1 /* SPDX-License-Identifier: LGPL-2.1+ */
3 Copyright 2011 Lennart Poettering
12 #include <sys/statvfs.h>
16 #include "alloc-util.h"
17 #include "btrfs-util.h"
18 #include "chattr-util.h"
22 #include "journal-authenticate.h"
23 #include "journal-def.h"
24 #include "journal-file.h"
26 #include "parse-util.h"
27 #include "path-util.h"
28 #include "random-util.h"
31 #include "stat-util.h"
32 #include "string-util.h"
34 #include "xattr-util.h"
36 #define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
37 #define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
39 #define DEFAULT_COMPRESS_THRESHOLD (512ULL)
40 #define MIN_COMPRESS_THRESHOLD (8ULL)
42 /* This is the minimum journal file size */
43 #define JOURNAL_FILE_SIZE_MIN (512ULL*1024ULL) /* 512 KiB */
45 /* These are the lower and upper bounds if we deduce the max_use value
46 * from the file system size */
47 #define DEFAULT_MAX_USE_LOWER (1ULL*1024ULL*1024ULL) /* 1 MiB */
48 #define DEFAULT_MAX_USE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
50 /* This is the default minimal use limit, how much we'll use even if keep_free suggests otherwise. */
51 #define DEFAULT_MIN_USE (1ULL*1024ULL*1024ULL) /* 1 MiB */
53 /* This is the upper bound if we deduce max_size from max_use */
54 #define DEFAULT_MAX_SIZE_UPPER (128ULL*1024ULL*1024ULL) /* 128 MiB */
56 /* This is the upper bound if we deduce the keep_free value from the
58 #define DEFAULT_KEEP_FREE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
60 /* This is the keep_free value when we can't determine the system
62 #define DEFAULT_KEEP_FREE (1024ULL*1024ULL) /* 1 MB */
64 /* This is the default maximum number of journal files to keep around. */
65 #define DEFAULT_N_MAX_FILES (100)
67 /* n_data was the first entry we added after the initial file format design */
68 #define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
70 /* How many entries to keep in the entry array chain cache at max */
71 #define CHAIN_CACHE_MAX 20
73 /* How much to increase the journal file size at once each time we allocate something new. */
74 #define FILE_SIZE_INCREASE (8ULL*1024ULL*1024ULL) /* 8MB */
76 /* Reread fstat() of the file for detecting deletions at least this often */
77 #define LAST_STAT_REFRESH_USEC (5*USEC_PER_SEC)
79 /* The mmap context to use for the header we pick as one above the last defined typed */
80 #define CONTEXT_HEADER _OBJECT_TYPE_MAX
83 # pragma GCC diagnostic ignored "-Waddress-of-packed-member"
86 /* This may be called from a separate thread to prevent blocking the caller for the duration of fsync().
87 * As a result we use atomic operations on f->offline_state for inter-thread communications with
88 * journal_file_set_offline() and journal_file_set_online(). */
89 static void journal_file_set_offline_internal(JournalFile
*f
) {
95 switch (f
->offline_state
) {
97 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_CANCEL
, OFFLINE_DONE
))
101 case OFFLINE_AGAIN_FROM_SYNCING
:
102 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_SYNCING
, OFFLINE_SYNCING
))
106 case OFFLINE_AGAIN_FROM_OFFLINING
:
107 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_OFFLINING
, OFFLINE_SYNCING
))
111 case OFFLINE_SYNCING
:
114 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_OFFLINING
))
117 f
->header
->state
= f
->archive
? STATE_ARCHIVED
: STATE_OFFLINE
;
121 case OFFLINE_OFFLINING
:
122 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_OFFLINING
, OFFLINE_DONE
))
129 log_debug("OFFLINE_JOINED unexpected offline state for journal_file_set_offline_internal()");
135 static void * journal_file_set_offline_thread(void *arg
) {
136 JournalFile
*f
= arg
;
138 (void) pthread_setname_np(pthread_self(), "journal-offline");
140 journal_file_set_offline_internal(f
);
145 static int journal_file_set_offline_thread_join(JournalFile
*f
) {
150 if (f
->offline_state
== OFFLINE_JOINED
)
153 r
= pthread_join(f
->offline_thread
, NULL
);
157 f
->offline_state
= OFFLINE_JOINED
;
159 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
165 /* Trigger a restart if the offline thread is mid-flight in a restartable state. */
166 static bool journal_file_set_offline_try_restart(JournalFile
*f
) {
168 switch (f
->offline_state
) {
169 case OFFLINE_AGAIN_FROM_SYNCING
:
170 case OFFLINE_AGAIN_FROM_OFFLINING
:
174 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_CANCEL
, OFFLINE_AGAIN_FROM_SYNCING
))
178 case OFFLINE_SYNCING
:
179 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_AGAIN_FROM_SYNCING
))
183 case OFFLINE_OFFLINING
:
184 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_OFFLINING
, OFFLINE_AGAIN_FROM_OFFLINING
))
194 /* Sets a journal offline.
196 * If wait is false then an offline is dispatched in a separate thread for a
197 * subsequent journal_file_set_offline() or journal_file_set_online() of the
198 * same journal to synchronize with.
200 * If wait is true, then either an existing offline thread will be restarted
201 * and joined, or if none exists the offline is simply performed in this
202 * context without involving another thread.
204 int journal_file_set_offline(JournalFile
*f
, bool wait
) {
213 if (f
->fd
< 0 || !f
->header
)
216 /* An offlining journal is implicitly online and may modify f->header->state,
217 * we must also join any potentially lingering offline thread when not online. */
218 if (!journal_file_is_offlining(f
) && f
->header
->state
!= STATE_ONLINE
)
219 return journal_file_set_offline_thread_join(f
);
221 /* Restart an in-flight offline thread and wait if needed, or join a lingering done one. */
222 restarted
= journal_file_set_offline_try_restart(f
);
223 if ((restarted
&& wait
) || !restarted
) {
224 r
= journal_file_set_offline_thread_join(f
);
232 /* Initiate a new offline. */
233 f
->offline_state
= OFFLINE_SYNCING
;
235 if (wait
) /* Without using a thread if waiting. */
236 journal_file_set_offline_internal(f
);
238 sigset_t ss
, saved_ss
;
241 if (sigfillset(&ss
) < 0)
244 r
= pthread_sigmask(SIG_BLOCK
, &ss
, &saved_ss
);
248 r
= pthread_create(&f
->offline_thread
, NULL
, journal_file_set_offline_thread
, f
);
250 k
= pthread_sigmask(SIG_SETMASK
, &saved_ss
, NULL
);
252 f
->offline_state
= OFFLINE_JOINED
;
262 static int journal_file_set_online(JournalFile
*f
) {
270 if (f
->fd
< 0 || !f
->header
)
274 switch (f
->offline_state
) {
276 /* No offline thread, no need to wait. */
280 case OFFLINE_SYNCING
:
281 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_CANCEL
))
283 /* Canceled syncing prior to offlining, no need to wait. */
287 case OFFLINE_AGAIN_FROM_SYNCING
:
288 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_SYNCING
, OFFLINE_CANCEL
))
290 /* Canceled restart from syncing, no need to wait. */
294 case OFFLINE_AGAIN_FROM_OFFLINING
:
295 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_OFFLINING
, OFFLINE_CANCEL
))
297 /* Canceled restart from offlining, must wait for offlining to complete however. */
302 r
= journal_file_set_offline_thread_join(f
);
312 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
315 switch (f
->header
->state
) {
320 f
->header
->state
= STATE_ONLINE
;
329 bool journal_file_is_offlining(JournalFile
*f
) {
332 __sync_synchronize();
334 if (IN_SET(f
->offline_state
, OFFLINE_DONE
, OFFLINE_JOINED
))
340 JournalFile
* journal_file_close(JournalFile
*f
) {
344 /* Write the final tag */
345 if (f
->seal
&& f
->writable
) {
348 r
= journal_file_append_tag(f
);
350 log_error_errno(r
, "Failed to append tag when closing journal: %m");
354 if (f
->post_change_timer
) {
357 if (sd_event_source_get_enabled(f
->post_change_timer
, &enabled
) >= 0)
358 if (enabled
== SD_EVENT_ONESHOT
)
359 journal_file_post_change(f
);
361 (void) sd_event_source_set_enabled(f
->post_change_timer
, SD_EVENT_OFF
);
362 sd_event_source_unref(f
->post_change_timer
);
365 journal_file_set_offline(f
, true);
367 if (f
->mmap
&& f
->cache_fd
)
368 mmap_cache_free_fd(f
->mmap
, f
->cache_fd
);
370 if (f
->fd
>= 0 && f
->defrag_on_close
) {
372 /* Be friendly to btrfs: turn COW back on again now,
373 * and defragment the file. We won't write to the file
374 * ever again, hence remove all fragmentation, and
375 * reenable all the good bits COW usually provides
376 * (such as data checksumming). */
378 (void) chattr_fd(f
->fd
, 0, FS_NOCOW_FL
);
379 (void) btrfs_defrag_fd(f
->fd
);
386 mmap_cache_unref(f
->mmap
);
388 ordered_hashmap_free_free(f
->chain_cache
);
390 #if HAVE_XZ || HAVE_LZ4
391 free(f
->compress_buffer
);
396 munmap(f
->fss_file
, PAGE_ALIGN(f
->fss_file_size
));
398 free(f
->fsprg_state
);
403 gcry_md_close(f
->hmac
);
409 static int journal_file_init_header(JournalFile
*f
, JournalFile
*template) {
416 memcpy(h
.signature
, HEADER_SIGNATURE
, 8);
417 h
.header_size
= htole64(ALIGN64(sizeof(h
)));
419 h
.incompatible_flags
|= htole32(
420 f
->compress_xz
* HEADER_INCOMPATIBLE_COMPRESSED_XZ
|
421 f
->compress_lz4
* HEADER_INCOMPATIBLE_COMPRESSED_LZ4
);
423 h
.compatible_flags
= htole32(
424 f
->seal
* HEADER_COMPATIBLE_SEALED
);
426 r
= sd_id128_randomize(&h
.file_id
);
431 h
.seqnum_id
= template->header
->seqnum_id
;
432 h
.tail_entry_seqnum
= template->header
->tail_entry_seqnum
;
434 h
.seqnum_id
= h
.file_id
;
436 k
= pwrite(f
->fd
, &h
, sizeof(h
), 0);
446 static int journal_file_refresh_header(JournalFile
*f
) {
453 r
= sd_id128_get_machine(&f
->header
->machine_id
);
454 if (IN_SET(r
, -ENOENT
, -ENOMEDIUM
))
455 /* We don't have a machine-id, let's continue without */
456 zero(f
->header
->machine_id
);
460 r
= sd_id128_get_boot(&boot_id
);
464 f
->header
->boot_id
= boot_id
;
466 r
= journal_file_set_online(f
);
468 /* Sync the online state to disk */
471 /* We likely just created a new file, also sync the directory this file is located in. */
472 (void) fsync_directory_of_file(f
->fd
);
477 static bool warn_wrong_flags(const JournalFile
*f
, bool compatible
) {
478 const uint32_t any
= compatible
? HEADER_COMPATIBLE_ANY
: HEADER_INCOMPATIBLE_ANY
,
479 supported
= compatible
? HEADER_COMPATIBLE_SUPPORTED
: HEADER_INCOMPATIBLE_SUPPORTED
;
480 const char *type
= compatible
? "compatible" : "incompatible";
483 flags
= le32toh(compatible
? f
->header
->compatible_flags
: f
->header
->incompatible_flags
);
485 if (flags
& ~supported
) {
487 log_debug("Journal file %s has unknown %s flags 0x%"PRIx32
,
488 f
->path
, type
, flags
& ~any
);
489 flags
= (flags
& any
) & ~supported
;
493 _cleanup_free_
char *t
= NULL
;
495 if (compatible
&& (flags
& HEADER_COMPATIBLE_SEALED
))
496 strv
[n
++] = "sealed";
497 if (!compatible
&& (flags
& HEADER_INCOMPATIBLE_COMPRESSED_XZ
))
498 strv
[n
++] = "xz-compressed";
499 if (!compatible
&& (flags
& HEADER_INCOMPATIBLE_COMPRESSED_LZ4
))
500 strv
[n
++] = "lz4-compressed";
502 assert(n
< ELEMENTSOF(strv
));
504 t
= strv_join((char**) strv
, ", ");
505 log_debug("Journal file %s uses %s %s %s disabled at compilation time.",
506 f
->path
, type
, n
> 1 ? "flags" : "flag", strnull(t
));
514 static int journal_file_verify_header(JournalFile
*f
) {
515 uint64_t arena_size
, header_size
;
520 if (memcmp(f
->header
->signature
, HEADER_SIGNATURE
, 8))
523 /* In both read and write mode we refuse to open files with incompatible
524 * flags we don't know. */
525 if (warn_wrong_flags(f
, false))
526 return -EPROTONOSUPPORT
;
528 /* When open for writing we refuse to open files with compatible flags, too. */
529 if (f
->writable
&& warn_wrong_flags(f
, true))
530 return -EPROTONOSUPPORT
;
532 if (f
->header
->state
>= _STATE_MAX
)
535 header_size
= le64toh(f
->header
->header_size
);
537 /* The first addition was n_data, so check that we are at least this large */
538 if (header_size
< HEADER_SIZE_MIN
)
541 if (JOURNAL_HEADER_SEALED(f
->header
) && !JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
544 arena_size
= le64toh(f
->header
->arena_size
);
546 if (UINT64_MAX
- header_size
< arena_size
|| header_size
+ arena_size
> (uint64_t) f
->last_stat
.st_size
)
549 if (le64toh(f
->header
->tail_object_offset
) > header_size
+ arena_size
)
552 if (!VALID64(le64toh(f
->header
->data_hash_table_offset
)) ||
553 !VALID64(le64toh(f
->header
->field_hash_table_offset
)) ||
554 !VALID64(le64toh(f
->header
->tail_object_offset
)) ||
555 !VALID64(le64toh(f
->header
->entry_array_offset
)))
559 sd_id128_t machine_id
;
563 r
= sd_id128_get_machine(&machine_id
);
567 if (!sd_id128_equal(machine_id
, f
->header
->machine_id
))
570 state
= f
->header
->state
;
572 if (state
== STATE_ARCHIVED
)
573 return -ESHUTDOWN
; /* Already archived */
574 else if (state
== STATE_ONLINE
) {
575 log_debug("Journal file %s is already online. Assuming unclean closing.", f
->path
);
577 } else if (state
!= STATE_OFFLINE
) {
578 log_debug("Journal file %s has unknown state %i.", f
->path
, state
);
582 if (f
->header
->field_hash_table_size
== 0 || f
->header
->data_hash_table_size
== 0)
585 /* Don't permit appending to files from the future. Because otherwise the realtime timestamps wouldn't
586 * be strictly ordered in the entries in the file anymore, and we can't have that since it breaks
588 if (le64toh(f
->header
->tail_entry_realtime
) > now(CLOCK_REALTIME
)) {
589 log_debug("Journal file %s is from the future, refusing to append new data to it that'd be older.", f
->path
);
594 f
->compress_xz
= JOURNAL_HEADER_COMPRESSED_XZ(f
->header
);
595 f
->compress_lz4
= JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
);
597 f
->seal
= JOURNAL_HEADER_SEALED(f
->header
);
602 static int journal_file_fstat(JournalFile
*f
) {
608 if (fstat(f
->fd
, &f
->last_stat
) < 0)
611 f
->last_stat_usec
= now(CLOCK_MONOTONIC
);
613 /* Refuse dealing with with files that aren't regular */
614 r
= stat_verify_regular(&f
->last_stat
);
618 /* Refuse appending to files that are already deleted */
619 if (f
->last_stat
.st_nlink
<= 0)
625 static int journal_file_allocate(JournalFile
*f
, uint64_t offset
, uint64_t size
) {
626 uint64_t old_size
, new_size
;
632 /* We assume that this file is not sparse, and we know that
633 * for sure, since we always call posix_fallocate()
636 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
640 le64toh(f
->header
->header_size
) +
641 le64toh(f
->header
->arena_size
);
643 new_size
= PAGE_ALIGN(offset
+ size
);
644 if (new_size
< le64toh(f
->header
->header_size
))
645 new_size
= le64toh(f
->header
->header_size
);
647 if (new_size
<= old_size
) {
649 /* We already pre-allocated enough space, but before
650 * we write to it, let's check with fstat() if the
651 * file got deleted, in order make sure we don't throw
652 * away the data immediately. Don't check fstat() for
653 * all writes though, but only once ever 10s. */
655 if (f
->last_stat_usec
+ LAST_STAT_REFRESH_USEC
> now(CLOCK_MONOTONIC
))
658 return journal_file_fstat(f
);
661 /* Allocate more space. */
663 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
666 if (new_size
> f
->metrics
.min_size
&& f
->metrics
.keep_free
> 0) {
669 if (fstatvfs(f
->fd
, &svfs
) >= 0) {
672 available
= LESS_BY((uint64_t) svfs
.f_bfree
* (uint64_t) svfs
.f_bsize
, f
->metrics
.keep_free
);
674 if (new_size
- old_size
> available
)
679 /* Increase by larger blocks at once */
680 new_size
= DIV_ROUND_UP(new_size
, FILE_SIZE_INCREASE
) * FILE_SIZE_INCREASE
;
681 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
682 new_size
= f
->metrics
.max_size
;
684 /* Note that the glibc fallocate() fallback is very
685 inefficient, hence we try to minimize the allocation area
687 r
= posix_fallocate(f
->fd
, old_size
, new_size
- old_size
);
691 f
->header
->arena_size
= htole64(new_size
- le64toh(f
->header
->header_size
));
693 return journal_file_fstat(f
);
696 static unsigned type_to_context(ObjectType type
) {
697 /* One context for each type, plus one catch-all for the rest */
698 assert_cc(_OBJECT_TYPE_MAX
<= MMAP_CACHE_MAX_CONTEXTS
);
699 assert_cc(CONTEXT_HEADER
< MMAP_CACHE_MAX_CONTEXTS
);
700 return type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
? type
: 0;
703 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
) {
712 /* Avoid SIGBUS on invalid accesses */
713 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
) {
714 /* Hmm, out of range? Let's refresh the fstat() data
715 * first, before we trust that check. */
717 r
= journal_file_fstat(f
);
721 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
)
722 return -EADDRNOTAVAIL
;
725 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
);
728 static uint64_t minimum_header_size(Object
*o
) {
730 static const uint64_t table
[] = {
731 [OBJECT_DATA
] = sizeof(DataObject
),
732 [OBJECT_FIELD
] = sizeof(FieldObject
),
733 [OBJECT_ENTRY
] = sizeof(EntryObject
),
734 [OBJECT_DATA_HASH_TABLE
] = sizeof(HashTableObject
),
735 [OBJECT_FIELD_HASH_TABLE
] = sizeof(HashTableObject
),
736 [OBJECT_ENTRY_ARRAY
] = sizeof(EntryArrayObject
),
737 [OBJECT_TAG
] = sizeof(TagObject
),
740 if (o
->object
.type
>= ELEMENTSOF(table
) || table
[o
->object
.type
] <= 0)
741 return sizeof(ObjectHeader
);
743 return table
[o
->object
.type
];
746 /* Lightweight object checks. We want this to be fast, so that we won't
747 * slowdown every journal_file_move_to_object() call too much. */
748 static int journal_file_check_object(JournalFile
*f
, uint64_t offset
, Object
*o
) {
752 switch (o
->object
.type
) {
755 if ((le64toh(o
->data
.entry_offset
) == 0) ^ (le64toh(o
->data
.n_entries
) == 0)) {
756 log_debug("Bad n_entries: %"PRIu64
": %"PRIu64
,
757 le64toh(o
->data
.n_entries
), offset
);
761 if (le64toh(o
->object
.size
) - offsetof(DataObject
, payload
) <= 0) {
762 log_debug("Bad object size (<= %zu): %"PRIu64
": %"PRIu64
,
763 offsetof(DataObject
, payload
),
764 le64toh(o
->object
.size
),
769 if (!VALID64(le64toh(o
->data
.next_hash_offset
)) ||
770 !VALID64(le64toh(o
->data
.next_field_offset
)) ||
771 !VALID64(le64toh(o
->data
.entry_offset
)) ||
772 !VALID64(le64toh(o
->data
.entry_array_offset
))) {
773 log_debug("Invalid offset, next_hash_offset="OFSfmt
", next_field_offset="OFSfmt
774 ", entry_offset="OFSfmt
", entry_array_offset="OFSfmt
": %"PRIu64
,
775 le64toh(o
->data
.next_hash_offset
),
776 le64toh(o
->data
.next_field_offset
),
777 le64toh(o
->data
.entry_offset
),
778 le64toh(o
->data
.entry_array_offset
),
787 if (le64toh(o
->object
.size
) - offsetof(FieldObject
, payload
) <= 0) {
789 "Bad field size (<= %zu): %"PRIu64
": %"PRIu64
,
790 offsetof(FieldObject
, payload
),
791 le64toh(o
->object
.size
),
796 if (!VALID64(le64toh(o
->field
.next_hash_offset
)) ||
797 !VALID64(le64toh(o
->field
.head_data_offset
))) {
799 "Invalid offset, next_hash_offset="OFSfmt
800 ", head_data_offset="OFSfmt
": %"PRIu64
,
801 le64toh(o
->field
.next_hash_offset
),
802 le64toh(o
->field
.head_data_offset
),
809 if ((le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) % sizeof(EntryItem
) != 0) {
811 "Bad entry size (<= %zu): %"PRIu64
": %"PRIu64
,
812 offsetof(EntryObject
, items
),
813 le64toh(o
->object
.size
),
818 if ((le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) / sizeof(EntryItem
) <= 0) {
820 "Invalid number items in entry: %"PRIu64
": %"PRIu64
,
821 (le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) / sizeof(EntryItem
),
826 if (le64toh(o
->entry
.seqnum
) <= 0) {
828 "Invalid entry seqnum: %"PRIx64
": %"PRIu64
,
829 le64toh(o
->entry
.seqnum
),
834 if (!VALID_REALTIME(le64toh(o
->entry
.realtime
))) {
836 "Invalid entry realtime timestamp: %"PRIu64
": %"PRIu64
,
837 le64toh(o
->entry
.realtime
),
842 if (!VALID_MONOTONIC(le64toh(o
->entry
.monotonic
))) {
844 "Invalid entry monotonic timestamp: %"PRIu64
": %"PRIu64
,
845 le64toh(o
->entry
.monotonic
),
852 case OBJECT_DATA_HASH_TABLE
:
853 case OBJECT_FIELD_HASH_TABLE
:
854 if ((le64toh(o
->object
.size
) - offsetof(HashTableObject
, items
)) % sizeof(HashItem
) != 0 ||
855 (le64toh(o
->object
.size
) - offsetof(HashTableObject
, items
)) / sizeof(HashItem
) <= 0) {
857 "Invalid %s hash table size: %"PRIu64
": %"PRIu64
,
858 o
->object
.type
== OBJECT_DATA_HASH_TABLE
? "data" : "field",
859 le64toh(o
->object
.size
),
866 case OBJECT_ENTRY_ARRAY
:
867 if ((le64toh(o
->object
.size
) - offsetof(EntryArrayObject
, items
)) % sizeof(le64_t
) != 0 ||
868 (le64toh(o
->object
.size
) - offsetof(EntryArrayObject
, items
)) / sizeof(le64_t
) <= 0) {
870 "Invalid object entry array size: %"PRIu64
": %"PRIu64
,
871 le64toh(o
->object
.size
),
876 if (!VALID64(le64toh(o
->entry_array
.next_entry_array_offset
))) {
878 "Invalid object entry array next_entry_array_offset: "OFSfmt
": %"PRIu64
,
879 le64toh(o
->entry_array
.next_entry_array_offset
),
887 if (le64toh(o
->object
.size
) != sizeof(TagObject
)) {
889 "Invalid object tag size: %"PRIu64
": %"PRIu64
,
890 le64toh(o
->object
.size
),
895 if (!VALID_EPOCH(le64toh(o
->tag
.epoch
))) {
897 "Invalid object tag epoch: %"PRIu64
": %"PRIu64
,
898 le64toh(o
->tag
.epoch
),
909 int journal_file_move_to_object(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
**ret
) {
919 /* Objects may only be located at multiple of 64 bit */
920 if (!VALID64(offset
)) {
921 log_debug("Attempt to move to object at non-64bit boundary: %" PRIu64
, offset
);
925 /* Object may not be located in the file header */
926 if (offset
< le64toh(f
->header
->header_size
)) {
927 log_debug("Attempt to move to object located in file header: %" PRIu64
, offset
);
931 r
= journal_file_move_to(f
, type
, false, offset
, sizeof(ObjectHeader
), &t
, &tsize
);
936 s
= le64toh(o
->object
.size
);
939 log_debug("Attempt to move to uninitialized object: %" PRIu64
, offset
);
942 if (s
< sizeof(ObjectHeader
)) {
943 log_debug("Attempt to move to overly short object: %" PRIu64
, offset
);
947 if (o
->object
.type
<= OBJECT_UNUSED
) {
948 log_debug("Attempt to move to object with invalid type: %" PRIu64
, offset
);
952 if (s
< minimum_header_size(o
)) {
953 log_debug("Attempt to move to truncated object: %" PRIu64
, offset
);
957 if (type
> OBJECT_UNUSED
&& o
->object
.type
!= type
) {
958 log_debug("Attempt to move to object of unexpected type: %" PRIu64
, offset
);
963 r
= journal_file_move_to(f
, type
, false, offset
, s
, &t
, NULL
);
970 r
= journal_file_check_object(f
, offset
, o
);
978 static uint64_t journal_file_entry_seqnum(JournalFile
*f
, uint64_t *seqnum
) {
984 r
= le64toh(f
->header
->tail_entry_seqnum
) + 1;
987 /* If an external seqnum counter was passed, we update
988 * both the local and the external one, and set it to
989 * the maximum of both */
997 f
->header
->tail_entry_seqnum
= htole64(r
);
999 if (f
->header
->head_entry_seqnum
== 0)
1000 f
->header
->head_entry_seqnum
= htole64(r
);
1005 int journal_file_append_object(JournalFile
*f
, ObjectType type
, uint64_t size
, Object
**ret
, uint64_t *offset
) {
1013 assert(type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
);
1014 assert(size
>= sizeof(ObjectHeader
));
1018 r
= journal_file_set_online(f
);
1022 p
= le64toh(f
->header
->tail_object_offset
);
1024 p
= le64toh(f
->header
->header_size
);
1026 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &tail
);
1030 p
+= ALIGN64(le64toh(tail
->object
.size
));
1033 r
= journal_file_allocate(f
, p
, size
);
1037 r
= journal_file_move_to(f
, type
, false, p
, size
, &t
, NULL
);
1044 o
->object
.type
= type
;
1045 o
->object
.size
= htole64(size
);
1047 f
->header
->tail_object_offset
= htole64(p
);
1048 f
->header
->n_objects
= htole64(le64toh(f
->header
->n_objects
) + 1);
1056 static int journal_file_setup_data_hash_table(JournalFile
*f
) {
1064 /* We estimate that we need 1 hash table entry per 768 bytes
1065 of journal file and we want to make sure we never get
1066 beyond 75% fill level. Calculate the hash table size for
1067 the maximum file size based on these metrics. */
1069 s
= (f
->metrics
.max_size
* 4 / 768 / 3) * sizeof(HashItem
);
1070 if (s
< DEFAULT_DATA_HASH_TABLE_SIZE
)
1071 s
= DEFAULT_DATA_HASH_TABLE_SIZE
;
1073 log_debug("Reserving %"PRIu64
" entries in hash table.", s
/ sizeof(HashItem
));
1075 r
= journal_file_append_object(f
,
1076 OBJECT_DATA_HASH_TABLE
,
1077 offsetof(Object
, hash_table
.items
) + s
,
1082 memzero(o
->hash_table
.items
, s
);
1084 f
->header
->data_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1085 f
->header
->data_hash_table_size
= htole64(s
);
1090 static int journal_file_setup_field_hash_table(JournalFile
*f
) {
1098 /* We use a fixed size hash table for the fields as this
1099 * number should grow very slowly only */
1101 s
= DEFAULT_FIELD_HASH_TABLE_SIZE
;
1102 r
= journal_file_append_object(f
,
1103 OBJECT_FIELD_HASH_TABLE
,
1104 offsetof(Object
, hash_table
.items
) + s
,
1109 memzero(o
->hash_table
.items
, s
);
1111 f
->header
->field_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1112 f
->header
->field_hash_table_size
= htole64(s
);
1117 int journal_file_map_data_hash_table(JournalFile
*f
) {
1125 if (f
->data_hash_table
)
1128 p
= le64toh(f
->header
->data_hash_table_offset
);
1129 s
= le64toh(f
->header
->data_hash_table_size
);
1131 r
= journal_file_move_to(f
,
1132 OBJECT_DATA_HASH_TABLE
,
1139 f
->data_hash_table
= t
;
1143 int journal_file_map_field_hash_table(JournalFile
*f
) {
1151 if (f
->field_hash_table
)
1154 p
= le64toh(f
->header
->field_hash_table_offset
);
1155 s
= le64toh(f
->header
->field_hash_table_size
);
1157 r
= journal_file_move_to(f
,
1158 OBJECT_FIELD_HASH_TABLE
,
1165 f
->field_hash_table
= t
;
1169 static int journal_file_link_field(
1180 assert(f
->field_hash_table
);
1184 if (o
->object
.type
!= OBJECT_FIELD
)
1187 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1191 /* This might alter the window we are looking at */
1192 o
->field
.next_hash_offset
= o
->field
.head_data_offset
= 0;
1195 p
= le64toh(f
->field_hash_table
[h
].tail_hash_offset
);
1197 f
->field_hash_table
[h
].head_hash_offset
= htole64(offset
);
1199 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1203 o
->field
.next_hash_offset
= htole64(offset
);
1206 f
->field_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1208 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
1209 f
->header
->n_fields
= htole64(le64toh(f
->header
->n_fields
) + 1);
1214 static int journal_file_link_data(
1225 assert(f
->data_hash_table
);
1229 if (o
->object
.type
!= OBJECT_DATA
)
1232 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1236 /* This might alter the window we are looking at */
1237 o
->data
.next_hash_offset
= o
->data
.next_field_offset
= 0;
1238 o
->data
.entry_offset
= o
->data
.entry_array_offset
= 0;
1239 o
->data
.n_entries
= 0;
1242 p
= le64toh(f
->data_hash_table
[h
].tail_hash_offset
);
1244 /* Only entry in the hash table is easy */
1245 f
->data_hash_table
[h
].head_hash_offset
= htole64(offset
);
1247 /* Move back to the previous data object, to patch in
1250 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1254 o
->data
.next_hash_offset
= htole64(offset
);
1257 f
->data_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1259 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
1260 f
->header
->n_data
= htole64(le64toh(f
->header
->n_data
) + 1);
1265 int journal_file_find_field_object_with_hash(
1267 const void *field
, uint64_t size
, uint64_t hash
,
1268 Object
**ret
, uint64_t *offset
) {
1270 uint64_t p
, osize
, h
, m
;
1275 assert(field
&& size
> 0);
1277 /* If the field hash table is empty, we can't find anything */
1278 if (le64toh(f
->header
->field_hash_table_size
) <= 0)
1281 /* Map the field hash table, if it isn't mapped yet. */
1282 r
= journal_file_map_field_hash_table(f
);
1286 osize
= offsetof(Object
, field
.payload
) + size
;
1288 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1293 p
= le64toh(f
->field_hash_table
[h
].head_hash_offset
);
1298 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1302 if (le64toh(o
->field
.hash
) == hash
&&
1303 le64toh(o
->object
.size
) == osize
&&
1304 memcmp(o
->field
.payload
, field
, size
) == 0) {
1314 p
= le64toh(o
->field
.next_hash_offset
);
1320 int journal_file_find_field_object(
1322 const void *field
, uint64_t size
,
1323 Object
**ret
, uint64_t *offset
) {
1328 assert(field
&& size
> 0);
1330 hash
= hash64(field
, size
);
1332 return journal_file_find_field_object_with_hash(f
,
1337 int journal_file_find_data_object_with_hash(
1339 const void *data
, uint64_t size
, uint64_t hash
,
1340 Object
**ret
, uint64_t *offset
) {
1342 uint64_t p
, osize
, h
, m
;
1347 assert(data
|| size
== 0);
1349 /* If there's no data hash table, then there's no entry. */
1350 if (le64toh(f
->header
->data_hash_table_size
) <= 0)
1353 /* Map the data hash table, if it isn't mapped yet. */
1354 r
= journal_file_map_data_hash_table(f
);
1358 osize
= offsetof(Object
, data
.payload
) + size
;
1360 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1365 p
= le64toh(f
->data_hash_table
[h
].head_hash_offset
);
1370 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1374 if (le64toh(o
->data
.hash
) != hash
)
1377 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
1378 #if HAVE_XZ || HAVE_LZ4
1382 l
= le64toh(o
->object
.size
);
1383 if (l
<= offsetof(Object
, data
.payload
))
1386 l
-= offsetof(Object
, data
.payload
);
1388 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
1389 o
->data
.payload
, l
, &f
->compress_buffer
, &f
->compress_buffer_size
, &rsize
, 0);
1393 if (rsize
== size
&&
1394 memcmp(f
->compress_buffer
, data
, size
) == 0) {
1405 return -EPROTONOSUPPORT
;
1407 } else if (le64toh(o
->object
.size
) == osize
&&
1408 memcmp(o
->data
.payload
, data
, size
) == 0) {
1420 p
= le64toh(o
->data
.next_hash_offset
);
1426 int journal_file_find_data_object(
1428 const void *data
, uint64_t size
,
1429 Object
**ret
, uint64_t *offset
) {
1434 assert(data
|| size
== 0);
1436 hash
= hash64(data
, size
);
1438 return journal_file_find_data_object_with_hash(f
,
1443 static int journal_file_append_field(
1445 const void *field
, uint64_t size
,
1446 Object
**ret
, uint64_t *offset
) {
1454 assert(field
&& size
> 0);
1456 hash
= hash64(field
, size
);
1458 r
= journal_file_find_field_object_with_hash(f
, field
, size
, hash
, &o
, &p
);
1472 osize
= offsetof(Object
, field
.payload
) + size
;
1473 r
= journal_file_append_object(f
, OBJECT_FIELD
, osize
, &o
, &p
);
1477 o
->field
.hash
= htole64(hash
);
1478 memcpy(o
->field
.payload
, field
, size
);
1480 r
= journal_file_link_field(f
, o
, p
, hash
);
1484 /* The linking might have altered the window, so let's
1485 * refresh our pointer */
1486 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1491 r
= journal_file_hmac_put_object(f
, OBJECT_FIELD
, o
, p
);
1505 static int journal_file_append_data(
1507 const void *data
, uint64_t size
,
1508 Object
**ret
, uint64_t *offset
) {
1513 int r
, compression
= 0;
1517 assert(data
|| size
== 0);
1519 hash
= hash64(data
, size
);
1521 r
= journal_file_find_data_object_with_hash(f
, data
, size
, hash
, &o
, &p
);
1535 osize
= offsetof(Object
, data
.payload
) + size
;
1536 r
= journal_file_append_object(f
, OBJECT_DATA
, osize
, &o
, &p
);
1540 o
->data
.hash
= htole64(hash
);
1542 #if HAVE_XZ || HAVE_LZ4
1543 if (JOURNAL_FILE_COMPRESS(f
) && size
>= f
->compress_threshold_bytes
) {
1546 compression
= compress_blob(data
, size
, o
->data
.payload
, size
- 1, &rsize
);
1548 if (compression
>= 0) {
1549 o
->object
.size
= htole64(offsetof(Object
, data
.payload
) + rsize
);
1550 o
->object
.flags
|= compression
;
1552 log_debug("Compressed data object %"PRIu64
" -> %zu using %s",
1553 size
, rsize
, object_compressed_to_string(compression
));
1555 /* Compression didn't work, we don't really care why, let's continue without compression */
1560 if (compression
== 0)
1561 memcpy_safe(o
->data
.payload
, data
, size
);
1563 r
= journal_file_link_data(f
, o
, p
, hash
);
1568 r
= journal_file_hmac_put_object(f
, OBJECT_DATA
, o
, p
);
1573 /* The linking might have altered the window, so let's
1574 * refresh our pointer */
1575 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1582 eq
= memchr(data
, '=', size
);
1583 if (eq
&& eq
> data
) {
1587 /* Create field object ... */
1588 r
= journal_file_append_field(f
, data
, (uint8_t*) eq
- (uint8_t*) data
, &fo
, &fp
);
1592 /* ... and link it in. */
1593 o
->data
.next_field_offset
= fo
->field
.head_data_offset
;
1594 fo
->field
.head_data_offset
= le64toh(p
);
1606 uint64_t journal_file_entry_n_items(Object
*o
) {
1609 if (o
->object
.type
!= OBJECT_ENTRY
)
1612 return (le64toh(o
->object
.size
) - offsetof(Object
, entry
.items
)) / sizeof(EntryItem
);
1615 uint64_t journal_file_entry_array_n_items(Object
*o
) {
1618 if (o
->object
.type
!= OBJECT_ENTRY_ARRAY
)
1621 return (le64toh(o
->object
.size
) - offsetof(Object
, entry_array
.items
)) / sizeof(uint64_t);
1624 uint64_t journal_file_hash_table_n_items(Object
*o
) {
1627 if (!IN_SET(o
->object
.type
, OBJECT_DATA_HASH_TABLE
, OBJECT_FIELD_HASH_TABLE
))
1630 return (le64toh(o
->object
.size
) - offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
);
1633 static int link_entry_into_array(JournalFile
*f
,
1638 uint64_t n
= 0, ap
= 0, q
, i
, a
, hidx
;
1647 a
= le64toh(*first
);
1648 i
= hidx
= le64toh(*idx
);
1651 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
1655 n
= journal_file_entry_array_n_items(o
);
1657 o
->entry_array
.items
[i
] = htole64(p
);
1658 *idx
= htole64(hidx
+ 1);
1664 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
1675 r
= journal_file_append_object(f
, OBJECT_ENTRY_ARRAY
,
1676 offsetof(Object
, entry_array
.items
) + n
* sizeof(uint64_t),
1682 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY_ARRAY
, o
, q
);
1687 o
->entry_array
.items
[i
] = htole64(p
);
1690 *first
= htole64(q
);
1692 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, ap
, &o
);
1696 o
->entry_array
.next_entry_array_offset
= htole64(q
);
1699 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
1700 f
->header
->n_entry_arrays
= htole64(le64toh(f
->header
->n_entry_arrays
) + 1);
1702 *idx
= htole64(hidx
+ 1);
1707 static int link_entry_into_array_plus_one(JournalFile
*f
,
1722 *extra
= htole64(p
);
1726 i
= htole64(le64toh(*idx
) - 1);
1727 r
= link_entry_into_array(f
, first
, &i
, p
);
1732 *idx
= htole64(le64toh(*idx
) + 1);
1736 static int journal_file_link_entry_item(JournalFile
*f
, Object
*o
, uint64_t offset
, uint64_t i
) {
1743 p
= le64toh(o
->entry
.items
[i
].object_offset
);
1747 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1751 return link_entry_into_array_plus_one(f
,
1752 &o
->data
.entry_offset
,
1753 &o
->data
.entry_array_offset
,
1758 static int journal_file_link_entry(JournalFile
*f
, Object
*o
, uint64_t offset
) {
1767 if (o
->object
.type
!= OBJECT_ENTRY
)
1770 __sync_synchronize();
1772 /* Link up the entry itself */
1773 r
= link_entry_into_array(f
,
1774 &f
->header
->entry_array_offset
,
1775 &f
->header
->n_entries
,
1780 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
1782 if (f
->header
->head_entry_realtime
== 0)
1783 f
->header
->head_entry_realtime
= o
->entry
.realtime
;
1785 f
->header
->tail_entry_realtime
= o
->entry
.realtime
;
1786 f
->header
->tail_entry_monotonic
= o
->entry
.monotonic
;
1788 /* Link up the items */
1789 n
= journal_file_entry_n_items(o
);
1790 for (i
= 0; i
< n
; i
++) {
1791 r
= journal_file_link_entry_item(f
, o
, offset
, i
);
1799 static int journal_file_append_entry_internal(
1801 const dual_timestamp
*ts
,
1802 const sd_id128_t
*boot_id
,
1804 const EntryItem items
[], unsigned n_items
,
1806 Object
**ret
, uint64_t *offset
) {
1814 assert(items
|| n_items
== 0);
1817 osize
= offsetof(Object
, entry
.items
) + (n_items
* sizeof(EntryItem
));
1819 r
= journal_file_append_object(f
, OBJECT_ENTRY
, osize
, &o
, &np
);
1823 o
->entry
.seqnum
= htole64(journal_file_entry_seqnum(f
, seqnum
));
1824 memcpy_safe(o
->entry
.items
, items
, n_items
* sizeof(EntryItem
));
1825 o
->entry
.realtime
= htole64(ts
->realtime
);
1826 o
->entry
.monotonic
= htole64(ts
->monotonic
);
1827 o
->entry
.xor_hash
= htole64(xor_hash
);
1828 o
->entry
.boot_id
= boot_id
? *boot_id
: f
->header
->boot_id
;
1831 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY
, o
, np
);
1836 r
= journal_file_link_entry(f
, o
, np
);
1849 void journal_file_post_change(JournalFile
*f
) {
1852 /* inotify() does not receive IN_MODIFY events from file
1853 * accesses done via mmap(). After each access we hence
1854 * trigger IN_MODIFY by truncating the journal file to its
1855 * current size which triggers IN_MODIFY. */
1857 __sync_synchronize();
1859 if (ftruncate(f
->fd
, f
->last_stat
.st_size
) < 0)
1860 log_debug_errno(errno
, "Failed to truncate file to its own size: %m");
1863 static int post_change_thunk(sd_event_source
*timer
, uint64_t usec
, void *userdata
) {
1866 journal_file_post_change(userdata
);
1871 static void schedule_post_change(JournalFile
*f
) {
1872 sd_event_source
*timer
;
1877 assert(f
->post_change_timer
);
1879 timer
= f
->post_change_timer
;
1881 r
= sd_event_source_get_enabled(timer
, &enabled
);
1883 log_debug_errno(r
, "Failed to get ftruncate timer state: %m");
1887 if (enabled
== SD_EVENT_ONESHOT
)
1890 r
= sd_event_now(sd_event_source_get_event(timer
), CLOCK_MONOTONIC
, &now
);
1892 log_debug_errno(r
, "Failed to get clock's now for scheduling ftruncate: %m");
1896 r
= sd_event_source_set_time(timer
, now
+f
->post_change_timer_period
);
1898 log_debug_errno(r
, "Failed to set time for scheduling ftruncate: %m");
1902 r
= sd_event_source_set_enabled(timer
, SD_EVENT_ONESHOT
);
1904 log_debug_errno(r
, "Failed to enable scheduled ftruncate: %m");
1911 /* On failure, let's simply post the change immediately. */
1912 journal_file_post_change(f
);
1915 /* Enable coalesced change posting in a timer on the provided sd_event instance */
1916 int journal_file_enable_post_change_timer(JournalFile
*f
, sd_event
*e
, usec_t t
) {
1917 _cleanup_(sd_event_source_unrefp
) sd_event_source
*timer
= NULL
;
1921 assert_return(!f
->post_change_timer
, -EINVAL
);
1925 r
= sd_event_add_time(e
, &timer
, CLOCK_MONOTONIC
, 0, 0, post_change_thunk
, f
);
1929 r
= sd_event_source_set_enabled(timer
, SD_EVENT_OFF
);
1933 f
->post_change_timer
= TAKE_PTR(timer
);
1934 f
->post_change_timer_period
= t
;
1939 static int entry_item_cmp(const void *_a
, const void *_b
) {
1940 const EntryItem
*a
= _a
, *b
= _b
;
1942 if (le64toh(a
->object_offset
) < le64toh(b
->object_offset
))
1944 if (le64toh(a
->object_offset
) > le64toh(b
->object_offset
))
1949 int journal_file_append_entry(
1951 const dual_timestamp
*ts
,
1952 const sd_id128_t
*boot_id
,
1953 const struct iovec iovec
[], unsigned n_iovec
,
1955 Object
**ret
, uint64_t *offset
) {
1960 uint64_t xor_hash
= 0;
1961 struct dual_timestamp _ts
;
1965 assert(iovec
|| n_iovec
== 0);
1968 if (!VALID_REALTIME(ts
->realtime
)) {
1969 log_debug("Invalid realtime timestamp %"PRIu64
", refusing entry.", ts
->realtime
);
1972 if (!VALID_MONOTONIC(ts
->monotonic
)) {
1973 log_debug("Invalid monotomic timestamp %"PRIu64
", refusing entry.", ts
->monotonic
);
1977 dual_timestamp_get(&_ts
);
1982 r
= journal_file_maybe_append_tag(f
, ts
->realtime
);
1987 /* alloca() can't take 0, hence let's allocate at least one */
1988 items
= newa(EntryItem
, MAX(1u, n_iovec
));
1990 for (i
= 0; i
< n_iovec
; i
++) {
1994 r
= journal_file_append_data(f
, iovec
[i
].iov_base
, iovec
[i
].iov_len
, &o
, &p
);
1998 xor_hash
^= le64toh(o
->data
.hash
);
1999 items
[i
].object_offset
= htole64(p
);
2000 items
[i
].hash
= o
->data
.hash
;
2003 /* Order by the position on disk, in order to improve seek
2004 * times for rotating media. */
2005 qsort_safe(items
, n_iovec
, sizeof(EntryItem
), entry_item_cmp
);
2007 r
= journal_file_append_entry_internal(f
, ts
, boot_id
, xor_hash
, items
, n_iovec
, seqnum
, ret
, offset
);
2009 /* If the memory mapping triggered a SIGBUS then we return an
2010 * IO error and ignore the error code passed down to us, since
2011 * it is very likely just an effect of a nullified replacement
2014 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
2017 if (f
->post_change_timer
)
2018 schedule_post_change(f
);
2020 journal_file_post_change(f
);
2025 typedef struct ChainCacheItem
{
2026 uint64_t first
; /* the array at the beginning of the chain */
2027 uint64_t array
; /* the cached array */
2028 uint64_t begin
; /* the first item in the cached array */
2029 uint64_t total
; /* the total number of items in all arrays before this one in the chain */
2030 uint64_t last_index
; /* the last index we looked at, to optimize locality when bisecting */
2033 static void chain_cache_put(
2040 uint64_t last_index
) {
2043 /* If the chain item to cache for this chain is the
2044 * first one it's not worth caching anything */
2048 if (ordered_hashmap_size(h
) >= CHAIN_CACHE_MAX
) {
2049 ci
= ordered_hashmap_steal_first(h
);
2052 ci
= new(ChainCacheItem
, 1);
2059 if (ordered_hashmap_put(h
, &ci
->first
, ci
) < 0) {
2064 assert(ci
->first
== first
);
2069 ci
->last_index
= last_index
;
2072 static int generic_array_get(
2076 Object
**ret
, uint64_t *offset
) {
2079 uint64_t p
= 0, a
, t
= 0;
2087 /* Try the chain cache first */
2088 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2089 if (ci
&& i
> ci
->total
) {
2098 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2102 k
= journal_file_entry_array_n_items(o
);
2104 p
= le64toh(o
->entry_array
.items
[i
]);
2110 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
2116 /* Let's cache this item for the next invocation */
2117 chain_cache_put(f
->chain_cache
, ci
, first
, a
, le64toh(o
->entry_array
.items
[0]), t
, i
);
2119 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2132 static int generic_array_get_plus_one(
2137 Object
**ret
, uint64_t *offset
) {
2146 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2159 return generic_array_get(f
, first
, i
-1, ret
, offset
);
2168 static int generic_array_bisect(
2173 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2174 direction_t direction
,
2179 uint64_t a
, p
, t
= 0, i
= 0, last_p
= 0, last_index
= (uint64_t) -1;
2180 bool subtract_one
= false;
2181 Object
*o
, *array
= NULL
;
2186 assert(test_object
);
2188 /* Start with the first array in the chain */
2191 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2192 if (ci
&& n
> ci
->total
&& ci
->begin
!= 0) {
2193 /* Ah, we have iterated this bisection array chain
2194 * previously! Let's see if we can skip ahead in the
2195 * chain, as far as the last time. But we can't jump
2196 * backwards in the chain, so let's check that
2199 r
= test_object(f
, ci
->begin
, needle
);
2203 if (r
== TEST_LEFT
) {
2204 /* OK, what we are looking for is right of the
2205 * begin of this EntryArray, so let's jump
2206 * straight to previously cached array in the
2212 last_index
= ci
->last_index
;
2217 uint64_t left
, right
, k
, lp
;
2219 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2223 k
= journal_file_entry_array_n_items(array
);
2229 lp
= p
= le64toh(array
->entry_array
.items
[i
]);
2233 r
= test_object(f
, p
, needle
);
2234 if (r
== -EBADMSG
) {
2235 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (1)");
2242 if (r
== TEST_FOUND
)
2243 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2245 if (r
== TEST_RIGHT
) {
2249 if (last_index
!= (uint64_t) -1) {
2250 assert(last_index
<= right
);
2252 /* If we cached the last index we
2253 * looked at, let's try to not to jump
2254 * too wildly around and see if we can
2255 * limit the range to look at early to
2256 * the immediate neighbors of the last
2257 * index we looked at. */
2259 if (last_index
> 0) {
2260 uint64_t x
= last_index
- 1;
2262 p
= le64toh(array
->entry_array
.items
[x
]);
2266 r
= test_object(f
, p
, needle
);
2270 if (r
== TEST_FOUND
)
2271 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2273 if (r
== TEST_RIGHT
)
2279 if (last_index
< right
) {
2280 uint64_t y
= last_index
+ 1;
2282 p
= le64toh(array
->entry_array
.items
[y
]);
2286 r
= test_object(f
, p
, needle
);
2290 if (r
== TEST_FOUND
)
2291 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2293 if (r
== TEST_RIGHT
)
2301 if (left
== right
) {
2302 if (direction
== DIRECTION_UP
)
2303 subtract_one
= true;
2309 assert(left
< right
);
2310 i
= (left
+ right
) / 2;
2312 p
= le64toh(array
->entry_array
.items
[i
]);
2316 r
= test_object(f
, p
, needle
);
2317 if (r
== -EBADMSG
) {
2318 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (2)");
2325 if (r
== TEST_FOUND
)
2326 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2328 if (r
== TEST_RIGHT
)
2336 if (direction
== DIRECTION_UP
) {
2338 subtract_one
= true;
2349 last_index
= (uint64_t) -1;
2350 a
= le64toh(array
->entry_array
.next_entry_array_offset
);
2356 if (subtract_one
&& t
== 0 && i
== 0)
2359 /* Let's cache this item for the next invocation */
2360 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
);
2362 if (subtract_one
&& i
== 0)
2364 else if (subtract_one
)
2365 p
= le64toh(array
->entry_array
.items
[i
-1]);
2367 p
= le64toh(array
->entry_array
.items
[i
]);
2369 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2380 *idx
= t
+ i
+ (subtract_one
? -1 : 0);
2385 static int generic_array_bisect_plus_one(
2391 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2392 direction_t direction
,
2398 bool step_back
= false;
2402 assert(test_object
);
2407 /* This bisects the array in object 'first', but first checks
2409 r
= test_object(f
, extra
, needle
);
2413 if (r
== TEST_FOUND
)
2414 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2416 /* if we are looking with DIRECTION_UP then we need to first
2417 see if in the actual array there is a matching entry, and
2418 return the last one of that. But if there isn't any we need
2419 to return this one. Hence remember this, and return it
2422 step_back
= direction
== DIRECTION_UP
;
2424 if (r
== TEST_RIGHT
) {
2425 if (direction
== DIRECTION_DOWN
)
2431 r
= generic_array_bisect(f
, first
, n
-1, needle
, test_object
, direction
, ret
, offset
, idx
);
2433 if (r
== 0 && step_back
)
2442 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2458 _pure_
static int test_object_offset(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2464 else if (p
< needle
)
2470 static int test_object_seqnum(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2477 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2481 if (le64toh(o
->entry
.seqnum
) == needle
)
2483 else if (le64toh(o
->entry
.seqnum
) < needle
)
2489 int journal_file_move_to_entry_by_seqnum(
2492 direction_t direction
,
2498 return generic_array_bisect(f
,
2499 le64toh(f
->header
->entry_array_offset
),
2500 le64toh(f
->header
->n_entries
),
2507 static int test_object_realtime(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2514 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2518 if (le64toh(o
->entry
.realtime
) == needle
)
2520 else if (le64toh(o
->entry
.realtime
) < needle
)
2526 int journal_file_move_to_entry_by_realtime(
2529 direction_t direction
,
2535 return generic_array_bisect(f
,
2536 le64toh(f
->header
->entry_array_offset
),
2537 le64toh(f
->header
->n_entries
),
2539 test_object_realtime
,
2544 static int test_object_monotonic(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2551 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2555 if (le64toh(o
->entry
.monotonic
) == needle
)
2557 else if (le64toh(o
->entry
.monotonic
) < needle
)
2563 static int find_data_object_by_boot_id(
2569 char t
[STRLEN("_BOOT_ID=") + 32 + 1] = "_BOOT_ID=";
2571 sd_id128_to_string(boot_id
, t
+ 9);
2572 return journal_file_find_data_object(f
, t
, sizeof(t
) - 1, o
, b
);
2575 int journal_file_move_to_entry_by_monotonic(
2579 direction_t direction
,
2588 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, NULL
);
2594 return generic_array_bisect_plus_one(f
,
2595 le64toh(o
->data
.entry_offset
),
2596 le64toh(o
->data
.entry_array_offset
),
2597 le64toh(o
->data
.n_entries
),
2599 test_object_monotonic
,
2604 void journal_file_reset_location(JournalFile
*f
) {
2605 f
->location_type
= LOCATION_HEAD
;
2606 f
->current_offset
= 0;
2607 f
->current_seqnum
= 0;
2608 f
->current_realtime
= 0;
2609 f
->current_monotonic
= 0;
2610 zero(f
->current_boot_id
);
2611 f
->current_xor_hash
= 0;
2614 void journal_file_save_location(JournalFile
*f
, Object
*o
, uint64_t offset
) {
2615 f
->location_type
= LOCATION_SEEK
;
2616 f
->current_offset
= offset
;
2617 f
->current_seqnum
= le64toh(o
->entry
.seqnum
);
2618 f
->current_realtime
= le64toh(o
->entry
.realtime
);
2619 f
->current_monotonic
= le64toh(o
->entry
.monotonic
);
2620 f
->current_boot_id
= o
->entry
.boot_id
;
2621 f
->current_xor_hash
= le64toh(o
->entry
.xor_hash
);
2624 int journal_file_compare_locations(JournalFile
*af
, JournalFile
*bf
) {
2629 assert(af
->location_type
== LOCATION_SEEK
);
2630 assert(bf
->location_type
== LOCATION_SEEK
);
2632 /* If contents and timestamps match, these entries are
2633 * identical, even if the seqnum does not match */
2634 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
) &&
2635 af
->current_monotonic
== bf
->current_monotonic
&&
2636 af
->current_realtime
== bf
->current_realtime
&&
2637 af
->current_xor_hash
== bf
->current_xor_hash
)
2640 if (sd_id128_equal(af
->header
->seqnum_id
, bf
->header
->seqnum_id
)) {
2642 /* If this is from the same seqnum source, compare
2644 if (af
->current_seqnum
< bf
->current_seqnum
)
2646 if (af
->current_seqnum
> bf
->current_seqnum
)
2649 /* Wow! This is weird, different data but the same
2650 * seqnums? Something is borked, but let's make the
2651 * best of it and compare by time. */
2654 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
)) {
2656 /* If the boot id matches, compare monotonic time */
2657 if (af
->current_monotonic
< bf
->current_monotonic
)
2659 if (af
->current_monotonic
> bf
->current_monotonic
)
2663 /* Otherwise, compare UTC time */
2664 if (af
->current_realtime
< bf
->current_realtime
)
2666 if (af
->current_realtime
> bf
->current_realtime
)
2669 /* Finally, compare by contents */
2670 if (af
->current_xor_hash
< bf
->current_xor_hash
)
2672 if (af
->current_xor_hash
> bf
->current_xor_hash
)
2678 static int bump_array_index(uint64_t *i
, direction_t direction
, uint64_t n
) {
2680 /* Increase or decrease the specified index, in the right direction. */
2682 if (direction
== DIRECTION_DOWN
) {
2697 static bool check_properly_ordered(uint64_t new_offset
, uint64_t old_offset
, direction_t direction
) {
2699 /* Consider it an error if any of the two offsets is uninitialized */
2700 if (old_offset
== 0 || new_offset
== 0)
2703 /* If we go down, the new offset must be larger than the old one. */
2704 return direction
== DIRECTION_DOWN
?
2705 new_offset
> old_offset
:
2706 new_offset
< old_offset
;
2709 int journal_file_next_entry(
2712 direction_t direction
,
2713 Object
**ret
, uint64_t *offset
) {
2721 n
= le64toh(f
->header
->n_entries
);
2726 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2728 r
= generic_array_bisect(f
,
2729 le64toh(f
->header
->entry_array_offset
),
2730 le64toh(f
->header
->n_entries
),
2739 r
= bump_array_index(&i
, direction
, n
);
2744 /* And jump to it */
2746 r
= generic_array_get(f
,
2747 le64toh(f
->header
->entry_array_offset
),
2755 /* OK, so this entry is borked. Most likely some entry didn't get synced to disk properly, let's see if
2756 * the next one might work for us instead. */
2757 log_debug_errno(r
, "Entry item %" PRIu64
" is bad, skipping over it.", i
);
2759 r
= bump_array_index(&i
, direction
, n
);
2764 /* Ensure our array is properly ordered. */
2765 if (p
> 0 && !check_properly_ordered(ofs
, p
, direction
)) {
2766 log_debug("%s: entry array not properly ordered at entry %" PRIu64
, f
->path
, i
);
2776 int journal_file_next_entry_for_data(
2778 Object
*o
, uint64_t p
,
2779 uint64_t data_offset
,
2780 direction_t direction
,
2781 Object
**ret
, uint64_t *offset
) {
2788 assert(p
> 0 || !o
);
2790 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2794 n
= le64toh(d
->data
.n_entries
);
2799 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2801 if (o
->object
.type
!= OBJECT_ENTRY
)
2804 r
= generic_array_bisect_plus_one(f
,
2805 le64toh(d
->data
.entry_offset
),
2806 le64toh(d
->data
.entry_array_offset
),
2807 le64toh(d
->data
.n_entries
),
2817 r
= bump_array_index(&i
, direction
, n
);
2823 r
= generic_array_get_plus_one(f
,
2824 le64toh(d
->data
.entry_offset
),
2825 le64toh(d
->data
.entry_array_offset
),
2833 log_debug_errno(r
, "Data entry item %" PRIu64
" is bad, skipping over it.", i
);
2835 r
= bump_array_index(&i
, direction
, n
);
2840 /* Ensure our array is properly ordered. */
2841 if (p
> 0 && check_properly_ordered(ofs
, p
, direction
)) {
2842 log_debug("%s data entry array not properly ordered at entry %" PRIu64
, f
->path
, i
);
2852 int journal_file_move_to_entry_by_offset_for_data(
2854 uint64_t data_offset
,
2856 direction_t direction
,
2857 Object
**ret
, uint64_t *offset
) {
2864 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2868 return generic_array_bisect_plus_one(f
,
2869 le64toh(d
->data
.entry_offset
),
2870 le64toh(d
->data
.entry_array_offset
),
2871 le64toh(d
->data
.n_entries
),
2878 int journal_file_move_to_entry_by_monotonic_for_data(
2880 uint64_t data_offset
,
2883 direction_t direction
,
2884 Object
**ret
, uint64_t *offset
) {
2892 /* First, seek by time */
2893 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &b
);
2899 r
= generic_array_bisect_plus_one(f
,
2900 le64toh(o
->data
.entry_offset
),
2901 le64toh(o
->data
.entry_array_offset
),
2902 le64toh(o
->data
.n_entries
),
2904 test_object_monotonic
,
2910 /* And now, continue seeking until we find an entry that
2911 * exists in both bisection arrays */
2917 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2921 r
= generic_array_bisect_plus_one(f
,
2922 le64toh(d
->data
.entry_offset
),
2923 le64toh(d
->data
.entry_array_offset
),
2924 le64toh(d
->data
.n_entries
),
2932 r
= journal_file_move_to_object(f
, OBJECT_DATA
, b
, &o
);
2936 r
= generic_array_bisect_plus_one(f
,
2937 le64toh(o
->data
.entry_offset
),
2938 le64toh(o
->data
.entry_array_offset
),
2939 le64toh(o
->data
.n_entries
),
2961 int journal_file_move_to_entry_by_seqnum_for_data(
2963 uint64_t data_offset
,
2965 direction_t direction
,
2966 Object
**ret
, uint64_t *offset
) {
2973 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2977 return generic_array_bisect_plus_one(f
,
2978 le64toh(d
->data
.entry_offset
),
2979 le64toh(d
->data
.entry_array_offset
),
2980 le64toh(d
->data
.n_entries
),
2987 int journal_file_move_to_entry_by_realtime_for_data(
2989 uint64_t data_offset
,
2991 direction_t direction
,
2992 Object
**ret
, uint64_t *offset
) {
2999 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
3003 return generic_array_bisect_plus_one(f
,
3004 le64toh(d
->data
.entry_offset
),
3005 le64toh(d
->data
.entry_array_offset
),
3006 le64toh(d
->data
.n_entries
),
3008 test_object_realtime
,
3013 void journal_file_dump(JournalFile
*f
) {
3021 journal_file_print_header(f
);
3023 p
= le64toh(f
->header
->header_size
);
3025 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &o
);
3029 switch (o
->object
.type
) {
3032 printf("Type: OBJECT_UNUSED\n");
3036 printf("Type: OBJECT_DATA\n");
3040 printf("Type: OBJECT_FIELD\n");
3044 printf("Type: OBJECT_ENTRY seqnum=%"PRIu64
" monotonic=%"PRIu64
" realtime=%"PRIu64
"\n",
3045 le64toh(o
->entry
.seqnum
),
3046 le64toh(o
->entry
.monotonic
),
3047 le64toh(o
->entry
.realtime
));
3050 case OBJECT_FIELD_HASH_TABLE
:
3051 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
3054 case OBJECT_DATA_HASH_TABLE
:
3055 printf("Type: OBJECT_DATA_HASH_TABLE\n");
3058 case OBJECT_ENTRY_ARRAY
:
3059 printf("Type: OBJECT_ENTRY_ARRAY\n");
3063 printf("Type: OBJECT_TAG seqnum=%"PRIu64
" epoch=%"PRIu64
"\n",
3064 le64toh(o
->tag
.seqnum
),
3065 le64toh(o
->tag
.epoch
));
3069 printf("Type: unknown (%i)\n", o
->object
.type
);
3073 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
)
3074 printf("Flags: %s\n",
3075 object_compressed_to_string(o
->object
.flags
& OBJECT_COMPRESSION_MASK
));
3077 if (p
== le64toh(f
->header
->tail_object_offset
))
3080 p
= p
+ ALIGN64(le64toh(o
->object
.size
));
3085 log_error("File corrupt");
3088 static const char* format_timestamp_safe(char *buf
, size_t l
, usec_t t
) {
3091 x
= format_timestamp(buf
, l
, t
);
3097 void journal_file_print_header(JournalFile
*f
) {
3098 char a
[33], b
[33], c
[33], d
[33];
3099 char x
[FORMAT_TIMESTAMP_MAX
], y
[FORMAT_TIMESTAMP_MAX
], z
[FORMAT_TIMESTAMP_MAX
];
3101 char bytes
[FORMAT_BYTES_MAX
];
3106 printf("File Path: %s\n"
3110 "Sequential Number ID: %s\n"
3112 "Compatible Flags:%s%s\n"
3113 "Incompatible Flags:%s%s%s\n"
3114 "Header size: %"PRIu64
"\n"
3115 "Arena size: %"PRIu64
"\n"
3116 "Data Hash Table Size: %"PRIu64
"\n"
3117 "Field Hash Table Size: %"PRIu64
"\n"
3118 "Rotate Suggested: %s\n"
3119 "Head Sequential Number: %"PRIu64
" (%"PRIx64
")\n"
3120 "Tail Sequential Number: %"PRIu64
" (%"PRIx64
")\n"
3121 "Head Realtime Timestamp: %s (%"PRIx64
")\n"
3122 "Tail Realtime Timestamp: %s (%"PRIx64
")\n"
3123 "Tail Monotonic Timestamp: %s (%"PRIx64
")\n"
3124 "Objects: %"PRIu64
"\n"
3125 "Entry Objects: %"PRIu64
"\n",
3127 sd_id128_to_string(f
->header
->file_id
, a
),
3128 sd_id128_to_string(f
->header
->machine_id
, b
),
3129 sd_id128_to_string(f
->header
->boot_id
, c
),
3130 sd_id128_to_string(f
->header
->seqnum_id
, d
),
3131 f
->header
->state
== STATE_OFFLINE
? "OFFLINE" :
3132 f
->header
->state
== STATE_ONLINE
? "ONLINE" :
3133 f
->header
->state
== STATE_ARCHIVED
? "ARCHIVED" : "UNKNOWN",
3134 JOURNAL_HEADER_SEALED(f
->header
) ? " SEALED" : "",
3135 (le32toh(f
->header
->compatible_flags
) & ~HEADER_COMPATIBLE_ANY
) ? " ???" : "",
3136 JOURNAL_HEADER_COMPRESSED_XZ(f
->header
) ? " COMPRESSED-XZ" : "",
3137 JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
) ? " COMPRESSED-LZ4" : "",
3138 (le32toh(f
->header
->incompatible_flags
) & ~HEADER_INCOMPATIBLE_ANY
) ? " ???" : "",
3139 le64toh(f
->header
->header_size
),
3140 le64toh(f
->header
->arena_size
),
3141 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3142 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
),
3143 yes_no(journal_file_rotate_suggested(f
, 0)),
3144 le64toh(f
->header
->head_entry_seqnum
), le64toh(f
->header
->head_entry_seqnum
),
3145 le64toh(f
->header
->tail_entry_seqnum
), le64toh(f
->header
->tail_entry_seqnum
),
3146 format_timestamp_safe(x
, sizeof(x
), le64toh(f
->header
->head_entry_realtime
)), le64toh(f
->header
->head_entry_realtime
),
3147 format_timestamp_safe(y
, sizeof(y
), le64toh(f
->header
->tail_entry_realtime
)), le64toh(f
->header
->tail_entry_realtime
),
3148 format_timespan(z
, sizeof(z
), le64toh(f
->header
->tail_entry_monotonic
), USEC_PER_MSEC
), le64toh(f
->header
->tail_entry_monotonic
),
3149 le64toh(f
->header
->n_objects
),
3150 le64toh(f
->header
->n_entries
));
3152 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3153 printf("Data Objects: %"PRIu64
"\n"
3154 "Data Hash Table Fill: %.1f%%\n",
3155 le64toh(f
->header
->n_data
),
3156 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))));
3158 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3159 printf("Field Objects: %"PRIu64
"\n"
3160 "Field Hash Table Fill: %.1f%%\n",
3161 le64toh(f
->header
->n_fields
),
3162 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))));
3164 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
))
3165 printf("Tag Objects: %"PRIu64
"\n",
3166 le64toh(f
->header
->n_tags
));
3167 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
3168 printf("Entry Array Objects: %"PRIu64
"\n",
3169 le64toh(f
->header
->n_entry_arrays
));
3171 if (fstat(f
->fd
, &st
) >= 0)
3172 printf("Disk usage: %s\n", format_bytes(bytes
, sizeof(bytes
), (uint64_t) st
.st_blocks
* 512ULL));
3175 static int journal_file_warn_btrfs(JournalFile
*f
) {
3181 /* Before we write anything, check if the COW logic is turned
3182 * off on btrfs. Given our write pattern that is quite
3183 * unfriendly to COW file systems this should greatly improve
3184 * performance on COW file systems, such as btrfs, at the
3185 * expense of data integrity features (which shouldn't be too
3186 * bad, given that we do our own checksumming). */
3188 r
= btrfs_is_filesystem(f
->fd
);
3190 return log_warning_errno(r
, "Failed to determine if journal is on btrfs: %m");
3194 r
= read_attr_fd(f
->fd
, &attrs
);
3196 return log_warning_errno(r
, "Failed to read file attributes: %m");
3198 if (attrs
& FS_NOCOW_FL
) {
3199 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3203 log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3204 "This is likely to slow down journal access substantially, please consider turning "
3205 "off the copy-on-write file attribute on the journal directory, using chattr +C.", f
->path
);
3210 int journal_file_open(
3216 uint64_t compress_threshold_bytes
,
3218 JournalMetrics
*metrics
,
3219 MMapCache
*mmap_cache
,
3220 Set
*deferred_closes
,
3221 JournalFile
*template,
3222 JournalFile
**ret
) {
3224 bool newly_created
= false;
3228 char bytes
[FORMAT_BYTES_MAX
];
3231 assert(fd
>= 0 || fname
);
3233 if (!IN_SET((flags
& O_ACCMODE
), O_RDONLY
, O_RDWR
))
3236 if (fname
&& (flags
& O_CREAT
) && !endswith(fname
, ".journal"))
3239 f
= new0(JournalFile
, 1);
3247 f
->prot
= prot_from_flags(flags
);
3248 f
->writable
= (flags
& O_ACCMODE
) != O_RDONLY
;
3250 f
->compress_lz4
= compress
;
3252 f
->compress_xz
= compress
;
3255 if (compress_threshold_bytes
== (uint64_t) -1)
3256 f
->compress_threshold_bytes
= DEFAULT_COMPRESS_THRESHOLD
;
3258 f
->compress_threshold_bytes
= MAX(MIN_COMPRESS_THRESHOLD
, compress_threshold_bytes
);
3264 log_debug("Journal effective settings seal=%s compress=%s compress_threshold_bytes=%s",
3265 yes_no(f
->seal
), yes_no(JOURNAL_FILE_COMPRESS(f
)),
3266 format_bytes(bytes
, sizeof(bytes
), f
->compress_threshold_bytes
));
3269 f
->mmap
= mmap_cache_ref(mmap_cache
);
3271 f
->mmap
= mmap_cache_new();
3279 f
->path
= strdup(fname
);
3287 /* If we don't know the path, fill in something explanatory and vaguely useful */
3288 if (asprintf(&f
->path
, "/proc/self/%i", fd
) < 0) {
3294 f
->chain_cache
= ordered_hashmap_new(&uint64_hash_ops
);
3295 if (!f
->chain_cache
) {
3301 /* We pass O_NONBLOCK here, so that in case somebody pointed us to some character device node or FIFO
3302 * or so, we likely fail quickly than block for long. For regular files O_NONBLOCK has no effect, hence
3303 * it doesn't hurt in that case. */
3305 f
->fd
= open(f
->path
, f
->flags
|O_CLOEXEC
|O_NONBLOCK
, f
->mode
);
3311 /* fds we opened here by us should also be closed by us. */
3314 r
= fd_nonblock(f
->fd
, false);
3319 f
->cache_fd
= mmap_cache_add_fd(f
->mmap
, f
->fd
);
3325 r
= journal_file_fstat(f
);
3329 if (f
->last_stat
.st_size
== 0 && f
->writable
) {
3331 (void) journal_file_warn_btrfs(f
);
3333 /* Let's attach the creation time to the journal file, so that the vacuuming code knows the age of this
3334 * file even if the file might end up corrupted one day... Ideally we'd just use the creation time many
3335 * file systems maintain for each file, but the API to query this is very new, hence let's emulate this
3336 * via extended attributes. If extended attributes are not supported we'll just skip this, and rely
3337 * solely on mtime/atime/ctime of the file. */
3338 (void) fd_setcrtime(f
->fd
, 0);
3341 /* Try to load the FSPRG state, and if we can't, then
3342 * just don't do sealing */
3344 r
= journal_file_fss_load(f
);
3350 r
= journal_file_init_header(f
, template);
3354 r
= journal_file_fstat(f
);
3358 newly_created
= true;
3361 if (f
->last_stat
.st_size
< (off_t
) HEADER_SIZE_MIN
) {
3366 r
= mmap_cache_get(f
->mmap
, f
->cache_fd
, f
->prot
, CONTEXT_HEADER
, true, 0, PAGE_ALIGN(sizeof(Header
)), &f
->last_stat
, &h
, NULL
);
3372 if (!newly_created
) {
3373 set_clear_with_destructor(deferred_closes
, journal_file_close
);
3375 r
= journal_file_verify_header(f
);
3381 if (!newly_created
&& f
->writable
) {
3382 r
= journal_file_fss_load(f
);
3390 journal_default_metrics(metrics
, f
->fd
);
3391 f
->metrics
= *metrics
;
3392 } else if (template)
3393 f
->metrics
= template->metrics
;
3395 r
= journal_file_refresh_header(f
);
3401 r
= journal_file_hmac_setup(f
);
3406 if (newly_created
) {
3407 r
= journal_file_setup_field_hash_table(f
);
3411 r
= journal_file_setup_data_hash_table(f
);
3416 r
= journal_file_append_first_tag(f
);
3422 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
)) {
3427 if (template && template->post_change_timer
) {
3428 r
= journal_file_enable_post_change_timer(
3430 sd_event_source_get_event(template->post_change_timer
),
3431 template->post_change_timer_period
);
3437 /* The file is opened now successfully, thus we take possession of any passed in fd. */
3444 if (f
->cache_fd
&& mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
3447 (void) journal_file_close(f
);
3452 int journal_file_rotate(JournalFile
**f
, bool compress
, uint64_t compress_threshold_bytes
, bool seal
, Set
*deferred_closes
) {
3453 _cleanup_free_
char *p
= NULL
;
3455 JournalFile
*old_file
, *new_file
= NULL
;
3463 if (!old_file
->writable
)
3466 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
3467 * rotation, since we don't know the actual path, and couldn't rename the file hence. */
3468 if (path_startswith(old_file
->path
, "/proc/self/fd"))
3471 if (!endswith(old_file
->path
, ".journal"))
3474 l
= strlen(old_file
->path
);
3475 r
= asprintf(&p
, "%.*s@" SD_ID128_FORMAT_STR
"-%016"PRIx64
"-%016"PRIx64
".journal",
3476 (int) l
- 8, old_file
->path
,
3477 SD_ID128_FORMAT_VAL(old_file
->header
->seqnum_id
),
3478 le64toh((*f
)->header
->head_entry_seqnum
),
3479 le64toh((*f
)->header
->head_entry_realtime
));
3483 /* Try to rename the file to the archived version. If the file
3484 * already was deleted, we'll get ENOENT, let's ignore that
3486 r
= rename(old_file
->path
, p
);
3487 if (r
< 0 && errno
!= ENOENT
)
3490 /* Sync the rename to disk */
3491 (void) fsync_directory_of_file(old_file
->fd
);
3493 /* Set as archive so offlining commits w/state=STATE_ARCHIVED.
3494 * Previously we would set old_file->header->state to STATE_ARCHIVED directly here,
3495 * but journal_file_set_offline() short-circuits when state != STATE_ONLINE, which
3496 * would result in the rotated journal never getting fsync() called before closing.
3497 * Now we simply queue the archive state by setting an archive bit, leaving the state
3498 * as STATE_ONLINE so proper offlining occurs. */
3499 old_file
->archive
= true;
3501 /* Currently, btrfs is not very good with out write patterns
3502 * and fragments heavily. Let's defrag our journal files when
3503 * we archive them */
3504 old_file
->defrag_on_close
= true;
3506 r
= journal_file_open(-1, old_file
->path
, old_file
->flags
, old_file
->mode
, compress
,
3507 compress_threshold_bytes
, seal
, NULL
, old_file
->mmap
, deferred_closes
,
3508 old_file
, &new_file
);
3510 if (deferred_closes
&&
3511 set_put(deferred_closes
, old_file
) >= 0)
3512 (void) journal_file_set_offline(old_file
, false);
3514 (void) journal_file_close(old_file
);
3520 int journal_file_open_reliably(
3525 uint64_t compress_threshold_bytes
,
3527 JournalMetrics
*metrics
,
3528 MMapCache
*mmap_cache
,
3529 Set
*deferred_closes
,
3530 JournalFile
*template,
3531 JournalFile
**ret
) {
3535 _cleanup_free_
char *p
= NULL
;
3537 r
= journal_file_open(-1, fname
, flags
, mode
, compress
, compress_threshold_bytes
, seal
, metrics
, mmap_cache
,
3538 deferred_closes
, template, ret
);
3540 -EBADMSG
, /* Corrupted */
3541 -ENODATA
, /* Truncated */
3542 -EHOSTDOWN
, /* Other machine */
3543 -EPROTONOSUPPORT
, /* Incompatible feature */
3544 -EBUSY
, /* Unclean shutdown */
3545 -ESHUTDOWN
, /* Already archived */
3546 -EIO
, /* IO error, including SIGBUS on mmap */
3547 -EIDRM
, /* File has been deleted */
3548 -ETXTBSY
)) /* File is from the future */
3551 if ((flags
& O_ACCMODE
) == O_RDONLY
)
3554 if (!(flags
& O_CREAT
))
3557 if (!endswith(fname
, ".journal"))
3560 /* The file is corrupted. Rotate it away and try it again (but only once) */
3563 if (asprintf(&p
, "%.*s@%016"PRIx64
"-%016"PRIx64
".journal~",
3565 now(CLOCK_REALTIME
),
3569 if (rename(fname
, p
) < 0)
3572 /* btrfs doesn't cope well with our write pattern and
3573 * fragments heavily. Let's defrag all files we rotate */
3575 (void) chattr_path(p
, 0, FS_NOCOW_FL
);
3576 (void) btrfs_defrag(p
);
3578 log_warning_errno(r
, "File %s corrupted or uncleanly shut down, renaming and replacing.", fname
);
3580 return journal_file_open(-1, fname
, flags
, mode
, compress
, compress_threshold_bytes
, seal
, metrics
, mmap_cache
,
3581 deferred_closes
, template, ret
);
3584 int journal_file_copy_entry(JournalFile
*from
, JournalFile
*to
, Object
*o
, uint64_t p
) {
3586 uint64_t q
, xor_hash
= 0;
3590 const sd_id128_t
*boot_id
;
3600 ts
.monotonic
= le64toh(o
->entry
.monotonic
);
3601 ts
.realtime
= le64toh(o
->entry
.realtime
);
3602 boot_id
= &o
->entry
.boot_id
;
3604 n
= journal_file_entry_n_items(o
);
3605 /* alloca() can't take 0, hence let's allocate at least one */
3606 items
= newa(EntryItem
, MAX(1u, n
));
3608 for (i
= 0; i
< n
; i
++) {
3615 q
= le64toh(o
->entry
.items
[i
].object_offset
);
3616 le_hash
= o
->entry
.items
[i
].hash
;
3618 r
= journal_file_move_to_object(from
, OBJECT_DATA
, q
, &o
);
3622 if (le_hash
!= o
->data
.hash
)
3625 l
= le64toh(o
->object
.size
) - offsetof(Object
, data
.payload
);
3628 /* We hit the limit on 32bit machines */
3629 if ((uint64_t) t
!= l
)
3632 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
3633 #if HAVE_XZ || HAVE_LZ4
3636 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
3637 o
->data
.payload
, l
, &from
->compress_buffer
, &from
->compress_buffer_size
, &rsize
, 0);
3641 data
= from
->compress_buffer
;
3644 return -EPROTONOSUPPORT
;
3647 data
= o
->data
.payload
;
3649 r
= journal_file_append_data(to
, data
, l
, &u
, &h
);
3653 xor_hash
^= le64toh(u
->data
.hash
);
3654 items
[i
].object_offset
= htole64(h
);
3655 items
[i
].hash
= u
->data
.hash
;
3657 r
= journal_file_move_to_object(from
, OBJECT_ENTRY
, p
, &o
);
3662 r
= journal_file_append_entry_internal(to
, &ts
, boot_id
, xor_hash
, items
, n
,
3665 if (mmap_cache_got_sigbus(to
->mmap
, to
->cache_fd
))
3671 void journal_reset_metrics(JournalMetrics
*m
) {
3674 /* Set everything to "pick automatic values". */
3676 *m
= (JournalMetrics
) {
3677 .min_use
= (uint64_t) -1,
3678 .max_use
= (uint64_t) -1,
3679 .min_size
= (uint64_t) -1,
3680 .max_size
= (uint64_t) -1,
3681 .keep_free
= (uint64_t) -1,
3682 .n_max_files
= (uint64_t) -1,
3686 void journal_default_metrics(JournalMetrics
*m
, int fd
) {
3687 char a
[FORMAT_BYTES_MAX
], b
[FORMAT_BYTES_MAX
], c
[FORMAT_BYTES_MAX
], d
[FORMAT_BYTES_MAX
], e
[FORMAT_BYTES_MAX
];
3694 if (fstatvfs(fd
, &ss
) >= 0)
3695 fs_size
= ss
.f_frsize
* ss
.f_blocks
;
3697 log_debug_errno(errno
, "Failed to determine disk size: %m");
3701 if (m
->max_use
== (uint64_t) -1) {
3704 m
->max_use
= PAGE_ALIGN(fs_size
/ 10); /* 10% of file system size */
3706 if (m
->max_use
> DEFAULT_MAX_USE_UPPER
)
3707 m
->max_use
= DEFAULT_MAX_USE_UPPER
;
3709 if (m
->max_use
< DEFAULT_MAX_USE_LOWER
)
3710 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3712 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3714 m
->max_use
= PAGE_ALIGN(m
->max_use
);
3716 if (m
->max_use
!= 0 && m
->max_use
< JOURNAL_FILE_SIZE_MIN
*2)
3717 m
->max_use
= JOURNAL_FILE_SIZE_MIN
*2;
3720 if (m
->min_use
== (uint64_t) -1)
3721 m
->min_use
= DEFAULT_MIN_USE
;
3723 if (m
->min_use
> m
->max_use
)
3724 m
->min_use
= m
->max_use
;
3726 if (m
->max_size
== (uint64_t) -1) {
3727 m
->max_size
= PAGE_ALIGN(m
->max_use
/ 8); /* 8 chunks */
3729 if (m
->max_size
> DEFAULT_MAX_SIZE_UPPER
)
3730 m
->max_size
= DEFAULT_MAX_SIZE_UPPER
;
3732 m
->max_size
= PAGE_ALIGN(m
->max_size
);
3734 if (m
->max_size
!= 0) {
3735 if (m
->max_size
< JOURNAL_FILE_SIZE_MIN
)
3736 m
->max_size
= JOURNAL_FILE_SIZE_MIN
;
3738 if (m
->max_use
!= 0 && m
->max_size
*2 > m
->max_use
)
3739 m
->max_use
= m
->max_size
*2;
3742 if (m
->min_size
== (uint64_t) -1)
3743 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3745 m
->min_size
= PAGE_ALIGN(m
->min_size
);
3747 if (m
->min_size
< JOURNAL_FILE_SIZE_MIN
)
3748 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3750 if (m
->max_size
!= 0 && m
->min_size
> m
->max_size
)
3751 m
->max_size
= m
->min_size
;
3754 if (m
->keep_free
== (uint64_t) -1) {
3757 m
->keep_free
= PAGE_ALIGN(fs_size
* 3 / 20); /* 15% of file system size */
3759 if (m
->keep_free
> DEFAULT_KEEP_FREE_UPPER
)
3760 m
->keep_free
= DEFAULT_KEEP_FREE_UPPER
;
3763 m
->keep_free
= DEFAULT_KEEP_FREE
;
3766 if (m
->n_max_files
== (uint64_t) -1)
3767 m
->n_max_files
= DEFAULT_N_MAX_FILES
;
3769 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64
,
3770 format_bytes(a
, sizeof(a
), m
->min_use
),
3771 format_bytes(b
, sizeof(b
), m
->max_use
),
3772 format_bytes(c
, sizeof(c
), m
->max_size
),
3773 format_bytes(d
, sizeof(d
), m
->min_size
),
3774 format_bytes(e
, sizeof(e
), m
->keep_free
),
3778 int journal_file_get_cutoff_realtime_usec(JournalFile
*f
, usec_t
*from
, usec_t
*to
) {
3784 if (f
->header
->head_entry_realtime
== 0)
3787 *from
= le64toh(f
->header
->head_entry_realtime
);
3791 if (f
->header
->tail_entry_realtime
== 0)
3794 *to
= le64toh(f
->header
->tail_entry_realtime
);
3800 int journal_file_get_cutoff_monotonic_usec(JournalFile
*f
, sd_id128_t boot_id
, usec_t
*from
, usec_t
*to
) {
3808 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &p
);
3812 if (le64toh(o
->data
.n_entries
) <= 0)
3816 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, le64toh(o
->data
.entry_offset
), &o
);
3820 *from
= le64toh(o
->entry
.monotonic
);
3824 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
3828 r
= generic_array_get_plus_one(f
,
3829 le64toh(o
->data
.entry_offset
),
3830 le64toh(o
->data
.entry_array_offset
),
3831 le64toh(o
->data
.n_entries
)-1,
3836 *to
= le64toh(o
->entry
.monotonic
);
3842 bool journal_file_rotate_suggested(JournalFile
*f
, usec_t max_file_usec
) {
3846 /* If we gained new header fields we gained new features,
3847 * hence suggest a rotation */
3848 if (le64toh(f
->header
->header_size
) < sizeof(Header
)) {
3849 log_debug("%s uses an outdated header, suggesting rotation.", f
->path
);
3853 /* Let's check if the hash tables grew over a certain fill
3854 * level (75%, borrowing this value from Java's hash table
3855 * implementation), and if so suggest a rotation. To calculate
3856 * the fill level we need the n_data field, which only exists
3857 * in newer versions. */
3859 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3860 if (le64toh(f
->header
->n_data
) * 4ULL > (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3861 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.",
3863 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))),
3864 le64toh(f
->header
->n_data
),
3865 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3866 (unsigned long long) f
->last_stat
.st_size
,
3867 f
->last_stat
.st_size
/ le64toh(f
->header
->n_data
));
3871 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3872 if (le64toh(f
->header
->n_fields
) * 4ULL > (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3873 log_debug("Field hash table of %s has a fill level at %.1f (%"PRIu64
" of %"PRIu64
" items), suggesting rotation.",
3875 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))),
3876 le64toh(f
->header
->n_fields
),
3877 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
));
3881 /* Are the data objects properly indexed by field objects? */
3882 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
3883 JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
3884 le64toh(f
->header
->n_data
) > 0 &&
3885 le64toh(f
->header
->n_fields
) == 0)
3888 if (max_file_usec
> 0) {
3891 h
= le64toh(f
->header
->head_entry_realtime
);
3892 t
= now(CLOCK_REALTIME
);
3894 if (h
> 0 && t
> h
+ max_file_usec
)