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"
21 #include "journal-authenticate.h"
22 #include "journal-def.h"
23 #include "journal-file.h"
25 #include "memory-util.h"
26 #include "parse-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
) {
346 /* Write the final tag */
347 if (f
->seal
&& f
->writable
) {
350 r
= journal_file_append_tag(f
);
352 log_error_errno(r
, "Failed to append tag when closing journal: %m");
356 if (f
->post_change_timer
) {
357 if (sd_event_source_get_enabled(f
->post_change_timer
, NULL
) > 0)
358 journal_file_post_change(f
);
360 sd_event_source_disable_unref(f
->post_change_timer
);
363 journal_file_set_offline(f
, true);
365 if (f
->mmap
&& f
->cache_fd
)
366 mmap_cache_free_fd(f
->mmap
, f
->cache_fd
);
368 if (f
->fd
>= 0 && f
->defrag_on_close
) {
370 /* Be friendly to btrfs: turn COW back on again now,
371 * and defragment the file. We won't write to the file
372 * ever again, hence remove all fragmentation, and
373 * reenable all the good bits COW usually provides
374 * (such as data checksumming). */
376 (void) chattr_fd(f
->fd
, 0, FS_NOCOW_FL
, NULL
);
377 (void) btrfs_defrag_fd(f
->fd
);
384 mmap_cache_unref(f
->mmap
);
386 ordered_hashmap_free_free(f
->chain_cache
);
388 #if HAVE_XZ || HAVE_LZ4
389 free(f
->compress_buffer
);
394 munmap(f
->fss_file
, PAGE_ALIGN(f
->fss_file_size
));
396 free(f
->fsprg_state
);
401 gcry_md_close(f
->hmac
);
407 static int journal_file_init_header(JournalFile
*f
, JournalFile
*template) {
414 memcpy(h
.signature
, HEADER_SIGNATURE
, 8);
415 h
.header_size
= htole64(ALIGN64(sizeof(h
)));
417 h
.incompatible_flags
|= htole32(
418 f
->compress_xz
* HEADER_INCOMPATIBLE_COMPRESSED_XZ
|
419 f
->compress_lz4
* HEADER_INCOMPATIBLE_COMPRESSED_LZ4
);
421 h
.compatible_flags
= htole32(
422 f
->seal
* HEADER_COMPATIBLE_SEALED
);
424 r
= sd_id128_randomize(&h
.file_id
);
429 h
.seqnum_id
= template->header
->seqnum_id
;
430 h
.tail_entry_seqnum
= template->header
->tail_entry_seqnum
;
432 h
.seqnum_id
= h
.file_id
;
434 k
= pwrite(f
->fd
, &h
, sizeof(h
), 0);
444 static int journal_file_refresh_header(JournalFile
*f
) {
451 r
= sd_id128_get_machine(&f
->header
->machine_id
);
452 if (IN_SET(r
, -ENOENT
, -ENOMEDIUM
))
453 /* We don't have a machine-id, let's continue without */
454 zero(f
->header
->machine_id
);
458 r
= sd_id128_get_boot(&boot_id
);
462 f
->header
->boot_id
= boot_id
;
464 r
= journal_file_set_online(f
);
466 /* Sync the online state to disk */
469 /* We likely just created a new file, also sync the directory this file is located in. */
470 (void) fsync_directory_of_file(f
->fd
);
475 static bool warn_wrong_flags(const JournalFile
*f
, bool compatible
) {
476 const uint32_t any
= compatible
? HEADER_COMPATIBLE_ANY
: HEADER_INCOMPATIBLE_ANY
,
477 supported
= compatible
? HEADER_COMPATIBLE_SUPPORTED
: HEADER_INCOMPATIBLE_SUPPORTED
;
478 const char *type
= compatible
? "compatible" : "incompatible";
481 flags
= le32toh(compatible
? f
->header
->compatible_flags
: f
->header
->incompatible_flags
);
483 if (flags
& ~supported
) {
485 log_debug("Journal file %s has unknown %s flags 0x%"PRIx32
,
486 f
->path
, type
, flags
& ~any
);
487 flags
= (flags
& any
) & ~supported
;
491 _cleanup_free_
char *t
= NULL
;
493 if (compatible
&& (flags
& HEADER_COMPATIBLE_SEALED
))
494 strv
[n
++] = "sealed";
495 if (!compatible
&& (flags
& HEADER_INCOMPATIBLE_COMPRESSED_XZ
))
496 strv
[n
++] = "xz-compressed";
497 if (!compatible
&& (flags
& HEADER_INCOMPATIBLE_COMPRESSED_LZ4
))
498 strv
[n
++] = "lz4-compressed";
500 assert(n
< ELEMENTSOF(strv
));
502 t
= strv_join((char**) strv
, ", ");
503 log_debug("Journal file %s uses %s %s %s disabled at compilation time.",
504 f
->path
, type
, n
> 1 ? "flags" : "flag", strnull(t
));
512 static int journal_file_verify_header(JournalFile
*f
) {
513 uint64_t arena_size
, header_size
;
518 if (memcmp(f
->header
->signature
, HEADER_SIGNATURE
, 8))
521 /* In both read and write mode we refuse to open files with incompatible
522 * flags we don't know. */
523 if (warn_wrong_flags(f
, false))
524 return -EPROTONOSUPPORT
;
526 /* When open for writing we refuse to open files with compatible flags, too. */
527 if (f
->writable
&& warn_wrong_flags(f
, true))
528 return -EPROTONOSUPPORT
;
530 if (f
->header
->state
>= _STATE_MAX
)
533 header_size
= le64toh(f
->header
->header_size
);
535 /* The first addition was n_data, so check that we are at least this large */
536 if (header_size
< HEADER_SIZE_MIN
)
539 if (JOURNAL_HEADER_SEALED(f
->header
) && !JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
542 arena_size
= le64toh(f
->header
->arena_size
);
544 if (UINT64_MAX
- header_size
< arena_size
|| header_size
+ arena_size
> (uint64_t) f
->last_stat
.st_size
)
547 if (le64toh(f
->header
->tail_object_offset
) > header_size
+ arena_size
)
550 if (!VALID64(le64toh(f
->header
->data_hash_table_offset
)) ||
551 !VALID64(le64toh(f
->header
->field_hash_table_offset
)) ||
552 !VALID64(le64toh(f
->header
->tail_object_offset
)) ||
553 !VALID64(le64toh(f
->header
->entry_array_offset
)))
557 sd_id128_t machine_id
;
561 r
= sd_id128_get_machine(&machine_id
);
565 if (!sd_id128_equal(machine_id
, f
->header
->machine_id
))
568 state
= f
->header
->state
;
570 if (state
== STATE_ARCHIVED
)
571 return -ESHUTDOWN
; /* Already archived */
572 else if (state
== STATE_ONLINE
)
573 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
),
574 "Journal file %s is already online. Assuming unclean closing.",
576 else if (state
!= STATE_OFFLINE
)
577 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
),
578 "Journal file %s has unknown state %i.",
581 if (f
->header
->field_hash_table_size
== 0 || f
->header
->data_hash_table_size
== 0)
584 /* Don't permit appending to files from the future. Because otherwise the realtime timestamps wouldn't
585 * be strictly ordered in the entries in the file anymore, and we can't have that since it breaks
587 if (le64toh(f
->header
->tail_entry_realtime
) > now(CLOCK_REALTIME
))
588 return log_debug_errno(SYNTHETIC_ERRNO(ETXTBSY
),
589 "Journal file %s is from the future, refusing to append new data to it that'd be older.",
593 f
->compress_xz
= JOURNAL_HEADER_COMPRESSED_XZ(f
->header
);
594 f
->compress_lz4
= JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
);
596 f
->seal
= JOURNAL_HEADER_SEALED(f
->header
);
601 static int journal_file_fstat(JournalFile
*f
) {
607 if (fstat(f
->fd
, &f
->last_stat
) < 0)
610 f
->last_stat_usec
= now(CLOCK_MONOTONIC
);
612 /* Refuse dealing with with files that aren't regular */
613 r
= stat_verify_regular(&f
->last_stat
);
617 /* Refuse appending to files that are already deleted */
618 if (f
->last_stat
.st_nlink
<= 0)
624 static int journal_file_allocate(JournalFile
*f
, uint64_t offset
, uint64_t size
) {
625 uint64_t old_size
, new_size
;
631 /* We assume that this file is not sparse, and we know that
632 * for sure, since we always call posix_fallocate()
635 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
639 le64toh(f
->header
->header_size
) +
640 le64toh(f
->header
->arena_size
);
642 new_size
= PAGE_ALIGN(offset
+ size
);
643 if (new_size
< le64toh(f
->header
->header_size
))
644 new_size
= le64toh(f
->header
->header_size
);
646 if (new_size
<= old_size
) {
648 /* We already pre-allocated enough space, but before
649 * we write to it, let's check with fstat() if the
650 * file got deleted, in order make sure we don't throw
651 * away the data immediately. Don't check fstat() for
652 * all writes though, but only once ever 10s. */
654 if (f
->last_stat_usec
+ LAST_STAT_REFRESH_USEC
> now(CLOCK_MONOTONIC
))
657 return journal_file_fstat(f
);
660 /* Allocate more space. */
662 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
665 if (new_size
> f
->metrics
.min_size
&& f
->metrics
.keep_free
> 0) {
668 if (fstatvfs(f
->fd
, &svfs
) >= 0) {
671 available
= LESS_BY((uint64_t) svfs
.f_bfree
* (uint64_t) svfs
.f_bsize
, f
->metrics
.keep_free
);
673 if (new_size
- old_size
> available
)
678 /* Increase by larger blocks at once */
679 new_size
= DIV_ROUND_UP(new_size
, FILE_SIZE_INCREASE
) * FILE_SIZE_INCREASE
;
680 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
681 new_size
= f
->metrics
.max_size
;
683 /* Note that the glibc fallocate() fallback is very
684 inefficient, hence we try to minimize the allocation area
686 r
= posix_fallocate(f
->fd
, old_size
, new_size
- old_size
);
690 f
->header
->arena_size
= htole64(new_size
- le64toh(f
->header
->header_size
));
692 return journal_file_fstat(f
);
695 static unsigned type_to_context(ObjectType type
) {
696 /* One context for each type, plus one catch-all for the rest */
697 assert_cc(_OBJECT_TYPE_MAX
<= MMAP_CACHE_MAX_CONTEXTS
);
698 assert_cc(CONTEXT_HEADER
< MMAP_CACHE_MAX_CONTEXTS
);
699 return type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
? type
: 0;
702 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
) {
711 /* Avoid SIGBUS on invalid accesses */
712 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
) {
713 /* Hmm, out of range? Let's refresh the fstat() data
714 * first, before we trust that check. */
716 r
= journal_file_fstat(f
);
720 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
)
721 return -EADDRNOTAVAIL
;
724 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
);
727 static uint64_t minimum_header_size(Object
*o
) {
729 static const uint64_t table
[] = {
730 [OBJECT_DATA
] = sizeof(DataObject
),
731 [OBJECT_FIELD
] = sizeof(FieldObject
),
732 [OBJECT_ENTRY
] = sizeof(EntryObject
),
733 [OBJECT_DATA_HASH_TABLE
] = sizeof(HashTableObject
),
734 [OBJECT_FIELD_HASH_TABLE
] = sizeof(HashTableObject
),
735 [OBJECT_ENTRY_ARRAY
] = sizeof(EntryArrayObject
),
736 [OBJECT_TAG
] = sizeof(TagObject
),
739 if (o
->object
.type
>= ELEMENTSOF(table
) || table
[o
->object
.type
] <= 0)
740 return sizeof(ObjectHeader
);
742 return table
[o
->object
.type
];
745 /* Lightweight object checks. We want this to be fast, so that we won't
746 * slowdown every journal_file_move_to_object() call too much. */
747 static int journal_file_check_object(JournalFile
*f
, uint64_t offset
, Object
*o
) {
751 switch (o
->object
.type
) {
754 if ((le64toh(o
->data
.entry_offset
) == 0) ^ (le64toh(o
->data
.n_entries
) == 0))
755 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
756 "Bad n_entries: %" PRIu64
": %" PRIu64
,
757 le64toh(o
->data
.n_entries
),
760 if (le64toh(o
->object
.size
) - offsetof(DataObject
, payload
) <= 0)
761 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
762 "Bad object size (<= %zu): %" PRIu64
": %" PRIu64
,
763 offsetof(DataObject
, payload
),
764 le64toh(o
->object
.size
),
767 if (!VALID64(le64toh(o
->data
.next_hash_offset
)) ||
768 !VALID64(le64toh(o
->data
.next_field_offset
)) ||
769 !VALID64(le64toh(o
->data
.entry_offset
)) ||
770 !VALID64(le64toh(o
->data
.entry_array_offset
)))
771 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
772 "Invalid offset, next_hash_offset=" OFSfmt
", next_field_offset=" OFSfmt
", entry_offset=" OFSfmt
", entry_array_offset=" OFSfmt
": %" PRIu64
,
773 le64toh(o
->data
.next_hash_offset
),
774 le64toh(o
->data
.next_field_offset
),
775 le64toh(o
->data
.entry_offset
),
776 le64toh(o
->data
.entry_array_offset
),
783 if (le64toh(o
->object
.size
) - offsetof(FieldObject
, payload
) <= 0)
784 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
785 "Bad field size (<= %zu): %" PRIu64
": %" PRIu64
,
786 offsetof(FieldObject
, payload
),
787 le64toh(o
->object
.size
),
790 if (!VALID64(le64toh(o
->field
.next_hash_offset
)) ||
791 !VALID64(le64toh(o
->field
.head_data_offset
)))
792 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
793 "Invalid offset, next_hash_offset=" OFSfmt
", head_data_offset=" OFSfmt
": %" PRIu64
,
794 le64toh(o
->field
.next_hash_offset
),
795 le64toh(o
->field
.head_data_offset
),
800 if ((le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) % sizeof(EntryItem
) != 0)
801 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
802 "Bad entry size (<= %zu): %" PRIu64
": %" PRIu64
,
803 offsetof(EntryObject
, items
),
804 le64toh(o
->object
.size
),
807 if ((le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) / sizeof(EntryItem
) <= 0)
808 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
809 "Invalid number items in entry: %" PRIu64
": %" PRIu64
,
810 (le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) / sizeof(EntryItem
),
813 if (le64toh(o
->entry
.seqnum
) <= 0)
814 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
815 "Invalid entry seqnum: %" PRIx64
": %" PRIu64
,
816 le64toh(o
->entry
.seqnum
),
819 if (!VALID_REALTIME(le64toh(o
->entry
.realtime
)))
820 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
821 "Invalid entry realtime timestamp: %" PRIu64
": %" PRIu64
,
822 le64toh(o
->entry
.realtime
),
825 if (!VALID_MONOTONIC(le64toh(o
->entry
.monotonic
)))
826 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
827 "Invalid entry monotonic timestamp: %" PRIu64
": %" PRIu64
,
828 le64toh(o
->entry
.monotonic
),
833 case OBJECT_DATA_HASH_TABLE
:
834 case OBJECT_FIELD_HASH_TABLE
:
835 if ((le64toh(o
->object
.size
) - offsetof(HashTableObject
, items
)) % sizeof(HashItem
) != 0 ||
836 (le64toh(o
->object
.size
) - offsetof(HashTableObject
, items
)) / sizeof(HashItem
) <= 0)
837 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
838 "Invalid %s hash table size: %" PRIu64
": %" PRIu64
,
839 o
->object
.type
== OBJECT_DATA_HASH_TABLE
? "data" : "field",
840 le64toh(o
->object
.size
),
845 case OBJECT_ENTRY_ARRAY
:
846 if ((le64toh(o
->object
.size
) - offsetof(EntryArrayObject
, items
)) % sizeof(le64_t
) != 0 ||
847 (le64toh(o
->object
.size
) - offsetof(EntryArrayObject
, items
)) / sizeof(le64_t
) <= 0)
848 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
849 "Invalid object entry array size: %" PRIu64
": %" PRIu64
,
850 le64toh(o
->object
.size
),
853 if (!VALID64(le64toh(o
->entry_array
.next_entry_array_offset
)))
854 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
855 "Invalid object entry array next_entry_array_offset: " OFSfmt
": %" PRIu64
,
856 le64toh(o
->entry_array
.next_entry_array_offset
),
862 if (le64toh(o
->object
.size
) != sizeof(TagObject
))
863 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
864 "Invalid object tag size: %" PRIu64
": %" PRIu64
,
865 le64toh(o
->object
.size
),
868 if (!VALID_EPOCH(le64toh(o
->tag
.epoch
)))
869 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
870 "Invalid object tag epoch: %" PRIu64
": %" PRIu64
,
871 le64toh(o
->tag
.epoch
), offset
);
879 int journal_file_move_to_object(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
**ret
) {
889 /* Objects may only be located at multiple of 64 bit */
890 if (!VALID64(offset
))
891 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
892 "Attempt to move to object at non-64bit boundary: %" PRIu64
,
895 /* Object may not be located in the file header */
896 if (offset
< le64toh(f
->header
->header_size
))
897 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
898 "Attempt to move to object located in file header: %" PRIu64
,
901 r
= journal_file_move_to(f
, type
, false, offset
, sizeof(ObjectHeader
), &t
, &tsize
);
906 s
= le64toh(o
->object
.size
);
909 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
910 "Attempt to move to uninitialized object: %" PRIu64
,
912 if (s
< sizeof(ObjectHeader
))
913 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
914 "Attempt to move to overly short object: %" PRIu64
,
917 if (o
->object
.type
<= OBJECT_UNUSED
)
918 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
919 "Attempt to move to object with invalid type: %" PRIu64
,
922 if (s
< minimum_header_size(o
))
923 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
924 "Attempt to move to truncated object: %" PRIu64
,
927 if (type
> OBJECT_UNUSED
&& o
->object
.type
!= type
)
928 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
929 "Attempt to move to object of unexpected type: %" PRIu64
,
933 r
= journal_file_move_to(f
, type
, false, offset
, s
, &t
, NULL
);
940 r
= journal_file_check_object(f
, offset
, o
);
948 static uint64_t journal_file_entry_seqnum(JournalFile
*f
, uint64_t *seqnum
) {
954 r
= le64toh(f
->header
->tail_entry_seqnum
) + 1;
957 /* If an external seqnum counter was passed, we update
958 * both the local and the external one, and set it to
959 * the maximum of both */
967 f
->header
->tail_entry_seqnum
= htole64(r
);
969 if (f
->header
->head_entry_seqnum
== 0)
970 f
->header
->head_entry_seqnum
= htole64(r
);
975 int journal_file_append_object(JournalFile
*f
, ObjectType type
, uint64_t size
, Object
**ret
, uint64_t *offset
) {
983 assert(type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
);
984 assert(size
>= sizeof(ObjectHeader
));
988 r
= journal_file_set_online(f
);
992 p
= le64toh(f
->header
->tail_object_offset
);
994 p
= le64toh(f
->header
->header_size
);
996 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &tail
);
1000 p
+= ALIGN64(le64toh(tail
->object
.size
));
1003 r
= journal_file_allocate(f
, p
, size
);
1007 r
= journal_file_move_to(f
, type
, false, p
, size
, &t
, NULL
);
1014 o
->object
.type
= type
;
1015 o
->object
.size
= htole64(size
);
1017 f
->header
->tail_object_offset
= htole64(p
);
1018 f
->header
->n_objects
= htole64(le64toh(f
->header
->n_objects
) + 1);
1026 static int journal_file_setup_data_hash_table(JournalFile
*f
) {
1034 /* We estimate that we need 1 hash table entry per 768 bytes
1035 of journal file and we want to make sure we never get
1036 beyond 75% fill level. Calculate the hash table size for
1037 the maximum file size based on these metrics. */
1039 s
= (f
->metrics
.max_size
* 4 / 768 / 3) * sizeof(HashItem
);
1040 if (s
< DEFAULT_DATA_HASH_TABLE_SIZE
)
1041 s
= DEFAULT_DATA_HASH_TABLE_SIZE
;
1043 log_debug("Reserving %"PRIu64
" entries in hash table.", s
/ sizeof(HashItem
));
1045 r
= journal_file_append_object(f
,
1046 OBJECT_DATA_HASH_TABLE
,
1047 offsetof(Object
, hash_table
.items
) + s
,
1052 memzero(o
->hash_table
.items
, s
);
1054 f
->header
->data_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1055 f
->header
->data_hash_table_size
= htole64(s
);
1060 static int journal_file_setup_field_hash_table(JournalFile
*f
) {
1068 /* We use a fixed size hash table for the fields as this
1069 * number should grow very slowly only */
1071 s
= DEFAULT_FIELD_HASH_TABLE_SIZE
;
1072 r
= journal_file_append_object(f
,
1073 OBJECT_FIELD_HASH_TABLE
,
1074 offsetof(Object
, hash_table
.items
) + s
,
1079 memzero(o
->hash_table
.items
, s
);
1081 f
->header
->field_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1082 f
->header
->field_hash_table_size
= htole64(s
);
1087 int journal_file_map_data_hash_table(JournalFile
*f
) {
1095 if (f
->data_hash_table
)
1098 p
= le64toh(f
->header
->data_hash_table_offset
);
1099 s
= le64toh(f
->header
->data_hash_table_size
);
1101 r
= journal_file_move_to(f
,
1102 OBJECT_DATA_HASH_TABLE
,
1109 f
->data_hash_table
= t
;
1113 int journal_file_map_field_hash_table(JournalFile
*f
) {
1121 if (f
->field_hash_table
)
1124 p
= le64toh(f
->header
->field_hash_table_offset
);
1125 s
= le64toh(f
->header
->field_hash_table_size
);
1127 r
= journal_file_move_to(f
,
1128 OBJECT_FIELD_HASH_TABLE
,
1135 f
->field_hash_table
= t
;
1139 static int journal_file_link_field(
1150 assert(f
->field_hash_table
);
1154 if (o
->object
.type
!= OBJECT_FIELD
)
1157 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1161 /* This might alter the window we are looking at */
1162 o
->field
.next_hash_offset
= o
->field
.head_data_offset
= 0;
1165 p
= le64toh(f
->field_hash_table
[h
].tail_hash_offset
);
1167 f
->field_hash_table
[h
].head_hash_offset
= htole64(offset
);
1169 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1173 o
->field
.next_hash_offset
= htole64(offset
);
1176 f
->field_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1178 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
1179 f
->header
->n_fields
= htole64(le64toh(f
->header
->n_fields
) + 1);
1184 static int journal_file_link_data(
1195 assert(f
->data_hash_table
);
1199 if (o
->object
.type
!= OBJECT_DATA
)
1202 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1206 /* This might alter the window we are looking at */
1207 o
->data
.next_hash_offset
= o
->data
.next_field_offset
= 0;
1208 o
->data
.entry_offset
= o
->data
.entry_array_offset
= 0;
1209 o
->data
.n_entries
= 0;
1212 p
= le64toh(f
->data_hash_table
[h
].tail_hash_offset
);
1214 /* Only entry in the hash table is easy */
1215 f
->data_hash_table
[h
].head_hash_offset
= htole64(offset
);
1217 /* Move back to the previous data object, to patch in
1220 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1224 o
->data
.next_hash_offset
= htole64(offset
);
1227 f
->data_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1229 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
1230 f
->header
->n_data
= htole64(le64toh(f
->header
->n_data
) + 1);
1235 int journal_file_find_field_object_with_hash(
1237 const void *field
, uint64_t size
, uint64_t hash
,
1238 Object
**ret
, uint64_t *offset
) {
1240 uint64_t p
, osize
, h
, m
;
1245 assert(field
&& size
> 0);
1247 /* If the field hash table is empty, we can't find anything */
1248 if (le64toh(f
->header
->field_hash_table_size
) <= 0)
1251 /* Map the field hash table, if it isn't mapped yet. */
1252 r
= journal_file_map_field_hash_table(f
);
1256 osize
= offsetof(Object
, field
.payload
) + size
;
1258 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1263 p
= le64toh(f
->field_hash_table
[h
].head_hash_offset
);
1268 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1272 if (le64toh(o
->field
.hash
) == hash
&&
1273 le64toh(o
->object
.size
) == osize
&&
1274 memcmp(o
->field
.payload
, field
, size
) == 0) {
1284 p
= le64toh(o
->field
.next_hash_offset
);
1290 int journal_file_find_field_object(
1292 const void *field
, uint64_t size
,
1293 Object
**ret
, uint64_t *offset
) {
1298 assert(field
&& size
> 0);
1300 hash
= hash64(field
, size
);
1302 return journal_file_find_field_object_with_hash(f
,
1307 int journal_file_find_data_object_with_hash(
1309 const void *data
, uint64_t size
, uint64_t hash
,
1310 Object
**ret
, uint64_t *offset
) {
1312 uint64_t p
, osize
, h
, m
;
1317 assert(data
|| size
== 0);
1319 /* If there's no data hash table, then there's no entry. */
1320 if (le64toh(f
->header
->data_hash_table_size
) <= 0)
1323 /* Map the data hash table, if it isn't mapped yet. */
1324 r
= journal_file_map_data_hash_table(f
);
1328 osize
= offsetof(Object
, data
.payload
) + size
;
1330 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1335 p
= le64toh(f
->data_hash_table
[h
].head_hash_offset
);
1340 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1344 if (le64toh(o
->data
.hash
) != hash
)
1347 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
1348 #if HAVE_XZ || HAVE_LZ4
1352 l
= le64toh(o
->object
.size
);
1353 if (l
<= offsetof(Object
, data
.payload
))
1356 l
-= offsetof(Object
, data
.payload
);
1358 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
1359 o
->data
.payload
, l
, &f
->compress_buffer
, &f
->compress_buffer_size
, &rsize
, 0);
1363 if (rsize
== size
&&
1364 memcmp(f
->compress_buffer
, data
, size
) == 0) {
1375 return -EPROTONOSUPPORT
;
1377 } else if (le64toh(o
->object
.size
) == osize
&&
1378 memcmp(o
->data
.payload
, data
, size
) == 0) {
1390 p
= le64toh(o
->data
.next_hash_offset
);
1396 int journal_file_find_data_object(
1398 const void *data
, uint64_t size
,
1399 Object
**ret
, uint64_t *offset
) {
1404 assert(data
|| size
== 0);
1406 hash
= hash64(data
, size
);
1408 return journal_file_find_data_object_with_hash(f
,
1413 static int journal_file_append_field(
1415 const void *field
, uint64_t size
,
1416 Object
**ret
, uint64_t *offset
) {
1424 assert(field
&& size
> 0);
1426 hash
= hash64(field
, size
);
1428 r
= journal_file_find_field_object_with_hash(f
, field
, size
, hash
, &o
, &p
);
1442 osize
= offsetof(Object
, field
.payload
) + size
;
1443 r
= journal_file_append_object(f
, OBJECT_FIELD
, osize
, &o
, &p
);
1447 o
->field
.hash
= htole64(hash
);
1448 memcpy(o
->field
.payload
, field
, size
);
1450 r
= journal_file_link_field(f
, o
, p
, hash
);
1454 /* The linking might have altered the window, so let's
1455 * refresh our pointer */
1456 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1461 r
= journal_file_hmac_put_object(f
, OBJECT_FIELD
, o
, p
);
1475 static int journal_file_append_data(
1477 const void *data
, uint64_t size
,
1478 Object
**ret
, uint64_t *offset
) {
1483 int r
, compression
= 0;
1487 assert(data
|| size
== 0);
1489 hash
= hash64(data
, size
);
1491 r
= journal_file_find_data_object_with_hash(f
, data
, size
, hash
, &o
, &p
);
1505 osize
= offsetof(Object
, data
.payload
) + size
;
1506 r
= journal_file_append_object(f
, OBJECT_DATA
, osize
, &o
, &p
);
1510 o
->data
.hash
= htole64(hash
);
1512 #if HAVE_XZ || HAVE_LZ4
1513 if (JOURNAL_FILE_COMPRESS(f
) && size
>= f
->compress_threshold_bytes
) {
1516 compression
= compress_blob(data
, size
, o
->data
.payload
, size
- 1, &rsize
);
1518 if (compression
>= 0) {
1519 o
->object
.size
= htole64(offsetof(Object
, data
.payload
) + rsize
);
1520 o
->object
.flags
|= compression
;
1522 log_debug("Compressed data object %"PRIu64
" -> %zu using %s",
1523 size
, rsize
, object_compressed_to_string(compression
));
1525 /* Compression didn't work, we don't really care why, let's continue without compression */
1530 if (compression
== 0)
1531 memcpy_safe(o
->data
.payload
, data
, size
);
1533 r
= journal_file_link_data(f
, o
, p
, hash
);
1538 r
= journal_file_hmac_put_object(f
, OBJECT_DATA
, o
, p
);
1543 /* The linking might have altered the window, so let's
1544 * refresh our pointer */
1545 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1552 eq
= memchr(data
, '=', size
);
1553 if (eq
&& eq
> data
) {
1557 /* Create field object ... */
1558 r
= journal_file_append_field(f
, data
, (uint8_t*) eq
- (uint8_t*) data
, &fo
, &fp
);
1562 /* ... and link it in. */
1563 o
->data
.next_field_offset
= fo
->field
.head_data_offset
;
1564 fo
->field
.head_data_offset
= le64toh(p
);
1576 uint64_t journal_file_entry_n_items(Object
*o
) {
1579 if (o
->object
.type
!= OBJECT_ENTRY
)
1582 return (le64toh(o
->object
.size
) - offsetof(Object
, entry
.items
)) / sizeof(EntryItem
);
1585 uint64_t journal_file_entry_array_n_items(Object
*o
) {
1588 if (o
->object
.type
!= OBJECT_ENTRY_ARRAY
)
1591 return (le64toh(o
->object
.size
) - offsetof(Object
, entry_array
.items
)) / sizeof(uint64_t);
1594 uint64_t journal_file_hash_table_n_items(Object
*o
) {
1597 if (!IN_SET(o
->object
.type
, OBJECT_DATA_HASH_TABLE
, OBJECT_FIELD_HASH_TABLE
))
1600 return (le64toh(o
->object
.size
) - offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
);
1603 static int link_entry_into_array(JournalFile
*f
,
1608 uint64_t n
= 0, ap
= 0, q
, i
, a
, hidx
;
1617 a
= le64toh(*first
);
1618 i
= hidx
= le64toh(*idx
);
1621 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
1625 n
= journal_file_entry_array_n_items(o
);
1627 o
->entry_array
.items
[i
] = htole64(p
);
1628 *idx
= htole64(hidx
+ 1);
1634 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
1645 r
= journal_file_append_object(f
, OBJECT_ENTRY_ARRAY
,
1646 offsetof(Object
, entry_array
.items
) + n
* sizeof(uint64_t),
1652 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY_ARRAY
, o
, q
);
1657 o
->entry_array
.items
[i
] = htole64(p
);
1660 *first
= htole64(q
);
1662 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, ap
, &o
);
1666 o
->entry_array
.next_entry_array_offset
= htole64(q
);
1669 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
1670 f
->header
->n_entry_arrays
= htole64(le64toh(f
->header
->n_entry_arrays
) + 1);
1672 *idx
= htole64(hidx
+ 1);
1677 static int link_entry_into_array_plus_one(JournalFile
*f
,
1692 *extra
= htole64(p
);
1696 i
= htole64(le64toh(*idx
) - 1);
1697 r
= link_entry_into_array(f
, first
, &i
, p
);
1702 *idx
= htole64(le64toh(*idx
) + 1);
1706 static int journal_file_link_entry_item(JournalFile
*f
, Object
*o
, uint64_t offset
, uint64_t i
) {
1713 p
= le64toh(o
->entry
.items
[i
].object_offset
);
1717 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1721 return link_entry_into_array_plus_one(f
,
1722 &o
->data
.entry_offset
,
1723 &o
->data
.entry_array_offset
,
1728 static int journal_file_link_entry(JournalFile
*f
, Object
*o
, uint64_t offset
) {
1737 if (o
->object
.type
!= OBJECT_ENTRY
)
1740 __sync_synchronize();
1742 /* Link up the entry itself */
1743 r
= link_entry_into_array(f
,
1744 &f
->header
->entry_array_offset
,
1745 &f
->header
->n_entries
,
1750 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
1752 if (f
->header
->head_entry_realtime
== 0)
1753 f
->header
->head_entry_realtime
= o
->entry
.realtime
;
1755 f
->header
->tail_entry_realtime
= o
->entry
.realtime
;
1756 f
->header
->tail_entry_monotonic
= o
->entry
.monotonic
;
1758 /* Link up the items */
1759 n
= journal_file_entry_n_items(o
);
1760 for (i
= 0; i
< n
; i
++) {
1761 r
= journal_file_link_entry_item(f
, o
, offset
, i
);
1769 static int journal_file_append_entry_internal(
1771 const dual_timestamp
*ts
,
1772 const sd_id128_t
*boot_id
,
1774 const EntryItem items
[], unsigned n_items
,
1776 Object
**ret
, uint64_t *offset
) {
1784 assert(items
|| n_items
== 0);
1787 osize
= offsetof(Object
, entry
.items
) + (n_items
* sizeof(EntryItem
));
1789 r
= journal_file_append_object(f
, OBJECT_ENTRY
, osize
, &o
, &np
);
1793 o
->entry
.seqnum
= htole64(journal_file_entry_seqnum(f
, seqnum
));
1794 memcpy_safe(o
->entry
.items
, items
, n_items
* sizeof(EntryItem
));
1795 o
->entry
.realtime
= htole64(ts
->realtime
);
1796 o
->entry
.monotonic
= htole64(ts
->monotonic
);
1797 o
->entry
.xor_hash
= htole64(xor_hash
);
1799 f
->header
->boot_id
= *boot_id
;
1800 o
->entry
.boot_id
= f
->header
->boot_id
;
1803 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY
, o
, np
);
1808 r
= journal_file_link_entry(f
, o
, np
);
1821 void journal_file_post_change(JournalFile
*f
) {
1827 /* inotify() does not receive IN_MODIFY events from file
1828 * accesses done via mmap(). After each access we hence
1829 * trigger IN_MODIFY by truncating the journal file to its
1830 * current size which triggers IN_MODIFY. */
1832 __sync_synchronize();
1834 if (ftruncate(f
->fd
, f
->last_stat
.st_size
) < 0)
1835 log_debug_errno(errno
, "Failed to truncate file to its own size: %m");
1838 static int post_change_thunk(sd_event_source
*timer
, uint64_t usec
, void *userdata
) {
1841 journal_file_post_change(userdata
);
1846 static void schedule_post_change(JournalFile
*f
) {
1851 assert(f
->post_change_timer
);
1853 r
= sd_event_source_get_enabled(f
->post_change_timer
, NULL
);
1855 log_debug_errno(r
, "Failed to get ftruncate timer state: %m");
1861 r
= sd_event_now(sd_event_source_get_event(f
->post_change_timer
), CLOCK_MONOTONIC
, &now
);
1863 log_debug_errno(r
, "Failed to get clock's now for scheduling ftruncate: %m");
1867 r
= sd_event_source_set_time(f
->post_change_timer
, now
+ f
->post_change_timer_period
);
1869 log_debug_errno(r
, "Failed to set time for scheduling ftruncate: %m");
1873 r
= sd_event_source_set_enabled(f
->post_change_timer
, SD_EVENT_ONESHOT
);
1875 log_debug_errno(r
, "Failed to enable scheduled ftruncate: %m");
1882 /* On failure, let's simply post the change immediately. */
1883 journal_file_post_change(f
);
1886 /* Enable coalesced change posting in a timer on the provided sd_event instance */
1887 int journal_file_enable_post_change_timer(JournalFile
*f
, sd_event
*e
, usec_t t
) {
1888 _cleanup_(sd_event_source_unrefp
) sd_event_source
*timer
= NULL
;
1892 assert_return(!f
->post_change_timer
, -EINVAL
);
1896 r
= sd_event_add_time(e
, &timer
, CLOCK_MONOTONIC
, 0, 0, post_change_thunk
, f
);
1900 r
= sd_event_source_set_enabled(timer
, SD_EVENT_OFF
);
1904 f
->post_change_timer
= TAKE_PTR(timer
);
1905 f
->post_change_timer_period
= t
;
1910 static int entry_item_cmp(const EntryItem
*a
, const EntryItem
*b
) {
1911 return CMP(le64toh(a
->object_offset
), le64toh(b
->object_offset
));
1914 int journal_file_append_entry(
1916 const dual_timestamp
*ts
,
1917 const sd_id128_t
*boot_id
,
1918 const struct iovec iovec
[], unsigned n_iovec
,
1920 Object
**ret
, uint64_t *offset
) {
1925 uint64_t xor_hash
= 0;
1926 struct dual_timestamp _ts
;
1930 assert(iovec
|| n_iovec
== 0);
1933 if (!VALID_REALTIME(ts
->realtime
))
1934 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1935 "Invalid realtime timestamp %" PRIu64
", refusing entry.",
1937 if (!VALID_MONOTONIC(ts
->monotonic
))
1938 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1939 "Invalid monotomic timestamp %" PRIu64
", refusing entry.",
1942 dual_timestamp_get(&_ts
);
1947 r
= journal_file_maybe_append_tag(f
, ts
->realtime
);
1952 /* alloca() can't take 0, hence let's allocate at least one */
1953 items
= newa(EntryItem
, MAX(1u, n_iovec
));
1955 for (i
= 0; i
< n_iovec
; i
++) {
1959 r
= journal_file_append_data(f
, iovec
[i
].iov_base
, iovec
[i
].iov_len
, &o
, &p
);
1963 xor_hash
^= le64toh(o
->data
.hash
);
1964 items
[i
].object_offset
= htole64(p
);
1965 items
[i
].hash
= o
->data
.hash
;
1968 /* Order by the position on disk, in order to improve seek
1969 * times for rotating media. */
1970 typesafe_qsort(items
, n_iovec
, entry_item_cmp
);
1972 r
= journal_file_append_entry_internal(f
, ts
, boot_id
, xor_hash
, items
, n_iovec
, seqnum
, ret
, offset
);
1974 /* If the memory mapping triggered a SIGBUS then we return an
1975 * IO error and ignore the error code passed down to us, since
1976 * it is very likely just an effect of a nullified replacement
1979 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
1982 if (f
->post_change_timer
)
1983 schedule_post_change(f
);
1985 journal_file_post_change(f
);
1990 typedef struct ChainCacheItem
{
1991 uint64_t first
; /* the array at the beginning of the chain */
1992 uint64_t array
; /* the cached array */
1993 uint64_t begin
; /* the first item in the cached array */
1994 uint64_t total
; /* the total number of items in all arrays before this one in the chain */
1995 uint64_t last_index
; /* the last index we looked at, to optimize locality when bisecting */
1998 static void chain_cache_put(
2005 uint64_t last_index
) {
2008 /* If the chain item to cache for this chain is the
2009 * first one it's not worth caching anything */
2013 if (ordered_hashmap_size(h
) >= CHAIN_CACHE_MAX
) {
2014 ci
= ordered_hashmap_steal_first(h
);
2017 ci
= new(ChainCacheItem
, 1);
2024 if (ordered_hashmap_put(h
, &ci
->first
, ci
) < 0) {
2029 assert(ci
->first
== first
);
2034 ci
->last_index
= last_index
;
2037 static int generic_array_get(
2041 Object
**ret
, uint64_t *offset
) {
2044 uint64_t p
= 0, a
, t
= 0;
2052 /* Try the chain cache first */
2053 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2054 if (ci
&& i
> ci
->total
) {
2063 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2067 k
= journal_file_entry_array_n_items(o
);
2069 p
= le64toh(o
->entry_array
.items
[i
]);
2075 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
2081 /* Let's cache this item for the next invocation */
2082 chain_cache_put(f
->chain_cache
, ci
, first
, a
, le64toh(o
->entry_array
.items
[0]), t
, i
);
2084 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2097 static int generic_array_get_plus_one(
2102 Object
**ret
, uint64_t *offset
) {
2111 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2124 return generic_array_get(f
, first
, i
-1, ret
, offset
);
2133 static int generic_array_bisect(
2138 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2139 direction_t direction
,
2144 uint64_t a
, p
, t
= 0, i
= 0, last_p
= 0, last_index
= (uint64_t) -1;
2145 bool subtract_one
= false;
2146 Object
*o
, *array
= NULL
;
2151 assert(test_object
);
2153 /* Start with the first array in the chain */
2156 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2157 if (ci
&& n
> ci
->total
&& ci
->begin
!= 0) {
2158 /* Ah, we have iterated this bisection array chain
2159 * previously! Let's see if we can skip ahead in the
2160 * chain, as far as the last time. But we can't jump
2161 * backwards in the chain, so let's check that
2164 r
= test_object(f
, ci
->begin
, needle
);
2168 if (r
== TEST_LEFT
) {
2169 /* OK, what we are looking for is right of the
2170 * begin of this EntryArray, so let's jump
2171 * straight to previously cached array in the
2177 last_index
= ci
->last_index
;
2182 uint64_t left
, right
, k
, lp
;
2184 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2188 k
= journal_file_entry_array_n_items(array
);
2194 lp
= p
= le64toh(array
->entry_array
.items
[i
]);
2198 r
= test_object(f
, p
, needle
);
2199 if (r
== -EBADMSG
) {
2200 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (1)");
2207 if (r
== TEST_FOUND
)
2208 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2210 if (r
== TEST_RIGHT
) {
2214 if (last_index
!= (uint64_t) -1) {
2215 assert(last_index
<= right
);
2217 /* If we cached the last index we
2218 * looked at, let's try to not to jump
2219 * too wildly around and see if we can
2220 * limit the range to look at early to
2221 * the immediate neighbors of the last
2222 * index we looked at. */
2224 if (last_index
> 0) {
2225 uint64_t x
= last_index
- 1;
2227 p
= le64toh(array
->entry_array
.items
[x
]);
2231 r
= test_object(f
, p
, needle
);
2235 if (r
== TEST_FOUND
)
2236 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2238 if (r
== TEST_RIGHT
)
2244 if (last_index
< right
) {
2245 uint64_t y
= last_index
+ 1;
2247 p
= le64toh(array
->entry_array
.items
[y
]);
2251 r
= test_object(f
, p
, needle
);
2255 if (r
== TEST_FOUND
)
2256 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2258 if (r
== TEST_RIGHT
)
2266 if (left
== right
) {
2267 if (direction
== DIRECTION_UP
)
2268 subtract_one
= true;
2274 assert(left
< right
);
2275 i
= (left
+ right
) / 2;
2277 p
= le64toh(array
->entry_array
.items
[i
]);
2281 r
= test_object(f
, p
, needle
);
2282 if (r
== -EBADMSG
) {
2283 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (2)");
2290 if (r
== TEST_FOUND
)
2291 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2293 if (r
== TEST_RIGHT
)
2301 if (direction
== DIRECTION_UP
) {
2303 subtract_one
= true;
2314 last_index
= (uint64_t) -1;
2315 a
= le64toh(array
->entry_array
.next_entry_array_offset
);
2321 if (subtract_one
&& t
== 0 && i
== 0)
2324 /* Let's cache this item for the next invocation */
2325 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
);
2327 if (subtract_one
&& i
== 0)
2329 else if (subtract_one
)
2330 p
= le64toh(array
->entry_array
.items
[i
-1]);
2332 p
= le64toh(array
->entry_array
.items
[i
]);
2334 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2345 *idx
= t
+ i
+ (subtract_one
? -1 : 0);
2350 static int generic_array_bisect_plus_one(
2356 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2357 direction_t direction
,
2363 bool step_back
= false;
2367 assert(test_object
);
2372 /* This bisects the array in object 'first', but first checks
2374 r
= test_object(f
, extra
, needle
);
2378 if (r
== TEST_FOUND
)
2379 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2381 /* if we are looking with DIRECTION_UP then we need to first
2382 see if in the actual array there is a matching entry, and
2383 return the last one of that. But if there isn't any we need
2384 to return this one. Hence remember this, and return it
2387 step_back
= direction
== DIRECTION_UP
;
2389 if (r
== TEST_RIGHT
) {
2390 if (direction
== DIRECTION_DOWN
)
2396 r
= generic_array_bisect(f
, first
, n
-1, needle
, test_object
, direction
, ret
, offset
, idx
);
2398 if (r
== 0 && step_back
)
2407 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2423 _pure_
static int test_object_offset(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2429 else if (p
< needle
)
2435 static int test_object_seqnum(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2442 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2446 if (le64toh(o
->entry
.seqnum
) == needle
)
2448 else if (le64toh(o
->entry
.seqnum
) < needle
)
2454 int journal_file_move_to_entry_by_seqnum(
2457 direction_t direction
,
2463 return generic_array_bisect(f
,
2464 le64toh(f
->header
->entry_array_offset
),
2465 le64toh(f
->header
->n_entries
),
2472 static int test_object_realtime(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2479 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2483 if (le64toh(o
->entry
.realtime
) == needle
)
2485 else if (le64toh(o
->entry
.realtime
) < needle
)
2491 int journal_file_move_to_entry_by_realtime(
2494 direction_t direction
,
2500 return generic_array_bisect(f
,
2501 le64toh(f
->header
->entry_array_offset
),
2502 le64toh(f
->header
->n_entries
),
2504 test_object_realtime
,
2509 static int test_object_monotonic(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2516 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2520 if (le64toh(o
->entry
.monotonic
) == needle
)
2522 else if (le64toh(o
->entry
.monotonic
) < needle
)
2528 static int find_data_object_by_boot_id(
2534 char t
[STRLEN("_BOOT_ID=") + 32 + 1] = "_BOOT_ID=";
2536 sd_id128_to_string(boot_id
, t
+ 9);
2537 return journal_file_find_data_object(f
, t
, sizeof(t
) - 1, o
, b
);
2540 int journal_file_move_to_entry_by_monotonic(
2544 direction_t direction
,
2553 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, NULL
);
2559 return generic_array_bisect_plus_one(f
,
2560 le64toh(o
->data
.entry_offset
),
2561 le64toh(o
->data
.entry_array_offset
),
2562 le64toh(o
->data
.n_entries
),
2564 test_object_monotonic
,
2569 void journal_file_reset_location(JournalFile
*f
) {
2570 f
->location_type
= LOCATION_HEAD
;
2571 f
->current_offset
= 0;
2572 f
->current_seqnum
= 0;
2573 f
->current_realtime
= 0;
2574 f
->current_monotonic
= 0;
2575 zero(f
->current_boot_id
);
2576 f
->current_xor_hash
= 0;
2579 void journal_file_save_location(JournalFile
*f
, Object
*o
, uint64_t offset
) {
2580 f
->location_type
= LOCATION_SEEK
;
2581 f
->current_offset
= offset
;
2582 f
->current_seqnum
= le64toh(o
->entry
.seqnum
);
2583 f
->current_realtime
= le64toh(o
->entry
.realtime
);
2584 f
->current_monotonic
= le64toh(o
->entry
.monotonic
);
2585 f
->current_boot_id
= o
->entry
.boot_id
;
2586 f
->current_xor_hash
= le64toh(o
->entry
.xor_hash
);
2589 int journal_file_compare_locations(JournalFile
*af
, JournalFile
*bf
) {
2596 assert(af
->location_type
== LOCATION_SEEK
);
2597 assert(bf
->location_type
== LOCATION_SEEK
);
2599 /* If contents and timestamps match, these entries are
2600 * identical, even if the seqnum does not match */
2601 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
) &&
2602 af
->current_monotonic
== bf
->current_monotonic
&&
2603 af
->current_realtime
== bf
->current_realtime
&&
2604 af
->current_xor_hash
== bf
->current_xor_hash
)
2607 if (sd_id128_equal(af
->header
->seqnum_id
, bf
->header
->seqnum_id
)) {
2609 /* If this is from the same seqnum source, compare
2611 r
= CMP(af
->current_seqnum
, bf
->current_seqnum
);
2615 /* Wow! This is weird, different data but the same
2616 * seqnums? Something is borked, but let's make the
2617 * best of it and compare by time. */
2620 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
)) {
2622 /* If the boot id matches, compare monotonic time */
2623 r
= CMP(af
->current_monotonic
, bf
->current_monotonic
);
2628 /* Otherwise, compare UTC time */
2629 r
= CMP(af
->current_realtime
, bf
->current_realtime
);
2633 /* Finally, compare by contents */
2634 return CMP(af
->current_xor_hash
, bf
->current_xor_hash
);
2637 static int bump_array_index(uint64_t *i
, direction_t direction
, uint64_t n
) {
2639 /* Increase or decrease the specified index, in the right direction. */
2641 if (direction
== DIRECTION_DOWN
) {
2656 static bool check_properly_ordered(uint64_t new_offset
, uint64_t old_offset
, direction_t direction
) {
2658 /* Consider it an error if any of the two offsets is uninitialized */
2659 if (old_offset
== 0 || new_offset
== 0)
2662 /* If we go down, the new offset must be larger than the old one. */
2663 return direction
== DIRECTION_DOWN
?
2664 new_offset
> old_offset
:
2665 new_offset
< old_offset
;
2668 int journal_file_next_entry(
2671 direction_t direction
,
2672 Object
**ret
, uint64_t *offset
) {
2680 n
= le64toh(f
->header
->n_entries
);
2685 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2687 r
= generic_array_bisect(f
,
2688 le64toh(f
->header
->entry_array_offset
),
2689 le64toh(f
->header
->n_entries
),
2698 r
= bump_array_index(&i
, direction
, n
);
2703 /* And jump to it */
2705 r
= generic_array_get(f
,
2706 le64toh(f
->header
->entry_array_offset
),
2714 /* OK, so this entry is borked. Most likely some entry didn't get synced to disk properly, let's see if
2715 * the next one might work for us instead. */
2716 log_debug_errno(r
, "Entry item %" PRIu64
" is bad, skipping over it.", i
);
2718 r
= bump_array_index(&i
, direction
, n
);
2723 /* Ensure our array is properly ordered. */
2724 if (p
> 0 && !check_properly_ordered(ofs
, p
, direction
))
2725 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
2726 "%s: entry array not properly ordered at entry %" PRIu64
,
2735 int journal_file_next_entry_for_data(
2737 Object
*o
, uint64_t p
,
2738 uint64_t data_offset
,
2739 direction_t direction
,
2740 Object
**ret
, uint64_t *offset
) {
2747 assert(p
> 0 || !o
);
2749 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2753 n
= le64toh(d
->data
.n_entries
);
2758 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2760 if (o
->object
.type
!= OBJECT_ENTRY
)
2763 r
= generic_array_bisect_plus_one(f
,
2764 le64toh(d
->data
.entry_offset
),
2765 le64toh(d
->data
.entry_array_offset
),
2766 le64toh(d
->data
.n_entries
),
2776 r
= bump_array_index(&i
, direction
, n
);
2782 r
= generic_array_get_plus_one(f
,
2783 le64toh(d
->data
.entry_offset
),
2784 le64toh(d
->data
.entry_array_offset
),
2792 log_debug_errno(r
, "Data entry item %" PRIu64
" is bad, skipping over it.", i
);
2794 r
= bump_array_index(&i
, direction
, n
);
2799 /* Ensure our array is properly ordered. */
2800 if (p
> 0 && check_properly_ordered(ofs
, p
, direction
))
2801 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
2802 "%s data entry array not properly ordered at entry %" PRIu64
,
2811 int journal_file_move_to_entry_by_offset_for_data(
2813 uint64_t data_offset
,
2815 direction_t direction
,
2816 Object
**ret
, uint64_t *offset
) {
2823 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2827 return generic_array_bisect_plus_one(f
,
2828 le64toh(d
->data
.entry_offset
),
2829 le64toh(d
->data
.entry_array_offset
),
2830 le64toh(d
->data
.n_entries
),
2837 int journal_file_move_to_entry_by_monotonic_for_data(
2839 uint64_t data_offset
,
2842 direction_t direction
,
2843 Object
**ret
, uint64_t *offset
) {
2851 /* First, seek by time */
2852 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &b
);
2858 r
= generic_array_bisect_plus_one(f
,
2859 le64toh(o
->data
.entry_offset
),
2860 le64toh(o
->data
.entry_array_offset
),
2861 le64toh(o
->data
.n_entries
),
2863 test_object_monotonic
,
2869 /* And now, continue seeking until we find an entry that
2870 * exists in both bisection arrays */
2876 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2880 r
= generic_array_bisect_plus_one(f
,
2881 le64toh(d
->data
.entry_offset
),
2882 le64toh(d
->data
.entry_array_offset
),
2883 le64toh(d
->data
.n_entries
),
2891 r
= journal_file_move_to_object(f
, OBJECT_DATA
, b
, &o
);
2895 r
= generic_array_bisect_plus_one(f
,
2896 le64toh(o
->data
.entry_offset
),
2897 le64toh(o
->data
.entry_array_offset
),
2898 le64toh(o
->data
.n_entries
),
2920 int journal_file_move_to_entry_by_seqnum_for_data(
2922 uint64_t data_offset
,
2924 direction_t direction
,
2925 Object
**ret
, uint64_t *offset
) {
2932 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2936 return generic_array_bisect_plus_one(f
,
2937 le64toh(d
->data
.entry_offset
),
2938 le64toh(d
->data
.entry_array_offset
),
2939 le64toh(d
->data
.n_entries
),
2946 int journal_file_move_to_entry_by_realtime_for_data(
2948 uint64_t data_offset
,
2950 direction_t direction
,
2951 Object
**ret
, uint64_t *offset
) {
2958 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2962 return generic_array_bisect_plus_one(f
,
2963 le64toh(d
->data
.entry_offset
),
2964 le64toh(d
->data
.entry_array_offset
),
2965 le64toh(d
->data
.n_entries
),
2967 test_object_realtime
,
2972 void journal_file_dump(JournalFile
*f
) {
2980 journal_file_print_header(f
);
2982 p
= le64toh(f
->header
->header_size
);
2984 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &o
);
2988 switch (o
->object
.type
) {
2991 printf("Type: OBJECT_UNUSED\n");
2995 printf("Type: OBJECT_DATA\n");
2999 printf("Type: OBJECT_FIELD\n");
3003 printf("Type: OBJECT_ENTRY seqnum=%"PRIu64
" monotonic=%"PRIu64
" realtime=%"PRIu64
"\n",
3004 le64toh(o
->entry
.seqnum
),
3005 le64toh(o
->entry
.monotonic
),
3006 le64toh(o
->entry
.realtime
));
3009 case OBJECT_FIELD_HASH_TABLE
:
3010 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
3013 case OBJECT_DATA_HASH_TABLE
:
3014 printf("Type: OBJECT_DATA_HASH_TABLE\n");
3017 case OBJECT_ENTRY_ARRAY
:
3018 printf("Type: OBJECT_ENTRY_ARRAY\n");
3022 printf("Type: OBJECT_TAG seqnum=%"PRIu64
" epoch=%"PRIu64
"\n",
3023 le64toh(o
->tag
.seqnum
),
3024 le64toh(o
->tag
.epoch
));
3028 printf("Type: unknown (%i)\n", o
->object
.type
);
3032 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
)
3033 printf("Flags: %s\n",
3034 object_compressed_to_string(o
->object
.flags
& OBJECT_COMPRESSION_MASK
));
3036 if (p
== le64toh(f
->header
->tail_object_offset
))
3039 p
= p
+ ALIGN64(le64toh(o
->object
.size
));
3044 log_error("File corrupt");
3047 static const char* format_timestamp_safe(char *buf
, size_t l
, usec_t t
) {
3050 x
= format_timestamp(buf
, l
, t
);
3056 void journal_file_print_header(JournalFile
*f
) {
3057 char a
[33], b
[33], c
[33], d
[33];
3058 char x
[FORMAT_TIMESTAMP_MAX
], y
[FORMAT_TIMESTAMP_MAX
], z
[FORMAT_TIMESTAMP_MAX
];
3060 char bytes
[FORMAT_BYTES_MAX
];
3065 printf("File Path: %s\n"
3069 "Sequential Number ID: %s\n"
3071 "Compatible Flags:%s%s\n"
3072 "Incompatible Flags:%s%s%s\n"
3073 "Header size: %"PRIu64
"\n"
3074 "Arena size: %"PRIu64
"\n"
3075 "Data Hash Table Size: %"PRIu64
"\n"
3076 "Field Hash Table Size: %"PRIu64
"\n"
3077 "Rotate Suggested: %s\n"
3078 "Head Sequential Number: %"PRIu64
" (%"PRIx64
")\n"
3079 "Tail Sequential Number: %"PRIu64
" (%"PRIx64
")\n"
3080 "Head Realtime Timestamp: %s (%"PRIx64
")\n"
3081 "Tail Realtime Timestamp: %s (%"PRIx64
")\n"
3082 "Tail Monotonic Timestamp: %s (%"PRIx64
")\n"
3083 "Objects: %"PRIu64
"\n"
3084 "Entry Objects: %"PRIu64
"\n",
3086 sd_id128_to_string(f
->header
->file_id
, a
),
3087 sd_id128_to_string(f
->header
->machine_id
, b
),
3088 sd_id128_to_string(f
->header
->boot_id
, c
),
3089 sd_id128_to_string(f
->header
->seqnum_id
, d
),
3090 f
->header
->state
== STATE_OFFLINE
? "OFFLINE" :
3091 f
->header
->state
== STATE_ONLINE
? "ONLINE" :
3092 f
->header
->state
== STATE_ARCHIVED
? "ARCHIVED" : "UNKNOWN",
3093 JOURNAL_HEADER_SEALED(f
->header
) ? " SEALED" : "",
3094 (le32toh(f
->header
->compatible_flags
) & ~HEADER_COMPATIBLE_ANY
) ? " ???" : "",
3095 JOURNAL_HEADER_COMPRESSED_XZ(f
->header
) ? " COMPRESSED-XZ" : "",
3096 JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
) ? " COMPRESSED-LZ4" : "",
3097 (le32toh(f
->header
->incompatible_flags
) & ~HEADER_INCOMPATIBLE_ANY
) ? " ???" : "",
3098 le64toh(f
->header
->header_size
),
3099 le64toh(f
->header
->arena_size
),
3100 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3101 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
),
3102 yes_no(journal_file_rotate_suggested(f
, 0)),
3103 le64toh(f
->header
->head_entry_seqnum
), le64toh(f
->header
->head_entry_seqnum
),
3104 le64toh(f
->header
->tail_entry_seqnum
), le64toh(f
->header
->tail_entry_seqnum
),
3105 format_timestamp_safe(x
, sizeof(x
), le64toh(f
->header
->head_entry_realtime
)), le64toh(f
->header
->head_entry_realtime
),
3106 format_timestamp_safe(y
, sizeof(y
), le64toh(f
->header
->tail_entry_realtime
)), le64toh(f
->header
->tail_entry_realtime
),
3107 format_timespan(z
, sizeof(z
), le64toh(f
->header
->tail_entry_monotonic
), USEC_PER_MSEC
), le64toh(f
->header
->tail_entry_monotonic
),
3108 le64toh(f
->header
->n_objects
),
3109 le64toh(f
->header
->n_entries
));
3111 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3112 printf("Data Objects: %"PRIu64
"\n"
3113 "Data Hash Table Fill: %.1f%%\n",
3114 le64toh(f
->header
->n_data
),
3115 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))));
3117 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3118 printf("Field Objects: %"PRIu64
"\n"
3119 "Field Hash Table Fill: %.1f%%\n",
3120 le64toh(f
->header
->n_fields
),
3121 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))));
3123 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
))
3124 printf("Tag Objects: %"PRIu64
"\n",
3125 le64toh(f
->header
->n_tags
));
3126 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
3127 printf("Entry Array Objects: %"PRIu64
"\n",
3128 le64toh(f
->header
->n_entry_arrays
));
3130 if (fstat(f
->fd
, &st
) >= 0)
3131 printf("Disk usage: %s\n", format_bytes(bytes
, sizeof(bytes
), (uint64_t) st
.st_blocks
* 512ULL));
3134 static int journal_file_warn_btrfs(JournalFile
*f
) {
3140 /* Before we write anything, check if the COW logic is turned
3141 * off on btrfs. Given our write pattern that is quite
3142 * unfriendly to COW file systems this should greatly improve
3143 * performance on COW file systems, such as btrfs, at the
3144 * expense of data integrity features (which shouldn't be too
3145 * bad, given that we do our own checksumming). */
3147 r
= btrfs_is_filesystem(f
->fd
);
3149 return log_warning_errno(r
, "Failed to determine if journal is on btrfs: %m");
3153 r
= read_attr_fd(f
->fd
, &attrs
);
3155 return log_warning_errno(r
, "Failed to read file attributes: %m");
3157 if (attrs
& FS_NOCOW_FL
) {
3158 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3162 log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3163 "This is likely to slow down journal access substantially, please consider turning "
3164 "off the copy-on-write file attribute on the journal directory, using chattr +C.", f
->path
);
3169 int journal_file_open(
3175 uint64_t compress_threshold_bytes
,
3177 JournalMetrics
*metrics
,
3178 MMapCache
*mmap_cache
,
3179 Set
*deferred_closes
,
3180 JournalFile
*template,
3181 JournalFile
**ret
) {
3183 bool newly_created
= false;
3187 char bytes
[FORMAT_BYTES_MAX
];
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 log_debug("Journal effective settings seal=%s compress=%s compress_threshold_bytes=%s",
3224 yes_no(f
->seal
), yes_no(JOURNAL_FILE_COMPRESS(f
)),
3225 format_bytes(bytes
, sizeof(bytes
), f
->compress_threshold_bytes
));
3228 f
->mmap
= mmap_cache_ref(mmap_cache
);
3230 f
->mmap
= mmap_cache_new();
3238 f
->path
= strdup(fname
);
3246 /* If we don't know the path, fill in something explanatory and vaguely useful */
3247 if (asprintf(&f
->path
, "/proc/self/%i", fd
) < 0) {
3253 f
->chain_cache
= ordered_hashmap_new(&uint64_hash_ops
);
3254 if (!f
->chain_cache
) {
3260 /* We pass O_NONBLOCK here, so that in case somebody pointed us to some character device node or FIFO
3261 * or so, we likely fail quickly than block for long. For regular files O_NONBLOCK has no effect, hence
3262 * it doesn't hurt in that case. */
3264 f
->fd
= open(f
->path
, f
->flags
|O_CLOEXEC
|O_NONBLOCK
, f
->mode
);
3270 /* fds we opened here by us should also be closed by us. */
3273 r
= fd_nonblock(f
->fd
, false);
3278 f
->cache_fd
= mmap_cache_add_fd(f
->mmap
, f
->fd
);
3284 r
= journal_file_fstat(f
);
3288 if (f
->last_stat
.st_size
== 0 && f
->writable
) {
3290 (void) journal_file_warn_btrfs(f
);
3292 /* Let's attach the creation time to the journal file, so that the vacuuming code knows the age of this
3293 * file even if the file might end up corrupted one day... Ideally we'd just use the creation time many
3294 * file systems maintain for each file, but the API to query this is very new, hence let's emulate this
3295 * via extended attributes. If extended attributes are not supported we'll just skip this, and rely
3296 * solely on mtime/atime/ctime of the file. */
3297 (void) fd_setcrtime(f
->fd
, 0);
3300 /* Try to load the FSPRG state, and if we can't, then
3301 * just don't do sealing */
3303 r
= journal_file_fss_load(f
);
3309 r
= journal_file_init_header(f
, template);
3313 r
= journal_file_fstat(f
);
3317 newly_created
= true;
3320 if (f
->last_stat
.st_size
< (off_t
) HEADER_SIZE_MIN
) {
3325 r
= mmap_cache_get(f
->mmap
, f
->cache_fd
, f
->prot
, CONTEXT_HEADER
, true, 0, PAGE_ALIGN(sizeof(Header
)), &f
->last_stat
, &h
, NULL
);
3327 /* Some file systems (jffs2 or p9fs) don't support mmap() properly (or only read-only
3328 * mmap()), and return EINVAL in that case. Let's propagate that as a more recognizable error
3338 if (!newly_created
) {
3339 set_clear_with_destructor(deferred_closes
, journal_file_close
);
3341 r
= journal_file_verify_header(f
);
3347 if (!newly_created
&& f
->writable
) {
3348 r
= journal_file_fss_load(f
);
3356 journal_default_metrics(metrics
, f
->fd
);
3357 f
->metrics
= *metrics
;
3358 } else if (template)
3359 f
->metrics
= template->metrics
;
3361 r
= journal_file_refresh_header(f
);
3367 r
= journal_file_hmac_setup(f
);
3372 if (newly_created
) {
3373 r
= journal_file_setup_field_hash_table(f
);
3377 r
= journal_file_setup_data_hash_table(f
);
3382 r
= journal_file_append_first_tag(f
);
3388 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
)) {
3393 if (template && template->post_change_timer
) {
3394 r
= journal_file_enable_post_change_timer(
3396 sd_event_source_get_event(template->post_change_timer
),
3397 template->post_change_timer_period
);
3403 /* The file is opened now successfully, thus we take possession of any passed in fd. */
3410 if (f
->cache_fd
&& mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
3413 (void) journal_file_close(f
);
3418 int journal_file_archive(JournalFile
*f
) {
3419 _cleanup_free_
char *p
= NULL
;
3426 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
3427 * rotation, since we don't know the actual path, and couldn't rename the file hence. */
3428 if (path_startswith(f
->path
, "/proc/self/fd"))
3431 if (!endswith(f
->path
, ".journal"))
3434 if (asprintf(&p
, "%.*s@" SD_ID128_FORMAT_STR
"-%016"PRIx64
"-%016"PRIx64
".journal",
3435 (int) strlen(f
->path
) - 8, f
->path
,
3436 SD_ID128_FORMAT_VAL(f
->header
->seqnum_id
),
3437 le64toh(f
->header
->head_entry_seqnum
),
3438 le64toh(f
->header
->head_entry_realtime
)) < 0)
3441 /* Try to rename the file to the archived version. If the file already was deleted, we'll get ENOENT, let's
3442 * ignore that case. */
3443 if (rename(f
->path
, p
) < 0 && errno
!= ENOENT
)
3446 /* Sync the rename to disk */
3447 (void) fsync_directory_of_file(f
->fd
);
3449 /* Set as archive so offlining commits w/state=STATE_ARCHIVED. Previously we would set old_file->header->state
3450 * to STATE_ARCHIVED directly here, but journal_file_set_offline() short-circuits when state != STATE_ONLINE,
3451 * which would result in the rotated journal never getting fsync() called before closing. Now we simply queue
3452 * the archive state by setting an archive bit, leaving the state as STATE_ONLINE so proper offlining
3456 /* Currently, btrfs is not very good with out write patterns and fragments heavily. Let's defrag our journal
3457 * files when we archive them */
3458 f
->defrag_on_close
= true;
3463 JournalFile
* journal_initiate_close(
3465 Set
*deferred_closes
) {
3471 if (deferred_closes
) {
3473 r
= set_put(deferred_closes
, f
);
3475 log_debug_errno(r
, "Failed to add file to deferred close set, closing immediately.");
3477 (void) journal_file_set_offline(f
, false);
3482 return journal_file_close(f
);
3485 int journal_file_rotate(
3488 uint64_t compress_threshold_bytes
,
3490 Set
*deferred_closes
) {
3492 JournalFile
*new_file
= NULL
;
3498 r
= journal_file_archive(*f
);
3502 r
= journal_file_open(
3508 compress_threshold_bytes
,
3516 journal_initiate_close(*f
, deferred_closes
);
3522 int journal_file_dispose(int dir_fd
, const char *fname
) {
3523 _cleanup_free_
char *p
= NULL
;
3524 _cleanup_close_
int fd
= -1;
3528 /* Renames a journal file to *.journal~, i.e. to mark it as corruped or otherwise uncleanly shutdown. Note that
3529 * this is done without looking into the file or changing any of its contents. The idea is that this is called
3530 * whenever something is suspicious and we want to move the file away and make clear that it is not accessed
3531 * for writing anymore. */
3533 if (!endswith(fname
, ".journal"))
3536 if (asprintf(&p
, "%.*s@%016" PRIx64
"-%016" PRIx64
".journal~",
3537 (int) strlen(fname
) - 8, fname
,
3538 now(CLOCK_REALTIME
),
3542 if (renameat(dir_fd
, fname
, dir_fd
, p
) < 0)
3545 /* btrfs doesn't cope well with our write pattern and fragments heavily. Let's defrag all files we rotate */
3546 fd
= openat(dir_fd
, p
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
|O_NOFOLLOW
);
3548 log_debug_errno(errno
, "Failed to open file for defragmentation/FS_NOCOW_FL, ignoring: %m");
3550 (void) chattr_fd(fd
, 0, FS_NOCOW_FL
, NULL
);
3551 (void) btrfs_defrag_fd(fd
);
3557 int journal_file_open_reliably(
3562 uint64_t compress_threshold_bytes
,
3564 JournalMetrics
*metrics
,
3565 MMapCache
*mmap_cache
,
3566 Set
*deferred_closes
,
3567 JournalFile
*template,
3568 JournalFile
**ret
) {
3572 r
= journal_file_open(-1, fname
, flags
, mode
, compress
, compress_threshold_bytes
, seal
, metrics
, mmap_cache
,
3573 deferred_closes
, template, ret
);
3575 -EBADMSG
, /* Corrupted */
3576 -ENODATA
, /* Truncated */
3577 -EHOSTDOWN
, /* Other machine */
3578 -EPROTONOSUPPORT
, /* Incompatible feature */
3579 -EBUSY
, /* Unclean shutdown */
3580 -ESHUTDOWN
, /* Already archived */
3581 -EIO
, /* IO error, including SIGBUS on mmap */
3582 -EIDRM
, /* File has been deleted */
3583 -ETXTBSY
)) /* File is from the future */
3586 if ((flags
& O_ACCMODE
) == O_RDONLY
)
3589 if (!(flags
& O_CREAT
))
3592 if (!endswith(fname
, ".journal"))
3595 /* The file is corrupted. Rotate it away and try it again (but only once) */
3596 log_warning_errno(r
, "File %s corrupted or uncleanly shut down, renaming and replacing.", fname
);
3598 r
= journal_file_dispose(AT_FDCWD
, fname
);
3602 return journal_file_open(-1, fname
, flags
, mode
, compress
, compress_threshold_bytes
, seal
, metrics
, mmap_cache
,
3603 deferred_closes
, template, ret
);
3606 int journal_file_copy_entry(JournalFile
*from
, JournalFile
*to
, Object
*o
, uint64_t p
) {
3608 uint64_t q
, xor_hash
= 0;
3612 const sd_id128_t
*boot_id
;
3622 ts
.monotonic
= le64toh(o
->entry
.monotonic
);
3623 ts
.realtime
= le64toh(o
->entry
.realtime
);
3624 boot_id
= &o
->entry
.boot_id
;
3626 n
= journal_file_entry_n_items(o
);
3627 /* alloca() can't take 0, hence let's allocate at least one */
3628 items
= newa(EntryItem
, MAX(1u, n
));
3630 for (i
= 0; i
< n
; i
++) {
3637 q
= le64toh(o
->entry
.items
[i
].object_offset
);
3638 le_hash
= o
->entry
.items
[i
].hash
;
3640 r
= journal_file_move_to_object(from
, OBJECT_DATA
, q
, &o
);
3644 if (le_hash
!= o
->data
.hash
)
3647 l
= le64toh(o
->object
.size
) - offsetof(Object
, data
.payload
);
3650 /* We hit the limit on 32bit machines */
3651 if ((uint64_t) t
!= l
)
3654 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
3655 #if HAVE_XZ || HAVE_LZ4
3658 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
3659 o
->data
.payload
, l
, &from
->compress_buffer
, &from
->compress_buffer_size
, &rsize
, 0);
3663 data
= from
->compress_buffer
;
3666 return -EPROTONOSUPPORT
;
3669 data
= o
->data
.payload
;
3671 r
= journal_file_append_data(to
, data
, l
, &u
, &h
);
3675 xor_hash
^= le64toh(u
->data
.hash
);
3676 items
[i
].object_offset
= htole64(h
);
3677 items
[i
].hash
= u
->data
.hash
;
3679 r
= journal_file_move_to_object(from
, OBJECT_ENTRY
, p
, &o
);
3684 r
= journal_file_append_entry_internal(to
, &ts
, boot_id
, xor_hash
, items
, n
,
3687 if (mmap_cache_got_sigbus(to
->mmap
, to
->cache_fd
))
3693 void journal_reset_metrics(JournalMetrics
*m
) {
3696 /* Set everything to "pick automatic values". */
3698 *m
= (JournalMetrics
) {
3699 .min_use
= (uint64_t) -1,
3700 .max_use
= (uint64_t) -1,
3701 .min_size
= (uint64_t) -1,
3702 .max_size
= (uint64_t) -1,
3703 .keep_free
= (uint64_t) -1,
3704 .n_max_files
= (uint64_t) -1,
3708 void journal_default_metrics(JournalMetrics
*m
, int fd
) {
3709 char a
[FORMAT_BYTES_MAX
], b
[FORMAT_BYTES_MAX
], c
[FORMAT_BYTES_MAX
], d
[FORMAT_BYTES_MAX
], e
[FORMAT_BYTES_MAX
];
3716 if (fstatvfs(fd
, &ss
) >= 0)
3717 fs_size
= ss
.f_frsize
* ss
.f_blocks
;
3719 log_debug_errno(errno
, "Failed to determine disk size: %m");
3723 if (m
->max_use
== (uint64_t) -1) {
3726 m
->max_use
= PAGE_ALIGN(fs_size
/ 10); /* 10% of file system size */
3728 if (m
->max_use
> DEFAULT_MAX_USE_UPPER
)
3729 m
->max_use
= DEFAULT_MAX_USE_UPPER
;
3731 if (m
->max_use
< DEFAULT_MAX_USE_LOWER
)
3732 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3734 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3736 m
->max_use
= PAGE_ALIGN(m
->max_use
);
3738 if (m
->max_use
!= 0 && m
->max_use
< JOURNAL_FILE_SIZE_MIN
*2)
3739 m
->max_use
= JOURNAL_FILE_SIZE_MIN
*2;
3742 if (m
->min_use
== (uint64_t) -1)
3743 m
->min_use
= DEFAULT_MIN_USE
;
3745 if (m
->min_use
> m
->max_use
)
3746 m
->min_use
= m
->max_use
;
3748 if (m
->max_size
== (uint64_t) -1) {
3749 m
->max_size
= PAGE_ALIGN(m
->max_use
/ 8); /* 8 chunks */
3751 if (m
->max_size
> DEFAULT_MAX_SIZE_UPPER
)
3752 m
->max_size
= DEFAULT_MAX_SIZE_UPPER
;
3754 m
->max_size
= PAGE_ALIGN(m
->max_size
);
3756 if (m
->max_size
!= 0) {
3757 if (m
->max_size
< JOURNAL_FILE_SIZE_MIN
)
3758 m
->max_size
= JOURNAL_FILE_SIZE_MIN
;
3760 if (m
->max_use
!= 0 && m
->max_size
*2 > m
->max_use
)
3761 m
->max_use
= m
->max_size
*2;
3764 if (m
->min_size
== (uint64_t) -1)
3765 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3767 m
->min_size
= PAGE_ALIGN(m
->min_size
);
3769 if (m
->min_size
< JOURNAL_FILE_SIZE_MIN
)
3770 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3772 if (m
->max_size
!= 0 && m
->min_size
> m
->max_size
)
3773 m
->max_size
= m
->min_size
;
3776 if (m
->keep_free
== (uint64_t) -1) {
3779 m
->keep_free
= PAGE_ALIGN(fs_size
* 3 / 20); /* 15% of file system size */
3781 if (m
->keep_free
> DEFAULT_KEEP_FREE_UPPER
)
3782 m
->keep_free
= DEFAULT_KEEP_FREE_UPPER
;
3785 m
->keep_free
= DEFAULT_KEEP_FREE
;
3788 if (m
->n_max_files
== (uint64_t) -1)
3789 m
->n_max_files
= DEFAULT_N_MAX_FILES
;
3791 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64
,
3792 format_bytes(a
, sizeof(a
), m
->min_use
),
3793 format_bytes(b
, sizeof(b
), m
->max_use
),
3794 format_bytes(c
, sizeof(c
), m
->max_size
),
3795 format_bytes(d
, sizeof(d
), m
->min_size
),
3796 format_bytes(e
, sizeof(e
), m
->keep_free
),
3800 int journal_file_get_cutoff_realtime_usec(JournalFile
*f
, usec_t
*from
, usec_t
*to
) {
3806 if (f
->header
->head_entry_realtime
== 0)
3809 *from
= le64toh(f
->header
->head_entry_realtime
);
3813 if (f
->header
->tail_entry_realtime
== 0)
3816 *to
= le64toh(f
->header
->tail_entry_realtime
);
3822 int journal_file_get_cutoff_monotonic_usec(JournalFile
*f
, sd_id128_t boot_id
, usec_t
*from
, usec_t
*to
) {
3830 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &p
);
3834 if (le64toh(o
->data
.n_entries
) <= 0)
3838 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, le64toh(o
->data
.entry_offset
), &o
);
3842 *from
= le64toh(o
->entry
.monotonic
);
3846 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
3850 r
= generic_array_get_plus_one(f
,
3851 le64toh(o
->data
.entry_offset
),
3852 le64toh(o
->data
.entry_array_offset
),
3853 le64toh(o
->data
.n_entries
)-1,
3858 *to
= le64toh(o
->entry
.monotonic
);
3864 bool journal_file_rotate_suggested(JournalFile
*f
, usec_t max_file_usec
) {
3868 /* If we gained new header fields we gained new features,
3869 * hence suggest a rotation */
3870 if (le64toh(f
->header
->header_size
) < sizeof(Header
)) {
3871 log_debug("%s uses an outdated header, suggesting rotation.", f
->path
);
3875 /* Let's check if the hash tables grew over a certain fill
3876 * level (75%, borrowing this value from Java's hash table
3877 * implementation), and if so suggest a rotation. To calculate
3878 * the fill level we need the n_data field, which only exists
3879 * in newer versions. */
3881 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3882 if (le64toh(f
->header
->n_data
) * 4ULL > (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3883 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.",
3885 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))),
3886 le64toh(f
->header
->n_data
),
3887 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3888 (unsigned long long) f
->last_stat
.st_size
,
3889 f
->last_stat
.st_size
/ le64toh(f
->header
->n_data
));
3893 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3894 if (le64toh(f
->header
->n_fields
) * 4ULL > (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3895 log_debug("Field hash table of %s has a fill level at %.1f (%"PRIu64
" of %"PRIu64
" items), suggesting rotation.",
3897 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))),
3898 le64toh(f
->header
->n_fields
),
3899 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
));
3903 /* Are the data objects properly indexed by field objects? */
3904 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
3905 JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
3906 le64toh(f
->header
->n_data
) > 0 &&
3907 le64toh(f
->header
->n_fields
) == 0)
3910 if (max_file_usec
> 0) {
3913 h
= le64toh(f
->header
->head_entry_realtime
);
3914 t
= now(CLOCK_REALTIME
);
3916 if (h
> 0 && t
> h
+ max_file_usec
)