1 /* SPDX-License-Identifier: LGPL-2.1+ */
9 #include <sys/statvfs.h>
15 #include "alloc-util.h"
16 #include "btrfs-util.h"
17 #include "chattr-util.h"
20 #include "format-util.h"
22 #include "journal-authenticate.h"
23 #include "journal-def.h"
24 #include "journal-file.h"
26 #include "memory-util.h"
27 #include "path-util.h"
28 #include "random-util.h"
30 #include "sort-util.h"
31 #include "stat-util.h"
32 #include "string-util.h"
34 #include "xattr-util.h"
36 #define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
37 #define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
39 #define DEFAULT_COMPRESS_THRESHOLD (512ULL)
40 #define MIN_COMPRESS_THRESHOLD (8ULL)
42 /* This is the minimum journal file size */
43 #define JOURNAL_FILE_SIZE_MIN (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 assert_se(sigfillset(&ss
) >= 0);
242 /* Don't block SIGBUS since the offlining thread accesses a memory mapped file.
243 * Asynchronous SIGBUS signals can safely be handled by either thread. */
244 assert_se(sigdelset(&ss
, SIGBUS
) >= 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
) {
347 /* Write the final tag */
348 if (f
->seal
&& f
->writable
) {
351 r
= journal_file_append_tag(f
);
353 log_error_errno(r
, "Failed to append tag when closing journal: %m");
357 if (f
->post_change_timer
) {
358 if (sd_event_source_get_enabled(f
->post_change_timer
, NULL
) > 0)
359 journal_file_post_change(f
);
361 sd_event_source_disable_unref(f
->post_change_timer
);
364 journal_file_set_offline(f
, true);
366 if (f
->mmap
&& f
->cache_fd
)
367 mmap_cache_free_fd(f
->mmap
, f
->cache_fd
);
369 if (f
->fd
>= 0 && f
->defrag_on_close
) {
371 /* Be friendly to btrfs: turn COW back on again now,
372 * and defragment the file. We won't write to the file
373 * ever again, hence remove all fragmentation, and
374 * reenable all the good bits COW usually provides
375 * (such as data checksumming). */
377 (void) chattr_fd(f
->fd
, 0, FS_NOCOW_FL
, NULL
);
378 (void) btrfs_defrag_fd(f
->fd
);
385 mmap_cache_unref(f
->mmap
);
387 ordered_hashmap_free_free(f
->chain_cache
);
389 #if HAVE_XZ || HAVE_LZ4
390 free(f
->compress_buffer
);
395 munmap(f
->fss_file
, PAGE_ALIGN(f
->fss_file_size
));
397 free(f
->fsprg_state
);
402 gcry_md_close(f
->hmac
);
408 static int journal_file_init_header(JournalFile
*f
, JournalFile
*template) {
415 memcpy(h
.signature
, HEADER_SIGNATURE
, 8);
416 h
.header_size
= htole64(ALIGN64(sizeof(h
)));
418 h
.incompatible_flags
|= htole32(
419 f
->compress_xz
* HEADER_INCOMPATIBLE_COMPRESSED_XZ
|
420 f
->compress_lz4
* HEADER_INCOMPATIBLE_COMPRESSED_LZ4
);
422 h
.compatible_flags
= htole32(
423 f
->seal
* HEADER_COMPATIBLE_SEALED
);
425 r
= sd_id128_randomize(&h
.file_id
);
430 h
.seqnum_id
= template->header
->seqnum_id
;
431 h
.tail_entry_seqnum
= template->header
->tail_entry_seqnum
;
433 h
.seqnum_id
= h
.file_id
;
435 k
= pwrite(f
->fd
, &h
, sizeof(h
), 0);
445 static int journal_file_refresh_header(JournalFile
*f
) {
452 r
= sd_id128_get_machine(&f
->header
->machine_id
);
453 if (IN_SET(r
, -ENOENT
, -ENOMEDIUM
))
454 /* We don't have a machine-id, let's continue without */
455 zero(f
->header
->machine_id
);
459 r
= sd_id128_get_boot(&boot_id
);
463 f
->header
->boot_id
= boot_id
;
465 r
= journal_file_set_online(f
);
467 /* Sync the online state to disk */
470 /* We likely just created a new file, also sync the directory this file is located in. */
471 (void) fsync_directory_of_file(f
->fd
);
476 static bool warn_wrong_flags(const JournalFile
*f
, bool compatible
) {
477 const uint32_t any
= compatible
? HEADER_COMPATIBLE_ANY
: HEADER_INCOMPATIBLE_ANY
,
478 supported
= compatible
? HEADER_COMPATIBLE_SUPPORTED
: HEADER_INCOMPATIBLE_SUPPORTED
;
479 const char *type
= compatible
? "compatible" : "incompatible";
482 flags
= le32toh(compatible
? f
->header
->compatible_flags
: f
->header
->incompatible_flags
);
484 if (flags
& ~supported
) {
486 log_debug("Journal file %s has unknown %s flags 0x%"PRIx32
,
487 f
->path
, type
, flags
& ~any
);
488 flags
= (flags
& any
) & ~supported
;
492 _cleanup_free_
char *t
= NULL
;
494 if (compatible
&& (flags
& HEADER_COMPATIBLE_SEALED
))
495 strv
[n
++] = "sealed";
496 if (!compatible
&& (flags
& HEADER_INCOMPATIBLE_COMPRESSED_XZ
))
497 strv
[n
++] = "xz-compressed";
498 if (!compatible
&& (flags
& HEADER_INCOMPATIBLE_COMPRESSED_LZ4
))
499 strv
[n
++] = "lz4-compressed";
501 assert(n
< ELEMENTSOF(strv
));
503 t
= strv_join((char**) strv
, ", ");
504 log_debug("Journal file %s uses %s %s %s disabled at compilation time.",
505 f
->path
, type
, n
> 1 ? "flags" : "flag", strnull(t
));
513 static int journal_file_verify_header(JournalFile
*f
) {
514 uint64_t arena_size
, header_size
;
519 if (memcmp(f
->header
->signature
, HEADER_SIGNATURE
, 8))
522 /* In both read and write mode we refuse to open files with incompatible
523 * flags we don't know. */
524 if (warn_wrong_flags(f
, false))
525 return -EPROTONOSUPPORT
;
527 /* When open for writing we refuse to open files with compatible flags, too. */
528 if (f
->writable
&& warn_wrong_flags(f
, true))
529 return -EPROTONOSUPPORT
;
531 if (f
->header
->state
>= _STATE_MAX
)
534 header_size
= le64toh(f
->header
->header_size
);
536 /* The first addition was n_data, so check that we are at least this large */
537 if (header_size
< HEADER_SIZE_MIN
)
540 if (JOURNAL_HEADER_SEALED(f
->header
) && !JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
543 arena_size
= le64toh(f
->header
->arena_size
);
545 if (UINT64_MAX
- header_size
< arena_size
|| header_size
+ arena_size
> (uint64_t) f
->last_stat
.st_size
)
548 if (le64toh(f
->header
->tail_object_offset
) > header_size
+ arena_size
)
551 if (!VALID64(le64toh(f
->header
->data_hash_table_offset
)) ||
552 !VALID64(le64toh(f
->header
->field_hash_table_offset
)) ||
553 !VALID64(le64toh(f
->header
->tail_object_offset
)) ||
554 !VALID64(le64toh(f
->header
->entry_array_offset
)))
558 sd_id128_t machine_id
;
562 r
= sd_id128_get_machine(&machine_id
);
566 if (!sd_id128_equal(machine_id
, f
->header
->machine_id
))
569 state
= f
->header
->state
;
571 if (state
== STATE_ARCHIVED
)
572 return -ESHUTDOWN
; /* Already archived */
573 else if (state
== STATE_ONLINE
)
574 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
),
575 "Journal file %s is already online. Assuming unclean closing.",
577 else if (state
!= STATE_OFFLINE
)
578 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
),
579 "Journal file %s has unknown state %i.",
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 return log_debug_errno(SYNTHETIC_ERRNO(ETXTBSY
),
590 "Journal file %s is from the future, refusing to append new data to it that'd be older.",
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 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
757 "Bad n_entries: %" PRIu64
": %" PRIu64
,
758 le64toh(o
->data
.n_entries
),
761 if (le64toh(o
->object
.size
) - offsetof(DataObject
, payload
) <= 0)
762 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
763 "Bad object size (<= %zu): %" PRIu64
": %" PRIu64
,
764 offsetof(DataObject
, payload
),
765 le64toh(o
->object
.size
),
768 if (!VALID64(le64toh(o
->data
.next_hash_offset
)) ||
769 !VALID64(le64toh(o
->data
.next_field_offset
)) ||
770 !VALID64(le64toh(o
->data
.entry_offset
)) ||
771 !VALID64(le64toh(o
->data
.entry_array_offset
)))
772 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
773 "Invalid offset, next_hash_offset=" OFSfmt
", next_field_offset=" OFSfmt
", entry_offset=" OFSfmt
", entry_array_offset=" OFSfmt
": %" PRIu64
,
774 le64toh(o
->data
.next_hash_offset
),
775 le64toh(o
->data
.next_field_offset
),
776 le64toh(o
->data
.entry_offset
),
777 le64toh(o
->data
.entry_array_offset
),
784 if (le64toh(o
->object
.size
) - offsetof(FieldObject
, payload
) <= 0)
785 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
786 "Bad field size (<= %zu): %" PRIu64
": %" PRIu64
,
787 offsetof(FieldObject
, payload
),
788 le64toh(o
->object
.size
),
791 if (!VALID64(le64toh(o
->field
.next_hash_offset
)) ||
792 !VALID64(le64toh(o
->field
.head_data_offset
)))
793 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
794 "Invalid offset, next_hash_offset=" OFSfmt
", head_data_offset=" OFSfmt
": %" PRIu64
,
795 le64toh(o
->field
.next_hash_offset
),
796 le64toh(o
->field
.head_data_offset
),
801 if ((le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) % sizeof(EntryItem
) != 0)
802 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
803 "Bad entry size (<= %zu): %" PRIu64
": %" PRIu64
,
804 offsetof(EntryObject
, items
),
805 le64toh(o
->object
.size
),
808 if ((le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) / sizeof(EntryItem
) <= 0)
809 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
810 "Invalid number items in entry: %" PRIu64
": %" PRIu64
,
811 (le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) / sizeof(EntryItem
),
814 if (le64toh(o
->entry
.seqnum
) <= 0)
815 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
816 "Invalid entry seqnum: %" PRIx64
": %" PRIu64
,
817 le64toh(o
->entry
.seqnum
),
820 if (!VALID_REALTIME(le64toh(o
->entry
.realtime
)))
821 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
822 "Invalid entry realtime timestamp: %" PRIu64
": %" PRIu64
,
823 le64toh(o
->entry
.realtime
),
826 if (!VALID_MONOTONIC(le64toh(o
->entry
.monotonic
)))
827 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
828 "Invalid entry monotonic timestamp: %" PRIu64
": %" PRIu64
,
829 le64toh(o
->entry
.monotonic
),
834 case OBJECT_DATA_HASH_TABLE
:
835 case OBJECT_FIELD_HASH_TABLE
:
836 if ((le64toh(o
->object
.size
) - offsetof(HashTableObject
, items
)) % sizeof(HashItem
) != 0 ||
837 (le64toh(o
->object
.size
) - offsetof(HashTableObject
, items
)) / sizeof(HashItem
) <= 0)
838 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
839 "Invalid %s hash table size: %" PRIu64
": %" PRIu64
,
840 o
->object
.type
== OBJECT_DATA_HASH_TABLE
? "data" : "field",
841 le64toh(o
->object
.size
),
846 case OBJECT_ENTRY_ARRAY
:
847 if ((le64toh(o
->object
.size
) - offsetof(EntryArrayObject
, items
)) % sizeof(le64_t
) != 0 ||
848 (le64toh(o
->object
.size
) - offsetof(EntryArrayObject
, items
)) / sizeof(le64_t
) <= 0)
849 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
850 "Invalid object entry array size: %" PRIu64
": %" PRIu64
,
851 le64toh(o
->object
.size
),
854 if (!VALID64(le64toh(o
->entry_array
.next_entry_array_offset
)))
855 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
856 "Invalid object entry array next_entry_array_offset: " OFSfmt
": %" PRIu64
,
857 le64toh(o
->entry_array
.next_entry_array_offset
),
863 if (le64toh(o
->object
.size
) != sizeof(TagObject
))
864 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
865 "Invalid object tag size: %" PRIu64
": %" PRIu64
,
866 le64toh(o
->object
.size
),
869 if (!VALID_EPOCH(le64toh(o
->tag
.epoch
)))
870 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
871 "Invalid object tag epoch: %" PRIu64
": %" PRIu64
,
872 le64toh(o
->tag
.epoch
), offset
);
880 int journal_file_move_to_object(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
**ret
) {
890 /* Objects may only be located at multiple of 64 bit */
891 if (!VALID64(offset
))
892 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
893 "Attempt to move to object at non-64bit boundary: %" PRIu64
,
896 /* Object may not be located in the file header */
897 if (offset
< le64toh(f
->header
->header_size
))
898 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
899 "Attempt to move to object located in file header: %" PRIu64
,
902 r
= journal_file_move_to(f
, type
, false, offset
, sizeof(ObjectHeader
), &t
, &tsize
);
907 s
= le64toh(o
->object
.size
);
910 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
911 "Attempt to move to uninitialized object: %" PRIu64
,
913 if (s
< sizeof(ObjectHeader
))
914 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
915 "Attempt to move to overly short object: %" PRIu64
,
918 if (o
->object
.type
<= OBJECT_UNUSED
)
919 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
920 "Attempt to move to object with invalid type: %" PRIu64
,
923 if (s
< minimum_header_size(o
))
924 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
925 "Attempt to move to truncated object: %" PRIu64
,
928 if (type
> OBJECT_UNUSED
&& o
->object
.type
!= type
)
929 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
930 "Attempt to move to object of unexpected type: %" PRIu64
,
934 r
= journal_file_move_to(f
, type
, false, offset
, s
, &t
, NULL
);
941 r
= journal_file_check_object(f
, offset
, o
);
949 static uint64_t journal_file_entry_seqnum(JournalFile
*f
, uint64_t *seqnum
) {
955 r
= le64toh(f
->header
->tail_entry_seqnum
) + 1;
958 /* If an external seqnum counter was passed, we update
959 * both the local and the external one, and set it to
960 * the maximum of both */
968 f
->header
->tail_entry_seqnum
= htole64(r
);
970 if (f
->header
->head_entry_seqnum
== 0)
971 f
->header
->head_entry_seqnum
= htole64(r
);
976 int journal_file_append_object(JournalFile
*f
, ObjectType type
, uint64_t size
, Object
**ret
, uint64_t *offset
) {
984 assert(type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
);
985 assert(size
>= sizeof(ObjectHeader
));
989 r
= journal_file_set_online(f
);
993 p
= le64toh(f
->header
->tail_object_offset
);
995 p
= le64toh(f
->header
->header_size
);
997 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &tail
);
1001 p
+= ALIGN64(le64toh(tail
->object
.size
));
1004 r
= journal_file_allocate(f
, p
, size
);
1008 r
= journal_file_move_to(f
, type
, false, p
, size
, &t
, NULL
);
1015 o
->object
.type
= type
;
1016 o
->object
.size
= htole64(size
);
1018 f
->header
->tail_object_offset
= htole64(p
);
1019 f
->header
->n_objects
= htole64(le64toh(f
->header
->n_objects
) + 1);
1027 static int journal_file_setup_data_hash_table(JournalFile
*f
) {
1035 /* We estimate that we need 1 hash table entry per 768 bytes
1036 of journal file and we want to make sure we never get
1037 beyond 75% fill level. Calculate the hash table size for
1038 the maximum file size based on these metrics. */
1040 s
= (f
->metrics
.max_size
* 4 / 768 / 3) * sizeof(HashItem
);
1041 if (s
< DEFAULT_DATA_HASH_TABLE_SIZE
)
1042 s
= DEFAULT_DATA_HASH_TABLE_SIZE
;
1044 log_debug("Reserving %"PRIu64
" entries in hash table.", s
/ sizeof(HashItem
));
1046 r
= journal_file_append_object(f
,
1047 OBJECT_DATA_HASH_TABLE
,
1048 offsetof(Object
, hash_table
.items
) + s
,
1053 memzero(o
->hash_table
.items
, s
);
1055 f
->header
->data_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1056 f
->header
->data_hash_table_size
= htole64(s
);
1061 static int journal_file_setup_field_hash_table(JournalFile
*f
) {
1069 /* We use a fixed size hash table for the fields as this
1070 * number should grow very slowly only */
1072 s
= DEFAULT_FIELD_HASH_TABLE_SIZE
;
1073 r
= journal_file_append_object(f
,
1074 OBJECT_FIELD_HASH_TABLE
,
1075 offsetof(Object
, hash_table
.items
) + s
,
1080 memzero(o
->hash_table
.items
, s
);
1082 f
->header
->field_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1083 f
->header
->field_hash_table_size
= htole64(s
);
1088 int journal_file_map_data_hash_table(JournalFile
*f
) {
1096 if (f
->data_hash_table
)
1099 p
= le64toh(f
->header
->data_hash_table_offset
);
1100 s
= le64toh(f
->header
->data_hash_table_size
);
1102 r
= journal_file_move_to(f
,
1103 OBJECT_DATA_HASH_TABLE
,
1110 f
->data_hash_table
= t
;
1114 int journal_file_map_field_hash_table(JournalFile
*f
) {
1122 if (f
->field_hash_table
)
1125 p
= le64toh(f
->header
->field_hash_table_offset
);
1126 s
= le64toh(f
->header
->field_hash_table_size
);
1128 r
= journal_file_move_to(f
,
1129 OBJECT_FIELD_HASH_TABLE
,
1136 f
->field_hash_table
= t
;
1140 static int journal_file_link_field(
1151 assert(f
->field_hash_table
);
1155 if (o
->object
.type
!= OBJECT_FIELD
)
1158 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1162 /* This might alter the window we are looking at */
1163 o
->field
.next_hash_offset
= o
->field
.head_data_offset
= 0;
1166 p
= le64toh(f
->field_hash_table
[h
].tail_hash_offset
);
1168 f
->field_hash_table
[h
].head_hash_offset
= htole64(offset
);
1170 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1174 o
->field
.next_hash_offset
= htole64(offset
);
1177 f
->field_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1179 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
1180 f
->header
->n_fields
= htole64(le64toh(f
->header
->n_fields
) + 1);
1185 static int journal_file_link_data(
1196 assert(f
->data_hash_table
);
1200 if (o
->object
.type
!= OBJECT_DATA
)
1203 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1207 /* This might alter the window we are looking at */
1208 o
->data
.next_hash_offset
= o
->data
.next_field_offset
= 0;
1209 o
->data
.entry_offset
= o
->data
.entry_array_offset
= 0;
1210 o
->data
.n_entries
= 0;
1213 p
= le64toh(f
->data_hash_table
[h
].tail_hash_offset
);
1215 /* Only entry in the hash table is easy */
1216 f
->data_hash_table
[h
].head_hash_offset
= htole64(offset
);
1218 /* Move back to the previous data object, to patch in
1221 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1225 o
->data
.next_hash_offset
= htole64(offset
);
1228 f
->data_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1230 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
1231 f
->header
->n_data
= htole64(le64toh(f
->header
->n_data
) + 1);
1236 int journal_file_find_field_object_with_hash(
1238 const void *field
, uint64_t size
, uint64_t hash
,
1239 Object
**ret
, uint64_t *offset
) {
1241 uint64_t p
, osize
, h
, m
;
1246 assert(field
&& size
> 0);
1248 /* If the field hash table is empty, we can't find anything */
1249 if (le64toh(f
->header
->field_hash_table_size
) <= 0)
1252 /* Map the field hash table, if it isn't mapped yet. */
1253 r
= journal_file_map_field_hash_table(f
);
1257 osize
= offsetof(Object
, field
.payload
) + size
;
1259 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1264 p
= le64toh(f
->field_hash_table
[h
].head_hash_offset
);
1269 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1273 if (le64toh(o
->field
.hash
) == hash
&&
1274 le64toh(o
->object
.size
) == osize
&&
1275 memcmp(o
->field
.payload
, field
, size
) == 0) {
1285 p
= le64toh(o
->field
.next_hash_offset
);
1291 int journal_file_find_field_object(
1293 const void *field
, uint64_t size
,
1294 Object
**ret
, uint64_t *offset
) {
1299 assert(field
&& size
> 0);
1301 hash
= hash64(field
, size
);
1303 return journal_file_find_field_object_with_hash(f
,
1308 int journal_file_find_data_object_with_hash(
1310 const void *data
, uint64_t size
, uint64_t hash
,
1311 Object
**ret
, uint64_t *offset
) {
1313 uint64_t p
, osize
, h
, m
;
1318 assert(data
|| size
== 0);
1320 /* If there's no data hash table, then there's no entry. */
1321 if (le64toh(f
->header
->data_hash_table_size
) <= 0)
1324 /* Map the data hash table, if it isn't mapped yet. */
1325 r
= journal_file_map_data_hash_table(f
);
1329 osize
= offsetof(Object
, data
.payload
) + size
;
1331 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1336 p
= le64toh(f
->data_hash_table
[h
].head_hash_offset
);
1341 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1345 if (le64toh(o
->data
.hash
) != hash
)
1348 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
1349 #if HAVE_XZ || HAVE_LZ4
1353 l
= le64toh(o
->object
.size
);
1354 if (l
<= offsetof(Object
, data
.payload
))
1357 l
-= offsetof(Object
, data
.payload
);
1359 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
1360 o
->data
.payload
, l
, &f
->compress_buffer
, &f
->compress_buffer_size
, &rsize
, 0);
1364 if (rsize
== size
&&
1365 memcmp(f
->compress_buffer
, data
, size
) == 0) {
1376 return -EPROTONOSUPPORT
;
1378 } else if (le64toh(o
->object
.size
) == osize
&&
1379 memcmp(o
->data
.payload
, data
, size
) == 0) {
1391 p
= le64toh(o
->data
.next_hash_offset
);
1397 int journal_file_find_data_object(
1399 const void *data
, uint64_t size
,
1400 Object
**ret
, uint64_t *offset
) {
1405 assert(data
|| size
== 0);
1407 hash
= hash64(data
, size
);
1409 return journal_file_find_data_object_with_hash(f
,
1414 static int journal_file_append_field(
1416 const void *field
, uint64_t size
,
1417 Object
**ret
, uint64_t *offset
) {
1425 assert(field
&& size
> 0);
1427 hash
= hash64(field
, size
);
1429 r
= journal_file_find_field_object_with_hash(f
, field
, size
, hash
, &o
, &p
);
1443 osize
= offsetof(Object
, field
.payload
) + size
;
1444 r
= journal_file_append_object(f
, OBJECT_FIELD
, osize
, &o
, &p
);
1448 o
->field
.hash
= htole64(hash
);
1449 memcpy(o
->field
.payload
, field
, size
);
1451 r
= journal_file_link_field(f
, o
, p
, hash
);
1455 /* The linking might have altered the window, so let's
1456 * refresh our pointer */
1457 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1462 r
= journal_file_hmac_put_object(f
, OBJECT_FIELD
, o
, p
);
1476 static int journal_file_append_data(
1478 const void *data
, uint64_t size
,
1479 Object
**ret
, uint64_t *offset
) {
1484 int r
, compression
= 0;
1488 assert(data
|| size
== 0);
1490 hash
= hash64(data
, size
);
1492 r
= journal_file_find_data_object_with_hash(f
, data
, size
, hash
, &o
, &p
);
1506 osize
= offsetof(Object
, data
.payload
) + size
;
1507 r
= journal_file_append_object(f
, OBJECT_DATA
, osize
, &o
, &p
);
1511 o
->data
.hash
= htole64(hash
);
1513 #if HAVE_XZ || HAVE_LZ4
1514 if (JOURNAL_FILE_COMPRESS(f
) && size
>= f
->compress_threshold_bytes
) {
1517 compression
= compress_blob(data
, size
, o
->data
.payload
, size
- 1, &rsize
);
1519 if (compression
>= 0) {
1520 o
->object
.size
= htole64(offsetof(Object
, data
.payload
) + rsize
);
1521 o
->object
.flags
|= compression
;
1523 log_debug("Compressed data object %"PRIu64
" -> %zu using %s",
1524 size
, rsize
, object_compressed_to_string(compression
));
1526 /* Compression didn't work, we don't really care why, let's continue without compression */
1531 if (compression
== 0)
1532 memcpy_safe(o
->data
.payload
, data
, size
);
1534 r
= journal_file_link_data(f
, o
, p
, hash
);
1539 r
= journal_file_hmac_put_object(f
, OBJECT_DATA
, o
, p
);
1544 /* The linking might have altered the window, so let's
1545 * refresh our pointer */
1546 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1553 eq
= memchr(data
, '=', size
);
1554 if (eq
&& eq
> data
) {
1558 /* Create field object ... */
1559 r
= journal_file_append_field(f
, data
, (uint8_t*) eq
- (uint8_t*) data
, &fo
, &fp
);
1563 /* ... and link it in. */
1564 o
->data
.next_field_offset
= fo
->field
.head_data_offset
;
1565 fo
->field
.head_data_offset
= le64toh(p
);
1577 uint64_t journal_file_entry_n_items(Object
*o
) {
1580 if (o
->object
.type
!= OBJECT_ENTRY
)
1583 return (le64toh(o
->object
.size
) - offsetof(Object
, entry
.items
)) / sizeof(EntryItem
);
1586 uint64_t journal_file_entry_array_n_items(Object
*o
) {
1589 if (o
->object
.type
!= OBJECT_ENTRY_ARRAY
)
1592 return (le64toh(o
->object
.size
) - offsetof(Object
, entry_array
.items
)) / sizeof(uint64_t);
1595 uint64_t journal_file_hash_table_n_items(Object
*o
) {
1598 if (!IN_SET(o
->object
.type
, OBJECT_DATA_HASH_TABLE
, OBJECT_FIELD_HASH_TABLE
))
1601 return (le64toh(o
->object
.size
) - offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
);
1604 static int link_entry_into_array(JournalFile
*f
,
1609 uint64_t n
= 0, ap
= 0, q
, i
, a
, hidx
;
1618 a
= le64toh(*first
);
1619 i
= hidx
= le64toh(*idx
);
1622 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
1626 n
= journal_file_entry_array_n_items(o
);
1628 o
->entry_array
.items
[i
] = htole64(p
);
1629 *idx
= htole64(hidx
+ 1);
1635 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
1646 r
= journal_file_append_object(f
, OBJECT_ENTRY_ARRAY
,
1647 offsetof(Object
, entry_array
.items
) + n
* sizeof(uint64_t),
1653 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY_ARRAY
, o
, q
);
1658 o
->entry_array
.items
[i
] = htole64(p
);
1661 *first
= htole64(q
);
1663 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, ap
, &o
);
1667 o
->entry_array
.next_entry_array_offset
= htole64(q
);
1670 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
1671 f
->header
->n_entry_arrays
= htole64(le64toh(f
->header
->n_entry_arrays
) + 1);
1673 *idx
= htole64(hidx
+ 1);
1678 static int link_entry_into_array_plus_one(JournalFile
*f
,
1693 *extra
= htole64(p
);
1697 i
= htole64(le64toh(*idx
) - 1);
1698 r
= link_entry_into_array(f
, first
, &i
, p
);
1703 *idx
= htole64(le64toh(*idx
) + 1);
1707 static int journal_file_link_entry_item(JournalFile
*f
, Object
*o
, uint64_t offset
, uint64_t i
) {
1714 p
= le64toh(o
->entry
.items
[i
].object_offset
);
1718 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1722 return link_entry_into_array_plus_one(f
,
1723 &o
->data
.entry_offset
,
1724 &o
->data
.entry_array_offset
,
1729 static int journal_file_link_entry(JournalFile
*f
, Object
*o
, uint64_t offset
) {
1738 if (o
->object
.type
!= OBJECT_ENTRY
)
1741 __sync_synchronize();
1743 /* Link up the entry itself */
1744 r
= link_entry_into_array(f
,
1745 &f
->header
->entry_array_offset
,
1746 &f
->header
->n_entries
,
1751 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
1753 if (f
->header
->head_entry_realtime
== 0)
1754 f
->header
->head_entry_realtime
= o
->entry
.realtime
;
1756 f
->header
->tail_entry_realtime
= o
->entry
.realtime
;
1757 f
->header
->tail_entry_monotonic
= o
->entry
.monotonic
;
1759 /* Link up the items */
1760 n
= journal_file_entry_n_items(o
);
1761 for (i
= 0; i
< n
; i
++) {
1762 r
= journal_file_link_entry_item(f
, o
, offset
, i
);
1770 static int journal_file_append_entry_internal(
1772 const dual_timestamp
*ts
,
1773 const sd_id128_t
*boot_id
,
1775 const EntryItem items
[], unsigned n_items
,
1777 Object
**ret
, uint64_t *offset
) {
1785 assert(items
|| n_items
== 0);
1788 osize
= offsetof(Object
, entry
.items
) + (n_items
* sizeof(EntryItem
));
1790 r
= journal_file_append_object(f
, OBJECT_ENTRY
, osize
, &o
, &np
);
1794 o
->entry
.seqnum
= htole64(journal_file_entry_seqnum(f
, seqnum
));
1795 memcpy_safe(o
->entry
.items
, items
, n_items
* sizeof(EntryItem
));
1796 o
->entry
.realtime
= htole64(ts
->realtime
);
1797 o
->entry
.monotonic
= htole64(ts
->monotonic
);
1798 o
->entry
.xor_hash
= htole64(xor_hash
);
1800 f
->header
->boot_id
= *boot_id
;
1801 o
->entry
.boot_id
= f
->header
->boot_id
;
1804 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY
, o
, np
);
1809 r
= journal_file_link_entry(f
, o
, np
);
1822 void journal_file_post_change(JournalFile
*f
) {
1828 /* inotify() does not receive IN_MODIFY events from file
1829 * accesses done via mmap(). After each access we hence
1830 * trigger IN_MODIFY by truncating the journal file to its
1831 * current size which triggers IN_MODIFY. */
1833 __sync_synchronize();
1835 if (ftruncate(f
->fd
, f
->last_stat
.st_size
) < 0)
1836 log_debug_errno(errno
, "Failed to truncate file to its own size: %m");
1839 static int post_change_thunk(sd_event_source
*timer
, uint64_t usec
, void *userdata
) {
1842 journal_file_post_change(userdata
);
1847 static void schedule_post_change(JournalFile
*f
) {
1852 assert(f
->post_change_timer
);
1854 r
= sd_event_source_get_enabled(f
->post_change_timer
, NULL
);
1856 log_debug_errno(r
, "Failed to get ftruncate timer state: %m");
1862 r
= sd_event_now(sd_event_source_get_event(f
->post_change_timer
), CLOCK_MONOTONIC
, &now
);
1864 log_debug_errno(r
, "Failed to get clock's now for scheduling ftruncate: %m");
1868 r
= sd_event_source_set_time(f
->post_change_timer
, now
+ f
->post_change_timer_period
);
1870 log_debug_errno(r
, "Failed to set time for scheduling ftruncate: %m");
1874 r
= sd_event_source_set_enabled(f
->post_change_timer
, SD_EVENT_ONESHOT
);
1876 log_debug_errno(r
, "Failed to enable scheduled ftruncate: %m");
1883 /* On failure, let's simply post the change immediately. */
1884 journal_file_post_change(f
);
1887 /* Enable coalesced change posting in a timer on the provided sd_event instance */
1888 int journal_file_enable_post_change_timer(JournalFile
*f
, sd_event
*e
, usec_t t
) {
1889 _cleanup_(sd_event_source_unrefp
) sd_event_source
*timer
= NULL
;
1893 assert_return(!f
->post_change_timer
, -EINVAL
);
1897 r
= sd_event_add_time(e
, &timer
, CLOCK_MONOTONIC
, 0, 0, post_change_thunk
, f
);
1901 r
= sd_event_source_set_enabled(timer
, SD_EVENT_OFF
);
1905 f
->post_change_timer
= TAKE_PTR(timer
);
1906 f
->post_change_timer_period
= t
;
1911 static int entry_item_cmp(const EntryItem
*a
, const EntryItem
*b
) {
1912 return CMP(le64toh(a
->object_offset
), le64toh(b
->object_offset
));
1915 int journal_file_append_entry(
1917 const dual_timestamp
*ts
,
1918 const sd_id128_t
*boot_id
,
1919 const struct iovec iovec
[], unsigned n_iovec
,
1921 Object
**ret
, uint64_t *offset
) {
1926 uint64_t xor_hash
= 0;
1927 struct dual_timestamp _ts
;
1931 assert(iovec
|| n_iovec
== 0);
1934 if (!VALID_REALTIME(ts
->realtime
))
1935 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1936 "Invalid realtime timestamp %" PRIu64
", refusing entry.",
1938 if (!VALID_MONOTONIC(ts
->monotonic
))
1939 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1940 "Invalid monotomic timestamp %" PRIu64
", refusing entry.",
1943 dual_timestamp_get(&_ts
);
1948 r
= journal_file_maybe_append_tag(f
, ts
->realtime
);
1953 /* alloca() can't take 0, hence let's allocate at least one */
1954 items
= newa(EntryItem
, MAX(1u, n_iovec
));
1956 for (i
= 0; i
< n_iovec
; i
++) {
1960 r
= journal_file_append_data(f
, iovec
[i
].iov_base
, iovec
[i
].iov_len
, &o
, &p
);
1964 xor_hash
^= le64toh(o
->data
.hash
);
1965 items
[i
].object_offset
= htole64(p
);
1966 items
[i
].hash
= o
->data
.hash
;
1969 /* Order by the position on disk, in order to improve seek
1970 * times for rotating media. */
1971 typesafe_qsort(items
, n_iovec
, entry_item_cmp
);
1973 r
= journal_file_append_entry_internal(f
, ts
, boot_id
, xor_hash
, items
, n_iovec
, seqnum
, ret
, offset
);
1975 /* If the memory mapping triggered a SIGBUS then we return an
1976 * IO error and ignore the error code passed down to us, since
1977 * it is very likely just an effect of a nullified replacement
1980 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
1983 if (f
->post_change_timer
)
1984 schedule_post_change(f
);
1986 journal_file_post_change(f
);
1991 typedef struct ChainCacheItem
{
1992 uint64_t first
; /* the array at the beginning of the chain */
1993 uint64_t array
; /* the cached array */
1994 uint64_t begin
; /* the first item in the cached array */
1995 uint64_t total
; /* the total number of items in all arrays before this one in the chain */
1996 uint64_t last_index
; /* the last index we looked at, to optimize locality when bisecting */
1999 static void chain_cache_put(
2006 uint64_t last_index
) {
2009 /* If the chain item to cache for this chain is the
2010 * first one it's not worth caching anything */
2014 if (ordered_hashmap_size(h
) >= CHAIN_CACHE_MAX
) {
2015 ci
= ordered_hashmap_steal_first(h
);
2018 ci
= new(ChainCacheItem
, 1);
2025 if (ordered_hashmap_put(h
, &ci
->first
, ci
) < 0) {
2030 assert(ci
->first
== first
);
2035 ci
->last_index
= last_index
;
2038 static int generic_array_get(
2042 Object
**ret
, uint64_t *offset
) {
2045 uint64_t p
= 0, a
, t
= 0;
2053 /* Try the chain cache first */
2054 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2055 if (ci
&& i
> ci
->total
) {
2064 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2068 k
= journal_file_entry_array_n_items(o
);
2070 p
= le64toh(o
->entry_array
.items
[i
]);
2076 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
2082 /* Let's cache this item for the next invocation */
2083 chain_cache_put(f
->chain_cache
, ci
, first
, a
, le64toh(o
->entry_array
.items
[0]), t
, i
);
2085 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2098 static int generic_array_get_plus_one(
2103 Object
**ret
, uint64_t *offset
) {
2112 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2125 return generic_array_get(f
, first
, i
-1, ret
, offset
);
2134 static int generic_array_bisect(
2139 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2140 direction_t direction
,
2145 uint64_t a
, p
, t
= 0, i
= 0, last_p
= 0, last_index
= (uint64_t) -1;
2146 bool subtract_one
= false;
2147 Object
*o
, *array
= NULL
;
2152 assert(test_object
);
2154 /* Start with the first array in the chain */
2157 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2158 if (ci
&& n
> ci
->total
&& ci
->begin
!= 0) {
2159 /* Ah, we have iterated this bisection array chain
2160 * previously! Let's see if we can skip ahead in the
2161 * chain, as far as the last time. But we can't jump
2162 * backwards in the chain, so let's check that
2165 r
= test_object(f
, ci
->begin
, needle
);
2169 if (r
== TEST_LEFT
) {
2170 /* OK, what we are looking for is right of the
2171 * begin of this EntryArray, so let's jump
2172 * straight to previously cached array in the
2178 last_index
= ci
->last_index
;
2183 uint64_t left
, right
, k
, lp
;
2185 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2189 k
= journal_file_entry_array_n_items(array
);
2195 lp
= p
= le64toh(array
->entry_array
.items
[i
]);
2199 r
= test_object(f
, p
, needle
);
2200 if (r
== -EBADMSG
) {
2201 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (1)");
2208 if (r
== TEST_FOUND
)
2209 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2211 if (r
== TEST_RIGHT
) {
2215 if (last_index
!= (uint64_t) -1) {
2216 assert(last_index
<= right
);
2218 /* If we cached the last index we
2219 * looked at, let's try to not to jump
2220 * too wildly around and see if we can
2221 * limit the range to look at early to
2222 * the immediate neighbors of the last
2223 * index we looked at. */
2225 if (last_index
> 0) {
2226 uint64_t x
= last_index
- 1;
2228 p
= le64toh(array
->entry_array
.items
[x
]);
2232 r
= test_object(f
, p
, needle
);
2236 if (r
== TEST_FOUND
)
2237 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2239 if (r
== TEST_RIGHT
)
2245 if (last_index
< right
) {
2246 uint64_t y
= last_index
+ 1;
2248 p
= le64toh(array
->entry_array
.items
[y
]);
2252 r
= test_object(f
, p
, needle
);
2256 if (r
== TEST_FOUND
)
2257 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2259 if (r
== TEST_RIGHT
)
2267 if (left
== right
) {
2268 if (direction
== DIRECTION_UP
)
2269 subtract_one
= true;
2275 assert(left
< right
);
2276 i
= (left
+ right
) / 2;
2278 p
= le64toh(array
->entry_array
.items
[i
]);
2282 r
= test_object(f
, p
, needle
);
2283 if (r
== -EBADMSG
) {
2284 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (2)");
2291 if (r
== TEST_FOUND
)
2292 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2294 if (r
== TEST_RIGHT
)
2302 if (direction
== DIRECTION_UP
) {
2304 subtract_one
= true;
2315 last_index
= (uint64_t) -1;
2316 a
= le64toh(array
->entry_array
.next_entry_array_offset
);
2322 if (subtract_one
&& t
== 0 && i
== 0)
2325 /* Let's cache this item for the next invocation */
2326 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
);
2328 if (subtract_one
&& i
== 0)
2330 else if (subtract_one
)
2331 p
= le64toh(array
->entry_array
.items
[i
-1]);
2333 p
= le64toh(array
->entry_array
.items
[i
]);
2335 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2346 *idx
= t
+ i
+ (subtract_one
? -1 : 0);
2351 static int generic_array_bisect_plus_one(
2357 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2358 direction_t direction
,
2364 bool step_back
= false;
2368 assert(test_object
);
2373 /* This bisects the array in object 'first', but first checks
2375 r
= test_object(f
, extra
, needle
);
2379 if (r
== TEST_FOUND
)
2380 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2382 /* if we are looking with DIRECTION_UP then we need to first
2383 see if in the actual array there is a matching entry, and
2384 return the last one of that. But if there isn't any we need
2385 to return this one. Hence remember this, and return it
2388 step_back
= direction
== DIRECTION_UP
;
2390 if (r
== TEST_RIGHT
) {
2391 if (direction
== DIRECTION_DOWN
)
2397 r
= generic_array_bisect(f
, first
, n
-1, needle
, test_object
, direction
, ret
, offset
, idx
);
2399 if (r
== 0 && step_back
)
2408 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2424 _pure_
static int test_object_offset(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2430 else if (p
< needle
)
2436 static int test_object_seqnum(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2443 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2447 if (le64toh(o
->entry
.seqnum
) == needle
)
2449 else if (le64toh(o
->entry
.seqnum
) < needle
)
2455 int journal_file_move_to_entry_by_seqnum(
2458 direction_t direction
,
2464 return generic_array_bisect(f
,
2465 le64toh(f
->header
->entry_array_offset
),
2466 le64toh(f
->header
->n_entries
),
2473 static int test_object_realtime(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2480 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2484 if (le64toh(o
->entry
.realtime
) == needle
)
2486 else if (le64toh(o
->entry
.realtime
) < needle
)
2492 int journal_file_move_to_entry_by_realtime(
2495 direction_t direction
,
2501 return generic_array_bisect(f
,
2502 le64toh(f
->header
->entry_array_offset
),
2503 le64toh(f
->header
->n_entries
),
2505 test_object_realtime
,
2510 static int test_object_monotonic(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2517 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2521 if (le64toh(o
->entry
.monotonic
) == needle
)
2523 else if (le64toh(o
->entry
.monotonic
) < needle
)
2529 static int find_data_object_by_boot_id(
2535 char t
[STRLEN("_BOOT_ID=") + 32 + 1] = "_BOOT_ID=";
2537 sd_id128_to_string(boot_id
, t
+ 9);
2538 return journal_file_find_data_object(f
, t
, sizeof(t
) - 1, o
, b
);
2541 int journal_file_move_to_entry_by_monotonic(
2545 direction_t direction
,
2554 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, NULL
);
2560 return generic_array_bisect_plus_one(f
,
2561 le64toh(o
->data
.entry_offset
),
2562 le64toh(o
->data
.entry_array_offset
),
2563 le64toh(o
->data
.n_entries
),
2565 test_object_monotonic
,
2570 void journal_file_reset_location(JournalFile
*f
) {
2571 f
->location_type
= LOCATION_HEAD
;
2572 f
->current_offset
= 0;
2573 f
->current_seqnum
= 0;
2574 f
->current_realtime
= 0;
2575 f
->current_monotonic
= 0;
2576 zero(f
->current_boot_id
);
2577 f
->current_xor_hash
= 0;
2580 void journal_file_save_location(JournalFile
*f
, Object
*o
, uint64_t offset
) {
2581 f
->location_type
= LOCATION_SEEK
;
2582 f
->current_offset
= offset
;
2583 f
->current_seqnum
= le64toh(o
->entry
.seqnum
);
2584 f
->current_realtime
= le64toh(o
->entry
.realtime
);
2585 f
->current_monotonic
= le64toh(o
->entry
.monotonic
);
2586 f
->current_boot_id
= o
->entry
.boot_id
;
2587 f
->current_xor_hash
= le64toh(o
->entry
.xor_hash
);
2590 int journal_file_compare_locations(JournalFile
*af
, JournalFile
*bf
) {
2597 assert(af
->location_type
== LOCATION_SEEK
);
2598 assert(bf
->location_type
== LOCATION_SEEK
);
2600 /* If contents and timestamps match, these entries are
2601 * identical, even if the seqnum does not match */
2602 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
) &&
2603 af
->current_monotonic
== bf
->current_monotonic
&&
2604 af
->current_realtime
== bf
->current_realtime
&&
2605 af
->current_xor_hash
== bf
->current_xor_hash
)
2608 if (sd_id128_equal(af
->header
->seqnum_id
, bf
->header
->seqnum_id
)) {
2610 /* If this is from the same seqnum source, compare
2612 r
= CMP(af
->current_seqnum
, bf
->current_seqnum
);
2616 /* Wow! This is weird, different data but the same
2617 * seqnums? Something is borked, but let's make the
2618 * best of it and compare by time. */
2621 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
)) {
2623 /* If the boot id matches, compare monotonic time */
2624 r
= CMP(af
->current_monotonic
, bf
->current_monotonic
);
2629 /* Otherwise, compare UTC time */
2630 r
= CMP(af
->current_realtime
, bf
->current_realtime
);
2634 /* Finally, compare by contents */
2635 return CMP(af
->current_xor_hash
, bf
->current_xor_hash
);
2638 static int bump_array_index(uint64_t *i
, direction_t direction
, uint64_t n
) {
2640 /* Increase or decrease the specified index, in the right direction. */
2642 if (direction
== DIRECTION_DOWN
) {
2657 static bool check_properly_ordered(uint64_t new_offset
, uint64_t old_offset
, direction_t direction
) {
2659 /* Consider it an error if any of the two offsets is uninitialized */
2660 if (old_offset
== 0 || new_offset
== 0)
2663 /* If we go down, the new offset must be larger than the old one. */
2664 return direction
== DIRECTION_DOWN
?
2665 new_offset
> old_offset
:
2666 new_offset
< old_offset
;
2669 int journal_file_next_entry(
2672 direction_t direction
,
2673 Object
**ret
, uint64_t *offset
) {
2681 n
= le64toh(f
->header
->n_entries
);
2686 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2688 r
= generic_array_bisect(f
,
2689 le64toh(f
->header
->entry_array_offset
),
2690 le64toh(f
->header
->n_entries
),
2699 r
= bump_array_index(&i
, direction
, n
);
2704 /* And jump to it */
2706 r
= generic_array_get(f
,
2707 le64toh(f
->header
->entry_array_offset
),
2715 /* OK, so this entry is borked. Most likely some entry didn't get synced to disk properly, let's see if
2716 * the next one might work for us instead. */
2717 log_debug_errno(r
, "Entry item %" PRIu64
" is bad, skipping over it.", i
);
2719 r
= bump_array_index(&i
, direction
, n
);
2724 /* Ensure our array is properly ordered. */
2725 if (p
> 0 && !check_properly_ordered(ofs
, p
, direction
))
2726 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
2727 "%s: entry array not properly ordered at entry %" PRIu64
,
2736 int journal_file_next_entry_for_data(
2738 Object
*o
, uint64_t p
,
2739 uint64_t data_offset
,
2740 direction_t direction
,
2741 Object
**ret
, uint64_t *offset
) {
2748 assert(p
> 0 || !o
);
2750 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2754 n
= le64toh(d
->data
.n_entries
);
2759 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2761 if (o
->object
.type
!= OBJECT_ENTRY
)
2764 r
= generic_array_bisect_plus_one(f
,
2765 le64toh(d
->data
.entry_offset
),
2766 le64toh(d
->data
.entry_array_offset
),
2767 le64toh(d
->data
.n_entries
),
2777 r
= bump_array_index(&i
, direction
, n
);
2783 r
= generic_array_get_plus_one(f
,
2784 le64toh(d
->data
.entry_offset
),
2785 le64toh(d
->data
.entry_array_offset
),
2793 log_debug_errno(r
, "Data entry item %" PRIu64
" is bad, skipping over it.", i
);
2795 r
= bump_array_index(&i
, direction
, n
);
2800 /* Ensure our array is properly ordered. */
2801 if (p
> 0 && check_properly_ordered(ofs
, p
, direction
))
2802 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
2803 "%s data entry array not properly ordered at entry %" PRIu64
,
2812 int journal_file_move_to_entry_by_offset_for_data(
2814 uint64_t data_offset
,
2816 direction_t direction
,
2817 Object
**ret
, uint64_t *offset
) {
2824 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2828 return generic_array_bisect_plus_one(f
,
2829 le64toh(d
->data
.entry_offset
),
2830 le64toh(d
->data
.entry_array_offset
),
2831 le64toh(d
->data
.n_entries
),
2838 int journal_file_move_to_entry_by_monotonic_for_data(
2840 uint64_t data_offset
,
2843 direction_t direction
,
2844 Object
**ret
, uint64_t *offset
) {
2852 /* First, seek by time */
2853 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &b
);
2859 r
= generic_array_bisect_plus_one(f
,
2860 le64toh(o
->data
.entry_offset
),
2861 le64toh(o
->data
.entry_array_offset
),
2862 le64toh(o
->data
.n_entries
),
2864 test_object_monotonic
,
2870 /* And now, continue seeking until we find an entry that
2871 * exists in both bisection arrays */
2877 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2881 r
= generic_array_bisect_plus_one(f
,
2882 le64toh(d
->data
.entry_offset
),
2883 le64toh(d
->data
.entry_array_offset
),
2884 le64toh(d
->data
.n_entries
),
2892 r
= journal_file_move_to_object(f
, OBJECT_DATA
, b
, &o
);
2896 r
= generic_array_bisect_plus_one(f
,
2897 le64toh(o
->data
.entry_offset
),
2898 le64toh(o
->data
.entry_array_offset
),
2899 le64toh(o
->data
.n_entries
),
2921 int journal_file_move_to_entry_by_seqnum_for_data(
2923 uint64_t data_offset
,
2925 direction_t direction
,
2926 Object
**ret
, uint64_t *offset
) {
2933 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2937 return generic_array_bisect_plus_one(f
,
2938 le64toh(d
->data
.entry_offset
),
2939 le64toh(d
->data
.entry_array_offset
),
2940 le64toh(d
->data
.n_entries
),
2947 int journal_file_move_to_entry_by_realtime_for_data(
2949 uint64_t data_offset
,
2951 direction_t direction
,
2952 Object
**ret
, uint64_t *offset
) {
2959 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2963 return generic_array_bisect_plus_one(f
,
2964 le64toh(d
->data
.entry_offset
),
2965 le64toh(d
->data
.entry_array_offset
),
2966 le64toh(d
->data
.n_entries
),
2968 test_object_realtime
,
2973 void journal_file_dump(JournalFile
*f
) {
2981 journal_file_print_header(f
);
2983 p
= le64toh(f
->header
->header_size
);
2985 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &o
);
2989 switch (o
->object
.type
) {
2992 printf("Type: OBJECT_UNUSED\n");
2996 printf("Type: OBJECT_DATA\n");
3000 printf("Type: OBJECT_FIELD\n");
3004 printf("Type: OBJECT_ENTRY seqnum=%"PRIu64
" monotonic=%"PRIu64
" realtime=%"PRIu64
"\n",
3005 le64toh(o
->entry
.seqnum
),
3006 le64toh(o
->entry
.monotonic
),
3007 le64toh(o
->entry
.realtime
));
3010 case OBJECT_FIELD_HASH_TABLE
:
3011 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
3014 case OBJECT_DATA_HASH_TABLE
:
3015 printf("Type: OBJECT_DATA_HASH_TABLE\n");
3018 case OBJECT_ENTRY_ARRAY
:
3019 printf("Type: OBJECT_ENTRY_ARRAY\n");
3023 printf("Type: OBJECT_TAG seqnum=%"PRIu64
" epoch=%"PRIu64
"\n",
3024 le64toh(o
->tag
.seqnum
),
3025 le64toh(o
->tag
.epoch
));
3029 printf("Type: unknown (%i)\n", o
->object
.type
);
3033 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
)
3034 printf("Flags: %s\n",
3035 object_compressed_to_string(o
->object
.flags
& OBJECT_COMPRESSION_MASK
));
3037 if (p
== le64toh(f
->header
->tail_object_offset
))
3040 p
= p
+ ALIGN64(le64toh(o
->object
.size
));
3045 log_error("File corrupt");
3048 static const char* format_timestamp_safe(char *buf
, size_t l
, usec_t t
) {
3051 x
= format_timestamp(buf
, l
, t
);
3057 void journal_file_print_header(JournalFile
*f
) {
3058 char a
[33], b
[33], c
[33], d
[33];
3059 char x
[FORMAT_TIMESTAMP_MAX
], y
[FORMAT_TIMESTAMP_MAX
], z
[FORMAT_TIMESTAMP_MAX
];
3061 char bytes
[FORMAT_BYTES_MAX
];
3066 printf("File Path: %s\n"
3070 "Sequential Number ID: %s\n"
3072 "Compatible Flags:%s%s\n"
3073 "Incompatible Flags:%s%s%s\n"
3074 "Header size: %"PRIu64
"\n"
3075 "Arena size: %"PRIu64
"\n"
3076 "Data Hash Table Size: %"PRIu64
"\n"
3077 "Field Hash Table Size: %"PRIu64
"\n"
3078 "Rotate Suggested: %s\n"
3079 "Head Sequential Number: %"PRIu64
" (%"PRIx64
")\n"
3080 "Tail Sequential Number: %"PRIu64
" (%"PRIx64
")\n"
3081 "Head Realtime Timestamp: %s (%"PRIx64
")\n"
3082 "Tail Realtime Timestamp: %s (%"PRIx64
")\n"
3083 "Tail Monotonic Timestamp: %s (%"PRIx64
")\n"
3084 "Objects: %"PRIu64
"\n"
3085 "Entry Objects: %"PRIu64
"\n",
3087 sd_id128_to_string(f
->header
->file_id
, a
),
3088 sd_id128_to_string(f
->header
->machine_id
, b
),
3089 sd_id128_to_string(f
->header
->boot_id
, c
),
3090 sd_id128_to_string(f
->header
->seqnum_id
, d
),
3091 f
->header
->state
== STATE_OFFLINE
? "OFFLINE" :
3092 f
->header
->state
== STATE_ONLINE
? "ONLINE" :
3093 f
->header
->state
== STATE_ARCHIVED
? "ARCHIVED" : "UNKNOWN",
3094 JOURNAL_HEADER_SEALED(f
->header
) ? " SEALED" : "",
3095 (le32toh(f
->header
->compatible_flags
) & ~HEADER_COMPATIBLE_ANY
) ? " ???" : "",
3096 JOURNAL_HEADER_COMPRESSED_XZ(f
->header
) ? " COMPRESSED-XZ" : "",
3097 JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
) ? " COMPRESSED-LZ4" : "",
3098 (le32toh(f
->header
->incompatible_flags
) & ~HEADER_INCOMPATIBLE_ANY
) ? " ???" : "",
3099 le64toh(f
->header
->header_size
),
3100 le64toh(f
->header
->arena_size
),
3101 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3102 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
),
3103 yes_no(journal_file_rotate_suggested(f
, 0)),
3104 le64toh(f
->header
->head_entry_seqnum
), le64toh(f
->header
->head_entry_seqnum
),
3105 le64toh(f
->header
->tail_entry_seqnum
), le64toh(f
->header
->tail_entry_seqnum
),
3106 format_timestamp_safe(x
, sizeof(x
), le64toh(f
->header
->head_entry_realtime
)), le64toh(f
->header
->head_entry_realtime
),
3107 format_timestamp_safe(y
, sizeof(y
), le64toh(f
->header
->tail_entry_realtime
)), le64toh(f
->header
->tail_entry_realtime
),
3108 format_timespan(z
, sizeof(z
), le64toh(f
->header
->tail_entry_monotonic
), USEC_PER_MSEC
), le64toh(f
->header
->tail_entry_monotonic
),
3109 le64toh(f
->header
->n_objects
),
3110 le64toh(f
->header
->n_entries
));
3112 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3113 printf("Data Objects: %"PRIu64
"\n"
3114 "Data Hash Table Fill: %.1f%%\n",
3115 le64toh(f
->header
->n_data
),
3116 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))));
3118 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3119 printf("Field Objects: %"PRIu64
"\n"
3120 "Field Hash Table Fill: %.1f%%\n",
3121 le64toh(f
->header
->n_fields
),
3122 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))));
3124 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
))
3125 printf("Tag Objects: %"PRIu64
"\n",
3126 le64toh(f
->header
->n_tags
));
3127 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
3128 printf("Entry Array Objects: %"PRIu64
"\n",
3129 le64toh(f
->header
->n_entry_arrays
));
3131 if (fstat(f
->fd
, &st
) >= 0)
3132 printf("Disk usage: %s\n", format_bytes(bytes
, sizeof(bytes
), (uint64_t) st
.st_blocks
* 512ULL));
3135 static int journal_file_warn_btrfs(JournalFile
*f
) {
3141 /* Before we write anything, check if the COW logic is turned
3142 * off on btrfs. Given our write pattern that is quite
3143 * unfriendly to COW file systems this should greatly improve
3144 * performance on COW file systems, such as btrfs, at the
3145 * expense of data integrity features (which shouldn't be too
3146 * bad, given that we do our own checksumming). */
3148 r
= btrfs_is_filesystem(f
->fd
);
3150 return log_warning_errno(r
, "Failed to determine if journal is on btrfs: %m");
3154 r
= read_attr_fd(f
->fd
, &attrs
);
3156 return log_warning_errno(r
, "Failed to read file attributes: %m");
3158 if (attrs
& FS_NOCOW_FL
) {
3159 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3163 log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3164 "This is likely to slow down journal access substantially, please consider turning "
3165 "off the copy-on-write file attribute on the journal directory, using chattr +C.", f
->path
);
3170 int journal_file_open(
3176 uint64_t compress_threshold_bytes
,
3178 JournalMetrics
*metrics
,
3179 MMapCache
*mmap_cache
,
3180 Set
*deferred_closes
,
3181 JournalFile
*template,
3182 JournalFile
**ret
) {
3184 bool newly_created
= false;
3190 assert(fd
>= 0 || fname
);
3192 if (!IN_SET((flags
& O_ACCMODE
), O_RDONLY
, O_RDWR
))
3195 if (fname
&& (flags
& O_CREAT
) && !endswith(fname
, ".journal"))
3198 f
= new(JournalFile
, 1);
3202 *f
= (JournalFile
) {
3207 .prot
= prot_from_flags(flags
),
3208 .writable
= (flags
& O_ACCMODE
) != O_RDONLY
,
3211 .compress_lz4
= compress
,
3213 .compress_xz
= compress
,
3215 .compress_threshold_bytes
= compress_threshold_bytes
== (uint64_t) -1 ?
3216 DEFAULT_COMPRESS_THRESHOLD
:
3217 MAX(MIN_COMPRESS_THRESHOLD
, compress_threshold_bytes
),
3223 if (DEBUG_LOGGING
) {
3224 static int last_seal
= -1, last_compress
= -1;
3225 static uint64_t last_bytes
= UINT64_MAX
;
3226 char bytes
[FORMAT_BYTES_MAX
];
3228 if (last_seal
!= f
->seal
||
3229 last_compress
!= JOURNAL_FILE_COMPRESS(f
) ||
3230 last_bytes
!= f
->compress_threshold_bytes
) {
3232 log_debug("Journal effective settings seal=%s compress=%s compress_threshold_bytes=%s",
3233 yes_no(f
->seal
), yes_no(JOURNAL_FILE_COMPRESS(f
)),
3234 format_bytes(bytes
, sizeof bytes
, f
->compress_threshold_bytes
));
3235 last_seal
= f
->seal
;
3236 last_compress
= JOURNAL_FILE_COMPRESS(f
);
3237 last_bytes
= f
->compress_threshold_bytes
;
3242 f
->mmap
= mmap_cache_ref(mmap_cache
);
3244 f
->mmap
= mmap_cache_new();
3252 f
->path
= strdup(fname
);
3260 /* If we don't know the path, fill in something explanatory and vaguely useful */
3261 if (asprintf(&f
->path
, "/proc/self/%i", fd
) < 0) {
3267 f
->chain_cache
= ordered_hashmap_new(&uint64_hash_ops
);
3268 if (!f
->chain_cache
) {
3274 /* We pass O_NONBLOCK here, so that in case somebody pointed us to some character device node or FIFO
3275 * or so, we likely fail quickly than block for long. For regular files O_NONBLOCK has no effect, hence
3276 * it doesn't hurt in that case. */
3278 f
->fd
= open(f
->path
, f
->flags
|O_CLOEXEC
|O_NONBLOCK
, f
->mode
);
3284 /* fds we opened here by us should also be closed by us. */
3287 r
= fd_nonblock(f
->fd
, false);
3292 f
->cache_fd
= mmap_cache_add_fd(f
->mmap
, f
->fd
);
3298 r
= journal_file_fstat(f
);
3302 if (f
->last_stat
.st_size
== 0 && f
->writable
) {
3304 (void) journal_file_warn_btrfs(f
);
3306 /* Let's attach the creation time to the journal file, so that the vacuuming code knows the age of this
3307 * file even if the file might end up corrupted one day... Ideally we'd just use the creation time many
3308 * file systems maintain for each file, but the API to query this is very new, hence let's emulate this
3309 * via extended attributes. If extended attributes are not supported we'll just skip this, and rely
3310 * solely on mtime/atime/ctime of the file. */
3311 (void) fd_setcrtime(f
->fd
, 0);
3314 /* Try to load the FSPRG state, and if we can't, then
3315 * just don't do sealing */
3317 r
= journal_file_fss_load(f
);
3323 r
= journal_file_init_header(f
, template);
3327 r
= journal_file_fstat(f
);
3331 newly_created
= true;
3334 if (f
->last_stat
.st_size
< (off_t
) HEADER_SIZE_MIN
) {
3339 r
= mmap_cache_get(f
->mmap
, f
->cache_fd
, f
->prot
, CONTEXT_HEADER
, true, 0, PAGE_ALIGN(sizeof(Header
)), &f
->last_stat
, &h
, NULL
);
3341 /* Some file systems (jffs2 or p9fs) don't support mmap() properly (or only read-only
3342 * mmap()), and return EINVAL in that case. Let's propagate that as a more recognizable error
3352 if (!newly_created
) {
3353 set_clear_with_destructor(deferred_closes
, journal_file_close
);
3355 r
= journal_file_verify_header(f
);
3361 if (!newly_created
&& f
->writable
) {
3362 r
= journal_file_fss_load(f
);
3370 journal_default_metrics(metrics
, f
->fd
);
3371 f
->metrics
= *metrics
;
3372 } else if (template)
3373 f
->metrics
= template->metrics
;
3375 r
= journal_file_refresh_header(f
);
3381 r
= journal_file_hmac_setup(f
);
3386 if (newly_created
) {
3387 r
= journal_file_setup_field_hash_table(f
);
3391 r
= journal_file_setup_data_hash_table(f
);
3396 r
= journal_file_append_first_tag(f
);
3402 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
)) {
3407 if (template && template->post_change_timer
) {
3408 r
= journal_file_enable_post_change_timer(
3410 sd_event_source_get_event(template->post_change_timer
),
3411 template->post_change_timer_period
);
3417 /* The file is opened now successfully, thus we take possession of any passed in fd. */
3424 if (f
->cache_fd
&& mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
3427 (void) journal_file_close(f
);
3432 int journal_file_archive(JournalFile
*f
) {
3433 _cleanup_free_
char *p
= NULL
;
3440 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
3441 * rotation, since we don't know the actual path, and couldn't rename the file hence. */
3442 if (path_startswith(f
->path
, "/proc/self/fd"))
3445 if (!endswith(f
->path
, ".journal"))
3448 if (asprintf(&p
, "%.*s@" SD_ID128_FORMAT_STR
"-%016"PRIx64
"-%016"PRIx64
".journal",
3449 (int) strlen(f
->path
) - 8, f
->path
,
3450 SD_ID128_FORMAT_VAL(f
->header
->seqnum_id
),
3451 le64toh(f
->header
->head_entry_seqnum
),
3452 le64toh(f
->header
->head_entry_realtime
)) < 0)
3455 /* Try to rename the file to the archived version. If the file already was deleted, we'll get ENOENT, let's
3456 * ignore that case. */
3457 if (rename(f
->path
, p
) < 0 && errno
!= ENOENT
)
3460 /* Sync the rename to disk */
3461 (void) fsync_directory_of_file(f
->fd
);
3463 /* Set as archive so offlining commits w/state=STATE_ARCHIVED. Previously we would set old_file->header->state
3464 * to STATE_ARCHIVED directly here, but journal_file_set_offline() short-circuits when state != STATE_ONLINE,
3465 * which would result in the rotated journal never getting fsync() called before closing. Now we simply queue
3466 * the archive state by setting an archive bit, leaving the state as STATE_ONLINE so proper offlining
3470 /* Currently, btrfs is not very good with out write patterns and fragments heavily. Let's defrag our journal
3471 * files when we archive them */
3472 f
->defrag_on_close
= true;
3477 JournalFile
* journal_initiate_close(
3479 Set
*deferred_closes
) {
3485 if (deferred_closes
) {
3487 r
= set_put(deferred_closes
, f
);
3489 log_debug_errno(r
, "Failed to add file to deferred close set, closing immediately.");
3491 (void) journal_file_set_offline(f
, false);
3496 return journal_file_close(f
);
3499 int journal_file_rotate(
3502 uint64_t compress_threshold_bytes
,
3504 Set
*deferred_closes
) {
3506 JournalFile
*new_file
= NULL
;
3512 r
= journal_file_archive(*f
);
3516 r
= journal_file_open(
3522 compress_threshold_bytes
,
3530 journal_initiate_close(*f
, deferred_closes
);
3536 int journal_file_dispose(int dir_fd
, const char *fname
) {
3537 _cleanup_free_
char *p
= NULL
;
3538 _cleanup_close_
int fd
= -1;
3542 /* Renames a journal file to *.journal~, i.e. to mark it as corruped or otherwise uncleanly shutdown. Note that
3543 * this is done without looking into the file or changing any of its contents. The idea is that this is called
3544 * whenever something is suspicious and we want to move the file away and make clear that it is not accessed
3545 * for writing anymore. */
3547 if (!endswith(fname
, ".journal"))
3550 if (asprintf(&p
, "%.*s@%016" PRIx64
"-%016" PRIx64
".journal~",
3551 (int) strlen(fname
) - 8, fname
,
3552 now(CLOCK_REALTIME
),
3556 if (renameat(dir_fd
, fname
, dir_fd
, p
) < 0)
3559 /* btrfs doesn't cope well with our write pattern and fragments heavily. Let's defrag all files we rotate */
3560 fd
= openat(dir_fd
, p
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
|O_NOFOLLOW
);
3562 log_debug_errno(errno
, "Failed to open file for defragmentation/FS_NOCOW_FL, ignoring: %m");
3564 (void) chattr_fd(fd
, 0, FS_NOCOW_FL
, NULL
);
3565 (void) btrfs_defrag_fd(fd
);
3571 int journal_file_open_reliably(
3576 uint64_t compress_threshold_bytes
,
3578 JournalMetrics
*metrics
,
3579 MMapCache
*mmap_cache
,
3580 Set
*deferred_closes
,
3581 JournalFile
*template,
3582 JournalFile
**ret
) {
3586 r
= journal_file_open(-1, fname
, flags
, mode
, compress
, compress_threshold_bytes
, seal
, metrics
, mmap_cache
,
3587 deferred_closes
, template, ret
);
3589 -EBADMSG
, /* Corrupted */
3590 -ENODATA
, /* Truncated */
3591 -EHOSTDOWN
, /* Other machine */
3592 -EPROTONOSUPPORT
, /* Incompatible feature */
3593 -EBUSY
, /* Unclean shutdown */
3594 -ESHUTDOWN
, /* Already archived */
3595 -EIO
, /* IO error, including SIGBUS on mmap */
3596 -EIDRM
, /* File has been deleted */
3597 -ETXTBSY
)) /* File is from the future */
3600 if ((flags
& O_ACCMODE
) == O_RDONLY
)
3603 if (!(flags
& O_CREAT
))
3606 if (!endswith(fname
, ".journal"))
3609 /* The file is corrupted. Rotate it away and try it again (but only once) */
3610 log_warning_errno(r
, "File %s corrupted or uncleanly shut down, renaming and replacing.", fname
);
3612 r
= journal_file_dispose(AT_FDCWD
, fname
);
3616 return journal_file_open(-1, fname
, flags
, mode
, compress
, compress_threshold_bytes
, seal
, metrics
, mmap_cache
,
3617 deferred_closes
, template, ret
);
3620 int journal_file_copy_entry(JournalFile
*from
, JournalFile
*to
, Object
*o
, uint64_t p
) {
3622 uint64_t q
, xor_hash
= 0;
3626 const sd_id128_t
*boot_id
;
3636 ts
.monotonic
= le64toh(o
->entry
.monotonic
);
3637 ts
.realtime
= le64toh(o
->entry
.realtime
);
3638 boot_id
= &o
->entry
.boot_id
;
3640 n
= journal_file_entry_n_items(o
);
3641 /* alloca() can't take 0, hence let's allocate at least one */
3642 items
= newa(EntryItem
, MAX(1u, n
));
3644 for (i
= 0; i
< n
; i
++) {
3651 q
= le64toh(o
->entry
.items
[i
].object_offset
);
3652 le_hash
= o
->entry
.items
[i
].hash
;
3654 r
= journal_file_move_to_object(from
, OBJECT_DATA
, q
, &o
);
3658 if (le_hash
!= o
->data
.hash
)
3661 l
= le64toh(o
->object
.size
) - offsetof(Object
, data
.payload
);
3664 /* We hit the limit on 32bit machines */
3665 if ((uint64_t) t
!= l
)
3668 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
3669 #if HAVE_XZ || HAVE_LZ4
3672 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
3673 o
->data
.payload
, l
, &from
->compress_buffer
, &from
->compress_buffer_size
, &rsize
, 0);
3677 data
= from
->compress_buffer
;
3680 return -EPROTONOSUPPORT
;
3683 data
= o
->data
.payload
;
3685 r
= journal_file_append_data(to
, data
, l
, &u
, &h
);
3689 xor_hash
^= le64toh(u
->data
.hash
);
3690 items
[i
].object_offset
= htole64(h
);
3691 items
[i
].hash
= u
->data
.hash
;
3693 r
= journal_file_move_to_object(from
, OBJECT_ENTRY
, p
, &o
);
3698 r
= journal_file_append_entry_internal(to
, &ts
, boot_id
, xor_hash
, items
, n
,
3701 if (mmap_cache_got_sigbus(to
->mmap
, to
->cache_fd
))
3707 void journal_reset_metrics(JournalMetrics
*m
) {
3710 /* Set everything to "pick automatic values". */
3712 *m
= (JournalMetrics
) {
3713 .min_use
= (uint64_t) -1,
3714 .max_use
= (uint64_t) -1,
3715 .min_size
= (uint64_t) -1,
3716 .max_size
= (uint64_t) -1,
3717 .keep_free
= (uint64_t) -1,
3718 .n_max_files
= (uint64_t) -1,
3722 void journal_default_metrics(JournalMetrics
*m
, int fd
) {
3723 char a
[FORMAT_BYTES_MAX
], b
[FORMAT_BYTES_MAX
], c
[FORMAT_BYTES_MAX
], d
[FORMAT_BYTES_MAX
], e
[FORMAT_BYTES_MAX
];
3730 if (fstatvfs(fd
, &ss
) >= 0)
3731 fs_size
= ss
.f_frsize
* ss
.f_blocks
;
3733 log_debug_errno(errno
, "Failed to determine disk size: %m");
3737 if (m
->max_use
== (uint64_t) -1) {
3740 m
->max_use
= PAGE_ALIGN(fs_size
/ 10); /* 10% of file system size */
3742 if (m
->max_use
> DEFAULT_MAX_USE_UPPER
)
3743 m
->max_use
= DEFAULT_MAX_USE_UPPER
;
3745 if (m
->max_use
< DEFAULT_MAX_USE_LOWER
)
3746 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3748 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3750 m
->max_use
= PAGE_ALIGN(m
->max_use
);
3752 if (m
->max_use
!= 0 && m
->max_use
< JOURNAL_FILE_SIZE_MIN
*2)
3753 m
->max_use
= JOURNAL_FILE_SIZE_MIN
*2;
3756 if (m
->min_use
== (uint64_t) -1)
3757 m
->min_use
= DEFAULT_MIN_USE
;
3759 if (m
->min_use
> m
->max_use
)
3760 m
->min_use
= m
->max_use
;
3762 if (m
->max_size
== (uint64_t) -1) {
3763 m
->max_size
= PAGE_ALIGN(m
->max_use
/ 8); /* 8 chunks */
3765 if (m
->max_size
> DEFAULT_MAX_SIZE_UPPER
)
3766 m
->max_size
= DEFAULT_MAX_SIZE_UPPER
;
3768 m
->max_size
= PAGE_ALIGN(m
->max_size
);
3770 if (m
->max_size
!= 0) {
3771 if (m
->max_size
< JOURNAL_FILE_SIZE_MIN
)
3772 m
->max_size
= JOURNAL_FILE_SIZE_MIN
;
3774 if (m
->max_use
!= 0 && m
->max_size
*2 > m
->max_use
)
3775 m
->max_use
= m
->max_size
*2;
3778 if (m
->min_size
== (uint64_t) -1)
3779 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3781 m
->min_size
= PAGE_ALIGN(m
->min_size
);
3783 if (m
->min_size
< JOURNAL_FILE_SIZE_MIN
)
3784 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3786 if (m
->max_size
!= 0 && m
->min_size
> m
->max_size
)
3787 m
->max_size
= m
->min_size
;
3790 if (m
->keep_free
== (uint64_t) -1) {
3793 m
->keep_free
= PAGE_ALIGN(fs_size
* 3 / 20); /* 15% of file system size */
3795 if (m
->keep_free
> DEFAULT_KEEP_FREE_UPPER
)
3796 m
->keep_free
= DEFAULT_KEEP_FREE_UPPER
;
3799 m
->keep_free
= DEFAULT_KEEP_FREE
;
3802 if (m
->n_max_files
== (uint64_t) -1)
3803 m
->n_max_files
= DEFAULT_N_MAX_FILES
;
3805 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64
,
3806 format_bytes(a
, sizeof(a
), m
->min_use
),
3807 format_bytes(b
, sizeof(b
), m
->max_use
),
3808 format_bytes(c
, sizeof(c
), m
->max_size
),
3809 format_bytes(d
, sizeof(d
), m
->min_size
),
3810 format_bytes(e
, sizeof(e
), m
->keep_free
),
3814 int journal_file_get_cutoff_realtime_usec(JournalFile
*f
, usec_t
*from
, usec_t
*to
) {
3820 if (f
->header
->head_entry_realtime
== 0)
3823 *from
= le64toh(f
->header
->head_entry_realtime
);
3827 if (f
->header
->tail_entry_realtime
== 0)
3830 *to
= le64toh(f
->header
->tail_entry_realtime
);
3836 int journal_file_get_cutoff_monotonic_usec(JournalFile
*f
, sd_id128_t boot_id
, usec_t
*from
, usec_t
*to
) {
3844 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &p
);
3848 if (le64toh(o
->data
.n_entries
) <= 0)
3852 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, le64toh(o
->data
.entry_offset
), &o
);
3856 *from
= le64toh(o
->entry
.monotonic
);
3860 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
3864 r
= generic_array_get_plus_one(f
,
3865 le64toh(o
->data
.entry_offset
),
3866 le64toh(o
->data
.entry_array_offset
),
3867 le64toh(o
->data
.n_entries
)-1,
3872 *to
= le64toh(o
->entry
.monotonic
);
3878 bool journal_file_rotate_suggested(JournalFile
*f
, usec_t max_file_usec
) {
3882 /* If we gained new header fields we gained new features,
3883 * hence suggest a rotation */
3884 if (le64toh(f
->header
->header_size
) < sizeof(Header
)) {
3885 log_debug("%s uses an outdated header, suggesting rotation.", f
->path
);
3889 /* Let's check if the hash tables grew over a certain fill
3890 * level (75%, borrowing this value from Java's hash table
3891 * implementation), and if so suggest a rotation. To calculate
3892 * the fill level we need the n_data field, which only exists
3893 * in newer versions. */
3895 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3896 if (le64toh(f
->header
->n_data
) * 4ULL > (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3897 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.",
3899 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))),
3900 le64toh(f
->header
->n_data
),
3901 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3902 (unsigned long long) f
->last_stat
.st_size
,
3903 f
->last_stat
.st_size
/ le64toh(f
->header
->n_data
));
3907 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3908 if (le64toh(f
->header
->n_fields
) * 4ULL > (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3909 log_debug("Field hash table of %s has a fill level at %.1f (%"PRIu64
" of %"PRIu64
" items), suggesting rotation.",
3911 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))),
3912 le64toh(f
->header
->n_fields
),
3913 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
));
3917 /* Are the data objects properly indexed by field objects? */
3918 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
3919 JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
3920 le64toh(f
->header
->n_data
) > 0 &&
3921 le64toh(f
->header
->n_fields
) == 0)
3924 if (max_file_usec
> 0) {
3927 h
= le64toh(f
->header
->head_entry_realtime
);
3928 t
= now(CLOCK_REALTIME
);
3930 if (h
> 0 && t
> h
+ max_file_usec
)