1 /* SPDX-License-Identifier: LGPL-2.1+ */
9 #include <sys/statvfs.h>
15 #include "alloc-util.h"
16 #include "btrfs-util.h"
17 #include "chattr-util.h"
20 #include "format-util.h"
22 #include "journal-authenticate.h"
23 #include "journal-def.h"
24 #include "journal-file.h"
26 #include "memory-util.h"
27 #include "path-util.h"
28 #include "random-util.h"
30 #include "sort-util.h"
31 #include "stat-util.h"
32 #include "string-util.h"
34 #include "xattr-util.h"
36 #define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
37 #define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
39 #define DEFAULT_COMPRESS_THRESHOLD (512ULL)
40 #define MIN_COMPRESS_THRESHOLD (8ULL)
42 /* This is the minimum journal file size */
43 #define JOURNAL_FILE_SIZE_MIN (512 * 1024ULL) /* 512 KiB */
45 /* These are the lower and upper bounds if we deduce the max_use value
46 * from the file system size */
47 #define MAX_USE_LOWER (1 * 1024 * 1024ULL) /* 1 MiB */
48 #define MAX_USE_UPPER (4 * 1024 * 1024 * 1024ULL) /* 4 GiB */
50 /* Those are the lower and upper bounds for the minimal use limit,
51 * i.e. how much we'll use even if keep_free suggests otherwise. */
52 #define MIN_USE_LOW (1 * 1024 * 1024ULL) /* 1 MiB */
53 #define MIN_USE_HIGH (16 * 1024 * 1024ULL) /* 16 MiB */
55 /* This is the upper bound if we deduce max_size from max_use */
56 #define MAX_SIZE_UPPER (128 * 1024 * 1024ULL) /* 128 MiB */
58 /* This is the upper bound if we deduce the keep_free value from the
60 #define KEEP_FREE_UPPER (4 * 1024 * 1024 * 1024ULL) /* 4 GiB */
62 /* This is the keep_free value when we can't determine the system
64 #define DEFAULT_KEEP_FREE (1024 * 1024ULL) /* 1 MB */
66 /* This is the default maximum number of journal files to keep around. */
67 #define DEFAULT_N_MAX_FILES 100
69 /* n_data was the first entry we added after the initial file format design */
70 #define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
72 /* How many entries to keep in the entry array chain cache at max */
73 #define CHAIN_CACHE_MAX 20
75 /* How much to increase the journal file size at once each time we allocate something new. */
76 #define FILE_SIZE_INCREASE (8 * 1024 * 1024ULL) /* 8MB */
78 /* Reread fstat() of the file for detecting deletions at least this often */
79 #define LAST_STAT_REFRESH_USEC (5*USEC_PER_SEC)
81 /* The mmap context to use for the header we pick as one above the last defined typed */
82 #define CONTEXT_HEADER _OBJECT_TYPE_MAX
85 # pragma GCC diagnostic ignored "-Waddress-of-packed-member"
88 /* This may be called from a separate thread to prevent blocking the caller for the duration of fsync().
89 * As a result we use atomic operations on f->offline_state for inter-thread communications with
90 * journal_file_set_offline() and journal_file_set_online(). */
91 static void journal_file_set_offline_internal(JournalFile
*f
) {
97 switch (f
->offline_state
) {
99 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_CANCEL
, OFFLINE_DONE
))
103 case OFFLINE_AGAIN_FROM_SYNCING
:
104 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_SYNCING
, OFFLINE_SYNCING
))
108 case OFFLINE_AGAIN_FROM_OFFLINING
:
109 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_OFFLINING
, OFFLINE_SYNCING
))
113 case OFFLINE_SYNCING
:
116 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_OFFLINING
))
119 f
->header
->state
= f
->archive
? STATE_ARCHIVED
: STATE_OFFLINE
;
123 case OFFLINE_OFFLINING
:
124 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_OFFLINING
, OFFLINE_DONE
))
131 log_debug("OFFLINE_JOINED unexpected offline state for journal_file_set_offline_internal()");
137 static void * journal_file_set_offline_thread(void *arg
) {
138 JournalFile
*f
= arg
;
140 (void) pthread_setname_np(pthread_self(), "journal-offline");
142 journal_file_set_offline_internal(f
);
147 static int journal_file_set_offline_thread_join(JournalFile
*f
) {
152 if (f
->offline_state
== OFFLINE_JOINED
)
155 r
= pthread_join(f
->offline_thread
, NULL
);
159 f
->offline_state
= OFFLINE_JOINED
;
161 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
167 /* Trigger a restart if the offline thread is mid-flight in a restartable state. */
168 static bool journal_file_set_offline_try_restart(JournalFile
*f
) {
170 switch (f
->offline_state
) {
171 case OFFLINE_AGAIN_FROM_SYNCING
:
172 case OFFLINE_AGAIN_FROM_OFFLINING
:
176 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_CANCEL
, OFFLINE_AGAIN_FROM_SYNCING
))
180 case OFFLINE_SYNCING
:
181 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_AGAIN_FROM_SYNCING
))
185 case OFFLINE_OFFLINING
:
186 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_OFFLINING
, OFFLINE_AGAIN_FROM_OFFLINING
))
196 /* Sets a journal offline.
198 * If wait is false then an offline is dispatched in a separate thread for a
199 * subsequent journal_file_set_offline() or journal_file_set_online() of the
200 * same journal to synchronize with.
202 * If wait is true, then either an existing offline thread will be restarted
203 * and joined, or if none exists the offline is simply performed in this
204 * context without involving another thread.
206 int journal_file_set_offline(JournalFile
*f
, bool wait
) {
215 if (f
->fd
< 0 || !f
->header
)
218 /* An offlining journal is implicitly online and may modify f->header->state,
219 * we must also join any potentially lingering offline thread when not online. */
220 if (!journal_file_is_offlining(f
) && f
->header
->state
!= STATE_ONLINE
)
221 return journal_file_set_offline_thread_join(f
);
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 sigset_t ss
, saved_ss
;
243 assert_se(sigfillset(&ss
) >= 0);
244 /* Don't block SIGBUS since the offlining thread accesses a memory mapped file.
245 * Asynchronous SIGBUS signals can safely be handled by either thread. */
246 assert_se(sigdelset(&ss
, SIGBUS
) >= 0);
248 r
= pthread_sigmask(SIG_BLOCK
, &ss
, &saved_ss
);
252 r
= pthread_create(&f
->offline_thread
, NULL
, journal_file_set_offline_thread
, f
);
254 k
= pthread_sigmask(SIG_SETMASK
, &saved_ss
, NULL
);
256 f
->offline_state
= OFFLINE_JOINED
;
266 static int journal_file_set_online(JournalFile
*f
) {
274 if (f
->fd
< 0 || !f
->header
)
278 switch (f
->offline_state
) {
280 /* No offline thread, no need to wait. */
284 case OFFLINE_SYNCING
:
285 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_CANCEL
))
287 /* Canceled syncing prior to offlining, no need to wait. */
291 case OFFLINE_AGAIN_FROM_SYNCING
:
292 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_SYNCING
, OFFLINE_CANCEL
))
294 /* Canceled restart from syncing, no need to wait. */
298 case OFFLINE_AGAIN_FROM_OFFLINING
:
299 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_OFFLINING
, OFFLINE_CANCEL
))
301 /* Canceled restart from offlining, must wait for offlining to complete however. */
306 r
= journal_file_set_offline_thread_join(f
);
316 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
319 switch (f
->header
->state
) {
324 f
->header
->state
= STATE_ONLINE
;
333 bool journal_file_is_offlining(JournalFile
*f
) {
336 __sync_synchronize();
338 if (IN_SET(f
->offline_state
, OFFLINE_DONE
, OFFLINE_JOINED
))
344 JournalFile
* journal_file_close(JournalFile
*f
) {
349 /* Write the final tag */
350 if (f
->seal
&& f
->writable
) {
353 r
= journal_file_append_tag(f
);
355 log_error_errno(r
, "Failed to append tag when closing journal: %m");
359 if (f
->post_change_timer
) {
360 if (sd_event_source_get_enabled(f
->post_change_timer
, NULL
) > 0)
361 journal_file_post_change(f
);
363 sd_event_source_disable_unref(f
->post_change_timer
);
366 journal_file_set_offline(f
, true);
368 if (f
->mmap
&& f
->cache_fd
)
369 mmap_cache_free_fd(f
->mmap
, f
->cache_fd
);
371 if (f
->fd
>= 0 && f
->defrag_on_close
) {
373 /* Be friendly to btrfs: turn COW back on again now,
374 * and defragment the file. We won't write to the file
375 * ever again, hence remove all fragmentation, and
376 * reenable all the good bits COW usually provides
377 * (such as data checksumming). */
379 (void) chattr_fd(f
->fd
, 0, FS_NOCOW_FL
, NULL
);
380 (void) btrfs_defrag_fd(f
->fd
);
387 mmap_cache_unref(f
->mmap
);
389 ordered_hashmap_free_free(f
->chain_cache
);
391 #if HAVE_XZ || HAVE_LZ4
392 free(f
->compress_buffer
);
397 munmap(f
->fss_file
, PAGE_ALIGN(f
->fss_file_size
));
399 free(f
->fsprg_state
);
404 gcry_md_close(f
->hmac
);
410 static int journal_file_init_header(JournalFile
*f
, JournalFile
*template) {
417 memcpy(h
.signature
, HEADER_SIGNATURE
, 8);
418 h
.header_size
= htole64(ALIGN64(sizeof(h
)));
420 h
.incompatible_flags
|= htole32(
421 f
->compress_xz
* HEADER_INCOMPATIBLE_COMPRESSED_XZ
|
422 f
->compress_lz4
* HEADER_INCOMPATIBLE_COMPRESSED_LZ4
);
424 h
.compatible_flags
= htole32(
425 f
->seal
* HEADER_COMPATIBLE_SEALED
);
427 r
= sd_id128_randomize(&h
.file_id
);
432 h
.seqnum_id
= template->header
->seqnum_id
;
433 h
.tail_entry_seqnum
= template->header
->tail_entry_seqnum
;
435 h
.seqnum_id
= h
.file_id
;
437 k
= pwrite(f
->fd
, &h
, sizeof(h
), 0);
447 static int journal_file_refresh_header(JournalFile
*f
) {
454 r
= sd_id128_get_machine(&f
->header
->machine_id
);
455 if (IN_SET(r
, -ENOENT
, -ENOMEDIUM
))
456 /* We don't have a machine-id, let's continue without */
457 zero(f
->header
->machine_id
);
461 r
= sd_id128_get_boot(&boot_id
);
465 f
->header
->boot_id
= boot_id
;
467 r
= journal_file_set_online(f
);
469 /* Sync the online state to disk */
472 /* We likely just created a new file, also sync the directory this file is located in. */
473 (void) fsync_directory_of_file(f
->fd
);
478 static bool warn_wrong_flags(const JournalFile
*f
, bool compatible
) {
479 const uint32_t any
= compatible
? HEADER_COMPATIBLE_ANY
: HEADER_INCOMPATIBLE_ANY
,
480 supported
= compatible
? HEADER_COMPATIBLE_SUPPORTED
: HEADER_INCOMPATIBLE_SUPPORTED
;
481 const char *type
= compatible
? "compatible" : "incompatible";
484 flags
= le32toh(compatible
? f
->header
->compatible_flags
: f
->header
->incompatible_flags
);
486 if (flags
& ~supported
) {
488 log_debug("Journal file %s has unknown %s flags 0x%"PRIx32
,
489 f
->path
, type
, flags
& ~any
);
490 flags
= (flags
& any
) & ~supported
;
494 _cleanup_free_
char *t
= NULL
;
496 if (compatible
&& (flags
& HEADER_COMPATIBLE_SEALED
))
497 strv
[n
++] = "sealed";
498 if (!compatible
&& (flags
& HEADER_INCOMPATIBLE_COMPRESSED_XZ
))
499 strv
[n
++] = "xz-compressed";
500 if (!compatible
&& (flags
& HEADER_INCOMPATIBLE_COMPRESSED_LZ4
))
501 strv
[n
++] = "lz4-compressed";
503 assert(n
< ELEMENTSOF(strv
));
505 t
= strv_join((char**) strv
, ", ");
506 log_debug("Journal file %s uses %s %s %s disabled at compilation time.",
507 f
->path
, type
, n
> 1 ? "flags" : "flag", strnull(t
));
515 static int journal_file_verify_header(JournalFile
*f
) {
516 uint64_t arena_size
, header_size
;
521 if (memcmp(f
->header
->signature
, HEADER_SIGNATURE
, 8))
524 /* In both read and write mode we refuse to open files with incompatible
525 * flags we don't know. */
526 if (warn_wrong_flags(f
, false))
527 return -EPROTONOSUPPORT
;
529 /* When open for writing we refuse to open files with compatible flags, too. */
530 if (f
->writable
&& warn_wrong_flags(f
, true))
531 return -EPROTONOSUPPORT
;
533 if (f
->header
->state
>= _STATE_MAX
)
536 header_size
= le64toh(READ_NOW(f
->header
->header_size
));
538 /* The first addition was n_data, so check that we are at least this large */
539 if (header_size
< HEADER_SIZE_MIN
)
542 if (JOURNAL_HEADER_SEALED(f
->header
) && !JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
545 arena_size
= le64toh(READ_NOW(f
->header
->arena_size
));
547 if (UINT64_MAX
- header_size
< arena_size
|| header_size
+ arena_size
> (uint64_t) f
->last_stat
.st_size
)
550 if (le64toh(f
->header
->tail_object_offset
) > header_size
+ arena_size
)
553 if (!VALID64(le64toh(f
->header
->data_hash_table_offset
)) ||
554 !VALID64(le64toh(f
->header
->field_hash_table_offset
)) ||
555 !VALID64(le64toh(f
->header
->tail_object_offset
)) ||
556 !VALID64(le64toh(f
->header
->entry_array_offset
)))
560 sd_id128_t machine_id
;
564 r
= sd_id128_get_machine(&machine_id
);
568 if (!sd_id128_equal(machine_id
, f
->header
->machine_id
))
571 state
= f
->header
->state
;
573 if (state
== STATE_ARCHIVED
)
574 return -ESHUTDOWN
; /* Already archived */
575 else if (state
== STATE_ONLINE
)
576 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
),
577 "Journal file %s is already online. Assuming unclean closing.",
579 else if (state
!= STATE_OFFLINE
)
580 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
),
581 "Journal file %s has unknown state %i.",
584 if (f
->header
->field_hash_table_size
== 0 || f
->header
->data_hash_table_size
== 0)
587 /* Don't permit appending to files from the future. Because otherwise the realtime timestamps wouldn't
588 * be strictly ordered in the entries in the file anymore, and we can't have that since it breaks
590 if (le64toh(f
->header
->tail_entry_realtime
) > now(CLOCK_REALTIME
))
591 return log_debug_errno(SYNTHETIC_ERRNO(ETXTBSY
),
592 "Journal file %s is from the future, refusing to append new data to it that'd be older.",
596 f
->compress_xz
= JOURNAL_HEADER_COMPRESSED_XZ(f
->header
);
597 f
->compress_lz4
= JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
);
599 f
->seal
= JOURNAL_HEADER_SEALED(f
->header
);
604 int journal_file_fstat(JournalFile
*f
) {
610 if (fstat(f
->fd
, &f
->last_stat
) < 0)
613 f
->last_stat_usec
= now(CLOCK_MONOTONIC
);
615 /* Refuse dealing with with files that aren't regular */
616 r
= stat_verify_regular(&f
->last_stat
);
620 /* Refuse appending to files that are already deleted */
621 if (f
->last_stat
.st_nlink
<= 0)
627 static int journal_file_allocate(JournalFile
*f
, uint64_t offset
, uint64_t size
) {
628 uint64_t old_size
, new_size
, old_header_size
, old_arena_size
;
634 /* We assume that this file is not sparse, and we know that for sure, since we always call
635 * posix_fallocate() ourselves */
637 if (size
> PAGE_ALIGN_DOWN(UINT64_MAX
) - offset
)
640 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
643 old_header_size
= le64toh(READ_NOW(f
->header
->header_size
));
644 old_arena_size
= le64toh(READ_NOW(f
->header
->arena_size
));
645 if (old_arena_size
> PAGE_ALIGN_DOWN(UINT64_MAX
) - old_header_size
)
648 old_size
= old_header_size
+ old_arena_size
;
650 new_size
= MAX(PAGE_ALIGN(offset
+ size
), old_header_size
);
652 if (new_size
<= old_size
) {
654 /* We already pre-allocated enough space, but before
655 * we write to it, let's check with fstat() if the
656 * file got deleted, in order make sure we don't throw
657 * away the data immediately. Don't check fstat() for
658 * all writes though, but only once ever 10s. */
660 if (f
->last_stat_usec
+ LAST_STAT_REFRESH_USEC
> now(CLOCK_MONOTONIC
))
663 return journal_file_fstat(f
);
666 /* Allocate more space. */
668 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
671 if (new_size
> f
->metrics
.min_size
&& f
->metrics
.keep_free
> 0) {
674 if (fstatvfs(f
->fd
, &svfs
) >= 0) {
677 available
= LESS_BY((uint64_t) svfs
.f_bfree
* (uint64_t) svfs
.f_bsize
, f
->metrics
.keep_free
);
679 if (new_size
- old_size
> available
)
684 /* Increase by larger blocks at once */
685 new_size
= DIV_ROUND_UP(new_size
, FILE_SIZE_INCREASE
) * FILE_SIZE_INCREASE
;
686 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
687 new_size
= f
->metrics
.max_size
;
689 /* Note that the glibc fallocate() fallback is very
690 inefficient, hence we try to minimize the allocation area
692 r
= posix_fallocate(f
->fd
, old_size
, new_size
- old_size
);
696 f
->header
->arena_size
= htole64(new_size
- old_header_size
);
698 return journal_file_fstat(f
);
701 static unsigned type_to_context(ObjectType type
) {
702 /* One context for each type, plus one catch-all for the rest */
703 assert_cc(_OBJECT_TYPE_MAX
<= MMAP_CACHE_MAX_CONTEXTS
);
704 assert_cc(CONTEXT_HEADER
< MMAP_CACHE_MAX_CONTEXTS
);
705 return type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
? type
: 0;
708 static int journal_file_move_to(
725 if (size
> UINT64_MAX
- offset
)
728 /* Avoid SIGBUS on invalid accesses */
729 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
) {
730 /* Hmm, out of range? Let's refresh the fstat() data
731 * first, before we trust that check. */
733 r
= journal_file_fstat(f
);
737 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
)
738 return -EADDRNOTAVAIL
;
741 return mmap_cache_get(f
->mmap
, f
->cache_fd
, f
->prot
, type_to_context(type
), keep_always
, offset
, size
, &f
->last_stat
, ret
, ret_size
);
744 static uint64_t minimum_header_size(Object
*o
) {
746 static const uint64_t table
[] = {
747 [OBJECT_DATA
] = sizeof(DataObject
),
748 [OBJECT_FIELD
] = sizeof(FieldObject
),
749 [OBJECT_ENTRY
] = sizeof(EntryObject
),
750 [OBJECT_DATA_HASH_TABLE
] = sizeof(HashTableObject
),
751 [OBJECT_FIELD_HASH_TABLE
] = sizeof(HashTableObject
),
752 [OBJECT_ENTRY_ARRAY
] = sizeof(EntryArrayObject
),
753 [OBJECT_TAG
] = sizeof(TagObject
),
756 if (o
->object
.type
>= ELEMENTSOF(table
) || table
[o
->object
.type
] <= 0)
757 return sizeof(ObjectHeader
);
759 return table
[o
->object
.type
];
762 /* Lightweight object checks. We want this to be fast, so that we won't
763 * slowdown every journal_file_move_to_object() call too much. */
764 static int journal_file_check_object(JournalFile
*f
, uint64_t offset
, Object
*o
) {
768 switch (o
->object
.type
) {
771 if ((le64toh(o
->data
.entry_offset
) == 0) ^ (le64toh(o
->data
.n_entries
) == 0))
772 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
773 "Bad n_entries: %" PRIu64
": %" PRIu64
,
774 le64toh(o
->data
.n_entries
),
777 if (le64toh(o
->object
.size
) <= offsetof(DataObject
, payload
))
778 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
779 "Bad object size (<= %zu): %" PRIu64
": %" PRIu64
,
780 offsetof(DataObject
, payload
),
781 le64toh(o
->object
.size
),
784 if (!VALID64(le64toh(o
->data
.next_hash_offset
)) ||
785 !VALID64(le64toh(o
->data
.next_field_offset
)) ||
786 !VALID64(le64toh(o
->data
.entry_offset
)) ||
787 !VALID64(le64toh(o
->data
.entry_array_offset
)))
788 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
789 "Invalid offset, next_hash_offset=" OFSfmt
", next_field_offset=" OFSfmt
", entry_offset=" OFSfmt
", entry_array_offset=" OFSfmt
": %" PRIu64
,
790 le64toh(o
->data
.next_hash_offset
),
791 le64toh(o
->data
.next_field_offset
),
792 le64toh(o
->data
.entry_offset
),
793 le64toh(o
->data
.entry_array_offset
),
799 if (le64toh(o
->object
.size
) <= offsetof(FieldObject
, payload
))
800 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
801 "Bad field size (<= %zu): %" PRIu64
": %" PRIu64
,
802 offsetof(FieldObject
, payload
),
803 le64toh(o
->object
.size
),
806 if (!VALID64(le64toh(o
->field
.next_hash_offset
)) ||
807 !VALID64(le64toh(o
->field
.head_data_offset
)))
808 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
809 "Invalid offset, next_hash_offset=" OFSfmt
", head_data_offset=" OFSfmt
": %" PRIu64
,
810 le64toh(o
->field
.next_hash_offset
),
811 le64toh(o
->field
.head_data_offset
),
818 sz
= le64toh(READ_NOW(o
->object
.size
));
819 if (sz
< offsetof(EntryObject
, items
) ||
820 (sz
- offsetof(EntryObject
, items
)) % sizeof(EntryItem
) != 0)
821 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
822 "Bad entry size (<= %zu): %" PRIu64
": %" PRIu64
,
823 offsetof(EntryObject
, items
),
827 if ((sz
- offsetof(EntryObject
, items
)) / sizeof(EntryItem
) <= 0)
828 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
829 "Invalid number items in entry: %" PRIu64
": %" PRIu64
,
830 (sz
- offsetof(EntryObject
, items
)) / sizeof(EntryItem
),
833 if (le64toh(o
->entry
.seqnum
) <= 0)
834 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
835 "Invalid entry seqnum: %" PRIx64
": %" PRIu64
,
836 le64toh(o
->entry
.seqnum
),
839 if (!VALID_REALTIME(le64toh(o
->entry
.realtime
)))
840 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
841 "Invalid entry realtime timestamp: %" PRIu64
": %" PRIu64
,
842 le64toh(o
->entry
.realtime
),
845 if (!VALID_MONOTONIC(le64toh(o
->entry
.monotonic
)))
846 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
847 "Invalid entry monotonic timestamp: %" PRIu64
": %" PRIu64
,
848 le64toh(o
->entry
.monotonic
),
854 case OBJECT_DATA_HASH_TABLE
:
855 case OBJECT_FIELD_HASH_TABLE
: {
858 sz
= le64toh(READ_NOW(o
->object
.size
));
859 if (sz
< offsetof(HashTableObject
, items
) ||
860 (sz
- offsetof(HashTableObject
, items
)) % sizeof(HashItem
) != 0 ||
861 (sz
- offsetof(HashTableObject
, items
)) / sizeof(HashItem
) <= 0)
862 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
863 "Invalid %s hash table size: %" PRIu64
": %" PRIu64
,
864 o
->object
.type
== OBJECT_DATA_HASH_TABLE
? "data" : "field",
871 case OBJECT_ENTRY_ARRAY
: {
874 sz
= le64toh(READ_NOW(o
->object
.size
));
875 if (sz
< offsetof(EntryArrayObject
, items
) ||
876 (sz
- offsetof(EntryArrayObject
, items
)) % sizeof(le64_t
) != 0 ||
877 (sz
- offsetof(EntryArrayObject
, items
)) / sizeof(le64_t
) <= 0)
878 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
879 "Invalid object entry array size: %" PRIu64
": %" PRIu64
,
883 if (!VALID64(le64toh(o
->entry_array
.next_entry_array_offset
)))
884 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
885 "Invalid object entry array next_entry_array_offset: " OFSfmt
": %" PRIu64
,
886 le64toh(o
->entry_array
.next_entry_array_offset
),
893 if (le64toh(o
->object
.size
) != sizeof(TagObject
))
894 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
895 "Invalid object tag size: %" PRIu64
": %" PRIu64
,
896 le64toh(o
->object
.size
),
899 if (!VALID_EPOCH(le64toh(o
->tag
.epoch
)))
900 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
901 "Invalid object tag epoch: %" PRIu64
": %" PRIu64
,
902 le64toh(o
->tag
.epoch
), offset
);
910 int journal_file_move_to_object(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
**ret
) {
920 /* Objects may only be located at multiple of 64 bit */
921 if (!VALID64(offset
))
922 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
923 "Attempt to move to object at non-64bit boundary: %" PRIu64
,
926 /* Object may not be located in the file header */
927 if (offset
< le64toh(f
->header
->header_size
))
928 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
929 "Attempt to move to object located in file header: %" PRIu64
,
932 r
= journal_file_move_to(f
, type
, false, offset
, sizeof(ObjectHeader
), &t
, &tsize
);
937 s
= le64toh(READ_NOW(o
->object
.size
));
940 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
941 "Attempt to move to uninitialized object: %" PRIu64
,
943 if (s
< sizeof(ObjectHeader
))
944 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
945 "Attempt to move to overly short object: %" PRIu64
,
948 if (o
->object
.type
<= OBJECT_UNUSED
)
949 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
950 "Attempt to move to object with invalid type: %" PRIu64
,
953 if (s
< minimum_header_size(o
))
954 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
955 "Attempt to move to truncated object: %" PRIu64
,
958 if (type
> OBJECT_UNUSED
&& o
->object
.type
!= type
)
959 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
960 "Attempt to move to object of unexpected type: %" PRIu64
,
964 r
= journal_file_move_to(f
, type
, false, offset
, s
, &t
, NULL
);
971 r
= journal_file_check_object(f
, offset
, o
);
979 static uint64_t journal_file_entry_seqnum(JournalFile
*f
, uint64_t *seqnum
) {
985 r
= le64toh(f
->header
->tail_entry_seqnum
) + 1;
988 /* If an external seqnum counter was passed, we update
989 * both the local and the external one, and set it to
990 * the maximum of both */
998 f
->header
->tail_entry_seqnum
= htole64(r
);
1000 if (f
->header
->head_entry_seqnum
== 0)
1001 f
->header
->head_entry_seqnum
= htole64(r
);
1006 int journal_file_append_object(JournalFile
*f
, ObjectType type
, uint64_t size
, Object
**ret
, uint64_t *offset
) {
1014 assert(type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
);
1015 assert(size
>= sizeof(ObjectHeader
));
1019 r
= journal_file_set_online(f
);
1023 p
= le64toh(f
->header
->tail_object_offset
);
1025 p
= le64toh(f
->header
->header_size
);
1029 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &tail
);
1033 sz
= le64toh(READ_NOW(tail
->object
.size
));
1034 if (sz
> UINT64_MAX
- sizeof(uint64_t) + 1)
1038 if (p
> UINT64_MAX
- sz
)
1044 r
= journal_file_allocate(f
, p
, size
);
1048 r
= journal_file_move_to(f
, type
, false, p
, size
, &t
, NULL
);
1053 o
->object
= (ObjectHeader
) {
1055 .size
= htole64(size
),
1058 f
->header
->tail_object_offset
= htole64(p
);
1059 f
->header
->n_objects
= htole64(le64toh(f
->header
->n_objects
) + 1);
1067 static int journal_file_setup_data_hash_table(JournalFile
*f
) {
1075 /* We estimate that we need 1 hash table entry per 768 bytes
1076 of journal file and we want to make sure we never get
1077 beyond 75% fill level. Calculate the hash table size for
1078 the maximum file size based on these metrics. */
1080 s
= (f
->metrics
.max_size
* 4 / 768 / 3) * sizeof(HashItem
);
1081 if (s
< DEFAULT_DATA_HASH_TABLE_SIZE
)
1082 s
= DEFAULT_DATA_HASH_TABLE_SIZE
;
1084 log_debug("Reserving %"PRIu64
" entries in hash table.", s
/ sizeof(HashItem
));
1086 r
= journal_file_append_object(f
,
1087 OBJECT_DATA_HASH_TABLE
,
1088 offsetof(Object
, hash_table
.items
) + s
,
1093 memzero(o
->hash_table
.items
, s
);
1095 f
->header
->data_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1096 f
->header
->data_hash_table_size
= htole64(s
);
1101 static int journal_file_setup_field_hash_table(JournalFile
*f
) {
1109 /* We use a fixed size hash table for the fields as this
1110 * number should grow very slowly only */
1112 s
= DEFAULT_FIELD_HASH_TABLE_SIZE
;
1113 r
= journal_file_append_object(f
,
1114 OBJECT_FIELD_HASH_TABLE
,
1115 offsetof(Object
, hash_table
.items
) + s
,
1120 memzero(o
->hash_table
.items
, s
);
1122 f
->header
->field_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1123 f
->header
->field_hash_table_size
= htole64(s
);
1128 int journal_file_map_data_hash_table(JournalFile
*f
) {
1136 if (f
->data_hash_table
)
1139 p
= le64toh(f
->header
->data_hash_table_offset
);
1140 s
= le64toh(f
->header
->data_hash_table_size
);
1142 r
= journal_file_move_to(f
,
1143 OBJECT_DATA_HASH_TABLE
,
1150 f
->data_hash_table
= t
;
1154 int journal_file_map_field_hash_table(JournalFile
*f
) {
1162 if (f
->field_hash_table
)
1165 p
= le64toh(f
->header
->field_hash_table_offset
);
1166 s
= le64toh(f
->header
->field_hash_table_size
);
1168 r
= journal_file_move_to(f
,
1169 OBJECT_FIELD_HASH_TABLE
,
1176 f
->field_hash_table
= t
;
1180 static int journal_file_link_field(
1191 assert(f
->field_hash_table
);
1195 if (o
->object
.type
!= OBJECT_FIELD
)
1198 m
= le64toh(READ_NOW(f
->header
->field_hash_table_size
)) / sizeof(HashItem
);
1202 /* This might alter the window we are looking at */
1203 o
->field
.next_hash_offset
= o
->field
.head_data_offset
= 0;
1206 p
= le64toh(f
->field_hash_table
[h
].tail_hash_offset
);
1208 f
->field_hash_table
[h
].head_hash_offset
= htole64(offset
);
1210 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1214 o
->field
.next_hash_offset
= htole64(offset
);
1217 f
->field_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1219 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
1220 f
->header
->n_fields
= htole64(le64toh(f
->header
->n_fields
) + 1);
1225 static int journal_file_link_data(
1236 assert(f
->data_hash_table
);
1240 if (o
->object
.type
!= OBJECT_DATA
)
1243 m
= le64toh(READ_NOW(f
->header
->data_hash_table_size
)) / sizeof(HashItem
);
1247 /* This might alter the window we are looking at */
1248 o
->data
.next_hash_offset
= o
->data
.next_field_offset
= 0;
1249 o
->data
.entry_offset
= o
->data
.entry_array_offset
= 0;
1250 o
->data
.n_entries
= 0;
1253 p
= le64toh(f
->data_hash_table
[h
].tail_hash_offset
);
1255 /* Only entry in the hash table is easy */
1256 f
->data_hash_table
[h
].head_hash_offset
= htole64(offset
);
1258 /* Move back to the previous data object, to patch in
1261 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1265 o
->data
.next_hash_offset
= htole64(offset
);
1268 f
->data_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1270 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
1271 f
->header
->n_data
= htole64(le64toh(f
->header
->n_data
) + 1);
1276 int journal_file_find_field_object_with_hash(
1278 const void *field
, uint64_t size
, uint64_t hash
,
1279 Object
**ret
, uint64_t *offset
) {
1281 uint64_t p
, osize
, h
, m
;
1286 assert(field
&& size
> 0);
1288 /* If the field hash table is empty, we can't find anything */
1289 if (le64toh(f
->header
->field_hash_table_size
) <= 0)
1292 /* Map the field hash table, if it isn't mapped yet. */
1293 r
= journal_file_map_field_hash_table(f
);
1297 osize
= offsetof(Object
, field
.payload
) + size
;
1299 m
= le64toh(READ_NOW(f
->header
->field_hash_table_size
)) / sizeof(HashItem
);
1304 p
= le64toh(f
->field_hash_table
[h
].head_hash_offset
);
1309 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1313 if (le64toh(o
->field
.hash
) == hash
&&
1314 le64toh(o
->object
.size
) == osize
&&
1315 memcmp(o
->field
.payload
, field
, size
) == 0) {
1325 p
= le64toh(o
->field
.next_hash_offset
);
1331 int journal_file_find_field_object(
1333 const void *field
, uint64_t size
,
1334 Object
**ret
, uint64_t *offset
) {
1339 assert(field
&& size
> 0);
1341 hash
= hash64(field
, size
);
1343 return journal_file_find_field_object_with_hash(f
,
1348 int journal_file_find_data_object_with_hash(
1350 const void *data
, uint64_t size
, uint64_t hash
,
1351 Object
**ret
, uint64_t *offset
) {
1353 uint64_t p
, osize
, h
, m
;
1358 assert(data
|| size
== 0);
1360 /* If there's no data hash table, then there's no entry. */
1361 if (le64toh(f
->header
->data_hash_table_size
) <= 0)
1364 /* Map the data hash table, if it isn't mapped yet. */
1365 r
= journal_file_map_data_hash_table(f
);
1369 osize
= offsetof(Object
, data
.payload
) + size
;
1371 m
= le64toh(READ_NOW(f
->header
->data_hash_table_size
)) / sizeof(HashItem
);
1376 p
= le64toh(f
->data_hash_table
[h
].head_hash_offset
);
1381 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1385 if (le64toh(o
->data
.hash
) != hash
)
1388 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
1389 #if HAVE_XZ || HAVE_LZ4
1393 l
= le64toh(READ_NOW(o
->object
.size
));
1394 if (l
<= offsetof(Object
, data
.payload
))
1397 l
-= offsetof(Object
, data
.payload
);
1399 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
1400 o
->data
.payload
, l
, &f
->compress_buffer
, &f
->compress_buffer_size
, &rsize
, 0);
1404 if (rsize
== size
&&
1405 memcmp(f
->compress_buffer
, data
, size
) == 0) {
1416 return -EPROTONOSUPPORT
;
1418 } else if (le64toh(o
->object
.size
) == osize
&&
1419 memcmp(o
->data
.payload
, data
, size
) == 0) {
1431 p
= le64toh(o
->data
.next_hash_offset
);
1437 int journal_file_find_data_object(
1439 const void *data
, uint64_t size
,
1440 Object
**ret
, uint64_t *offset
) {
1445 assert(data
|| size
== 0);
1447 hash
= hash64(data
, size
);
1449 return journal_file_find_data_object_with_hash(f
,
1454 static int journal_file_append_field(
1456 const void *field
, uint64_t size
,
1457 Object
**ret
, uint64_t *offset
) {
1465 assert(field
&& size
> 0);
1467 hash
= hash64(field
, size
);
1469 r
= journal_file_find_field_object_with_hash(f
, field
, size
, hash
, &o
, &p
);
1483 osize
= offsetof(Object
, field
.payload
) + size
;
1484 r
= journal_file_append_object(f
, OBJECT_FIELD
, osize
, &o
, &p
);
1488 o
->field
.hash
= htole64(hash
);
1489 memcpy(o
->field
.payload
, field
, size
);
1491 r
= journal_file_link_field(f
, o
, p
, hash
);
1495 /* The linking might have altered the window, so let's
1496 * refresh our pointer */
1497 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1502 r
= journal_file_hmac_put_object(f
, OBJECT_FIELD
, o
, p
);
1516 static int journal_file_append_data(
1518 const void *data
, uint64_t size
,
1519 Object
**ret
, uint64_t *offset
) {
1524 int r
, compression
= 0;
1528 assert(data
|| size
== 0);
1530 hash
= hash64(data
, size
);
1532 r
= journal_file_find_data_object_with_hash(f
, data
, size
, hash
, &o
, &p
);
1546 osize
= offsetof(Object
, data
.payload
) + size
;
1547 r
= journal_file_append_object(f
, OBJECT_DATA
, osize
, &o
, &p
);
1551 o
->data
.hash
= htole64(hash
);
1553 #if HAVE_XZ || HAVE_LZ4
1554 if (JOURNAL_FILE_COMPRESS(f
) && size
>= f
->compress_threshold_bytes
) {
1557 compression
= compress_blob(data
, size
, o
->data
.payload
, size
- 1, &rsize
);
1559 if (compression
>= 0) {
1560 o
->object
.size
= htole64(offsetof(Object
, data
.payload
) + rsize
);
1561 o
->object
.flags
|= compression
;
1563 log_debug("Compressed data object %"PRIu64
" -> %zu using %s",
1564 size
, rsize
, object_compressed_to_string(compression
));
1566 /* Compression didn't work, we don't really care why, let's continue without compression */
1571 if (compression
== 0)
1572 memcpy_safe(o
->data
.payload
, data
, size
);
1574 r
= journal_file_link_data(f
, o
, p
, hash
);
1579 r
= journal_file_hmac_put_object(f
, OBJECT_DATA
, o
, p
);
1584 /* The linking might have altered the window, so let's
1585 * refresh our pointer */
1586 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1593 eq
= memchr(data
, '=', size
);
1594 if (eq
&& eq
> data
) {
1598 /* Create field object ... */
1599 r
= journal_file_append_field(f
, data
, (uint8_t*) eq
- (uint8_t*) data
, &fo
, &fp
);
1603 /* ... and link it in. */
1604 o
->data
.next_field_offset
= fo
->field
.head_data_offset
;
1605 fo
->field
.head_data_offset
= le64toh(p
);
1617 uint64_t journal_file_entry_n_items(Object
*o
) {
1621 if (o
->object
.type
!= OBJECT_ENTRY
)
1624 sz
= le64toh(READ_NOW(o
->object
.size
));
1625 if (sz
< offsetof(Object
, entry
.items
))
1628 return (sz
- offsetof(Object
, entry
.items
)) / sizeof(EntryItem
);
1631 uint64_t journal_file_entry_array_n_items(Object
*o
) {
1636 if (o
->object
.type
!= OBJECT_ENTRY_ARRAY
)
1639 sz
= le64toh(READ_NOW(o
->object
.size
));
1640 if (sz
< offsetof(Object
, entry_array
.items
))
1643 return (sz
- offsetof(Object
, entry_array
.items
)) / sizeof(uint64_t);
1646 uint64_t journal_file_hash_table_n_items(Object
*o
) {
1651 if (!IN_SET(o
->object
.type
, OBJECT_DATA_HASH_TABLE
, OBJECT_FIELD_HASH_TABLE
))
1654 sz
= le64toh(READ_NOW(o
->object
.size
));
1655 if (sz
< offsetof(Object
, hash_table
.items
))
1658 return (sz
- offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
);
1661 static int link_entry_into_array(JournalFile
*f
,
1666 uint64_t n
= 0, ap
= 0, q
, i
, a
, hidx
;
1675 a
= le64toh(*first
);
1676 i
= hidx
= le64toh(READ_NOW(*idx
));
1679 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
1683 n
= journal_file_entry_array_n_items(o
);
1685 o
->entry_array
.items
[i
] = htole64(p
);
1686 *idx
= htole64(hidx
+ 1);
1692 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
1703 r
= journal_file_append_object(f
, OBJECT_ENTRY_ARRAY
,
1704 offsetof(Object
, entry_array
.items
) + n
* sizeof(uint64_t),
1710 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY_ARRAY
, o
, q
);
1715 o
->entry_array
.items
[i
] = htole64(p
);
1718 *first
= htole64(q
);
1720 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, ap
, &o
);
1724 o
->entry_array
.next_entry_array_offset
= htole64(q
);
1727 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
1728 f
->header
->n_entry_arrays
= htole64(le64toh(f
->header
->n_entry_arrays
) + 1);
1730 *idx
= htole64(hidx
+ 1);
1735 static int link_entry_into_array_plus_one(JournalFile
*f
,
1750 hidx
= le64toh(READ_NOW(*idx
));
1751 if (hidx
== UINT64_MAX
)
1754 *extra
= htole64(p
);
1758 i
= htole64(hidx
- 1);
1759 r
= link_entry_into_array(f
, first
, &i
, p
);
1764 *idx
= htole64(hidx
+ 1);
1768 static int journal_file_link_entry_item(JournalFile
*f
, Object
*o
, uint64_t offset
, uint64_t i
) {
1776 p
= le64toh(o
->entry
.items
[i
].object_offset
);
1777 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1781 return link_entry_into_array_plus_one(f
,
1782 &o
->data
.entry_offset
,
1783 &o
->data
.entry_array_offset
,
1788 static int journal_file_link_entry(JournalFile
*f
, Object
*o
, uint64_t offset
) {
1797 if (o
->object
.type
!= OBJECT_ENTRY
)
1800 __sync_synchronize();
1802 /* Link up the entry itself */
1803 r
= link_entry_into_array(f
,
1804 &f
->header
->entry_array_offset
,
1805 &f
->header
->n_entries
,
1810 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
1812 if (f
->header
->head_entry_realtime
== 0)
1813 f
->header
->head_entry_realtime
= o
->entry
.realtime
;
1815 f
->header
->tail_entry_realtime
= o
->entry
.realtime
;
1816 f
->header
->tail_entry_monotonic
= o
->entry
.monotonic
;
1818 /* Link up the items */
1819 n
= journal_file_entry_n_items(o
);
1820 for (i
= 0; i
< n
; i
++) {
1821 r
= journal_file_link_entry_item(f
, o
, offset
, i
);
1829 static int journal_file_append_entry_internal(
1831 const dual_timestamp
*ts
,
1832 const sd_id128_t
*boot_id
,
1834 const EntryItem items
[], unsigned n_items
,
1836 Object
**ret
, uint64_t *offset
) {
1844 assert(items
|| n_items
== 0);
1847 osize
= offsetof(Object
, entry
.items
) + (n_items
* sizeof(EntryItem
));
1849 r
= journal_file_append_object(f
, OBJECT_ENTRY
, osize
, &o
, &np
);
1853 o
->entry
.seqnum
= htole64(journal_file_entry_seqnum(f
, seqnum
));
1854 memcpy_safe(o
->entry
.items
, items
, n_items
* sizeof(EntryItem
));
1855 o
->entry
.realtime
= htole64(ts
->realtime
);
1856 o
->entry
.monotonic
= htole64(ts
->monotonic
);
1857 o
->entry
.xor_hash
= htole64(xor_hash
);
1859 f
->header
->boot_id
= *boot_id
;
1860 o
->entry
.boot_id
= f
->header
->boot_id
;
1863 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY
, o
, np
);
1868 r
= journal_file_link_entry(f
, o
, np
);
1881 void journal_file_post_change(JournalFile
*f
) {
1887 /* inotify() does not receive IN_MODIFY events from file
1888 * accesses done via mmap(). After each access we hence
1889 * trigger IN_MODIFY by truncating the journal file to its
1890 * current size which triggers IN_MODIFY. */
1892 __sync_synchronize();
1894 if (ftruncate(f
->fd
, f
->last_stat
.st_size
) < 0)
1895 log_debug_errno(errno
, "Failed to truncate file to its own size: %m");
1898 static int post_change_thunk(sd_event_source
*timer
, uint64_t usec
, void *userdata
) {
1901 journal_file_post_change(userdata
);
1906 static void schedule_post_change(JournalFile
*f
) {
1911 assert(f
->post_change_timer
);
1913 r
= sd_event_source_get_enabled(f
->post_change_timer
, NULL
);
1915 log_debug_errno(r
, "Failed to get ftruncate timer state: %m");
1921 r
= sd_event_now(sd_event_source_get_event(f
->post_change_timer
), CLOCK_MONOTONIC
, &now
);
1923 log_debug_errno(r
, "Failed to get clock's now for scheduling ftruncate: %m");
1927 r
= sd_event_source_set_time(f
->post_change_timer
, now
+ f
->post_change_timer_period
);
1929 log_debug_errno(r
, "Failed to set time for scheduling ftruncate: %m");
1933 r
= sd_event_source_set_enabled(f
->post_change_timer
, SD_EVENT_ONESHOT
);
1935 log_debug_errno(r
, "Failed to enable scheduled ftruncate: %m");
1942 /* On failure, let's simply post the change immediately. */
1943 journal_file_post_change(f
);
1946 /* Enable coalesced change posting in a timer on the provided sd_event instance */
1947 int journal_file_enable_post_change_timer(JournalFile
*f
, sd_event
*e
, usec_t t
) {
1948 _cleanup_(sd_event_source_unrefp
) sd_event_source
*timer
= NULL
;
1952 assert_return(!f
->post_change_timer
, -EINVAL
);
1956 r
= sd_event_add_time(e
, &timer
, CLOCK_MONOTONIC
, 0, 0, post_change_thunk
, f
);
1960 r
= sd_event_source_set_enabled(timer
, SD_EVENT_OFF
);
1964 f
->post_change_timer
= TAKE_PTR(timer
);
1965 f
->post_change_timer_period
= t
;
1970 static int entry_item_cmp(const EntryItem
*a
, const EntryItem
*b
) {
1971 return CMP(le64toh(a
->object_offset
), le64toh(b
->object_offset
));
1974 int journal_file_append_entry(
1976 const dual_timestamp
*ts
,
1977 const sd_id128_t
*boot_id
,
1978 const struct iovec iovec
[], unsigned n_iovec
,
1980 Object
**ret
, uint64_t *offset
) {
1985 uint64_t xor_hash
= 0;
1986 struct dual_timestamp _ts
;
1990 assert(iovec
|| n_iovec
== 0);
1993 if (!VALID_REALTIME(ts
->realtime
))
1994 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1995 "Invalid realtime timestamp %" PRIu64
", refusing entry.",
1997 if (!VALID_MONOTONIC(ts
->monotonic
))
1998 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1999 "Invalid monotomic timestamp %" PRIu64
", refusing entry.",
2002 dual_timestamp_get(&_ts
);
2007 r
= journal_file_maybe_append_tag(f
, ts
->realtime
);
2012 /* alloca() can't take 0, hence let's allocate at least one */
2013 items
= newa(EntryItem
, MAX(1u, n_iovec
));
2015 for (i
= 0; i
< n_iovec
; i
++) {
2019 r
= journal_file_append_data(f
, iovec
[i
].iov_base
, iovec
[i
].iov_len
, &o
, &p
);
2023 xor_hash
^= le64toh(o
->data
.hash
);
2024 items
[i
].object_offset
= htole64(p
);
2025 items
[i
].hash
= o
->data
.hash
;
2028 /* Order by the position on disk, in order to improve seek
2029 * times for rotating media. */
2030 typesafe_qsort(items
, n_iovec
, entry_item_cmp
);
2032 r
= journal_file_append_entry_internal(f
, ts
, boot_id
, xor_hash
, items
, n_iovec
, seqnum
, ret
, offset
);
2034 /* If the memory mapping triggered a SIGBUS then we return an
2035 * IO error and ignore the error code passed down to us, since
2036 * it is very likely just an effect of a nullified replacement
2039 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
2042 if (f
->post_change_timer
)
2043 schedule_post_change(f
);
2045 journal_file_post_change(f
);
2050 typedef struct ChainCacheItem
{
2051 uint64_t first
; /* the array at the beginning of the chain */
2052 uint64_t array
; /* the cached array */
2053 uint64_t begin
; /* the first item in the cached array */
2054 uint64_t total
; /* the total number of items in all arrays before this one in the chain */
2055 uint64_t last_index
; /* the last index we looked at, to optimize locality when bisecting */
2058 static void chain_cache_put(
2065 uint64_t last_index
) {
2068 /* If the chain item to cache for this chain is the
2069 * first one it's not worth caching anything */
2073 if (ordered_hashmap_size(h
) >= CHAIN_CACHE_MAX
) {
2074 ci
= ordered_hashmap_steal_first(h
);
2077 ci
= new(ChainCacheItem
, 1);
2084 if (ordered_hashmap_put(h
, &ci
->first
, ci
) < 0) {
2089 assert(ci
->first
== first
);
2094 ci
->last_index
= last_index
;
2097 static int generic_array_get(
2101 Object
**ret
, uint64_t *offset
) {
2104 uint64_t p
= 0, a
, t
= 0;
2112 /* Try the chain cache first */
2113 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2114 if (ci
&& i
> ci
->total
) {
2123 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2127 k
= journal_file_entry_array_n_items(o
);
2129 p
= le64toh(o
->entry_array
.items
[i
]);
2135 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
2141 /* Let's cache this item for the next invocation */
2142 chain_cache_put(f
->chain_cache
, ci
, first
, a
, le64toh(o
->entry_array
.items
[0]), t
, i
);
2144 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2157 static int generic_array_get_plus_one(
2162 Object
**ret
, uint64_t *offset
) {
2171 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2184 return generic_array_get(f
, first
, i
-1, ret
, offset
);
2193 static int generic_array_bisect(
2198 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2199 direction_t direction
,
2204 uint64_t a
, p
, t
= 0, i
= 0, last_p
= 0, last_index
= (uint64_t) -1;
2205 bool subtract_one
= false;
2206 Object
*o
, *array
= NULL
;
2211 assert(test_object
);
2213 /* Start with the first array in the chain */
2216 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2217 if (ci
&& n
> ci
->total
&& ci
->begin
!= 0) {
2218 /* Ah, we have iterated this bisection array chain
2219 * previously! Let's see if we can skip ahead in the
2220 * chain, as far as the last time. But we can't jump
2221 * backwards in the chain, so let's check that
2224 r
= test_object(f
, ci
->begin
, needle
);
2228 if (r
== TEST_LEFT
) {
2229 /* OK, what we are looking for is right of the
2230 * begin of this EntryArray, so let's jump
2231 * straight to previously cached array in the
2237 last_index
= ci
->last_index
;
2242 uint64_t left
, right
, k
, lp
;
2244 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2248 k
= journal_file_entry_array_n_items(array
);
2254 lp
= p
= le64toh(array
->entry_array
.items
[i
]);
2258 r
= test_object(f
, p
, needle
);
2259 if (r
== -EBADMSG
) {
2260 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (1)");
2267 if (r
== TEST_FOUND
)
2268 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2270 if (r
== TEST_RIGHT
) {
2274 if (last_index
!= (uint64_t) -1) {
2275 assert(last_index
<= right
);
2277 /* If we cached the last index we
2278 * looked at, let's try to not to jump
2279 * too wildly around and see if we can
2280 * limit the range to look at early to
2281 * the immediate neighbors of the last
2282 * index we looked at. */
2284 if (last_index
> 0) {
2285 uint64_t x
= last_index
- 1;
2287 p
= le64toh(array
->entry_array
.items
[x
]);
2291 r
= test_object(f
, p
, needle
);
2295 if (r
== TEST_FOUND
)
2296 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2298 if (r
== TEST_RIGHT
)
2304 if (last_index
< right
) {
2305 uint64_t y
= last_index
+ 1;
2307 p
= le64toh(array
->entry_array
.items
[y
]);
2311 r
= test_object(f
, p
, needle
);
2315 if (r
== TEST_FOUND
)
2316 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2318 if (r
== TEST_RIGHT
)
2326 if (left
== right
) {
2327 if (direction
== DIRECTION_UP
)
2328 subtract_one
= true;
2334 assert(left
< right
);
2335 i
= (left
+ right
) / 2;
2337 p
= le64toh(array
->entry_array
.items
[i
]);
2341 r
= test_object(f
, p
, needle
);
2342 if (r
== -EBADMSG
) {
2343 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (2)");
2350 if (r
== TEST_FOUND
)
2351 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2353 if (r
== TEST_RIGHT
)
2361 if (direction
== DIRECTION_UP
) {
2363 subtract_one
= true;
2374 last_index
= (uint64_t) -1;
2375 a
= le64toh(array
->entry_array
.next_entry_array_offset
);
2381 if (subtract_one
&& t
== 0 && i
== 0)
2384 /* Let's cache this item for the next invocation */
2385 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
);
2387 if (subtract_one
&& i
== 0)
2389 else if (subtract_one
)
2390 p
= le64toh(array
->entry_array
.items
[i
-1]);
2392 p
= le64toh(array
->entry_array
.items
[i
]);
2394 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2405 *idx
= t
+ i
+ (subtract_one
? -1 : 0);
2410 static int generic_array_bisect_plus_one(
2416 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2417 direction_t direction
,
2423 bool step_back
= false;
2427 assert(test_object
);
2432 /* This bisects the array in object 'first', but first checks
2434 r
= test_object(f
, extra
, needle
);
2438 if (r
== TEST_FOUND
)
2439 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2441 /* if we are looking with DIRECTION_UP then we need to first
2442 see if in the actual array there is a matching entry, and
2443 return the last one of that. But if there isn't any we need
2444 to return this one. Hence remember this, and return it
2447 step_back
= direction
== DIRECTION_UP
;
2449 if (r
== TEST_RIGHT
) {
2450 if (direction
== DIRECTION_DOWN
)
2456 r
= generic_array_bisect(f
, first
, n
-1, needle
, test_object
, direction
, ret
, offset
, idx
);
2458 if (r
== 0 && step_back
)
2467 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2483 _pure_
static int test_object_offset(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2489 else if (p
< needle
)
2495 static int test_object_seqnum(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2503 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2507 sq
= le64toh(READ_NOW(o
->entry
.seqnum
));
2510 else if (sq
< needle
)
2516 int journal_file_move_to_entry_by_seqnum(
2519 direction_t direction
,
2525 return generic_array_bisect(f
,
2526 le64toh(f
->header
->entry_array_offset
),
2527 le64toh(f
->header
->n_entries
),
2534 static int test_object_realtime(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2542 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2546 rt
= le64toh(READ_NOW(o
->entry
.realtime
));
2549 else if (rt
< needle
)
2555 int journal_file_move_to_entry_by_realtime(
2558 direction_t direction
,
2564 return generic_array_bisect(f
,
2565 le64toh(f
->header
->entry_array_offset
),
2566 le64toh(f
->header
->n_entries
),
2568 test_object_realtime
,
2573 static int test_object_monotonic(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2581 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2585 m
= le64toh(READ_NOW(o
->entry
.monotonic
));
2588 else if (m
< needle
)
2594 static int find_data_object_by_boot_id(
2600 char t
[STRLEN("_BOOT_ID=") + 32 + 1] = "_BOOT_ID=";
2602 sd_id128_to_string(boot_id
, t
+ 9);
2603 return journal_file_find_data_object(f
, t
, sizeof(t
) - 1, o
, b
);
2606 int journal_file_move_to_entry_by_monotonic(
2610 direction_t direction
,
2619 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, NULL
);
2625 return generic_array_bisect_plus_one(f
,
2626 le64toh(o
->data
.entry_offset
),
2627 le64toh(o
->data
.entry_array_offset
),
2628 le64toh(o
->data
.n_entries
),
2630 test_object_monotonic
,
2635 void journal_file_reset_location(JournalFile
*f
) {
2636 f
->location_type
= LOCATION_HEAD
;
2637 f
->current_offset
= 0;
2638 f
->current_seqnum
= 0;
2639 f
->current_realtime
= 0;
2640 f
->current_monotonic
= 0;
2641 zero(f
->current_boot_id
);
2642 f
->current_xor_hash
= 0;
2645 void journal_file_save_location(JournalFile
*f
, Object
*o
, uint64_t offset
) {
2646 f
->location_type
= LOCATION_SEEK
;
2647 f
->current_offset
= offset
;
2648 f
->current_seqnum
= le64toh(o
->entry
.seqnum
);
2649 f
->current_realtime
= le64toh(o
->entry
.realtime
);
2650 f
->current_monotonic
= le64toh(o
->entry
.monotonic
);
2651 f
->current_boot_id
= o
->entry
.boot_id
;
2652 f
->current_xor_hash
= le64toh(o
->entry
.xor_hash
);
2655 int journal_file_compare_locations(JournalFile
*af
, JournalFile
*bf
) {
2662 assert(af
->location_type
== LOCATION_SEEK
);
2663 assert(bf
->location_type
== LOCATION_SEEK
);
2665 /* If contents and timestamps match, these entries are
2666 * identical, even if the seqnum does not match */
2667 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
) &&
2668 af
->current_monotonic
== bf
->current_monotonic
&&
2669 af
->current_realtime
== bf
->current_realtime
&&
2670 af
->current_xor_hash
== bf
->current_xor_hash
)
2673 if (sd_id128_equal(af
->header
->seqnum_id
, bf
->header
->seqnum_id
)) {
2675 /* If this is from the same seqnum source, compare
2677 r
= CMP(af
->current_seqnum
, bf
->current_seqnum
);
2681 /* Wow! This is weird, different data but the same
2682 * seqnums? Something is borked, but let's make the
2683 * best of it and compare by time. */
2686 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
)) {
2688 /* If the boot id matches, compare monotonic time */
2689 r
= CMP(af
->current_monotonic
, bf
->current_monotonic
);
2694 /* Otherwise, compare UTC time */
2695 r
= CMP(af
->current_realtime
, bf
->current_realtime
);
2699 /* Finally, compare by contents */
2700 return CMP(af
->current_xor_hash
, bf
->current_xor_hash
);
2703 static int bump_array_index(uint64_t *i
, direction_t direction
, uint64_t n
) {
2705 /* Increase or decrease the specified index, in the right direction. */
2707 if (direction
== DIRECTION_DOWN
) {
2722 static bool check_properly_ordered(uint64_t new_offset
, uint64_t old_offset
, direction_t direction
) {
2724 /* Consider it an error if any of the two offsets is uninitialized */
2725 if (old_offset
== 0 || new_offset
== 0)
2728 /* If we go down, the new offset must be larger than the old one. */
2729 return direction
== DIRECTION_DOWN
?
2730 new_offset
> old_offset
:
2731 new_offset
< old_offset
;
2734 int journal_file_next_entry(
2737 direction_t direction
,
2738 Object
**ret
, uint64_t *offset
) {
2746 n
= le64toh(READ_NOW(f
->header
->n_entries
));
2751 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2753 r
= generic_array_bisect(f
,
2754 le64toh(f
->header
->entry_array_offset
),
2755 le64toh(f
->header
->n_entries
),
2764 r
= bump_array_index(&i
, direction
, n
);
2769 /* And jump to it */
2771 r
= generic_array_get(f
,
2772 le64toh(f
->header
->entry_array_offset
),
2780 /* OK, so this entry is borked. Most likely some entry didn't get synced to disk properly, let's see if
2781 * the next one might work for us instead. */
2782 log_debug_errno(r
, "Entry item %" PRIu64
" is bad, skipping over it.", i
);
2784 r
= bump_array_index(&i
, direction
, n
);
2789 /* Ensure our array is properly ordered. */
2790 if (p
> 0 && !check_properly_ordered(ofs
, p
, direction
))
2791 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
2792 "%s: entry array not properly ordered at entry %" PRIu64
,
2801 int journal_file_next_entry_for_data(
2803 Object
*o
, uint64_t p
,
2804 uint64_t data_offset
,
2805 direction_t direction
,
2806 Object
**ret
, uint64_t *offset
) {
2813 assert(p
> 0 || !o
);
2815 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2819 n
= le64toh(READ_NOW(d
->data
.n_entries
));
2824 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2826 if (o
->object
.type
!= OBJECT_ENTRY
)
2829 r
= generic_array_bisect_plus_one(f
,
2830 le64toh(d
->data
.entry_offset
),
2831 le64toh(d
->data
.entry_array_offset
),
2832 le64toh(d
->data
.n_entries
),
2842 r
= bump_array_index(&i
, direction
, n
);
2848 r
= generic_array_get_plus_one(f
,
2849 le64toh(d
->data
.entry_offset
),
2850 le64toh(d
->data
.entry_array_offset
),
2858 log_debug_errno(r
, "Data entry item %" PRIu64
" is bad, skipping over it.", i
);
2860 r
= bump_array_index(&i
, direction
, n
);
2865 /* Ensure our array is properly ordered. */
2866 if (p
> 0 && check_properly_ordered(ofs
, p
, direction
))
2867 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
2868 "%s data entry array not properly ordered at entry %" PRIu64
,
2877 int journal_file_move_to_entry_by_offset_for_data(
2879 uint64_t data_offset
,
2881 direction_t direction
,
2882 Object
**ret
, uint64_t *offset
) {
2889 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2893 return generic_array_bisect_plus_one(f
,
2894 le64toh(d
->data
.entry_offset
),
2895 le64toh(d
->data
.entry_array_offset
),
2896 le64toh(d
->data
.n_entries
),
2903 int journal_file_move_to_entry_by_monotonic_for_data(
2905 uint64_t data_offset
,
2908 direction_t direction
,
2909 Object
**ret
, uint64_t *offset
) {
2917 /* First, seek by time */
2918 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &b
);
2924 r
= generic_array_bisect_plus_one(f
,
2925 le64toh(o
->data
.entry_offset
),
2926 le64toh(o
->data
.entry_array_offset
),
2927 le64toh(o
->data
.n_entries
),
2929 test_object_monotonic
,
2935 /* And now, continue seeking until we find an entry that
2936 * exists in both bisection arrays */
2942 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2946 r
= generic_array_bisect_plus_one(f
,
2947 le64toh(d
->data
.entry_offset
),
2948 le64toh(d
->data
.entry_array_offset
),
2949 le64toh(d
->data
.n_entries
),
2957 r
= journal_file_move_to_object(f
, OBJECT_DATA
, b
, &o
);
2961 r
= generic_array_bisect_plus_one(f
,
2962 le64toh(o
->data
.entry_offset
),
2963 le64toh(o
->data
.entry_array_offset
),
2964 le64toh(o
->data
.n_entries
),
2986 int journal_file_move_to_entry_by_seqnum_for_data(
2988 uint64_t data_offset
,
2990 direction_t direction
,
2991 Object
**ret
, uint64_t *offset
) {
2998 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
3002 return generic_array_bisect_plus_one(f
,
3003 le64toh(d
->data
.entry_offset
),
3004 le64toh(d
->data
.entry_array_offset
),
3005 le64toh(d
->data
.n_entries
),
3012 int journal_file_move_to_entry_by_realtime_for_data(
3014 uint64_t data_offset
,
3016 direction_t direction
,
3017 Object
**ret
, uint64_t *offset
) {
3024 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
3028 return generic_array_bisect_plus_one(f
,
3029 le64toh(d
->data
.entry_offset
),
3030 le64toh(d
->data
.entry_array_offset
),
3031 le64toh(d
->data
.n_entries
),
3033 test_object_realtime
,
3038 void journal_file_dump(JournalFile
*f
) {
3046 journal_file_print_header(f
);
3048 p
= le64toh(READ_NOW(f
->header
->header_size
));
3050 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &o
);
3054 switch (o
->object
.type
) {
3057 printf("Type: OBJECT_UNUSED\n");
3061 printf("Type: OBJECT_DATA\n");
3065 printf("Type: OBJECT_FIELD\n");
3069 printf("Type: OBJECT_ENTRY seqnum=%"PRIu64
" monotonic=%"PRIu64
" realtime=%"PRIu64
"\n",
3070 le64toh(o
->entry
.seqnum
),
3071 le64toh(o
->entry
.monotonic
),
3072 le64toh(o
->entry
.realtime
));
3075 case OBJECT_FIELD_HASH_TABLE
:
3076 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
3079 case OBJECT_DATA_HASH_TABLE
:
3080 printf("Type: OBJECT_DATA_HASH_TABLE\n");
3083 case OBJECT_ENTRY_ARRAY
:
3084 printf("Type: OBJECT_ENTRY_ARRAY\n");
3088 printf("Type: OBJECT_TAG seqnum=%"PRIu64
" epoch=%"PRIu64
"\n",
3089 le64toh(o
->tag
.seqnum
),
3090 le64toh(o
->tag
.epoch
));
3094 printf("Type: unknown (%i)\n", o
->object
.type
);
3098 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
)
3099 printf("Flags: %s\n",
3100 object_compressed_to_string(o
->object
.flags
& OBJECT_COMPRESSION_MASK
));
3102 if (p
== le64toh(f
->header
->tail_object_offset
))
3105 p
+= ALIGN64(le64toh(o
->object
.size
));
3110 log_error("File corrupt");
3113 static const char* format_timestamp_safe(char *buf
, size_t l
, usec_t t
) {
3116 x
= format_timestamp(buf
, l
, t
);
3122 void journal_file_print_header(JournalFile
*f
) {
3123 char a
[SD_ID128_STRING_MAX
], b
[SD_ID128_STRING_MAX
], c
[SD_ID128_STRING_MAX
], d
[SD_ID128_STRING_MAX
];
3124 char x
[FORMAT_TIMESTAMP_MAX
], y
[FORMAT_TIMESTAMP_MAX
], z
[FORMAT_TIMESTAMP_MAX
];
3126 char bytes
[FORMAT_BYTES_MAX
];
3131 printf("File path: %s\n"
3135 "Sequential number ID: %s\n"
3137 "Compatible flags:%s%s\n"
3138 "Incompatible flags:%s%s%s\n"
3139 "Header size: %"PRIu64
"\n"
3140 "Arena size: %"PRIu64
"\n"
3141 "Data hash table size: %"PRIu64
"\n"
3142 "Field hash table size: %"PRIu64
"\n"
3143 "Rotate suggested: %s\n"
3144 "Head sequential number: %"PRIu64
" (%"PRIx64
")\n"
3145 "Tail sequential number: %"PRIu64
" (%"PRIx64
")\n"
3146 "Head realtime timestamp: %s (%"PRIx64
")\n"
3147 "Tail realtime timestamp: %s (%"PRIx64
")\n"
3148 "Tail monotonic timestamp: %s (%"PRIx64
")\n"
3149 "Objects: %"PRIu64
"\n"
3150 "Entry objects: %"PRIu64
"\n",
3152 sd_id128_to_string(f
->header
->file_id
, a
),
3153 sd_id128_to_string(f
->header
->machine_id
, b
),
3154 sd_id128_to_string(f
->header
->boot_id
, c
),
3155 sd_id128_to_string(f
->header
->seqnum_id
, d
),
3156 f
->header
->state
== STATE_OFFLINE
? "OFFLINE" :
3157 f
->header
->state
== STATE_ONLINE
? "ONLINE" :
3158 f
->header
->state
== STATE_ARCHIVED
? "ARCHIVED" : "UNKNOWN",
3159 JOURNAL_HEADER_SEALED(f
->header
) ? " SEALED" : "",
3160 (le32toh(f
->header
->compatible_flags
) & ~HEADER_COMPATIBLE_ANY
) ? " ???" : "",
3161 JOURNAL_HEADER_COMPRESSED_XZ(f
->header
) ? " COMPRESSED-XZ" : "",
3162 JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
) ? " COMPRESSED-LZ4" : "",
3163 (le32toh(f
->header
->incompatible_flags
) & ~HEADER_INCOMPATIBLE_ANY
) ? " ???" : "",
3164 le64toh(f
->header
->header_size
),
3165 le64toh(f
->header
->arena_size
),
3166 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3167 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
),
3168 yes_no(journal_file_rotate_suggested(f
, 0)),
3169 le64toh(f
->header
->head_entry_seqnum
), le64toh(f
->header
->head_entry_seqnum
),
3170 le64toh(f
->header
->tail_entry_seqnum
), le64toh(f
->header
->tail_entry_seqnum
),
3171 format_timestamp_safe(x
, sizeof(x
), le64toh(f
->header
->head_entry_realtime
)), le64toh(f
->header
->head_entry_realtime
),
3172 format_timestamp_safe(y
, sizeof(y
), le64toh(f
->header
->tail_entry_realtime
)), le64toh(f
->header
->tail_entry_realtime
),
3173 format_timespan(z
, sizeof(z
), le64toh(f
->header
->tail_entry_monotonic
), USEC_PER_MSEC
), le64toh(f
->header
->tail_entry_monotonic
),
3174 le64toh(f
->header
->n_objects
),
3175 le64toh(f
->header
->n_entries
));
3177 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3178 printf("Data objects: %"PRIu64
"\n"
3179 "Data hash table fill: %.1f%%\n",
3180 le64toh(f
->header
->n_data
),
3181 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))));
3183 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3184 printf("Field objects: %"PRIu64
"\n"
3185 "Field hash table fill: %.1f%%\n",
3186 le64toh(f
->header
->n_fields
),
3187 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))));
3189 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
))
3190 printf("Tag objects: %"PRIu64
"\n",
3191 le64toh(f
->header
->n_tags
));
3192 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
3193 printf("Entry array objects: %"PRIu64
"\n",
3194 le64toh(f
->header
->n_entry_arrays
));
3196 if (fstat(f
->fd
, &st
) >= 0)
3197 printf("Disk usage: %s\n", format_bytes(bytes
, sizeof(bytes
), (uint64_t) st
.st_blocks
* 512ULL));
3200 static int journal_file_warn_btrfs(JournalFile
*f
) {
3206 /* Before we write anything, check if the COW logic is turned
3207 * off on btrfs. Given our write pattern that is quite
3208 * unfriendly to COW file systems this should greatly improve
3209 * performance on COW file systems, such as btrfs, at the
3210 * expense of data integrity features (which shouldn't be too
3211 * bad, given that we do our own checksumming). */
3213 r
= btrfs_is_filesystem(f
->fd
);
3215 return log_warning_errno(r
, "Failed to determine if journal is on btrfs: %m");
3219 r
= read_attr_fd(f
->fd
, &attrs
);
3221 return log_warning_errno(r
, "Failed to read file attributes: %m");
3223 if (attrs
& FS_NOCOW_FL
) {
3224 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3228 log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3229 "This is likely to slow down journal access substantially, please consider turning "
3230 "off the copy-on-write file attribute on the journal directory, using chattr +C.", f
->path
);
3235 int journal_file_open(
3241 uint64_t compress_threshold_bytes
,
3243 JournalMetrics
*metrics
,
3244 MMapCache
*mmap_cache
,
3245 Set
*deferred_closes
,
3246 JournalFile
*template,
3247 JournalFile
**ret
) {
3249 bool newly_created
= false;
3255 assert(fd
>= 0 || fname
);
3257 if (!IN_SET((flags
& O_ACCMODE
), O_RDONLY
, O_RDWR
))
3260 if (fname
&& (flags
& O_CREAT
) && !endswith(fname
, ".journal"))
3263 f
= new(JournalFile
, 1);
3267 *f
= (JournalFile
) {
3272 .prot
= prot_from_flags(flags
),
3273 .writable
= (flags
& O_ACCMODE
) != O_RDONLY
,
3276 .compress_lz4
= compress
,
3278 .compress_xz
= compress
,
3280 .compress_threshold_bytes
= compress_threshold_bytes
== (uint64_t) -1 ?
3281 DEFAULT_COMPRESS_THRESHOLD
:
3282 MAX(MIN_COMPRESS_THRESHOLD
, compress_threshold_bytes
),
3288 if (DEBUG_LOGGING
) {
3289 static int last_seal
= -1, last_compress
= -1;
3290 static uint64_t last_bytes
= UINT64_MAX
;
3291 char bytes
[FORMAT_BYTES_MAX
];
3293 if (last_seal
!= f
->seal
||
3294 last_compress
!= JOURNAL_FILE_COMPRESS(f
) ||
3295 last_bytes
!= f
->compress_threshold_bytes
) {
3297 log_debug("Journal effective settings seal=%s compress=%s compress_threshold_bytes=%s",
3298 yes_no(f
->seal
), yes_no(JOURNAL_FILE_COMPRESS(f
)),
3299 format_bytes(bytes
, sizeof bytes
, f
->compress_threshold_bytes
));
3300 last_seal
= f
->seal
;
3301 last_compress
= JOURNAL_FILE_COMPRESS(f
);
3302 last_bytes
= f
->compress_threshold_bytes
;
3307 f
->mmap
= mmap_cache_ref(mmap_cache
);
3309 f
->mmap
= mmap_cache_new();
3317 f
->path
= strdup(fname
);
3325 /* If we don't know the path, fill in something explanatory and vaguely useful */
3326 if (asprintf(&f
->path
, "/proc/self/%i", fd
) < 0) {
3332 f
->chain_cache
= ordered_hashmap_new(&uint64_hash_ops
);
3333 if (!f
->chain_cache
) {
3339 /* We pass O_NONBLOCK here, so that in case somebody pointed us to some character device node or FIFO
3340 * or so, we likely fail quickly than block for long. For regular files O_NONBLOCK has no effect, hence
3341 * it doesn't hurt in that case. */
3343 f
->fd
= open(f
->path
, f
->flags
|O_CLOEXEC
|O_NONBLOCK
, f
->mode
);
3349 /* fds we opened here by us should also be closed by us. */
3352 r
= fd_nonblock(f
->fd
, false);
3357 f
->cache_fd
= mmap_cache_add_fd(f
->mmap
, f
->fd
);
3363 r
= journal_file_fstat(f
);
3367 if (f
->last_stat
.st_size
== 0 && f
->writable
) {
3369 (void) journal_file_warn_btrfs(f
);
3371 /* Let's attach the creation time to the journal file, so that the vacuuming code knows the age of this
3372 * file even if the file might end up corrupted one day... Ideally we'd just use the creation time many
3373 * file systems maintain for each file, but the API to query this is very new, hence let's emulate this
3374 * via extended attributes. If extended attributes are not supported we'll just skip this, and rely
3375 * solely on mtime/atime/ctime of the file. */
3376 (void) fd_setcrtime(f
->fd
, 0);
3379 /* Try to load the FSPRG state, and if we can't, then
3380 * just don't do sealing */
3382 r
= journal_file_fss_load(f
);
3388 r
= journal_file_init_header(f
, template);
3392 r
= journal_file_fstat(f
);
3396 newly_created
= true;
3399 if (f
->last_stat
.st_size
< (off_t
) HEADER_SIZE_MIN
) {
3404 r
= mmap_cache_get(f
->mmap
, f
->cache_fd
, f
->prot
, CONTEXT_HEADER
, true, 0, PAGE_ALIGN(sizeof(Header
)), &f
->last_stat
, &h
, NULL
);
3406 /* Some file systems (jffs2 or p9fs) don't support mmap() properly (or only read-only
3407 * mmap()), and return EINVAL in that case. Let's propagate that as a more recognizable error
3417 if (!newly_created
) {
3418 set_clear_with_destructor(deferred_closes
, journal_file_close
);
3420 r
= journal_file_verify_header(f
);
3426 if (!newly_created
&& f
->writable
) {
3427 r
= journal_file_fss_load(f
);
3435 journal_default_metrics(metrics
, f
->fd
);
3436 f
->metrics
= *metrics
;
3437 } else if (template)
3438 f
->metrics
= template->metrics
;
3440 r
= journal_file_refresh_header(f
);
3446 r
= journal_file_hmac_setup(f
);
3451 if (newly_created
) {
3452 r
= journal_file_setup_field_hash_table(f
);
3456 r
= journal_file_setup_data_hash_table(f
);
3461 r
= journal_file_append_first_tag(f
);
3467 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
)) {
3472 if (template && template->post_change_timer
) {
3473 r
= journal_file_enable_post_change_timer(
3475 sd_event_source_get_event(template->post_change_timer
),
3476 template->post_change_timer_period
);
3482 /* The file is opened now successfully, thus we take possession of any passed in fd. */
3489 if (f
->cache_fd
&& mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
3492 (void) journal_file_close(f
);
3497 int journal_file_archive(JournalFile
*f
) {
3498 _cleanup_free_
char *p
= NULL
;
3505 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
3506 * rotation, since we don't know the actual path, and couldn't rename the file hence. */
3507 if (path_startswith(f
->path
, "/proc/self/fd"))
3510 if (!endswith(f
->path
, ".journal"))
3513 if (asprintf(&p
, "%.*s@" SD_ID128_FORMAT_STR
"-%016"PRIx64
"-%016"PRIx64
".journal",
3514 (int) strlen(f
->path
) - 8, f
->path
,
3515 SD_ID128_FORMAT_VAL(f
->header
->seqnum_id
),
3516 le64toh(f
->header
->head_entry_seqnum
),
3517 le64toh(f
->header
->head_entry_realtime
)) < 0)
3520 /* Try to rename the file to the archived version. If the file already was deleted, we'll get ENOENT, let's
3521 * ignore that case. */
3522 if (rename(f
->path
, p
) < 0 && errno
!= ENOENT
)
3525 /* Sync the rename to disk */
3526 (void) fsync_directory_of_file(f
->fd
);
3528 /* Set as archive so offlining commits w/state=STATE_ARCHIVED. Previously we would set old_file->header->state
3529 * to STATE_ARCHIVED directly here, but journal_file_set_offline() short-circuits when state != STATE_ONLINE,
3530 * which would result in the rotated journal never getting fsync() called before closing. Now we simply queue
3531 * the archive state by setting an archive bit, leaving the state as STATE_ONLINE so proper offlining
3535 /* Currently, btrfs is not very good with out write patterns and fragments heavily. Let's defrag our journal
3536 * files when we archive them */
3537 f
->defrag_on_close
= true;
3542 JournalFile
* journal_initiate_close(
3544 Set
*deferred_closes
) {
3550 if (deferred_closes
) {
3552 r
= set_put(deferred_closes
, f
);
3554 log_debug_errno(r
, "Failed to add file to deferred close set, closing immediately.");
3556 (void) journal_file_set_offline(f
, false);
3561 return journal_file_close(f
);
3564 int journal_file_rotate(
3567 uint64_t compress_threshold_bytes
,
3569 Set
*deferred_closes
) {
3571 JournalFile
*new_file
= NULL
;
3577 r
= journal_file_archive(*f
);
3581 r
= journal_file_open(
3587 compress_threshold_bytes
,
3595 journal_initiate_close(*f
, deferred_closes
);
3601 int journal_file_dispose(int dir_fd
, const char *fname
) {
3602 _cleanup_free_
char *p
= NULL
;
3603 _cleanup_close_
int fd
= -1;
3607 /* Renames a journal file to *.journal~, i.e. to mark it as corruped or otherwise uncleanly shutdown. Note that
3608 * this is done without looking into the file or changing any of its contents. The idea is that this is called
3609 * whenever something is suspicious and we want to move the file away and make clear that it is not accessed
3610 * for writing anymore. */
3612 if (!endswith(fname
, ".journal"))
3615 if (asprintf(&p
, "%.*s@%016" PRIx64
"-%016" PRIx64
".journal~",
3616 (int) strlen(fname
) - 8, fname
,
3617 now(CLOCK_REALTIME
),
3621 if (renameat(dir_fd
, fname
, dir_fd
, p
) < 0)
3624 /* btrfs doesn't cope well with our write pattern and fragments heavily. Let's defrag all files we rotate */
3625 fd
= openat(dir_fd
, p
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
|O_NOFOLLOW
);
3627 log_debug_errno(errno
, "Failed to open file for defragmentation/FS_NOCOW_FL, ignoring: %m");
3629 (void) chattr_fd(fd
, 0, FS_NOCOW_FL
, NULL
);
3630 (void) btrfs_defrag_fd(fd
);
3636 int journal_file_open_reliably(
3641 uint64_t compress_threshold_bytes
,
3643 JournalMetrics
*metrics
,
3644 MMapCache
*mmap_cache
,
3645 Set
*deferred_closes
,
3646 JournalFile
*template,
3647 JournalFile
**ret
) {
3651 r
= journal_file_open(-1, fname
, flags
, mode
, compress
, compress_threshold_bytes
, seal
, metrics
, mmap_cache
,
3652 deferred_closes
, template, ret
);
3654 -EBADMSG
, /* Corrupted */
3655 -ENODATA
, /* Truncated */
3656 -EHOSTDOWN
, /* Other machine */
3657 -EPROTONOSUPPORT
, /* Incompatible feature */
3658 -EBUSY
, /* Unclean shutdown */
3659 -ESHUTDOWN
, /* Already archived */
3660 -EIO
, /* IO error, including SIGBUS on mmap */
3661 -EIDRM
, /* File has been deleted */
3662 -ETXTBSY
)) /* File is from the future */
3665 if ((flags
& O_ACCMODE
) == O_RDONLY
)
3668 if (!(flags
& O_CREAT
))
3671 if (!endswith(fname
, ".journal"))
3674 /* The file is corrupted. Rotate it away and try it again (but only once) */
3675 log_warning_errno(r
, "File %s corrupted or uncleanly shut down, renaming and replacing.", fname
);
3677 r
= journal_file_dispose(AT_FDCWD
, fname
);
3681 return journal_file_open(-1, fname
, flags
, mode
, compress
, compress_threshold_bytes
, seal
, metrics
, mmap_cache
,
3682 deferred_closes
, template, ret
);
3685 int journal_file_copy_entry(JournalFile
*from
, JournalFile
*to
, Object
*o
, uint64_t p
) {
3687 uint64_t q
, xor_hash
= 0;
3691 const sd_id128_t
*boot_id
;
3701 ts
.monotonic
= le64toh(o
->entry
.monotonic
);
3702 ts
.realtime
= le64toh(o
->entry
.realtime
);
3703 boot_id
= &o
->entry
.boot_id
;
3705 n
= journal_file_entry_n_items(o
);
3706 /* alloca() can't take 0, hence let's allocate at least one */
3707 items
= newa(EntryItem
, MAX(1u, n
));
3709 for (i
= 0; i
< n
; i
++) {
3716 q
= le64toh(o
->entry
.items
[i
].object_offset
);
3717 le_hash
= o
->entry
.items
[i
].hash
;
3719 r
= journal_file_move_to_object(from
, OBJECT_DATA
, q
, &o
);
3723 if (le_hash
!= o
->data
.hash
)
3726 l
= le64toh(READ_NOW(o
->object
.size
));
3727 if (l
< offsetof(Object
, data
.payload
))
3730 l
-= offsetof(Object
, data
.payload
);
3733 /* We hit the limit on 32bit machines */
3734 if ((uint64_t) t
!= l
)
3737 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
3738 #if HAVE_XZ || HAVE_LZ4
3741 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
3742 o
->data
.payload
, l
, &from
->compress_buffer
, &from
->compress_buffer_size
, &rsize
, 0);
3746 data
= from
->compress_buffer
;
3749 return -EPROTONOSUPPORT
;
3752 data
= o
->data
.payload
;
3754 r
= journal_file_append_data(to
, data
, l
, &u
, &h
);
3758 xor_hash
^= le64toh(u
->data
.hash
);
3759 items
[i
].object_offset
= htole64(h
);
3760 items
[i
].hash
= u
->data
.hash
;
3762 r
= journal_file_move_to_object(from
, OBJECT_ENTRY
, p
, &o
);
3767 r
= journal_file_append_entry_internal(to
, &ts
, boot_id
, xor_hash
, items
, n
,
3770 if (mmap_cache_got_sigbus(to
->mmap
, to
->cache_fd
))
3776 void journal_reset_metrics(JournalMetrics
*m
) {
3779 /* Set everything to "pick automatic values". */
3781 *m
= (JournalMetrics
) {
3782 .min_use
= (uint64_t) -1,
3783 .max_use
= (uint64_t) -1,
3784 .min_size
= (uint64_t) -1,
3785 .max_size
= (uint64_t) -1,
3786 .keep_free
= (uint64_t) -1,
3787 .n_max_files
= (uint64_t) -1,
3791 void journal_default_metrics(JournalMetrics
*m
, int fd
) {
3792 char a
[FORMAT_BYTES_MAX
], b
[FORMAT_BYTES_MAX
], c
[FORMAT_BYTES_MAX
], d
[FORMAT_BYTES_MAX
], e
[FORMAT_BYTES_MAX
];
3794 uint64_t fs_size
= 0;
3799 if (fstatvfs(fd
, &ss
) >= 0)
3800 fs_size
= ss
.f_frsize
* ss
.f_blocks
;
3802 log_debug_errno(errno
, "Failed to determine disk size: %m");
3804 if (m
->max_use
== (uint64_t) -1) {
3807 m
->max_use
= CLAMP(PAGE_ALIGN(fs_size
/ 10), /* 10% of file system size */
3808 MAX_USE_LOWER
, MAX_USE_UPPER
);
3810 m
->max_use
= MAX_USE_LOWER
;
3812 m
->max_use
= PAGE_ALIGN(m
->max_use
);
3814 if (m
->max_use
!= 0 && m
->max_use
< JOURNAL_FILE_SIZE_MIN
*2)
3815 m
->max_use
= JOURNAL_FILE_SIZE_MIN
*2;
3818 if (m
->min_use
== (uint64_t) -1) {
3820 m
->min_use
= CLAMP(PAGE_ALIGN(fs_size
/ 50), /* 2% of file system size */
3821 MIN_USE_LOW
, MIN_USE_HIGH
);
3823 m
->min_use
= MIN_USE_LOW
;
3826 if (m
->min_use
> m
->max_use
)
3827 m
->min_use
= m
->max_use
;
3829 if (m
->max_size
== (uint64_t) -1)
3830 m
->max_size
= MIN(PAGE_ALIGN(m
->max_use
/ 8), /* 8 chunks */
3833 m
->max_size
= PAGE_ALIGN(m
->max_size
);
3835 if (m
->max_size
!= 0) {
3836 if (m
->max_size
< JOURNAL_FILE_SIZE_MIN
)
3837 m
->max_size
= JOURNAL_FILE_SIZE_MIN
;
3839 if (m
->max_use
!= 0 && m
->max_size
*2 > m
->max_use
)
3840 m
->max_use
= m
->max_size
*2;
3843 if (m
->min_size
== (uint64_t) -1)
3844 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3846 m
->min_size
= CLAMP(PAGE_ALIGN(m
->min_size
),
3847 JOURNAL_FILE_SIZE_MIN
,
3848 m
->max_size
?: UINT64_MAX
);
3850 if (m
->keep_free
== (uint64_t) -1) {
3852 m
->keep_free
= MIN(PAGE_ALIGN(fs_size
/ 20), /* 5% of file system size */
3855 m
->keep_free
= DEFAULT_KEEP_FREE
;
3858 if (m
->n_max_files
== (uint64_t) -1)
3859 m
->n_max_files
= DEFAULT_N_MAX_FILES
;
3861 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64
,
3862 format_bytes(a
, sizeof(a
), m
->min_use
),
3863 format_bytes(b
, sizeof(b
), m
->max_use
),
3864 format_bytes(c
, sizeof(c
), m
->max_size
),
3865 format_bytes(d
, sizeof(d
), m
->min_size
),
3866 format_bytes(e
, sizeof(e
), m
->keep_free
),
3870 int journal_file_get_cutoff_realtime_usec(JournalFile
*f
, usec_t
*from
, usec_t
*to
) {
3876 if (f
->header
->head_entry_realtime
== 0)
3879 *from
= le64toh(f
->header
->head_entry_realtime
);
3883 if (f
->header
->tail_entry_realtime
== 0)
3886 *to
= le64toh(f
->header
->tail_entry_realtime
);
3892 int journal_file_get_cutoff_monotonic_usec(JournalFile
*f
, sd_id128_t boot_id
, usec_t
*from
, usec_t
*to
) {
3900 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &p
);
3904 if (le64toh(o
->data
.n_entries
) <= 0)
3908 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, le64toh(o
->data
.entry_offset
), &o
);
3912 *from
= le64toh(o
->entry
.monotonic
);
3916 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
3920 r
= generic_array_get_plus_one(f
,
3921 le64toh(o
->data
.entry_offset
),
3922 le64toh(o
->data
.entry_array_offset
),
3923 le64toh(o
->data
.n_entries
)-1,
3928 *to
= le64toh(o
->entry
.monotonic
);
3934 bool journal_file_rotate_suggested(JournalFile
*f
, usec_t max_file_usec
) {
3938 /* If we gained new header fields we gained new features,
3939 * hence suggest a rotation */
3940 if (le64toh(f
->header
->header_size
) < sizeof(Header
)) {
3941 log_debug("%s uses an outdated header, suggesting rotation.", f
->path
);
3945 /* Let's check if the hash tables grew over a certain fill
3946 * level (75%, borrowing this value from Java's hash table
3947 * implementation), and if so suggest a rotation. To calculate
3948 * the fill level we need the n_data field, which only exists
3949 * in newer versions. */
3951 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3952 if (le64toh(f
->header
->n_data
) * 4ULL > (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3953 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.",
3955 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))),
3956 le64toh(f
->header
->n_data
),
3957 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3958 (unsigned long long) f
->last_stat
.st_size
,
3959 f
->last_stat
.st_size
/ le64toh(f
->header
->n_data
));
3963 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3964 if (le64toh(f
->header
->n_fields
) * 4ULL > (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3965 log_debug("Field hash table of %s has a fill level at %.1f (%"PRIu64
" of %"PRIu64
" items), suggesting rotation.",
3967 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))),
3968 le64toh(f
->header
->n_fields
),
3969 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
));
3973 /* Are the data objects properly indexed by field objects? */
3974 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
3975 JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
3976 le64toh(f
->header
->n_data
) > 0 &&
3977 le64toh(f
->header
->n_fields
) == 0)
3980 if (max_file_usec
> 0) {
3983 h
= le64toh(f
->header
->head_entry_realtime
);
3984 t
= now(CLOCK_REALTIME
);
3986 if (h
> 0 && t
> h
+ max_file_usec
)