1 /* SPDX-License-Identifier: LGPL-2.1+ */
3 This file is part of systemd.
5 Copyright 2011 Lennart Poettering
14 #include <sys/statvfs.h>
18 #include "alloc-util.h"
19 #include "btrfs-util.h"
20 #include "chattr-util.h"
24 #include "journal-authenticate.h"
25 #include "journal-def.h"
26 #include "journal-file.h"
28 #include "parse-util.h"
29 #include "path-util.h"
30 #include "random-util.h"
33 #include "stat-util.h"
34 #include "string-util.h"
36 #include "xattr-util.h"
38 #define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
39 #define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
41 #define DEFAULT_COMPRESS_THRESHOLD (512ULL)
42 #define MIN_COMPRESS_THRESHOLD (8ULL)
44 /* This is the minimum journal file size */
45 #define JOURNAL_FILE_SIZE_MIN (512ULL*1024ULL) /* 512 KiB */
47 /* These are the lower and upper bounds if we deduce the max_use value
48 * from the file system size */
49 #define DEFAULT_MAX_USE_LOWER (1ULL*1024ULL*1024ULL) /* 1 MiB */
50 #define DEFAULT_MAX_USE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
52 /* This is the default minimal use limit, how much we'll use even if keep_free suggests otherwise. */
53 #define DEFAULT_MIN_USE (1ULL*1024ULL*1024ULL) /* 1 MiB */
55 /* This is the upper bound if we deduce max_size from max_use */
56 #define DEFAULT_MAX_SIZE_UPPER (128ULL*1024ULL*1024ULL) /* 128 MiB */
58 /* This is the upper bound if we deduce the keep_free value from the
60 #define DEFAULT_KEEP_FREE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
62 /* This is the keep_free value when we can't determine the system
64 #define DEFAULT_KEEP_FREE (1024ULL*1024ULL) /* 1 MB */
66 /* This is the default maximum number of journal files to keep around. */
67 #define DEFAULT_N_MAX_FILES (100)
69 /* n_data was the first entry we added after the initial file format design */
70 #define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
72 /* How many entries to keep in the entry array chain cache at max */
73 #define CHAIN_CACHE_MAX 20
75 /* How much to increase the journal file size at once each time we allocate something new. */
76 #define FILE_SIZE_INCREASE (8ULL*1024ULL*1024ULL) /* 8MB */
78 /* Reread fstat() of the file for detecting deletions at least this often */
79 #define LAST_STAT_REFRESH_USEC (5*USEC_PER_SEC)
81 /* The mmap context to use for the header we pick as one above the last defined typed */
82 #define CONTEXT_HEADER _OBJECT_TYPE_MAX
85 # pragma GCC diagnostic ignored "-Waddress-of-packed-member"
88 /* This may be called from a separate thread to prevent blocking the caller for the duration of fsync().
89 * As a result we use atomic operations on f->offline_state for inter-thread communications with
90 * journal_file_set_offline() and journal_file_set_online(). */
91 static void journal_file_set_offline_internal(JournalFile
*f
) {
97 switch (f
->offline_state
) {
99 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_CANCEL
, OFFLINE_DONE
))
103 case OFFLINE_AGAIN_FROM_SYNCING
:
104 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_SYNCING
, OFFLINE_SYNCING
))
108 case OFFLINE_AGAIN_FROM_OFFLINING
:
109 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_OFFLINING
, OFFLINE_SYNCING
))
113 case OFFLINE_SYNCING
:
116 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_OFFLINING
))
119 f
->header
->state
= f
->archive
? STATE_ARCHIVED
: STATE_OFFLINE
;
123 case OFFLINE_OFFLINING
:
124 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_OFFLINING
, OFFLINE_DONE
))
131 log_debug("OFFLINE_JOINED unexpected offline state for journal_file_set_offline_internal()");
137 static void * journal_file_set_offline_thread(void *arg
) {
138 JournalFile
*f
= arg
;
140 (void) pthread_setname_np(pthread_self(), "journal-offline");
142 journal_file_set_offline_internal(f
);
147 static int journal_file_set_offline_thread_join(JournalFile
*f
) {
152 if (f
->offline_state
== OFFLINE_JOINED
)
155 r
= pthread_join(f
->offline_thread
, NULL
);
159 f
->offline_state
= OFFLINE_JOINED
;
161 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
167 /* Trigger a restart if the offline thread is mid-flight in a restartable state. */
168 static bool journal_file_set_offline_try_restart(JournalFile
*f
) {
170 switch (f
->offline_state
) {
171 case OFFLINE_AGAIN_FROM_SYNCING
:
172 case OFFLINE_AGAIN_FROM_OFFLINING
:
176 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_CANCEL
, OFFLINE_AGAIN_FROM_SYNCING
))
180 case OFFLINE_SYNCING
:
181 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_AGAIN_FROM_SYNCING
))
185 case OFFLINE_OFFLINING
:
186 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_OFFLINING
, OFFLINE_AGAIN_FROM_OFFLINING
))
196 /* Sets a journal offline.
198 * If wait is false then an offline is dispatched in a separate thread for a
199 * subsequent journal_file_set_offline() or journal_file_set_online() of the
200 * same journal to synchronize with.
202 * If wait is true, then either an existing offline thread will be restarted
203 * and joined, or if none exists the offline is simply performed in this
204 * context without involving another thread.
206 int journal_file_set_offline(JournalFile
*f
, bool wait
) {
215 if (!(f
->fd
>= 0 && f
->header
))
218 /* An offlining journal is implicitly online and may modify f->header->state,
219 * we must also join any potentially lingering offline thread when not online. */
220 if (!journal_file_is_offlining(f
) && f
->header
->state
!= STATE_ONLINE
)
221 return journal_file_set_offline_thread_join(f
);
223 /* Restart an in-flight offline thread and wait if needed, or join a lingering done one. */
224 restarted
= journal_file_set_offline_try_restart(f
);
225 if ((restarted
&& wait
) || !restarted
) {
226 r
= journal_file_set_offline_thread_join(f
);
234 /* Initiate a new offline. */
235 f
->offline_state
= OFFLINE_SYNCING
;
237 if (wait
) /* Without using a thread if waiting. */
238 journal_file_set_offline_internal(f
);
240 sigset_t ss
, saved_ss
;
243 if (sigfillset(&ss
) < 0)
246 r
= pthread_sigmask(SIG_BLOCK
, &ss
, &saved_ss
);
250 r
= pthread_create(&f
->offline_thread
, NULL
, journal_file_set_offline_thread
, f
);
252 k
= pthread_sigmask(SIG_SETMASK
, &saved_ss
, NULL
);
254 f
->offline_state
= OFFLINE_JOINED
;
264 static int journal_file_set_online(JournalFile
*f
) {
272 if (!(f
->fd
>= 0 && f
->header
))
276 switch (f
->offline_state
) {
278 /* No offline thread, no need to wait. */
282 case OFFLINE_SYNCING
:
283 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_CANCEL
))
285 /* Canceled syncing prior to offlining, no need to wait. */
289 case OFFLINE_AGAIN_FROM_SYNCING
:
290 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_SYNCING
, OFFLINE_CANCEL
))
292 /* Canceled restart from syncing, no need to wait. */
296 case OFFLINE_AGAIN_FROM_OFFLINING
:
297 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_OFFLINING
, OFFLINE_CANCEL
))
299 /* Canceled restart from offlining, must wait for offlining to complete however. */
304 r
= journal_file_set_offline_thread_join(f
);
314 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
317 switch (f
->header
->state
) {
322 f
->header
->state
= STATE_ONLINE
;
331 bool journal_file_is_offlining(JournalFile
*f
) {
334 __sync_synchronize();
336 if (IN_SET(f
->offline_state
, OFFLINE_DONE
, OFFLINE_JOINED
))
342 JournalFile
* journal_file_close(JournalFile
*f
) {
346 /* Write the final tag */
347 if (f
->seal
&& f
->writable
) {
350 r
= journal_file_append_tag(f
);
352 log_error_errno(r
, "Failed to append tag when closing journal: %m");
356 if (f
->post_change_timer
) {
359 if (sd_event_source_get_enabled(f
->post_change_timer
, &enabled
) >= 0)
360 if (enabled
== SD_EVENT_ONESHOT
)
361 journal_file_post_change(f
);
363 (void) sd_event_source_set_enabled(f
->post_change_timer
, SD_EVENT_OFF
);
364 sd_event_source_unref(f
->post_change_timer
);
367 journal_file_set_offline(f
, true);
369 if (f
->mmap
&& f
->cache_fd
)
370 mmap_cache_free_fd(f
->mmap
, f
->cache_fd
);
372 if (f
->fd
>= 0 && f
->defrag_on_close
) {
374 /* Be friendly to btrfs: turn COW back on again now,
375 * and defragment the file. We won't write to the file
376 * ever again, hence remove all fragmentation, and
377 * reenable all the good bits COW usually provides
378 * (such as data checksumming). */
380 (void) chattr_fd(f
->fd
, 0, FS_NOCOW_FL
);
381 (void) btrfs_defrag_fd(f
->fd
);
388 mmap_cache_unref(f
->mmap
);
390 ordered_hashmap_free_free(f
->chain_cache
);
392 #if HAVE_XZ || HAVE_LZ4
393 free(f
->compress_buffer
);
398 munmap(f
->fss_file
, PAGE_ALIGN(f
->fss_file_size
));
400 free(f
->fsprg_state
);
405 gcry_md_close(f
->hmac
);
411 static int journal_file_init_header(JournalFile
*f
, JournalFile
*template) {
418 memcpy(h
.signature
, HEADER_SIGNATURE
, 8);
419 h
.header_size
= htole64(ALIGN64(sizeof(h
)));
421 h
.incompatible_flags
|= htole32(
422 f
->compress_xz
* HEADER_INCOMPATIBLE_COMPRESSED_XZ
|
423 f
->compress_lz4
* HEADER_INCOMPATIBLE_COMPRESSED_LZ4
);
425 h
.compatible_flags
= htole32(
426 f
->seal
* HEADER_COMPATIBLE_SEALED
);
428 r
= sd_id128_randomize(&h
.file_id
);
433 h
.seqnum_id
= template->header
->seqnum_id
;
434 h
.tail_entry_seqnum
= template->header
->tail_entry_seqnum
;
436 h
.seqnum_id
= h
.file_id
;
438 k
= pwrite(f
->fd
, &h
, sizeof(h
), 0);
448 static int journal_file_refresh_header(JournalFile
*f
) {
455 r
= sd_id128_get_machine(&f
->header
->machine_id
);
456 if (IN_SET(r
, -ENOENT
, -ENOMEDIUM
))
457 /* We don't have a machine-id, let's continue without */
458 zero(f
->header
->machine_id
);
462 r
= sd_id128_get_boot(&boot_id
);
466 f
->header
->boot_id
= boot_id
;
468 r
= journal_file_set_online(f
);
470 /* Sync the online state to disk */
473 /* We likely just created a new file, also sync the directory this file is located in. */
474 (void) fsync_directory_of_file(f
->fd
);
479 static bool warn_wrong_flags(const JournalFile
*f
, bool compatible
) {
480 const uint32_t any
= compatible
? HEADER_COMPATIBLE_ANY
: HEADER_INCOMPATIBLE_ANY
,
481 supported
= compatible
? HEADER_COMPATIBLE_SUPPORTED
: HEADER_INCOMPATIBLE_SUPPORTED
;
482 const char *type
= compatible
? "compatible" : "incompatible";
485 flags
= le32toh(compatible
? f
->header
->compatible_flags
: f
->header
->incompatible_flags
);
487 if (flags
& ~supported
) {
489 log_debug("Journal file %s has unknown %s flags 0x%"PRIx32
,
490 f
->path
, type
, flags
& ~any
);
491 flags
= (flags
& any
) & ~supported
;
495 _cleanup_free_
char *t
= NULL
;
497 if (compatible
&& (flags
& HEADER_COMPATIBLE_SEALED
))
498 strv
[n
++] = "sealed";
499 if (!compatible
&& (flags
& HEADER_INCOMPATIBLE_COMPRESSED_XZ
))
500 strv
[n
++] = "xz-compressed";
501 if (!compatible
&& (flags
& HEADER_INCOMPATIBLE_COMPRESSED_LZ4
))
502 strv
[n
++] = "lz4-compressed";
504 assert(n
< ELEMENTSOF(strv
));
506 t
= strv_join((char**) strv
, ", ");
507 log_debug("Journal file %s uses %s %s %s disabled at compilation time.",
508 f
->path
, type
, n
> 1 ? "flags" : "flag", strnull(t
));
516 static int journal_file_verify_header(JournalFile
*f
) {
517 uint64_t arena_size
, header_size
;
522 if (memcmp(f
->header
->signature
, HEADER_SIGNATURE
, 8))
525 /* In both read and write mode we refuse to open files with incompatible
526 * flags we don't know. */
527 if (warn_wrong_flags(f
, false))
528 return -EPROTONOSUPPORT
;
530 /* When open for writing we refuse to open files with compatible flags, too. */
531 if (f
->writable
&& warn_wrong_flags(f
, true))
532 return -EPROTONOSUPPORT
;
534 if (f
->header
->state
>= _STATE_MAX
)
537 header_size
= le64toh(f
->header
->header_size
);
539 /* The first addition was n_data, so check that we are at least this large */
540 if (header_size
< HEADER_SIZE_MIN
)
543 if (JOURNAL_HEADER_SEALED(f
->header
) && !JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
546 arena_size
= le64toh(f
->header
->arena_size
);
548 if (UINT64_MAX
- header_size
< arena_size
|| header_size
+ arena_size
> (uint64_t) f
->last_stat
.st_size
)
551 if (le64toh(f
->header
->tail_object_offset
) > header_size
+ arena_size
)
554 if (!VALID64(le64toh(f
->header
->data_hash_table_offset
)) ||
555 !VALID64(le64toh(f
->header
->field_hash_table_offset
)) ||
556 !VALID64(le64toh(f
->header
->tail_object_offset
)) ||
557 !VALID64(le64toh(f
->header
->entry_array_offset
)))
561 sd_id128_t machine_id
;
565 r
= sd_id128_get_machine(&machine_id
);
569 if (!sd_id128_equal(machine_id
, f
->header
->machine_id
))
572 state
= f
->header
->state
;
574 if (state
== STATE_ARCHIVED
)
575 return -ESHUTDOWN
; /* Already archived */
576 else if (state
== STATE_ONLINE
) {
577 log_debug("Journal file %s is already online. Assuming unclean closing.", f
->path
);
579 } else if (state
!= STATE_OFFLINE
) {
580 log_debug("Journal file %s has unknown state %i.", f
->path
, state
);
584 if (f
->header
->field_hash_table_size
== 0 || f
->header
->data_hash_table_size
== 0)
587 /* Don't permit appending to files from the future. Because otherwise the realtime timestamps wouldn't
588 * be strictly ordered in the entries in the file anymore, and we can't have that since it breaks
590 if (le64toh(f
->header
->tail_entry_realtime
) > now(CLOCK_REALTIME
)) {
591 log_debug("Journal file %s is from the future, refusing to append new data to it that'd be older.", f
->path
);
596 f
->compress_xz
= JOURNAL_HEADER_COMPRESSED_XZ(f
->header
);
597 f
->compress_lz4
= JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
);
599 f
->seal
= JOURNAL_HEADER_SEALED(f
->header
);
604 static int journal_file_fstat(JournalFile
*f
) {
610 if (fstat(f
->fd
, &f
->last_stat
) < 0)
613 f
->last_stat_usec
= now(CLOCK_MONOTONIC
);
615 /* Refuse dealing with with files that aren't regular */
616 r
= stat_verify_regular(&f
->last_stat
);
620 /* Refuse appending to files that are already deleted */
621 if (f
->last_stat
.st_nlink
<= 0)
627 static int journal_file_allocate(JournalFile
*f
, uint64_t offset
, uint64_t size
) {
628 uint64_t old_size
, new_size
;
634 /* We assume that this file is not sparse, and we know that
635 * for sure, since we always call posix_fallocate()
638 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
642 le64toh(f
->header
->header_size
) +
643 le64toh(f
->header
->arena_size
);
645 new_size
= PAGE_ALIGN(offset
+ size
);
646 if (new_size
< le64toh(f
->header
->header_size
))
647 new_size
= le64toh(f
->header
->header_size
);
649 if (new_size
<= old_size
) {
651 /* We already pre-allocated enough space, but before
652 * we write to it, let's check with fstat() if the
653 * file got deleted, in order make sure we don't throw
654 * away the data immediately. Don't check fstat() for
655 * all writes though, but only once ever 10s. */
657 if (f
->last_stat_usec
+ LAST_STAT_REFRESH_USEC
> now(CLOCK_MONOTONIC
))
660 return journal_file_fstat(f
);
663 /* Allocate more space. */
665 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
668 if (new_size
> f
->metrics
.min_size
&& f
->metrics
.keep_free
> 0) {
671 if (fstatvfs(f
->fd
, &svfs
) >= 0) {
674 available
= LESS_BY((uint64_t) svfs
.f_bfree
* (uint64_t) svfs
.f_bsize
, f
->metrics
.keep_free
);
676 if (new_size
- old_size
> available
)
681 /* Increase by larger blocks at once */
682 new_size
= DIV_ROUND_UP(new_size
, FILE_SIZE_INCREASE
) * FILE_SIZE_INCREASE
;
683 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
684 new_size
= f
->metrics
.max_size
;
686 /* Note that the glibc fallocate() fallback is very
687 inefficient, hence we try to minimize the allocation area
689 r
= posix_fallocate(f
->fd
, old_size
, new_size
- old_size
);
693 f
->header
->arena_size
= htole64(new_size
- le64toh(f
->header
->header_size
));
695 return journal_file_fstat(f
);
698 static unsigned type_to_context(ObjectType type
) {
699 /* One context for each type, plus one catch-all for the rest */
700 assert_cc(_OBJECT_TYPE_MAX
<= MMAP_CACHE_MAX_CONTEXTS
);
701 assert_cc(CONTEXT_HEADER
< MMAP_CACHE_MAX_CONTEXTS
);
702 return type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
? type
: 0;
705 static int journal_file_move_to(JournalFile
*f
, ObjectType type
, bool keep_always
, uint64_t offset
, uint64_t size
, void **ret
, size_t *ret_size
) {
714 /* Avoid SIGBUS on invalid accesses */
715 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
) {
716 /* Hmm, out of range? Let's refresh the fstat() data
717 * first, before we trust that check. */
719 r
= journal_file_fstat(f
);
723 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
)
724 return -EADDRNOTAVAIL
;
727 return mmap_cache_get(f
->mmap
, f
->cache_fd
, f
->prot
, type_to_context(type
), keep_always
, offset
, size
, &f
->last_stat
, ret
, ret_size
);
730 static uint64_t minimum_header_size(Object
*o
) {
732 static const uint64_t table
[] = {
733 [OBJECT_DATA
] = sizeof(DataObject
),
734 [OBJECT_FIELD
] = sizeof(FieldObject
),
735 [OBJECT_ENTRY
] = sizeof(EntryObject
),
736 [OBJECT_DATA_HASH_TABLE
] = sizeof(HashTableObject
),
737 [OBJECT_FIELD_HASH_TABLE
] = sizeof(HashTableObject
),
738 [OBJECT_ENTRY_ARRAY
] = sizeof(EntryArrayObject
),
739 [OBJECT_TAG
] = sizeof(TagObject
),
742 if (o
->object
.type
>= ELEMENTSOF(table
) || table
[o
->object
.type
] <= 0)
743 return sizeof(ObjectHeader
);
745 return table
[o
->object
.type
];
748 /* Lightweight object checks. We want this to be fast, so that we won't
749 * slowdown every journal_file_move_to_object() call too much. */
750 static int journal_file_check_object(JournalFile
*f
, uint64_t offset
, Object
*o
) {
754 switch (o
->object
.type
) {
757 if ((le64toh(o
->data
.entry_offset
) == 0) ^ (le64toh(o
->data
.n_entries
) == 0)) {
758 log_debug("Bad n_entries: %"PRIu64
": %"PRIu64
,
759 le64toh(o
->data
.n_entries
), offset
);
763 if (le64toh(o
->object
.size
) - offsetof(DataObject
, payload
) <= 0) {
764 log_debug("Bad object size (<= %zu): %"PRIu64
": %"PRIu64
,
765 offsetof(DataObject
, payload
),
766 le64toh(o
->object
.size
),
771 if (!VALID64(le64toh(o
->data
.next_hash_offset
)) ||
772 !VALID64(le64toh(o
->data
.next_field_offset
)) ||
773 !VALID64(le64toh(o
->data
.entry_offset
)) ||
774 !VALID64(le64toh(o
->data
.entry_array_offset
))) {
775 log_debug("Invalid offset, next_hash_offset="OFSfmt
", next_field_offset="OFSfmt
776 ", entry_offset="OFSfmt
", entry_array_offset="OFSfmt
": %"PRIu64
,
777 le64toh(o
->data
.next_hash_offset
),
778 le64toh(o
->data
.next_field_offset
),
779 le64toh(o
->data
.entry_offset
),
780 le64toh(o
->data
.entry_array_offset
),
789 if (le64toh(o
->object
.size
) - offsetof(FieldObject
, payload
) <= 0) {
791 "Bad field size (<= %zu): %"PRIu64
": %"PRIu64
,
792 offsetof(FieldObject
, payload
),
793 le64toh(o
->object
.size
),
798 if (!VALID64(le64toh(o
->field
.next_hash_offset
)) ||
799 !VALID64(le64toh(o
->field
.head_data_offset
))) {
801 "Invalid offset, next_hash_offset="OFSfmt
802 ", head_data_offset="OFSfmt
": %"PRIu64
,
803 le64toh(o
->field
.next_hash_offset
),
804 le64toh(o
->field
.head_data_offset
),
811 if ((le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) % sizeof(EntryItem
) != 0) {
813 "Bad entry size (<= %zu): %"PRIu64
": %"PRIu64
,
814 offsetof(EntryObject
, items
),
815 le64toh(o
->object
.size
),
820 if ((le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) / sizeof(EntryItem
) <= 0) {
822 "Invalid number items in entry: %"PRIu64
": %"PRIu64
,
823 (le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) / sizeof(EntryItem
),
828 if (le64toh(o
->entry
.seqnum
) <= 0) {
830 "Invalid entry seqnum: %"PRIx64
": %"PRIu64
,
831 le64toh(o
->entry
.seqnum
),
836 if (!VALID_REALTIME(le64toh(o
->entry
.realtime
))) {
838 "Invalid entry realtime timestamp: %"PRIu64
": %"PRIu64
,
839 le64toh(o
->entry
.realtime
),
844 if (!VALID_MONOTONIC(le64toh(o
->entry
.monotonic
))) {
846 "Invalid entry monotonic timestamp: %"PRIu64
": %"PRIu64
,
847 le64toh(o
->entry
.monotonic
),
854 case OBJECT_DATA_HASH_TABLE
:
855 case OBJECT_FIELD_HASH_TABLE
:
856 if ((le64toh(o
->object
.size
) - offsetof(HashTableObject
, items
)) % sizeof(HashItem
) != 0 ||
857 (le64toh(o
->object
.size
) - offsetof(HashTableObject
, items
)) / sizeof(HashItem
) <= 0) {
859 "Invalid %s hash table size: %"PRIu64
": %"PRIu64
,
860 o
->object
.type
== OBJECT_DATA_HASH_TABLE
? "data" : "field",
861 le64toh(o
->object
.size
),
868 case OBJECT_ENTRY_ARRAY
:
869 if ((le64toh(o
->object
.size
) - offsetof(EntryArrayObject
, items
)) % sizeof(le64_t
) != 0 ||
870 (le64toh(o
->object
.size
) - offsetof(EntryArrayObject
, items
)) / sizeof(le64_t
) <= 0) {
872 "Invalid object entry array size: %"PRIu64
": %"PRIu64
,
873 le64toh(o
->object
.size
),
878 if (!VALID64(le64toh(o
->entry_array
.next_entry_array_offset
))) {
880 "Invalid object entry array next_entry_array_offset: "OFSfmt
": %"PRIu64
,
881 le64toh(o
->entry_array
.next_entry_array_offset
),
889 if (le64toh(o
->object
.size
) != sizeof(TagObject
)) {
891 "Invalid object tag size: %"PRIu64
": %"PRIu64
,
892 le64toh(o
->object
.size
),
897 if (!VALID_EPOCH(le64toh(o
->tag
.epoch
))) {
899 "Invalid object tag epoch: %"PRIu64
": %"PRIu64
,
900 le64toh(o
->tag
.epoch
),
911 int journal_file_move_to_object(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
**ret
) {
921 /* Objects may only be located at multiple of 64 bit */
922 if (!VALID64(offset
)) {
923 log_debug("Attempt to move to object at non-64bit boundary: %" PRIu64
, offset
);
927 /* Object may not be located in the file header */
928 if (offset
< le64toh(f
->header
->header_size
)) {
929 log_debug("Attempt to move to object located in file header: %" PRIu64
, offset
);
933 r
= journal_file_move_to(f
, type
, false, offset
, sizeof(ObjectHeader
), &t
, &tsize
);
938 s
= le64toh(o
->object
.size
);
941 log_debug("Attempt to move to uninitialized object: %" PRIu64
, offset
);
944 if (s
< sizeof(ObjectHeader
)) {
945 log_debug("Attempt to move to overly short object: %" PRIu64
, offset
);
949 if (o
->object
.type
<= OBJECT_UNUSED
) {
950 log_debug("Attempt to move to object with invalid type: %" PRIu64
, offset
);
954 if (s
< minimum_header_size(o
)) {
955 log_debug("Attempt to move to truncated object: %" PRIu64
, offset
);
959 if (type
> OBJECT_UNUSED
&& o
->object
.type
!= type
) {
960 log_debug("Attempt to move to object of unexpected type: %" PRIu64
, offset
);
965 r
= journal_file_move_to(f
, type
, false, offset
, s
, &t
, NULL
);
972 r
= journal_file_check_object(f
, offset
, o
);
980 static uint64_t journal_file_entry_seqnum(JournalFile
*f
, uint64_t *seqnum
) {
986 r
= le64toh(f
->header
->tail_entry_seqnum
) + 1;
989 /* If an external seqnum counter was passed, we update
990 * both the local and the external one, and set it to
991 * the maximum of both */
999 f
->header
->tail_entry_seqnum
= htole64(r
);
1001 if (f
->header
->head_entry_seqnum
== 0)
1002 f
->header
->head_entry_seqnum
= htole64(r
);
1007 int journal_file_append_object(JournalFile
*f
, ObjectType type
, uint64_t size
, Object
**ret
, uint64_t *offset
) {
1015 assert(type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
);
1016 assert(size
>= sizeof(ObjectHeader
));
1020 r
= journal_file_set_online(f
);
1024 p
= le64toh(f
->header
->tail_object_offset
);
1026 p
= le64toh(f
->header
->header_size
);
1028 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &tail
);
1032 p
+= ALIGN64(le64toh(tail
->object
.size
));
1035 r
= journal_file_allocate(f
, p
, size
);
1039 r
= journal_file_move_to(f
, type
, false, p
, size
, &t
, NULL
);
1046 o
->object
.type
= type
;
1047 o
->object
.size
= htole64(size
);
1049 f
->header
->tail_object_offset
= htole64(p
);
1050 f
->header
->n_objects
= htole64(le64toh(f
->header
->n_objects
) + 1);
1058 static int journal_file_setup_data_hash_table(JournalFile
*f
) {
1066 /* We estimate that we need 1 hash table entry per 768 bytes
1067 of journal file and we want to make sure we never get
1068 beyond 75% fill level. Calculate the hash table size for
1069 the maximum file size based on these metrics. */
1071 s
= (f
->metrics
.max_size
* 4 / 768 / 3) * sizeof(HashItem
);
1072 if (s
< DEFAULT_DATA_HASH_TABLE_SIZE
)
1073 s
= DEFAULT_DATA_HASH_TABLE_SIZE
;
1075 log_debug("Reserving %"PRIu64
" entries in hash table.", s
/ sizeof(HashItem
));
1077 r
= journal_file_append_object(f
,
1078 OBJECT_DATA_HASH_TABLE
,
1079 offsetof(Object
, hash_table
.items
) + s
,
1084 memzero(o
->hash_table
.items
, s
);
1086 f
->header
->data_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1087 f
->header
->data_hash_table_size
= htole64(s
);
1092 static int journal_file_setup_field_hash_table(JournalFile
*f
) {
1100 /* We use a fixed size hash table for the fields as this
1101 * number should grow very slowly only */
1103 s
= DEFAULT_FIELD_HASH_TABLE_SIZE
;
1104 r
= journal_file_append_object(f
,
1105 OBJECT_FIELD_HASH_TABLE
,
1106 offsetof(Object
, hash_table
.items
) + s
,
1111 memzero(o
->hash_table
.items
, s
);
1113 f
->header
->field_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1114 f
->header
->field_hash_table_size
= htole64(s
);
1119 int journal_file_map_data_hash_table(JournalFile
*f
) {
1127 if (f
->data_hash_table
)
1130 p
= le64toh(f
->header
->data_hash_table_offset
);
1131 s
= le64toh(f
->header
->data_hash_table_size
);
1133 r
= journal_file_move_to(f
,
1134 OBJECT_DATA_HASH_TABLE
,
1141 f
->data_hash_table
= t
;
1145 int journal_file_map_field_hash_table(JournalFile
*f
) {
1153 if (f
->field_hash_table
)
1156 p
= le64toh(f
->header
->field_hash_table_offset
);
1157 s
= le64toh(f
->header
->field_hash_table_size
);
1159 r
= journal_file_move_to(f
,
1160 OBJECT_FIELD_HASH_TABLE
,
1167 f
->field_hash_table
= t
;
1171 static int journal_file_link_field(
1182 assert(f
->field_hash_table
);
1186 if (o
->object
.type
!= OBJECT_FIELD
)
1189 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1193 /* This might alter the window we are looking at */
1194 o
->field
.next_hash_offset
= o
->field
.head_data_offset
= 0;
1197 p
= le64toh(f
->field_hash_table
[h
].tail_hash_offset
);
1199 f
->field_hash_table
[h
].head_hash_offset
= htole64(offset
);
1201 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1205 o
->field
.next_hash_offset
= htole64(offset
);
1208 f
->field_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1210 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
1211 f
->header
->n_fields
= htole64(le64toh(f
->header
->n_fields
) + 1);
1216 static int journal_file_link_data(
1227 assert(f
->data_hash_table
);
1231 if (o
->object
.type
!= OBJECT_DATA
)
1234 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1238 /* This might alter the window we are looking at */
1239 o
->data
.next_hash_offset
= o
->data
.next_field_offset
= 0;
1240 o
->data
.entry_offset
= o
->data
.entry_array_offset
= 0;
1241 o
->data
.n_entries
= 0;
1244 p
= le64toh(f
->data_hash_table
[h
].tail_hash_offset
);
1246 /* Only entry in the hash table is easy */
1247 f
->data_hash_table
[h
].head_hash_offset
= htole64(offset
);
1249 /* Move back to the previous data object, to patch in
1252 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1256 o
->data
.next_hash_offset
= htole64(offset
);
1259 f
->data_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1261 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
1262 f
->header
->n_data
= htole64(le64toh(f
->header
->n_data
) + 1);
1267 int journal_file_find_field_object_with_hash(
1269 const void *field
, uint64_t size
, uint64_t hash
,
1270 Object
**ret
, uint64_t *offset
) {
1272 uint64_t p
, osize
, h
, m
;
1277 assert(field
&& size
> 0);
1279 /* If the field hash table is empty, we can't find anything */
1280 if (le64toh(f
->header
->field_hash_table_size
) <= 0)
1283 /* Map the field hash table, if it isn't mapped yet. */
1284 r
= journal_file_map_field_hash_table(f
);
1288 osize
= offsetof(Object
, field
.payload
) + size
;
1290 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1295 p
= le64toh(f
->field_hash_table
[h
].head_hash_offset
);
1300 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1304 if (le64toh(o
->field
.hash
) == hash
&&
1305 le64toh(o
->object
.size
) == osize
&&
1306 memcmp(o
->field
.payload
, field
, size
) == 0) {
1316 p
= le64toh(o
->field
.next_hash_offset
);
1322 int journal_file_find_field_object(
1324 const void *field
, uint64_t size
,
1325 Object
**ret
, uint64_t *offset
) {
1330 assert(field
&& size
> 0);
1332 hash
= hash64(field
, size
);
1334 return journal_file_find_field_object_with_hash(f
,
1339 int journal_file_find_data_object_with_hash(
1341 const void *data
, uint64_t size
, uint64_t hash
,
1342 Object
**ret
, uint64_t *offset
) {
1344 uint64_t p
, osize
, h
, m
;
1349 assert(data
|| size
== 0);
1351 /* If there's no data hash table, then there's no entry. */
1352 if (le64toh(f
->header
->data_hash_table_size
) <= 0)
1355 /* Map the data hash table, if it isn't mapped yet. */
1356 r
= journal_file_map_data_hash_table(f
);
1360 osize
= offsetof(Object
, data
.payload
) + size
;
1362 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1367 p
= le64toh(f
->data_hash_table
[h
].head_hash_offset
);
1372 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1376 if (le64toh(o
->data
.hash
) != hash
)
1379 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
1380 #if HAVE_XZ || HAVE_LZ4
1384 l
= le64toh(o
->object
.size
);
1385 if (l
<= offsetof(Object
, data
.payload
))
1388 l
-= offsetof(Object
, data
.payload
);
1390 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
1391 o
->data
.payload
, l
, &f
->compress_buffer
, &f
->compress_buffer_size
, &rsize
, 0);
1395 if (rsize
== size
&&
1396 memcmp(f
->compress_buffer
, data
, size
) == 0) {
1407 return -EPROTONOSUPPORT
;
1409 } else if (le64toh(o
->object
.size
) == osize
&&
1410 memcmp(o
->data
.payload
, data
, size
) == 0) {
1422 p
= le64toh(o
->data
.next_hash_offset
);
1428 int journal_file_find_data_object(
1430 const void *data
, uint64_t size
,
1431 Object
**ret
, uint64_t *offset
) {
1436 assert(data
|| size
== 0);
1438 hash
= hash64(data
, size
);
1440 return journal_file_find_data_object_with_hash(f
,
1445 static int journal_file_append_field(
1447 const void *field
, uint64_t size
,
1448 Object
**ret
, uint64_t *offset
) {
1456 assert(field
&& size
> 0);
1458 hash
= hash64(field
, size
);
1460 r
= journal_file_find_field_object_with_hash(f
, field
, size
, hash
, &o
, &p
);
1474 osize
= offsetof(Object
, field
.payload
) + size
;
1475 r
= journal_file_append_object(f
, OBJECT_FIELD
, osize
, &o
, &p
);
1479 o
->field
.hash
= htole64(hash
);
1480 memcpy(o
->field
.payload
, field
, size
);
1482 r
= journal_file_link_field(f
, o
, p
, hash
);
1486 /* The linking might have altered the window, so let's
1487 * refresh our pointer */
1488 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1493 r
= journal_file_hmac_put_object(f
, OBJECT_FIELD
, o
, p
);
1507 static int journal_file_append_data(
1509 const void *data
, uint64_t size
,
1510 Object
**ret
, uint64_t *offset
) {
1515 int r
, compression
= 0;
1519 assert(data
|| size
== 0);
1521 hash
= hash64(data
, size
);
1523 r
= journal_file_find_data_object_with_hash(f
, data
, size
, hash
, &o
, &p
);
1537 osize
= offsetof(Object
, data
.payload
) + size
;
1538 r
= journal_file_append_object(f
, OBJECT_DATA
, osize
, &o
, &p
);
1542 o
->data
.hash
= htole64(hash
);
1544 #if HAVE_XZ || HAVE_LZ4
1545 if (JOURNAL_FILE_COMPRESS(f
) && size
>= f
->compress_threshold_bytes
) {
1548 compression
= compress_blob(data
, size
, o
->data
.payload
, size
- 1, &rsize
);
1550 if (compression
>= 0) {
1551 o
->object
.size
= htole64(offsetof(Object
, data
.payload
) + rsize
);
1552 o
->object
.flags
|= compression
;
1554 log_debug("Compressed data object %"PRIu64
" -> %zu using %s",
1555 size
, rsize
, object_compressed_to_string(compression
));
1557 /* Compression didn't work, we don't really care why, let's continue without compression */
1562 if (compression
== 0)
1563 memcpy_safe(o
->data
.payload
, data
, size
);
1565 r
= journal_file_link_data(f
, o
, p
, hash
);
1570 r
= journal_file_hmac_put_object(f
, OBJECT_DATA
, o
, p
);
1575 /* The linking might have altered the window, so let's
1576 * refresh our pointer */
1577 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1584 eq
= memchr(data
, '=', size
);
1585 if (eq
&& eq
> data
) {
1589 /* Create field object ... */
1590 r
= journal_file_append_field(f
, data
, (uint8_t*) eq
- (uint8_t*) data
, &fo
, &fp
);
1594 /* ... and link it in. */
1595 o
->data
.next_field_offset
= fo
->field
.head_data_offset
;
1596 fo
->field
.head_data_offset
= le64toh(p
);
1608 uint64_t journal_file_entry_n_items(Object
*o
) {
1611 if (o
->object
.type
!= OBJECT_ENTRY
)
1614 return (le64toh(o
->object
.size
) - offsetof(Object
, entry
.items
)) / sizeof(EntryItem
);
1617 uint64_t journal_file_entry_array_n_items(Object
*o
) {
1620 if (o
->object
.type
!= OBJECT_ENTRY_ARRAY
)
1623 return (le64toh(o
->object
.size
) - offsetof(Object
, entry_array
.items
)) / sizeof(uint64_t);
1626 uint64_t journal_file_hash_table_n_items(Object
*o
) {
1629 if (!IN_SET(o
->object
.type
, OBJECT_DATA_HASH_TABLE
, OBJECT_FIELD_HASH_TABLE
))
1632 return (le64toh(o
->object
.size
) - offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
);
1635 static int link_entry_into_array(JournalFile
*f
,
1640 uint64_t n
= 0, ap
= 0, q
, i
, a
, hidx
;
1649 a
= le64toh(*first
);
1650 i
= hidx
= le64toh(*idx
);
1653 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
1657 n
= journal_file_entry_array_n_items(o
);
1659 o
->entry_array
.items
[i
] = htole64(p
);
1660 *idx
= htole64(hidx
+ 1);
1666 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
1677 r
= journal_file_append_object(f
, OBJECT_ENTRY_ARRAY
,
1678 offsetof(Object
, entry_array
.items
) + n
* sizeof(uint64_t),
1684 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY_ARRAY
, o
, q
);
1689 o
->entry_array
.items
[i
] = htole64(p
);
1692 *first
= htole64(q
);
1694 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, ap
, &o
);
1698 o
->entry_array
.next_entry_array_offset
= htole64(q
);
1701 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
1702 f
->header
->n_entry_arrays
= htole64(le64toh(f
->header
->n_entry_arrays
) + 1);
1704 *idx
= htole64(hidx
+ 1);
1709 static int link_entry_into_array_plus_one(JournalFile
*f
,
1724 *extra
= htole64(p
);
1728 i
= htole64(le64toh(*idx
) - 1);
1729 r
= link_entry_into_array(f
, first
, &i
, p
);
1734 *idx
= htole64(le64toh(*idx
) + 1);
1738 static int journal_file_link_entry_item(JournalFile
*f
, Object
*o
, uint64_t offset
, uint64_t i
) {
1745 p
= le64toh(o
->entry
.items
[i
].object_offset
);
1749 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1753 return link_entry_into_array_plus_one(f
,
1754 &o
->data
.entry_offset
,
1755 &o
->data
.entry_array_offset
,
1760 static int journal_file_link_entry(JournalFile
*f
, Object
*o
, uint64_t offset
) {
1769 if (o
->object
.type
!= OBJECT_ENTRY
)
1772 __sync_synchronize();
1774 /* Link up the entry itself */
1775 r
= link_entry_into_array(f
,
1776 &f
->header
->entry_array_offset
,
1777 &f
->header
->n_entries
,
1782 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
1784 if (f
->header
->head_entry_realtime
== 0)
1785 f
->header
->head_entry_realtime
= o
->entry
.realtime
;
1787 f
->header
->tail_entry_realtime
= o
->entry
.realtime
;
1788 f
->header
->tail_entry_monotonic
= o
->entry
.monotonic
;
1790 /* Link up the items */
1791 n
= journal_file_entry_n_items(o
);
1792 for (i
= 0; i
< n
; i
++) {
1793 r
= journal_file_link_entry_item(f
, o
, offset
, i
);
1801 static int journal_file_append_entry_internal(
1803 const dual_timestamp
*ts
,
1804 const sd_id128_t
*boot_id
,
1806 const EntryItem items
[], unsigned n_items
,
1808 Object
**ret
, uint64_t *offset
) {
1816 assert(items
|| n_items
== 0);
1819 osize
= offsetof(Object
, entry
.items
) + (n_items
* sizeof(EntryItem
));
1821 r
= journal_file_append_object(f
, OBJECT_ENTRY
, osize
, &o
, &np
);
1825 o
->entry
.seqnum
= htole64(journal_file_entry_seqnum(f
, seqnum
));
1826 memcpy_safe(o
->entry
.items
, items
, n_items
* sizeof(EntryItem
));
1827 o
->entry
.realtime
= htole64(ts
->realtime
);
1828 o
->entry
.monotonic
= htole64(ts
->monotonic
);
1829 o
->entry
.xor_hash
= htole64(xor_hash
);
1830 o
->entry
.boot_id
= boot_id
? *boot_id
: f
->header
->boot_id
;
1833 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY
, o
, np
);
1838 r
= journal_file_link_entry(f
, o
, np
);
1851 void journal_file_post_change(JournalFile
*f
) {
1854 /* inotify() does not receive IN_MODIFY events from file
1855 * accesses done via mmap(). After each access we hence
1856 * trigger IN_MODIFY by truncating the journal file to its
1857 * current size which triggers IN_MODIFY. */
1859 __sync_synchronize();
1861 if (ftruncate(f
->fd
, f
->last_stat
.st_size
) < 0)
1862 log_debug_errno(errno
, "Failed to truncate file to its own size: %m");
1865 static int post_change_thunk(sd_event_source
*timer
, uint64_t usec
, void *userdata
) {
1868 journal_file_post_change(userdata
);
1873 static void schedule_post_change(JournalFile
*f
) {
1874 sd_event_source
*timer
;
1879 assert(f
->post_change_timer
);
1881 timer
= f
->post_change_timer
;
1883 r
= sd_event_source_get_enabled(timer
, &enabled
);
1885 log_debug_errno(r
, "Failed to get ftruncate timer state: %m");
1889 if (enabled
== SD_EVENT_ONESHOT
)
1892 r
= sd_event_now(sd_event_source_get_event(timer
), CLOCK_MONOTONIC
, &now
);
1894 log_debug_errno(r
, "Failed to get clock's now for scheduling ftruncate: %m");
1898 r
= sd_event_source_set_time(timer
, now
+f
->post_change_timer_period
);
1900 log_debug_errno(r
, "Failed to set time for scheduling ftruncate: %m");
1904 r
= sd_event_source_set_enabled(timer
, SD_EVENT_ONESHOT
);
1906 log_debug_errno(r
, "Failed to enable scheduled ftruncate: %m");
1913 /* On failure, let's simply post the change immediately. */
1914 journal_file_post_change(f
);
1917 /* Enable coalesced change posting in a timer on the provided sd_event instance */
1918 int journal_file_enable_post_change_timer(JournalFile
*f
, sd_event
*e
, usec_t t
) {
1919 _cleanup_(sd_event_source_unrefp
) sd_event_source
*timer
= NULL
;
1923 assert_return(!f
->post_change_timer
, -EINVAL
);
1927 r
= sd_event_add_time(e
, &timer
, CLOCK_MONOTONIC
, 0, 0, post_change_thunk
, f
);
1931 r
= sd_event_source_set_enabled(timer
, SD_EVENT_OFF
);
1935 f
->post_change_timer
= TAKE_PTR(timer
);
1936 f
->post_change_timer_period
= t
;
1941 static int entry_item_cmp(const void *_a
, const void *_b
) {
1942 const EntryItem
*a
= _a
, *b
= _b
;
1944 if (le64toh(a
->object_offset
) < le64toh(b
->object_offset
))
1946 if (le64toh(a
->object_offset
) > le64toh(b
->object_offset
))
1951 int journal_file_append_entry(
1953 const dual_timestamp
*ts
,
1954 const sd_id128_t
*boot_id
,
1955 const struct iovec iovec
[], unsigned n_iovec
,
1957 Object
**ret
, uint64_t *offset
) {
1962 uint64_t xor_hash
= 0;
1963 struct dual_timestamp _ts
;
1967 assert(iovec
|| n_iovec
== 0);
1970 if (!VALID_REALTIME(ts
->realtime
)) {
1971 log_debug("Invalid realtime timestamp %"PRIu64
", refusing entry.", ts
->realtime
);
1974 if (!VALID_MONOTONIC(ts
->monotonic
)) {
1975 log_debug("Invalid monotomic timestamp %"PRIu64
", refusing entry.", ts
->monotonic
);
1979 dual_timestamp_get(&_ts
);
1984 r
= journal_file_maybe_append_tag(f
, ts
->realtime
);
1989 /* alloca() can't take 0, hence let's allocate at least one */
1990 items
= newa(EntryItem
, MAX(1u, n_iovec
));
1992 for (i
= 0; i
< n_iovec
; i
++) {
1996 r
= journal_file_append_data(f
, iovec
[i
].iov_base
, iovec
[i
].iov_len
, &o
, &p
);
2000 xor_hash
^= le64toh(o
->data
.hash
);
2001 items
[i
].object_offset
= htole64(p
);
2002 items
[i
].hash
= o
->data
.hash
;
2005 /* Order by the position on disk, in order to improve seek
2006 * times for rotating media. */
2007 qsort_safe(items
, n_iovec
, sizeof(EntryItem
), entry_item_cmp
);
2009 r
= journal_file_append_entry_internal(f
, ts
, boot_id
, xor_hash
, items
, n_iovec
, seqnum
, ret
, offset
);
2011 /* If the memory mapping triggered a SIGBUS then we return an
2012 * IO error and ignore the error code passed down to us, since
2013 * it is very likely just an effect of a nullified replacement
2016 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
2019 if (f
->post_change_timer
)
2020 schedule_post_change(f
);
2022 journal_file_post_change(f
);
2027 typedef struct ChainCacheItem
{
2028 uint64_t first
; /* the array at the beginning of the chain */
2029 uint64_t array
; /* the cached array */
2030 uint64_t begin
; /* the first item in the cached array */
2031 uint64_t total
; /* the total number of items in all arrays before this one in the chain */
2032 uint64_t last_index
; /* the last index we looked at, to optimize locality when bisecting */
2035 static void chain_cache_put(
2042 uint64_t last_index
) {
2045 /* If the chain item to cache for this chain is the
2046 * first one it's not worth caching anything */
2050 if (ordered_hashmap_size(h
) >= CHAIN_CACHE_MAX
) {
2051 ci
= ordered_hashmap_steal_first(h
);
2054 ci
= new(ChainCacheItem
, 1);
2061 if (ordered_hashmap_put(h
, &ci
->first
, ci
) < 0) {
2066 assert(ci
->first
== first
);
2071 ci
->last_index
= last_index
;
2074 static int generic_array_get(
2078 Object
**ret
, uint64_t *offset
) {
2081 uint64_t p
= 0, a
, t
= 0;
2089 /* Try the chain cache first */
2090 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2091 if (ci
&& i
> ci
->total
) {
2100 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2104 k
= journal_file_entry_array_n_items(o
);
2106 p
= le64toh(o
->entry_array
.items
[i
]);
2112 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
2118 /* Let's cache this item for the next invocation */
2119 chain_cache_put(f
->chain_cache
, ci
, first
, a
, le64toh(o
->entry_array
.items
[0]), t
, i
);
2121 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2134 static int generic_array_get_plus_one(
2139 Object
**ret
, uint64_t *offset
) {
2148 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2161 return generic_array_get(f
, first
, i
-1, ret
, offset
);
2170 static int generic_array_bisect(
2175 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2176 direction_t direction
,
2181 uint64_t a
, p
, t
= 0, i
= 0, last_p
= 0, last_index
= (uint64_t) -1;
2182 bool subtract_one
= false;
2183 Object
*o
, *array
= NULL
;
2188 assert(test_object
);
2190 /* Start with the first array in the chain */
2193 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2194 if (ci
&& n
> ci
->total
&& ci
->begin
!= 0) {
2195 /* Ah, we have iterated this bisection array chain
2196 * previously! Let's see if we can skip ahead in the
2197 * chain, as far as the last time. But we can't jump
2198 * backwards in the chain, so let's check that
2201 r
= test_object(f
, ci
->begin
, needle
);
2205 if (r
== TEST_LEFT
) {
2206 /* OK, what we are looking for is right of the
2207 * begin of this EntryArray, so let's jump
2208 * straight to previously cached array in the
2214 last_index
= ci
->last_index
;
2219 uint64_t left
, right
, k
, lp
;
2221 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2225 k
= journal_file_entry_array_n_items(array
);
2231 lp
= p
= le64toh(array
->entry_array
.items
[i
]);
2235 r
= test_object(f
, p
, needle
);
2236 if (r
== -EBADMSG
) {
2237 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (1)");
2244 if (r
== TEST_FOUND
)
2245 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2247 if (r
== TEST_RIGHT
) {
2251 if (last_index
!= (uint64_t) -1) {
2252 assert(last_index
<= right
);
2254 /* If we cached the last index we
2255 * looked at, let's try to not to jump
2256 * too wildly around and see if we can
2257 * limit the range to look at early to
2258 * the immediate neighbors of the last
2259 * index we looked at. */
2261 if (last_index
> 0) {
2262 uint64_t x
= last_index
- 1;
2264 p
= le64toh(array
->entry_array
.items
[x
]);
2268 r
= test_object(f
, p
, needle
);
2272 if (r
== TEST_FOUND
)
2273 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2275 if (r
== TEST_RIGHT
)
2281 if (last_index
< right
) {
2282 uint64_t y
= last_index
+ 1;
2284 p
= le64toh(array
->entry_array
.items
[y
]);
2288 r
= test_object(f
, p
, needle
);
2292 if (r
== TEST_FOUND
)
2293 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2295 if (r
== TEST_RIGHT
)
2303 if (left
== right
) {
2304 if (direction
== DIRECTION_UP
)
2305 subtract_one
= true;
2311 assert(left
< right
);
2312 i
= (left
+ right
) / 2;
2314 p
= le64toh(array
->entry_array
.items
[i
]);
2318 r
= test_object(f
, p
, needle
);
2319 if (r
== -EBADMSG
) {
2320 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (2)");
2327 if (r
== TEST_FOUND
)
2328 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2330 if (r
== TEST_RIGHT
)
2338 if (direction
== DIRECTION_UP
) {
2340 subtract_one
= true;
2351 last_index
= (uint64_t) -1;
2352 a
= le64toh(array
->entry_array
.next_entry_array_offset
);
2358 if (subtract_one
&& t
== 0 && i
== 0)
2361 /* Let's cache this item for the next invocation */
2362 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
);
2364 if (subtract_one
&& i
== 0)
2366 else if (subtract_one
)
2367 p
= le64toh(array
->entry_array
.items
[i
-1]);
2369 p
= le64toh(array
->entry_array
.items
[i
]);
2371 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2382 *idx
= t
+ i
+ (subtract_one
? -1 : 0);
2387 static int generic_array_bisect_plus_one(
2393 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2394 direction_t direction
,
2400 bool step_back
= false;
2404 assert(test_object
);
2409 /* This bisects the array in object 'first', but first checks
2411 r
= test_object(f
, extra
, needle
);
2415 if (r
== TEST_FOUND
)
2416 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2418 /* if we are looking with DIRECTION_UP then we need to first
2419 see if in the actual array there is a matching entry, and
2420 return the last one of that. But if there isn't any we need
2421 to return this one. Hence remember this, and return it
2424 step_back
= direction
== DIRECTION_UP
;
2426 if (r
== TEST_RIGHT
) {
2427 if (direction
== DIRECTION_DOWN
)
2433 r
= generic_array_bisect(f
, first
, n
-1, needle
, test_object
, direction
, ret
, offset
, idx
);
2435 if (r
== 0 && step_back
)
2444 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2460 _pure_
static int test_object_offset(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2466 else if (p
< needle
)
2472 static int test_object_seqnum(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2479 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2483 if (le64toh(o
->entry
.seqnum
) == needle
)
2485 else if (le64toh(o
->entry
.seqnum
) < needle
)
2491 int journal_file_move_to_entry_by_seqnum(
2494 direction_t direction
,
2500 return generic_array_bisect(f
,
2501 le64toh(f
->header
->entry_array_offset
),
2502 le64toh(f
->header
->n_entries
),
2509 static int test_object_realtime(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2516 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2520 if (le64toh(o
->entry
.realtime
) == needle
)
2522 else if (le64toh(o
->entry
.realtime
) < needle
)
2528 int journal_file_move_to_entry_by_realtime(
2531 direction_t direction
,
2537 return generic_array_bisect(f
,
2538 le64toh(f
->header
->entry_array_offset
),
2539 le64toh(f
->header
->n_entries
),
2541 test_object_realtime
,
2546 static int test_object_monotonic(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2553 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2557 if (le64toh(o
->entry
.monotonic
) == needle
)
2559 else if (le64toh(o
->entry
.monotonic
) < needle
)
2565 static int find_data_object_by_boot_id(
2571 char t
[STRLEN("_BOOT_ID=") + 32 + 1] = "_BOOT_ID=";
2573 sd_id128_to_string(boot_id
, t
+ 9);
2574 return journal_file_find_data_object(f
, t
, sizeof(t
) - 1, o
, b
);
2577 int journal_file_move_to_entry_by_monotonic(
2581 direction_t direction
,
2590 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, NULL
);
2596 return generic_array_bisect_plus_one(f
,
2597 le64toh(o
->data
.entry_offset
),
2598 le64toh(o
->data
.entry_array_offset
),
2599 le64toh(o
->data
.n_entries
),
2601 test_object_monotonic
,
2606 void journal_file_reset_location(JournalFile
*f
) {
2607 f
->location_type
= LOCATION_HEAD
;
2608 f
->current_offset
= 0;
2609 f
->current_seqnum
= 0;
2610 f
->current_realtime
= 0;
2611 f
->current_monotonic
= 0;
2612 zero(f
->current_boot_id
);
2613 f
->current_xor_hash
= 0;
2616 void journal_file_save_location(JournalFile
*f
, Object
*o
, uint64_t offset
) {
2617 f
->location_type
= LOCATION_SEEK
;
2618 f
->current_offset
= offset
;
2619 f
->current_seqnum
= le64toh(o
->entry
.seqnum
);
2620 f
->current_realtime
= le64toh(o
->entry
.realtime
);
2621 f
->current_monotonic
= le64toh(o
->entry
.monotonic
);
2622 f
->current_boot_id
= o
->entry
.boot_id
;
2623 f
->current_xor_hash
= le64toh(o
->entry
.xor_hash
);
2626 int journal_file_compare_locations(JournalFile
*af
, JournalFile
*bf
) {
2631 assert(af
->location_type
== LOCATION_SEEK
);
2632 assert(bf
->location_type
== LOCATION_SEEK
);
2634 /* If contents and timestamps match, these entries are
2635 * identical, even if the seqnum does not match */
2636 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
) &&
2637 af
->current_monotonic
== bf
->current_monotonic
&&
2638 af
->current_realtime
== bf
->current_realtime
&&
2639 af
->current_xor_hash
== bf
->current_xor_hash
)
2642 if (sd_id128_equal(af
->header
->seqnum_id
, bf
->header
->seqnum_id
)) {
2644 /* If this is from the same seqnum source, compare
2646 if (af
->current_seqnum
< bf
->current_seqnum
)
2648 if (af
->current_seqnum
> bf
->current_seqnum
)
2651 /* Wow! This is weird, different data but the same
2652 * seqnums? Something is borked, but let's make the
2653 * best of it and compare by time. */
2656 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
)) {
2658 /* If the boot id matches, compare monotonic time */
2659 if (af
->current_monotonic
< bf
->current_monotonic
)
2661 if (af
->current_monotonic
> bf
->current_monotonic
)
2665 /* Otherwise, compare UTC time */
2666 if (af
->current_realtime
< bf
->current_realtime
)
2668 if (af
->current_realtime
> bf
->current_realtime
)
2671 /* Finally, compare by contents */
2672 if (af
->current_xor_hash
< bf
->current_xor_hash
)
2674 if (af
->current_xor_hash
> bf
->current_xor_hash
)
2680 static int bump_array_index(uint64_t *i
, direction_t direction
, uint64_t n
) {
2682 /* Increase or decrease the specified index, in the right direction. */
2684 if (direction
== DIRECTION_DOWN
) {
2699 static bool check_properly_ordered(uint64_t new_offset
, uint64_t old_offset
, direction_t direction
) {
2701 /* Consider it an error if any of the two offsets is uninitialized */
2702 if (old_offset
== 0 || new_offset
== 0)
2705 /* If we go down, the new offset must be larger than the old one. */
2706 return direction
== DIRECTION_DOWN
?
2707 new_offset
> old_offset
:
2708 new_offset
< old_offset
;
2711 int journal_file_next_entry(
2714 direction_t direction
,
2715 Object
**ret
, uint64_t *offset
) {
2723 n
= le64toh(f
->header
->n_entries
);
2728 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2730 r
= generic_array_bisect(f
,
2731 le64toh(f
->header
->entry_array_offset
),
2732 le64toh(f
->header
->n_entries
),
2741 r
= bump_array_index(&i
, direction
, n
);
2746 /* And jump to it */
2748 r
= generic_array_get(f
,
2749 le64toh(f
->header
->entry_array_offset
),
2757 /* OK, so this entry is borked. Most likely some entry didn't get synced to disk properly, let's see if
2758 * the next one might work for us instead. */
2759 log_debug_errno(r
, "Entry item %" PRIu64
" is bad, skipping over it.", i
);
2761 r
= bump_array_index(&i
, direction
, n
);
2766 /* Ensure our array is properly ordered. */
2767 if (p
> 0 && !check_properly_ordered(ofs
, p
, direction
)) {
2768 log_debug("%s: entry array not properly ordered at entry %" PRIu64
, f
->path
, i
);
2778 int journal_file_next_entry_for_data(
2780 Object
*o
, uint64_t p
,
2781 uint64_t data_offset
,
2782 direction_t direction
,
2783 Object
**ret
, uint64_t *offset
) {
2790 assert(p
> 0 || !o
);
2792 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2796 n
= le64toh(d
->data
.n_entries
);
2801 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2803 if (o
->object
.type
!= OBJECT_ENTRY
)
2806 r
= generic_array_bisect_plus_one(f
,
2807 le64toh(d
->data
.entry_offset
),
2808 le64toh(d
->data
.entry_array_offset
),
2809 le64toh(d
->data
.n_entries
),
2819 r
= bump_array_index(&i
, direction
, n
);
2825 r
= generic_array_get_plus_one(f
,
2826 le64toh(d
->data
.entry_offset
),
2827 le64toh(d
->data
.entry_array_offset
),
2835 log_debug_errno(r
, "Data entry item %" PRIu64
" is bad, skipping over it.", i
);
2837 r
= bump_array_index(&i
, direction
, n
);
2842 /* Ensure our array is properly ordered. */
2843 if (p
> 0 && check_properly_ordered(ofs
, p
, direction
)) {
2844 log_debug("%s data entry array not properly ordered at entry %" PRIu64
, f
->path
, i
);
2854 int journal_file_move_to_entry_by_offset_for_data(
2856 uint64_t data_offset
,
2858 direction_t direction
,
2859 Object
**ret
, uint64_t *offset
) {
2866 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2870 return generic_array_bisect_plus_one(f
,
2871 le64toh(d
->data
.entry_offset
),
2872 le64toh(d
->data
.entry_array_offset
),
2873 le64toh(d
->data
.n_entries
),
2880 int journal_file_move_to_entry_by_monotonic_for_data(
2882 uint64_t data_offset
,
2885 direction_t direction
,
2886 Object
**ret
, uint64_t *offset
) {
2894 /* First, seek by time */
2895 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &b
);
2901 r
= generic_array_bisect_plus_one(f
,
2902 le64toh(o
->data
.entry_offset
),
2903 le64toh(o
->data
.entry_array_offset
),
2904 le64toh(o
->data
.n_entries
),
2906 test_object_monotonic
,
2912 /* And now, continue seeking until we find an entry that
2913 * exists in both bisection arrays */
2919 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2923 r
= generic_array_bisect_plus_one(f
,
2924 le64toh(d
->data
.entry_offset
),
2925 le64toh(d
->data
.entry_array_offset
),
2926 le64toh(d
->data
.n_entries
),
2934 r
= journal_file_move_to_object(f
, OBJECT_DATA
, b
, &o
);
2938 r
= generic_array_bisect_plus_one(f
,
2939 le64toh(o
->data
.entry_offset
),
2940 le64toh(o
->data
.entry_array_offset
),
2941 le64toh(o
->data
.n_entries
),
2963 int journal_file_move_to_entry_by_seqnum_for_data(
2965 uint64_t data_offset
,
2967 direction_t direction
,
2968 Object
**ret
, uint64_t *offset
) {
2975 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2979 return generic_array_bisect_plus_one(f
,
2980 le64toh(d
->data
.entry_offset
),
2981 le64toh(d
->data
.entry_array_offset
),
2982 le64toh(d
->data
.n_entries
),
2989 int journal_file_move_to_entry_by_realtime_for_data(
2991 uint64_t data_offset
,
2993 direction_t direction
,
2994 Object
**ret
, uint64_t *offset
) {
3001 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
3005 return generic_array_bisect_plus_one(f
,
3006 le64toh(d
->data
.entry_offset
),
3007 le64toh(d
->data
.entry_array_offset
),
3008 le64toh(d
->data
.n_entries
),
3010 test_object_realtime
,
3015 void journal_file_dump(JournalFile
*f
) {
3023 journal_file_print_header(f
);
3025 p
= le64toh(f
->header
->header_size
);
3027 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &o
);
3031 switch (o
->object
.type
) {
3034 printf("Type: OBJECT_UNUSED\n");
3038 printf("Type: OBJECT_DATA\n");
3042 printf("Type: OBJECT_FIELD\n");
3046 printf("Type: OBJECT_ENTRY seqnum=%"PRIu64
" monotonic=%"PRIu64
" realtime=%"PRIu64
"\n",
3047 le64toh(o
->entry
.seqnum
),
3048 le64toh(o
->entry
.monotonic
),
3049 le64toh(o
->entry
.realtime
));
3052 case OBJECT_FIELD_HASH_TABLE
:
3053 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
3056 case OBJECT_DATA_HASH_TABLE
:
3057 printf("Type: OBJECT_DATA_HASH_TABLE\n");
3060 case OBJECT_ENTRY_ARRAY
:
3061 printf("Type: OBJECT_ENTRY_ARRAY\n");
3065 printf("Type: OBJECT_TAG seqnum=%"PRIu64
" epoch=%"PRIu64
"\n",
3066 le64toh(o
->tag
.seqnum
),
3067 le64toh(o
->tag
.epoch
));
3071 printf("Type: unknown (%i)\n", o
->object
.type
);
3075 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
)
3076 printf("Flags: %s\n",
3077 object_compressed_to_string(o
->object
.flags
& OBJECT_COMPRESSION_MASK
));
3079 if (p
== le64toh(f
->header
->tail_object_offset
))
3082 p
= p
+ ALIGN64(le64toh(o
->object
.size
));
3087 log_error("File corrupt");
3090 static const char* format_timestamp_safe(char *buf
, size_t l
, usec_t t
) {
3093 x
= format_timestamp(buf
, l
, t
);
3099 void journal_file_print_header(JournalFile
*f
) {
3100 char a
[33], b
[33], c
[33], d
[33];
3101 char x
[FORMAT_TIMESTAMP_MAX
], y
[FORMAT_TIMESTAMP_MAX
], z
[FORMAT_TIMESTAMP_MAX
];
3103 char bytes
[FORMAT_BYTES_MAX
];
3108 printf("File Path: %s\n"
3112 "Sequential Number ID: %s\n"
3114 "Compatible Flags:%s%s\n"
3115 "Incompatible Flags:%s%s%s\n"
3116 "Header size: %"PRIu64
"\n"
3117 "Arena size: %"PRIu64
"\n"
3118 "Data Hash Table Size: %"PRIu64
"\n"
3119 "Field Hash Table Size: %"PRIu64
"\n"
3120 "Rotate Suggested: %s\n"
3121 "Head Sequential Number: %"PRIu64
" (%"PRIx64
")\n"
3122 "Tail Sequential Number: %"PRIu64
" (%"PRIx64
")\n"
3123 "Head Realtime Timestamp: %s (%"PRIx64
")\n"
3124 "Tail Realtime Timestamp: %s (%"PRIx64
")\n"
3125 "Tail Monotonic Timestamp: %s (%"PRIx64
")\n"
3126 "Objects: %"PRIu64
"\n"
3127 "Entry Objects: %"PRIu64
"\n",
3129 sd_id128_to_string(f
->header
->file_id
, a
),
3130 sd_id128_to_string(f
->header
->machine_id
, b
),
3131 sd_id128_to_string(f
->header
->boot_id
, c
),
3132 sd_id128_to_string(f
->header
->seqnum_id
, d
),
3133 f
->header
->state
== STATE_OFFLINE
? "OFFLINE" :
3134 f
->header
->state
== STATE_ONLINE
? "ONLINE" :
3135 f
->header
->state
== STATE_ARCHIVED
? "ARCHIVED" : "UNKNOWN",
3136 JOURNAL_HEADER_SEALED(f
->header
) ? " SEALED" : "",
3137 (le32toh(f
->header
->compatible_flags
) & ~HEADER_COMPATIBLE_ANY
) ? " ???" : "",
3138 JOURNAL_HEADER_COMPRESSED_XZ(f
->header
) ? " COMPRESSED-XZ" : "",
3139 JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
) ? " COMPRESSED-LZ4" : "",
3140 (le32toh(f
->header
->incompatible_flags
) & ~HEADER_INCOMPATIBLE_ANY
) ? " ???" : "",
3141 le64toh(f
->header
->header_size
),
3142 le64toh(f
->header
->arena_size
),
3143 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3144 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
),
3145 yes_no(journal_file_rotate_suggested(f
, 0)),
3146 le64toh(f
->header
->head_entry_seqnum
), le64toh(f
->header
->head_entry_seqnum
),
3147 le64toh(f
->header
->tail_entry_seqnum
), le64toh(f
->header
->tail_entry_seqnum
),
3148 format_timestamp_safe(x
, sizeof(x
), le64toh(f
->header
->head_entry_realtime
)), le64toh(f
->header
->head_entry_realtime
),
3149 format_timestamp_safe(y
, sizeof(y
), le64toh(f
->header
->tail_entry_realtime
)), le64toh(f
->header
->tail_entry_realtime
),
3150 format_timespan(z
, sizeof(z
), le64toh(f
->header
->tail_entry_monotonic
), USEC_PER_MSEC
), le64toh(f
->header
->tail_entry_monotonic
),
3151 le64toh(f
->header
->n_objects
),
3152 le64toh(f
->header
->n_entries
));
3154 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3155 printf("Data Objects: %"PRIu64
"\n"
3156 "Data Hash Table Fill: %.1f%%\n",
3157 le64toh(f
->header
->n_data
),
3158 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))));
3160 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3161 printf("Field Objects: %"PRIu64
"\n"
3162 "Field Hash Table Fill: %.1f%%\n",
3163 le64toh(f
->header
->n_fields
),
3164 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))));
3166 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
))
3167 printf("Tag Objects: %"PRIu64
"\n",
3168 le64toh(f
->header
->n_tags
));
3169 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
3170 printf("Entry Array Objects: %"PRIu64
"\n",
3171 le64toh(f
->header
->n_entry_arrays
));
3173 if (fstat(f
->fd
, &st
) >= 0)
3174 printf("Disk usage: %s\n", format_bytes(bytes
, sizeof(bytes
), (uint64_t) st
.st_blocks
* 512ULL));
3177 static int journal_file_warn_btrfs(JournalFile
*f
) {
3183 /* Before we write anything, check if the COW logic is turned
3184 * off on btrfs. Given our write pattern that is quite
3185 * unfriendly to COW file systems this should greatly improve
3186 * performance on COW file systems, such as btrfs, at the
3187 * expense of data integrity features (which shouldn't be too
3188 * bad, given that we do our own checksumming). */
3190 r
= btrfs_is_filesystem(f
->fd
);
3192 return log_warning_errno(r
, "Failed to determine if journal is on btrfs: %m");
3196 r
= read_attr_fd(f
->fd
, &attrs
);
3198 return log_warning_errno(r
, "Failed to read file attributes: %m");
3200 if (attrs
& FS_NOCOW_FL
) {
3201 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3205 log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3206 "This is likely to slow down journal access substantially, please consider turning "
3207 "off the copy-on-write file attribute on the journal directory, using chattr +C.", f
->path
);
3212 int journal_file_open(
3218 uint64_t compress_threshold_bytes
,
3220 JournalMetrics
*metrics
,
3221 MMapCache
*mmap_cache
,
3222 Set
*deferred_closes
,
3223 JournalFile
*template,
3224 JournalFile
**ret
) {
3226 bool newly_created
= false;
3230 char bytes
[FORMAT_BYTES_MAX
];
3233 assert(fd
>= 0 || fname
);
3235 if (!IN_SET((flags
& O_ACCMODE
), O_RDONLY
, O_RDWR
))
3238 if (fname
&& (flags
& O_CREAT
) && !endswith(fname
, ".journal"))
3241 f
= new0(JournalFile
, 1);
3249 f
->prot
= prot_from_flags(flags
);
3250 f
->writable
= (flags
& O_ACCMODE
) != O_RDONLY
;
3252 f
->compress_lz4
= compress
;
3254 f
->compress_xz
= compress
;
3257 if (compress_threshold_bytes
== (uint64_t) -1)
3258 f
->compress_threshold_bytes
= DEFAULT_COMPRESS_THRESHOLD
;
3260 f
->compress_threshold_bytes
= MAX(MIN_COMPRESS_THRESHOLD
, compress_threshold_bytes
);
3266 log_debug("Journal effective settings seal=%s compress=%s compress_threshold_bytes=%s",
3267 yes_no(f
->seal
), yes_no(JOURNAL_FILE_COMPRESS(f
)),
3268 format_bytes(bytes
, sizeof(bytes
), f
->compress_threshold_bytes
));
3271 f
->mmap
= mmap_cache_ref(mmap_cache
);
3273 f
->mmap
= mmap_cache_new();
3281 f
->path
= strdup(fname
);
3289 /* If we don't know the path, fill in something explanatory and vaguely useful */
3290 if (asprintf(&f
->path
, "/proc/self/%i", fd
) < 0) {
3296 f
->chain_cache
= ordered_hashmap_new(&uint64_hash_ops
);
3297 if (!f
->chain_cache
) {
3303 /* We pass O_NONBLOCK here, so that in case somebody pointed us to some character device node or FIFO
3304 * or so, we likely fail quickly than block for long. For regular files O_NONBLOCK has no effect, hence
3305 * it doesn't hurt in that case. */
3307 f
->fd
= open(f
->path
, f
->flags
|O_CLOEXEC
|O_NONBLOCK
, f
->mode
);
3313 /* fds we opened here by us should also be closed by us. */
3316 r
= fd_nonblock(f
->fd
, false);
3321 f
->cache_fd
= mmap_cache_add_fd(f
->mmap
, f
->fd
);
3327 r
= journal_file_fstat(f
);
3331 if (f
->last_stat
.st_size
== 0 && f
->writable
) {
3333 (void) journal_file_warn_btrfs(f
);
3335 /* Let's attach the creation time to the journal file, so that the vacuuming code knows the age of this
3336 * file even if the file might end up corrupted one day... Ideally we'd just use the creation time many
3337 * file systems maintain for each file, but the API to query this is very new, hence let's emulate this
3338 * via extended attributes. If extended attributes are not supported we'll just skip this, and rely
3339 * solely on mtime/atime/ctime of the file. */
3340 (void) fd_setcrtime(f
->fd
, 0);
3343 /* Try to load the FSPRG state, and if we can't, then
3344 * just don't do sealing */
3346 r
= journal_file_fss_load(f
);
3352 r
= journal_file_init_header(f
, template);
3356 r
= journal_file_fstat(f
);
3360 newly_created
= true;
3363 if (f
->last_stat
.st_size
< (off_t
) HEADER_SIZE_MIN
) {
3368 r
= mmap_cache_get(f
->mmap
, f
->cache_fd
, f
->prot
, CONTEXT_HEADER
, true, 0, PAGE_ALIGN(sizeof(Header
)), &f
->last_stat
, &h
, NULL
);
3374 if (!newly_created
) {
3375 set_clear_with_destructor(deferred_closes
, journal_file_close
);
3377 r
= journal_file_verify_header(f
);
3383 if (!newly_created
&& f
->writable
) {
3384 r
= journal_file_fss_load(f
);
3392 journal_default_metrics(metrics
, f
->fd
);
3393 f
->metrics
= *metrics
;
3394 } else if (template)
3395 f
->metrics
= template->metrics
;
3397 r
= journal_file_refresh_header(f
);
3403 r
= journal_file_hmac_setup(f
);
3408 if (newly_created
) {
3409 r
= journal_file_setup_field_hash_table(f
);
3413 r
= journal_file_setup_data_hash_table(f
);
3418 r
= journal_file_append_first_tag(f
);
3424 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
)) {
3429 if (template && template->post_change_timer
) {
3430 r
= journal_file_enable_post_change_timer(
3432 sd_event_source_get_event(template->post_change_timer
),
3433 template->post_change_timer_period
);
3439 /* The file is opened now successfully, thus we take possession of any passed in fd. */
3446 if (f
->cache_fd
&& mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
3449 (void) journal_file_close(f
);
3454 int journal_file_rotate(JournalFile
**f
, bool compress
, uint64_t compress_threshold_bytes
, bool seal
, Set
*deferred_closes
) {
3455 _cleanup_free_
char *p
= NULL
;
3457 JournalFile
*old_file
, *new_file
= NULL
;
3465 if (!old_file
->writable
)
3468 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
3469 * rotation, since we don't know the actual path, and couldn't rename the file hence. */
3470 if (path_startswith(old_file
->path
, "/proc/self/fd"))
3473 if (!endswith(old_file
->path
, ".journal"))
3476 l
= strlen(old_file
->path
);
3477 r
= asprintf(&p
, "%.*s@" SD_ID128_FORMAT_STR
"-%016"PRIx64
"-%016"PRIx64
".journal",
3478 (int) l
- 8, old_file
->path
,
3479 SD_ID128_FORMAT_VAL(old_file
->header
->seqnum_id
),
3480 le64toh((*f
)->header
->head_entry_seqnum
),
3481 le64toh((*f
)->header
->head_entry_realtime
));
3485 /* Try to rename the file to the archived version. If the file
3486 * already was deleted, we'll get ENOENT, let's ignore that
3488 r
= rename(old_file
->path
, p
);
3489 if (r
< 0 && errno
!= ENOENT
)
3492 /* Sync the rename to disk */
3493 (void) fsync_directory_of_file(old_file
->fd
);
3495 /* Set as archive so offlining commits w/state=STATE_ARCHIVED.
3496 * Previously we would set old_file->header->state to STATE_ARCHIVED directly here,
3497 * but journal_file_set_offline() short-circuits when state != STATE_ONLINE, which
3498 * would result in the rotated journal never getting fsync() called before closing.
3499 * Now we simply queue the archive state by setting an archive bit, leaving the state
3500 * as STATE_ONLINE so proper offlining occurs. */
3501 old_file
->archive
= true;
3503 /* Currently, btrfs is not very good with out write patterns
3504 * and fragments heavily. Let's defrag our journal files when
3505 * we archive them */
3506 old_file
->defrag_on_close
= true;
3508 r
= journal_file_open(-1, old_file
->path
, old_file
->flags
, old_file
->mode
, compress
,
3509 compress_threshold_bytes
, seal
, NULL
, old_file
->mmap
, deferred_closes
,
3510 old_file
, &new_file
);
3512 if (deferred_closes
&&
3513 set_put(deferred_closes
, old_file
) >= 0)
3514 (void) journal_file_set_offline(old_file
, false);
3516 (void) journal_file_close(old_file
);
3522 int journal_file_open_reliably(
3527 uint64_t compress_threshold_bytes
,
3529 JournalMetrics
*metrics
,
3530 MMapCache
*mmap_cache
,
3531 Set
*deferred_closes
,
3532 JournalFile
*template,
3533 JournalFile
**ret
) {
3537 _cleanup_free_
char *p
= NULL
;
3539 r
= journal_file_open(-1, fname
, flags
, mode
, compress
, compress_threshold_bytes
, seal
, metrics
, mmap_cache
,
3540 deferred_closes
, template, ret
);
3542 -EBADMSG
, /* Corrupted */
3543 -ENODATA
, /* Truncated */
3544 -EHOSTDOWN
, /* Other machine */
3545 -EPROTONOSUPPORT
, /* Incompatible feature */
3546 -EBUSY
, /* Unclean shutdown */
3547 -ESHUTDOWN
, /* Already archived */
3548 -EIO
, /* IO error, including SIGBUS on mmap */
3549 -EIDRM
, /* File has been deleted */
3550 -ETXTBSY
)) /* File is from the future */
3553 if ((flags
& O_ACCMODE
) == O_RDONLY
)
3556 if (!(flags
& O_CREAT
))
3559 if (!endswith(fname
, ".journal"))
3562 /* The file is corrupted. Rotate it away and try it again (but only once) */
3565 if (asprintf(&p
, "%.*s@%016"PRIx64
"-%016"PRIx64
".journal~",
3567 now(CLOCK_REALTIME
),
3571 if (rename(fname
, p
) < 0)
3574 /* btrfs doesn't cope well with our write pattern and
3575 * fragments heavily. Let's defrag all files we rotate */
3577 (void) chattr_path(p
, 0, FS_NOCOW_FL
);
3578 (void) btrfs_defrag(p
);
3580 log_warning_errno(r
, "File %s corrupted or uncleanly shut down, renaming and replacing.", fname
);
3582 return journal_file_open(-1, fname
, flags
, mode
, compress
, compress_threshold_bytes
, seal
, metrics
, mmap_cache
,
3583 deferred_closes
, template, ret
);
3586 int journal_file_copy_entry(JournalFile
*from
, JournalFile
*to
, Object
*o
, uint64_t p
) {
3588 uint64_t q
, xor_hash
= 0;
3592 const sd_id128_t
*boot_id
;
3602 ts
.monotonic
= le64toh(o
->entry
.monotonic
);
3603 ts
.realtime
= le64toh(o
->entry
.realtime
);
3604 boot_id
= &o
->entry
.boot_id
;
3606 n
= journal_file_entry_n_items(o
);
3607 /* alloca() can't take 0, hence let's allocate at least one */
3608 items
= newa(EntryItem
, MAX(1u, n
));
3610 for (i
= 0; i
< n
; i
++) {
3617 q
= le64toh(o
->entry
.items
[i
].object_offset
);
3618 le_hash
= o
->entry
.items
[i
].hash
;
3620 r
= journal_file_move_to_object(from
, OBJECT_DATA
, q
, &o
);
3624 if (le_hash
!= o
->data
.hash
)
3627 l
= le64toh(o
->object
.size
) - offsetof(Object
, data
.payload
);
3630 /* We hit the limit on 32bit machines */
3631 if ((uint64_t) t
!= l
)
3634 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
3635 #if HAVE_XZ || HAVE_LZ4
3638 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
3639 o
->data
.payload
, l
, &from
->compress_buffer
, &from
->compress_buffer_size
, &rsize
, 0);
3643 data
= from
->compress_buffer
;
3646 return -EPROTONOSUPPORT
;
3649 data
= o
->data
.payload
;
3651 r
= journal_file_append_data(to
, data
, l
, &u
, &h
);
3655 xor_hash
^= le64toh(u
->data
.hash
);
3656 items
[i
].object_offset
= htole64(h
);
3657 items
[i
].hash
= u
->data
.hash
;
3659 r
= journal_file_move_to_object(from
, OBJECT_ENTRY
, p
, &o
);
3664 r
= journal_file_append_entry_internal(to
, &ts
, boot_id
, xor_hash
, items
, n
,
3667 if (mmap_cache_got_sigbus(to
->mmap
, to
->cache_fd
))
3673 void journal_reset_metrics(JournalMetrics
*m
) {
3676 /* Set everything to "pick automatic values". */
3678 *m
= (JournalMetrics
) {
3679 .min_use
= (uint64_t) -1,
3680 .max_use
= (uint64_t) -1,
3681 .min_size
= (uint64_t) -1,
3682 .max_size
= (uint64_t) -1,
3683 .keep_free
= (uint64_t) -1,
3684 .n_max_files
= (uint64_t) -1,
3688 void journal_default_metrics(JournalMetrics
*m
, int fd
) {
3689 char a
[FORMAT_BYTES_MAX
], b
[FORMAT_BYTES_MAX
], c
[FORMAT_BYTES_MAX
], d
[FORMAT_BYTES_MAX
], e
[FORMAT_BYTES_MAX
];
3696 if (fstatvfs(fd
, &ss
) >= 0)
3697 fs_size
= ss
.f_frsize
* ss
.f_blocks
;
3699 log_debug_errno(errno
, "Failed to determine disk size: %m");
3703 if (m
->max_use
== (uint64_t) -1) {
3706 m
->max_use
= PAGE_ALIGN(fs_size
/ 10); /* 10% of file system size */
3708 if (m
->max_use
> DEFAULT_MAX_USE_UPPER
)
3709 m
->max_use
= DEFAULT_MAX_USE_UPPER
;
3711 if (m
->max_use
< DEFAULT_MAX_USE_LOWER
)
3712 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3714 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3716 m
->max_use
= PAGE_ALIGN(m
->max_use
);
3718 if (m
->max_use
!= 0 && m
->max_use
< JOURNAL_FILE_SIZE_MIN
*2)
3719 m
->max_use
= JOURNAL_FILE_SIZE_MIN
*2;
3722 if (m
->min_use
== (uint64_t) -1)
3723 m
->min_use
= DEFAULT_MIN_USE
;
3725 if (m
->min_use
> m
->max_use
)
3726 m
->min_use
= m
->max_use
;
3728 if (m
->max_size
== (uint64_t) -1) {
3729 m
->max_size
= PAGE_ALIGN(m
->max_use
/ 8); /* 8 chunks */
3731 if (m
->max_size
> DEFAULT_MAX_SIZE_UPPER
)
3732 m
->max_size
= DEFAULT_MAX_SIZE_UPPER
;
3734 m
->max_size
= PAGE_ALIGN(m
->max_size
);
3736 if (m
->max_size
!= 0) {
3737 if (m
->max_size
< JOURNAL_FILE_SIZE_MIN
)
3738 m
->max_size
= JOURNAL_FILE_SIZE_MIN
;
3740 if (m
->max_use
!= 0 && m
->max_size
*2 > m
->max_use
)
3741 m
->max_use
= m
->max_size
*2;
3744 if (m
->min_size
== (uint64_t) -1)
3745 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3747 m
->min_size
= PAGE_ALIGN(m
->min_size
);
3749 if (m
->min_size
< JOURNAL_FILE_SIZE_MIN
)
3750 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3752 if (m
->max_size
!= 0 && m
->min_size
> m
->max_size
)
3753 m
->max_size
= m
->min_size
;
3756 if (m
->keep_free
== (uint64_t) -1) {
3759 m
->keep_free
= PAGE_ALIGN(fs_size
* 3 / 20); /* 15% of file system size */
3761 if (m
->keep_free
> DEFAULT_KEEP_FREE_UPPER
)
3762 m
->keep_free
= DEFAULT_KEEP_FREE_UPPER
;
3765 m
->keep_free
= DEFAULT_KEEP_FREE
;
3768 if (m
->n_max_files
== (uint64_t) -1)
3769 m
->n_max_files
= DEFAULT_N_MAX_FILES
;
3771 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64
,
3772 format_bytes(a
, sizeof(a
), m
->min_use
),
3773 format_bytes(b
, sizeof(b
), m
->max_use
),
3774 format_bytes(c
, sizeof(c
), m
->max_size
),
3775 format_bytes(d
, sizeof(d
), m
->min_size
),
3776 format_bytes(e
, sizeof(e
), m
->keep_free
),
3780 int journal_file_get_cutoff_realtime_usec(JournalFile
*f
, usec_t
*from
, usec_t
*to
) {
3786 if (f
->header
->head_entry_realtime
== 0)
3789 *from
= le64toh(f
->header
->head_entry_realtime
);
3793 if (f
->header
->tail_entry_realtime
== 0)
3796 *to
= le64toh(f
->header
->tail_entry_realtime
);
3802 int journal_file_get_cutoff_monotonic_usec(JournalFile
*f
, sd_id128_t boot_id
, usec_t
*from
, usec_t
*to
) {
3810 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &p
);
3814 if (le64toh(o
->data
.n_entries
) <= 0)
3818 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, le64toh(o
->data
.entry_offset
), &o
);
3822 *from
= le64toh(o
->entry
.monotonic
);
3826 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
3830 r
= generic_array_get_plus_one(f
,
3831 le64toh(o
->data
.entry_offset
),
3832 le64toh(o
->data
.entry_array_offset
),
3833 le64toh(o
->data
.n_entries
)-1,
3838 *to
= le64toh(o
->entry
.monotonic
);
3844 bool journal_file_rotate_suggested(JournalFile
*f
, usec_t max_file_usec
) {
3848 /* If we gained new header fields we gained new features,
3849 * hence suggest a rotation */
3850 if (le64toh(f
->header
->header_size
) < sizeof(Header
)) {
3851 log_debug("%s uses an outdated header, suggesting rotation.", f
->path
);
3855 /* Let's check if the hash tables grew over a certain fill
3856 * level (75%, borrowing this value from Java's hash table
3857 * implementation), and if so suggest a rotation. To calculate
3858 * the fill level we need the n_data field, which only exists
3859 * in newer versions. */
3861 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3862 if (le64toh(f
->header
->n_data
) * 4ULL > (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3863 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.",
3865 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))),
3866 le64toh(f
->header
->n_data
),
3867 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3868 (unsigned long long) f
->last_stat
.st_size
,
3869 f
->last_stat
.st_size
/ le64toh(f
->header
->n_data
));
3873 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3874 if (le64toh(f
->header
->n_fields
) * 4ULL > (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3875 log_debug("Field hash table of %s has a fill level at %.1f (%"PRIu64
" of %"PRIu64
" items), suggesting rotation.",
3877 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))),
3878 le64toh(f
->header
->n_fields
),
3879 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
));
3883 /* Are the data objects properly indexed by field objects? */
3884 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
3885 JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
3886 le64toh(f
->header
->n_data
) > 0 &&
3887 le64toh(f
->header
->n_fields
) == 0)
3890 if (max_file_usec
> 0) {
3893 h
= le64toh(f
->header
->head_entry_realtime
);
3894 t
= now(CLOCK_REALTIME
);
3896 if (h
> 0 && t
> h
+ max_file_usec
)