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
) {
352 if (sd_event_source_get_enabled(f
->post_change_timer
, NULL
) > 0)
353 journal_file_post_change(f
);
355 (void) sd_event_source_set_enabled(f
->post_change_timer
, SD_EVENT_OFF
);
356 sd_event_source_unref(f
->post_change_timer
);
359 journal_file_set_offline(f
, true);
361 if (f
->mmap
&& f
->cache_fd
)
362 mmap_cache_free_fd(f
->mmap
, f
->cache_fd
);
364 if (f
->fd
>= 0 && f
->defrag_on_close
) {
366 /* Be friendly to btrfs: turn COW back on again now,
367 * and defragment the file. We won't write to the file
368 * ever again, hence remove all fragmentation, and
369 * reenable all the good bits COW usually provides
370 * (such as data checksumming). */
372 (void) chattr_fd(f
->fd
, 0, FS_NOCOW_FL
, NULL
);
373 (void) btrfs_defrag_fd(f
->fd
);
380 mmap_cache_unref(f
->mmap
);
382 ordered_hashmap_free_free(f
->chain_cache
);
384 #if HAVE_XZ || HAVE_LZ4
385 free(f
->compress_buffer
);
390 munmap(f
->fss_file
, PAGE_ALIGN(f
->fss_file_size
));
392 free(f
->fsprg_state
);
397 gcry_md_close(f
->hmac
);
403 static int journal_file_init_header(JournalFile
*f
, JournalFile
*template) {
410 memcpy(h
.signature
, HEADER_SIGNATURE
, 8);
411 h
.header_size
= htole64(ALIGN64(sizeof(h
)));
413 h
.incompatible_flags
|= htole32(
414 f
->compress_xz
* HEADER_INCOMPATIBLE_COMPRESSED_XZ
|
415 f
->compress_lz4
* HEADER_INCOMPATIBLE_COMPRESSED_LZ4
);
417 h
.compatible_flags
= htole32(
418 f
->seal
* HEADER_COMPATIBLE_SEALED
);
420 r
= sd_id128_randomize(&h
.file_id
);
425 h
.seqnum_id
= template->header
->seqnum_id
;
426 h
.tail_entry_seqnum
= template->header
->tail_entry_seqnum
;
428 h
.seqnum_id
= h
.file_id
;
430 k
= pwrite(f
->fd
, &h
, sizeof(h
), 0);
440 static int journal_file_refresh_header(JournalFile
*f
) {
447 r
= sd_id128_get_machine(&f
->header
->machine_id
);
448 if (IN_SET(r
, -ENOENT
, -ENOMEDIUM
))
449 /* We don't have a machine-id, let's continue without */
450 zero(f
->header
->machine_id
);
454 r
= sd_id128_get_boot(&boot_id
);
458 f
->header
->boot_id
= boot_id
;
460 r
= journal_file_set_online(f
);
462 /* Sync the online state to disk */
465 /* We likely just created a new file, also sync the directory this file is located in. */
466 (void) fsync_directory_of_file(f
->fd
);
471 static bool warn_wrong_flags(const JournalFile
*f
, bool compatible
) {
472 const uint32_t any
= compatible
? HEADER_COMPATIBLE_ANY
: HEADER_INCOMPATIBLE_ANY
,
473 supported
= compatible
? HEADER_COMPATIBLE_SUPPORTED
: HEADER_INCOMPATIBLE_SUPPORTED
;
474 const char *type
= compatible
? "compatible" : "incompatible";
477 flags
= le32toh(compatible
? f
->header
->compatible_flags
: f
->header
->incompatible_flags
);
479 if (flags
& ~supported
) {
481 log_debug("Journal file %s has unknown %s flags 0x%"PRIx32
,
482 f
->path
, type
, flags
& ~any
);
483 flags
= (flags
& any
) & ~supported
;
487 _cleanup_free_
char *t
= NULL
;
489 if (compatible
&& (flags
& HEADER_COMPATIBLE_SEALED
))
490 strv
[n
++] = "sealed";
491 if (!compatible
&& (flags
& HEADER_INCOMPATIBLE_COMPRESSED_XZ
))
492 strv
[n
++] = "xz-compressed";
493 if (!compatible
&& (flags
& HEADER_INCOMPATIBLE_COMPRESSED_LZ4
))
494 strv
[n
++] = "lz4-compressed";
496 assert(n
< ELEMENTSOF(strv
));
498 t
= strv_join((char**) strv
, ", ");
499 log_debug("Journal file %s uses %s %s %s disabled at compilation time.",
500 f
->path
, type
, n
> 1 ? "flags" : "flag", strnull(t
));
508 static int journal_file_verify_header(JournalFile
*f
) {
509 uint64_t arena_size
, header_size
;
514 if (memcmp(f
->header
->signature
, HEADER_SIGNATURE
, 8))
517 /* In both read and write mode we refuse to open files with incompatible
518 * flags we don't know. */
519 if (warn_wrong_flags(f
, false))
520 return -EPROTONOSUPPORT
;
522 /* When open for writing we refuse to open files with compatible flags, too. */
523 if (f
->writable
&& warn_wrong_flags(f
, true))
524 return -EPROTONOSUPPORT
;
526 if (f
->header
->state
>= _STATE_MAX
)
529 header_size
= le64toh(f
->header
->header_size
);
531 /* The first addition was n_data, so check that we are at least this large */
532 if (header_size
< HEADER_SIZE_MIN
)
535 if (JOURNAL_HEADER_SEALED(f
->header
) && !JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
538 arena_size
= le64toh(f
->header
->arena_size
);
540 if (UINT64_MAX
- header_size
< arena_size
|| header_size
+ arena_size
> (uint64_t) f
->last_stat
.st_size
)
543 if (le64toh(f
->header
->tail_object_offset
) > header_size
+ arena_size
)
546 if (!VALID64(le64toh(f
->header
->data_hash_table_offset
)) ||
547 !VALID64(le64toh(f
->header
->field_hash_table_offset
)) ||
548 !VALID64(le64toh(f
->header
->tail_object_offset
)) ||
549 !VALID64(le64toh(f
->header
->entry_array_offset
)))
553 sd_id128_t machine_id
;
557 r
= sd_id128_get_machine(&machine_id
);
561 if (!sd_id128_equal(machine_id
, f
->header
->machine_id
))
564 state
= f
->header
->state
;
566 if (state
== STATE_ARCHIVED
)
567 return -ESHUTDOWN
; /* Already archived */
568 else if (state
== STATE_ONLINE
) {
569 log_debug("Journal file %s is already online. Assuming unclean closing.", f
->path
);
571 } else if (state
!= STATE_OFFLINE
) {
572 log_debug("Journal file %s has unknown state %i.", f
->path
, state
);
576 if (f
->header
->field_hash_table_size
== 0 || f
->header
->data_hash_table_size
== 0)
579 /* Don't permit appending to files from the future. Because otherwise the realtime timestamps wouldn't
580 * be strictly ordered in the entries in the file anymore, and we can't have that since it breaks
582 if (le64toh(f
->header
->tail_entry_realtime
) > now(CLOCK_REALTIME
)) {
583 log_debug("Journal file %s is from the future, refusing to append new data to it that'd be older.", f
->path
);
588 f
->compress_xz
= JOURNAL_HEADER_COMPRESSED_XZ(f
->header
);
589 f
->compress_lz4
= JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
);
591 f
->seal
= JOURNAL_HEADER_SEALED(f
->header
);
596 static int journal_file_fstat(JournalFile
*f
) {
602 if (fstat(f
->fd
, &f
->last_stat
) < 0)
605 f
->last_stat_usec
= now(CLOCK_MONOTONIC
);
607 /* Refuse dealing with with files that aren't regular */
608 r
= stat_verify_regular(&f
->last_stat
);
612 /* Refuse appending to files that are already deleted */
613 if (f
->last_stat
.st_nlink
<= 0)
619 static int journal_file_allocate(JournalFile
*f
, uint64_t offset
, uint64_t size
) {
620 uint64_t old_size
, new_size
;
626 /* We assume that this file is not sparse, and we know that
627 * for sure, since we always call posix_fallocate()
630 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
634 le64toh(f
->header
->header_size
) +
635 le64toh(f
->header
->arena_size
);
637 new_size
= PAGE_ALIGN(offset
+ size
);
638 if (new_size
< le64toh(f
->header
->header_size
))
639 new_size
= le64toh(f
->header
->header_size
);
641 if (new_size
<= old_size
) {
643 /* We already pre-allocated enough space, but before
644 * we write to it, let's check with fstat() if the
645 * file got deleted, in order make sure we don't throw
646 * away the data immediately. Don't check fstat() for
647 * all writes though, but only once ever 10s. */
649 if (f
->last_stat_usec
+ LAST_STAT_REFRESH_USEC
> now(CLOCK_MONOTONIC
))
652 return journal_file_fstat(f
);
655 /* Allocate more space. */
657 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
660 if (new_size
> f
->metrics
.min_size
&& f
->metrics
.keep_free
> 0) {
663 if (fstatvfs(f
->fd
, &svfs
) >= 0) {
666 available
= LESS_BY((uint64_t) svfs
.f_bfree
* (uint64_t) svfs
.f_bsize
, f
->metrics
.keep_free
);
668 if (new_size
- old_size
> available
)
673 /* Increase by larger blocks at once */
674 new_size
= DIV_ROUND_UP(new_size
, FILE_SIZE_INCREASE
) * FILE_SIZE_INCREASE
;
675 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
676 new_size
= f
->metrics
.max_size
;
678 /* Note that the glibc fallocate() fallback is very
679 inefficient, hence we try to minimize the allocation area
681 r
= posix_fallocate(f
->fd
, old_size
, new_size
- old_size
);
685 f
->header
->arena_size
= htole64(new_size
- le64toh(f
->header
->header_size
));
687 return journal_file_fstat(f
);
690 static unsigned type_to_context(ObjectType type
) {
691 /* One context for each type, plus one catch-all for the rest */
692 assert_cc(_OBJECT_TYPE_MAX
<= MMAP_CACHE_MAX_CONTEXTS
);
693 assert_cc(CONTEXT_HEADER
< MMAP_CACHE_MAX_CONTEXTS
);
694 return type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
? type
: 0;
697 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
) {
706 /* Avoid SIGBUS on invalid accesses */
707 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
) {
708 /* Hmm, out of range? Let's refresh the fstat() data
709 * first, before we trust that check. */
711 r
= journal_file_fstat(f
);
715 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
)
716 return -EADDRNOTAVAIL
;
719 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
);
722 static uint64_t minimum_header_size(Object
*o
) {
724 static const uint64_t table
[] = {
725 [OBJECT_DATA
] = sizeof(DataObject
),
726 [OBJECT_FIELD
] = sizeof(FieldObject
),
727 [OBJECT_ENTRY
] = sizeof(EntryObject
),
728 [OBJECT_DATA_HASH_TABLE
] = sizeof(HashTableObject
),
729 [OBJECT_FIELD_HASH_TABLE
] = sizeof(HashTableObject
),
730 [OBJECT_ENTRY_ARRAY
] = sizeof(EntryArrayObject
),
731 [OBJECT_TAG
] = sizeof(TagObject
),
734 if (o
->object
.type
>= ELEMENTSOF(table
) || table
[o
->object
.type
] <= 0)
735 return sizeof(ObjectHeader
);
737 return table
[o
->object
.type
];
740 /* Lightweight object checks. We want this to be fast, so that we won't
741 * slowdown every journal_file_move_to_object() call too much. */
742 static int journal_file_check_object(JournalFile
*f
, uint64_t offset
, Object
*o
) {
746 switch (o
->object
.type
) {
749 if ((le64toh(o
->data
.entry_offset
) == 0) ^ (le64toh(o
->data
.n_entries
) == 0)) {
750 log_debug("Bad n_entries: %"PRIu64
": %"PRIu64
,
751 le64toh(o
->data
.n_entries
), offset
);
755 if (le64toh(o
->object
.size
) - offsetof(DataObject
, payload
) <= 0) {
756 log_debug("Bad object size (<= %zu): %"PRIu64
": %"PRIu64
,
757 offsetof(DataObject
, payload
),
758 le64toh(o
->object
.size
),
763 if (!VALID64(le64toh(o
->data
.next_hash_offset
)) ||
764 !VALID64(le64toh(o
->data
.next_field_offset
)) ||
765 !VALID64(le64toh(o
->data
.entry_offset
)) ||
766 !VALID64(le64toh(o
->data
.entry_array_offset
))) {
767 log_debug("Invalid offset, next_hash_offset="OFSfmt
", next_field_offset="OFSfmt
768 ", entry_offset="OFSfmt
", entry_array_offset="OFSfmt
": %"PRIu64
,
769 le64toh(o
->data
.next_hash_offset
),
770 le64toh(o
->data
.next_field_offset
),
771 le64toh(o
->data
.entry_offset
),
772 le64toh(o
->data
.entry_array_offset
),
781 if (le64toh(o
->object
.size
) - offsetof(FieldObject
, payload
) <= 0) {
783 "Bad field size (<= %zu): %"PRIu64
": %"PRIu64
,
784 offsetof(FieldObject
, payload
),
785 le64toh(o
->object
.size
),
790 if (!VALID64(le64toh(o
->field
.next_hash_offset
)) ||
791 !VALID64(le64toh(o
->field
.head_data_offset
))) {
793 "Invalid offset, next_hash_offset="OFSfmt
794 ", head_data_offset="OFSfmt
": %"PRIu64
,
795 le64toh(o
->field
.next_hash_offset
),
796 le64toh(o
->field
.head_data_offset
),
803 if ((le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) % sizeof(EntryItem
) != 0) {
805 "Bad entry size (<= %zu): %"PRIu64
": %"PRIu64
,
806 offsetof(EntryObject
, items
),
807 le64toh(o
->object
.size
),
812 if ((le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) / sizeof(EntryItem
) <= 0) {
814 "Invalid number items in entry: %"PRIu64
": %"PRIu64
,
815 (le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) / sizeof(EntryItem
),
820 if (le64toh(o
->entry
.seqnum
) <= 0) {
822 "Invalid entry seqnum: %"PRIx64
": %"PRIu64
,
823 le64toh(o
->entry
.seqnum
),
828 if (!VALID_REALTIME(le64toh(o
->entry
.realtime
))) {
830 "Invalid entry realtime timestamp: %"PRIu64
": %"PRIu64
,
831 le64toh(o
->entry
.realtime
),
836 if (!VALID_MONOTONIC(le64toh(o
->entry
.monotonic
))) {
838 "Invalid entry monotonic timestamp: %"PRIu64
": %"PRIu64
,
839 le64toh(o
->entry
.monotonic
),
846 case OBJECT_DATA_HASH_TABLE
:
847 case OBJECT_FIELD_HASH_TABLE
:
848 if ((le64toh(o
->object
.size
) - offsetof(HashTableObject
, items
)) % sizeof(HashItem
) != 0 ||
849 (le64toh(o
->object
.size
) - offsetof(HashTableObject
, items
)) / sizeof(HashItem
) <= 0) {
851 "Invalid %s hash table size: %"PRIu64
": %"PRIu64
,
852 o
->object
.type
== OBJECT_DATA_HASH_TABLE
? "data" : "field",
853 le64toh(o
->object
.size
),
860 case OBJECT_ENTRY_ARRAY
:
861 if ((le64toh(o
->object
.size
) - offsetof(EntryArrayObject
, items
)) % sizeof(le64_t
) != 0 ||
862 (le64toh(o
->object
.size
) - offsetof(EntryArrayObject
, items
)) / sizeof(le64_t
) <= 0) {
864 "Invalid object entry array size: %"PRIu64
": %"PRIu64
,
865 le64toh(o
->object
.size
),
870 if (!VALID64(le64toh(o
->entry_array
.next_entry_array_offset
))) {
872 "Invalid object entry array next_entry_array_offset: "OFSfmt
": %"PRIu64
,
873 le64toh(o
->entry_array
.next_entry_array_offset
),
881 if (le64toh(o
->object
.size
) != sizeof(TagObject
)) {
883 "Invalid object tag size: %"PRIu64
": %"PRIu64
,
884 le64toh(o
->object
.size
),
889 if (!VALID_EPOCH(le64toh(o
->tag
.epoch
))) {
891 "Invalid object tag epoch: %"PRIu64
": %"PRIu64
,
892 le64toh(o
->tag
.epoch
),
903 int journal_file_move_to_object(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
**ret
) {
913 /* Objects may only be located at multiple of 64 bit */
914 if (!VALID64(offset
)) {
915 log_debug("Attempt to move to object at non-64bit boundary: %" PRIu64
, offset
);
919 /* Object may not be located in the file header */
920 if (offset
< le64toh(f
->header
->header_size
)) {
921 log_debug("Attempt to move to object located in file header: %" PRIu64
, offset
);
925 r
= journal_file_move_to(f
, type
, false, offset
, sizeof(ObjectHeader
), &t
, &tsize
);
930 s
= le64toh(o
->object
.size
);
933 log_debug("Attempt to move to uninitialized object: %" PRIu64
, offset
);
936 if (s
< sizeof(ObjectHeader
)) {
937 log_debug("Attempt to move to overly short object: %" PRIu64
, offset
);
941 if (o
->object
.type
<= OBJECT_UNUSED
) {
942 log_debug("Attempt to move to object with invalid type: %" PRIu64
, offset
);
946 if (s
< minimum_header_size(o
)) {
947 log_debug("Attempt to move to truncated object: %" PRIu64
, offset
);
951 if (type
> OBJECT_UNUSED
&& o
->object
.type
!= type
) {
952 log_debug("Attempt to move to object of unexpected type: %" PRIu64
, offset
);
957 r
= journal_file_move_to(f
, type
, false, offset
, s
, &t
, NULL
);
964 r
= journal_file_check_object(f
, offset
, o
);
972 static uint64_t journal_file_entry_seqnum(JournalFile
*f
, uint64_t *seqnum
) {
978 r
= le64toh(f
->header
->tail_entry_seqnum
) + 1;
981 /* If an external seqnum counter was passed, we update
982 * both the local and the external one, and set it to
983 * the maximum of both */
991 f
->header
->tail_entry_seqnum
= htole64(r
);
993 if (f
->header
->head_entry_seqnum
== 0)
994 f
->header
->head_entry_seqnum
= htole64(r
);
999 int journal_file_append_object(JournalFile
*f
, ObjectType type
, uint64_t size
, Object
**ret
, uint64_t *offset
) {
1007 assert(type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
);
1008 assert(size
>= sizeof(ObjectHeader
));
1012 r
= journal_file_set_online(f
);
1016 p
= le64toh(f
->header
->tail_object_offset
);
1018 p
= le64toh(f
->header
->header_size
);
1020 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &tail
);
1024 p
+= ALIGN64(le64toh(tail
->object
.size
));
1027 r
= journal_file_allocate(f
, p
, size
);
1031 r
= journal_file_move_to(f
, type
, false, p
, size
, &t
, NULL
);
1038 o
->object
.type
= type
;
1039 o
->object
.size
= htole64(size
);
1041 f
->header
->tail_object_offset
= htole64(p
);
1042 f
->header
->n_objects
= htole64(le64toh(f
->header
->n_objects
) + 1);
1050 static int journal_file_setup_data_hash_table(JournalFile
*f
) {
1058 /* We estimate that we need 1 hash table entry per 768 bytes
1059 of journal file and we want to make sure we never get
1060 beyond 75% fill level. Calculate the hash table size for
1061 the maximum file size based on these metrics. */
1063 s
= (f
->metrics
.max_size
* 4 / 768 / 3) * sizeof(HashItem
);
1064 if (s
< DEFAULT_DATA_HASH_TABLE_SIZE
)
1065 s
= DEFAULT_DATA_HASH_TABLE_SIZE
;
1067 log_debug("Reserving %"PRIu64
" entries in hash table.", s
/ sizeof(HashItem
));
1069 r
= journal_file_append_object(f
,
1070 OBJECT_DATA_HASH_TABLE
,
1071 offsetof(Object
, hash_table
.items
) + s
,
1076 memzero(o
->hash_table
.items
, s
);
1078 f
->header
->data_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1079 f
->header
->data_hash_table_size
= htole64(s
);
1084 static int journal_file_setup_field_hash_table(JournalFile
*f
) {
1092 /* We use a fixed size hash table for the fields as this
1093 * number should grow very slowly only */
1095 s
= DEFAULT_FIELD_HASH_TABLE_SIZE
;
1096 r
= journal_file_append_object(f
,
1097 OBJECT_FIELD_HASH_TABLE
,
1098 offsetof(Object
, hash_table
.items
) + s
,
1103 memzero(o
->hash_table
.items
, s
);
1105 f
->header
->field_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1106 f
->header
->field_hash_table_size
= htole64(s
);
1111 int journal_file_map_data_hash_table(JournalFile
*f
) {
1119 if (f
->data_hash_table
)
1122 p
= le64toh(f
->header
->data_hash_table_offset
);
1123 s
= le64toh(f
->header
->data_hash_table_size
);
1125 r
= journal_file_move_to(f
,
1126 OBJECT_DATA_HASH_TABLE
,
1133 f
->data_hash_table
= t
;
1137 int journal_file_map_field_hash_table(JournalFile
*f
) {
1145 if (f
->field_hash_table
)
1148 p
= le64toh(f
->header
->field_hash_table_offset
);
1149 s
= le64toh(f
->header
->field_hash_table_size
);
1151 r
= journal_file_move_to(f
,
1152 OBJECT_FIELD_HASH_TABLE
,
1159 f
->field_hash_table
= t
;
1163 static int journal_file_link_field(
1174 assert(f
->field_hash_table
);
1178 if (o
->object
.type
!= OBJECT_FIELD
)
1181 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1185 /* This might alter the window we are looking at */
1186 o
->field
.next_hash_offset
= o
->field
.head_data_offset
= 0;
1189 p
= le64toh(f
->field_hash_table
[h
].tail_hash_offset
);
1191 f
->field_hash_table
[h
].head_hash_offset
= htole64(offset
);
1193 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1197 o
->field
.next_hash_offset
= htole64(offset
);
1200 f
->field_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1202 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
1203 f
->header
->n_fields
= htole64(le64toh(f
->header
->n_fields
) + 1);
1208 static int journal_file_link_data(
1219 assert(f
->data_hash_table
);
1223 if (o
->object
.type
!= OBJECT_DATA
)
1226 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1230 /* This might alter the window we are looking at */
1231 o
->data
.next_hash_offset
= o
->data
.next_field_offset
= 0;
1232 o
->data
.entry_offset
= o
->data
.entry_array_offset
= 0;
1233 o
->data
.n_entries
= 0;
1236 p
= le64toh(f
->data_hash_table
[h
].tail_hash_offset
);
1238 /* Only entry in the hash table is easy */
1239 f
->data_hash_table
[h
].head_hash_offset
= htole64(offset
);
1241 /* Move back to the previous data object, to patch in
1244 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1248 o
->data
.next_hash_offset
= htole64(offset
);
1251 f
->data_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1253 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
1254 f
->header
->n_data
= htole64(le64toh(f
->header
->n_data
) + 1);
1259 int journal_file_find_field_object_with_hash(
1261 const void *field
, uint64_t size
, uint64_t hash
,
1262 Object
**ret
, uint64_t *offset
) {
1264 uint64_t p
, osize
, h
, m
;
1269 assert(field
&& size
> 0);
1271 /* If the field hash table is empty, we can't find anything */
1272 if (le64toh(f
->header
->field_hash_table_size
) <= 0)
1275 /* Map the field hash table, if it isn't mapped yet. */
1276 r
= journal_file_map_field_hash_table(f
);
1280 osize
= offsetof(Object
, field
.payload
) + size
;
1282 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1287 p
= le64toh(f
->field_hash_table
[h
].head_hash_offset
);
1292 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1296 if (le64toh(o
->field
.hash
) == hash
&&
1297 le64toh(o
->object
.size
) == osize
&&
1298 memcmp(o
->field
.payload
, field
, size
) == 0) {
1308 p
= le64toh(o
->field
.next_hash_offset
);
1314 int journal_file_find_field_object(
1316 const void *field
, uint64_t size
,
1317 Object
**ret
, uint64_t *offset
) {
1322 assert(field
&& size
> 0);
1324 hash
= hash64(field
, size
);
1326 return journal_file_find_field_object_with_hash(f
,
1331 int journal_file_find_data_object_with_hash(
1333 const void *data
, uint64_t size
, uint64_t hash
,
1334 Object
**ret
, uint64_t *offset
) {
1336 uint64_t p
, osize
, h
, m
;
1341 assert(data
|| size
== 0);
1343 /* If there's no data hash table, then there's no entry. */
1344 if (le64toh(f
->header
->data_hash_table_size
) <= 0)
1347 /* Map the data hash table, if it isn't mapped yet. */
1348 r
= journal_file_map_data_hash_table(f
);
1352 osize
= offsetof(Object
, data
.payload
) + size
;
1354 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1359 p
= le64toh(f
->data_hash_table
[h
].head_hash_offset
);
1364 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1368 if (le64toh(o
->data
.hash
) != hash
)
1371 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
1372 #if HAVE_XZ || HAVE_LZ4
1376 l
= le64toh(o
->object
.size
);
1377 if (l
<= offsetof(Object
, data
.payload
))
1380 l
-= offsetof(Object
, data
.payload
);
1382 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
1383 o
->data
.payload
, l
, &f
->compress_buffer
, &f
->compress_buffer_size
, &rsize
, 0);
1387 if (rsize
== size
&&
1388 memcmp(f
->compress_buffer
, data
, size
) == 0) {
1399 return -EPROTONOSUPPORT
;
1401 } else if (le64toh(o
->object
.size
) == osize
&&
1402 memcmp(o
->data
.payload
, data
, size
) == 0) {
1414 p
= le64toh(o
->data
.next_hash_offset
);
1420 int journal_file_find_data_object(
1422 const void *data
, uint64_t size
,
1423 Object
**ret
, uint64_t *offset
) {
1428 assert(data
|| size
== 0);
1430 hash
= hash64(data
, size
);
1432 return journal_file_find_data_object_with_hash(f
,
1437 static int journal_file_append_field(
1439 const void *field
, uint64_t size
,
1440 Object
**ret
, uint64_t *offset
) {
1448 assert(field
&& size
> 0);
1450 hash
= hash64(field
, size
);
1452 r
= journal_file_find_field_object_with_hash(f
, field
, size
, hash
, &o
, &p
);
1466 osize
= offsetof(Object
, field
.payload
) + size
;
1467 r
= journal_file_append_object(f
, OBJECT_FIELD
, osize
, &o
, &p
);
1471 o
->field
.hash
= htole64(hash
);
1472 memcpy(o
->field
.payload
, field
, size
);
1474 r
= journal_file_link_field(f
, o
, p
, hash
);
1478 /* The linking might have altered the window, so let's
1479 * refresh our pointer */
1480 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1485 r
= journal_file_hmac_put_object(f
, OBJECT_FIELD
, o
, p
);
1499 static int journal_file_append_data(
1501 const void *data
, uint64_t size
,
1502 Object
**ret
, uint64_t *offset
) {
1507 int r
, compression
= 0;
1511 assert(data
|| size
== 0);
1513 hash
= hash64(data
, size
);
1515 r
= journal_file_find_data_object_with_hash(f
, data
, size
, hash
, &o
, &p
);
1529 osize
= offsetof(Object
, data
.payload
) + size
;
1530 r
= journal_file_append_object(f
, OBJECT_DATA
, osize
, &o
, &p
);
1534 o
->data
.hash
= htole64(hash
);
1536 #if HAVE_XZ || HAVE_LZ4
1537 if (JOURNAL_FILE_COMPRESS(f
) && size
>= f
->compress_threshold_bytes
) {
1540 compression
= compress_blob(data
, size
, o
->data
.payload
, size
- 1, &rsize
);
1542 if (compression
>= 0) {
1543 o
->object
.size
= htole64(offsetof(Object
, data
.payload
) + rsize
);
1544 o
->object
.flags
|= compression
;
1546 log_debug("Compressed data object %"PRIu64
" -> %zu using %s",
1547 size
, rsize
, object_compressed_to_string(compression
));
1549 /* Compression didn't work, we don't really care why, let's continue without compression */
1554 if (compression
== 0)
1555 memcpy_safe(o
->data
.payload
, data
, size
);
1557 r
= journal_file_link_data(f
, o
, p
, hash
);
1562 r
= journal_file_hmac_put_object(f
, OBJECT_DATA
, o
, p
);
1567 /* The linking might have altered the window, so let's
1568 * refresh our pointer */
1569 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1576 eq
= memchr(data
, '=', size
);
1577 if (eq
&& eq
> data
) {
1581 /* Create field object ... */
1582 r
= journal_file_append_field(f
, data
, (uint8_t*) eq
- (uint8_t*) data
, &fo
, &fp
);
1586 /* ... and link it in. */
1587 o
->data
.next_field_offset
= fo
->field
.head_data_offset
;
1588 fo
->field
.head_data_offset
= le64toh(p
);
1600 uint64_t journal_file_entry_n_items(Object
*o
) {
1603 if (o
->object
.type
!= OBJECT_ENTRY
)
1606 return (le64toh(o
->object
.size
) - offsetof(Object
, entry
.items
)) / sizeof(EntryItem
);
1609 uint64_t journal_file_entry_array_n_items(Object
*o
) {
1612 if (o
->object
.type
!= OBJECT_ENTRY_ARRAY
)
1615 return (le64toh(o
->object
.size
) - offsetof(Object
, entry_array
.items
)) / sizeof(uint64_t);
1618 uint64_t journal_file_hash_table_n_items(Object
*o
) {
1621 if (!IN_SET(o
->object
.type
, OBJECT_DATA_HASH_TABLE
, OBJECT_FIELD_HASH_TABLE
))
1624 return (le64toh(o
->object
.size
) - offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
);
1627 static int link_entry_into_array(JournalFile
*f
,
1632 uint64_t n
= 0, ap
= 0, q
, i
, a
, hidx
;
1641 a
= le64toh(*first
);
1642 i
= hidx
= le64toh(*idx
);
1645 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
1649 n
= journal_file_entry_array_n_items(o
);
1651 o
->entry_array
.items
[i
] = htole64(p
);
1652 *idx
= htole64(hidx
+ 1);
1658 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
1669 r
= journal_file_append_object(f
, OBJECT_ENTRY_ARRAY
,
1670 offsetof(Object
, entry_array
.items
) + n
* sizeof(uint64_t),
1676 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY_ARRAY
, o
, q
);
1681 o
->entry_array
.items
[i
] = htole64(p
);
1684 *first
= htole64(q
);
1686 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, ap
, &o
);
1690 o
->entry_array
.next_entry_array_offset
= htole64(q
);
1693 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
1694 f
->header
->n_entry_arrays
= htole64(le64toh(f
->header
->n_entry_arrays
) + 1);
1696 *idx
= htole64(hidx
+ 1);
1701 static int link_entry_into_array_plus_one(JournalFile
*f
,
1716 *extra
= htole64(p
);
1720 i
= htole64(le64toh(*idx
) - 1);
1721 r
= link_entry_into_array(f
, first
, &i
, p
);
1726 *idx
= htole64(le64toh(*idx
) + 1);
1730 static int journal_file_link_entry_item(JournalFile
*f
, Object
*o
, uint64_t offset
, uint64_t i
) {
1737 p
= le64toh(o
->entry
.items
[i
].object_offset
);
1741 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1745 return link_entry_into_array_plus_one(f
,
1746 &o
->data
.entry_offset
,
1747 &o
->data
.entry_array_offset
,
1752 static int journal_file_link_entry(JournalFile
*f
, Object
*o
, uint64_t offset
) {
1761 if (o
->object
.type
!= OBJECT_ENTRY
)
1764 __sync_synchronize();
1766 /* Link up the entry itself */
1767 r
= link_entry_into_array(f
,
1768 &f
->header
->entry_array_offset
,
1769 &f
->header
->n_entries
,
1774 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
1776 if (f
->header
->head_entry_realtime
== 0)
1777 f
->header
->head_entry_realtime
= o
->entry
.realtime
;
1779 f
->header
->tail_entry_realtime
= o
->entry
.realtime
;
1780 f
->header
->tail_entry_monotonic
= o
->entry
.monotonic
;
1782 /* Link up the items */
1783 n
= journal_file_entry_n_items(o
);
1784 for (i
= 0; i
< n
; i
++) {
1785 r
= journal_file_link_entry_item(f
, o
, offset
, i
);
1793 static int journal_file_append_entry_internal(
1795 const dual_timestamp
*ts
,
1796 const sd_id128_t
*boot_id
,
1798 const EntryItem items
[], unsigned n_items
,
1800 Object
**ret
, uint64_t *offset
) {
1808 assert(items
|| n_items
== 0);
1811 osize
= offsetof(Object
, entry
.items
) + (n_items
* sizeof(EntryItem
));
1813 r
= journal_file_append_object(f
, OBJECT_ENTRY
, osize
, &o
, &np
);
1817 o
->entry
.seqnum
= htole64(journal_file_entry_seqnum(f
, seqnum
));
1818 memcpy_safe(o
->entry
.items
, items
, n_items
* sizeof(EntryItem
));
1819 o
->entry
.realtime
= htole64(ts
->realtime
);
1820 o
->entry
.monotonic
= htole64(ts
->monotonic
);
1821 o
->entry
.xor_hash
= htole64(xor_hash
);
1822 o
->entry
.boot_id
= boot_id
? *boot_id
: f
->header
->boot_id
;
1825 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY
, o
, np
);
1830 r
= journal_file_link_entry(f
, o
, np
);
1843 void journal_file_post_change(JournalFile
*f
) {
1849 /* inotify() does not receive IN_MODIFY events from file
1850 * accesses done via mmap(). After each access we hence
1851 * trigger IN_MODIFY by truncating the journal file to its
1852 * current size which triggers IN_MODIFY. */
1854 __sync_synchronize();
1856 if (ftruncate(f
->fd
, f
->last_stat
.st_size
) < 0)
1857 log_debug_errno(errno
, "Failed to truncate file to its own size: %m");
1860 static int post_change_thunk(sd_event_source
*timer
, uint64_t usec
, void *userdata
) {
1863 journal_file_post_change(userdata
);
1868 static void schedule_post_change(JournalFile
*f
) {
1873 assert(f
->post_change_timer
);
1875 r
= sd_event_source_get_enabled(f
->post_change_timer
, NULL
);
1877 log_debug_errno(r
, "Failed to get ftruncate timer state: %m");
1883 r
= sd_event_now(sd_event_source_get_event(f
->post_change_timer
), CLOCK_MONOTONIC
, &now
);
1885 log_debug_errno(r
, "Failed to get clock's now for scheduling ftruncate: %m");
1889 r
= sd_event_source_set_time(f
->post_change_timer
, now
+ f
->post_change_timer_period
);
1891 log_debug_errno(r
, "Failed to set time for scheduling ftruncate: %m");
1895 r
= sd_event_source_set_enabled(f
->post_change_timer
, SD_EVENT_ONESHOT
);
1897 log_debug_errno(r
, "Failed to enable scheduled ftruncate: %m");
1904 /* On failure, let's simply post the change immediately. */
1905 journal_file_post_change(f
);
1908 /* Enable coalesced change posting in a timer on the provided sd_event instance */
1909 int journal_file_enable_post_change_timer(JournalFile
*f
, sd_event
*e
, usec_t t
) {
1910 _cleanup_(sd_event_source_unrefp
) sd_event_source
*timer
= NULL
;
1914 assert_return(!f
->post_change_timer
, -EINVAL
);
1918 r
= sd_event_add_time(e
, &timer
, CLOCK_MONOTONIC
, 0, 0, post_change_thunk
, f
);
1922 r
= sd_event_source_set_enabled(timer
, SD_EVENT_OFF
);
1926 f
->post_change_timer
= TAKE_PTR(timer
);
1927 f
->post_change_timer_period
= t
;
1932 static int entry_item_cmp(const EntryItem
*a
, const EntryItem
*b
) {
1933 return CMP(le64toh(a
->object_offset
), le64toh(b
->object_offset
));
1936 int journal_file_append_entry(
1938 const dual_timestamp
*ts
,
1939 const sd_id128_t
*boot_id
,
1940 const struct iovec iovec
[], unsigned n_iovec
,
1942 Object
**ret
, uint64_t *offset
) {
1947 uint64_t xor_hash
= 0;
1948 struct dual_timestamp _ts
;
1952 assert(iovec
|| n_iovec
== 0);
1955 if (!VALID_REALTIME(ts
->realtime
)) {
1956 log_debug("Invalid realtime timestamp %"PRIu64
", refusing entry.", ts
->realtime
);
1959 if (!VALID_MONOTONIC(ts
->monotonic
)) {
1960 log_debug("Invalid monotomic timestamp %"PRIu64
", refusing entry.", ts
->monotonic
);
1964 dual_timestamp_get(&_ts
);
1969 r
= journal_file_maybe_append_tag(f
, ts
->realtime
);
1974 /* alloca() can't take 0, hence let's allocate at least one */
1975 items
= newa(EntryItem
, MAX(1u, n_iovec
));
1977 for (i
= 0; i
< n_iovec
; i
++) {
1981 r
= journal_file_append_data(f
, iovec
[i
].iov_base
, iovec
[i
].iov_len
, &o
, &p
);
1985 xor_hash
^= le64toh(o
->data
.hash
);
1986 items
[i
].object_offset
= htole64(p
);
1987 items
[i
].hash
= o
->data
.hash
;
1990 /* Order by the position on disk, in order to improve seek
1991 * times for rotating media. */
1992 typesafe_qsort(items
, n_iovec
, entry_item_cmp
);
1994 r
= journal_file_append_entry_internal(f
, ts
, boot_id
, xor_hash
, items
, n_iovec
, seqnum
, ret
, offset
);
1996 /* If the memory mapping triggered a SIGBUS then we return an
1997 * IO error and ignore the error code passed down to us, since
1998 * it is very likely just an effect of a nullified replacement
2001 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
2004 if (f
->post_change_timer
)
2005 schedule_post_change(f
);
2007 journal_file_post_change(f
);
2012 typedef struct ChainCacheItem
{
2013 uint64_t first
; /* the array at the beginning of the chain */
2014 uint64_t array
; /* the cached array */
2015 uint64_t begin
; /* the first item in the cached array */
2016 uint64_t total
; /* the total number of items in all arrays before this one in the chain */
2017 uint64_t last_index
; /* the last index we looked at, to optimize locality when bisecting */
2020 static void chain_cache_put(
2027 uint64_t last_index
) {
2030 /* If the chain item to cache for this chain is the
2031 * first one it's not worth caching anything */
2035 if (ordered_hashmap_size(h
) >= CHAIN_CACHE_MAX
) {
2036 ci
= ordered_hashmap_steal_first(h
);
2039 ci
= new(ChainCacheItem
, 1);
2046 if (ordered_hashmap_put(h
, &ci
->first
, ci
) < 0) {
2051 assert(ci
->first
== first
);
2056 ci
->last_index
= last_index
;
2059 static int generic_array_get(
2063 Object
**ret
, uint64_t *offset
) {
2066 uint64_t p
= 0, a
, t
= 0;
2074 /* Try the chain cache first */
2075 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2076 if (ci
&& i
> ci
->total
) {
2085 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2089 k
= journal_file_entry_array_n_items(o
);
2091 p
= le64toh(o
->entry_array
.items
[i
]);
2097 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
2103 /* Let's cache this item for the next invocation */
2104 chain_cache_put(f
->chain_cache
, ci
, first
, a
, le64toh(o
->entry_array
.items
[0]), t
, i
);
2106 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2119 static int generic_array_get_plus_one(
2124 Object
**ret
, uint64_t *offset
) {
2133 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2146 return generic_array_get(f
, first
, i
-1, ret
, offset
);
2155 static int generic_array_bisect(
2160 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2161 direction_t direction
,
2166 uint64_t a
, p
, t
= 0, i
= 0, last_p
= 0, last_index
= (uint64_t) -1;
2167 bool subtract_one
= false;
2168 Object
*o
, *array
= NULL
;
2173 assert(test_object
);
2175 /* Start with the first array in the chain */
2178 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2179 if (ci
&& n
> ci
->total
&& ci
->begin
!= 0) {
2180 /* Ah, we have iterated this bisection array chain
2181 * previously! Let's see if we can skip ahead in the
2182 * chain, as far as the last time. But we can't jump
2183 * backwards in the chain, so let's check that
2186 r
= test_object(f
, ci
->begin
, needle
);
2190 if (r
== TEST_LEFT
) {
2191 /* OK, what we are looking for is right of the
2192 * begin of this EntryArray, so let's jump
2193 * straight to previously cached array in the
2199 last_index
= ci
->last_index
;
2204 uint64_t left
, right
, k
, lp
;
2206 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2210 k
= journal_file_entry_array_n_items(array
);
2216 lp
= p
= le64toh(array
->entry_array
.items
[i
]);
2220 r
= test_object(f
, p
, needle
);
2221 if (r
== -EBADMSG
) {
2222 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (1)");
2229 if (r
== TEST_FOUND
)
2230 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2232 if (r
== TEST_RIGHT
) {
2236 if (last_index
!= (uint64_t) -1) {
2237 assert(last_index
<= right
);
2239 /* If we cached the last index we
2240 * looked at, let's try to not to jump
2241 * too wildly around and see if we can
2242 * limit the range to look at early to
2243 * the immediate neighbors of the last
2244 * index we looked at. */
2246 if (last_index
> 0) {
2247 uint64_t x
= last_index
- 1;
2249 p
= le64toh(array
->entry_array
.items
[x
]);
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
)
2266 if (last_index
< right
) {
2267 uint64_t y
= last_index
+ 1;
2269 p
= le64toh(array
->entry_array
.items
[y
]);
2273 r
= test_object(f
, p
, needle
);
2277 if (r
== TEST_FOUND
)
2278 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2280 if (r
== TEST_RIGHT
)
2288 if (left
== right
) {
2289 if (direction
== DIRECTION_UP
)
2290 subtract_one
= true;
2296 assert(left
< right
);
2297 i
= (left
+ right
) / 2;
2299 p
= le64toh(array
->entry_array
.items
[i
]);
2303 r
= test_object(f
, p
, needle
);
2304 if (r
== -EBADMSG
) {
2305 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (2)");
2312 if (r
== TEST_FOUND
)
2313 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2315 if (r
== TEST_RIGHT
)
2323 if (direction
== DIRECTION_UP
) {
2325 subtract_one
= true;
2336 last_index
= (uint64_t) -1;
2337 a
= le64toh(array
->entry_array
.next_entry_array_offset
);
2343 if (subtract_one
&& t
== 0 && i
== 0)
2346 /* Let's cache this item for the next invocation */
2347 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
);
2349 if (subtract_one
&& i
== 0)
2351 else if (subtract_one
)
2352 p
= le64toh(array
->entry_array
.items
[i
-1]);
2354 p
= le64toh(array
->entry_array
.items
[i
]);
2356 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2367 *idx
= t
+ i
+ (subtract_one
? -1 : 0);
2372 static int generic_array_bisect_plus_one(
2378 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2379 direction_t direction
,
2385 bool step_back
= false;
2389 assert(test_object
);
2394 /* This bisects the array in object 'first', but first checks
2396 r
= test_object(f
, extra
, needle
);
2400 if (r
== TEST_FOUND
)
2401 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2403 /* if we are looking with DIRECTION_UP then we need to first
2404 see if in the actual array there is a matching entry, and
2405 return the last one of that. But if there isn't any we need
2406 to return this one. Hence remember this, and return it
2409 step_back
= direction
== DIRECTION_UP
;
2411 if (r
== TEST_RIGHT
) {
2412 if (direction
== DIRECTION_DOWN
)
2418 r
= generic_array_bisect(f
, first
, n
-1, needle
, test_object
, direction
, ret
, offset
, idx
);
2420 if (r
== 0 && step_back
)
2429 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2445 _pure_
static int test_object_offset(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2451 else if (p
< needle
)
2457 static int test_object_seqnum(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2464 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2468 if (le64toh(o
->entry
.seqnum
) == needle
)
2470 else if (le64toh(o
->entry
.seqnum
) < needle
)
2476 int journal_file_move_to_entry_by_seqnum(
2479 direction_t direction
,
2485 return generic_array_bisect(f
,
2486 le64toh(f
->header
->entry_array_offset
),
2487 le64toh(f
->header
->n_entries
),
2494 static int test_object_realtime(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2501 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2505 if (le64toh(o
->entry
.realtime
) == needle
)
2507 else if (le64toh(o
->entry
.realtime
) < needle
)
2513 int journal_file_move_to_entry_by_realtime(
2516 direction_t direction
,
2522 return generic_array_bisect(f
,
2523 le64toh(f
->header
->entry_array_offset
),
2524 le64toh(f
->header
->n_entries
),
2526 test_object_realtime
,
2531 static int test_object_monotonic(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2538 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2542 if (le64toh(o
->entry
.monotonic
) == needle
)
2544 else if (le64toh(o
->entry
.monotonic
) < needle
)
2550 static int find_data_object_by_boot_id(
2556 char t
[STRLEN("_BOOT_ID=") + 32 + 1] = "_BOOT_ID=";
2558 sd_id128_to_string(boot_id
, t
+ 9);
2559 return journal_file_find_data_object(f
, t
, sizeof(t
) - 1, o
, b
);
2562 int journal_file_move_to_entry_by_monotonic(
2566 direction_t direction
,
2575 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, NULL
);
2581 return generic_array_bisect_plus_one(f
,
2582 le64toh(o
->data
.entry_offset
),
2583 le64toh(o
->data
.entry_array_offset
),
2584 le64toh(o
->data
.n_entries
),
2586 test_object_monotonic
,
2591 void journal_file_reset_location(JournalFile
*f
) {
2592 f
->location_type
= LOCATION_HEAD
;
2593 f
->current_offset
= 0;
2594 f
->current_seqnum
= 0;
2595 f
->current_realtime
= 0;
2596 f
->current_monotonic
= 0;
2597 zero(f
->current_boot_id
);
2598 f
->current_xor_hash
= 0;
2601 void journal_file_save_location(JournalFile
*f
, Object
*o
, uint64_t offset
) {
2602 f
->location_type
= LOCATION_SEEK
;
2603 f
->current_offset
= offset
;
2604 f
->current_seqnum
= le64toh(o
->entry
.seqnum
);
2605 f
->current_realtime
= le64toh(o
->entry
.realtime
);
2606 f
->current_monotonic
= le64toh(o
->entry
.monotonic
);
2607 f
->current_boot_id
= o
->entry
.boot_id
;
2608 f
->current_xor_hash
= le64toh(o
->entry
.xor_hash
);
2611 int journal_file_compare_locations(JournalFile
*af
, JournalFile
*bf
) {
2618 assert(af
->location_type
== LOCATION_SEEK
);
2619 assert(bf
->location_type
== LOCATION_SEEK
);
2621 /* If contents and timestamps match, these entries are
2622 * identical, even if the seqnum does not match */
2623 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
) &&
2624 af
->current_monotonic
== bf
->current_monotonic
&&
2625 af
->current_realtime
== bf
->current_realtime
&&
2626 af
->current_xor_hash
== bf
->current_xor_hash
)
2629 if (sd_id128_equal(af
->header
->seqnum_id
, bf
->header
->seqnum_id
)) {
2631 /* If this is from the same seqnum source, compare
2633 r
= CMP(af
->current_seqnum
, bf
->current_seqnum
);
2637 /* Wow! This is weird, different data but the same
2638 * seqnums? Something is borked, but let's make the
2639 * best of it and compare by time. */
2642 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
)) {
2644 /* If the boot id matches, compare monotonic time */
2645 r
= CMP(af
->current_monotonic
, bf
->current_monotonic
);
2650 /* Otherwise, compare UTC time */
2651 r
= CMP(af
->current_realtime
, bf
->current_realtime
);
2655 /* Finally, compare by contents */
2656 return CMP(af
->current_xor_hash
, bf
->current_xor_hash
);
2659 static int bump_array_index(uint64_t *i
, direction_t direction
, uint64_t n
) {
2661 /* Increase or decrease the specified index, in the right direction. */
2663 if (direction
== DIRECTION_DOWN
) {
2678 static bool check_properly_ordered(uint64_t new_offset
, uint64_t old_offset
, direction_t direction
) {
2680 /* Consider it an error if any of the two offsets is uninitialized */
2681 if (old_offset
== 0 || new_offset
== 0)
2684 /* If we go down, the new offset must be larger than the old one. */
2685 return direction
== DIRECTION_DOWN
?
2686 new_offset
> old_offset
:
2687 new_offset
< old_offset
;
2690 int journal_file_next_entry(
2693 direction_t direction
,
2694 Object
**ret
, uint64_t *offset
) {
2702 n
= le64toh(f
->header
->n_entries
);
2707 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2709 r
= generic_array_bisect(f
,
2710 le64toh(f
->header
->entry_array_offset
),
2711 le64toh(f
->header
->n_entries
),
2720 r
= bump_array_index(&i
, direction
, n
);
2725 /* And jump to it */
2727 r
= generic_array_get(f
,
2728 le64toh(f
->header
->entry_array_offset
),
2736 /* OK, so this entry is borked. Most likely some entry didn't get synced to disk properly, let's see if
2737 * the next one might work for us instead. */
2738 log_debug_errno(r
, "Entry item %" PRIu64
" is bad, skipping over it.", i
);
2740 r
= bump_array_index(&i
, direction
, n
);
2745 /* Ensure our array is properly ordered. */
2746 if (p
> 0 && !check_properly_ordered(ofs
, p
, direction
)) {
2747 log_debug("%s: entry array not properly ordered at entry %" PRIu64
, f
->path
, i
);
2757 int journal_file_next_entry_for_data(
2759 Object
*o
, uint64_t p
,
2760 uint64_t data_offset
,
2761 direction_t direction
,
2762 Object
**ret
, uint64_t *offset
) {
2769 assert(p
> 0 || !o
);
2771 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2775 n
= le64toh(d
->data
.n_entries
);
2780 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2782 if (o
->object
.type
!= OBJECT_ENTRY
)
2785 r
= generic_array_bisect_plus_one(f
,
2786 le64toh(d
->data
.entry_offset
),
2787 le64toh(d
->data
.entry_array_offset
),
2788 le64toh(d
->data
.n_entries
),
2798 r
= bump_array_index(&i
, direction
, n
);
2804 r
= generic_array_get_plus_one(f
,
2805 le64toh(d
->data
.entry_offset
),
2806 le64toh(d
->data
.entry_array_offset
),
2814 log_debug_errno(r
, "Data entry item %" PRIu64
" is bad, skipping over it.", i
);
2816 r
= bump_array_index(&i
, direction
, n
);
2821 /* Ensure our array is properly ordered. */
2822 if (p
> 0 && check_properly_ordered(ofs
, p
, direction
)) {
2823 log_debug("%s data entry array not properly ordered at entry %" PRIu64
, f
->path
, i
);
2833 int journal_file_move_to_entry_by_offset_for_data(
2835 uint64_t data_offset
,
2837 direction_t direction
,
2838 Object
**ret
, uint64_t *offset
) {
2845 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2849 return generic_array_bisect_plus_one(f
,
2850 le64toh(d
->data
.entry_offset
),
2851 le64toh(d
->data
.entry_array_offset
),
2852 le64toh(d
->data
.n_entries
),
2859 int journal_file_move_to_entry_by_monotonic_for_data(
2861 uint64_t data_offset
,
2864 direction_t direction
,
2865 Object
**ret
, uint64_t *offset
) {
2873 /* First, seek by time */
2874 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &b
);
2880 r
= generic_array_bisect_plus_one(f
,
2881 le64toh(o
->data
.entry_offset
),
2882 le64toh(o
->data
.entry_array_offset
),
2883 le64toh(o
->data
.n_entries
),
2885 test_object_monotonic
,
2891 /* And now, continue seeking until we find an entry that
2892 * exists in both bisection arrays */
2898 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2902 r
= generic_array_bisect_plus_one(f
,
2903 le64toh(d
->data
.entry_offset
),
2904 le64toh(d
->data
.entry_array_offset
),
2905 le64toh(d
->data
.n_entries
),
2913 r
= journal_file_move_to_object(f
, OBJECT_DATA
, b
, &o
);
2917 r
= generic_array_bisect_plus_one(f
,
2918 le64toh(o
->data
.entry_offset
),
2919 le64toh(o
->data
.entry_array_offset
),
2920 le64toh(o
->data
.n_entries
),
2942 int journal_file_move_to_entry_by_seqnum_for_data(
2944 uint64_t data_offset
,
2946 direction_t direction
,
2947 Object
**ret
, uint64_t *offset
) {
2954 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2958 return generic_array_bisect_plus_one(f
,
2959 le64toh(d
->data
.entry_offset
),
2960 le64toh(d
->data
.entry_array_offset
),
2961 le64toh(d
->data
.n_entries
),
2968 int journal_file_move_to_entry_by_realtime_for_data(
2970 uint64_t data_offset
,
2972 direction_t direction
,
2973 Object
**ret
, uint64_t *offset
) {
2980 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2984 return generic_array_bisect_plus_one(f
,
2985 le64toh(d
->data
.entry_offset
),
2986 le64toh(d
->data
.entry_array_offset
),
2987 le64toh(d
->data
.n_entries
),
2989 test_object_realtime
,
2994 void journal_file_dump(JournalFile
*f
) {
3002 journal_file_print_header(f
);
3004 p
= le64toh(f
->header
->header_size
);
3006 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &o
);
3010 switch (o
->object
.type
) {
3013 printf("Type: OBJECT_UNUSED\n");
3017 printf("Type: OBJECT_DATA\n");
3021 printf("Type: OBJECT_FIELD\n");
3025 printf("Type: OBJECT_ENTRY seqnum=%"PRIu64
" monotonic=%"PRIu64
" realtime=%"PRIu64
"\n",
3026 le64toh(o
->entry
.seqnum
),
3027 le64toh(o
->entry
.monotonic
),
3028 le64toh(o
->entry
.realtime
));
3031 case OBJECT_FIELD_HASH_TABLE
:
3032 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
3035 case OBJECT_DATA_HASH_TABLE
:
3036 printf("Type: OBJECT_DATA_HASH_TABLE\n");
3039 case OBJECT_ENTRY_ARRAY
:
3040 printf("Type: OBJECT_ENTRY_ARRAY\n");
3044 printf("Type: OBJECT_TAG seqnum=%"PRIu64
" epoch=%"PRIu64
"\n",
3045 le64toh(o
->tag
.seqnum
),
3046 le64toh(o
->tag
.epoch
));
3050 printf("Type: unknown (%i)\n", o
->object
.type
);
3054 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
)
3055 printf("Flags: %s\n",
3056 object_compressed_to_string(o
->object
.flags
& OBJECT_COMPRESSION_MASK
));
3058 if (p
== le64toh(f
->header
->tail_object_offset
))
3061 p
= p
+ ALIGN64(le64toh(o
->object
.size
));
3066 log_error("File corrupt");
3069 static const char* format_timestamp_safe(char *buf
, size_t l
, usec_t t
) {
3072 x
= format_timestamp(buf
, l
, t
);
3078 void journal_file_print_header(JournalFile
*f
) {
3079 char a
[33], b
[33], c
[33], d
[33];
3080 char x
[FORMAT_TIMESTAMP_MAX
], y
[FORMAT_TIMESTAMP_MAX
], z
[FORMAT_TIMESTAMP_MAX
];
3082 char bytes
[FORMAT_BYTES_MAX
];
3087 printf("File Path: %s\n"
3091 "Sequential Number ID: %s\n"
3093 "Compatible Flags:%s%s\n"
3094 "Incompatible Flags:%s%s%s\n"
3095 "Header size: %"PRIu64
"\n"
3096 "Arena size: %"PRIu64
"\n"
3097 "Data Hash Table Size: %"PRIu64
"\n"
3098 "Field Hash Table Size: %"PRIu64
"\n"
3099 "Rotate Suggested: %s\n"
3100 "Head Sequential Number: %"PRIu64
" (%"PRIx64
")\n"
3101 "Tail Sequential Number: %"PRIu64
" (%"PRIx64
")\n"
3102 "Head Realtime Timestamp: %s (%"PRIx64
")\n"
3103 "Tail Realtime Timestamp: %s (%"PRIx64
")\n"
3104 "Tail Monotonic Timestamp: %s (%"PRIx64
")\n"
3105 "Objects: %"PRIu64
"\n"
3106 "Entry Objects: %"PRIu64
"\n",
3108 sd_id128_to_string(f
->header
->file_id
, a
),
3109 sd_id128_to_string(f
->header
->machine_id
, b
),
3110 sd_id128_to_string(f
->header
->boot_id
, c
),
3111 sd_id128_to_string(f
->header
->seqnum_id
, d
),
3112 f
->header
->state
== STATE_OFFLINE
? "OFFLINE" :
3113 f
->header
->state
== STATE_ONLINE
? "ONLINE" :
3114 f
->header
->state
== STATE_ARCHIVED
? "ARCHIVED" : "UNKNOWN",
3115 JOURNAL_HEADER_SEALED(f
->header
) ? " SEALED" : "",
3116 (le32toh(f
->header
->compatible_flags
) & ~HEADER_COMPATIBLE_ANY
) ? " ???" : "",
3117 JOURNAL_HEADER_COMPRESSED_XZ(f
->header
) ? " COMPRESSED-XZ" : "",
3118 JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
) ? " COMPRESSED-LZ4" : "",
3119 (le32toh(f
->header
->incompatible_flags
) & ~HEADER_INCOMPATIBLE_ANY
) ? " ???" : "",
3120 le64toh(f
->header
->header_size
),
3121 le64toh(f
->header
->arena_size
),
3122 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3123 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
),
3124 yes_no(journal_file_rotate_suggested(f
, 0)),
3125 le64toh(f
->header
->head_entry_seqnum
), le64toh(f
->header
->head_entry_seqnum
),
3126 le64toh(f
->header
->tail_entry_seqnum
), le64toh(f
->header
->tail_entry_seqnum
),
3127 format_timestamp_safe(x
, sizeof(x
), le64toh(f
->header
->head_entry_realtime
)), le64toh(f
->header
->head_entry_realtime
),
3128 format_timestamp_safe(y
, sizeof(y
), le64toh(f
->header
->tail_entry_realtime
)), le64toh(f
->header
->tail_entry_realtime
),
3129 format_timespan(z
, sizeof(z
), le64toh(f
->header
->tail_entry_monotonic
), USEC_PER_MSEC
), le64toh(f
->header
->tail_entry_monotonic
),
3130 le64toh(f
->header
->n_objects
),
3131 le64toh(f
->header
->n_entries
));
3133 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3134 printf("Data Objects: %"PRIu64
"\n"
3135 "Data Hash Table Fill: %.1f%%\n",
3136 le64toh(f
->header
->n_data
),
3137 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))));
3139 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3140 printf("Field Objects: %"PRIu64
"\n"
3141 "Field Hash Table Fill: %.1f%%\n",
3142 le64toh(f
->header
->n_fields
),
3143 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))));
3145 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
))
3146 printf("Tag Objects: %"PRIu64
"\n",
3147 le64toh(f
->header
->n_tags
));
3148 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
3149 printf("Entry Array Objects: %"PRIu64
"\n",
3150 le64toh(f
->header
->n_entry_arrays
));
3152 if (fstat(f
->fd
, &st
) >= 0)
3153 printf("Disk usage: %s\n", format_bytes(bytes
, sizeof(bytes
), (uint64_t) st
.st_blocks
* 512ULL));
3156 static int journal_file_warn_btrfs(JournalFile
*f
) {
3162 /* Before we write anything, check if the COW logic is turned
3163 * off on btrfs. Given our write pattern that is quite
3164 * unfriendly to COW file systems this should greatly improve
3165 * performance on COW file systems, such as btrfs, at the
3166 * expense of data integrity features (which shouldn't be too
3167 * bad, given that we do our own checksumming). */
3169 r
= btrfs_is_filesystem(f
->fd
);
3171 return log_warning_errno(r
, "Failed to determine if journal is on btrfs: %m");
3175 r
= read_attr_fd(f
->fd
, &attrs
);
3177 return log_warning_errno(r
, "Failed to read file attributes: %m");
3179 if (attrs
& FS_NOCOW_FL
) {
3180 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3184 log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3185 "This is likely to slow down journal access substantially, please consider turning "
3186 "off the copy-on-write file attribute on the journal directory, using chattr +C.", f
->path
);
3191 int journal_file_open(
3197 uint64_t compress_threshold_bytes
,
3199 JournalMetrics
*metrics
,
3200 MMapCache
*mmap_cache
,
3201 Set
*deferred_closes
,
3202 JournalFile
*template,
3203 JournalFile
**ret
) {
3205 bool newly_created
= false;
3209 char bytes
[FORMAT_BYTES_MAX
];
3212 assert(fd
>= 0 || fname
);
3214 if (!IN_SET((flags
& O_ACCMODE
), O_RDONLY
, O_RDWR
))
3217 if (fname
&& (flags
& O_CREAT
) && !endswith(fname
, ".journal"))
3220 f
= new(JournalFile
, 1);
3224 *f
= (JournalFile
) {
3229 .prot
= prot_from_flags(flags
),
3230 .writable
= (flags
& O_ACCMODE
) != O_RDONLY
,
3233 .compress_lz4
= compress
,
3235 .compress_xz
= compress
,
3237 .compress_threshold_bytes
= compress_threshold_bytes
== (uint64_t) -1 ?
3238 DEFAULT_COMPRESS_THRESHOLD
:
3239 MAX(MIN_COMPRESS_THRESHOLD
, compress_threshold_bytes
),
3245 log_debug("Journal effective settings seal=%s compress=%s compress_threshold_bytes=%s",
3246 yes_no(f
->seal
), yes_no(JOURNAL_FILE_COMPRESS(f
)),
3247 format_bytes(bytes
, sizeof(bytes
), f
->compress_threshold_bytes
));
3250 f
->mmap
= mmap_cache_ref(mmap_cache
);
3252 f
->mmap
= mmap_cache_new();
3260 f
->path
= strdup(fname
);
3268 /* If we don't know the path, fill in something explanatory and vaguely useful */
3269 if (asprintf(&f
->path
, "/proc/self/%i", fd
) < 0) {
3275 f
->chain_cache
= ordered_hashmap_new(&uint64_hash_ops
);
3276 if (!f
->chain_cache
) {
3282 /* We pass O_NONBLOCK here, so that in case somebody pointed us to some character device node or FIFO
3283 * or so, we likely fail quickly than block for long. For regular files O_NONBLOCK has no effect, hence
3284 * it doesn't hurt in that case. */
3286 f
->fd
= open(f
->path
, f
->flags
|O_CLOEXEC
|O_NONBLOCK
, f
->mode
);
3292 /* fds we opened here by us should also be closed by us. */
3295 r
= fd_nonblock(f
->fd
, false);
3300 f
->cache_fd
= mmap_cache_add_fd(f
->mmap
, f
->fd
);
3306 r
= journal_file_fstat(f
);
3310 if (f
->last_stat
.st_size
== 0 && f
->writable
) {
3312 (void) journal_file_warn_btrfs(f
);
3314 /* Let's attach the creation time to the journal file, so that the vacuuming code knows the age of this
3315 * file even if the file might end up corrupted one day... Ideally we'd just use the creation time many
3316 * file systems maintain for each file, but the API to query this is very new, hence let's emulate this
3317 * via extended attributes. If extended attributes are not supported we'll just skip this, and rely
3318 * solely on mtime/atime/ctime of the file. */
3319 (void) fd_setcrtime(f
->fd
, 0);
3322 /* Try to load the FSPRG state, and if we can't, then
3323 * just don't do sealing */
3325 r
= journal_file_fss_load(f
);
3331 r
= journal_file_init_header(f
, template);
3335 r
= journal_file_fstat(f
);
3339 newly_created
= true;
3342 if (f
->last_stat
.st_size
< (off_t
) HEADER_SIZE_MIN
) {
3347 r
= mmap_cache_get(f
->mmap
, f
->cache_fd
, f
->prot
, CONTEXT_HEADER
, true, 0, PAGE_ALIGN(sizeof(Header
)), &f
->last_stat
, &h
, NULL
);
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
];
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");
3738 if (m
->max_use
== (uint64_t) -1) {
3741 m
->max_use
= PAGE_ALIGN(fs_size
/ 10); /* 10% of file system size */
3743 if (m
->max_use
> DEFAULT_MAX_USE_UPPER
)
3744 m
->max_use
= DEFAULT_MAX_USE_UPPER
;
3746 if (m
->max_use
< DEFAULT_MAX_USE_LOWER
)
3747 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3749 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3751 m
->max_use
= PAGE_ALIGN(m
->max_use
);
3753 if (m
->max_use
!= 0 && m
->max_use
< JOURNAL_FILE_SIZE_MIN
*2)
3754 m
->max_use
= JOURNAL_FILE_SIZE_MIN
*2;
3757 if (m
->min_use
== (uint64_t) -1)
3758 m
->min_use
= DEFAULT_MIN_USE
;
3760 if (m
->min_use
> m
->max_use
)
3761 m
->min_use
= m
->max_use
;
3763 if (m
->max_size
== (uint64_t) -1) {
3764 m
->max_size
= PAGE_ALIGN(m
->max_use
/ 8); /* 8 chunks */
3766 if (m
->max_size
> DEFAULT_MAX_SIZE_UPPER
)
3767 m
->max_size
= DEFAULT_MAX_SIZE_UPPER
;
3769 m
->max_size
= PAGE_ALIGN(m
->max_size
);
3771 if (m
->max_size
!= 0) {
3772 if (m
->max_size
< JOURNAL_FILE_SIZE_MIN
)
3773 m
->max_size
= JOURNAL_FILE_SIZE_MIN
;
3775 if (m
->max_use
!= 0 && m
->max_size
*2 > m
->max_use
)
3776 m
->max_use
= m
->max_size
*2;
3779 if (m
->min_size
== (uint64_t) -1)
3780 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3782 m
->min_size
= PAGE_ALIGN(m
->min_size
);
3784 if (m
->min_size
< JOURNAL_FILE_SIZE_MIN
)
3785 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3787 if (m
->max_size
!= 0 && m
->min_size
> m
->max_size
)
3788 m
->max_size
= m
->min_size
;
3791 if (m
->keep_free
== (uint64_t) -1) {
3794 m
->keep_free
= PAGE_ALIGN(fs_size
* 3 / 20); /* 15% of file system size */
3796 if (m
->keep_free
> DEFAULT_KEEP_FREE_UPPER
)
3797 m
->keep_free
= DEFAULT_KEEP_FREE_UPPER
;
3800 m
->keep_free
= DEFAULT_KEEP_FREE
;
3803 if (m
->n_max_files
== (uint64_t) -1)
3804 m
->n_max_files
= DEFAULT_N_MAX_FILES
;
3806 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64
,
3807 format_bytes(a
, sizeof(a
), m
->min_use
),
3808 format_bytes(b
, sizeof(b
), m
->max_use
),
3809 format_bytes(c
, sizeof(c
), m
->max_size
),
3810 format_bytes(d
, sizeof(d
), m
->min_size
),
3811 format_bytes(e
, sizeof(e
), m
->keep_free
),
3815 int journal_file_get_cutoff_realtime_usec(JournalFile
*f
, usec_t
*from
, usec_t
*to
) {
3821 if (f
->header
->head_entry_realtime
== 0)
3824 *from
= le64toh(f
->header
->head_entry_realtime
);
3828 if (f
->header
->tail_entry_realtime
== 0)
3831 *to
= le64toh(f
->header
->tail_entry_realtime
);
3837 int journal_file_get_cutoff_monotonic_usec(JournalFile
*f
, sd_id128_t boot_id
, usec_t
*from
, usec_t
*to
) {
3845 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &p
);
3849 if (le64toh(o
->data
.n_entries
) <= 0)
3853 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, le64toh(o
->data
.entry_offset
), &o
);
3857 *from
= le64toh(o
->entry
.monotonic
);
3861 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
3865 r
= generic_array_get_plus_one(f
,
3866 le64toh(o
->data
.entry_offset
),
3867 le64toh(o
->data
.entry_array_offset
),
3868 le64toh(o
->data
.n_entries
)-1,
3873 *to
= le64toh(o
->entry
.monotonic
);
3879 bool journal_file_rotate_suggested(JournalFile
*f
, usec_t max_file_usec
) {
3883 /* If we gained new header fields we gained new features,
3884 * hence suggest a rotation */
3885 if (le64toh(f
->header
->header_size
) < sizeof(Header
)) {
3886 log_debug("%s uses an outdated header, suggesting rotation.", f
->path
);
3890 /* Let's check if the hash tables grew over a certain fill
3891 * level (75%, borrowing this value from Java's hash table
3892 * implementation), and if so suggest a rotation. To calculate
3893 * the fill level we need the n_data field, which only exists
3894 * in newer versions. */
3896 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3897 if (le64toh(f
->header
->n_data
) * 4ULL > (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3898 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.",
3900 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))),
3901 le64toh(f
->header
->n_data
),
3902 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3903 (unsigned long long) f
->last_stat
.st_size
,
3904 f
->last_stat
.st_size
/ le64toh(f
->header
->n_data
));
3908 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3909 if (le64toh(f
->header
->n_fields
) * 4ULL > (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3910 log_debug("Field hash table of %s has a fill level at %.1f (%"PRIu64
" of %"PRIu64
" items), suggesting rotation.",
3912 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))),
3913 le64toh(f
->header
->n_fields
),
3914 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
));
3918 /* Are the data objects properly indexed by field objects? */
3919 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
3920 JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
3921 le64toh(f
->header
->n_data
) > 0 &&
3922 le64toh(f
->header
->n_fields
) == 0)
3925 if (max_file_usec
> 0) {
3928 h
= le64toh(f
->header
->head_entry_realtime
);
3929 t
= now(CLOCK_REALTIME
);
3931 if (h
> 0 && t
> h
+ max_file_usec
)