1 /* SPDX-License-Identifier: LGPL-2.1+ */
9 #include <sys/statvfs.h>
13 #include "alloc-util.h"
14 #include "btrfs-util.h"
15 #include "chattr-util.h"
19 #include "journal-authenticate.h"
20 #include "journal-def.h"
21 #include "journal-file.h"
23 #include "parse-util.h"
24 #include "path-util.h"
25 #include "random-util.h"
28 #include "stat-util.h"
29 #include "string-util.h"
31 #include "xattr-util.h"
33 #define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
34 #define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
36 #define DEFAULT_COMPRESS_THRESHOLD (512ULL)
37 #define MIN_COMPRESS_THRESHOLD (8ULL)
39 /* This is the minimum journal file size */
40 #define JOURNAL_FILE_SIZE_MIN (512ULL*1024ULL) /* 512 KiB */
42 /* These are the lower and upper bounds if we deduce the max_use value
43 * from the file system size */
44 #define DEFAULT_MAX_USE_LOWER (1ULL*1024ULL*1024ULL) /* 1 MiB */
45 #define DEFAULT_MAX_USE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
47 /* This is the default minimal use limit, how much we'll use even if keep_free suggests otherwise. */
48 #define DEFAULT_MIN_USE (1ULL*1024ULL*1024ULL) /* 1 MiB */
50 /* This is the upper bound if we deduce max_size from max_use */
51 #define DEFAULT_MAX_SIZE_UPPER (128ULL*1024ULL*1024ULL) /* 128 MiB */
53 /* This is the upper bound if we deduce the keep_free value from the
55 #define DEFAULT_KEEP_FREE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
57 /* This is the keep_free value when we can't determine the system
59 #define DEFAULT_KEEP_FREE (1024ULL*1024ULL) /* 1 MB */
61 /* This is the default maximum number of journal files to keep around. */
62 #define DEFAULT_N_MAX_FILES (100)
64 /* n_data was the first entry we added after the initial file format design */
65 #define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
67 /* How many entries to keep in the entry array chain cache at max */
68 #define CHAIN_CACHE_MAX 20
70 /* How much to increase the journal file size at once each time we allocate something new. */
71 #define FILE_SIZE_INCREASE (8ULL*1024ULL*1024ULL) /* 8MB */
73 /* Reread fstat() of the file for detecting deletions at least this often */
74 #define LAST_STAT_REFRESH_USEC (5*USEC_PER_SEC)
76 /* The mmap context to use for the header we pick as one above the last defined typed */
77 #define CONTEXT_HEADER _OBJECT_TYPE_MAX
80 # pragma GCC diagnostic ignored "-Waddress-of-packed-member"
83 /* This may be called from a separate thread to prevent blocking the caller for the duration of fsync().
84 * As a result we use atomic operations on f->offline_state for inter-thread communications with
85 * journal_file_set_offline() and journal_file_set_online(). */
86 static void journal_file_set_offline_internal(JournalFile
*f
) {
92 switch (f
->offline_state
) {
94 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_CANCEL
, OFFLINE_DONE
))
98 case OFFLINE_AGAIN_FROM_SYNCING
:
99 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_SYNCING
, OFFLINE_SYNCING
))
103 case OFFLINE_AGAIN_FROM_OFFLINING
:
104 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_OFFLINING
, OFFLINE_SYNCING
))
108 case OFFLINE_SYNCING
:
111 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_OFFLINING
))
114 f
->header
->state
= f
->archive
? STATE_ARCHIVED
: STATE_OFFLINE
;
118 case OFFLINE_OFFLINING
:
119 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_OFFLINING
, OFFLINE_DONE
))
126 log_debug("OFFLINE_JOINED unexpected offline state for journal_file_set_offline_internal()");
132 static void * journal_file_set_offline_thread(void *arg
) {
133 JournalFile
*f
= arg
;
135 (void) pthread_setname_np(pthread_self(), "journal-offline");
137 journal_file_set_offline_internal(f
);
142 static int journal_file_set_offline_thread_join(JournalFile
*f
) {
147 if (f
->offline_state
== OFFLINE_JOINED
)
150 r
= pthread_join(f
->offline_thread
, NULL
);
154 f
->offline_state
= OFFLINE_JOINED
;
156 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
162 /* Trigger a restart if the offline thread is mid-flight in a restartable state. */
163 static bool journal_file_set_offline_try_restart(JournalFile
*f
) {
165 switch (f
->offline_state
) {
166 case OFFLINE_AGAIN_FROM_SYNCING
:
167 case OFFLINE_AGAIN_FROM_OFFLINING
:
171 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_CANCEL
, OFFLINE_AGAIN_FROM_SYNCING
))
175 case OFFLINE_SYNCING
:
176 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_AGAIN_FROM_SYNCING
))
180 case OFFLINE_OFFLINING
:
181 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_OFFLINING
, OFFLINE_AGAIN_FROM_OFFLINING
))
191 /* Sets a journal offline.
193 * If wait is false then an offline is dispatched in a separate thread for a
194 * subsequent journal_file_set_offline() or journal_file_set_online() of the
195 * same journal to synchronize with.
197 * If wait is true, then either an existing offline thread will be restarted
198 * and joined, or if none exists the offline is simply performed in this
199 * context without involving another thread.
201 int journal_file_set_offline(JournalFile
*f
, bool wait
) {
210 if (f
->fd
< 0 || !f
->header
)
213 /* An offlining journal is implicitly online and may modify f->header->state,
214 * we must also join any potentially lingering offline thread when not online. */
215 if (!journal_file_is_offlining(f
) && f
->header
->state
!= STATE_ONLINE
)
216 return journal_file_set_offline_thread_join(f
);
218 /* Restart an in-flight offline thread and wait if needed, or join a lingering done one. */
219 restarted
= journal_file_set_offline_try_restart(f
);
220 if ((restarted
&& wait
) || !restarted
) {
221 r
= journal_file_set_offline_thread_join(f
);
229 /* Initiate a new offline. */
230 f
->offline_state
= OFFLINE_SYNCING
;
232 if (wait
) /* Without using a thread if waiting. */
233 journal_file_set_offline_internal(f
);
235 sigset_t ss
, saved_ss
;
238 if (sigfillset(&ss
) < 0)
241 r
= pthread_sigmask(SIG_BLOCK
, &ss
, &saved_ss
);
245 r
= pthread_create(&f
->offline_thread
, NULL
, journal_file_set_offline_thread
, f
);
247 k
= pthread_sigmask(SIG_SETMASK
, &saved_ss
, NULL
);
249 f
->offline_state
= OFFLINE_JOINED
;
259 static int journal_file_set_online(JournalFile
*f
) {
267 if (f
->fd
< 0 || !f
->header
)
271 switch (f
->offline_state
) {
273 /* No offline thread, no need to wait. */
277 case OFFLINE_SYNCING
:
278 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_CANCEL
))
280 /* Canceled syncing prior to offlining, no need to wait. */
284 case OFFLINE_AGAIN_FROM_SYNCING
:
285 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_SYNCING
, OFFLINE_CANCEL
))
287 /* Canceled restart from syncing, no need to wait. */
291 case OFFLINE_AGAIN_FROM_OFFLINING
:
292 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_OFFLINING
, OFFLINE_CANCEL
))
294 /* Canceled restart from offlining, must wait for offlining to complete however. */
299 r
= journal_file_set_offline_thread_join(f
);
309 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
312 switch (f
->header
->state
) {
317 f
->header
->state
= STATE_ONLINE
;
326 bool journal_file_is_offlining(JournalFile
*f
) {
329 __sync_synchronize();
331 if (IN_SET(f
->offline_state
, OFFLINE_DONE
, OFFLINE_JOINED
))
337 JournalFile
* journal_file_close(JournalFile
*f
) {
341 /* Write the final tag */
342 if (f
->seal
&& f
->writable
) {
345 r
= journal_file_append_tag(f
);
347 log_error_errno(r
, "Failed to append tag when closing journal: %m");
351 if (f
->post_change_timer
) {
354 if (sd_event_source_get_enabled(f
->post_change_timer
, &enabled
) >= 0)
355 if (enabled
== SD_EVENT_ONESHOT
)
356 journal_file_post_change(f
);
358 (void) sd_event_source_set_enabled(f
->post_change_timer
, SD_EVENT_OFF
);
359 sd_event_source_unref(f
->post_change_timer
);
362 journal_file_set_offline(f
, true);
364 if (f
->mmap
&& f
->cache_fd
)
365 mmap_cache_free_fd(f
->mmap
, f
->cache_fd
);
367 if (f
->fd
>= 0 && f
->defrag_on_close
) {
369 /* Be friendly to btrfs: turn COW back on again now,
370 * and defragment the file. We won't write to the file
371 * ever again, hence remove all fragmentation, and
372 * reenable all the good bits COW usually provides
373 * (such as data checksumming). */
375 (void) chattr_fd(f
->fd
, 0, FS_NOCOW_FL
, NULL
);
376 (void) btrfs_defrag_fd(f
->fd
);
383 mmap_cache_unref(f
->mmap
);
385 ordered_hashmap_free_free(f
->chain_cache
);
387 #if HAVE_XZ || HAVE_LZ4
388 free(f
->compress_buffer
);
393 munmap(f
->fss_file
, PAGE_ALIGN(f
->fss_file_size
));
395 free(f
->fsprg_state
);
400 gcry_md_close(f
->hmac
);
406 static int journal_file_init_header(JournalFile
*f
, JournalFile
*template) {
413 memcpy(h
.signature
, HEADER_SIGNATURE
, 8);
414 h
.header_size
= htole64(ALIGN64(sizeof(h
)));
416 h
.incompatible_flags
|= htole32(
417 f
->compress_xz
* HEADER_INCOMPATIBLE_COMPRESSED_XZ
|
418 f
->compress_lz4
* HEADER_INCOMPATIBLE_COMPRESSED_LZ4
);
420 h
.compatible_flags
= htole32(
421 f
->seal
* HEADER_COMPATIBLE_SEALED
);
423 r
= sd_id128_randomize(&h
.file_id
);
428 h
.seqnum_id
= template->header
->seqnum_id
;
429 h
.tail_entry_seqnum
= template->header
->tail_entry_seqnum
;
431 h
.seqnum_id
= h
.file_id
;
433 k
= pwrite(f
->fd
, &h
, sizeof(h
), 0);
443 static int journal_file_refresh_header(JournalFile
*f
) {
450 r
= sd_id128_get_machine(&f
->header
->machine_id
);
451 if (IN_SET(r
, -ENOENT
, -ENOMEDIUM
))
452 /* We don't have a machine-id, let's continue without */
453 zero(f
->header
->machine_id
);
457 r
= sd_id128_get_boot(&boot_id
);
461 f
->header
->boot_id
= boot_id
;
463 r
= journal_file_set_online(f
);
465 /* Sync the online state to disk */
468 /* We likely just created a new file, also sync the directory this file is located in. */
469 (void) fsync_directory_of_file(f
->fd
);
474 static bool warn_wrong_flags(const JournalFile
*f
, bool compatible
) {
475 const uint32_t any
= compatible
? HEADER_COMPATIBLE_ANY
: HEADER_INCOMPATIBLE_ANY
,
476 supported
= compatible
? HEADER_COMPATIBLE_SUPPORTED
: HEADER_INCOMPATIBLE_SUPPORTED
;
477 const char *type
= compatible
? "compatible" : "incompatible";
480 flags
= le32toh(compatible
? f
->header
->compatible_flags
: f
->header
->incompatible_flags
);
482 if (flags
& ~supported
) {
484 log_debug("Journal file %s has unknown %s flags 0x%"PRIx32
,
485 f
->path
, type
, flags
& ~any
);
486 flags
= (flags
& any
) & ~supported
;
490 _cleanup_free_
char *t
= NULL
;
492 if (compatible
&& (flags
& HEADER_COMPATIBLE_SEALED
))
493 strv
[n
++] = "sealed";
494 if (!compatible
&& (flags
& HEADER_INCOMPATIBLE_COMPRESSED_XZ
))
495 strv
[n
++] = "xz-compressed";
496 if (!compatible
&& (flags
& HEADER_INCOMPATIBLE_COMPRESSED_LZ4
))
497 strv
[n
++] = "lz4-compressed";
499 assert(n
< ELEMENTSOF(strv
));
501 t
= strv_join((char**) strv
, ", ");
502 log_debug("Journal file %s uses %s %s %s disabled at compilation time.",
503 f
->path
, type
, n
> 1 ? "flags" : "flag", strnull(t
));
511 static int journal_file_verify_header(JournalFile
*f
) {
512 uint64_t arena_size
, header_size
;
517 if (memcmp(f
->header
->signature
, HEADER_SIGNATURE
, 8))
520 /* In both read and write mode we refuse to open files with incompatible
521 * flags we don't know. */
522 if (warn_wrong_flags(f
, false))
523 return -EPROTONOSUPPORT
;
525 /* When open for writing we refuse to open files with compatible flags, too. */
526 if (f
->writable
&& warn_wrong_flags(f
, true))
527 return -EPROTONOSUPPORT
;
529 if (f
->header
->state
>= _STATE_MAX
)
532 header_size
= le64toh(f
->header
->header_size
);
534 /* The first addition was n_data, so check that we are at least this large */
535 if (header_size
< HEADER_SIZE_MIN
)
538 if (JOURNAL_HEADER_SEALED(f
->header
) && !JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
541 arena_size
= le64toh(f
->header
->arena_size
);
543 if (UINT64_MAX
- header_size
< arena_size
|| header_size
+ arena_size
> (uint64_t) f
->last_stat
.st_size
)
546 if (le64toh(f
->header
->tail_object_offset
) > header_size
+ arena_size
)
549 if (!VALID64(le64toh(f
->header
->data_hash_table_offset
)) ||
550 !VALID64(le64toh(f
->header
->field_hash_table_offset
)) ||
551 !VALID64(le64toh(f
->header
->tail_object_offset
)) ||
552 !VALID64(le64toh(f
->header
->entry_array_offset
)))
556 sd_id128_t machine_id
;
560 r
= sd_id128_get_machine(&machine_id
);
564 if (!sd_id128_equal(machine_id
, f
->header
->machine_id
))
567 state
= f
->header
->state
;
569 if (state
== STATE_ARCHIVED
)
570 return -ESHUTDOWN
; /* Already archived */
571 else if (state
== STATE_ONLINE
) {
572 log_debug("Journal file %s is already online. Assuming unclean closing.", f
->path
);
574 } else if (state
!= STATE_OFFLINE
) {
575 log_debug("Journal file %s has unknown state %i.", f
->path
, state
);
579 if (f
->header
->field_hash_table_size
== 0 || f
->header
->data_hash_table_size
== 0)
582 /* Don't permit appending to files from the future. Because otherwise the realtime timestamps wouldn't
583 * be strictly ordered in the entries in the file anymore, and we can't have that since it breaks
585 if (le64toh(f
->header
->tail_entry_realtime
) > now(CLOCK_REALTIME
)) {
586 log_debug("Journal file %s is from the future, refusing to append new data to it that'd be older.", f
->path
);
591 f
->compress_xz
= JOURNAL_HEADER_COMPRESSED_XZ(f
->header
);
592 f
->compress_lz4
= JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
);
594 f
->seal
= JOURNAL_HEADER_SEALED(f
->header
);
599 static int journal_file_fstat(JournalFile
*f
) {
605 if (fstat(f
->fd
, &f
->last_stat
) < 0)
608 f
->last_stat_usec
= now(CLOCK_MONOTONIC
);
610 /* Refuse dealing with with files that aren't regular */
611 r
= stat_verify_regular(&f
->last_stat
);
615 /* Refuse appending to files that are already deleted */
616 if (f
->last_stat
.st_nlink
<= 0)
622 static int journal_file_allocate(JournalFile
*f
, uint64_t offset
, uint64_t size
) {
623 uint64_t old_size
, new_size
;
629 /* We assume that this file is not sparse, and we know that
630 * for sure, since we always call posix_fallocate()
633 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
637 le64toh(f
->header
->header_size
) +
638 le64toh(f
->header
->arena_size
);
640 new_size
= PAGE_ALIGN(offset
+ size
);
641 if (new_size
< le64toh(f
->header
->header_size
))
642 new_size
= le64toh(f
->header
->header_size
);
644 if (new_size
<= old_size
) {
646 /* We already pre-allocated enough space, but before
647 * we write to it, let's check with fstat() if the
648 * file got deleted, in order make sure we don't throw
649 * away the data immediately. Don't check fstat() for
650 * all writes though, but only once ever 10s. */
652 if (f
->last_stat_usec
+ LAST_STAT_REFRESH_USEC
> now(CLOCK_MONOTONIC
))
655 return journal_file_fstat(f
);
658 /* Allocate more space. */
660 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
663 if (new_size
> f
->metrics
.min_size
&& f
->metrics
.keep_free
> 0) {
666 if (fstatvfs(f
->fd
, &svfs
) >= 0) {
669 available
= LESS_BY((uint64_t) svfs
.f_bfree
* (uint64_t) svfs
.f_bsize
, f
->metrics
.keep_free
);
671 if (new_size
- old_size
> available
)
676 /* Increase by larger blocks at once */
677 new_size
= DIV_ROUND_UP(new_size
, FILE_SIZE_INCREASE
) * FILE_SIZE_INCREASE
;
678 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
679 new_size
= f
->metrics
.max_size
;
681 /* Note that the glibc fallocate() fallback is very
682 inefficient, hence we try to minimize the allocation area
684 r
= posix_fallocate(f
->fd
, old_size
, new_size
- old_size
);
688 f
->header
->arena_size
= htole64(new_size
- le64toh(f
->header
->header_size
));
690 return journal_file_fstat(f
);
693 static unsigned type_to_context(ObjectType type
) {
694 /* One context for each type, plus one catch-all for the rest */
695 assert_cc(_OBJECT_TYPE_MAX
<= MMAP_CACHE_MAX_CONTEXTS
);
696 assert_cc(CONTEXT_HEADER
< MMAP_CACHE_MAX_CONTEXTS
);
697 return type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
? type
: 0;
700 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
) {
709 /* Avoid SIGBUS on invalid accesses */
710 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
) {
711 /* Hmm, out of range? Let's refresh the fstat() data
712 * first, before we trust that check. */
714 r
= journal_file_fstat(f
);
718 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
)
719 return -EADDRNOTAVAIL
;
722 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
);
725 static uint64_t minimum_header_size(Object
*o
) {
727 static const uint64_t table
[] = {
728 [OBJECT_DATA
] = sizeof(DataObject
),
729 [OBJECT_FIELD
] = sizeof(FieldObject
),
730 [OBJECT_ENTRY
] = sizeof(EntryObject
),
731 [OBJECT_DATA_HASH_TABLE
] = sizeof(HashTableObject
),
732 [OBJECT_FIELD_HASH_TABLE
] = sizeof(HashTableObject
),
733 [OBJECT_ENTRY_ARRAY
] = sizeof(EntryArrayObject
),
734 [OBJECT_TAG
] = sizeof(TagObject
),
737 if (o
->object
.type
>= ELEMENTSOF(table
) || table
[o
->object
.type
] <= 0)
738 return sizeof(ObjectHeader
);
740 return table
[o
->object
.type
];
743 /* Lightweight object checks. We want this to be fast, so that we won't
744 * slowdown every journal_file_move_to_object() call too much. */
745 static int journal_file_check_object(JournalFile
*f
, uint64_t offset
, Object
*o
) {
749 switch (o
->object
.type
) {
752 if ((le64toh(o
->data
.entry_offset
) == 0) ^ (le64toh(o
->data
.n_entries
) == 0)) {
753 log_debug("Bad n_entries: %"PRIu64
": %"PRIu64
,
754 le64toh(o
->data
.n_entries
), offset
);
758 if (le64toh(o
->object
.size
) - offsetof(DataObject
, payload
) <= 0) {
759 log_debug("Bad object size (<= %zu): %"PRIu64
": %"PRIu64
,
760 offsetof(DataObject
, payload
),
761 le64toh(o
->object
.size
),
766 if (!VALID64(le64toh(o
->data
.next_hash_offset
)) ||
767 !VALID64(le64toh(o
->data
.next_field_offset
)) ||
768 !VALID64(le64toh(o
->data
.entry_offset
)) ||
769 !VALID64(le64toh(o
->data
.entry_array_offset
))) {
770 log_debug("Invalid offset, next_hash_offset="OFSfmt
", next_field_offset="OFSfmt
771 ", entry_offset="OFSfmt
", entry_array_offset="OFSfmt
": %"PRIu64
,
772 le64toh(o
->data
.next_hash_offset
),
773 le64toh(o
->data
.next_field_offset
),
774 le64toh(o
->data
.entry_offset
),
775 le64toh(o
->data
.entry_array_offset
),
784 if (le64toh(o
->object
.size
) - offsetof(FieldObject
, payload
) <= 0) {
786 "Bad field size (<= %zu): %"PRIu64
": %"PRIu64
,
787 offsetof(FieldObject
, payload
),
788 le64toh(o
->object
.size
),
793 if (!VALID64(le64toh(o
->field
.next_hash_offset
)) ||
794 !VALID64(le64toh(o
->field
.head_data_offset
))) {
796 "Invalid offset, next_hash_offset="OFSfmt
797 ", head_data_offset="OFSfmt
": %"PRIu64
,
798 le64toh(o
->field
.next_hash_offset
),
799 le64toh(o
->field
.head_data_offset
),
806 if ((le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) % sizeof(EntryItem
) != 0) {
808 "Bad entry size (<= %zu): %"PRIu64
": %"PRIu64
,
809 offsetof(EntryObject
, items
),
810 le64toh(o
->object
.size
),
815 if ((le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) / sizeof(EntryItem
) <= 0) {
817 "Invalid number items in entry: %"PRIu64
": %"PRIu64
,
818 (le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) / sizeof(EntryItem
),
823 if (le64toh(o
->entry
.seqnum
) <= 0) {
825 "Invalid entry seqnum: %"PRIx64
": %"PRIu64
,
826 le64toh(o
->entry
.seqnum
),
831 if (!VALID_REALTIME(le64toh(o
->entry
.realtime
))) {
833 "Invalid entry realtime timestamp: %"PRIu64
": %"PRIu64
,
834 le64toh(o
->entry
.realtime
),
839 if (!VALID_MONOTONIC(le64toh(o
->entry
.monotonic
))) {
841 "Invalid entry monotonic timestamp: %"PRIu64
": %"PRIu64
,
842 le64toh(o
->entry
.monotonic
),
849 case OBJECT_DATA_HASH_TABLE
:
850 case OBJECT_FIELD_HASH_TABLE
:
851 if ((le64toh(o
->object
.size
) - offsetof(HashTableObject
, items
)) % sizeof(HashItem
) != 0 ||
852 (le64toh(o
->object
.size
) - offsetof(HashTableObject
, items
)) / sizeof(HashItem
) <= 0) {
854 "Invalid %s hash table size: %"PRIu64
": %"PRIu64
,
855 o
->object
.type
== OBJECT_DATA_HASH_TABLE
? "data" : "field",
856 le64toh(o
->object
.size
),
863 case OBJECT_ENTRY_ARRAY
:
864 if ((le64toh(o
->object
.size
) - offsetof(EntryArrayObject
, items
)) % sizeof(le64_t
) != 0 ||
865 (le64toh(o
->object
.size
) - offsetof(EntryArrayObject
, items
)) / sizeof(le64_t
) <= 0) {
867 "Invalid object entry array size: %"PRIu64
": %"PRIu64
,
868 le64toh(o
->object
.size
),
873 if (!VALID64(le64toh(o
->entry_array
.next_entry_array_offset
))) {
875 "Invalid object entry array next_entry_array_offset: "OFSfmt
": %"PRIu64
,
876 le64toh(o
->entry_array
.next_entry_array_offset
),
884 if (le64toh(o
->object
.size
) != sizeof(TagObject
)) {
886 "Invalid object tag size: %"PRIu64
": %"PRIu64
,
887 le64toh(o
->object
.size
),
892 if (!VALID_EPOCH(le64toh(o
->tag
.epoch
))) {
894 "Invalid object tag epoch: %"PRIu64
": %"PRIu64
,
895 le64toh(o
->tag
.epoch
),
906 int journal_file_move_to_object(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
**ret
) {
916 /* Objects may only be located at multiple of 64 bit */
917 if (!VALID64(offset
)) {
918 log_debug("Attempt to move to object at non-64bit boundary: %" PRIu64
, offset
);
922 /* Object may not be located in the file header */
923 if (offset
< le64toh(f
->header
->header_size
)) {
924 log_debug("Attempt to move to object located in file header: %" PRIu64
, offset
);
928 r
= journal_file_move_to(f
, type
, false, offset
, sizeof(ObjectHeader
), &t
, &tsize
);
933 s
= le64toh(o
->object
.size
);
936 log_debug("Attempt to move to uninitialized object: %" PRIu64
, offset
);
939 if (s
< sizeof(ObjectHeader
)) {
940 log_debug("Attempt to move to overly short object: %" PRIu64
, offset
);
944 if (o
->object
.type
<= OBJECT_UNUSED
) {
945 log_debug("Attempt to move to object with invalid type: %" PRIu64
, offset
);
949 if (s
< minimum_header_size(o
)) {
950 log_debug("Attempt to move to truncated object: %" PRIu64
, offset
);
954 if (type
> OBJECT_UNUSED
&& o
->object
.type
!= type
) {
955 log_debug("Attempt to move to object of unexpected type: %" PRIu64
, offset
);
960 r
= journal_file_move_to(f
, type
, false, offset
, s
, &t
, NULL
);
967 r
= journal_file_check_object(f
, offset
, o
);
975 static uint64_t journal_file_entry_seqnum(JournalFile
*f
, uint64_t *seqnum
) {
981 r
= le64toh(f
->header
->tail_entry_seqnum
) + 1;
984 /* If an external seqnum counter was passed, we update
985 * both the local and the external one, and set it to
986 * the maximum of both */
994 f
->header
->tail_entry_seqnum
= htole64(r
);
996 if (f
->header
->head_entry_seqnum
== 0)
997 f
->header
->head_entry_seqnum
= htole64(r
);
1002 int journal_file_append_object(JournalFile
*f
, ObjectType type
, uint64_t size
, Object
**ret
, uint64_t *offset
) {
1010 assert(type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
);
1011 assert(size
>= sizeof(ObjectHeader
));
1015 r
= journal_file_set_online(f
);
1019 p
= le64toh(f
->header
->tail_object_offset
);
1021 p
= le64toh(f
->header
->header_size
);
1023 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &tail
);
1027 p
+= ALIGN64(le64toh(tail
->object
.size
));
1030 r
= journal_file_allocate(f
, p
, size
);
1034 r
= journal_file_move_to(f
, type
, false, p
, size
, &t
, NULL
);
1041 o
->object
.type
= type
;
1042 o
->object
.size
= htole64(size
);
1044 f
->header
->tail_object_offset
= htole64(p
);
1045 f
->header
->n_objects
= htole64(le64toh(f
->header
->n_objects
) + 1);
1053 static int journal_file_setup_data_hash_table(JournalFile
*f
) {
1061 /* We estimate that we need 1 hash table entry per 768 bytes
1062 of journal file and we want to make sure we never get
1063 beyond 75% fill level. Calculate the hash table size for
1064 the maximum file size based on these metrics. */
1066 s
= (f
->metrics
.max_size
* 4 / 768 / 3) * sizeof(HashItem
);
1067 if (s
< DEFAULT_DATA_HASH_TABLE_SIZE
)
1068 s
= DEFAULT_DATA_HASH_TABLE_SIZE
;
1070 log_debug("Reserving %"PRIu64
" entries in hash table.", s
/ sizeof(HashItem
));
1072 r
= journal_file_append_object(f
,
1073 OBJECT_DATA_HASH_TABLE
,
1074 offsetof(Object
, hash_table
.items
) + s
,
1079 memzero(o
->hash_table
.items
, s
);
1081 f
->header
->data_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1082 f
->header
->data_hash_table_size
= htole64(s
);
1087 static int journal_file_setup_field_hash_table(JournalFile
*f
) {
1095 /* We use a fixed size hash table for the fields as this
1096 * number should grow very slowly only */
1098 s
= DEFAULT_FIELD_HASH_TABLE_SIZE
;
1099 r
= journal_file_append_object(f
,
1100 OBJECT_FIELD_HASH_TABLE
,
1101 offsetof(Object
, hash_table
.items
) + s
,
1106 memzero(o
->hash_table
.items
, s
);
1108 f
->header
->field_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1109 f
->header
->field_hash_table_size
= htole64(s
);
1114 int journal_file_map_data_hash_table(JournalFile
*f
) {
1122 if (f
->data_hash_table
)
1125 p
= le64toh(f
->header
->data_hash_table_offset
);
1126 s
= le64toh(f
->header
->data_hash_table_size
);
1128 r
= journal_file_move_to(f
,
1129 OBJECT_DATA_HASH_TABLE
,
1136 f
->data_hash_table
= t
;
1140 int journal_file_map_field_hash_table(JournalFile
*f
) {
1148 if (f
->field_hash_table
)
1151 p
= le64toh(f
->header
->field_hash_table_offset
);
1152 s
= le64toh(f
->header
->field_hash_table_size
);
1154 r
= journal_file_move_to(f
,
1155 OBJECT_FIELD_HASH_TABLE
,
1162 f
->field_hash_table
= t
;
1166 static int journal_file_link_field(
1177 assert(f
->field_hash_table
);
1181 if (o
->object
.type
!= OBJECT_FIELD
)
1184 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1188 /* This might alter the window we are looking at */
1189 o
->field
.next_hash_offset
= o
->field
.head_data_offset
= 0;
1192 p
= le64toh(f
->field_hash_table
[h
].tail_hash_offset
);
1194 f
->field_hash_table
[h
].head_hash_offset
= htole64(offset
);
1196 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1200 o
->field
.next_hash_offset
= htole64(offset
);
1203 f
->field_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1205 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
1206 f
->header
->n_fields
= htole64(le64toh(f
->header
->n_fields
) + 1);
1211 static int journal_file_link_data(
1222 assert(f
->data_hash_table
);
1226 if (o
->object
.type
!= OBJECT_DATA
)
1229 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1233 /* This might alter the window we are looking at */
1234 o
->data
.next_hash_offset
= o
->data
.next_field_offset
= 0;
1235 o
->data
.entry_offset
= o
->data
.entry_array_offset
= 0;
1236 o
->data
.n_entries
= 0;
1239 p
= le64toh(f
->data_hash_table
[h
].tail_hash_offset
);
1241 /* Only entry in the hash table is easy */
1242 f
->data_hash_table
[h
].head_hash_offset
= htole64(offset
);
1244 /* Move back to the previous data object, to patch in
1247 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1251 o
->data
.next_hash_offset
= htole64(offset
);
1254 f
->data_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1256 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
1257 f
->header
->n_data
= htole64(le64toh(f
->header
->n_data
) + 1);
1262 int journal_file_find_field_object_with_hash(
1264 const void *field
, uint64_t size
, uint64_t hash
,
1265 Object
**ret
, uint64_t *offset
) {
1267 uint64_t p
, osize
, h
, m
;
1272 assert(field
&& size
> 0);
1274 /* If the field hash table is empty, we can't find anything */
1275 if (le64toh(f
->header
->field_hash_table_size
) <= 0)
1278 /* Map the field hash table, if it isn't mapped yet. */
1279 r
= journal_file_map_field_hash_table(f
);
1283 osize
= offsetof(Object
, field
.payload
) + size
;
1285 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1290 p
= le64toh(f
->field_hash_table
[h
].head_hash_offset
);
1295 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1299 if (le64toh(o
->field
.hash
) == hash
&&
1300 le64toh(o
->object
.size
) == osize
&&
1301 memcmp(o
->field
.payload
, field
, size
) == 0) {
1311 p
= le64toh(o
->field
.next_hash_offset
);
1317 int journal_file_find_field_object(
1319 const void *field
, uint64_t size
,
1320 Object
**ret
, uint64_t *offset
) {
1325 assert(field
&& size
> 0);
1327 hash
= hash64(field
, size
);
1329 return journal_file_find_field_object_with_hash(f
,
1334 int journal_file_find_data_object_with_hash(
1336 const void *data
, uint64_t size
, uint64_t hash
,
1337 Object
**ret
, uint64_t *offset
) {
1339 uint64_t p
, osize
, h
, m
;
1344 assert(data
|| size
== 0);
1346 /* If there's no data hash table, then there's no entry. */
1347 if (le64toh(f
->header
->data_hash_table_size
) <= 0)
1350 /* Map the data hash table, if it isn't mapped yet. */
1351 r
= journal_file_map_data_hash_table(f
);
1355 osize
= offsetof(Object
, data
.payload
) + size
;
1357 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1362 p
= le64toh(f
->data_hash_table
[h
].head_hash_offset
);
1367 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1371 if (le64toh(o
->data
.hash
) != hash
)
1374 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
1375 #if HAVE_XZ || HAVE_LZ4
1379 l
= le64toh(o
->object
.size
);
1380 if (l
<= offsetof(Object
, data
.payload
))
1383 l
-= offsetof(Object
, data
.payload
);
1385 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
1386 o
->data
.payload
, l
, &f
->compress_buffer
, &f
->compress_buffer_size
, &rsize
, 0);
1390 if (rsize
== size
&&
1391 memcmp(f
->compress_buffer
, data
, size
) == 0) {
1402 return -EPROTONOSUPPORT
;
1404 } else if (le64toh(o
->object
.size
) == osize
&&
1405 memcmp(o
->data
.payload
, data
, size
) == 0) {
1417 p
= le64toh(o
->data
.next_hash_offset
);
1423 int journal_file_find_data_object(
1425 const void *data
, uint64_t size
,
1426 Object
**ret
, uint64_t *offset
) {
1431 assert(data
|| size
== 0);
1433 hash
= hash64(data
, size
);
1435 return journal_file_find_data_object_with_hash(f
,
1440 static int journal_file_append_field(
1442 const void *field
, uint64_t size
,
1443 Object
**ret
, uint64_t *offset
) {
1451 assert(field
&& size
> 0);
1453 hash
= hash64(field
, size
);
1455 r
= journal_file_find_field_object_with_hash(f
, field
, size
, hash
, &o
, &p
);
1469 osize
= offsetof(Object
, field
.payload
) + size
;
1470 r
= journal_file_append_object(f
, OBJECT_FIELD
, osize
, &o
, &p
);
1474 o
->field
.hash
= htole64(hash
);
1475 memcpy(o
->field
.payload
, field
, size
);
1477 r
= journal_file_link_field(f
, o
, p
, hash
);
1481 /* The linking might have altered the window, so let's
1482 * refresh our pointer */
1483 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1488 r
= journal_file_hmac_put_object(f
, OBJECT_FIELD
, o
, p
);
1502 static int journal_file_append_data(
1504 const void *data
, uint64_t size
,
1505 Object
**ret
, uint64_t *offset
) {
1510 int r
, compression
= 0;
1514 assert(data
|| size
== 0);
1516 hash
= hash64(data
, size
);
1518 r
= journal_file_find_data_object_with_hash(f
, data
, size
, hash
, &o
, &p
);
1532 osize
= offsetof(Object
, data
.payload
) + size
;
1533 r
= journal_file_append_object(f
, OBJECT_DATA
, osize
, &o
, &p
);
1537 o
->data
.hash
= htole64(hash
);
1539 #if HAVE_XZ || HAVE_LZ4
1540 if (JOURNAL_FILE_COMPRESS(f
) && size
>= f
->compress_threshold_bytes
) {
1543 compression
= compress_blob(data
, size
, o
->data
.payload
, size
- 1, &rsize
);
1545 if (compression
>= 0) {
1546 o
->object
.size
= htole64(offsetof(Object
, data
.payload
) + rsize
);
1547 o
->object
.flags
|= compression
;
1549 log_debug("Compressed data object %"PRIu64
" -> %zu using %s",
1550 size
, rsize
, object_compressed_to_string(compression
));
1552 /* Compression didn't work, we don't really care why, let's continue without compression */
1557 if (compression
== 0)
1558 memcpy_safe(o
->data
.payload
, data
, size
);
1560 r
= journal_file_link_data(f
, o
, p
, hash
);
1565 r
= journal_file_hmac_put_object(f
, OBJECT_DATA
, o
, p
);
1570 /* The linking might have altered the window, so let's
1571 * refresh our pointer */
1572 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1579 eq
= memchr(data
, '=', size
);
1580 if (eq
&& eq
> data
) {
1584 /* Create field object ... */
1585 r
= journal_file_append_field(f
, data
, (uint8_t*) eq
- (uint8_t*) data
, &fo
, &fp
);
1589 /* ... and link it in. */
1590 o
->data
.next_field_offset
= fo
->field
.head_data_offset
;
1591 fo
->field
.head_data_offset
= le64toh(p
);
1603 uint64_t journal_file_entry_n_items(Object
*o
) {
1606 if (o
->object
.type
!= OBJECT_ENTRY
)
1609 return (le64toh(o
->object
.size
) - offsetof(Object
, entry
.items
)) / sizeof(EntryItem
);
1612 uint64_t journal_file_entry_array_n_items(Object
*o
) {
1615 if (o
->object
.type
!= OBJECT_ENTRY_ARRAY
)
1618 return (le64toh(o
->object
.size
) - offsetof(Object
, entry_array
.items
)) / sizeof(uint64_t);
1621 uint64_t journal_file_hash_table_n_items(Object
*o
) {
1624 if (!IN_SET(o
->object
.type
, OBJECT_DATA_HASH_TABLE
, OBJECT_FIELD_HASH_TABLE
))
1627 return (le64toh(o
->object
.size
) - offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
);
1630 static int link_entry_into_array(JournalFile
*f
,
1635 uint64_t n
= 0, ap
= 0, q
, i
, a
, hidx
;
1644 a
= le64toh(*first
);
1645 i
= hidx
= le64toh(*idx
);
1648 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
1652 n
= journal_file_entry_array_n_items(o
);
1654 o
->entry_array
.items
[i
] = htole64(p
);
1655 *idx
= htole64(hidx
+ 1);
1661 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
1672 r
= journal_file_append_object(f
, OBJECT_ENTRY_ARRAY
,
1673 offsetof(Object
, entry_array
.items
) + n
* sizeof(uint64_t),
1679 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY_ARRAY
, o
, q
);
1684 o
->entry_array
.items
[i
] = htole64(p
);
1687 *first
= htole64(q
);
1689 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, ap
, &o
);
1693 o
->entry_array
.next_entry_array_offset
= htole64(q
);
1696 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
1697 f
->header
->n_entry_arrays
= htole64(le64toh(f
->header
->n_entry_arrays
) + 1);
1699 *idx
= htole64(hidx
+ 1);
1704 static int link_entry_into_array_plus_one(JournalFile
*f
,
1719 *extra
= htole64(p
);
1723 i
= htole64(le64toh(*idx
) - 1);
1724 r
= link_entry_into_array(f
, first
, &i
, p
);
1729 *idx
= htole64(le64toh(*idx
) + 1);
1733 static int journal_file_link_entry_item(JournalFile
*f
, Object
*o
, uint64_t offset
, uint64_t i
) {
1740 p
= le64toh(o
->entry
.items
[i
].object_offset
);
1744 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1748 return link_entry_into_array_plus_one(f
,
1749 &o
->data
.entry_offset
,
1750 &o
->data
.entry_array_offset
,
1755 static int journal_file_link_entry(JournalFile
*f
, Object
*o
, uint64_t offset
) {
1764 if (o
->object
.type
!= OBJECT_ENTRY
)
1767 __sync_synchronize();
1769 /* Link up the entry itself */
1770 r
= link_entry_into_array(f
,
1771 &f
->header
->entry_array_offset
,
1772 &f
->header
->n_entries
,
1777 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
1779 if (f
->header
->head_entry_realtime
== 0)
1780 f
->header
->head_entry_realtime
= o
->entry
.realtime
;
1782 f
->header
->tail_entry_realtime
= o
->entry
.realtime
;
1783 f
->header
->tail_entry_monotonic
= o
->entry
.monotonic
;
1785 /* Link up the items */
1786 n
= journal_file_entry_n_items(o
);
1787 for (i
= 0; i
< n
; i
++) {
1788 r
= journal_file_link_entry_item(f
, o
, offset
, i
);
1796 static int journal_file_append_entry_internal(
1798 const dual_timestamp
*ts
,
1799 const sd_id128_t
*boot_id
,
1801 const EntryItem items
[], unsigned n_items
,
1803 Object
**ret
, uint64_t *offset
) {
1811 assert(items
|| n_items
== 0);
1814 osize
= offsetof(Object
, entry
.items
) + (n_items
* sizeof(EntryItem
));
1816 r
= journal_file_append_object(f
, OBJECT_ENTRY
, osize
, &o
, &np
);
1820 o
->entry
.seqnum
= htole64(journal_file_entry_seqnum(f
, seqnum
));
1821 memcpy_safe(o
->entry
.items
, items
, n_items
* sizeof(EntryItem
));
1822 o
->entry
.realtime
= htole64(ts
->realtime
);
1823 o
->entry
.monotonic
= htole64(ts
->monotonic
);
1824 o
->entry
.xor_hash
= htole64(xor_hash
);
1825 o
->entry
.boot_id
= boot_id
? *boot_id
: f
->header
->boot_id
;
1828 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY
, o
, np
);
1833 r
= journal_file_link_entry(f
, o
, np
);
1846 void journal_file_post_change(JournalFile
*f
) {
1852 /* inotify() does not receive IN_MODIFY events from file
1853 * accesses done via mmap(). After each access we hence
1854 * trigger IN_MODIFY by truncating the journal file to its
1855 * current size which triggers IN_MODIFY. */
1857 __sync_synchronize();
1859 if (ftruncate(f
->fd
, f
->last_stat
.st_size
) < 0)
1860 log_debug_errno(errno
, "Failed to truncate file to its own size: %m");
1863 static int post_change_thunk(sd_event_source
*timer
, uint64_t usec
, void *userdata
) {
1866 journal_file_post_change(userdata
);
1871 static void schedule_post_change(JournalFile
*f
) {
1872 sd_event_source
*timer
;
1877 assert(f
->post_change_timer
);
1879 timer
= f
->post_change_timer
;
1881 r
= sd_event_source_get_enabled(timer
, &enabled
);
1883 log_debug_errno(r
, "Failed to get ftruncate timer state: %m");
1887 if (enabled
== SD_EVENT_ONESHOT
)
1890 r
= sd_event_now(sd_event_source_get_event(timer
), CLOCK_MONOTONIC
, &now
);
1892 log_debug_errno(r
, "Failed to get clock's now for scheduling ftruncate: %m");
1896 r
= sd_event_source_set_time(timer
, now
+f
->post_change_timer_period
);
1898 log_debug_errno(r
, "Failed to set time for scheduling ftruncate: %m");
1902 r
= sd_event_source_set_enabled(timer
, SD_EVENT_ONESHOT
);
1904 log_debug_errno(r
, "Failed to enable scheduled ftruncate: %m");
1911 /* On failure, let's simply post the change immediately. */
1912 journal_file_post_change(f
);
1915 /* Enable coalesced change posting in a timer on the provided sd_event instance */
1916 int journal_file_enable_post_change_timer(JournalFile
*f
, sd_event
*e
, usec_t t
) {
1917 _cleanup_(sd_event_source_unrefp
) sd_event_source
*timer
= NULL
;
1921 assert_return(!f
->post_change_timer
, -EINVAL
);
1925 r
= sd_event_add_time(e
, &timer
, CLOCK_MONOTONIC
, 0, 0, post_change_thunk
, f
);
1929 r
= sd_event_source_set_enabled(timer
, SD_EVENT_OFF
);
1933 f
->post_change_timer
= TAKE_PTR(timer
);
1934 f
->post_change_timer_period
= t
;
1939 static int entry_item_cmp(const EntryItem
*a
, const EntryItem
*b
) {
1940 return CMP(le64toh(a
->object_offset
), le64toh(b
->object_offset
));
1943 int journal_file_append_entry(
1945 const dual_timestamp
*ts
,
1946 const sd_id128_t
*boot_id
,
1947 const struct iovec iovec
[], unsigned n_iovec
,
1949 Object
**ret
, uint64_t *offset
) {
1954 uint64_t xor_hash
= 0;
1955 struct dual_timestamp _ts
;
1959 assert(iovec
|| n_iovec
== 0);
1962 if (!VALID_REALTIME(ts
->realtime
)) {
1963 log_debug("Invalid realtime timestamp %"PRIu64
", refusing entry.", ts
->realtime
);
1966 if (!VALID_MONOTONIC(ts
->monotonic
)) {
1967 log_debug("Invalid monotomic timestamp %"PRIu64
", refusing entry.", ts
->monotonic
);
1971 dual_timestamp_get(&_ts
);
1976 r
= journal_file_maybe_append_tag(f
, ts
->realtime
);
1981 /* alloca() can't take 0, hence let's allocate at least one */
1982 items
= newa(EntryItem
, MAX(1u, n_iovec
));
1984 for (i
= 0; i
< n_iovec
; i
++) {
1988 r
= journal_file_append_data(f
, iovec
[i
].iov_base
, iovec
[i
].iov_len
, &o
, &p
);
1992 xor_hash
^= le64toh(o
->data
.hash
);
1993 items
[i
].object_offset
= htole64(p
);
1994 items
[i
].hash
= o
->data
.hash
;
1997 /* Order by the position on disk, in order to improve seek
1998 * times for rotating media. */
1999 typesafe_qsort(items
, n_iovec
, entry_item_cmp
);
2001 r
= journal_file_append_entry_internal(f
, ts
, boot_id
, xor_hash
, items
, n_iovec
, seqnum
, ret
, offset
);
2003 /* If the memory mapping triggered a SIGBUS then we return an
2004 * IO error and ignore the error code passed down to us, since
2005 * it is very likely just an effect of a nullified replacement
2008 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
2011 if (f
->post_change_timer
)
2012 schedule_post_change(f
);
2014 journal_file_post_change(f
);
2019 typedef struct ChainCacheItem
{
2020 uint64_t first
; /* the array at the beginning of the chain */
2021 uint64_t array
; /* the cached array */
2022 uint64_t begin
; /* the first item in the cached array */
2023 uint64_t total
; /* the total number of items in all arrays before this one in the chain */
2024 uint64_t last_index
; /* the last index we looked at, to optimize locality when bisecting */
2027 static void chain_cache_put(
2034 uint64_t last_index
) {
2037 /* If the chain item to cache for this chain is the
2038 * first one it's not worth caching anything */
2042 if (ordered_hashmap_size(h
) >= CHAIN_CACHE_MAX
) {
2043 ci
= ordered_hashmap_steal_first(h
);
2046 ci
= new(ChainCacheItem
, 1);
2053 if (ordered_hashmap_put(h
, &ci
->first
, ci
) < 0) {
2058 assert(ci
->first
== first
);
2063 ci
->last_index
= last_index
;
2066 static int generic_array_get(
2070 Object
**ret
, uint64_t *offset
) {
2073 uint64_t p
= 0, a
, t
= 0;
2081 /* Try the chain cache first */
2082 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2083 if (ci
&& i
> ci
->total
) {
2092 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2096 k
= journal_file_entry_array_n_items(o
);
2098 p
= le64toh(o
->entry_array
.items
[i
]);
2104 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
2110 /* Let's cache this item for the next invocation */
2111 chain_cache_put(f
->chain_cache
, ci
, first
, a
, le64toh(o
->entry_array
.items
[0]), t
, i
);
2113 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2126 static int generic_array_get_plus_one(
2131 Object
**ret
, uint64_t *offset
) {
2140 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2153 return generic_array_get(f
, first
, i
-1, ret
, offset
);
2162 static int generic_array_bisect(
2167 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2168 direction_t direction
,
2173 uint64_t a
, p
, t
= 0, i
= 0, last_p
= 0, last_index
= (uint64_t) -1;
2174 bool subtract_one
= false;
2175 Object
*o
, *array
= NULL
;
2180 assert(test_object
);
2182 /* Start with the first array in the chain */
2185 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2186 if (ci
&& n
> ci
->total
&& ci
->begin
!= 0) {
2187 /* Ah, we have iterated this bisection array chain
2188 * previously! Let's see if we can skip ahead in the
2189 * chain, as far as the last time. But we can't jump
2190 * backwards in the chain, so let's check that
2193 r
= test_object(f
, ci
->begin
, needle
);
2197 if (r
== TEST_LEFT
) {
2198 /* OK, what we are looking for is right of the
2199 * begin of this EntryArray, so let's jump
2200 * straight to previously cached array in the
2206 last_index
= ci
->last_index
;
2211 uint64_t left
, right
, k
, lp
;
2213 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2217 k
= journal_file_entry_array_n_items(array
);
2223 lp
= p
= le64toh(array
->entry_array
.items
[i
]);
2227 r
= test_object(f
, p
, needle
);
2228 if (r
== -EBADMSG
) {
2229 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (1)");
2236 if (r
== TEST_FOUND
)
2237 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2239 if (r
== TEST_RIGHT
) {
2243 if (last_index
!= (uint64_t) -1) {
2244 assert(last_index
<= right
);
2246 /* If we cached the last index we
2247 * looked at, let's try to not to jump
2248 * too wildly around and see if we can
2249 * limit the range to look at early to
2250 * the immediate neighbors of the last
2251 * index we looked at. */
2253 if (last_index
> 0) {
2254 uint64_t x
= last_index
- 1;
2256 p
= le64toh(array
->entry_array
.items
[x
]);
2260 r
= test_object(f
, p
, needle
);
2264 if (r
== TEST_FOUND
)
2265 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2267 if (r
== TEST_RIGHT
)
2273 if (last_index
< right
) {
2274 uint64_t y
= last_index
+ 1;
2276 p
= le64toh(array
->entry_array
.items
[y
]);
2280 r
= test_object(f
, p
, needle
);
2284 if (r
== TEST_FOUND
)
2285 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2287 if (r
== TEST_RIGHT
)
2295 if (left
== right
) {
2296 if (direction
== DIRECTION_UP
)
2297 subtract_one
= true;
2303 assert(left
< right
);
2304 i
= (left
+ right
) / 2;
2306 p
= le64toh(array
->entry_array
.items
[i
]);
2310 r
= test_object(f
, p
, needle
);
2311 if (r
== -EBADMSG
) {
2312 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (2)");
2319 if (r
== TEST_FOUND
)
2320 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2322 if (r
== TEST_RIGHT
)
2330 if (direction
== DIRECTION_UP
) {
2332 subtract_one
= true;
2343 last_index
= (uint64_t) -1;
2344 a
= le64toh(array
->entry_array
.next_entry_array_offset
);
2350 if (subtract_one
&& t
== 0 && i
== 0)
2353 /* Let's cache this item for the next invocation */
2354 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
);
2356 if (subtract_one
&& i
== 0)
2358 else if (subtract_one
)
2359 p
= le64toh(array
->entry_array
.items
[i
-1]);
2361 p
= le64toh(array
->entry_array
.items
[i
]);
2363 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2374 *idx
= t
+ i
+ (subtract_one
? -1 : 0);
2379 static int generic_array_bisect_plus_one(
2385 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2386 direction_t direction
,
2392 bool step_back
= false;
2396 assert(test_object
);
2401 /* This bisects the array in object 'first', but first checks
2403 r
= test_object(f
, extra
, needle
);
2407 if (r
== TEST_FOUND
)
2408 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2410 /* if we are looking with DIRECTION_UP then we need to first
2411 see if in the actual array there is a matching entry, and
2412 return the last one of that. But if there isn't any we need
2413 to return this one. Hence remember this, and return it
2416 step_back
= direction
== DIRECTION_UP
;
2418 if (r
== TEST_RIGHT
) {
2419 if (direction
== DIRECTION_DOWN
)
2425 r
= generic_array_bisect(f
, first
, n
-1, needle
, test_object
, direction
, ret
, offset
, idx
);
2427 if (r
== 0 && step_back
)
2436 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2452 _pure_
static int test_object_offset(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2458 else if (p
< needle
)
2464 static int test_object_seqnum(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2471 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2475 if (le64toh(o
->entry
.seqnum
) == needle
)
2477 else if (le64toh(o
->entry
.seqnum
) < needle
)
2483 int journal_file_move_to_entry_by_seqnum(
2486 direction_t direction
,
2492 return generic_array_bisect(f
,
2493 le64toh(f
->header
->entry_array_offset
),
2494 le64toh(f
->header
->n_entries
),
2501 static int test_object_realtime(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2508 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2512 if (le64toh(o
->entry
.realtime
) == needle
)
2514 else if (le64toh(o
->entry
.realtime
) < needle
)
2520 int journal_file_move_to_entry_by_realtime(
2523 direction_t direction
,
2529 return generic_array_bisect(f
,
2530 le64toh(f
->header
->entry_array_offset
),
2531 le64toh(f
->header
->n_entries
),
2533 test_object_realtime
,
2538 static int test_object_monotonic(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2545 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2549 if (le64toh(o
->entry
.monotonic
) == needle
)
2551 else if (le64toh(o
->entry
.monotonic
) < needle
)
2557 static int find_data_object_by_boot_id(
2563 char t
[STRLEN("_BOOT_ID=") + 32 + 1] = "_BOOT_ID=";
2565 sd_id128_to_string(boot_id
, t
+ 9);
2566 return journal_file_find_data_object(f
, t
, sizeof(t
) - 1, o
, b
);
2569 int journal_file_move_to_entry_by_monotonic(
2573 direction_t direction
,
2582 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, NULL
);
2588 return generic_array_bisect_plus_one(f
,
2589 le64toh(o
->data
.entry_offset
),
2590 le64toh(o
->data
.entry_array_offset
),
2591 le64toh(o
->data
.n_entries
),
2593 test_object_monotonic
,
2598 void journal_file_reset_location(JournalFile
*f
) {
2599 f
->location_type
= LOCATION_HEAD
;
2600 f
->current_offset
= 0;
2601 f
->current_seqnum
= 0;
2602 f
->current_realtime
= 0;
2603 f
->current_monotonic
= 0;
2604 zero(f
->current_boot_id
);
2605 f
->current_xor_hash
= 0;
2608 void journal_file_save_location(JournalFile
*f
, Object
*o
, uint64_t offset
) {
2609 f
->location_type
= LOCATION_SEEK
;
2610 f
->current_offset
= offset
;
2611 f
->current_seqnum
= le64toh(o
->entry
.seqnum
);
2612 f
->current_realtime
= le64toh(o
->entry
.realtime
);
2613 f
->current_monotonic
= le64toh(o
->entry
.monotonic
);
2614 f
->current_boot_id
= o
->entry
.boot_id
;
2615 f
->current_xor_hash
= le64toh(o
->entry
.xor_hash
);
2618 int journal_file_compare_locations(JournalFile
*af
, JournalFile
*bf
) {
2625 assert(af
->location_type
== LOCATION_SEEK
);
2626 assert(bf
->location_type
== LOCATION_SEEK
);
2628 /* If contents and timestamps match, these entries are
2629 * identical, even if the seqnum does not match */
2630 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
) &&
2631 af
->current_monotonic
== bf
->current_monotonic
&&
2632 af
->current_realtime
== bf
->current_realtime
&&
2633 af
->current_xor_hash
== bf
->current_xor_hash
)
2636 if (sd_id128_equal(af
->header
->seqnum_id
, bf
->header
->seqnum_id
)) {
2638 /* If this is from the same seqnum source, compare
2640 r
= CMP(af
->current_seqnum
, bf
->current_seqnum
);
2644 /* Wow! This is weird, different data but the same
2645 * seqnums? Something is borked, but let's make the
2646 * best of it and compare by time. */
2649 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
)) {
2651 /* If the boot id matches, compare monotonic time */
2652 r
= CMP(af
->current_monotonic
, bf
->current_monotonic
);
2657 /* Otherwise, compare UTC time */
2658 r
= CMP(af
->current_realtime
, bf
->current_realtime
);
2662 /* Finally, compare by contents */
2663 return CMP(af
->current_xor_hash
, bf
->current_xor_hash
);
2666 static int bump_array_index(uint64_t *i
, direction_t direction
, uint64_t n
) {
2668 /* Increase or decrease the specified index, in the right direction. */
2670 if (direction
== DIRECTION_DOWN
) {
2685 static bool check_properly_ordered(uint64_t new_offset
, uint64_t old_offset
, direction_t direction
) {
2687 /* Consider it an error if any of the two offsets is uninitialized */
2688 if (old_offset
== 0 || new_offset
== 0)
2691 /* If we go down, the new offset must be larger than the old one. */
2692 return direction
== DIRECTION_DOWN
?
2693 new_offset
> old_offset
:
2694 new_offset
< old_offset
;
2697 int journal_file_next_entry(
2700 direction_t direction
,
2701 Object
**ret
, uint64_t *offset
) {
2709 n
= le64toh(f
->header
->n_entries
);
2714 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2716 r
= generic_array_bisect(f
,
2717 le64toh(f
->header
->entry_array_offset
),
2718 le64toh(f
->header
->n_entries
),
2727 r
= bump_array_index(&i
, direction
, n
);
2732 /* And jump to it */
2734 r
= generic_array_get(f
,
2735 le64toh(f
->header
->entry_array_offset
),
2743 /* OK, so this entry is borked. Most likely some entry didn't get synced to disk properly, let's see if
2744 * the next one might work for us instead. */
2745 log_debug_errno(r
, "Entry item %" PRIu64
" is bad, skipping over it.", i
);
2747 r
= bump_array_index(&i
, direction
, n
);
2752 /* Ensure our array is properly ordered. */
2753 if (p
> 0 && !check_properly_ordered(ofs
, p
, direction
)) {
2754 log_debug("%s: entry array not properly ordered at entry %" PRIu64
, f
->path
, i
);
2764 int journal_file_next_entry_for_data(
2766 Object
*o
, uint64_t p
,
2767 uint64_t data_offset
,
2768 direction_t direction
,
2769 Object
**ret
, uint64_t *offset
) {
2776 assert(p
> 0 || !o
);
2778 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2782 n
= le64toh(d
->data
.n_entries
);
2787 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2789 if (o
->object
.type
!= OBJECT_ENTRY
)
2792 r
= generic_array_bisect_plus_one(f
,
2793 le64toh(d
->data
.entry_offset
),
2794 le64toh(d
->data
.entry_array_offset
),
2795 le64toh(d
->data
.n_entries
),
2805 r
= bump_array_index(&i
, direction
, n
);
2811 r
= generic_array_get_plus_one(f
,
2812 le64toh(d
->data
.entry_offset
),
2813 le64toh(d
->data
.entry_array_offset
),
2821 log_debug_errno(r
, "Data entry item %" PRIu64
" is bad, skipping over it.", i
);
2823 r
= bump_array_index(&i
, direction
, n
);
2828 /* Ensure our array is properly ordered. */
2829 if (p
> 0 && check_properly_ordered(ofs
, p
, direction
)) {
2830 log_debug("%s data entry array not properly ordered at entry %" PRIu64
, f
->path
, i
);
2840 int journal_file_move_to_entry_by_offset_for_data(
2842 uint64_t data_offset
,
2844 direction_t direction
,
2845 Object
**ret
, uint64_t *offset
) {
2852 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2856 return generic_array_bisect_plus_one(f
,
2857 le64toh(d
->data
.entry_offset
),
2858 le64toh(d
->data
.entry_array_offset
),
2859 le64toh(d
->data
.n_entries
),
2866 int journal_file_move_to_entry_by_monotonic_for_data(
2868 uint64_t data_offset
,
2871 direction_t direction
,
2872 Object
**ret
, uint64_t *offset
) {
2880 /* First, seek by time */
2881 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &b
);
2887 r
= generic_array_bisect_plus_one(f
,
2888 le64toh(o
->data
.entry_offset
),
2889 le64toh(o
->data
.entry_array_offset
),
2890 le64toh(o
->data
.n_entries
),
2892 test_object_monotonic
,
2898 /* And now, continue seeking until we find an entry that
2899 * exists in both bisection arrays */
2905 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2909 r
= generic_array_bisect_plus_one(f
,
2910 le64toh(d
->data
.entry_offset
),
2911 le64toh(d
->data
.entry_array_offset
),
2912 le64toh(d
->data
.n_entries
),
2920 r
= journal_file_move_to_object(f
, OBJECT_DATA
, b
, &o
);
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
),
2949 int journal_file_move_to_entry_by_seqnum_for_data(
2951 uint64_t data_offset
,
2953 direction_t direction
,
2954 Object
**ret
, uint64_t *offset
) {
2961 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2965 return generic_array_bisect_plus_one(f
,
2966 le64toh(d
->data
.entry_offset
),
2967 le64toh(d
->data
.entry_array_offset
),
2968 le64toh(d
->data
.n_entries
),
2975 int journal_file_move_to_entry_by_realtime_for_data(
2977 uint64_t data_offset
,
2979 direction_t direction
,
2980 Object
**ret
, uint64_t *offset
) {
2987 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2991 return generic_array_bisect_plus_one(f
,
2992 le64toh(d
->data
.entry_offset
),
2993 le64toh(d
->data
.entry_array_offset
),
2994 le64toh(d
->data
.n_entries
),
2996 test_object_realtime
,
3001 void journal_file_dump(JournalFile
*f
) {
3009 journal_file_print_header(f
);
3011 p
= le64toh(f
->header
->header_size
);
3013 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &o
);
3017 switch (o
->object
.type
) {
3020 printf("Type: OBJECT_UNUSED\n");
3024 printf("Type: OBJECT_DATA\n");
3028 printf("Type: OBJECT_FIELD\n");
3032 printf("Type: OBJECT_ENTRY seqnum=%"PRIu64
" monotonic=%"PRIu64
" realtime=%"PRIu64
"\n",
3033 le64toh(o
->entry
.seqnum
),
3034 le64toh(o
->entry
.monotonic
),
3035 le64toh(o
->entry
.realtime
));
3038 case OBJECT_FIELD_HASH_TABLE
:
3039 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
3042 case OBJECT_DATA_HASH_TABLE
:
3043 printf("Type: OBJECT_DATA_HASH_TABLE\n");
3046 case OBJECT_ENTRY_ARRAY
:
3047 printf("Type: OBJECT_ENTRY_ARRAY\n");
3051 printf("Type: OBJECT_TAG seqnum=%"PRIu64
" epoch=%"PRIu64
"\n",
3052 le64toh(o
->tag
.seqnum
),
3053 le64toh(o
->tag
.epoch
));
3057 printf("Type: unknown (%i)\n", o
->object
.type
);
3061 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
)
3062 printf("Flags: %s\n",
3063 object_compressed_to_string(o
->object
.flags
& OBJECT_COMPRESSION_MASK
));
3065 if (p
== le64toh(f
->header
->tail_object_offset
))
3068 p
= p
+ ALIGN64(le64toh(o
->object
.size
));
3073 log_error("File corrupt");
3076 static const char* format_timestamp_safe(char *buf
, size_t l
, usec_t t
) {
3079 x
= format_timestamp(buf
, l
, t
);
3085 void journal_file_print_header(JournalFile
*f
) {
3086 char a
[33], b
[33], c
[33], d
[33];
3087 char x
[FORMAT_TIMESTAMP_MAX
], y
[FORMAT_TIMESTAMP_MAX
], z
[FORMAT_TIMESTAMP_MAX
];
3089 char bytes
[FORMAT_BYTES_MAX
];
3094 printf("File Path: %s\n"
3098 "Sequential Number ID: %s\n"
3100 "Compatible Flags:%s%s\n"
3101 "Incompatible Flags:%s%s%s\n"
3102 "Header size: %"PRIu64
"\n"
3103 "Arena size: %"PRIu64
"\n"
3104 "Data Hash Table Size: %"PRIu64
"\n"
3105 "Field Hash Table Size: %"PRIu64
"\n"
3106 "Rotate Suggested: %s\n"
3107 "Head Sequential Number: %"PRIu64
" (%"PRIx64
")\n"
3108 "Tail Sequential Number: %"PRIu64
" (%"PRIx64
")\n"
3109 "Head Realtime Timestamp: %s (%"PRIx64
")\n"
3110 "Tail Realtime Timestamp: %s (%"PRIx64
")\n"
3111 "Tail Monotonic Timestamp: %s (%"PRIx64
")\n"
3112 "Objects: %"PRIu64
"\n"
3113 "Entry Objects: %"PRIu64
"\n",
3115 sd_id128_to_string(f
->header
->file_id
, a
),
3116 sd_id128_to_string(f
->header
->machine_id
, b
),
3117 sd_id128_to_string(f
->header
->boot_id
, c
),
3118 sd_id128_to_string(f
->header
->seqnum_id
, d
),
3119 f
->header
->state
== STATE_OFFLINE
? "OFFLINE" :
3120 f
->header
->state
== STATE_ONLINE
? "ONLINE" :
3121 f
->header
->state
== STATE_ARCHIVED
? "ARCHIVED" : "UNKNOWN",
3122 JOURNAL_HEADER_SEALED(f
->header
) ? " SEALED" : "",
3123 (le32toh(f
->header
->compatible_flags
) & ~HEADER_COMPATIBLE_ANY
) ? " ???" : "",
3124 JOURNAL_HEADER_COMPRESSED_XZ(f
->header
) ? " COMPRESSED-XZ" : "",
3125 JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
) ? " COMPRESSED-LZ4" : "",
3126 (le32toh(f
->header
->incompatible_flags
) & ~HEADER_INCOMPATIBLE_ANY
) ? " ???" : "",
3127 le64toh(f
->header
->header_size
),
3128 le64toh(f
->header
->arena_size
),
3129 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3130 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
),
3131 yes_no(journal_file_rotate_suggested(f
, 0)),
3132 le64toh(f
->header
->head_entry_seqnum
), le64toh(f
->header
->head_entry_seqnum
),
3133 le64toh(f
->header
->tail_entry_seqnum
), le64toh(f
->header
->tail_entry_seqnum
),
3134 format_timestamp_safe(x
, sizeof(x
), le64toh(f
->header
->head_entry_realtime
)), le64toh(f
->header
->head_entry_realtime
),
3135 format_timestamp_safe(y
, sizeof(y
), le64toh(f
->header
->tail_entry_realtime
)), le64toh(f
->header
->tail_entry_realtime
),
3136 format_timespan(z
, sizeof(z
), le64toh(f
->header
->tail_entry_monotonic
), USEC_PER_MSEC
), le64toh(f
->header
->tail_entry_monotonic
),
3137 le64toh(f
->header
->n_objects
),
3138 le64toh(f
->header
->n_entries
));
3140 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3141 printf("Data Objects: %"PRIu64
"\n"
3142 "Data Hash Table Fill: %.1f%%\n",
3143 le64toh(f
->header
->n_data
),
3144 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))));
3146 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3147 printf("Field Objects: %"PRIu64
"\n"
3148 "Field Hash Table Fill: %.1f%%\n",
3149 le64toh(f
->header
->n_fields
),
3150 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))));
3152 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
))
3153 printf("Tag Objects: %"PRIu64
"\n",
3154 le64toh(f
->header
->n_tags
));
3155 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
3156 printf("Entry Array Objects: %"PRIu64
"\n",
3157 le64toh(f
->header
->n_entry_arrays
));
3159 if (fstat(f
->fd
, &st
) >= 0)
3160 printf("Disk usage: %s\n", format_bytes(bytes
, sizeof(bytes
), (uint64_t) st
.st_blocks
* 512ULL));
3163 static int journal_file_warn_btrfs(JournalFile
*f
) {
3169 /* Before we write anything, check if the COW logic is turned
3170 * off on btrfs. Given our write pattern that is quite
3171 * unfriendly to COW file systems this should greatly improve
3172 * performance on COW file systems, such as btrfs, at the
3173 * expense of data integrity features (which shouldn't be too
3174 * bad, given that we do our own checksumming). */
3176 r
= btrfs_is_filesystem(f
->fd
);
3178 return log_warning_errno(r
, "Failed to determine if journal is on btrfs: %m");
3182 r
= read_attr_fd(f
->fd
, &attrs
);
3184 return log_warning_errno(r
, "Failed to read file attributes: %m");
3186 if (attrs
& FS_NOCOW_FL
) {
3187 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3191 log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3192 "This is likely to slow down journal access substantially, please consider turning "
3193 "off the copy-on-write file attribute on the journal directory, using chattr +C.", f
->path
);
3198 int journal_file_open(
3204 uint64_t compress_threshold_bytes
,
3206 JournalMetrics
*metrics
,
3207 MMapCache
*mmap_cache
,
3208 Set
*deferred_closes
,
3209 JournalFile
*template,
3210 JournalFile
**ret
) {
3212 bool newly_created
= false;
3216 char bytes
[FORMAT_BYTES_MAX
];
3219 assert(fd
>= 0 || fname
);
3221 if (!IN_SET((flags
& O_ACCMODE
), O_RDONLY
, O_RDWR
))
3224 if (fname
&& (flags
& O_CREAT
) && !endswith(fname
, ".journal"))
3227 f
= new(JournalFile
, 1);
3231 *f
= (JournalFile
) {
3236 .prot
= prot_from_flags(flags
),
3237 .writable
= (flags
& O_ACCMODE
) != O_RDONLY
,
3240 .compress_lz4
= compress
,
3242 .compress_xz
= compress
,
3244 .compress_threshold_bytes
= compress_threshold_bytes
== (uint64_t) -1 ?
3245 DEFAULT_COMPRESS_THRESHOLD
:
3246 MAX(MIN_COMPRESS_THRESHOLD
, compress_threshold_bytes
),
3252 log_debug("Journal effective settings seal=%s compress=%s compress_threshold_bytes=%s",
3253 yes_no(f
->seal
), yes_no(JOURNAL_FILE_COMPRESS(f
)),
3254 format_bytes(bytes
, sizeof(bytes
), f
->compress_threshold_bytes
));
3257 f
->mmap
= mmap_cache_ref(mmap_cache
);
3259 f
->mmap
= mmap_cache_new();
3267 f
->path
= strdup(fname
);
3275 /* If we don't know the path, fill in something explanatory and vaguely useful */
3276 if (asprintf(&f
->path
, "/proc/self/%i", fd
) < 0) {
3282 f
->chain_cache
= ordered_hashmap_new(&uint64_hash_ops
);
3283 if (!f
->chain_cache
) {
3289 /* We pass O_NONBLOCK here, so that in case somebody pointed us to some character device node or FIFO
3290 * or so, we likely fail quickly than block for long. For regular files O_NONBLOCK has no effect, hence
3291 * it doesn't hurt in that case. */
3293 f
->fd
= open(f
->path
, f
->flags
|O_CLOEXEC
|O_NONBLOCK
, f
->mode
);
3299 /* fds we opened here by us should also be closed by us. */
3302 r
= fd_nonblock(f
->fd
, false);
3307 f
->cache_fd
= mmap_cache_add_fd(f
->mmap
, f
->fd
);
3313 r
= journal_file_fstat(f
);
3317 if (f
->last_stat
.st_size
== 0 && f
->writable
) {
3319 (void) journal_file_warn_btrfs(f
);
3321 /* Let's attach the creation time to the journal file, so that the vacuuming code knows the age of this
3322 * file even if the file might end up corrupted one day... Ideally we'd just use the creation time many
3323 * file systems maintain for each file, but the API to query this is very new, hence let's emulate this
3324 * via extended attributes. If extended attributes are not supported we'll just skip this, and rely
3325 * solely on mtime/atime/ctime of the file. */
3326 (void) fd_setcrtime(f
->fd
, 0);
3329 /* Try to load the FSPRG state, and if we can't, then
3330 * just don't do sealing */
3332 r
= journal_file_fss_load(f
);
3338 r
= journal_file_init_header(f
, template);
3342 r
= journal_file_fstat(f
);
3346 newly_created
= true;
3349 if (f
->last_stat
.st_size
< (off_t
) HEADER_SIZE_MIN
) {
3354 r
= mmap_cache_get(f
->mmap
, f
->cache_fd
, f
->prot
, CONTEXT_HEADER
, true, 0, PAGE_ALIGN(sizeof(Header
)), &f
->last_stat
, &h
, NULL
);
3360 if (!newly_created
) {
3361 set_clear_with_destructor(deferred_closes
, journal_file_close
);
3363 r
= journal_file_verify_header(f
);
3369 if (!newly_created
&& f
->writable
) {
3370 r
= journal_file_fss_load(f
);
3378 journal_default_metrics(metrics
, f
->fd
);
3379 f
->metrics
= *metrics
;
3380 } else if (template)
3381 f
->metrics
= template->metrics
;
3383 r
= journal_file_refresh_header(f
);
3389 r
= journal_file_hmac_setup(f
);
3394 if (newly_created
) {
3395 r
= journal_file_setup_field_hash_table(f
);
3399 r
= journal_file_setup_data_hash_table(f
);
3404 r
= journal_file_append_first_tag(f
);
3410 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
)) {
3415 if (template && template->post_change_timer
) {
3416 r
= journal_file_enable_post_change_timer(
3418 sd_event_source_get_event(template->post_change_timer
),
3419 template->post_change_timer_period
);
3425 /* The file is opened now successfully, thus we take possession of any passed in fd. */
3432 if (f
->cache_fd
&& mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
3435 (void) journal_file_close(f
);
3440 int journal_file_archive(JournalFile
*f
) {
3441 _cleanup_free_
char *p
= NULL
;
3448 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
3449 * rotation, since we don't know the actual path, and couldn't rename the file hence. */
3450 if (path_startswith(f
->path
, "/proc/self/fd"))
3453 if (!endswith(f
->path
, ".journal"))
3456 if (asprintf(&p
, "%.*s@" SD_ID128_FORMAT_STR
"-%016"PRIx64
"-%016"PRIx64
".journal",
3457 (int) strlen(f
->path
) - 8, f
->path
,
3458 SD_ID128_FORMAT_VAL(f
->header
->seqnum_id
),
3459 le64toh(f
->header
->head_entry_seqnum
),
3460 le64toh(f
->header
->head_entry_realtime
)) < 0)
3463 /* Try to rename the file to the archived version. If the file already was deleted, we'll get ENOENT, let's
3464 * ignore that case. */
3465 if (rename(f
->path
, p
) < 0 && errno
!= ENOENT
)
3468 /* Sync the rename to disk */
3469 (void) fsync_directory_of_file(f
->fd
);
3471 /* Set as archive so offlining commits w/state=STATE_ARCHIVED. Previously we would set old_file->header->state
3472 * to STATE_ARCHIVED directly here, but journal_file_set_offline() short-circuits when state != STATE_ONLINE,
3473 * which would result in the rotated journal never getting fsync() called before closing. Now we simply queue
3474 * the archive state by setting an archive bit, leaving the state as STATE_ONLINE so proper offlining
3478 /* Currently, btrfs is not very good with out write patterns and fragments heavily. Let's defrag our journal
3479 * files when we archive them */
3480 f
->defrag_on_close
= true;
3485 JournalFile
* journal_initiate_close(
3487 Set
*deferred_closes
) {
3493 if (deferred_closes
) {
3495 r
= set_put(deferred_closes
, f
);
3497 log_debug_errno(r
, "Failed to add file to deferred close set, closing immediately.");
3499 (void) journal_file_set_offline(f
, false);
3504 return journal_file_close(f
);
3507 int journal_file_rotate(
3510 uint64_t compress_threshold_bytes
,
3512 Set
*deferred_closes
) {
3514 JournalFile
*new_file
= NULL
;
3520 r
= journal_file_archive(*f
);
3524 r
= journal_file_open(
3530 compress_threshold_bytes
,
3538 journal_initiate_close(*f
, deferred_closes
);
3544 int journal_file_dispose(int dir_fd
, const char *fname
) {
3545 _cleanup_free_
char *p
= NULL
;
3546 _cleanup_close_
int fd
= -1;
3550 /* Renames a journal file to *.journal~, i.e. to mark it as corruped or otherwise uncleanly shutdown. Note that
3551 * this is done without looking into the file or changing any of its contents. The idea is that this is called
3552 * whenever something is suspicious and we want to move the file away and make clear that it is not accessed
3553 * for writing anymore. */
3555 if (!endswith(fname
, ".journal"))
3558 if (asprintf(&p
, "%.*s@%016" PRIx64
"-%016" PRIx64
".journal~",
3559 (int) strlen(fname
) - 8, fname
,
3560 now(CLOCK_REALTIME
),
3564 if (renameat(dir_fd
, fname
, dir_fd
, p
) < 0)
3567 /* btrfs doesn't cope well with our write pattern and fragments heavily. Let's defrag all files we rotate */
3568 fd
= openat(dir_fd
, p
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
|O_NOFOLLOW
);
3570 log_debug_errno(errno
, "Failed to open file for defragmentation/FS_NOCOW_FL, ignoring: %m");
3572 (void) chattr_fd(fd
, 0, FS_NOCOW_FL
, NULL
);
3573 (void) btrfs_defrag_fd(fd
);
3579 int journal_file_open_reliably(
3584 uint64_t compress_threshold_bytes
,
3586 JournalMetrics
*metrics
,
3587 MMapCache
*mmap_cache
,
3588 Set
*deferred_closes
,
3589 JournalFile
*template,
3590 JournalFile
**ret
) {
3592 _cleanup_free_
char *p
= NULL
;
3595 r
= journal_file_open(-1, fname
, flags
, mode
, compress
, compress_threshold_bytes
, seal
, metrics
, mmap_cache
,
3596 deferred_closes
, template, ret
);
3598 -EBADMSG
, /* Corrupted */
3599 -ENODATA
, /* Truncated */
3600 -EHOSTDOWN
, /* Other machine */
3601 -EPROTONOSUPPORT
, /* Incompatible feature */
3602 -EBUSY
, /* Unclean shutdown */
3603 -ESHUTDOWN
, /* Already archived */
3604 -EIO
, /* IO error, including SIGBUS on mmap */
3605 -EIDRM
, /* File has been deleted */
3606 -ETXTBSY
)) /* File is from the future */
3609 if ((flags
& O_ACCMODE
) == O_RDONLY
)
3612 if (!(flags
& O_CREAT
))
3615 if (!endswith(fname
, ".journal"))
3618 /* The file is corrupted. Rotate it away and try it again (but only once) */
3619 log_warning_errno(r
, "File %s corrupted or uncleanly shut down, renaming and replacing.", fname
);
3621 r
= journal_file_dispose(AT_FDCWD
, fname
);
3625 return journal_file_open(-1, fname
, flags
, mode
, compress
, compress_threshold_bytes
, seal
, metrics
, mmap_cache
,
3626 deferred_closes
, template, ret
);
3629 int journal_file_copy_entry(JournalFile
*from
, JournalFile
*to
, Object
*o
, uint64_t p
) {
3631 uint64_t q
, xor_hash
= 0;
3635 const sd_id128_t
*boot_id
;
3645 ts
.monotonic
= le64toh(o
->entry
.monotonic
);
3646 ts
.realtime
= le64toh(o
->entry
.realtime
);
3647 boot_id
= &o
->entry
.boot_id
;
3649 n
= journal_file_entry_n_items(o
);
3650 /* alloca() can't take 0, hence let's allocate at least one */
3651 items
= newa(EntryItem
, MAX(1u, n
));
3653 for (i
= 0; i
< n
; i
++) {
3660 q
= le64toh(o
->entry
.items
[i
].object_offset
);
3661 le_hash
= o
->entry
.items
[i
].hash
;
3663 r
= journal_file_move_to_object(from
, OBJECT_DATA
, q
, &o
);
3667 if (le_hash
!= o
->data
.hash
)
3670 l
= le64toh(o
->object
.size
) - offsetof(Object
, data
.payload
);
3673 /* We hit the limit on 32bit machines */
3674 if ((uint64_t) t
!= l
)
3677 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
3678 #if HAVE_XZ || HAVE_LZ4
3681 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
3682 o
->data
.payload
, l
, &from
->compress_buffer
, &from
->compress_buffer_size
, &rsize
, 0);
3686 data
= from
->compress_buffer
;
3689 return -EPROTONOSUPPORT
;
3692 data
= o
->data
.payload
;
3694 r
= journal_file_append_data(to
, data
, l
, &u
, &h
);
3698 xor_hash
^= le64toh(u
->data
.hash
);
3699 items
[i
].object_offset
= htole64(h
);
3700 items
[i
].hash
= u
->data
.hash
;
3702 r
= journal_file_move_to_object(from
, OBJECT_ENTRY
, p
, &o
);
3707 r
= journal_file_append_entry_internal(to
, &ts
, boot_id
, xor_hash
, items
, n
,
3710 if (mmap_cache_got_sigbus(to
->mmap
, to
->cache_fd
))
3716 void journal_reset_metrics(JournalMetrics
*m
) {
3719 /* Set everything to "pick automatic values". */
3721 *m
= (JournalMetrics
) {
3722 .min_use
= (uint64_t) -1,
3723 .max_use
= (uint64_t) -1,
3724 .min_size
= (uint64_t) -1,
3725 .max_size
= (uint64_t) -1,
3726 .keep_free
= (uint64_t) -1,
3727 .n_max_files
= (uint64_t) -1,
3731 void journal_default_metrics(JournalMetrics
*m
, int fd
) {
3732 char a
[FORMAT_BYTES_MAX
], b
[FORMAT_BYTES_MAX
], c
[FORMAT_BYTES_MAX
], d
[FORMAT_BYTES_MAX
], e
[FORMAT_BYTES_MAX
];
3739 if (fstatvfs(fd
, &ss
) >= 0)
3740 fs_size
= ss
.f_frsize
* ss
.f_blocks
;
3742 log_debug_errno(errno
, "Failed to determine disk size: %m");
3746 if (m
->max_use
== (uint64_t) -1) {
3749 m
->max_use
= PAGE_ALIGN(fs_size
/ 10); /* 10% of file system size */
3751 if (m
->max_use
> DEFAULT_MAX_USE_UPPER
)
3752 m
->max_use
= DEFAULT_MAX_USE_UPPER
;
3754 if (m
->max_use
< DEFAULT_MAX_USE_LOWER
)
3755 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3757 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3759 m
->max_use
= PAGE_ALIGN(m
->max_use
);
3761 if (m
->max_use
!= 0 && m
->max_use
< JOURNAL_FILE_SIZE_MIN
*2)
3762 m
->max_use
= JOURNAL_FILE_SIZE_MIN
*2;
3765 if (m
->min_use
== (uint64_t) -1)
3766 m
->min_use
= DEFAULT_MIN_USE
;
3768 if (m
->min_use
> m
->max_use
)
3769 m
->min_use
= m
->max_use
;
3771 if (m
->max_size
== (uint64_t) -1) {
3772 m
->max_size
= PAGE_ALIGN(m
->max_use
/ 8); /* 8 chunks */
3774 if (m
->max_size
> DEFAULT_MAX_SIZE_UPPER
)
3775 m
->max_size
= DEFAULT_MAX_SIZE_UPPER
;
3777 m
->max_size
= PAGE_ALIGN(m
->max_size
);
3779 if (m
->max_size
!= 0) {
3780 if (m
->max_size
< JOURNAL_FILE_SIZE_MIN
)
3781 m
->max_size
= JOURNAL_FILE_SIZE_MIN
;
3783 if (m
->max_use
!= 0 && m
->max_size
*2 > m
->max_use
)
3784 m
->max_use
= m
->max_size
*2;
3787 if (m
->min_size
== (uint64_t) -1)
3788 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3790 m
->min_size
= PAGE_ALIGN(m
->min_size
);
3792 if (m
->min_size
< JOURNAL_FILE_SIZE_MIN
)
3793 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3795 if (m
->max_size
!= 0 && m
->min_size
> m
->max_size
)
3796 m
->max_size
= m
->min_size
;
3799 if (m
->keep_free
== (uint64_t) -1) {
3802 m
->keep_free
= PAGE_ALIGN(fs_size
* 3 / 20); /* 15% of file system size */
3804 if (m
->keep_free
> DEFAULT_KEEP_FREE_UPPER
)
3805 m
->keep_free
= DEFAULT_KEEP_FREE_UPPER
;
3808 m
->keep_free
= DEFAULT_KEEP_FREE
;
3811 if (m
->n_max_files
== (uint64_t) -1)
3812 m
->n_max_files
= DEFAULT_N_MAX_FILES
;
3814 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64
,
3815 format_bytes(a
, sizeof(a
), m
->min_use
),
3816 format_bytes(b
, sizeof(b
), m
->max_use
),
3817 format_bytes(c
, sizeof(c
), m
->max_size
),
3818 format_bytes(d
, sizeof(d
), m
->min_size
),
3819 format_bytes(e
, sizeof(e
), m
->keep_free
),
3823 int journal_file_get_cutoff_realtime_usec(JournalFile
*f
, usec_t
*from
, usec_t
*to
) {
3829 if (f
->header
->head_entry_realtime
== 0)
3832 *from
= le64toh(f
->header
->head_entry_realtime
);
3836 if (f
->header
->tail_entry_realtime
== 0)
3839 *to
= le64toh(f
->header
->tail_entry_realtime
);
3845 int journal_file_get_cutoff_monotonic_usec(JournalFile
*f
, sd_id128_t boot_id
, usec_t
*from
, usec_t
*to
) {
3853 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &p
);
3857 if (le64toh(o
->data
.n_entries
) <= 0)
3861 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, le64toh(o
->data
.entry_offset
), &o
);
3865 *from
= le64toh(o
->entry
.monotonic
);
3869 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
3873 r
= generic_array_get_plus_one(f
,
3874 le64toh(o
->data
.entry_offset
),
3875 le64toh(o
->data
.entry_array_offset
),
3876 le64toh(o
->data
.n_entries
)-1,
3881 *to
= le64toh(o
->entry
.monotonic
);
3887 bool journal_file_rotate_suggested(JournalFile
*f
, usec_t max_file_usec
) {
3891 /* If we gained new header fields we gained new features,
3892 * hence suggest a rotation */
3893 if (le64toh(f
->header
->header_size
) < sizeof(Header
)) {
3894 log_debug("%s uses an outdated header, suggesting rotation.", f
->path
);
3898 /* Let's check if the hash tables grew over a certain fill
3899 * level (75%, borrowing this value from Java's hash table
3900 * implementation), and if so suggest a rotation. To calculate
3901 * the fill level we need the n_data field, which only exists
3902 * in newer versions. */
3904 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3905 if (le64toh(f
->header
->n_data
) * 4ULL > (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3906 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.",
3908 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))),
3909 le64toh(f
->header
->n_data
),
3910 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3911 (unsigned long long) f
->last_stat
.st_size
,
3912 f
->last_stat
.st_size
/ le64toh(f
->header
->n_data
));
3916 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3917 if (le64toh(f
->header
->n_fields
) * 4ULL > (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3918 log_debug("Field hash table of %s has a fill level at %.1f (%"PRIu64
" of %"PRIu64
" items), suggesting rotation.",
3920 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))),
3921 le64toh(f
->header
->n_fields
),
3922 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
));
3926 /* Are the data objects properly indexed by field objects? */
3927 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
3928 JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
3929 le64toh(f
->header
->n_data
) > 0 &&
3930 le64toh(f
->header
->n_fields
) == 0)
3933 if (max_file_usec
> 0) {
3936 h
= le64toh(f
->header
->head_entry_realtime
);
3937 t
= now(CLOCK_REALTIME
);
3939 if (h
> 0 && t
> h
+ max_file_usec
)