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(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(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 static 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
;
634 /* We assume that this file is not sparse, and we know that
635 * for sure, since we always call posix_fallocate()
638 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
642 le64toh(f
->header
->header_size
) +
643 le64toh(f
->header
->arena_size
);
645 new_size
= PAGE_ALIGN(offset
+ size
);
646 if (new_size
< le64toh(f
->header
->header_size
))
647 new_size
= le64toh(f
->header
->header_size
);
649 if (new_size
<= old_size
) {
651 /* We already pre-allocated enough space, but before
652 * we write to it, let's check with fstat() if the
653 * file got deleted, in order make sure we don't throw
654 * away the data immediately. Don't check fstat() for
655 * all writes though, but only once ever 10s. */
657 if (f
->last_stat_usec
+ LAST_STAT_REFRESH_USEC
> now(CLOCK_MONOTONIC
))
660 return journal_file_fstat(f
);
663 /* Allocate more space. */
665 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
668 if (new_size
> f
->metrics
.min_size
&& f
->metrics
.keep_free
> 0) {
671 if (fstatvfs(f
->fd
, &svfs
) >= 0) {
674 available
= LESS_BY((uint64_t) svfs
.f_bfree
* (uint64_t) svfs
.f_bsize
, f
->metrics
.keep_free
);
676 if (new_size
- old_size
> available
)
681 /* Increase by larger blocks at once */
682 new_size
= DIV_ROUND_UP(new_size
, FILE_SIZE_INCREASE
) * FILE_SIZE_INCREASE
;
683 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
684 new_size
= f
->metrics
.max_size
;
686 /* Note that the glibc fallocate() fallback is very
687 inefficient, hence we try to minimize the allocation area
689 r
= posix_fallocate(f
->fd
, old_size
, new_size
- old_size
);
693 f
->header
->arena_size
= htole64(new_size
- le64toh(f
->header
->header_size
));
695 return journal_file_fstat(f
);
698 static unsigned type_to_context(ObjectType type
) {
699 /* One context for each type, plus one catch-all for the rest */
700 assert_cc(_OBJECT_TYPE_MAX
<= MMAP_CACHE_MAX_CONTEXTS
);
701 assert_cc(CONTEXT_HEADER
< MMAP_CACHE_MAX_CONTEXTS
);
702 return type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
? type
: 0;
705 static int journal_file_move_to(JournalFile
*f
, ObjectType type
, bool keep_always
, uint64_t offset
, uint64_t size
, void **ret
, size_t *ret_size
) {
714 /* Avoid SIGBUS on invalid accesses */
715 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
) {
716 /* Hmm, out of range? Let's refresh the fstat() data
717 * first, before we trust that check. */
719 r
= journal_file_fstat(f
);
723 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
)
724 return -EADDRNOTAVAIL
;
727 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
);
730 static uint64_t minimum_header_size(Object
*o
) {
732 static const uint64_t table
[] = {
733 [OBJECT_DATA
] = sizeof(DataObject
),
734 [OBJECT_FIELD
] = sizeof(FieldObject
),
735 [OBJECT_ENTRY
] = sizeof(EntryObject
),
736 [OBJECT_DATA_HASH_TABLE
] = sizeof(HashTableObject
),
737 [OBJECT_FIELD_HASH_TABLE
] = sizeof(HashTableObject
),
738 [OBJECT_ENTRY_ARRAY
] = sizeof(EntryArrayObject
),
739 [OBJECT_TAG
] = sizeof(TagObject
),
742 if (o
->object
.type
>= ELEMENTSOF(table
) || table
[o
->object
.type
] <= 0)
743 return sizeof(ObjectHeader
);
745 return table
[o
->object
.type
];
748 /* Lightweight object checks. We want this to be fast, so that we won't
749 * slowdown every journal_file_move_to_object() call too much. */
750 static int journal_file_check_object(JournalFile
*f
, uint64_t offset
, Object
*o
) {
754 switch (o
->object
.type
) {
757 if ((le64toh(o
->data
.entry_offset
) == 0) ^ (le64toh(o
->data
.n_entries
) == 0))
758 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
759 "Bad n_entries: %" PRIu64
": %" PRIu64
,
760 le64toh(o
->data
.n_entries
),
763 if (le64toh(o
->object
.size
) - offsetof(DataObject
, payload
) <= 0)
764 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
765 "Bad object size (<= %zu): %" PRIu64
": %" PRIu64
,
766 offsetof(DataObject
, payload
),
767 le64toh(o
->object
.size
),
770 if (!VALID64(le64toh(o
->data
.next_hash_offset
)) ||
771 !VALID64(le64toh(o
->data
.next_field_offset
)) ||
772 !VALID64(le64toh(o
->data
.entry_offset
)) ||
773 !VALID64(le64toh(o
->data
.entry_array_offset
)))
774 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
775 "Invalid offset, next_hash_offset=" OFSfmt
", next_field_offset=" OFSfmt
", entry_offset=" OFSfmt
", entry_array_offset=" OFSfmt
": %" PRIu64
,
776 le64toh(o
->data
.next_hash_offset
),
777 le64toh(o
->data
.next_field_offset
),
778 le64toh(o
->data
.entry_offset
),
779 le64toh(o
->data
.entry_array_offset
),
785 if (le64toh(o
->object
.size
) - offsetof(FieldObject
, payload
) <= 0)
786 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
787 "Bad field size (<= %zu): %" PRIu64
": %" PRIu64
,
788 offsetof(FieldObject
, payload
),
789 le64toh(o
->object
.size
),
792 if (!VALID64(le64toh(o
->field
.next_hash_offset
)) ||
793 !VALID64(le64toh(o
->field
.head_data_offset
)))
794 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
795 "Invalid offset, next_hash_offset=" OFSfmt
", head_data_offset=" OFSfmt
": %" PRIu64
,
796 le64toh(o
->field
.next_hash_offset
),
797 le64toh(o
->field
.head_data_offset
),
802 if ((le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) % sizeof(EntryItem
) != 0)
803 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
804 "Bad entry size (<= %zu): %" PRIu64
": %" PRIu64
,
805 offsetof(EntryObject
, items
),
806 le64toh(o
->object
.size
),
809 if ((le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) / sizeof(EntryItem
) <= 0)
810 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
811 "Invalid number items in entry: %" PRIu64
": %" PRIu64
,
812 (le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) / sizeof(EntryItem
),
815 if (le64toh(o
->entry
.seqnum
) <= 0)
816 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
817 "Invalid entry seqnum: %" PRIx64
": %" PRIu64
,
818 le64toh(o
->entry
.seqnum
),
821 if (!VALID_REALTIME(le64toh(o
->entry
.realtime
)))
822 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
823 "Invalid entry realtime timestamp: %" PRIu64
": %" PRIu64
,
824 le64toh(o
->entry
.realtime
),
827 if (!VALID_MONOTONIC(le64toh(o
->entry
.monotonic
)))
828 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
829 "Invalid entry monotonic timestamp: %" PRIu64
": %" PRIu64
,
830 le64toh(o
->entry
.monotonic
),
835 case OBJECT_DATA_HASH_TABLE
:
836 case OBJECT_FIELD_HASH_TABLE
:
837 if ((le64toh(o
->object
.size
) - offsetof(HashTableObject
, items
)) % sizeof(HashItem
) != 0 ||
838 (le64toh(o
->object
.size
) - offsetof(HashTableObject
, items
)) / sizeof(HashItem
) <= 0)
839 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
840 "Invalid %s hash table size: %" PRIu64
": %" PRIu64
,
841 o
->object
.type
== OBJECT_DATA_HASH_TABLE
? "data" : "field",
842 le64toh(o
->object
.size
),
847 case OBJECT_ENTRY_ARRAY
:
848 if ((le64toh(o
->object
.size
) - offsetof(EntryArrayObject
, items
)) % sizeof(le64_t
) != 0 ||
849 (le64toh(o
->object
.size
) - offsetof(EntryArrayObject
, items
)) / sizeof(le64_t
) <= 0)
850 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
851 "Invalid object entry array size: %" PRIu64
": %" PRIu64
,
852 le64toh(o
->object
.size
),
855 if (!VALID64(le64toh(o
->entry_array
.next_entry_array_offset
)))
856 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
857 "Invalid object entry array next_entry_array_offset: " OFSfmt
": %" PRIu64
,
858 le64toh(o
->entry_array
.next_entry_array_offset
),
864 if (le64toh(o
->object
.size
) != sizeof(TagObject
))
865 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
866 "Invalid object tag size: %" PRIu64
": %" PRIu64
,
867 le64toh(o
->object
.size
),
870 if (!VALID_EPOCH(le64toh(o
->tag
.epoch
)))
871 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
872 "Invalid object tag epoch: %" PRIu64
": %" PRIu64
,
873 le64toh(o
->tag
.epoch
), offset
);
881 int journal_file_move_to_object(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
**ret
) {
891 /* Objects may only be located at multiple of 64 bit */
892 if (!VALID64(offset
))
893 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
894 "Attempt to move to object at non-64bit boundary: %" PRIu64
,
897 /* Object may not be located in the file header */
898 if (offset
< le64toh(f
->header
->header_size
))
899 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
900 "Attempt to move to object located in file header: %" PRIu64
,
903 r
= journal_file_move_to(f
, type
, false, offset
, sizeof(ObjectHeader
), &t
, &tsize
);
908 s
= le64toh(o
->object
.size
);
911 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
912 "Attempt to move to uninitialized object: %" PRIu64
,
914 if (s
< sizeof(ObjectHeader
))
915 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
916 "Attempt to move to overly short object: %" PRIu64
,
919 if (o
->object
.type
<= OBJECT_UNUSED
)
920 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
921 "Attempt to move to object with invalid type: %" PRIu64
,
924 if (s
< minimum_header_size(o
))
925 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
926 "Attempt to move to truncated object: %" PRIu64
,
929 if (type
> OBJECT_UNUSED
&& o
->object
.type
!= type
)
930 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
931 "Attempt to move to object of unexpected type: %" PRIu64
,
935 r
= journal_file_move_to(f
, type
, false, offset
, s
, &t
, NULL
);
942 r
= journal_file_check_object(f
, offset
, o
);
950 static uint64_t journal_file_entry_seqnum(JournalFile
*f
, uint64_t *seqnum
) {
956 r
= le64toh(f
->header
->tail_entry_seqnum
) + 1;
959 /* If an external seqnum counter was passed, we update
960 * both the local and the external one, and set it to
961 * the maximum of both */
969 f
->header
->tail_entry_seqnum
= htole64(r
);
971 if (f
->header
->head_entry_seqnum
== 0)
972 f
->header
->head_entry_seqnum
= htole64(r
);
977 int journal_file_append_object(JournalFile
*f
, ObjectType type
, uint64_t size
, Object
**ret
, uint64_t *offset
) {
985 assert(type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
);
986 assert(size
>= sizeof(ObjectHeader
));
990 r
= journal_file_set_online(f
);
994 p
= le64toh(f
->header
->tail_object_offset
);
996 p
= le64toh(f
->header
->header_size
);
998 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &tail
);
1002 p
+= ALIGN64(le64toh(tail
->object
.size
));
1005 r
= journal_file_allocate(f
, p
, size
);
1009 r
= journal_file_move_to(f
, type
, false, p
, size
, &t
, NULL
);
1016 o
->object
.type
= type
;
1017 o
->object
.size
= htole64(size
);
1019 f
->header
->tail_object_offset
= htole64(p
);
1020 f
->header
->n_objects
= htole64(le64toh(f
->header
->n_objects
) + 1);
1028 static int journal_file_setup_data_hash_table(JournalFile
*f
) {
1036 /* We estimate that we need 1 hash table entry per 768 bytes
1037 of journal file and we want to make sure we never get
1038 beyond 75% fill level. Calculate the hash table size for
1039 the maximum file size based on these metrics. */
1041 s
= (f
->metrics
.max_size
* 4 / 768 / 3) * sizeof(HashItem
);
1042 if (s
< DEFAULT_DATA_HASH_TABLE_SIZE
)
1043 s
= DEFAULT_DATA_HASH_TABLE_SIZE
;
1045 log_debug("Reserving %"PRIu64
" entries in hash table.", s
/ sizeof(HashItem
));
1047 r
= journal_file_append_object(f
,
1048 OBJECT_DATA_HASH_TABLE
,
1049 offsetof(Object
, hash_table
.items
) + s
,
1054 memzero(o
->hash_table
.items
, s
);
1056 f
->header
->data_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1057 f
->header
->data_hash_table_size
= htole64(s
);
1062 static int journal_file_setup_field_hash_table(JournalFile
*f
) {
1070 /* We use a fixed size hash table for the fields as this
1071 * number should grow very slowly only */
1073 s
= DEFAULT_FIELD_HASH_TABLE_SIZE
;
1074 r
= journal_file_append_object(f
,
1075 OBJECT_FIELD_HASH_TABLE
,
1076 offsetof(Object
, hash_table
.items
) + s
,
1081 memzero(o
->hash_table
.items
, s
);
1083 f
->header
->field_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1084 f
->header
->field_hash_table_size
= htole64(s
);
1089 int journal_file_map_data_hash_table(JournalFile
*f
) {
1097 if (f
->data_hash_table
)
1100 p
= le64toh(f
->header
->data_hash_table_offset
);
1101 s
= le64toh(f
->header
->data_hash_table_size
);
1103 r
= journal_file_move_to(f
,
1104 OBJECT_DATA_HASH_TABLE
,
1111 f
->data_hash_table
= t
;
1115 int journal_file_map_field_hash_table(JournalFile
*f
) {
1123 if (f
->field_hash_table
)
1126 p
= le64toh(f
->header
->field_hash_table_offset
);
1127 s
= le64toh(f
->header
->field_hash_table_size
);
1129 r
= journal_file_move_to(f
,
1130 OBJECT_FIELD_HASH_TABLE
,
1137 f
->field_hash_table
= t
;
1141 static int journal_file_link_field(
1152 assert(f
->field_hash_table
);
1156 if (o
->object
.type
!= OBJECT_FIELD
)
1159 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1163 /* This might alter the window we are looking at */
1164 o
->field
.next_hash_offset
= o
->field
.head_data_offset
= 0;
1167 p
= le64toh(f
->field_hash_table
[h
].tail_hash_offset
);
1169 f
->field_hash_table
[h
].head_hash_offset
= htole64(offset
);
1171 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1175 o
->field
.next_hash_offset
= htole64(offset
);
1178 f
->field_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1180 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
1181 f
->header
->n_fields
= htole64(le64toh(f
->header
->n_fields
) + 1);
1186 static int journal_file_link_data(
1197 assert(f
->data_hash_table
);
1201 if (o
->object
.type
!= OBJECT_DATA
)
1204 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1208 /* This might alter the window we are looking at */
1209 o
->data
.next_hash_offset
= o
->data
.next_field_offset
= 0;
1210 o
->data
.entry_offset
= o
->data
.entry_array_offset
= 0;
1211 o
->data
.n_entries
= 0;
1214 p
= le64toh(f
->data_hash_table
[h
].tail_hash_offset
);
1216 /* Only entry in the hash table is easy */
1217 f
->data_hash_table
[h
].head_hash_offset
= htole64(offset
);
1219 /* Move back to the previous data object, to patch in
1222 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1226 o
->data
.next_hash_offset
= htole64(offset
);
1229 f
->data_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1231 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
1232 f
->header
->n_data
= htole64(le64toh(f
->header
->n_data
) + 1);
1237 int journal_file_find_field_object_with_hash(
1239 const void *field
, uint64_t size
, uint64_t hash
,
1240 Object
**ret
, uint64_t *offset
) {
1242 uint64_t p
, osize
, h
, m
;
1247 assert(field
&& size
> 0);
1249 /* If the field hash table is empty, we can't find anything */
1250 if (le64toh(f
->header
->field_hash_table_size
) <= 0)
1253 /* Map the field hash table, if it isn't mapped yet. */
1254 r
= journal_file_map_field_hash_table(f
);
1258 osize
= offsetof(Object
, field
.payload
) + size
;
1260 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1265 p
= le64toh(f
->field_hash_table
[h
].head_hash_offset
);
1270 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1274 if (le64toh(o
->field
.hash
) == hash
&&
1275 le64toh(o
->object
.size
) == osize
&&
1276 memcmp(o
->field
.payload
, field
, size
) == 0) {
1286 p
= le64toh(o
->field
.next_hash_offset
);
1292 int journal_file_find_field_object(
1294 const void *field
, uint64_t size
,
1295 Object
**ret
, uint64_t *offset
) {
1300 assert(field
&& size
> 0);
1302 hash
= hash64(field
, size
);
1304 return journal_file_find_field_object_with_hash(f
,
1309 int journal_file_find_data_object_with_hash(
1311 const void *data
, uint64_t size
, uint64_t hash
,
1312 Object
**ret
, uint64_t *offset
) {
1314 uint64_t p
, osize
, h
, m
;
1319 assert(data
|| size
== 0);
1321 /* If there's no data hash table, then there's no entry. */
1322 if (le64toh(f
->header
->data_hash_table_size
) <= 0)
1325 /* Map the data hash table, if it isn't mapped yet. */
1326 r
= journal_file_map_data_hash_table(f
);
1330 osize
= offsetof(Object
, data
.payload
) + size
;
1332 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1337 p
= le64toh(f
->data_hash_table
[h
].head_hash_offset
);
1342 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1346 if (le64toh(o
->data
.hash
) != hash
)
1349 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
1350 #if HAVE_XZ || HAVE_LZ4
1354 l
= le64toh(o
->object
.size
);
1355 if (l
<= offsetof(Object
, data
.payload
))
1358 l
-= offsetof(Object
, data
.payload
);
1360 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
1361 o
->data
.payload
, l
, &f
->compress_buffer
, &f
->compress_buffer_size
, &rsize
, 0);
1365 if (rsize
== size
&&
1366 memcmp(f
->compress_buffer
, data
, size
) == 0) {
1377 return -EPROTONOSUPPORT
;
1379 } else if (le64toh(o
->object
.size
) == osize
&&
1380 memcmp(o
->data
.payload
, data
, size
) == 0) {
1392 p
= le64toh(o
->data
.next_hash_offset
);
1398 int journal_file_find_data_object(
1400 const void *data
, uint64_t size
,
1401 Object
**ret
, uint64_t *offset
) {
1406 assert(data
|| size
== 0);
1408 hash
= hash64(data
, size
);
1410 return journal_file_find_data_object_with_hash(f
,
1415 static int journal_file_append_field(
1417 const void *field
, uint64_t size
,
1418 Object
**ret
, uint64_t *offset
) {
1426 assert(field
&& size
> 0);
1428 hash
= hash64(field
, size
);
1430 r
= journal_file_find_field_object_with_hash(f
, field
, size
, hash
, &o
, &p
);
1444 osize
= offsetof(Object
, field
.payload
) + size
;
1445 r
= journal_file_append_object(f
, OBJECT_FIELD
, osize
, &o
, &p
);
1449 o
->field
.hash
= htole64(hash
);
1450 memcpy(o
->field
.payload
, field
, size
);
1452 r
= journal_file_link_field(f
, o
, p
, hash
);
1456 /* The linking might have altered the window, so let's
1457 * refresh our pointer */
1458 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1463 r
= journal_file_hmac_put_object(f
, OBJECT_FIELD
, o
, p
);
1477 static int journal_file_append_data(
1479 const void *data
, uint64_t size
,
1480 Object
**ret
, uint64_t *offset
) {
1485 int r
, compression
= 0;
1489 assert(data
|| size
== 0);
1491 hash
= hash64(data
, size
);
1493 r
= journal_file_find_data_object_with_hash(f
, data
, size
, hash
, &o
, &p
);
1507 osize
= offsetof(Object
, data
.payload
) + size
;
1508 r
= journal_file_append_object(f
, OBJECT_DATA
, osize
, &o
, &p
);
1512 o
->data
.hash
= htole64(hash
);
1514 #if HAVE_XZ || HAVE_LZ4
1515 if (JOURNAL_FILE_COMPRESS(f
) && size
>= f
->compress_threshold_bytes
) {
1518 compression
= compress_blob(data
, size
, o
->data
.payload
, size
- 1, &rsize
);
1520 if (compression
>= 0) {
1521 o
->object
.size
= htole64(offsetof(Object
, data
.payload
) + rsize
);
1522 o
->object
.flags
|= compression
;
1524 log_debug("Compressed data object %"PRIu64
" -> %zu using %s",
1525 size
, rsize
, object_compressed_to_string(compression
));
1527 /* Compression didn't work, we don't really care why, let's continue without compression */
1532 if (compression
== 0)
1533 memcpy_safe(o
->data
.payload
, data
, size
);
1535 r
= journal_file_link_data(f
, o
, p
, hash
);
1540 r
= journal_file_hmac_put_object(f
, OBJECT_DATA
, o
, p
);
1545 /* The linking might have altered the window, so let's
1546 * refresh our pointer */
1547 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1554 eq
= memchr(data
, '=', size
);
1555 if (eq
&& eq
> data
) {
1559 /* Create field object ... */
1560 r
= journal_file_append_field(f
, data
, (uint8_t*) eq
- (uint8_t*) data
, &fo
, &fp
);
1564 /* ... and link it in. */
1565 o
->data
.next_field_offset
= fo
->field
.head_data_offset
;
1566 fo
->field
.head_data_offset
= le64toh(p
);
1578 uint64_t journal_file_entry_n_items(Object
*o
) {
1581 if (o
->object
.type
!= OBJECT_ENTRY
)
1584 return (le64toh(o
->object
.size
) - offsetof(Object
, entry
.items
)) / sizeof(EntryItem
);
1587 uint64_t journal_file_entry_array_n_items(Object
*o
) {
1590 if (o
->object
.type
!= OBJECT_ENTRY_ARRAY
)
1593 return (le64toh(o
->object
.size
) - offsetof(Object
, entry_array
.items
)) / sizeof(uint64_t);
1596 uint64_t journal_file_hash_table_n_items(Object
*o
) {
1599 if (!IN_SET(o
->object
.type
, OBJECT_DATA_HASH_TABLE
, OBJECT_FIELD_HASH_TABLE
))
1602 return (le64toh(o
->object
.size
) - offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
);
1605 static int link_entry_into_array(JournalFile
*f
,
1610 uint64_t n
= 0, ap
= 0, q
, i
, a
, hidx
;
1619 a
= le64toh(*first
);
1620 i
= hidx
= le64toh(*idx
);
1623 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
1627 n
= journal_file_entry_array_n_items(o
);
1629 o
->entry_array
.items
[i
] = htole64(p
);
1630 *idx
= htole64(hidx
+ 1);
1636 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
1647 r
= journal_file_append_object(f
, OBJECT_ENTRY_ARRAY
,
1648 offsetof(Object
, entry_array
.items
) + n
* sizeof(uint64_t),
1654 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY_ARRAY
, o
, q
);
1659 o
->entry_array
.items
[i
] = htole64(p
);
1662 *first
= htole64(q
);
1664 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, ap
, &o
);
1668 o
->entry_array
.next_entry_array_offset
= htole64(q
);
1671 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
1672 f
->header
->n_entry_arrays
= htole64(le64toh(f
->header
->n_entry_arrays
) + 1);
1674 *idx
= htole64(hidx
+ 1);
1679 static int link_entry_into_array_plus_one(JournalFile
*f
,
1694 *extra
= htole64(p
);
1698 i
= htole64(le64toh(*idx
) - 1);
1699 r
= link_entry_into_array(f
, first
, &i
, p
);
1704 *idx
= htole64(le64toh(*idx
) + 1);
1708 static int journal_file_link_entry_item(JournalFile
*f
, Object
*o
, uint64_t offset
, uint64_t i
) {
1715 p
= le64toh(o
->entry
.items
[i
].object_offset
);
1719 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1723 return link_entry_into_array_plus_one(f
,
1724 &o
->data
.entry_offset
,
1725 &o
->data
.entry_array_offset
,
1730 static int journal_file_link_entry(JournalFile
*f
, Object
*o
, uint64_t offset
) {
1739 if (o
->object
.type
!= OBJECT_ENTRY
)
1742 __sync_synchronize();
1744 /* Link up the entry itself */
1745 r
= link_entry_into_array(f
,
1746 &f
->header
->entry_array_offset
,
1747 &f
->header
->n_entries
,
1752 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
1754 if (f
->header
->head_entry_realtime
== 0)
1755 f
->header
->head_entry_realtime
= o
->entry
.realtime
;
1757 f
->header
->tail_entry_realtime
= o
->entry
.realtime
;
1758 f
->header
->tail_entry_monotonic
= o
->entry
.monotonic
;
1760 /* Link up the items */
1761 n
= journal_file_entry_n_items(o
);
1762 for (i
= 0; i
< n
; i
++) {
1763 r
= journal_file_link_entry_item(f
, o
, offset
, i
);
1771 static int journal_file_append_entry_internal(
1773 const dual_timestamp
*ts
,
1774 const sd_id128_t
*boot_id
,
1776 const EntryItem items
[], unsigned n_items
,
1778 Object
**ret
, uint64_t *offset
) {
1786 assert(items
|| n_items
== 0);
1789 osize
= offsetof(Object
, entry
.items
) + (n_items
* sizeof(EntryItem
));
1791 r
= journal_file_append_object(f
, OBJECT_ENTRY
, osize
, &o
, &np
);
1795 o
->entry
.seqnum
= htole64(journal_file_entry_seqnum(f
, seqnum
));
1796 memcpy_safe(o
->entry
.items
, items
, n_items
* sizeof(EntryItem
));
1797 o
->entry
.realtime
= htole64(ts
->realtime
);
1798 o
->entry
.monotonic
= htole64(ts
->monotonic
);
1799 o
->entry
.xor_hash
= htole64(xor_hash
);
1801 f
->header
->boot_id
= *boot_id
;
1802 o
->entry
.boot_id
= f
->header
->boot_id
;
1805 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY
, o
, np
);
1810 r
= journal_file_link_entry(f
, o
, np
);
1823 void journal_file_post_change(JournalFile
*f
) {
1829 /* inotify() does not receive IN_MODIFY events from file
1830 * accesses done via mmap(). After each access we hence
1831 * trigger IN_MODIFY by truncating the journal file to its
1832 * current size which triggers IN_MODIFY. */
1834 __sync_synchronize();
1836 if (ftruncate(f
->fd
, f
->last_stat
.st_size
) < 0)
1837 log_debug_errno(errno
, "Failed to truncate file to its own size: %m");
1840 static int post_change_thunk(sd_event_source
*timer
, uint64_t usec
, void *userdata
) {
1843 journal_file_post_change(userdata
);
1848 static void schedule_post_change(JournalFile
*f
) {
1853 assert(f
->post_change_timer
);
1855 r
= sd_event_source_get_enabled(f
->post_change_timer
, NULL
);
1857 log_debug_errno(r
, "Failed to get ftruncate timer state: %m");
1863 r
= sd_event_now(sd_event_source_get_event(f
->post_change_timer
), CLOCK_MONOTONIC
, &now
);
1865 log_debug_errno(r
, "Failed to get clock's now for scheduling ftruncate: %m");
1869 r
= sd_event_source_set_time(f
->post_change_timer
, now
+ f
->post_change_timer_period
);
1871 log_debug_errno(r
, "Failed to set time for scheduling ftruncate: %m");
1875 r
= sd_event_source_set_enabled(f
->post_change_timer
, SD_EVENT_ONESHOT
);
1877 log_debug_errno(r
, "Failed to enable scheduled ftruncate: %m");
1884 /* On failure, let's simply post the change immediately. */
1885 journal_file_post_change(f
);
1888 /* Enable coalesced change posting in a timer on the provided sd_event instance */
1889 int journal_file_enable_post_change_timer(JournalFile
*f
, sd_event
*e
, usec_t t
) {
1890 _cleanup_(sd_event_source_unrefp
) sd_event_source
*timer
= NULL
;
1894 assert_return(!f
->post_change_timer
, -EINVAL
);
1898 r
= sd_event_add_time(e
, &timer
, CLOCK_MONOTONIC
, 0, 0, post_change_thunk
, f
);
1902 r
= sd_event_source_set_enabled(timer
, SD_EVENT_OFF
);
1906 f
->post_change_timer
= TAKE_PTR(timer
);
1907 f
->post_change_timer_period
= t
;
1912 static int entry_item_cmp(const EntryItem
*a
, const EntryItem
*b
) {
1913 return CMP(le64toh(a
->object_offset
), le64toh(b
->object_offset
));
1916 int journal_file_append_entry(
1918 const dual_timestamp
*ts
,
1919 const sd_id128_t
*boot_id
,
1920 const struct iovec iovec
[], unsigned n_iovec
,
1922 Object
**ret
, uint64_t *offset
) {
1927 uint64_t xor_hash
= 0;
1928 struct dual_timestamp _ts
;
1932 assert(iovec
|| n_iovec
== 0);
1935 if (!VALID_REALTIME(ts
->realtime
))
1936 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1937 "Invalid realtime timestamp %" PRIu64
", refusing entry.",
1939 if (!VALID_MONOTONIC(ts
->monotonic
))
1940 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1941 "Invalid monotomic timestamp %" PRIu64
", refusing entry.",
1944 dual_timestamp_get(&_ts
);
1949 r
= journal_file_maybe_append_tag(f
, ts
->realtime
);
1954 /* alloca() can't take 0, hence let's allocate at least one */
1955 items
= newa(EntryItem
, MAX(1u, n_iovec
));
1957 for (i
= 0; i
< n_iovec
; i
++) {
1961 r
= journal_file_append_data(f
, iovec
[i
].iov_base
, iovec
[i
].iov_len
, &o
, &p
);
1965 xor_hash
^= le64toh(o
->data
.hash
);
1966 items
[i
].object_offset
= htole64(p
);
1967 items
[i
].hash
= o
->data
.hash
;
1970 /* Order by the position on disk, in order to improve seek
1971 * times for rotating media. */
1972 typesafe_qsort(items
, n_iovec
, entry_item_cmp
);
1974 r
= journal_file_append_entry_internal(f
, ts
, boot_id
, xor_hash
, items
, n_iovec
, seqnum
, ret
, offset
);
1976 /* If the memory mapping triggered a SIGBUS then we return an
1977 * IO error and ignore the error code passed down to us, since
1978 * it is very likely just an effect of a nullified replacement
1981 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
1984 if (f
->post_change_timer
)
1985 schedule_post_change(f
);
1987 journal_file_post_change(f
);
1992 typedef struct ChainCacheItem
{
1993 uint64_t first
; /* the array at the beginning of the chain */
1994 uint64_t array
; /* the cached array */
1995 uint64_t begin
; /* the first item in the cached array */
1996 uint64_t total
; /* the total number of items in all arrays before this one in the chain */
1997 uint64_t last_index
; /* the last index we looked at, to optimize locality when bisecting */
2000 static void chain_cache_put(
2007 uint64_t last_index
) {
2010 /* If the chain item to cache for this chain is the
2011 * first one it's not worth caching anything */
2015 if (ordered_hashmap_size(h
) >= CHAIN_CACHE_MAX
) {
2016 ci
= ordered_hashmap_steal_first(h
);
2019 ci
= new(ChainCacheItem
, 1);
2026 if (ordered_hashmap_put(h
, &ci
->first
, ci
) < 0) {
2031 assert(ci
->first
== first
);
2036 ci
->last_index
= last_index
;
2039 static int generic_array_get(
2043 Object
**ret
, uint64_t *offset
) {
2046 uint64_t p
= 0, a
, t
= 0;
2054 /* Try the chain cache first */
2055 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2056 if (ci
&& i
> ci
->total
) {
2065 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2069 k
= journal_file_entry_array_n_items(o
);
2071 p
= le64toh(o
->entry_array
.items
[i
]);
2077 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
2083 /* Let's cache this item for the next invocation */
2084 chain_cache_put(f
->chain_cache
, ci
, first
, a
, le64toh(o
->entry_array
.items
[0]), t
, i
);
2086 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2099 static int generic_array_get_plus_one(
2104 Object
**ret
, uint64_t *offset
) {
2113 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2126 return generic_array_get(f
, first
, i
-1, ret
, offset
);
2135 static int generic_array_bisect(
2140 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2141 direction_t direction
,
2146 uint64_t a
, p
, t
= 0, i
= 0, last_p
= 0, last_index
= (uint64_t) -1;
2147 bool subtract_one
= false;
2148 Object
*o
, *array
= NULL
;
2153 assert(test_object
);
2155 /* Start with the first array in the chain */
2158 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2159 if (ci
&& n
> ci
->total
&& ci
->begin
!= 0) {
2160 /* Ah, we have iterated this bisection array chain
2161 * previously! Let's see if we can skip ahead in the
2162 * chain, as far as the last time. But we can't jump
2163 * backwards in the chain, so let's check that
2166 r
= test_object(f
, ci
->begin
, needle
);
2170 if (r
== TEST_LEFT
) {
2171 /* OK, what we are looking for is right of the
2172 * begin of this EntryArray, so let's jump
2173 * straight to previously cached array in the
2179 last_index
= ci
->last_index
;
2184 uint64_t left
, right
, k
, lp
;
2186 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2190 k
= journal_file_entry_array_n_items(array
);
2196 lp
= p
= le64toh(array
->entry_array
.items
[i
]);
2200 r
= test_object(f
, p
, needle
);
2201 if (r
== -EBADMSG
) {
2202 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (1)");
2209 if (r
== TEST_FOUND
)
2210 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2212 if (r
== TEST_RIGHT
) {
2216 if (last_index
!= (uint64_t) -1) {
2217 assert(last_index
<= right
);
2219 /* If we cached the last index we
2220 * looked at, let's try to not to jump
2221 * too wildly around and see if we can
2222 * limit the range to look at early to
2223 * the immediate neighbors of the last
2224 * index we looked at. */
2226 if (last_index
> 0) {
2227 uint64_t x
= last_index
- 1;
2229 p
= le64toh(array
->entry_array
.items
[x
]);
2233 r
= test_object(f
, p
, needle
);
2237 if (r
== TEST_FOUND
)
2238 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2240 if (r
== TEST_RIGHT
)
2246 if (last_index
< right
) {
2247 uint64_t y
= last_index
+ 1;
2249 p
= le64toh(array
->entry_array
.items
[y
]);
2253 r
= test_object(f
, p
, needle
);
2257 if (r
== TEST_FOUND
)
2258 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2260 if (r
== TEST_RIGHT
)
2268 if (left
== right
) {
2269 if (direction
== DIRECTION_UP
)
2270 subtract_one
= true;
2276 assert(left
< right
);
2277 i
= (left
+ right
) / 2;
2279 p
= le64toh(array
->entry_array
.items
[i
]);
2283 r
= test_object(f
, p
, needle
);
2284 if (r
== -EBADMSG
) {
2285 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (2)");
2292 if (r
== TEST_FOUND
)
2293 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2295 if (r
== TEST_RIGHT
)
2303 if (direction
== DIRECTION_UP
) {
2305 subtract_one
= true;
2316 last_index
= (uint64_t) -1;
2317 a
= le64toh(array
->entry_array
.next_entry_array_offset
);
2323 if (subtract_one
&& t
== 0 && i
== 0)
2326 /* Let's cache this item for the next invocation */
2327 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
);
2329 if (subtract_one
&& i
== 0)
2331 else if (subtract_one
)
2332 p
= le64toh(array
->entry_array
.items
[i
-1]);
2334 p
= le64toh(array
->entry_array
.items
[i
]);
2336 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2347 *idx
= t
+ i
+ (subtract_one
? -1 : 0);
2352 static int generic_array_bisect_plus_one(
2358 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2359 direction_t direction
,
2365 bool step_back
= false;
2369 assert(test_object
);
2374 /* This bisects the array in object 'first', but first checks
2376 r
= test_object(f
, extra
, needle
);
2380 if (r
== TEST_FOUND
)
2381 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2383 /* if we are looking with DIRECTION_UP then we need to first
2384 see if in the actual array there is a matching entry, and
2385 return the last one of that. But if there isn't any we need
2386 to return this one. Hence remember this, and return it
2389 step_back
= direction
== DIRECTION_UP
;
2391 if (r
== TEST_RIGHT
) {
2392 if (direction
== DIRECTION_DOWN
)
2398 r
= generic_array_bisect(f
, first
, n
-1, needle
, test_object
, direction
, ret
, offset
, idx
);
2400 if (r
== 0 && step_back
)
2409 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2425 _pure_
static int test_object_offset(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2431 else if (p
< needle
)
2437 static int test_object_seqnum(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2444 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2448 if (le64toh(o
->entry
.seqnum
) == needle
)
2450 else if (le64toh(o
->entry
.seqnum
) < needle
)
2456 int journal_file_move_to_entry_by_seqnum(
2459 direction_t direction
,
2465 return generic_array_bisect(f
,
2466 le64toh(f
->header
->entry_array_offset
),
2467 le64toh(f
->header
->n_entries
),
2474 static int test_object_realtime(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2481 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2485 if (le64toh(o
->entry
.realtime
) == needle
)
2487 else if (le64toh(o
->entry
.realtime
) < needle
)
2493 int journal_file_move_to_entry_by_realtime(
2496 direction_t direction
,
2502 return generic_array_bisect(f
,
2503 le64toh(f
->header
->entry_array_offset
),
2504 le64toh(f
->header
->n_entries
),
2506 test_object_realtime
,
2511 static int test_object_monotonic(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2518 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2522 if (le64toh(o
->entry
.monotonic
) == needle
)
2524 else if (le64toh(o
->entry
.monotonic
) < needle
)
2530 static int find_data_object_by_boot_id(
2536 char t
[STRLEN("_BOOT_ID=") + 32 + 1] = "_BOOT_ID=";
2538 sd_id128_to_string(boot_id
, t
+ 9);
2539 return journal_file_find_data_object(f
, t
, sizeof(t
) - 1, o
, b
);
2542 int journal_file_move_to_entry_by_monotonic(
2546 direction_t direction
,
2555 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, NULL
);
2561 return generic_array_bisect_plus_one(f
,
2562 le64toh(o
->data
.entry_offset
),
2563 le64toh(o
->data
.entry_array_offset
),
2564 le64toh(o
->data
.n_entries
),
2566 test_object_monotonic
,
2571 void journal_file_reset_location(JournalFile
*f
) {
2572 f
->location_type
= LOCATION_HEAD
;
2573 f
->current_offset
= 0;
2574 f
->current_seqnum
= 0;
2575 f
->current_realtime
= 0;
2576 f
->current_monotonic
= 0;
2577 zero(f
->current_boot_id
);
2578 f
->current_xor_hash
= 0;
2581 void journal_file_save_location(JournalFile
*f
, Object
*o
, uint64_t offset
) {
2582 f
->location_type
= LOCATION_SEEK
;
2583 f
->current_offset
= offset
;
2584 f
->current_seqnum
= le64toh(o
->entry
.seqnum
);
2585 f
->current_realtime
= le64toh(o
->entry
.realtime
);
2586 f
->current_monotonic
= le64toh(o
->entry
.monotonic
);
2587 f
->current_boot_id
= o
->entry
.boot_id
;
2588 f
->current_xor_hash
= le64toh(o
->entry
.xor_hash
);
2591 int journal_file_compare_locations(JournalFile
*af
, JournalFile
*bf
) {
2598 assert(af
->location_type
== LOCATION_SEEK
);
2599 assert(bf
->location_type
== LOCATION_SEEK
);
2601 /* If contents and timestamps match, these entries are
2602 * identical, even if the seqnum does not match */
2603 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
) &&
2604 af
->current_monotonic
== bf
->current_monotonic
&&
2605 af
->current_realtime
== bf
->current_realtime
&&
2606 af
->current_xor_hash
== bf
->current_xor_hash
)
2609 if (sd_id128_equal(af
->header
->seqnum_id
, bf
->header
->seqnum_id
)) {
2611 /* If this is from the same seqnum source, compare
2613 r
= CMP(af
->current_seqnum
, bf
->current_seqnum
);
2617 /* Wow! This is weird, different data but the same
2618 * seqnums? Something is borked, but let's make the
2619 * best of it and compare by time. */
2622 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
)) {
2624 /* If the boot id matches, compare monotonic time */
2625 r
= CMP(af
->current_monotonic
, bf
->current_monotonic
);
2630 /* Otherwise, compare UTC time */
2631 r
= CMP(af
->current_realtime
, bf
->current_realtime
);
2635 /* Finally, compare by contents */
2636 return CMP(af
->current_xor_hash
, bf
->current_xor_hash
);
2639 static int bump_array_index(uint64_t *i
, direction_t direction
, uint64_t n
) {
2641 /* Increase or decrease the specified index, in the right direction. */
2643 if (direction
== DIRECTION_DOWN
) {
2658 static bool check_properly_ordered(uint64_t new_offset
, uint64_t old_offset
, direction_t direction
) {
2660 /* Consider it an error if any of the two offsets is uninitialized */
2661 if (old_offset
== 0 || new_offset
== 0)
2664 /* If we go down, the new offset must be larger than the old one. */
2665 return direction
== DIRECTION_DOWN
?
2666 new_offset
> old_offset
:
2667 new_offset
< old_offset
;
2670 int journal_file_next_entry(
2673 direction_t direction
,
2674 Object
**ret
, uint64_t *offset
) {
2682 n
= le64toh(f
->header
->n_entries
);
2687 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2689 r
= generic_array_bisect(f
,
2690 le64toh(f
->header
->entry_array_offset
),
2691 le64toh(f
->header
->n_entries
),
2700 r
= bump_array_index(&i
, direction
, n
);
2705 /* And jump to it */
2707 r
= generic_array_get(f
,
2708 le64toh(f
->header
->entry_array_offset
),
2716 /* OK, so this entry is borked. Most likely some entry didn't get synced to disk properly, let's see if
2717 * the next one might work for us instead. */
2718 log_debug_errno(r
, "Entry item %" PRIu64
" is bad, skipping over it.", i
);
2720 r
= bump_array_index(&i
, direction
, n
);
2725 /* Ensure our array is properly ordered. */
2726 if (p
> 0 && !check_properly_ordered(ofs
, p
, direction
))
2727 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
2728 "%s: entry array not properly ordered at entry %" PRIu64
,
2737 int journal_file_next_entry_for_data(
2739 Object
*o
, uint64_t p
,
2740 uint64_t data_offset
,
2741 direction_t direction
,
2742 Object
**ret
, uint64_t *offset
) {
2749 assert(p
> 0 || !o
);
2751 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2755 n
= le64toh(d
->data
.n_entries
);
2760 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2762 if (o
->object
.type
!= OBJECT_ENTRY
)
2765 r
= generic_array_bisect_plus_one(f
,
2766 le64toh(d
->data
.entry_offset
),
2767 le64toh(d
->data
.entry_array_offset
),
2768 le64toh(d
->data
.n_entries
),
2778 r
= bump_array_index(&i
, direction
, n
);
2784 r
= generic_array_get_plus_one(f
,
2785 le64toh(d
->data
.entry_offset
),
2786 le64toh(d
->data
.entry_array_offset
),
2794 log_debug_errno(r
, "Data entry item %" PRIu64
" is bad, skipping over it.", i
);
2796 r
= bump_array_index(&i
, direction
, n
);
2801 /* Ensure our array is properly ordered. */
2802 if (p
> 0 && check_properly_ordered(ofs
, p
, direction
))
2803 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
2804 "%s data entry array not properly ordered at entry %" PRIu64
,
2813 int journal_file_move_to_entry_by_offset_for_data(
2815 uint64_t data_offset
,
2817 direction_t direction
,
2818 Object
**ret
, uint64_t *offset
) {
2825 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2829 return 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
),
2839 int journal_file_move_to_entry_by_monotonic_for_data(
2841 uint64_t data_offset
,
2844 direction_t direction
,
2845 Object
**ret
, uint64_t *offset
) {
2853 /* First, seek by time */
2854 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &b
);
2860 r
= generic_array_bisect_plus_one(f
,
2861 le64toh(o
->data
.entry_offset
),
2862 le64toh(o
->data
.entry_array_offset
),
2863 le64toh(o
->data
.n_entries
),
2865 test_object_monotonic
,
2871 /* And now, continue seeking until we find an entry that
2872 * exists in both bisection arrays */
2878 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2882 r
= generic_array_bisect_plus_one(f
,
2883 le64toh(d
->data
.entry_offset
),
2884 le64toh(d
->data
.entry_array_offset
),
2885 le64toh(d
->data
.n_entries
),
2893 r
= journal_file_move_to_object(f
, OBJECT_DATA
, b
, &o
);
2897 r
= generic_array_bisect_plus_one(f
,
2898 le64toh(o
->data
.entry_offset
),
2899 le64toh(o
->data
.entry_array_offset
),
2900 le64toh(o
->data
.n_entries
),
2922 int journal_file_move_to_entry_by_seqnum_for_data(
2924 uint64_t data_offset
,
2926 direction_t direction
,
2927 Object
**ret
, uint64_t *offset
) {
2934 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2938 return generic_array_bisect_plus_one(f
,
2939 le64toh(d
->data
.entry_offset
),
2940 le64toh(d
->data
.entry_array_offset
),
2941 le64toh(d
->data
.n_entries
),
2948 int journal_file_move_to_entry_by_realtime_for_data(
2950 uint64_t data_offset
,
2952 direction_t direction
,
2953 Object
**ret
, uint64_t *offset
) {
2960 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2964 return generic_array_bisect_plus_one(f
,
2965 le64toh(d
->data
.entry_offset
),
2966 le64toh(d
->data
.entry_array_offset
),
2967 le64toh(d
->data
.n_entries
),
2969 test_object_realtime
,
2974 void journal_file_dump(JournalFile
*f
) {
2982 journal_file_print_header(f
);
2984 p
= le64toh(f
->header
->header_size
);
2986 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &o
);
2990 switch (o
->object
.type
) {
2993 printf("Type: OBJECT_UNUSED\n");
2997 printf("Type: OBJECT_DATA\n");
3001 printf("Type: OBJECT_FIELD\n");
3005 printf("Type: OBJECT_ENTRY seqnum=%"PRIu64
" monotonic=%"PRIu64
" realtime=%"PRIu64
"\n",
3006 le64toh(o
->entry
.seqnum
),
3007 le64toh(o
->entry
.monotonic
),
3008 le64toh(o
->entry
.realtime
));
3011 case OBJECT_FIELD_HASH_TABLE
:
3012 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
3015 case OBJECT_DATA_HASH_TABLE
:
3016 printf("Type: OBJECT_DATA_HASH_TABLE\n");
3019 case OBJECT_ENTRY_ARRAY
:
3020 printf("Type: OBJECT_ENTRY_ARRAY\n");
3024 printf("Type: OBJECT_TAG seqnum=%"PRIu64
" epoch=%"PRIu64
"\n",
3025 le64toh(o
->tag
.seqnum
),
3026 le64toh(o
->tag
.epoch
));
3030 printf("Type: unknown (%i)\n", o
->object
.type
);
3034 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
)
3035 printf("Flags: %s\n",
3036 object_compressed_to_string(o
->object
.flags
& OBJECT_COMPRESSION_MASK
));
3038 if (p
== le64toh(f
->header
->tail_object_offset
))
3041 p
= p
+ ALIGN64(le64toh(o
->object
.size
));
3046 log_error("File corrupt");
3049 static const char* format_timestamp_safe(char *buf
, size_t l
, usec_t t
) {
3052 x
= format_timestamp(buf
, l
, t
);
3058 void journal_file_print_header(JournalFile
*f
) {
3059 char a
[SD_ID128_STRING_MAX
], b
[SD_ID128_STRING_MAX
], c
[SD_ID128_STRING_MAX
], d
[SD_ID128_STRING_MAX
];
3060 char x
[FORMAT_TIMESTAMP_MAX
], y
[FORMAT_TIMESTAMP_MAX
], z
[FORMAT_TIMESTAMP_MAX
];
3062 char bytes
[FORMAT_BYTES_MAX
];
3067 printf("File path: %s\n"
3071 "Sequential number ID: %s\n"
3073 "Compatible flags:%s%s\n"
3074 "Incompatible flags:%s%s%s\n"
3075 "Header size: %"PRIu64
"\n"
3076 "Arena size: %"PRIu64
"\n"
3077 "Data hash table size: %"PRIu64
"\n"
3078 "Field hash table size: %"PRIu64
"\n"
3079 "Rotate suggested: %s\n"
3080 "Head sequential number: %"PRIu64
" (%"PRIx64
")\n"
3081 "Tail sequential number: %"PRIu64
" (%"PRIx64
")\n"
3082 "Head realtime timestamp: %s (%"PRIx64
")\n"
3083 "Tail realtime timestamp: %s (%"PRIx64
")\n"
3084 "Tail monotonic timestamp: %s (%"PRIx64
")\n"
3085 "Objects: %"PRIu64
"\n"
3086 "Entry objects: %"PRIu64
"\n",
3088 sd_id128_to_string(f
->header
->file_id
, a
),
3089 sd_id128_to_string(f
->header
->machine_id
, b
),
3090 sd_id128_to_string(f
->header
->boot_id
, c
),
3091 sd_id128_to_string(f
->header
->seqnum_id
, d
),
3092 f
->header
->state
== STATE_OFFLINE
? "OFFLINE" :
3093 f
->header
->state
== STATE_ONLINE
? "ONLINE" :
3094 f
->header
->state
== STATE_ARCHIVED
? "ARCHIVED" : "UNKNOWN",
3095 JOURNAL_HEADER_SEALED(f
->header
) ? " SEALED" : "",
3096 (le32toh(f
->header
->compatible_flags
) & ~HEADER_COMPATIBLE_ANY
) ? " ???" : "",
3097 JOURNAL_HEADER_COMPRESSED_XZ(f
->header
) ? " COMPRESSED-XZ" : "",
3098 JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
) ? " COMPRESSED-LZ4" : "",
3099 (le32toh(f
->header
->incompatible_flags
) & ~HEADER_INCOMPATIBLE_ANY
) ? " ???" : "",
3100 le64toh(f
->header
->header_size
),
3101 le64toh(f
->header
->arena_size
),
3102 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3103 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
),
3104 yes_no(journal_file_rotate_suggested(f
, 0)),
3105 le64toh(f
->header
->head_entry_seqnum
), le64toh(f
->header
->head_entry_seqnum
),
3106 le64toh(f
->header
->tail_entry_seqnum
), le64toh(f
->header
->tail_entry_seqnum
),
3107 format_timestamp_safe(x
, sizeof(x
), le64toh(f
->header
->head_entry_realtime
)), le64toh(f
->header
->head_entry_realtime
),
3108 format_timestamp_safe(y
, sizeof(y
), le64toh(f
->header
->tail_entry_realtime
)), le64toh(f
->header
->tail_entry_realtime
),
3109 format_timespan(z
, sizeof(z
), le64toh(f
->header
->tail_entry_monotonic
), USEC_PER_MSEC
), le64toh(f
->header
->tail_entry_monotonic
),
3110 le64toh(f
->header
->n_objects
),
3111 le64toh(f
->header
->n_entries
));
3113 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3114 printf("Data objects: %"PRIu64
"\n"
3115 "Data hash table fill: %.1f%%\n",
3116 le64toh(f
->header
->n_data
),
3117 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))));
3119 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3120 printf("Field objects: %"PRIu64
"\n"
3121 "Field hash table fill: %.1f%%\n",
3122 le64toh(f
->header
->n_fields
),
3123 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))));
3125 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
))
3126 printf("Tag objects: %"PRIu64
"\n",
3127 le64toh(f
->header
->n_tags
));
3128 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
3129 printf("Entry array objects: %"PRIu64
"\n",
3130 le64toh(f
->header
->n_entry_arrays
));
3132 if (fstat(f
->fd
, &st
) >= 0)
3133 printf("Disk usage: %s\n", format_bytes(bytes
, sizeof(bytes
), (uint64_t) st
.st_blocks
* 512ULL));
3136 static int journal_file_warn_btrfs(JournalFile
*f
) {
3142 /* Before we write anything, check if the COW logic is turned
3143 * off on btrfs. Given our write pattern that is quite
3144 * unfriendly to COW file systems this should greatly improve
3145 * performance on COW file systems, such as btrfs, at the
3146 * expense of data integrity features (which shouldn't be too
3147 * bad, given that we do our own checksumming). */
3149 r
= btrfs_is_filesystem(f
->fd
);
3151 return log_warning_errno(r
, "Failed to determine if journal is on btrfs: %m");
3155 r
= read_attr_fd(f
->fd
, &attrs
);
3157 return log_warning_errno(r
, "Failed to read file attributes: %m");
3159 if (attrs
& FS_NOCOW_FL
) {
3160 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3164 log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3165 "This is likely to slow down journal access substantially, please consider turning "
3166 "off the copy-on-write file attribute on the journal directory, using chattr +C.", f
->path
);
3171 int journal_file_open(
3177 uint64_t compress_threshold_bytes
,
3179 JournalMetrics
*metrics
,
3180 MMapCache
*mmap_cache
,
3181 Set
*deferred_closes
,
3182 JournalFile
*template,
3183 JournalFile
**ret
) {
3185 bool newly_created
= false;
3191 assert(fd
>= 0 || fname
);
3193 if (!IN_SET((flags
& O_ACCMODE
), O_RDONLY
, O_RDWR
))
3196 if (fname
&& (flags
& O_CREAT
) && !endswith(fname
, ".journal"))
3199 f
= new(JournalFile
, 1);
3203 *f
= (JournalFile
) {
3208 .prot
= prot_from_flags(flags
),
3209 .writable
= (flags
& O_ACCMODE
) != O_RDONLY
,
3212 .compress_lz4
= compress
,
3214 .compress_xz
= compress
,
3216 .compress_threshold_bytes
= compress_threshold_bytes
== (uint64_t) -1 ?
3217 DEFAULT_COMPRESS_THRESHOLD
:
3218 MAX(MIN_COMPRESS_THRESHOLD
, compress_threshold_bytes
),
3224 if (DEBUG_LOGGING
) {
3225 static int last_seal
= -1, last_compress
= -1;
3226 static uint64_t last_bytes
= UINT64_MAX
;
3227 char bytes
[FORMAT_BYTES_MAX
];
3229 if (last_seal
!= f
->seal
||
3230 last_compress
!= JOURNAL_FILE_COMPRESS(f
) ||
3231 last_bytes
!= f
->compress_threshold_bytes
) {
3233 log_debug("Journal effective settings seal=%s compress=%s compress_threshold_bytes=%s",
3234 yes_no(f
->seal
), yes_no(JOURNAL_FILE_COMPRESS(f
)),
3235 format_bytes(bytes
, sizeof bytes
, f
->compress_threshold_bytes
));
3236 last_seal
= f
->seal
;
3237 last_compress
= JOURNAL_FILE_COMPRESS(f
);
3238 last_bytes
= f
->compress_threshold_bytes
;
3243 f
->mmap
= mmap_cache_ref(mmap_cache
);
3245 f
->mmap
= mmap_cache_new();
3253 f
->path
= strdup(fname
);
3261 /* If we don't know the path, fill in something explanatory and vaguely useful */
3262 if (asprintf(&f
->path
, "/proc/self/%i", fd
) < 0) {
3268 f
->chain_cache
= ordered_hashmap_new(&uint64_hash_ops
);
3269 if (!f
->chain_cache
) {
3275 /* We pass O_NONBLOCK here, so that in case somebody pointed us to some character device node or FIFO
3276 * or so, we likely fail quickly than block for long. For regular files O_NONBLOCK has no effect, hence
3277 * it doesn't hurt in that case. */
3279 f
->fd
= open(f
->path
, f
->flags
|O_CLOEXEC
|O_NONBLOCK
, f
->mode
);
3285 /* fds we opened here by us should also be closed by us. */
3288 r
= fd_nonblock(f
->fd
, false);
3293 f
->cache_fd
= mmap_cache_add_fd(f
->mmap
, f
->fd
);
3299 r
= journal_file_fstat(f
);
3303 if (f
->last_stat
.st_size
== 0 && f
->writable
) {
3305 (void) journal_file_warn_btrfs(f
);
3307 /* Let's attach the creation time to the journal file, so that the vacuuming code knows the age of this
3308 * file even if the file might end up corrupted one day... Ideally we'd just use the creation time many
3309 * file systems maintain for each file, but the API to query this is very new, hence let's emulate this
3310 * via extended attributes. If extended attributes are not supported we'll just skip this, and rely
3311 * solely on mtime/atime/ctime of the file. */
3312 (void) fd_setcrtime(f
->fd
, 0);
3315 /* Try to load the FSPRG state, and if we can't, then
3316 * just don't do sealing */
3318 r
= journal_file_fss_load(f
);
3324 r
= journal_file_init_header(f
, template);
3328 r
= journal_file_fstat(f
);
3332 newly_created
= true;
3335 if (f
->last_stat
.st_size
< (off_t
) HEADER_SIZE_MIN
) {
3340 r
= mmap_cache_get(f
->mmap
, f
->cache_fd
, f
->prot
, CONTEXT_HEADER
, true, 0, PAGE_ALIGN(sizeof(Header
)), &f
->last_stat
, &h
, NULL
);
3342 /* Some file systems (jffs2 or p9fs) don't support mmap() properly (or only read-only
3343 * mmap()), and return EINVAL in that case. Let's propagate that as a more recognizable error
3353 if (!newly_created
) {
3354 set_clear_with_destructor(deferred_closes
, journal_file_close
);
3356 r
= journal_file_verify_header(f
);
3362 if (!newly_created
&& f
->writable
) {
3363 r
= journal_file_fss_load(f
);
3371 journal_default_metrics(metrics
, f
->fd
);
3372 f
->metrics
= *metrics
;
3373 } else if (template)
3374 f
->metrics
= template->metrics
;
3376 r
= journal_file_refresh_header(f
);
3382 r
= journal_file_hmac_setup(f
);
3387 if (newly_created
) {
3388 r
= journal_file_setup_field_hash_table(f
);
3392 r
= journal_file_setup_data_hash_table(f
);
3397 r
= journal_file_append_first_tag(f
);
3403 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
)) {
3408 if (template && template->post_change_timer
) {
3409 r
= journal_file_enable_post_change_timer(
3411 sd_event_source_get_event(template->post_change_timer
),
3412 template->post_change_timer_period
);
3418 /* The file is opened now successfully, thus we take possession of any passed in fd. */
3425 if (f
->cache_fd
&& mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
3428 (void) journal_file_close(f
);
3433 int journal_file_archive(JournalFile
*f
) {
3434 _cleanup_free_
char *p
= NULL
;
3441 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
3442 * rotation, since we don't know the actual path, and couldn't rename the file hence. */
3443 if (path_startswith(f
->path
, "/proc/self/fd"))
3446 if (!endswith(f
->path
, ".journal"))
3449 if (asprintf(&p
, "%.*s@" SD_ID128_FORMAT_STR
"-%016"PRIx64
"-%016"PRIx64
".journal",
3450 (int) strlen(f
->path
) - 8, f
->path
,
3451 SD_ID128_FORMAT_VAL(f
->header
->seqnum_id
),
3452 le64toh(f
->header
->head_entry_seqnum
),
3453 le64toh(f
->header
->head_entry_realtime
)) < 0)
3456 /* Try to rename the file to the archived version. If the file already was deleted, we'll get ENOENT, let's
3457 * ignore that case. */
3458 if (rename(f
->path
, p
) < 0 && errno
!= ENOENT
)
3461 /* Sync the rename to disk */
3462 (void) fsync_directory_of_file(f
->fd
);
3464 /* Set as archive so offlining commits w/state=STATE_ARCHIVED. Previously we would set old_file->header->state
3465 * to STATE_ARCHIVED directly here, but journal_file_set_offline() short-circuits when state != STATE_ONLINE,
3466 * which would result in the rotated journal never getting fsync() called before closing. Now we simply queue
3467 * the archive state by setting an archive bit, leaving the state as STATE_ONLINE so proper offlining
3471 /* Currently, btrfs is not very good with out write patterns and fragments heavily. Let's defrag our journal
3472 * files when we archive them */
3473 f
->defrag_on_close
= true;
3478 JournalFile
* journal_initiate_close(
3480 Set
*deferred_closes
) {
3486 if (deferred_closes
) {
3488 r
= set_put(deferred_closes
, f
);
3490 log_debug_errno(r
, "Failed to add file to deferred close set, closing immediately.");
3492 (void) journal_file_set_offline(f
, false);
3497 return journal_file_close(f
);
3500 int journal_file_rotate(
3503 uint64_t compress_threshold_bytes
,
3505 Set
*deferred_closes
) {
3507 JournalFile
*new_file
= NULL
;
3513 r
= journal_file_archive(*f
);
3517 r
= journal_file_open(
3523 compress_threshold_bytes
,
3531 journal_initiate_close(*f
, deferred_closes
);
3537 int journal_file_dispose(int dir_fd
, const char *fname
) {
3538 _cleanup_free_
char *p
= NULL
;
3539 _cleanup_close_
int fd
= -1;
3543 /* Renames a journal file to *.journal~, i.e. to mark it as corruped or otherwise uncleanly shutdown. Note that
3544 * this is done without looking into the file or changing any of its contents. The idea is that this is called
3545 * whenever something is suspicious and we want to move the file away and make clear that it is not accessed
3546 * for writing anymore. */
3548 if (!endswith(fname
, ".journal"))
3551 if (asprintf(&p
, "%.*s@%016" PRIx64
"-%016" PRIx64
".journal~",
3552 (int) strlen(fname
) - 8, fname
,
3553 now(CLOCK_REALTIME
),
3557 if (renameat(dir_fd
, fname
, dir_fd
, p
) < 0)
3560 /* btrfs doesn't cope well with our write pattern and fragments heavily. Let's defrag all files we rotate */
3561 fd
= openat(dir_fd
, p
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
|O_NOFOLLOW
);
3563 log_debug_errno(errno
, "Failed to open file for defragmentation/FS_NOCOW_FL, ignoring: %m");
3565 (void) chattr_fd(fd
, 0, FS_NOCOW_FL
, NULL
);
3566 (void) btrfs_defrag_fd(fd
);
3572 int journal_file_open_reliably(
3577 uint64_t compress_threshold_bytes
,
3579 JournalMetrics
*metrics
,
3580 MMapCache
*mmap_cache
,
3581 Set
*deferred_closes
,
3582 JournalFile
*template,
3583 JournalFile
**ret
) {
3587 r
= journal_file_open(-1, fname
, flags
, mode
, compress
, compress_threshold_bytes
, seal
, metrics
, mmap_cache
,
3588 deferred_closes
, template, ret
);
3590 -EBADMSG
, /* Corrupted */
3591 -ENODATA
, /* Truncated */
3592 -EHOSTDOWN
, /* Other machine */
3593 -EPROTONOSUPPORT
, /* Incompatible feature */
3594 -EBUSY
, /* Unclean shutdown */
3595 -ESHUTDOWN
, /* Already archived */
3596 -EIO
, /* IO error, including SIGBUS on mmap */
3597 -EIDRM
, /* File has been deleted */
3598 -ETXTBSY
)) /* File is from the future */
3601 if ((flags
& O_ACCMODE
) == O_RDONLY
)
3604 if (!(flags
& O_CREAT
))
3607 if (!endswith(fname
, ".journal"))
3610 /* The file is corrupted. Rotate it away and try it again (but only once) */
3611 log_warning_errno(r
, "File %s corrupted or uncleanly shut down, renaming and replacing.", fname
);
3613 r
= journal_file_dispose(AT_FDCWD
, fname
);
3617 return journal_file_open(-1, fname
, flags
, mode
, compress
, compress_threshold_bytes
, seal
, metrics
, mmap_cache
,
3618 deferred_closes
, template, ret
);
3621 int journal_file_copy_entry(JournalFile
*from
, JournalFile
*to
, Object
*o
, uint64_t p
) {
3623 uint64_t q
, xor_hash
= 0;
3627 const sd_id128_t
*boot_id
;
3637 ts
.monotonic
= le64toh(o
->entry
.monotonic
);
3638 ts
.realtime
= le64toh(o
->entry
.realtime
);
3639 boot_id
= &o
->entry
.boot_id
;
3641 n
= journal_file_entry_n_items(o
);
3642 /* alloca() can't take 0, hence let's allocate at least one */
3643 items
= newa(EntryItem
, MAX(1u, n
));
3645 for (i
= 0; i
< n
; i
++) {
3652 q
= le64toh(o
->entry
.items
[i
].object_offset
);
3653 le_hash
= o
->entry
.items
[i
].hash
;
3655 r
= journal_file_move_to_object(from
, OBJECT_DATA
, q
, &o
);
3659 if (le_hash
!= o
->data
.hash
)
3662 l
= le64toh(o
->object
.size
) - offsetof(Object
, data
.payload
);
3665 /* We hit the limit on 32bit machines */
3666 if ((uint64_t) t
!= l
)
3669 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
3670 #if HAVE_XZ || HAVE_LZ4
3673 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
3674 o
->data
.payload
, l
, &from
->compress_buffer
, &from
->compress_buffer_size
, &rsize
, 0);
3678 data
= from
->compress_buffer
;
3681 return -EPROTONOSUPPORT
;
3684 data
= o
->data
.payload
;
3686 r
= journal_file_append_data(to
, data
, l
, &u
, &h
);
3690 xor_hash
^= le64toh(u
->data
.hash
);
3691 items
[i
].object_offset
= htole64(h
);
3692 items
[i
].hash
= u
->data
.hash
;
3694 r
= journal_file_move_to_object(from
, OBJECT_ENTRY
, p
, &o
);
3699 r
= journal_file_append_entry_internal(to
, &ts
, boot_id
, xor_hash
, items
, n
,
3702 if (mmap_cache_got_sigbus(to
->mmap
, to
->cache_fd
))
3708 void journal_reset_metrics(JournalMetrics
*m
) {
3711 /* Set everything to "pick automatic values". */
3713 *m
= (JournalMetrics
) {
3714 .min_use
= (uint64_t) -1,
3715 .max_use
= (uint64_t) -1,
3716 .min_size
= (uint64_t) -1,
3717 .max_size
= (uint64_t) -1,
3718 .keep_free
= (uint64_t) -1,
3719 .n_max_files
= (uint64_t) -1,
3723 void journal_default_metrics(JournalMetrics
*m
, int fd
) {
3724 char a
[FORMAT_BYTES_MAX
], b
[FORMAT_BYTES_MAX
], c
[FORMAT_BYTES_MAX
], d
[FORMAT_BYTES_MAX
], e
[FORMAT_BYTES_MAX
];
3726 uint64_t fs_size
= 0;
3731 if (fstatvfs(fd
, &ss
) >= 0)
3732 fs_size
= ss
.f_frsize
* ss
.f_blocks
;
3734 log_debug_errno(errno
, "Failed to determine disk size: %m");
3736 if (m
->max_use
== (uint64_t) -1) {
3739 m
->max_use
= CLAMP(PAGE_ALIGN(fs_size
/ 10), /* 10% of file system size */
3740 MAX_USE_LOWER
, MAX_USE_UPPER
);
3742 m
->max_use
= MAX_USE_LOWER
;
3744 m
->max_use
= PAGE_ALIGN(m
->max_use
);
3746 if (m
->max_use
!= 0 && m
->max_use
< JOURNAL_FILE_SIZE_MIN
*2)
3747 m
->max_use
= JOURNAL_FILE_SIZE_MIN
*2;
3750 if (m
->min_use
== (uint64_t) -1) {
3752 m
->min_use
= CLAMP(PAGE_ALIGN(fs_size
/ 50), /* 2% of file system size */
3753 MIN_USE_LOW
, MIN_USE_HIGH
);
3755 m
->min_use
= MIN_USE_LOW
;
3758 if (m
->min_use
> m
->max_use
)
3759 m
->min_use
= m
->max_use
;
3761 if (m
->max_size
== (uint64_t) -1)
3762 m
->max_size
= MIN(PAGE_ALIGN(m
->max_use
/ 8), /* 8 chunks */
3765 m
->max_size
= PAGE_ALIGN(m
->max_size
);
3767 if (m
->max_size
!= 0) {
3768 if (m
->max_size
< JOURNAL_FILE_SIZE_MIN
)
3769 m
->max_size
= JOURNAL_FILE_SIZE_MIN
;
3771 if (m
->max_use
!= 0 && m
->max_size
*2 > m
->max_use
)
3772 m
->max_use
= m
->max_size
*2;
3775 if (m
->min_size
== (uint64_t) -1)
3776 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3778 m
->min_size
= CLAMP(PAGE_ALIGN(m
->min_size
),
3779 JOURNAL_FILE_SIZE_MIN
,
3780 m
->max_size
?: UINT64_MAX
);
3782 if (m
->keep_free
== (uint64_t) -1) {
3784 m
->keep_free
= MIN(PAGE_ALIGN(fs_size
/ 20), /* 5% of file system size */
3787 m
->keep_free
= DEFAULT_KEEP_FREE
;
3790 if (m
->n_max_files
== (uint64_t) -1)
3791 m
->n_max_files
= DEFAULT_N_MAX_FILES
;
3793 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64
,
3794 format_bytes(a
, sizeof(a
), m
->min_use
),
3795 format_bytes(b
, sizeof(b
), m
->max_use
),
3796 format_bytes(c
, sizeof(c
), m
->max_size
),
3797 format_bytes(d
, sizeof(d
), m
->min_size
),
3798 format_bytes(e
, sizeof(e
), m
->keep_free
),
3802 int journal_file_get_cutoff_realtime_usec(JournalFile
*f
, usec_t
*from
, usec_t
*to
) {
3808 if (f
->header
->head_entry_realtime
== 0)
3811 *from
= le64toh(f
->header
->head_entry_realtime
);
3815 if (f
->header
->tail_entry_realtime
== 0)
3818 *to
= le64toh(f
->header
->tail_entry_realtime
);
3824 int journal_file_get_cutoff_monotonic_usec(JournalFile
*f
, sd_id128_t boot_id
, usec_t
*from
, usec_t
*to
) {
3832 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &p
);
3836 if (le64toh(o
->data
.n_entries
) <= 0)
3840 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, le64toh(o
->data
.entry_offset
), &o
);
3844 *from
= le64toh(o
->entry
.monotonic
);
3848 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
3852 r
= generic_array_get_plus_one(f
,
3853 le64toh(o
->data
.entry_offset
),
3854 le64toh(o
->data
.entry_array_offset
),
3855 le64toh(o
->data
.n_entries
)-1,
3860 *to
= le64toh(o
->entry
.monotonic
);
3866 bool journal_file_rotate_suggested(JournalFile
*f
, usec_t max_file_usec
) {
3870 /* If we gained new header fields we gained new features,
3871 * hence suggest a rotation */
3872 if (le64toh(f
->header
->header_size
) < sizeof(Header
)) {
3873 log_debug("%s uses an outdated header, suggesting rotation.", f
->path
);
3877 /* Let's check if the hash tables grew over a certain fill
3878 * level (75%, borrowing this value from Java's hash table
3879 * implementation), and if so suggest a rotation. To calculate
3880 * the fill level we need the n_data field, which only exists
3881 * in newer versions. */
3883 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3884 if (le64toh(f
->header
->n_data
) * 4ULL > (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3885 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.",
3887 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))),
3888 le64toh(f
->header
->n_data
),
3889 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3890 (unsigned long long) f
->last_stat
.st_size
,
3891 f
->last_stat
.st_size
/ le64toh(f
->header
->n_data
));
3895 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3896 if (le64toh(f
->header
->n_fields
) * 4ULL > (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3897 log_debug("Field hash table of %s has a fill level at %.1f (%"PRIu64
" of %"PRIu64
" items), suggesting rotation.",
3899 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))),
3900 le64toh(f
->header
->n_fields
),
3901 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
));
3905 /* Are the data objects properly indexed by field objects? */
3906 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
3907 JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
3908 le64toh(f
->header
->n_data
) > 0 &&
3909 le64toh(f
->header
->n_fields
) == 0)
3912 if (max_file_usec
> 0) {
3915 h
= le64toh(f
->header
->head_entry_realtime
);
3916 t
= now(CLOCK_REALTIME
);
3918 if (h
> 0 && t
> h
+ max_file_usec
)