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);
243 r
= pthread_sigmask(SIG_BLOCK
, &ss
, &saved_ss
);
247 r
= pthread_create(&f
->offline_thread
, NULL
, journal_file_set_offline_thread
, f
);
249 k
= pthread_sigmask(SIG_SETMASK
, &saved_ss
, NULL
);
251 f
->offline_state
= OFFLINE_JOINED
;
261 static int journal_file_set_online(JournalFile
*f
) {
269 if (f
->fd
< 0 || !f
->header
)
273 switch (f
->offline_state
) {
275 /* No offline thread, no need to wait. */
279 case OFFLINE_SYNCING
:
280 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_CANCEL
))
282 /* Canceled syncing prior to offlining, no need to wait. */
286 case OFFLINE_AGAIN_FROM_SYNCING
:
287 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_SYNCING
, OFFLINE_CANCEL
))
289 /* Canceled restart from syncing, no need to wait. */
293 case OFFLINE_AGAIN_FROM_OFFLINING
:
294 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_OFFLINING
, OFFLINE_CANCEL
))
296 /* Canceled restart from offlining, must wait for offlining to complete however. */
301 r
= journal_file_set_offline_thread_join(f
);
311 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
314 switch (f
->header
->state
) {
319 f
->header
->state
= STATE_ONLINE
;
328 bool journal_file_is_offlining(JournalFile
*f
) {
331 __sync_synchronize();
333 if (IN_SET(f
->offline_state
, OFFLINE_DONE
, OFFLINE_JOINED
))
339 JournalFile
* journal_file_close(JournalFile
*f
) {
343 /* Write the final tag */
344 if (f
->seal
&& f
->writable
) {
347 r
= journal_file_append_tag(f
);
349 log_error_errno(r
, "Failed to append tag when closing journal: %m");
353 if (f
->post_change_timer
) {
354 if (sd_event_source_get_enabled(f
->post_change_timer
, NULL
) > 0)
355 journal_file_post_change(f
);
357 (void) sd_event_source_set_enabled(f
->post_change_timer
, SD_EVENT_OFF
);
358 sd_event_source_unref(f
->post_change_timer
);
361 journal_file_set_offline(f
, true);
363 if (f
->mmap
&& f
->cache_fd
)
364 mmap_cache_free_fd(f
->mmap
, f
->cache_fd
);
366 if (f
->fd
>= 0 && f
->defrag_on_close
) {
368 /* Be friendly to btrfs: turn COW back on again now,
369 * and defragment the file. We won't write to the file
370 * ever again, hence remove all fragmentation, and
371 * reenable all the good bits COW usually provides
372 * (such as data checksumming). */
374 (void) chattr_fd(f
->fd
, 0, FS_NOCOW_FL
, NULL
);
375 (void) btrfs_defrag_fd(f
->fd
);
382 mmap_cache_unref(f
->mmap
);
384 ordered_hashmap_free_free(f
->chain_cache
);
386 #if HAVE_XZ || HAVE_LZ4
387 free(f
->compress_buffer
);
392 munmap(f
->fss_file
, PAGE_ALIGN(f
->fss_file_size
));
394 free(f
->fsprg_state
);
399 gcry_md_close(f
->hmac
);
405 static int journal_file_init_header(JournalFile
*f
, JournalFile
*template) {
412 memcpy(h
.signature
, HEADER_SIGNATURE
, 8);
413 h
.header_size
= htole64(ALIGN64(sizeof(h
)));
415 h
.incompatible_flags
|= htole32(
416 f
->compress_xz
* HEADER_INCOMPATIBLE_COMPRESSED_XZ
|
417 f
->compress_lz4
* HEADER_INCOMPATIBLE_COMPRESSED_LZ4
);
419 h
.compatible_flags
= htole32(
420 f
->seal
* HEADER_COMPATIBLE_SEALED
);
422 r
= sd_id128_randomize(&h
.file_id
);
427 h
.seqnum_id
= template->header
->seqnum_id
;
428 h
.tail_entry_seqnum
= template->header
->tail_entry_seqnum
;
430 h
.seqnum_id
= h
.file_id
;
432 k
= pwrite(f
->fd
, &h
, sizeof(h
), 0);
442 static int journal_file_refresh_header(JournalFile
*f
) {
449 r
= sd_id128_get_machine(&f
->header
->machine_id
);
450 if (IN_SET(r
, -ENOENT
, -ENOMEDIUM
))
451 /* We don't have a machine-id, let's continue without */
452 zero(f
->header
->machine_id
);
456 r
= sd_id128_get_boot(&boot_id
);
460 f
->header
->boot_id
= boot_id
;
462 r
= journal_file_set_online(f
);
464 /* Sync the online state to disk */
467 /* We likely just created a new file, also sync the directory this file is located in. */
468 (void) fsync_directory_of_file(f
->fd
);
473 static bool warn_wrong_flags(const JournalFile
*f
, bool compatible
) {
474 const uint32_t any
= compatible
? HEADER_COMPATIBLE_ANY
: HEADER_INCOMPATIBLE_ANY
,
475 supported
= compatible
? HEADER_COMPATIBLE_SUPPORTED
: HEADER_INCOMPATIBLE_SUPPORTED
;
476 const char *type
= compatible
? "compatible" : "incompatible";
479 flags
= le32toh(compatible
? f
->header
->compatible_flags
: f
->header
->incompatible_flags
);
481 if (flags
& ~supported
) {
483 log_debug("Journal file %s has unknown %s flags 0x%"PRIx32
,
484 f
->path
, type
, flags
& ~any
);
485 flags
= (flags
& any
) & ~supported
;
489 _cleanup_free_
char *t
= NULL
;
491 if (compatible
&& (flags
& HEADER_COMPATIBLE_SEALED
))
492 strv
[n
++] = "sealed";
493 if (!compatible
&& (flags
& HEADER_INCOMPATIBLE_COMPRESSED_XZ
))
494 strv
[n
++] = "xz-compressed";
495 if (!compatible
&& (flags
& HEADER_INCOMPATIBLE_COMPRESSED_LZ4
))
496 strv
[n
++] = "lz4-compressed";
498 assert(n
< ELEMENTSOF(strv
));
500 t
= strv_join((char**) strv
, ", ");
501 log_debug("Journal file %s uses %s %s %s disabled at compilation time.",
502 f
->path
, type
, n
> 1 ? "flags" : "flag", strnull(t
));
510 static int journal_file_verify_header(JournalFile
*f
) {
511 uint64_t arena_size
, header_size
;
516 if (memcmp(f
->header
->signature
, HEADER_SIGNATURE
, 8))
519 /* In both read and write mode we refuse to open files with incompatible
520 * flags we don't know. */
521 if (warn_wrong_flags(f
, false))
522 return -EPROTONOSUPPORT
;
524 /* When open for writing we refuse to open files with compatible flags, too. */
525 if (f
->writable
&& warn_wrong_flags(f
, true))
526 return -EPROTONOSUPPORT
;
528 if (f
->header
->state
>= _STATE_MAX
)
531 header_size
= le64toh(f
->header
->header_size
);
533 /* The first addition was n_data, so check that we are at least this large */
534 if (header_size
< HEADER_SIZE_MIN
)
537 if (JOURNAL_HEADER_SEALED(f
->header
) && !JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
540 arena_size
= le64toh(f
->header
->arena_size
);
542 if (UINT64_MAX
- header_size
< arena_size
|| header_size
+ arena_size
> (uint64_t) f
->last_stat
.st_size
)
545 if (le64toh(f
->header
->tail_object_offset
) > header_size
+ arena_size
)
548 if (!VALID64(le64toh(f
->header
->data_hash_table_offset
)) ||
549 !VALID64(le64toh(f
->header
->field_hash_table_offset
)) ||
550 !VALID64(le64toh(f
->header
->tail_object_offset
)) ||
551 !VALID64(le64toh(f
->header
->entry_array_offset
)))
555 sd_id128_t machine_id
;
559 r
= sd_id128_get_machine(&machine_id
);
563 if (!sd_id128_equal(machine_id
, f
->header
->machine_id
))
566 state
= f
->header
->state
;
568 if (state
== STATE_ARCHIVED
)
569 return -ESHUTDOWN
; /* Already archived */
570 else if (state
== STATE_ONLINE
)
571 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
),
572 "Journal file %s is already online. Assuming unclean closing.",
574 else if (state
!= STATE_OFFLINE
)
575 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
),
576 "Journal file %s has unknown state %i.",
579 if (f
->header
->field_hash_table_size
== 0 || f
->header
->data_hash_table_size
== 0)
582 /* Don't permit appending to files from the future. Because otherwise the realtime timestamps wouldn't
583 * be strictly ordered in the entries in the file anymore, and we can't have that since it breaks
585 if (le64toh(f
->header
->tail_entry_realtime
) > now(CLOCK_REALTIME
))
586 return log_debug_errno(SYNTHETIC_ERRNO(ETXTBSY
),
587 "Journal file %s is from the future, refusing to append new data to it that'd be older.",
591 f
->compress_xz
= JOURNAL_HEADER_COMPRESSED_XZ(f
->header
);
592 f
->compress_lz4
= JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
);
594 f
->seal
= JOURNAL_HEADER_SEALED(f
->header
);
599 static int journal_file_fstat(JournalFile
*f
) {
605 if (fstat(f
->fd
, &f
->last_stat
) < 0)
608 f
->last_stat_usec
= now(CLOCK_MONOTONIC
);
610 /* Refuse dealing with with files that aren't regular */
611 r
= stat_verify_regular(&f
->last_stat
);
615 /* Refuse appending to files that are already deleted */
616 if (f
->last_stat
.st_nlink
<= 0)
622 static int journal_file_allocate(JournalFile
*f
, uint64_t offset
, uint64_t size
) {
623 uint64_t old_size
, new_size
;
629 /* We assume that this file is not sparse, and we know that
630 * for sure, since we always call posix_fallocate()
633 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
637 le64toh(f
->header
->header_size
) +
638 le64toh(f
->header
->arena_size
);
640 new_size
= PAGE_ALIGN(offset
+ size
);
641 if (new_size
< le64toh(f
->header
->header_size
))
642 new_size
= le64toh(f
->header
->header_size
);
644 if (new_size
<= old_size
) {
646 /* We already pre-allocated enough space, but before
647 * we write to it, let's check with fstat() if the
648 * file got deleted, in order make sure we don't throw
649 * away the data immediately. Don't check fstat() for
650 * all writes though, but only once ever 10s. */
652 if (f
->last_stat_usec
+ LAST_STAT_REFRESH_USEC
> now(CLOCK_MONOTONIC
))
655 return journal_file_fstat(f
);
658 /* Allocate more space. */
660 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
663 if (new_size
> f
->metrics
.min_size
&& f
->metrics
.keep_free
> 0) {
666 if (fstatvfs(f
->fd
, &svfs
) >= 0) {
669 available
= LESS_BY((uint64_t) svfs
.f_bfree
* (uint64_t) svfs
.f_bsize
, f
->metrics
.keep_free
);
671 if (new_size
- old_size
> available
)
676 /* Increase by larger blocks at once */
677 new_size
= DIV_ROUND_UP(new_size
, FILE_SIZE_INCREASE
) * FILE_SIZE_INCREASE
;
678 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
679 new_size
= f
->metrics
.max_size
;
681 /* Note that the glibc fallocate() fallback is very
682 inefficient, hence we try to minimize the allocation area
684 r
= posix_fallocate(f
->fd
, old_size
, new_size
- old_size
);
688 f
->header
->arena_size
= htole64(new_size
- le64toh(f
->header
->header_size
));
690 return journal_file_fstat(f
);
693 static unsigned type_to_context(ObjectType type
) {
694 /* One context for each type, plus one catch-all for the rest */
695 assert_cc(_OBJECT_TYPE_MAX
<= MMAP_CACHE_MAX_CONTEXTS
);
696 assert_cc(CONTEXT_HEADER
< MMAP_CACHE_MAX_CONTEXTS
);
697 return type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
? type
: 0;
700 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
) {
709 /* Avoid SIGBUS on invalid accesses */
710 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
) {
711 /* Hmm, out of range? Let's refresh the fstat() data
712 * first, before we trust that check. */
714 r
= journal_file_fstat(f
);
718 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
)
719 return -EADDRNOTAVAIL
;
722 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
);
725 static uint64_t minimum_header_size(Object
*o
) {
727 static const uint64_t table
[] = {
728 [OBJECT_DATA
] = sizeof(DataObject
),
729 [OBJECT_FIELD
] = sizeof(FieldObject
),
730 [OBJECT_ENTRY
] = sizeof(EntryObject
),
731 [OBJECT_DATA_HASH_TABLE
] = sizeof(HashTableObject
),
732 [OBJECT_FIELD_HASH_TABLE
] = sizeof(HashTableObject
),
733 [OBJECT_ENTRY_ARRAY
] = sizeof(EntryArrayObject
),
734 [OBJECT_TAG
] = sizeof(TagObject
),
737 if (o
->object
.type
>= ELEMENTSOF(table
) || table
[o
->object
.type
] <= 0)
738 return sizeof(ObjectHeader
);
740 return table
[o
->object
.type
];
743 /* Lightweight object checks. We want this to be fast, so that we won't
744 * slowdown every journal_file_move_to_object() call too much. */
745 static int journal_file_check_object(JournalFile
*f
, uint64_t offset
, Object
*o
) {
749 switch (o
->object
.type
) {
752 if ((le64toh(o
->data
.entry_offset
) == 0) ^ (le64toh(o
->data
.n_entries
) == 0))
753 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
754 "Bad n_entries: %" PRIu64
": %" PRIu64
,
755 le64toh(o
->data
.n_entries
),
758 if (le64toh(o
->object
.size
) - offsetof(DataObject
, payload
) <= 0)
759 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
760 "Bad object size (<= %zu): %" PRIu64
": %" PRIu64
,
761 offsetof(DataObject
, payload
),
762 le64toh(o
->object
.size
),
765 if (!VALID64(le64toh(o
->data
.next_hash_offset
)) ||
766 !VALID64(le64toh(o
->data
.next_field_offset
)) ||
767 !VALID64(le64toh(o
->data
.entry_offset
)) ||
768 !VALID64(le64toh(o
->data
.entry_array_offset
)))
769 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
770 "Invalid offset, next_hash_offset=" OFSfmt
", next_field_offset=" OFSfmt
", entry_offset=" OFSfmt
", entry_array_offset=" OFSfmt
": %" PRIu64
,
771 le64toh(o
->data
.next_hash_offset
),
772 le64toh(o
->data
.next_field_offset
),
773 le64toh(o
->data
.entry_offset
),
774 le64toh(o
->data
.entry_array_offset
),
781 if (le64toh(o
->object
.size
) - offsetof(FieldObject
, payload
) <= 0)
782 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
783 "Bad field size (<= %zu): %" PRIu64
": %" PRIu64
,
784 offsetof(FieldObject
, payload
),
785 le64toh(o
->object
.size
),
788 if (!VALID64(le64toh(o
->field
.next_hash_offset
)) ||
789 !VALID64(le64toh(o
->field
.head_data_offset
)))
790 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
791 "Invalid offset, next_hash_offset=" OFSfmt
", head_data_offset=" OFSfmt
": %" PRIu64
,
792 le64toh(o
->field
.next_hash_offset
),
793 le64toh(o
->field
.head_data_offset
),
798 if ((le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) % sizeof(EntryItem
) != 0)
799 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
800 "Bad entry size (<= %zu): %" PRIu64
": %" PRIu64
,
801 offsetof(EntryObject
, items
),
802 le64toh(o
->object
.size
),
805 if ((le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) / sizeof(EntryItem
) <= 0)
806 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
807 "Invalid number items in entry: %" PRIu64
": %" PRIu64
,
808 (le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) / sizeof(EntryItem
),
811 if (le64toh(o
->entry
.seqnum
) <= 0)
812 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
813 "Invalid entry seqnum: %" PRIx64
": %" PRIu64
,
814 le64toh(o
->entry
.seqnum
),
817 if (!VALID_REALTIME(le64toh(o
->entry
.realtime
)))
818 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
819 "Invalid entry realtime timestamp: %" PRIu64
": %" PRIu64
,
820 le64toh(o
->entry
.realtime
),
823 if (!VALID_MONOTONIC(le64toh(o
->entry
.monotonic
)))
824 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
825 "Invalid entry monotonic timestamp: %" PRIu64
": %" PRIu64
,
826 le64toh(o
->entry
.monotonic
),
831 case OBJECT_DATA_HASH_TABLE
:
832 case OBJECT_FIELD_HASH_TABLE
:
833 if ((le64toh(o
->object
.size
) - offsetof(HashTableObject
, items
)) % sizeof(HashItem
) != 0 ||
834 (le64toh(o
->object
.size
) - offsetof(HashTableObject
, items
)) / sizeof(HashItem
) <= 0)
835 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
836 "Invalid %s hash table size: %" PRIu64
": %" PRIu64
,
837 o
->object
.type
== OBJECT_DATA_HASH_TABLE
? "data" : "field",
838 le64toh(o
->object
.size
),
843 case OBJECT_ENTRY_ARRAY
:
844 if ((le64toh(o
->object
.size
) - offsetof(EntryArrayObject
, items
)) % sizeof(le64_t
) != 0 ||
845 (le64toh(o
->object
.size
) - offsetof(EntryArrayObject
, items
)) / sizeof(le64_t
) <= 0)
846 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
847 "Invalid object entry array size: %" PRIu64
": %" PRIu64
,
848 le64toh(o
->object
.size
),
851 if (!VALID64(le64toh(o
->entry_array
.next_entry_array_offset
)))
852 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
853 "Invalid object entry array next_entry_array_offset: " OFSfmt
": %" PRIu64
,
854 le64toh(o
->entry_array
.next_entry_array_offset
),
860 if (le64toh(o
->object
.size
) != sizeof(TagObject
))
861 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
862 "Invalid object tag size: %" PRIu64
": %" PRIu64
,
863 le64toh(o
->object
.size
),
866 if (!VALID_EPOCH(le64toh(o
->tag
.epoch
)))
867 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
868 "Invalid object tag epoch: %" PRIu64
": %" PRIu64
,
869 le64toh(o
->tag
.epoch
), offset
);
877 int journal_file_move_to_object(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
**ret
) {
887 /* Objects may only be located at multiple of 64 bit */
888 if (!VALID64(offset
))
889 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
890 "Attempt to move to object at non-64bit boundary: %" PRIu64
,
893 /* Object may not be located in the file header */
894 if (offset
< le64toh(f
->header
->header_size
))
895 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
896 "Attempt to move to object located in file header: %" PRIu64
,
899 r
= journal_file_move_to(f
, type
, false, offset
, sizeof(ObjectHeader
), &t
, &tsize
);
904 s
= le64toh(o
->object
.size
);
907 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
908 "Attempt to move to uninitialized object: %" PRIu64
,
910 if (s
< sizeof(ObjectHeader
))
911 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
912 "Attempt to move to overly short object: %" PRIu64
,
915 if (o
->object
.type
<= OBJECT_UNUSED
)
916 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
917 "Attempt to move to object with invalid type: %" PRIu64
,
920 if (s
< minimum_header_size(o
))
921 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
922 "Attempt to move to truncated object: %" PRIu64
,
925 if (type
> OBJECT_UNUSED
&& o
->object
.type
!= type
)
926 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
927 "Attempt to move to object of unexpected type: %" PRIu64
,
931 r
= journal_file_move_to(f
, type
, false, offset
, s
, &t
, NULL
);
938 r
= journal_file_check_object(f
, offset
, o
);
946 static uint64_t journal_file_entry_seqnum(JournalFile
*f
, uint64_t *seqnum
) {
952 r
= le64toh(f
->header
->tail_entry_seqnum
) + 1;
955 /* If an external seqnum counter was passed, we update
956 * both the local and the external one, and set it to
957 * the maximum of both */
965 f
->header
->tail_entry_seqnum
= htole64(r
);
967 if (f
->header
->head_entry_seqnum
== 0)
968 f
->header
->head_entry_seqnum
= htole64(r
);
973 int journal_file_append_object(JournalFile
*f
, ObjectType type
, uint64_t size
, Object
**ret
, uint64_t *offset
) {
981 assert(type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
);
982 assert(size
>= sizeof(ObjectHeader
));
986 r
= journal_file_set_online(f
);
990 p
= le64toh(f
->header
->tail_object_offset
);
992 p
= le64toh(f
->header
->header_size
);
994 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &tail
);
998 p
+= ALIGN64(le64toh(tail
->object
.size
));
1001 r
= journal_file_allocate(f
, p
, size
);
1005 r
= journal_file_move_to(f
, type
, false, p
, size
, &t
, NULL
);
1012 o
->object
.type
= type
;
1013 o
->object
.size
= htole64(size
);
1015 f
->header
->tail_object_offset
= htole64(p
);
1016 f
->header
->n_objects
= htole64(le64toh(f
->header
->n_objects
) + 1);
1024 static int journal_file_setup_data_hash_table(JournalFile
*f
) {
1032 /* We estimate that we need 1 hash table entry per 768 bytes
1033 of journal file and we want to make sure we never get
1034 beyond 75% fill level. Calculate the hash table size for
1035 the maximum file size based on these metrics. */
1037 s
= (f
->metrics
.max_size
* 4 / 768 / 3) * sizeof(HashItem
);
1038 if (s
< DEFAULT_DATA_HASH_TABLE_SIZE
)
1039 s
= DEFAULT_DATA_HASH_TABLE_SIZE
;
1041 log_debug("Reserving %"PRIu64
" entries in hash table.", s
/ sizeof(HashItem
));
1043 r
= journal_file_append_object(f
,
1044 OBJECT_DATA_HASH_TABLE
,
1045 offsetof(Object
, hash_table
.items
) + s
,
1050 memzero(o
->hash_table
.items
, s
);
1052 f
->header
->data_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1053 f
->header
->data_hash_table_size
= htole64(s
);
1058 static int journal_file_setup_field_hash_table(JournalFile
*f
) {
1066 /* We use a fixed size hash table for the fields as this
1067 * number should grow very slowly only */
1069 s
= DEFAULT_FIELD_HASH_TABLE_SIZE
;
1070 r
= journal_file_append_object(f
,
1071 OBJECT_FIELD_HASH_TABLE
,
1072 offsetof(Object
, hash_table
.items
) + s
,
1077 memzero(o
->hash_table
.items
, s
);
1079 f
->header
->field_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1080 f
->header
->field_hash_table_size
= htole64(s
);
1085 int journal_file_map_data_hash_table(JournalFile
*f
) {
1093 if (f
->data_hash_table
)
1096 p
= le64toh(f
->header
->data_hash_table_offset
);
1097 s
= le64toh(f
->header
->data_hash_table_size
);
1099 r
= journal_file_move_to(f
,
1100 OBJECT_DATA_HASH_TABLE
,
1107 f
->data_hash_table
= t
;
1111 int journal_file_map_field_hash_table(JournalFile
*f
) {
1119 if (f
->field_hash_table
)
1122 p
= le64toh(f
->header
->field_hash_table_offset
);
1123 s
= le64toh(f
->header
->field_hash_table_size
);
1125 r
= journal_file_move_to(f
,
1126 OBJECT_FIELD_HASH_TABLE
,
1133 f
->field_hash_table
= t
;
1137 static int journal_file_link_field(
1148 assert(f
->field_hash_table
);
1152 if (o
->object
.type
!= OBJECT_FIELD
)
1155 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1159 /* This might alter the window we are looking at */
1160 o
->field
.next_hash_offset
= o
->field
.head_data_offset
= 0;
1163 p
= le64toh(f
->field_hash_table
[h
].tail_hash_offset
);
1165 f
->field_hash_table
[h
].head_hash_offset
= htole64(offset
);
1167 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1171 o
->field
.next_hash_offset
= htole64(offset
);
1174 f
->field_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1176 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
1177 f
->header
->n_fields
= htole64(le64toh(f
->header
->n_fields
) + 1);
1182 static int journal_file_link_data(
1193 assert(f
->data_hash_table
);
1197 if (o
->object
.type
!= OBJECT_DATA
)
1200 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1204 /* This might alter the window we are looking at */
1205 o
->data
.next_hash_offset
= o
->data
.next_field_offset
= 0;
1206 o
->data
.entry_offset
= o
->data
.entry_array_offset
= 0;
1207 o
->data
.n_entries
= 0;
1210 p
= le64toh(f
->data_hash_table
[h
].tail_hash_offset
);
1212 /* Only entry in the hash table is easy */
1213 f
->data_hash_table
[h
].head_hash_offset
= htole64(offset
);
1215 /* Move back to the previous data object, to patch in
1218 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1222 o
->data
.next_hash_offset
= htole64(offset
);
1225 f
->data_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1227 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
1228 f
->header
->n_data
= htole64(le64toh(f
->header
->n_data
) + 1);
1233 int journal_file_find_field_object_with_hash(
1235 const void *field
, uint64_t size
, uint64_t hash
,
1236 Object
**ret
, uint64_t *offset
) {
1238 uint64_t p
, osize
, h
, m
;
1243 assert(field
&& size
> 0);
1245 /* If the field hash table is empty, we can't find anything */
1246 if (le64toh(f
->header
->field_hash_table_size
) <= 0)
1249 /* Map the field hash table, if it isn't mapped yet. */
1250 r
= journal_file_map_field_hash_table(f
);
1254 osize
= offsetof(Object
, field
.payload
) + size
;
1256 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1261 p
= le64toh(f
->field_hash_table
[h
].head_hash_offset
);
1266 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1270 if (le64toh(o
->field
.hash
) == hash
&&
1271 le64toh(o
->object
.size
) == osize
&&
1272 memcmp(o
->field
.payload
, field
, size
) == 0) {
1282 p
= le64toh(o
->field
.next_hash_offset
);
1288 int journal_file_find_field_object(
1290 const void *field
, uint64_t size
,
1291 Object
**ret
, uint64_t *offset
) {
1296 assert(field
&& size
> 0);
1298 hash
= hash64(field
, size
);
1300 return journal_file_find_field_object_with_hash(f
,
1305 int journal_file_find_data_object_with_hash(
1307 const void *data
, uint64_t size
, uint64_t hash
,
1308 Object
**ret
, uint64_t *offset
) {
1310 uint64_t p
, osize
, h
, m
;
1315 assert(data
|| size
== 0);
1317 /* If there's no data hash table, then there's no entry. */
1318 if (le64toh(f
->header
->data_hash_table_size
) <= 0)
1321 /* Map the data hash table, if it isn't mapped yet. */
1322 r
= journal_file_map_data_hash_table(f
);
1326 osize
= offsetof(Object
, data
.payload
) + size
;
1328 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1333 p
= le64toh(f
->data_hash_table
[h
].head_hash_offset
);
1338 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1342 if (le64toh(o
->data
.hash
) != hash
)
1345 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
1346 #if HAVE_XZ || HAVE_LZ4
1350 l
= le64toh(o
->object
.size
);
1351 if (l
<= offsetof(Object
, data
.payload
))
1354 l
-= offsetof(Object
, data
.payload
);
1356 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
1357 o
->data
.payload
, l
, &f
->compress_buffer
, &f
->compress_buffer_size
, &rsize
, 0);
1361 if (rsize
== size
&&
1362 memcmp(f
->compress_buffer
, data
, size
) == 0) {
1373 return -EPROTONOSUPPORT
;
1375 } else if (le64toh(o
->object
.size
) == osize
&&
1376 memcmp(o
->data
.payload
, data
, size
) == 0) {
1388 p
= le64toh(o
->data
.next_hash_offset
);
1394 int journal_file_find_data_object(
1396 const void *data
, uint64_t size
,
1397 Object
**ret
, uint64_t *offset
) {
1402 assert(data
|| size
== 0);
1404 hash
= hash64(data
, size
);
1406 return journal_file_find_data_object_with_hash(f
,
1411 static int journal_file_append_field(
1413 const void *field
, uint64_t size
,
1414 Object
**ret
, uint64_t *offset
) {
1422 assert(field
&& size
> 0);
1424 hash
= hash64(field
, size
);
1426 r
= journal_file_find_field_object_with_hash(f
, field
, size
, hash
, &o
, &p
);
1440 osize
= offsetof(Object
, field
.payload
) + size
;
1441 r
= journal_file_append_object(f
, OBJECT_FIELD
, osize
, &o
, &p
);
1445 o
->field
.hash
= htole64(hash
);
1446 memcpy(o
->field
.payload
, field
, size
);
1448 r
= journal_file_link_field(f
, o
, p
, hash
);
1452 /* The linking might have altered the window, so let's
1453 * refresh our pointer */
1454 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1459 r
= journal_file_hmac_put_object(f
, OBJECT_FIELD
, o
, p
);
1473 static int journal_file_append_data(
1475 const void *data
, uint64_t size
,
1476 Object
**ret
, uint64_t *offset
) {
1481 int r
, compression
= 0;
1485 assert(data
|| size
== 0);
1487 hash
= hash64(data
, size
);
1489 r
= journal_file_find_data_object_with_hash(f
, data
, size
, hash
, &o
, &p
);
1503 osize
= offsetof(Object
, data
.payload
) + size
;
1504 r
= journal_file_append_object(f
, OBJECT_DATA
, osize
, &o
, &p
);
1508 o
->data
.hash
= htole64(hash
);
1510 #if HAVE_XZ || HAVE_LZ4
1511 if (JOURNAL_FILE_COMPRESS(f
) && size
>= f
->compress_threshold_bytes
) {
1514 compression
= compress_blob(data
, size
, o
->data
.payload
, size
- 1, &rsize
);
1516 if (compression
>= 0) {
1517 o
->object
.size
= htole64(offsetof(Object
, data
.payload
) + rsize
);
1518 o
->object
.flags
|= compression
;
1520 log_debug("Compressed data object %"PRIu64
" -> %zu using %s",
1521 size
, rsize
, object_compressed_to_string(compression
));
1523 /* Compression didn't work, we don't really care why, let's continue without compression */
1528 if (compression
== 0)
1529 memcpy_safe(o
->data
.payload
, data
, size
);
1531 r
= journal_file_link_data(f
, o
, p
, hash
);
1536 r
= journal_file_hmac_put_object(f
, OBJECT_DATA
, o
, p
);
1541 /* The linking might have altered the window, so let's
1542 * refresh our pointer */
1543 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1550 eq
= memchr(data
, '=', size
);
1551 if (eq
&& eq
> data
) {
1555 /* Create field object ... */
1556 r
= journal_file_append_field(f
, data
, (uint8_t*) eq
- (uint8_t*) data
, &fo
, &fp
);
1560 /* ... and link it in. */
1561 o
->data
.next_field_offset
= fo
->field
.head_data_offset
;
1562 fo
->field
.head_data_offset
= le64toh(p
);
1574 uint64_t journal_file_entry_n_items(Object
*o
) {
1577 if (o
->object
.type
!= OBJECT_ENTRY
)
1580 return (le64toh(o
->object
.size
) - offsetof(Object
, entry
.items
)) / sizeof(EntryItem
);
1583 uint64_t journal_file_entry_array_n_items(Object
*o
) {
1586 if (o
->object
.type
!= OBJECT_ENTRY_ARRAY
)
1589 return (le64toh(o
->object
.size
) - offsetof(Object
, entry_array
.items
)) / sizeof(uint64_t);
1592 uint64_t journal_file_hash_table_n_items(Object
*o
) {
1595 if (!IN_SET(o
->object
.type
, OBJECT_DATA_HASH_TABLE
, OBJECT_FIELD_HASH_TABLE
))
1598 return (le64toh(o
->object
.size
) - offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
);
1601 static int link_entry_into_array(JournalFile
*f
,
1606 uint64_t n
= 0, ap
= 0, q
, i
, a
, hidx
;
1615 a
= le64toh(*first
);
1616 i
= hidx
= le64toh(*idx
);
1619 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
1623 n
= journal_file_entry_array_n_items(o
);
1625 o
->entry_array
.items
[i
] = htole64(p
);
1626 *idx
= htole64(hidx
+ 1);
1632 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
1643 r
= journal_file_append_object(f
, OBJECT_ENTRY_ARRAY
,
1644 offsetof(Object
, entry_array
.items
) + n
* sizeof(uint64_t),
1650 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY_ARRAY
, o
, q
);
1655 o
->entry_array
.items
[i
] = htole64(p
);
1658 *first
= htole64(q
);
1660 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, ap
, &o
);
1664 o
->entry_array
.next_entry_array_offset
= htole64(q
);
1667 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
1668 f
->header
->n_entry_arrays
= htole64(le64toh(f
->header
->n_entry_arrays
) + 1);
1670 *idx
= htole64(hidx
+ 1);
1675 static int link_entry_into_array_plus_one(JournalFile
*f
,
1690 *extra
= htole64(p
);
1694 i
= htole64(le64toh(*idx
) - 1);
1695 r
= link_entry_into_array(f
, first
, &i
, p
);
1700 *idx
= htole64(le64toh(*idx
) + 1);
1704 static int journal_file_link_entry_item(JournalFile
*f
, Object
*o
, uint64_t offset
, uint64_t i
) {
1711 p
= le64toh(o
->entry
.items
[i
].object_offset
);
1715 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1719 return link_entry_into_array_plus_one(f
,
1720 &o
->data
.entry_offset
,
1721 &o
->data
.entry_array_offset
,
1726 static int journal_file_link_entry(JournalFile
*f
, Object
*o
, uint64_t offset
) {
1735 if (o
->object
.type
!= OBJECT_ENTRY
)
1738 __sync_synchronize();
1740 /* Link up the entry itself */
1741 r
= link_entry_into_array(f
,
1742 &f
->header
->entry_array_offset
,
1743 &f
->header
->n_entries
,
1748 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
1750 if (f
->header
->head_entry_realtime
== 0)
1751 f
->header
->head_entry_realtime
= o
->entry
.realtime
;
1753 f
->header
->tail_entry_realtime
= o
->entry
.realtime
;
1754 f
->header
->tail_entry_monotonic
= o
->entry
.monotonic
;
1756 /* Link up the items */
1757 n
= journal_file_entry_n_items(o
);
1758 for (i
= 0; i
< n
; i
++) {
1759 r
= journal_file_link_entry_item(f
, o
, offset
, i
);
1767 static int journal_file_append_entry_internal(
1769 const dual_timestamp
*ts
,
1770 const sd_id128_t
*boot_id
,
1772 const EntryItem items
[], unsigned n_items
,
1774 Object
**ret
, uint64_t *offset
) {
1782 assert(items
|| n_items
== 0);
1785 osize
= offsetof(Object
, entry
.items
) + (n_items
* sizeof(EntryItem
));
1787 r
= journal_file_append_object(f
, OBJECT_ENTRY
, osize
, &o
, &np
);
1791 o
->entry
.seqnum
= htole64(journal_file_entry_seqnum(f
, seqnum
));
1792 memcpy_safe(o
->entry
.items
, items
, n_items
* sizeof(EntryItem
));
1793 o
->entry
.realtime
= htole64(ts
->realtime
);
1794 o
->entry
.monotonic
= htole64(ts
->monotonic
);
1795 o
->entry
.xor_hash
= htole64(xor_hash
);
1796 o
->entry
.boot_id
= boot_id
? *boot_id
: f
->header
->boot_id
;
1799 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY
, o
, np
);
1804 r
= journal_file_link_entry(f
, o
, np
);
1817 void journal_file_post_change(JournalFile
*f
) {
1823 /* inotify() does not receive IN_MODIFY events from file
1824 * accesses done via mmap(). After each access we hence
1825 * trigger IN_MODIFY by truncating the journal file to its
1826 * current size which triggers IN_MODIFY. */
1828 __sync_synchronize();
1830 if (ftruncate(f
->fd
, f
->last_stat
.st_size
) < 0)
1831 log_debug_errno(errno
, "Failed to truncate file to its own size: %m");
1834 static int post_change_thunk(sd_event_source
*timer
, uint64_t usec
, void *userdata
) {
1837 journal_file_post_change(userdata
);
1842 static void schedule_post_change(JournalFile
*f
) {
1847 assert(f
->post_change_timer
);
1849 r
= sd_event_source_get_enabled(f
->post_change_timer
, NULL
);
1851 log_debug_errno(r
, "Failed to get ftruncate timer state: %m");
1857 r
= sd_event_now(sd_event_source_get_event(f
->post_change_timer
), CLOCK_MONOTONIC
, &now
);
1859 log_debug_errno(r
, "Failed to get clock's now for scheduling ftruncate: %m");
1863 r
= sd_event_source_set_time(f
->post_change_timer
, now
+ f
->post_change_timer_period
);
1865 log_debug_errno(r
, "Failed to set time for scheduling ftruncate: %m");
1869 r
= sd_event_source_set_enabled(f
->post_change_timer
, SD_EVENT_ONESHOT
);
1871 log_debug_errno(r
, "Failed to enable scheduled ftruncate: %m");
1878 /* On failure, let's simply post the change immediately. */
1879 journal_file_post_change(f
);
1882 /* Enable coalesced change posting in a timer on the provided sd_event instance */
1883 int journal_file_enable_post_change_timer(JournalFile
*f
, sd_event
*e
, usec_t t
) {
1884 _cleanup_(sd_event_source_unrefp
) sd_event_source
*timer
= NULL
;
1888 assert_return(!f
->post_change_timer
, -EINVAL
);
1892 r
= sd_event_add_time(e
, &timer
, CLOCK_MONOTONIC
, 0, 0, post_change_thunk
, f
);
1896 r
= sd_event_source_set_enabled(timer
, SD_EVENT_OFF
);
1900 f
->post_change_timer
= TAKE_PTR(timer
);
1901 f
->post_change_timer_period
= t
;
1906 static int entry_item_cmp(const EntryItem
*a
, const EntryItem
*b
) {
1907 return CMP(le64toh(a
->object_offset
), le64toh(b
->object_offset
));
1910 int journal_file_append_entry(
1912 const dual_timestamp
*ts
,
1913 const sd_id128_t
*boot_id
,
1914 const struct iovec iovec
[], unsigned n_iovec
,
1916 Object
**ret
, uint64_t *offset
) {
1921 uint64_t xor_hash
= 0;
1922 struct dual_timestamp _ts
;
1926 assert(iovec
|| n_iovec
== 0);
1929 if (!VALID_REALTIME(ts
->realtime
))
1930 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1931 "Invalid realtime timestamp %" PRIu64
", refusing entry.",
1933 if (!VALID_MONOTONIC(ts
->monotonic
))
1934 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1935 "Invalid monotomic timestamp %" PRIu64
", refusing entry.",
1938 dual_timestamp_get(&_ts
);
1943 r
= journal_file_maybe_append_tag(f
, ts
->realtime
);
1948 /* alloca() can't take 0, hence let's allocate at least one */
1949 items
= newa(EntryItem
, MAX(1u, n_iovec
));
1951 for (i
= 0; i
< n_iovec
; i
++) {
1955 r
= journal_file_append_data(f
, iovec
[i
].iov_base
, iovec
[i
].iov_len
, &o
, &p
);
1959 xor_hash
^= le64toh(o
->data
.hash
);
1960 items
[i
].object_offset
= htole64(p
);
1961 items
[i
].hash
= o
->data
.hash
;
1964 /* Order by the position on disk, in order to improve seek
1965 * times for rotating media. */
1966 typesafe_qsort(items
, n_iovec
, entry_item_cmp
);
1968 r
= journal_file_append_entry_internal(f
, ts
, boot_id
, xor_hash
, items
, n_iovec
, seqnum
, ret
, offset
);
1970 /* If the memory mapping triggered a SIGBUS then we return an
1971 * IO error and ignore the error code passed down to us, since
1972 * it is very likely just an effect of a nullified replacement
1975 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
1978 if (f
->post_change_timer
)
1979 schedule_post_change(f
);
1981 journal_file_post_change(f
);
1986 typedef struct ChainCacheItem
{
1987 uint64_t first
; /* the array at the beginning of the chain */
1988 uint64_t array
; /* the cached array */
1989 uint64_t begin
; /* the first item in the cached array */
1990 uint64_t total
; /* the total number of items in all arrays before this one in the chain */
1991 uint64_t last_index
; /* the last index we looked at, to optimize locality when bisecting */
1994 static void chain_cache_put(
2001 uint64_t last_index
) {
2004 /* If the chain item to cache for this chain is the
2005 * first one it's not worth caching anything */
2009 if (ordered_hashmap_size(h
) >= CHAIN_CACHE_MAX
) {
2010 ci
= ordered_hashmap_steal_first(h
);
2013 ci
= new(ChainCacheItem
, 1);
2020 if (ordered_hashmap_put(h
, &ci
->first
, ci
) < 0) {
2025 assert(ci
->first
== first
);
2030 ci
->last_index
= last_index
;
2033 static int generic_array_get(
2037 Object
**ret
, uint64_t *offset
) {
2040 uint64_t p
= 0, a
, t
= 0;
2048 /* Try the chain cache first */
2049 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2050 if (ci
&& i
> ci
->total
) {
2059 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2063 k
= journal_file_entry_array_n_items(o
);
2065 p
= le64toh(o
->entry_array
.items
[i
]);
2071 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
2077 /* Let's cache this item for the next invocation */
2078 chain_cache_put(f
->chain_cache
, ci
, first
, a
, le64toh(o
->entry_array
.items
[0]), t
, i
);
2080 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2093 static int generic_array_get_plus_one(
2098 Object
**ret
, uint64_t *offset
) {
2107 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2120 return generic_array_get(f
, first
, i
-1, ret
, offset
);
2129 static int generic_array_bisect(
2134 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2135 direction_t direction
,
2140 uint64_t a
, p
, t
= 0, i
= 0, last_p
= 0, last_index
= (uint64_t) -1;
2141 bool subtract_one
= false;
2142 Object
*o
, *array
= NULL
;
2147 assert(test_object
);
2149 /* Start with the first array in the chain */
2152 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2153 if (ci
&& n
> ci
->total
&& ci
->begin
!= 0) {
2154 /* Ah, we have iterated this bisection array chain
2155 * previously! Let's see if we can skip ahead in the
2156 * chain, as far as the last time. But we can't jump
2157 * backwards in the chain, so let's check that
2160 r
= test_object(f
, ci
->begin
, needle
);
2164 if (r
== TEST_LEFT
) {
2165 /* OK, what we are looking for is right of the
2166 * begin of this EntryArray, so let's jump
2167 * straight to previously cached array in the
2173 last_index
= ci
->last_index
;
2178 uint64_t left
, right
, k
, lp
;
2180 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2184 k
= journal_file_entry_array_n_items(array
);
2190 lp
= p
= le64toh(array
->entry_array
.items
[i
]);
2194 r
= test_object(f
, p
, needle
);
2195 if (r
== -EBADMSG
) {
2196 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (1)");
2203 if (r
== TEST_FOUND
)
2204 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2206 if (r
== TEST_RIGHT
) {
2210 if (last_index
!= (uint64_t) -1) {
2211 assert(last_index
<= right
);
2213 /* If we cached the last index we
2214 * looked at, let's try to not to jump
2215 * too wildly around and see if we can
2216 * limit the range to look at early to
2217 * the immediate neighbors of the last
2218 * index we looked at. */
2220 if (last_index
> 0) {
2221 uint64_t x
= last_index
- 1;
2223 p
= le64toh(array
->entry_array
.items
[x
]);
2227 r
= test_object(f
, p
, needle
);
2231 if (r
== TEST_FOUND
)
2232 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2234 if (r
== TEST_RIGHT
)
2240 if (last_index
< right
) {
2241 uint64_t y
= last_index
+ 1;
2243 p
= le64toh(array
->entry_array
.items
[y
]);
2247 r
= test_object(f
, p
, needle
);
2251 if (r
== TEST_FOUND
)
2252 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2254 if (r
== TEST_RIGHT
)
2262 if (left
== right
) {
2263 if (direction
== DIRECTION_UP
)
2264 subtract_one
= true;
2270 assert(left
< right
);
2271 i
= (left
+ right
) / 2;
2273 p
= le64toh(array
->entry_array
.items
[i
]);
2277 r
= test_object(f
, p
, needle
);
2278 if (r
== -EBADMSG
) {
2279 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (2)");
2286 if (r
== TEST_FOUND
)
2287 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2289 if (r
== TEST_RIGHT
)
2297 if (direction
== DIRECTION_UP
) {
2299 subtract_one
= true;
2310 last_index
= (uint64_t) -1;
2311 a
= le64toh(array
->entry_array
.next_entry_array_offset
);
2317 if (subtract_one
&& t
== 0 && i
== 0)
2320 /* Let's cache this item for the next invocation */
2321 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
);
2323 if (subtract_one
&& i
== 0)
2325 else if (subtract_one
)
2326 p
= le64toh(array
->entry_array
.items
[i
-1]);
2328 p
= le64toh(array
->entry_array
.items
[i
]);
2330 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2341 *idx
= t
+ i
+ (subtract_one
? -1 : 0);
2346 static int generic_array_bisect_plus_one(
2352 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2353 direction_t direction
,
2359 bool step_back
= false;
2363 assert(test_object
);
2368 /* This bisects the array in object 'first', but first checks
2370 r
= test_object(f
, extra
, needle
);
2374 if (r
== TEST_FOUND
)
2375 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2377 /* if we are looking with DIRECTION_UP then we need to first
2378 see if in the actual array there is a matching entry, and
2379 return the last one of that. But if there isn't any we need
2380 to return this one. Hence remember this, and return it
2383 step_back
= direction
== DIRECTION_UP
;
2385 if (r
== TEST_RIGHT
) {
2386 if (direction
== DIRECTION_DOWN
)
2392 r
= generic_array_bisect(f
, first
, n
-1, needle
, test_object
, direction
, ret
, offset
, idx
);
2394 if (r
== 0 && step_back
)
2403 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2419 _pure_
static int test_object_offset(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2425 else if (p
< needle
)
2431 static int test_object_seqnum(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2438 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2442 if (le64toh(o
->entry
.seqnum
) == needle
)
2444 else if (le64toh(o
->entry
.seqnum
) < needle
)
2450 int journal_file_move_to_entry_by_seqnum(
2453 direction_t direction
,
2459 return generic_array_bisect(f
,
2460 le64toh(f
->header
->entry_array_offset
),
2461 le64toh(f
->header
->n_entries
),
2468 static int test_object_realtime(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2475 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2479 if (le64toh(o
->entry
.realtime
) == needle
)
2481 else if (le64toh(o
->entry
.realtime
) < needle
)
2487 int journal_file_move_to_entry_by_realtime(
2490 direction_t direction
,
2496 return generic_array_bisect(f
,
2497 le64toh(f
->header
->entry_array_offset
),
2498 le64toh(f
->header
->n_entries
),
2500 test_object_realtime
,
2505 static int test_object_monotonic(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2512 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2516 if (le64toh(o
->entry
.monotonic
) == needle
)
2518 else if (le64toh(o
->entry
.monotonic
) < needle
)
2524 static int find_data_object_by_boot_id(
2530 char t
[STRLEN("_BOOT_ID=") + 32 + 1] = "_BOOT_ID=";
2532 sd_id128_to_string(boot_id
, t
+ 9);
2533 return journal_file_find_data_object(f
, t
, sizeof(t
) - 1, o
, b
);
2536 int journal_file_move_to_entry_by_monotonic(
2540 direction_t direction
,
2549 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, NULL
);
2555 return generic_array_bisect_plus_one(f
,
2556 le64toh(o
->data
.entry_offset
),
2557 le64toh(o
->data
.entry_array_offset
),
2558 le64toh(o
->data
.n_entries
),
2560 test_object_monotonic
,
2565 void journal_file_reset_location(JournalFile
*f
) {
2566 f
->location_type
= LOCATION_HEAD
;
2567 f
->current_offset
= 0;
2568 f
->current_seqnum
= 0;
2569 f
->current_realtime
= 0;
2570 f
->current_monotonic
= 0;
2571 zero(f
->current_boot_id
);
2572 f
->current_xor_hash
= 0;
2575 void journal_file_save_location(JournalFile
*f
, Object
*o
, uint64_t offset
) {
2576 f
->location_type
= LOCATION_SEEK
;
2577 f
->current_offset
= offset
;
2578 f
->current_seqnum
= le64toh(o
->entry
.seqnum
);
2579 f
->current_realtime
= le64toh(o
->entry
.realtime
);
2580 f
->current_monotonic
= le64toh(o
->entry
.monotonic
);
2581 f
->current_boot_id
= o
->entry
.boot_id
;
2582 f
->current_xor_hash
= le64toh(o
->entry
.xor_hash
);
2585 int journal_file_compare_locations(JournalFile
*af
, JournalFile
*bf
) {
2592 assert(af
->location_type
== LOCATION_SEEK
);
2593 assert(bf
->location_type
== LOCATION_SEEK
);
2595 /* If contents and timestamps match, these entries are
2596 * identical, even if the seqnum does not match */
2597 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
) &&
2598 af
->current_monotonic
== bf
->current_monotonic
&&
2599 af
->current_realtime
== bf
->current_realtime
&&
2600 af
->current_xor_hash
== bf
->current_xor_hash
)
2603 if (sd_id128_equal(af
->header
->seqnum_id
, bf
->header
->seqnum_id
)) {
2605 /* If this is from the same seqnum source, compare
2607 r
= CMP(af
->current_seqnum
, bf
->current_seqnum
);
2611 /* Wow! This is weird, different data but the same
2612 * seqnums? Something is borked, but let's make the
2613 * best of it and compare by time. */
2616 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
)) {
2618 /* If the boot id matches, compare monotonic time */
2619 r
= CMP(af
->current_monotonic
, bf
->current_monotonic
);
2624 /* Otherwise, compare UTC time */
2625 r
= CMP(af
->current_realtime
, bf
->current_realtime
);
2629 /* Finally, compare by contents */
2630 return CMP(af
->current_xor_hash
, bf
->current_xor_hash
);
2633 static int bump_array_index(uint64_t *i
, direction_t direction
, uint64_t n
) {
2635 /* Increase or decrease the specified index, in the right direction. */
2637 if (direction
== DIRECTION_DOWN
) {
2652 static bool check_properly_ordered(uint64_t new_offset
, uint64_t old_offset
, direction_t direction
) {
2654 /* Consider it an error if any of the two offsets is uninitialized */
2655 if (old_offset
== 0 || new_offset
== 0)
2658 /* If we go down, the new offset must be larger than the old one. */
2659 return direction
== DIRECTION_DOWN
?
2660 new_offset
> old_offset
:
2661 new_offset
< old_offset
;
2664 int journal_file_next_entry(
2667 direction_t direction
,
2668 Object
**ret
, uint64_t *offset
) {
2676 n
= le64toh(f
->header
->n_entries
);
2681 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2683 r
= generic_array_bisect(f
,
2684 le64toh(f
->header
->entry_array_offset
),
2685 le64toh(f
->header
->n_entries
),
2694 r
= bump_array_index(&i
, direction
, n
);
2699 /* And jump to it */
2701 r
= generic_array_get(f
,
2702 le64toh(f
->header
->entry_array_offset
),
2710 /* OK, so this entry is borked. Most likely some entry didn't get synced to disk properly, let's see if
2711 * the next one might work for us instead. */
2712 log_debug_errno(r
, "Entry item %" PRIu64
" is bad, skipping over it.", i
);
2714 r
= bump_array_index(&i
, direction
, n
);
2719 /* Ensure our array is properly ordered. */
2720 if (p
> 0 && !check_properly_ordered(ofs
, p
, direction
))
2721 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
2722 "%s: entry array not properly ordered at entry %" PRIu64
,
2731 int journal_file_next_entry_for_data(
2733 Object
*o
, uint64_t p
,
2734 uint64_t data_offset
,
2735 direction_t direction
,
2736 Object
**ret
, uint64_t *offset
) {
2743 assert(p
> 0 || !o
);
2745 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2749 n
= le64toh(d
->data
.n_entries
);
2754 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2756 if (o
->object
.type
!= OBJECT_ENTRY
)
2759 r
= generic_array_bisect_plus_one(f
,
2760 le64toh(d
->data
.entry_offset
),
2761 le64toh(d
->data
.entry_array_offset
),
2762 le64toh(d
->data
.n_entries
),
2772 r
= bump_array_index(&i
, direction
, n
);
2778 r
= generic_array_get_plus_one(f
,
2779 le64toh(d
->data
.entry_offset
),
2780 le64toh(d
->data
.entry_array_offset
),
2788 log_debug_errno(r
, "Data entry item %" PRIu64
" is bad, skipping over it.", i
);
2790 r
= bump_array_index(&i
, direction
, n
);
2795 /* Ensure our array is properly ordered. */
2796 if (p
> 0 && check_properly_ordered(ofs
, p
, direction
))
2797 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
2798 "%s data entry array not properly ordered at entry %" PRIu64
,
2807 int journal_file_move_to_entry_by_offset_for_data(
2809 uint64_t data_offset
,
2811 direction_t direction
,
2812 Object
**ret
, uint64_t *offset
) {
2819 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2823 return generic_array_bisect_plus_one(f
,
2824 le64toh(d
->data
.entry_offset
),
2825 le64toh(d
->data
.entry_array_offset
),
2826 le64toh(d
->data
.n_entries
),
2833 int journal_file_move_to_entry_by_monotonic_for_data(
2835 uint64_t data_offset
,
2838 direction_t direction
,
2839 Object
**ret
, uint64_t *offset
) {
2847 /* First, seek by time */
2848 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &b
);
2854 r
= generic_array_bisect_plus_one(f
,
2855 le64toh(o
->data
.entry_offset
),
2856 le64toh(o
->data
.entry_array_offset
),
2857 le64toh(o
->data
.n_entries
),
2859 test_object_monotonic
,
2865 /* And now, continue seeking until we find an entry that
2866 * exists in both bisection arrays */
2872 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2876 r
= generic_array_bisect_plus_one(f
,
2877 le64toh(d
->data
.entry_offset
),
2878 le64toh(d
->data
.entry_array_offset
),
2879 le64toh(d
->data
.n_entries
),
2887 r
= journal_file_move_to_object(f
, OBJECT_DATA
, b
, &o
);
2891 r
= generic_array_bisect_plus_one(f
,
2892 le64toh(o
->data
.entry_offset
),
2893 le64toh(o
->data
.entry_array_offset
),
2894 le64toh(o
->data
.n_entries
),
2916 int journal_file_move_to_entry_by_seqnum_for_data(
2918 uint64_t data_offset
,
2920 direction_t direction
,
2921 Object
**ret
, uint64_t *offset
) {
2928 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2932 return generic_array_bisect_plus_one(f
,
2933 le64toh(d
->data
.entry_offset
),
2934 le64toh(d
->data
.entry_array_offset
),
2935 le64toh(d
->data
.n_entries
),
2942 int journal_file_move_to_entry_by_realtime_for_data(
2944 uint64_t data_offset
,
2946 direction_t direction
,
2947 Object
**ret
, uint64_t *offset
) {
2954 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2958 return generic_array_bisect_plus_one(f
,
2959 le64toh(d
->data
.entry_offset
),
2960 le64toh(d
->data
.entry_array_offset
),
2961 le64toh(d
->data
.n_entries
),
2963 test_object_realtime
,
2968 void journal_file_dump(JournalFile
*f
) {
2976 journal_file_print_header(f
);
2978 p
= le64toh(f
->header
->header_size
);
2980 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &o
);
2984 switch (o
->object
.type
) {
2987 printf("Type: OBJECT_UNUSED\n");
2991 printf("Type: OBJECT_DATA\n");
2995 printf("Type: OBJECT_FIELD\n");
2999 printf("Type: OBJECT_ENTRY seqnum=%"PRIu64
" monotonic=%"PRIu64
" realtime=%"PRIu64
"\n",
3000 le64toh(o
->entry
.seqnum
),
3001 le64toh(o
->entry
.monotonic
),
3002 le64toh(o
->entry
.realtime
));
3005 case OBJECT_FIELD_HASH_TABLE
:
3006 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
3009 case OBJECT_DATA_HASH_TABLE
:
3010 printf("Type: OBJECT_DATA_HASH_TABLE\n");
3013 case OBJECT_ENTRY_ARRAY
:
3014 printf("Type: OBJECT_ENTRY_ARRAY\n");
3018 printf("Type: OBJECT_TAG seqnum=%"PRIu64
" epoch=%"PRIu64
"\n",
3019 le64toh(o
->tag
.seqnum
),
3020 le64toh(o
->tag
.epoch
));
3024 printf("Type: unknown (%i)\n", o
->object
.type
);
3028 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
)
3029 printf("Flags: %s\n",
3030 object_compressed_to_string(o
->object
.flags
& OBJECT_COMPRESSION_MASK
));
3032 if (p
== le64toh(f
->header
->tail_object_offset
))
3035 p
= p
+ ALIGN64(le64toh(o
->object
.size
));
3040 log_error("File corrupt");
3043 static const char* format_timestamp_safe(char *buf
, size_t l
, usec_t t
) {
3046 x
= format_timestamp(buf
, l
, t
);
3052 void journal_file_print_header(JournalFile
*f
) {
3053 char a
[33], b
[33], c
[33], d
[33];
3054 char x
[FORMAT_TIMESTAMP_MAX
], y
[FORMAT_TIMESTAMP_MAX
], z
[FORMAT_TIMESTAMP_MAX
];
3056 char bytes
[FORMAT_BYTES_MAX
];
3061 printf("File Path: %s\n"
3065 "Sequential Number ID: %s\n"
3067 "Compatible Flags:%s%s\n"
3068 "Incompatible Flags:%s%s%s\n"
3069 "Header size: %"PRIu64
"\n"
3070 "Arena size: %"PRIu64
"\n"
3071 "Data Hash Table Size: %"PRIu64
"\n"
3072 "Field Hash Table Size: %"PRIu64
"\n"
3073 "Rotate Suggested: %s\n"
3074 "Head Sequential Number: %"PRIu64
" (%"PRIx64
")\n"
3075 "Tail Sequential Number: %"PRIu64
" (%"PRIx64
")\n"
3076 "Head Realtime Timestamp: %s (%"PRIx64
")\n"
3077 "Tail Realtime Timestamp: %s (%"PRIx64
")\n"
3078 "Tail Monotonic Timestamp: %s (%"PRIx64
")\n"
3079 "Objects: %"PRIu64
"\n"
3080 "Entry Objects: %"PRIu64
"\n",
3082 sd_id128_to_string(f
->header
->file_id
, a
),
3083 sd_id128_to_string(f
->header
->machine_id
, b
),
3084 sd_id128_to_string(f
->header
->boot_id
, c
),
3085 sd_id128_to_string(f
->header
->seqnum_id
, d
),
3086 f
->header
->state
== STATE_OFFLINE
? "OFFLINE" :
3087 f
->header
->state
== STATE_ONLINE
? "ONLINE" :
3088 f
->header
->state
== STATE_ARCHIVED
? "ARCHIVED" : "UNKNOWN",
3089 JOURNAL_HEADER_SEALED(f
->header
) ? " SEALED" : "",
3090 (le32toh(f
->header
->compatible_flags
) & ~HEADER_COMPATIBLE_ANY
) ? " ???" : "",
3091 JOURNAL_HEADER_COMPRESSED_XZ(f
->header
) ? " COMPRESSED-XZ" : "",
3092 JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
) ? " COMPRESSED-LZ4" : "",
3093 (le32toh(f
->header
->incompatible_flags
) & ~HEADER_INCOMPATIBLE_ANY
) ? " ???" : "",
3094 le64toh(f
->header
->header_size
),
3095 le64toh(f
->header
->arena_size
),
3096 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3097 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
),
3098 yes_no(journal_file_rotate_suggested(f
, 0)),
3099 le64toh(f
->header
->head_entry_seqnum
), le64toh(f
->header
->head_entry_seqnum
),
3100 le64toh(f
->header
->tail_entry_seqnum
), le64toh(f
->header
->tail_entry_seqnum
),
3101 format_timestamp_safe(x
, sizeof(x
), le64toh(f
->header
->head_entry_realtime
)), le64toh(f
->header
->head_entry_realtime
),
3102 format_timestamp_safe(y
, sizeof(y
), le64toh(f
->header
->tail_entry_realtime
)), le64toh(f
->header
->tail_entry_realtime
),
3103 format_timespan(z
, sizeof(z
), le64toh(f
->header
->tail_entry_monotonic
), USEC_PER_MSEC
), le64toh(f
->header
->tail_entry_monotonic
),
3104 le64toh(f
->header
->n_objects
),
3105 le64toh(f
->header
->n_entries
));
3107 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3108 printf("Data Objects: %"PRIu64
"\n"
3109 "Data Hash Table Fill: %.1f%%\n",
3110 le64toh(f
->header
->n_data
),
3111 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))));
3113 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3114 printf("Field Objects: %"PRIu64
"\n"
3115 "Field Hash Table Fill: %.1f%%\n",
3116 le64toh(f
->header
->n_fields
),
3117 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))));
3119 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
))
3120 printf("Tag Objects: %"PRIu64
"\n",
3121 le64toh(f
->header
->n_tags
));
3122 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
3123 printf("Entry Array Objects: %"PRIu64
"\n",
3124 le64toh(f
->header
->n_entry_arrays
));
3126 if (fstat(f
->fd
, &st
) >= 0)
3127 printf("Disk usage: %s\n", format_bytes(bytes
, sizeof(bytes
), (uint64_t) st
.st_blocks
* 512ULL));
3130 static int journal_file_warn_btrfs(JournalFile
*f
) {
3136 /* Before we write anything, check if the COW logic is turned
3137 * off on btrfs. Given our write pattern that is quite
3138 * unfriendly to COW file systems this should greatly improve
3139 * performance on COW file systems, such as btrfs, at the
3140 * expense of data integrity features (which shouldn't be too
3141 * bad, given that we do our own checksumming). */
3143 r
= btrfs_is_filesystem(f
->fd
);
3145 return log_warning_errno(r
, "Failed to determine if journal is on btrfs: %m");
3149 r
= read_attr_fd(f
->fd
, &attrs
);
3151 return log_warning_errno(r
, "Failed to read file attributes: %m");
3153 if (attrs
& FS_NOCOW_FL
) {
3154 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3158 log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3159 "This is likely to slow down journal access substantially, please consider turning "
3160 "off the copy-on-write file attribute on the journal directory, using chattr +C.", f
->path
);
3165 int journal_file_open(
3171 uint64_t compress_threshold_bytes
,
3173 JournalMetrics
*metrics
,
3174 MMapCache
*mmap_cache
,
3175 Set
*deferred_closes
,
3176 JournalFile
*template,
3177 JournalFile
**ret
) {
3179 bool newly_created
= false;
3183 char bytes
[FORMAT_BYTES_MAX
];
3186 assert(fd
>= 0 || fname
);
3188 if (!IN_SET((flags
& O_ACCMODE
), O_RDONLY
, O_RDWR
))
3191 if (fname
&& (flags
& O_CREAT
) && !endswith(fname
, ".journal"))
3194 f
= new(JournalFile
, 1);
3198 *f
= (JournalFile
) {
3203 .prot
= prot_from_flags(flags
),
3204 .writable
= (flags
& O_ACCMODE
) != O_RDONLY
,
3207 .compress_lz4
= compress
,
3209 .compress_xz
= compress
,
3211 .compress_threshold_bytes
= compress_threshold_bytes
== (uint64_t) -1 ?
3212 DEFAULT_COMPRESS_THRESHOLD
:
3213 MAX(MIN_COMPRESS_THRESHOLD
, compress_threshold_bytes
),
3219 log_debug("Journal effective settings seal=%s compress=%s compress_threshold_bytes=%s",
3220 yes_no(f
->seal
), yes_no(JOURNAL_FILE_COMPRESS(f
)),
3221 format_bytes(bytes
, sizeof(bytes
), f
->compress_threshold_bytes
));
3224 f
->mmap
= mmap_cache_ref(mmap_cache
);
3226 f
->mmap
= mmap_cache_new();
3234 f
->path
= strdup(fname
);
3242 /* If we don't know the path, fill in something explanatory and vaguely useful */
3243 if (asprintf(&f
->path
, "/proc/self/%i", fd
) < 0) {
3249 f
->chain_cache
= ordered_hashmap_new(&uint64_hash_ops
);
3250 if (!f
->chain_cache
) {
3256 /* We pass O_NONBLOCK here, so that in case somebody pointed us to some character device node or FIFO
3257 * or so, we likely fail quickly than block for long. For regular files O_NONBLOCK has no effect, hence
3258 * it doesn't hurt in that case. */
3260 f
->fd
= open(f
->path
, f
->flags
|O_CLOEXEC
|O_NONBLOCK
, f
->mode
);
3266 /* fds we opened here by us should also be closed by us. */
3269 r
= fd_nonblock(f
->fd
, false);
3274 f
->cache_fd
= mmap_cache_add_fd(f
->mmap
, f
->fd
);
3280 r
= journal_file_fstat(f
);
3284 if (f
->last_stat
.st_size
== 0 && f
->writable
) {
3286 (void) journal_file_warn_btrfs(f
);
3288 /* Let's attach the creation time to the journal file, so that the vacuuming code knows the age of this
3289 * file even if the file might end up corrupted one day... Ideally we'd just use the creation time many
3290 * file systems maintain for each file, but the API to query this is very new, hence let's emulate this
3291 * via extended attributes. If extended attributes are not supported we'll just skip this, and rely
3292 * solely on mtime/atime/ctime of the file. */
3293 (void) fd_setcrtime(f
->fd
, 0);
3296 /* Try to load the FSPRG state, and if we can't, then
3297 * just don't do sealing */
3299 r
= journal_file_fss_load(f
);
3305 r
= journal_file_init_header(f
, template);
3309 r
= journal_file_fstat(f
);
3313 newly_created
= true;
3316 if (f
->last_stat
.st_size
< (off_t
) HEADER_SIZE_MIN
) {
3321 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 if (!newly_created
) {
3328 set_clear_with_destructor(deferred_closes
, journal_file_close
);
3330 r
= journal_file_verify_header(f
);
3336 if (!newly_created
&& f
->writable
) {
3337 r
= journal_file_fss_load(f
);
3345 journal_default_metrics(metrics
, f
->fd
);
3346 f
->metrics
= *metrics
;
3347 } else if (template)
3348 f
->metrics
= template->metrics
;
3350 r
= journal_file_refresh_header(f
);
3356 r
= journal_file_hmac_setup(f
);
3361 if (newly_created
) {
3362 r
= journal_file_setup_field_hash_table(f
);
3366 r
= journal_file_setup_data_hash_table(f
);
3371 r
= journal_file_append_first_tag(f
);
3377 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
)) {
3382 if (template && template->post_change_timer
) {
3383 r
= journal_file_enable_post_change_timer(
3385 sd_event_source_get_event(template->post_change_timer
),
3386 template->post_change_timer_period
);
3392 /* The file is opened now successfully, thus we take possession of any passed in fd. */
3399 if (f
->cache_fd
&& mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
3402 (void) journal_file_close(f
);
3407 int journal_file_archive(JournalFile
*f
) {
3408 _cleanup_free_
char *p
= NULL
;
3415 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
3416 * rotation, since we don't know the actual path, and couldn't rename the file hence. */
3417 if (path_startswith(f
->path
, "/proc/self/fd"))
3420 if (!endswith(f
->path
, ".journal"))
3423 if (asprintf(&p
, "%.*s@" SD_ID128_FORMAT_STR
"-%016"PRIx64
"-%016"PRIx64
".journal",
3424 (int) strlen(f
->path
) - 8, f
->path
,
3425 SD_ID128_FORMAT_VAL(f
->header
->seqnum_id
),
3426 le64toh(f
->header
->head_entry_seqnum
),
3427 le64toh(f
->header
->head_entry_realtime
)) < 0)
3430 /* Try to rename the file to the archived version. If the file already was deleted, we'll get ENOENT, let's
3431 * ignore that case. */
3432 if (rename(f
->path
, p
) < 0 && errno
!= ENOENT
)
3435 /* Sync the rename to disk */
3436 (void) fsync_directory_of_file(f
->fd
);
3438 /* Set as archive so offlining commits w/state=STATE_ARCHIVED. Previously we would set old_file->header->state
3439 * to STATE_ARCHIVED directly here, but journal_file_set_offline() short-circuits when state != STATE_ONLINE,
3440 * which would result in the rotated journal never getting fsync() called before closing. Now we simply queue
3441 * the archive state by setting an archive bit, leaving the state as STATE_ONLINE so proper offlining
3445 /* Currently, btrfs is not very good with out write patterns and fragments heavily. Let's defrag our journal
3446 * files when we archive them */
3447 f
->defrag_on_close
= true;
3452 JournalFile
* journal_initiate_close(
3454 Set
*deferred_closes
) {
3460 if (deferred_closes
) {
3462 r
= set_put(deferred_closes
, f
);
3464 log_debug_errno(r
, "Failed to add file to deferred close set, closing immediately.");
3466 (void) journal_file_set_offline(f
, false);
3471 return journal_file_close(f
);
3474 int journal_file_rotate(
3477 uint64_t compress_threshold_bytes
,
3479 Set
*deferred_closes
) {
3481 JournalFile
*new_file
= NULL
;
3487 r
= journal_file_archive(*f
);
3491 r
= journal_file_open(
3497 compress_threshold_bytes
,
3505 journal_initiate_close(*f
, deferred_closes
);
3511 int journal_file_dispose(int dir_fd
, const char *fname
) {
3512 _cleanup_free_
char *p
= NULL
;
3513 _cleanup_close_
int fd
= -1;
3517 /* Renames a journal file to *.journal~, i.e. to mark it as corruped or otherwise uncleanly shutdown. Note that
3518 * this is done without looking into the file or changing any of its contents. The idea is that this is called
3519 * whenever something is suspicious and we want to move the file away and make clear that it is not accessed
3520 * for writing anymore. */
3522 if (!endswith(fname
, ".journal"))
3525 if (asprintf(&p
, "%.*s@%016" PRIx64
"-%016" PRIx64
".journal~",
3526 (int) strlen(fname
) - 8, fname
,
3527 now(CLOCK_REALTIME
),
3531 if (renameat(dir_fd
, fname
, dir_fd
, p
) < 0)
3534 /* btrfs doesn't cope well with our write pattern and fragments heavily. Let's defrag all files we rotate */
3535 fd
= openat(dir_fd
, p
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
|O_NOFOLLOW
);
3537 log_debug_errno(errno
, "Failed to open file for defragmentation/FS_NOCOW_FL, ignoring: %m");
3539 (void) chattr_fd(fd
, 0, FS_NOCOW_FL
, NULL
);
3540 (void) btrfs_defrag_fd(fd
);
3546 int journal_file_open_reliably(
3551 uint64_t compress_threshold_bytes
,
3553 JournalMetrics
*metrics
,
3554 MMapCache
*mmap_cache
,
3555 Set
*deferred_closes
,
3556 JournalFile
*template,
3557 JournalFile
**ret
) {
3561 r
= journal_file_open(-1, fname
, flags
, mode
, compress
, compress_threshold_bytes
, seal
, metrics
, mmap_cache
,
3562 deferred_closes
, template, ret
);
3564 -EBADMSG
, /* Corrupted */
3565 -ENODATA
, /* Truncated */
3566 -EHOSTDOWN
, /* Other machine */
3567 -EPROTONOSUPPORT
, /* Incompatible feature */
3568 -EBUSY
, /* Unclean shutdown */
3569 -ESHUTDOWN
, /* Already archived */
3570 -EIO
, /* IO error, including SIGBUS on mmap */
3571 -EIDRM
, /* File has been deleted */
3572 -ETXTBSY
)) /* File is from the future */
3575 if ((flags
& O_ACCMODE
) == O_RDONLY
)
3578 if (!(flags
& O_CREAT
))
3581 if (!endswith(fname
, ".journal"))
3584 /* The file is corrupted. Rotate it away and try it again (but only once) */
3585 log_warning_errno(r
, "File %s corrupted or uncleanly shut down, renaming and replacing.", fname
);
3587 r
= journal_file_dispose(AT_FDCWD
, fname
);
3591 return journal_file_open(-1, fname
, flags
, mode
, compress
, compress_threshold_bytes
, seal
, metrics
, mmap_cache
,
3592 deferred_closes
, template, ret
);
3595 int journal_file_copy_entry(JournalFile
*from
, JournalFile
*to
, Object
*o
, uint64_t p
) {
3597 uint64_t q
, xor_hash
= 0;
3601 const sd_id128_t
*boot_id
;
3611 ts
.monotonic
= le64toh(o
->entry
.monotonic
);
3612 ts
.realtime
= le64toh(o
->entry
.realtime
);
3613 boot_id
= &o
->entry
.boot_id
;
3615 n
= journal_file_entry_n_items(o
);
3616 /* alloca() can't take 0, hence let's allocate at least one */
3617 items
= newa(EntryItem
, MAX(1u, n
));
3619 for (i
= 0; i
< n
; i
++) {
3626 q
= le64toh(o
->entry
.items
[i
].object_offset
);
3627 le_hash
= o
->entry
.items
[i
].hash
;
3629 r
= journal_file_move_to_object(from
, OBJECT_DATA
, q
, &o
);
3633 if (le_hash
!= o
->data
.hash
)
3636 l
= le64toh(o
->object
.size
) - offsetof(Object
, data
.payload
);
3639 /* We hit the limit on 32bit machines */
3640 if ((uint64_t) t
!= l
)
3643 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
3644 #if HAVE_XZ || HAVE_LZ4
3647 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
3648 o
->data
.payload
, l
, &from
->compress_buffer
, &from
->compress_buffer_size
, &rsize
, 0);
3652 data
= from
->compress_buffer
;
3655 return -EPROTONOSUPPORT
;
3658 data
= o
->data
.payload
;
3660 r
= journal_file_append_data(to
, data
, l
, &u
, &h
);
3664 xor_hash
^= le64toh(u
->data
.hash
);
3665 items
[i
].object_offset
= htole64(h
);
3666 items
[i
].hash
= u
->data
.hash
;
3668 r
= journal_file_move_to_object(from
, OBJECT_ENTRY
, p
, &o
);
3673 r
= journal_file_append_entry_internal(to
, &ts
, boot_id
, xor_hash
, items
, n
,
3676 if (mmap_cache_got_sigbus(to
->mmap
, to
->cache_fd
))
3682 void journal_reset_metrics(JournalMetrics
*m
) {
3685 /* Set everything to "pick automatic values". */
3687 *m
= (JournalMetrics
) {
3688 .min_use
= (uint64_t) -1,
3689 .max_use
= (uint64_t) -1,
3690 .min_size
= (uint64_t) -1,
3691 .max_size
= (uint64_t) -1,
3692 .keep_free
= (uint64_t) -1,
3693 .n_max_files
= (uint64_t) -1,
3697 void journal_default_metrics(JournalMetrics
*m
, int fd
) {
3698 char a
[FORMAT_BYTES_MAX
], b
[FORMAT_BYTES_MAX
], c
[FORMAT_BYTES_MAX
], d
[FORMAT_BYTES_MAX
], e
[FORMAT_BYTES_MAX
];
3705 if (fstatvfs(fd
, &ss
) >= 0)
3706 fs_size
= ss
.f_frsize
* ss
.f_blocks
;
3708 log_debug_errno(errno
, "Failed to determine disk size: %m");
3712 if (m
->max_use
== (uint64_t) -1) {
3715 m
->max_use
= PAGE_ALIGN(fs_size
/ 10); /* 10% of file system size */
3717 if (m
->max_use
> DEFAULT_MAX_USE_UPPER
)
3718 m
->max_use
= DEFAULT_MAX_USE_UPPER
;
3720 if (m
->max_use
< DEFAULT_MAX_USE_LOWER
)
3721 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3723 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3725 m
->max_use
= PAGE_ALIGN(m
->max_use
);
3727 if (m
->max_use
!= 0 && m
->max_use
< JOURNAL_FILE_SIZE_MIN
*2)
3728 m
->max_use
= JOURNAL_FILE_SIZE_MIN
*2;
3731 if (m
->min_use
== (uint64_t) -1)
3732 m
->min_use
= DEFAULT_MIN_USE
;
3734 if (m
->min_use
> m
->max_use
)
3735 m
->min_use
= m
->max_use
;
3737 if (m
->max_size
== (uint64_t) -1) {
3738 m
->max_size
= PAGE_ALIGN(m
->max_use
/ 8); /* 8 chunks */
3740 if (m
->max_size
> DEFAULT_MAX_SIZE_UPPER
)
3741 m
->max_size
= DEFAULT_MAX_SIZE_UPPER
;
3743 m
->max_size
= PAGE_ALIGN(m
->max_size
);
3745 if (m
->max_size
!= 0) {
3746 if (m
->max_size
< JOURNAL_FILE_SIZE_MIN
)
3747 m
->max_size
= JOURNAL_FILE_SIZE_MIN
;
3749 if (m
->max_use
!= 0 && m
->max_size
*2 > m
->max_use
)
3750 m
->max_use
= m
->max_size
*2;
3753 if (m
->min_size
== (uint64_t) -1)
3754 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3756 m
->min_size
= PAGE_ALIGN(m
->min_size
);
3758 if (m
->min_size
< JOURNAL_FILE_SIZE_MIN
)
3759 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3761 if (m
->max_size
!= 0 && m
->min_size
> m
->max_size
)
3762 m
->max_size
= m
->min_size
;
3765 if (m
->keep_free
== (uint64_t) -1) {
3768 m
->keep_free
= PAGE_ALIGN(fs_size
* 3 / 20); /* 15% of file system size */
3770 if (m
->keep_free
> DEFAULT_KEEP_FREE_UPPER
)
3771 m
->keep_free
= DEFAULT_KEEP_FREE_UPPER
;
3774 m
->keep_free
= DEFAULT_KEEP_FREE
;
3777 if (m
->n_max_files
== (uint64_t) -1)
3778 m
->n_max_files
= DEFAULT_N_MAX_FILES
;
3780 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64
,
3781 format_bytes(a
, sizeof(a
), m
->min_use
),
3782 format_bytes(b
, sizeof(b
), m
->max_use
),
3783 format_bytes(c
, sizeof(c
), m
->max_size
),
3784 format_bytes(d
, sizeof(d
), m
->min_size
),
3785 format_bytes(e
, sizeof(e
), m
->keep_free
),
3789 int journal_file_get_cutoff_realtime_usec(JournalFile
*f
, usec_t
*from
, usec_t
*to
) {
3795 if (f
->header
->head_entry_realtime
== 0)
3798 *from
= le64toh(f
->header
->head_entry_realtime
);
3802 if (f
->header
->tail_entry_realtime
== 0)
3805 *to
= le64toh(f
->header
->tail_entry_realtime
);
3811 int journal_file_get_cutoff_monotonic_usec(JournalFile
*f
, sd_id128_t boot_id
, usec_t
*from
, usec_t
*to
) {
3819 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &p
);
3823 if (le64toh(o
->data
.n_entries
) <= 0)
3827 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, le64toh(o
->data
.entry_offset
), &o
);
3831 *from
= le64toh(o
->entry
.monotonic
);
3835 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
3839 r
= generic_array_get_plus_one(f
,
3840 le64toh(o
->data
.entry_offset
),
3841 le64toh(o
->data
.entry_array_offset
),
3842 le64toh(o
->data
.n_entries
)-1,
3847 *to
= le64toh(o
->entry
.monotonic
);
3853 bool journal_file_rotate_suggested(JournalFile
*f
, usec_t max_file_usec
) {
3857 /* If we gained new header fields we gained new features,
3858 * hence suggest a rotation */
3859 if (le64toh(f
->header
->header_size
) < sizeof(Header
)) {
3860 log_debug("%s uses an outdated header, suggesting rotation.", f
->path
);
3864 /* Let's check if the hash tables grew over a certain fill
3865 * level (75%, borrowing this value from Java's hash table
3866 * implementation), and if so suggest a rotation. To calculate
3867 * the fill level we need the n_data field, which only exists
3868 * in newer versions. */
3870 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3871 if (le64toh(f
->header
->n_data
) * 4ULL > (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3872 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.",
3874 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))),
3875 le64toh(f
->header
->n_data
),
3876 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3877 (unsigned long long) f
->last_stat
.st_size
,
3878 f
->last_stat
.st_size
/ le64toh(f
->header
->n_data
));
3882 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3883 if (le64toh(f
->header
->n_fields
) * 4ULL > (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3884 log_debug("Field hash table of %s has a fill level at %.1f (%"PRIu64
" of %"PRIu64
" items), suggesting rotation.",
3886 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))),
3887 le64toh(f
->header
->n_fields
),
3888 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
));
3892 /* Are the data objects properly indexed by field objects? */
3893 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
3894 JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
3895 le64toh(f
->header
->n_data
) > 0 &&
3896 le64toh(f
->header
->n_fields
) == 0)
3899 if (max_file_usec
> 0) {
3902 h
= le64toh(f
->header
->head_entry_realtime
);
3903 t
= now(CLOCK_REALTIME
);
3905 if (h
> 0 && t
> h
+ max_file_usec
)