->data
.entry_offset
),
2591 le64toh(o
->data
.entry_array_offset
),
2592 le64toh(o
->data
.n_entries
),
2594 test_object_monotonic
,
2600 /* And now, continue seeking until we find an entry that
2601 * exists in both bisection arrays */
2607 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2611 r
= generic_array_bisect_plus_one(f
,
2612 le64toh(d
->data
.entry_offset
),
2613 le64toh(d
->data
.entry_array_offset
),
2614 le64toh(d
->data
.n_entries
),
2622 r
= journal_file_move_to_object(f
, OBJECT_DATA
, b
, &o
);
2626 r
= generic_array_bisect_plus_one(f
,
2627 le64toh(o
->data
.entry_offset
),
2628 le64toh(o
->data
.entry_array_offset
),
2629 le64toh(o
->data
.n_entries
),
2651 int journal_file_move_to_entry_by_seqnum_for_data(
2653 uint64_t data_offset
,
2655 direction_t direction
,
2656 Object
**ret
, uint64_t *offset
) {
2663 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2667 return generic_array_bisect_plus_one(f
,
2668 le64toh(d
->data
.entry_offset
),
2669 le64toh(d
->data
.entry_array_offset
),
2670 le64toh(d
->data
.n_entries
),
2677 int journal_file_move_to_entry_by_realtime_for_data(
2679 uint64_t data_offset
,
2681 direction_t direction
,
2682 Object
**ret
, uint64_t *offset
) {
2689 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2693 return generic_array_bisect_plus_one(f
,
2694 le64toh(d
->data
.entry_offset
),
2695 le64toh(d
->data
.entry_array_offset
),
2696 le64toh(d
->data
.n_entries
),
2698 test_object_realtime
,
2703 void journal_file_dump(JournalFile
*f
) {
2711 journal_file_print_header(f
);
2713 p
= le64toh(f
->header
->header_size
);
2715 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &o
);
2719 switch (o
->object
.type
) {
2722 printf("Type: OBJECT_UNUSED\n");
2726 printf("Type: OBJECT_DATA\n");
2730 printf("Type: OBJECT_FIELD\n");
2734 printf("Type: OBJECT_ENTRY seqnum=%"PRIu64
" monotonic=%"PRIu64
" realtime=%"PRIu64
"\n",
2735 le64toh(o
->entry
.seqnum
),
2736 le64toh(o
->entry
.monotonic
),
2737 le64toh(o
->entry
.realtime
));
2740 case OBJECT_FIELD_HASH_TABLE
:
2741 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
2744 case OBJECT_DATA_HASH_TABLE
:
2745 printf("Type: OBJECT_DATA_HASH_TABLE\n");
2748 case OBJECT_ENTRY_ARRAY
:
2749 printf("Type: OBJECT_ENTRY_ARRAY\n");
2753 printf("Type: OBJECT_TAG seqnum=%"PRIu64
" epoch=%"PRIu64
"\n",
2754 le64toh(o
->tag
.seqnum
),
2755 le64toh(o
->tag
.epoch
));
2759 printf("Type: unknown (%i)\n", o
->object
.type
);
2763 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
)
2764 printf("Flags: %s\n",
2765 object_compressed_to_string(o
->object
.flags
& OBJECT_COMPRESSION_MASK
));
2767 if (p
== le64toh(f
->header
->tail_object_offset
))
2770 p
= p
+ ALIGN64(le64toh(o
->object
.size
));
2775 log_error("File corrupt");
2778 static const char* format_timestamp_safe(char *buf
, size_t l
, usec_t t
) {
2781 x
= format_timestamp(buf
, l
, t
);
2787 void journal_file_print_header(JournalFile
*f
) {
2788 char a
[33], b
[33], c
[33], d
[33];
2789 char x
[FORMAT_TIMESTAMP_MAX
], y
[FORMAT_TIMESTAMP_MAX
], z
[FORMAT_TIMESTAMP_MAX
];
2791 char bytes
[FORMAT_BYTES_MAX
];
2796 printf("File Path: %s\n"
2800 "Sequential Number ID: %s\n"
2802 "Compatible Flags:%s%s\n"
2803 "Incompatible Flags:%s%s%s\n"
2804 "Header size: %"PRIu64
"\n"
2805 "Arena size: %"PRIu64
"\n"
2806 "Data Hash Table Size: %"PRIu64
"\n"
2807 "Field Hash Table Size: %"PRIu64
"\n"
2808 "Rotate Suggested: %s\n"
2809 "Head Sequential Number: %"PRIu64
"\n"
2810 "Tail Sequential Number: %"PRIu64
"\n"
2811 "Head Realtime Timestamp: %s\n"
2812 "Tail Realtime Timestamp: %s\n"
2813 "Tail Monotonic Timestamp: %s\n"
2814 "Objects: %"PRIu64
"\n"
2815 "Entry Objects: %"PRIu64
"\n",
2817 sd_id128_to_string(f
->header
->file_id
, a
),
2818 sd_id128_to_string(f
->header
->machine_id
, b
),
2819 sd_id128_to_string(f
->header
->boot_id
, c
),
2820 sd_id128_to_string(f
->header
->seqnum_id
, d
),
2821 f
->header
->state
== STATE_OFFLINE
? "OFFLINE" :
2822 f
->header
->state
== STATE_ONLINE
? "ONLINE" :
2823 f
->header
->state
== STATE_ARCHIVED
? "ARCHIVED" : "UNKNOWN",
2824 JOURNAL_HEADER_SEALED(f
->header
) ? " SEALED" : "",
2825 (le32toh(f
->header
->compatible_flags
) & ~HEADER_COMPATIBLE_ANY
) ? " ???" : "",
2826 JOURNAL_HEADER_COMPRESSED_XZ(f
->header
) ? " COMPRESSED-XZ" : "",
2827 JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
) ? " COMPRESSED-LZ4" : "",
2828 (le32toh(f
->header
->incompatible_flags
) & ~HEADER_INCOMPATIBLE_ANY
) ? " ???" : "",
2829 le64toh(f
->header
->header_size
),
2830 le64toh(f
->header
->arena_size
),
2831 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
2832 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
),
2833 yes_no(journal_file_rotate_suggested(f
, 0)),
2834 le64toh(f
->header
->head_entry_seqnum
),
2835 le64toh(f
->header
->tail_entry_seqnum
),
2836 format_timestamp_safe(x
, sizeof(x
), le64toh(f
->header
->head_entry_realtime
)),
2837 format_timestamp_safe(y
, sizeof(y
), le64toh(f
->header
->tail_entry_realtime
)),
2838 format_timespan(z
, sizeof(z
), le64toh(f
->header
->tail_entry_monotonic
), USEC_PER_MSEC
),
2839 le64toh(f
->header
->n_objects
),
2840 le64toh(f
->header
->n_entries
));
2842 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
2843 printf("Data Objects: %"PRIu64
"\n"
2844 "Data Hash Table Fill: %.1f%%\n",
2845 le64toh(f
->header
->n_data
),
2846 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))));
2848 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
2849 printf("Field Objects: %"PRIu64
"\n"
2850 "Field Hash Table Fill: %.1f%%\n",
2851 le64toh(f
->header
->n_fields
),
2852 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))));
2854 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
))
2855 printf("Tag Objects: %"PRIu64
"\n",
2856 le64toh(f
->header
->n_tags
));
2857 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
2858 printf("Entry Array Objects: %"PRIu64
"\n",
2859 le64toh(f
->header
->n_entry_arrays
));
2861 if (fstat(f
->fd
, &st
) >= 0)
2862 printf("Disk usage: %s\n", format_bytes(bytes
, sizeof(bytes
), (uint64_t) st
.st_blocks
* 512ULL));
2865 static int journal_file_warn_btrfs(JournalFile
*f
) {
2871 /* Before we write anything, check if the COW logic is turned
2872 * off on btrfs. Given our write pattern that is quite
2873 * unfriendly to COW file systems this should greatly improve
2874 * performance on COW file systems, such as btrfs, at the
2875 * expense of data integrity features (which shouldn't be too
2876 * bad, given that we do our own checksumming). */
2878 r
= btrfs_is_filesystem(f
->fd
);
2880 return log_warning_errno(r
, "Failed to determine if journal is on btrfs: %m");
2884 r
= read_attr_fd(f
->fd
, &attrs
);
2886 return log_warning_errno(r
, "Failed to read file attributes: %m");
2888 if (attrs
& FS_NOCOW_FL
) {
2889 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
2893 log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
2894 "This is likely to slow down journal access substantially, please consider turning "
2895 "off the copy-on-write file attribute on the journal directory, using chattr +C.", f
->path
);
2900 int journal_file_open(
2906 JournalMetrics
*metrics
,
2907 MMapCache
*mmap_cache
,
2908 Set
*deferred_closes
,
2909 JournalFile
*template,
2910 JournalFile
**ret
) {
2912 bool newly_created
= false;
2920 if ((flags
& O_ACCMODE
) != O_RDONLY
&&
2921 (flags
& O_ACCMODE
) != O_RDWR
)
2924 if (!endswith(fname
, ".journal") &&
2925 !endswith(fname
, ".journal~"))
2928 f
= new0(JournalFile
, 1);
2936 f
->prot
= prot_from_flags(flags
);
2937 f
->writable
= (flags
& O_ACCMODE
) != O_RDONLY
;
2938 #if defined(HAVE_LZ4)
2939 f
->compress_lz4
= compress
;
2940 #elif defined(HAVE_XZ)
2941 f
->compress_xz
= compress
;
2948 f
->mmap
= mmap_cache_ref(mmap_cache
);
2950 f
->mmap
= mmap_cache_new();
2957 f
->path
= strdup(fname
);
2963 f
->chain_cache
= ordered_hashmap_new(&uint64_hash_ops
);
2964 if (!f
->chain_cache
) {
2969 f
->fd
= open(f
->path
, f
->flags
|O_CLOEXEC
, f
->mode
);
2975 r
= journal_file_fstat(f
);
2979 if (f
->last_stat
.st_size
== 0 && f
->writable
) {
2981 (void) journal_file_warn_btrfs(f
);
2983 /* Let's attach the creation time to the journal file,
2984 * so that the vacuuming code knows the age of this
2985 * file even if the file might end up corrupted one
2986 * day... Ideally we'd just use the creation time many
2987 * file systems maintain for each file, but there is
2988 * currently no usable API to query this, hence let's
2989 * emulate this via extended attributes. If extended
2990 * attributes are not supported we'll just skip this,
2991 * and rely solely on mtime/atime/ctime of the file. */
2993 fd_setcrtime(f
->fd
, 0);
2996 /* Try to load the FSPRG state, and if we can't, then
2997 * just don't do sealing */
2999 r
= journal_file_fss_load(f
);
3005 r
= journal_file_init_header(f
, template);
3009 r
= journal_file_fstat(f
);
3013 newly_created
= true;
3016 if (f
->last_stat
.st_size
< (off_t
) HEADER_SIZE_MIN
) {
3021 r
= mmap_cache_get(f
->mmap
, f
->fd
, f
->prot
, CONTEXT_HEADER
, true, 0, PAGE_ALIGN(sizeof(Header
)), &f
->last_stat
, &h
);
3027 if (!newly_created
) {
3028 if (deferred_closes
)
3029 journal_file_close_set(deferred_closes
);
3031 r
= journal_file_verify_header(f
);
3037 if (!newly_created
&& f
->writable
) {
3038 r
= journal_file_fss_load(f
);
3046 journal_default_metrics(metrics
, f
->fd
);
3047 f
->metrics
= *metrics
;
3048 } else if (template)
3049 f
->metrics
= template->metrics
;
3051 r
= journal_file_refresh_header(f
);
3057 r
= journal_file_hmac_setup(f
);
3062 if (newly_created
) {
3063 r
= journal_file_setup_field_hash_table(f
);
3067 r
= journal_file_setup_data_hash_table(f
);
3072 r
= journal_file_append_first_tag(f
);
3078 if (mmap_cache_got_sigbus(f
->mmap
, f
->fd
)) {
3083 if (template && template->post_change_timer
) {
3084 r
= journal_file_enable_post_change_timer(
3086 sd_event_source_get_event(template->post_change_timer
),
3087 template->post_change_timer_period
);
3097 if (f
->fd
>= 0 && mmap_cache_got_sigbus(f
->mmap
, f
->fd
))
3100 (void) journal_file_close(f
);
3105 int journal_file_rotate(JournalFile
**f
, bool compress
, bool seal
, Set
*deferred_closes
) {
3106 _cleanup_free_
char *p
= NULL
;
3108 JournalFile
*old_file
, *new_file
= NULL
;
3116 if (!old_file
->writable
)
3119 if (!endswith(old_file
->path
, ".journal"))
3122 l
= strlen(old_file
->path
);
3123 r
= asprintf(&p
, "%.*s@" SD_ID128_FORMAT_STR
"-%016"PRIx64
"-%016"PRIx64
".journal",
3124 (int) l
- 8, old_file
->path
,
3125 SD_ID128_FORMAT_VAL(old_file
->header
->seqnum_id
),
3126 le64toh((*f
)->header
->head_entry_seqnum
),
3127 le64toh((*f
)->header
->head_entry_realtime
));
3131 /* Try to rename the file to the archived version. If the file
3132 * already was deleted, we'll get ENOENT, let's ignore that
3134 r
= rename(old_file
->path
, p
);
3135 if (r
< 0 && errno
!= ENOENT
)
3138 old_file
->header
->state
= STATE_ARCHIVED
;
3140 /* Currently, btrfs is not very good with out write patterns
3141 * and fragments heavily. Let's defrag our journal files when
3142 * we archive them */
3143 old_file
->defrag_on_close
= true;
3145 r
= journal_file_open(old_file
->path
, old_file
->flags
, old_file
->mode
, compress
, seal
, NULL
, old_file
->mmap
, deferred_closes
, old_file
, &new_file
);
3147 if (deferred_closes
&&
3148 set_put(deferred_closes
, old_file
) >= 0)
3149 (void) journal_file_set_offline(old_file
, false);
3151 (void) journal_file_close(old_file
);
3157 int journal_file_open_reliably(
3163 JournalMetrics
*metrics
,
3164 MMapCache
*mmap_cache
,
3165 Set
*deferred_closes
,
3166 JournalFile
*template,
3167 JournalFile
**ret
) {
3171 _cleanup_free_
char *p
= NULL
;
3173 r
= journal_file_open(fname
, flags
, mode
, compress
, seal
, metrics
, mmap_cache
, deferred_closes
, template, ret
);
3175 -EBADMSG
, /* corrupted */
3176 -ENODATA
, /* truncated */
3177 -EHOSTDOWN
, /* other machine */
3178 -EPROTONOSUPPORT
, /* incompatible feature */
3179 -EBUSY
, /* unclean shutdown */
3180 -ESHUTDOWN
, /* already archived */
3181 -EIO
, /* IO error, including SIGBUS on mmap */
3182 -EIDRM
/* File has been deleted */))
3185 if ((flags
& O_ACCMODE
) == O_RDONLY
)
3188 if (!(flags
& O_CREAT
))
3191 if (!endswith(fname
, ".journal"))
3194 /* The file is corrupted. Rotate it away and try it again (but only once) */
3197 if (asprintf(&p
, "%.*s@%016"PRIx64
"-%016"PRIx64
".journal~",
3199 now(CLOCK_REALTIME
),
3203 if (rename(fname
, p
) < 0)
3206 /* btrfs doesn't cope well with our write pattern and
3207 * fragments heavily. Let's defrag all files we rotate */
3209 (void) chattr_path(p
, false, FS_NOCOW_FL
);
3210 (void) btrfs_defrag(p
);
3212 log_warning_errno(r
, "File %s corrupted or uncleanly shut down, renaming and replacing.", fname
);
3214 return journal_file_open(fname
, flags
, mode
, compress
, seal
, metrics
, mmap_cache
, deferred_closes
, template, ret
);
3217 int journal_file_copy_entry(JournalFile
*from
, JournalFile
*to
, Object
*o
, uint64_t p
, uint64_t *seqnum
, Object
**ret
, uint64_t *offset
) {
3219 uint64_t q
, xor_hash
= 0;
3232 ts
.monotonic
= le64toh(o
->entry
.monotonic
);
3233 ts
.realtime
= le64toh(o
->entry
.realtime
);
3235 n
= journal_file_entry_n_items(o
);
3236 /* alloca() can't take 0, hence let's allocate at least one */
3237 items
= alloca(sizeof(EntryItem
) * MAX(1u, n
));
3239 for (i
= 0; i
< n
; i
++) {
3246 q
= le64toh(o
->entry
.items
[i
].object_offset
);
3247 le_hash
= o
->entry
.items
[i
].hash
;
3249 r
= journal_file_move_to_object(from
, OBJECT_DATA
, q
, &o
);
3253 if (le_hash
!= o
->data
.hash
)
3256 l
= le64toh(o
->object
.size
) - offsetof(Object
, data
.payload
);
3259 /* We hit the limit on 32bit machines */
3260 if ((uint64_t) t
!= l
)
3263 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
3264 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
3267 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
3268 o
->data
.payload
, l
, &from
->compress_buffer
, &from
->compress_buffer_size
, &rsize
, 0);
3272 data
= from
->compress_buffer
;
3275 return -EPROTONOSUPPORT
;
3278 data
= o
->data
.payload
;
3280 r
= journal_file_append_data(to
, data
, l
, &u
, &h
);
3284 xor_hash
^= le64toh(u
->data
.hash
);
3285 items
[i
].object_offset
= htole64(h
);
3286 items
[i
].hash
= u
->data
.hash
;
3288 r
= journal_file_move_to_object(from
, OBJECT_ENTRY
, p
, &o
);
3293 r
= journal_file_append_entry_internal(to
, &ts
, xor_hash
, items
, n
, seqnum
, ret
, offset
);
3295 if (mmap_cache_got_sigbus(to
->mmap
, to
->fd
))
3301 void journal_reset_metrics(JournalMetrics
*m
) {
3304 /* Set everything to "pick automatic values". */
3306 *m
= (JournalMetrics
) {
3307 .min_use
= (uint64_t) -1,
3308 .max_use
= (uint64_t) -1,
3309 .min_size
= (uint64_t) -1,
3310 .max_size
= (uint64_t) -1,
3311 .keep_free
= (uint64_t) -1,
3312 .n_max_files
= (uint64_t) -1,
3316 void journal_default_metrics(JournalMetrics
*m
, int fd
) {
3317 char a
[FORMAT_BYTES_MAX
], b
[FORMAT_BYTES_MAX
], c
[FORMAT_BYTES_MAX
], d
[FORMAT_BYTES_MAX
], e
[FORMAT_BYTES_MAX
];
3324 if (fstatvfs(fd
, &ss
) >= 0)
3325 fs_size
= ss
.f_frsize
* ss
.f_blocks
;
3327 log_debug_errno(errno
, "Failed to detremine disk size: %m");
3331 if (m
->max_use
== (uint64_t) -1) {
3334 m
->max_use
= PAGE_ALIGN(fs_size
/ 10); /* 10% of file system size */
3336 if (m
->max_use
> DEFAULT_MAX_USE_UPPER
)
3337 m
->max_use
= DEFAULT_MAX_USE_UPPER
;
3339 if (m
->max_use
< DEFAULT_MAX_USE_LOWER
)
3340 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3342 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3344 m
->max_use
= PAGE_ALIGN(m
->max_use
);
3346 if (m
->max_use
!= 0 && m
->max_use
< JOURNAL_FILE_SIZE_MIN
*2)
3347 m
->max_use
= JOURNAL_FILE_SIZE_MIN
*2;
3350 if (m
->min_use
== (uint64_t) -1)
3351 m
->min_use
= DEFAULT_MIN_USE
;
3353 if (m
->min_use
> m
->max_use
)
3354 m
->min_use
= m
->max_use
;
3356 if (m
->max_size
== (uint64_t) -1) {
3357 m
->max_size
= PAGE_ALIGN(m
->max_use
/ 8); /* 8 chunks */
3359 if (m
->max_size
> DEFAULT_MAX_SIZE_UPPER
)
3360 m
->max_size
= DEFAULT_MAX_SIZE_UPPER
;
3362 m
->max_size
= PAGE_ALIGN(m
->max_size
);
3364 if (m
->max_size
!= 0) {
3365 if (m
->max_size
< JOURNAL_FILE_SIZE_MIN
)
3366 m
->max_size
= JOURNAL_FILE_SIZE_MIN
;
3368 if (m
->max_use
!= 0 && m
->max_size
*2 > m
->max_use
)
3369 m
->max_use
= m
->max_size
*2;
3372 if (m
->min_size
== (uint64_t) -1)
3373 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3375 m
->min_size
= PAGE_ALIGN(m
->min_size
);
3377 if (m
->min_size
< JOURNAL_FILE_SIZE_MIN
)
3378 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3380 if (m
->max_size
!= 0 && m
->min_size
> m
->max_size
)
3381 m
->max_size
= m
->min_size
;
3384 if (m
->keep_free
== (uint64_t) -1) {
3387 m
->keep_free
= PAGE_ALIGN(fs_size
* 3 / 20); /* 15% of file system size */
3389 if (m
->keep_free
> DEFAULT_KEEP_FREE_UPPER
)
3390 m
->keep_free
= DEFAULT_KEEP_FREE_UPPER
;
3393 m
->keep_free
= DEFAULT_KEEP_FREE
;
3396 if (m
->n_max_files
== (uint64_t) -1)
3397 m
->n_max_files
= DEFAULT_N_MAX_FILES
;
3399 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64
,
3400 format_bytes(a
, sizeof(a
), m
->min_use
),
3401 format_bytes(b
, sizeof(b
), m
->max_use
),
3402 format_bytes(c
, sizeof(c
), m
->max_size
),
3403 format_bytes(d
, sizeof(d
), m
->min_size
),
3404 format_bytes(e
, sizeof(e
), m
->keep_free
),
3408 int journal_file_get_cutoff_realtime_usec(JournalFile
*f
, usec_t
*from
, usec_t
*to
) {
3414 if (f
->header
->head_entry_realtime
== 0)
3417 *from
= le64toh(f
->header
->head_entry_realtime
);
3421 if (f
->header
->tail_entry_realtime
== 0)
3424 *to
= le64toh(f
->header
->tail_entry_realtime
);
3430 int journal_file_get_cutoff_monotonic_usec(JournalFile
*f
, sd_id128_t boot_id
, usec_t
*from
, usec_t
*to
) {
3438 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &p
);
3442 if (le64toh(o
->data
.n_entries
) <= 0)
3446 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, le64toh(o
->data
.entry_offset
), &o
);
3450 *from
= le64toh(o
->entry
.monotonic
);
3454 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
3458 r
= generic_array_get_plus_one(f
,
3459 le64toh(o
->data
.entry_offset
),
3460 le64toh(o
->data
.entry_array_offset
),
3461 le64toh(o
->data
.n_entries
)-1,
3466 *to
= le64toh(o
->entry
.monotonic
);
3472 bool journal_file_rotate_suggested(JournalFile
*f
, usec_t max_file_usec
) {
3476 /* If we gained new header fields we gained new features,
3477 * hence suggest a rotation */
3478 if (le64toh(f
->header
->header_size
) < sizeof(Header
)) {
3479 log_debug("%s uses an outdated header, suggesting rotation.", f
->path
);
3483 /* Let's check if the hash tables grew over a certain fill
3484 * level (75%, borrowing this value from Java's hash table
3485 * implementation), and if so suggest a rotation. To calculate
3486 * the fill level we need the n_data field, which only exists
3487 * in newer versions. */
3489 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3490 if (le64toh(f
->header
->n_data
) * 4ULL > (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3491 log_debug("Data hash table of %s has a fill level at %.1f (%"PRIu64
" of %"PRIu64
" items, %llu file size, %"PRIu64
" bytes per hash table item), suggesting rotation.",
3493 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))),
3494 le64toh(f
->header
->n_data
),
3495 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3496 (unsigned long long) f
->last_stat
.st_size
,
3497 f
->last_stat
.st_size
/ le64toh(f
->header
->n_data
));
3501 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3502 if (le64toh(f
->header
->n_fields
) * 4ULL > (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3503 log_debug("Field hash table of %s has a fill level at %.1f (%"PRIu64
" of %"PRIu64
" items), suggesting rotation.",
3505 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))),
3506 le64toh(f
->header
->n_fields
),
3507 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
));
3511 /* Are the data objects properly indexed by field objects? */
3512 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
3513 JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
3514 le64toh(f
->header
->n_data
) > 0 &&
3515 le64toh(f
->header
->n_fields
) == 0)
3518 if (max_file_usec
> 0) {
3521 h
= le64toh(f
->header
->head_entry_realtime
);
3522 t
= now(CLOCK_REALTIME
);
3524 if (h
> 0 && t
> h
+ max_file_usec
)