1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
6 #include <linux/magic.h>
10 #include <sys/statvfs.h>
16 #include "alloc-util.h"
17 #include "chattr-util.h"
21 #include "format-util.h"
23 #include "id128-util.h"
24 #include "journal-authenticate.h"
25 #include "journal-def.h"
26 #include "journal-file.h"
27 #include "journal-internal.h"
29 #include "memory-util.h"
30 #include "missing_threads.h"
31 #include "path-util.h"
33 #include "random-util.h"
35 #include "sort-util.h"
36 #include "stat-util.h"
37 #include "string-table.h"
38 #include "string-util.h"
40 #include "sync-util.h"
41 #include "user-util.h"
42 #include "xattr-util.h"
44 #define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
45 #define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
47 #define DEFAULT_COMPRESS_THRESHOLD (512ULL)
48 #define MIN_COMPRESS_THRESHOLD (8ULL)
50 #define U64_KB UINT64_C(1024)
51 #define U64_MB (UINT64_C(1024) * U64_KB)
52 #define U64_GB (UINT64_C(1024) * U64_MB)
54 /* This is the minimum journal file size */
55 #define JOURNAL_FILE_SIZE_MIN (512 * U64_KB) /* 512 KiB */
56 #define JOURNAL_COMPACT_SIZE_MAX ((uint64_t) UINT32_MAX) /* 4 GiB */
58 /* These are the lower and upper bounds if we deduce the max_use value from the file system size */
59 #define MAX_USE_LOWER (1 * U64_MB) /* 1 MiB */
60 #define MAX_USE_UPPER (4 * U64_GB) /* 4 GiB */
62 /* Those are the lower and upper bounds for the minimal use limit,
63 * i.e. how much we'll use even if keep_free suggests otherwise. */
64 #define MIN_USE_LOW (1 * U64_MB) /* 1 MiB */
65 #define MIN_USE_HIGH (16 * U64_MB) /* 16 MiB */
67 /* This is the upper bound if we deduce max_size from max_use */
68 #define MAX_SIZE_UPPER (128 * U64_MB) /* 128 MiB */
70 /* This is the upper bound if we deduce the keep_free value from the file system size */
71 #define KEEP_FREE_UPPER (4 * U64_GB) /* 4 GiB */
73 /* This is the keep_free value when we can't determine the system size */
74 #define DEFAULT_KEEP_FREE (1 * U64_MB) /* 1 MB */
76 /* This is the default maximum number of journal files to keep around. */
77 #define DEFAULT_N_MAX_FILES 100
79 /* n_data was the first entry we added after the initial file format design */
80 #define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
82 /* How many entries to keep in the entry array chain cache at max */
83 #define CHAIN_CACHE_MAX 20
85 /* How much to increase the journal file size at once each time we allocate something new. */
86 #define FILE_SIZE_INCREASE (8 * U64_MB) /* 8MB */
88 /* Reread fstat() of the file for detecting deletions at least this often */
89 #define LAST_STAT_REFRESH_USEC (5*USEC_PER_SEC)
91 /* Longest hash chain to rotate after */
92 #define HASH_CHAIN_DEPTH_MAX 100
95 # pragma GCC diagnostic ignored "-Waddress-of-packed-member"
98 static int mmap_prot_from_open_flags(int flags
) {
99 switch (flags
& O_ACCMODE
) {
105 return PROT_READ
|PROT_WRITE
;
107 assert_not_reached();
111 int journal_file_tail_end_by_pread(JournalFile
*f
, uint64_t *ret_offset
) {
119 /* Same as journal_file_tail_end_by_mmap() below, but operates with pread() to avoid the mmap cache
120 * (and thus is thread safe) */
122 p
= le64toh(f
->header
->tail_object_offset
);
124 p
= le64toh(f
->header
->header_size
);
129 r
= journal_file_read_object_header(f
, OBJECT_UNUSED
, p
, &tail
);
133 sz
= le64toh(tail
.object
.size
);
134 if (sz
> UINT64_MAX
- sizeof(uint64_t) + 1)
138 if (p
> UINT64_MAX
- sz
)
149 int journal_file_tail_end_by_mmap(JournalFile
*f
, uint64_t *ret_offset
) {
157 /* Same as journal_file_tail_end_by_pread() above, but operates with the usual mmap logic */
159 p
= le64toh(f
->header
->tail_object_offset
);
161 p
= le64toh(f
->header
->header_size
);
166 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &tail
);
170 sz
= le64toh(READ_NOW(tail
->object
.size
));
171 if (sz
> UINT64_MAX
- sizeof(uint64_t) + 1)
175 if (p
> UINT64_MAX
- sz
)
186 int journal_file_set_offline_thread_join(JournalFile
*f
) {
191 if (f
->offline_state
== OFFLINE_JOINED
)
194 r
= pthread_join(f
->offline_thread
, NULL
);
198 f
->offline_state
= OFFLINE_JOINED
;
200 if (mmap_cache_fd_got_sigbus(f
->cache_fd
))
206 static int journal_file_set_online(JournalFile
*f
) {
211 if (!journal_file_writable(f
))
214 if (f
->fd
< 0 || !f
->header
)
218 switch (f
->offline_state
) {
220 /* No offline thread, no need to wait. */
224 case OFFLINE_SYNCING
: {
225 OfflineState tmp_state
= OFFLINE_SYNCING
;
226 if (!__atomic_compare_exchange_n(&f
->offline_state
, &tmp_state
, OFFLINE_CANCEL
,
227 false, __ATOMIC_SEQ_CST
, __ATOMIC_SEQ_CST
))
230 /* Canceled syncing prior to offlining, no need to wait. */
234 case OFFLINE_AGAIN_FROM_SYNCING
: {
235 OfflineState tmp_state
= OFFLINE_AGAIN_FROM_SYNCING
;
236 if (!__atomic_compare_exchange_n(&f
->offline_state
, &tmp_state
, OFFLINE_CANCEL
,
237 false, __ATOMIC_SEQ_CST
, __ATOMIC_SEQ_CST
))
240 /* Canceled restart from syncing, no need to wait. */
244 case OFFLINE_AGAIN_FROM_OFFLINING
: {
245 OfflineState tmp_state
= OFFLINE_AGAIN_FROM_OFFLINING
;
246 if (!__atomic_compare_exchange_n(&f
->offline_state
, &tmp_state
, OFFLINE_CANCEL
,
247 false, __ATOMIC_SEQ_CST
, __ATOMIC_SEQ_CST
))
250 /* Canceled restart from offlining, must wait for offlining to complete however. */
255 r
= journal_file_set_offline_thread_join(f
);
265 if (mmap_cache_fd_got_sigbus(f
->cache_fd
))
268 switch (f
->header
->state
) {
273 f
->header
->state
= STATE_ONLINE
;
282 JournalFile
* journal_file_close(JournalFile
*f
) {
286 assert(f
->newest_boot_id_prioq_idx
== PRIOQ_IDX_NULL
);
289 mmap_cache_fd_free(f
->cache_fd
);
295 ordered_hashmap_free_free(f
->chain_cache
);
298 free(f
->compress_buffer
);
303 size_t sz
= PAGE_ALIGN(f
->fss_file_size
);
304 assert(sz
< SIZE_MAX
);
305 munmap(f
->fss_file
, sz
);
307 free(f
->fsprg_state
);
312 gcry_md_close(f
->hmac
);
318 static bool keyed_hash_requested(void) {
319 static thread_local
int cached
= -1;
323 r
= getenv_bool("SYSTEMD_JOURNAL_KEYED_HASH");
326 log_debug_errno(r
, "Failed to parse $SYSTEMD_JOURNAL_KEYED_HASH environment variable, ignoring: %m");
335 static bool compact_mode_requested(void) {
336 static thread_local
int cached
= -1;
340 r
= getenv_bool("SYSTEMD_JOURNAL_COMPACT");
343 log_debug_errno(r
, "Failed to parse $SYSTEMD_JOURNAL_COMPACT environment variable, ignoring: %m");
353 static Compression
getenv_compression(void) {
358 e
= getenv("SYSTEMD_JOURNAL_COMPRESS");
360 return DEFAULT_COMPRESSION
;
362 r
= parse_boolean(e
);
364 return r
? DEFAULT_COMPRESSION
: COMPRESSION_NONE
;
366 c
= compression_from_string(e
);
368 log_debug_errno(c
, "Failed to parse SYSTEMD_JOURNAL_COMPRESS value, ignoring: %s", e
);
369 return DEFAULT_COMPRESSION
;
372 if (!compression_supported(c
)) {
373 log_debug("Unsupported compression algorithm specified, ignoring: %s", e
);
374 return DEFAULT_COMPRESSION
;
381 static Compression
compression_requested(void) {
383 static thread_local Compression cached
= _COMPRESSION_INVALID
;
386 cached
= getenv_compression();
390 return COMPRESSION_NONE
;
394 static int journal_file_init_header(
396 JournalFileFlags file_flags
,
397 JournalFile
*template) {
406 /* Try to load the FSPRG state, and if we can't, then just don't do sealing */
407 seal
= FLAGS_SET(file_flags
, JOURNAL_SEAL
) && journal_file_fss_load(f
) >= 0;
411 .header_size
= htole64(ALIGN64(sizeof(h
))),
412 .incompatible_flags
= htole32(
413 FLAGS_SET(file_flags
, JOURNAL_COMPRESS
) * COMPRESSION_TO_HEADER_INCOMPATIBLE_FLAG(compression_requested()) |
414 keyed_hash_requested() * HEADER_INCOMPATIBLE_KEYED_HASH
|
415 compact_mode_requested() * HEADER_INCOMPATIBLE_COMPACT
),
416 .compatible_flags
= htole32(
417 (seal
* HEADER_COMPATIBLE_SEALED
) |
418 HEADER_COMPATIBLE_TAIL_ENTRY_BOOT_ID
),
421 assert_cc(sizeof(h
.signature
) == sizeof(HEADER_SIGNATURE
));
422 memcpy(h
.signature
, HEADER_SIGNATURE
, sizeof(HEADER_SIGNATURE
));
424 r
= sd_id128_randomize(&h
.file_id
);
428 r
= sd_id128_get_machine(&h
.machine_id
);
429 if (r
< 0 && !ERRNO_IS_MACHINE_ID_UNSET(r
))
430 return r
; /* If we have no valid machine ID (test environment?), let's simply leave the
431 * machine ID field all zeroes. */
434 h
.seqnum_id
= template->header
->seqnum_id
;
435 h
.tail_entry_seqnum
= template->header
->tail_entry_seqnum
;
437 h
.seqnum_id
= h
.file_id
;
439 k
= pwrite(f
->fd
, &h
, sizeof(h
), 0);
448 static int journal_file_refresh_header(JournalFile
*f
) {
454 /* We used to update the header's boot ID field here, but we don't do that anymore, as per
455 * HEADER_COMPATIBLE_TAIL_ENTRY_BOOT_ID */
457 r
= journal_file_set_online(f
);
459 /* Sync the online state to disk; likely just created a new file, also sync the directory this file
461 (void) fsync_full(f
->fd
);
466 static bool warn_wrong_flags(const JournalFile
*f
, bool compatible
) {
467 const uint32_t any
= compatible
? HEADER_COMPATIBLE_ANY
: HEADER_INCOMPATIBLE_ANY
,
468 supported
= compatible
? HEADER_COMPATIBLE_SUPPORTED
: HEADER_INCOMPATIBLE_SUPPORTED
;
469 const char *type
= compatible
? "compatible" : "incompatible";
475 flags
= le32toh(compatible
? f
->header
->compatible_flags
: f
->header
->incompatible_flags
);
477 if (flags
& ~supported
) {
479 log_debug("Journal file %s has unknown %s flags 0x%"PRIx32
,
480 f
->path
, type
, flags
& ~any
);
481 flags
= (flags
& any
) & ~supported
;
485 _cleanup_free_
char *t
= NULL
;
488 if (flags
& HEADER_COMPATIBLE_SEALED
)
489 strv
[n
++] = "sealed";
491 if (flags
& HEADER_INCOMPATIBLE_COMPRESSED_XZ
)
492 strv
[n
++] = "xz-compressed";
493 if (flags
& HEADER_INCOMPATIBLE_COMPRESSED_LZ4
)
494 strv
[n
++] = "lz4-compressed";
495 if (flags
& HEADER_INCOMPATIBLE_COMPRESSED_ZSTD
)
496 strv
[n
++] = "zstd-compressed";
497 if (flags
& HEADER_INCOMPATIBLE_KEYED_HASH
)
498 strv
[n
++] = "keyed-hash";
499 if (flags
& HEADER_INCOMPATIBLE_COMPACT
)
500 strv
[n
++] = "compact";
503 assert(n
< ELEMENTSOF(strv
));
505 t
= strv_join((char**) strv
, ", ");
506 log_debug("Journal file %s uses %s %s %s disabled at compilation time.",
507 f
->path
, type
, n
> 1 ? "flags" : "flag", strnull(t
));
515 static bool offset_is_valid(uint64_t offset
, uint64_t header_size
, uint64_t tail_object_offset
) {
518 if (!VALID64(offset
))
520 if (offset
< header_size
)
522 if (offset
> tail_object_offset
)
527 static bool hash_table_is_valid(uint64_t offset
, uint64_t size
, uint64_t header_size
, uint64_t arena_size
, uint64_t tail_object_offset
) {
528 if ((offset
== 0) != (size
== 0))
532 if (offset
<= offsetof(Object
, hash_table
.items
))
534 offset
-= offsetof(Object
, hash_table
.items
);
535 if (!offset_is_valid(offset
, header_size
, tail_object_offset
))
537 assert(offset
<= header_size
+ arena_size
);
538 if (size
> header_size
+ arena_size
- offset
)
543 static int journal_file_verify_header(JournalFile
*f
) {
544 uint64_t arena_size
, header_size
;
549 if (memcmp(f
->header
->signature
, HEADER_SIGNATURE
, 8))
552 /* In both read and write mode we refuse to open files with incompatible
553 * flags we don't know. */
554 if (warn_wrong_flags(f
, false))
555 return -EPROTONOSUPPORT
;
557 /* When open for writing we refuse to open files with compatible flags, too. */
558 if (journal_file_writable(f
) && warn_wrong_flags(f
, true))
559 return -EPROTONOSUPPORT
;
561 if (f
->header
->state
>= _STATE_MAX
)
564 header_size
= le64toh(READ_NOW(f
->header
->header_size
));
566 /* The first addition was n_data, so check that we are at least this large */
567 if (header_size
< HEADER_SIZE_MIN
)
570 /* When open for writing we refuse to open files with a mismatch of the header size, i.e. writing to
571 * files implementing older or new header structures. */
572 if (journal_file_writable(f
) && header_size
!= sizeof(Header
))
573 return -EPROTONOSUPPORT
;
575 /* Don't write to journal files without the new boot ID update behavior guarantee. */
576 if (journal_file_writable(f
) && !JOURNAL_HEADER_TAIL_ENTRY_BOOT_ID(f
->header
))
577 return -EPROTONOSUPPORT
;
579 if (JOURNAL_HEADER_SEALED(f
->header
) && !JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
582 arena_size
= le64toh(READ_NOW(f
->header
->arena_size
));
584 if (UINT64_MAX
- header_size
< arena_size
|| header_size
+ arena_size
> (uint64_t) f
->last_stat
.st_size
)
587 uint64_t tail_object_offset
= le64toh(f
->header
->tail_object_offset
);
588 if (!offset_is_valid(tail_object_offset
, header_size
, UINT64_MAX
))
590 if (header_size
+ arena_size
< tail_object_offset
)
592 if (header_size
+ arena_size
- tail_object_offset
< sizeof(ObjectHeader
))
595 if (!hash_table_is_valid(le64toh(f
->header
->data_hash_table_offset
),
596 le64toh(f
->header
->data_hash_table_size
),
597 header_size
, arena_size
, tail_object_offset
))
600 if (!hash_table_is_valid(le64toh(f
->header
->field_hash_table_offset
),
601 le64toh(f
->header
->field_hash_table_size
),
602 header_size
, arena_size
, tail_object_offset
))
605 uint64_t entry_array_offset
= le64toh(f
->header
->entry_array_offset
);
606 if (!offset_is_valid(entry_array_offset
, header_size
, tail_object_offset
))
609 if (JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_array_offset
)) {
610 uint32_t offset
= le32toh(f
->header
->tail_entry_array_offset
);
611 uint32_t n
= le32toh(f
->header
->tail_entry_array_n_entries
);
613 if (!offset_is_valid(offset
, header_size
, tail_object_offset
))
615 if (entry_array_offset
> offset
)
617 if (entry_array_offset
== 0 && offset
!= 0)
619 if ((offset
== 0) != (n
== 0))
621 assert(offset
<= header_size
+ arena_size
);
622 if ((uint64_t) n
* journal_file_entry_array_item_size(f
) > header_size
+ arena_size
- offset
)
626 if (JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_offset
)) {
627 uint64_t offset
= le64toh(f
->header
->tail_entry_offset
);
629 if (!offset_is_valid(offset
, header_size
, tail_object_offset
))
633 /* When there is an entry object, then these fields must be filled. */
634 if (sd_id128_is_null(f
->header
->tail_entry_boot_id
))
636 if (!VALID_REALTIME(le64toh(f
->header
->head_entry_realtime
)))
638 if (!VALID_REALTIME(le64toh(f
->header
->tail_entry_realtime
)))
640 if (!VALID_MONOTONIC(le64toh(f
->header
->tail_entry_realtime
)))
643 /* Otherwise, the fields must be zero. */
644 if (JOURNAL_HEADER_TAIL_ENTRY_BOOT_ID(f
->header
) &&
645 !sd_id128_is_null(f
->header
->tail_entry_boot_id
))
647 if (f
->header
->head_entry_realtime
!= 0)
649 if (f
->header
->tail_entry_realtime
!= 0)
651 if (f
->header
->tail_entry_realtime
!= 0)
656 /* Verify number of objects */
657 uint64_t n_objects
= le64toh(f
->header
->n_objects
);
658 if (n_objects
> arena_size
/ sizeof(ObjectHeader
))
661 uint64_t n_entries
= le64toh(f
->header
->n_entries
);
662 if (n_entries
> n_objects
)
665 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
666 le64toh(f
->header
->n_data
) > n_objects
)
669 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
670 le64toh(f
->header
->n_fields
) > n_objects
)
673 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
) &&
674 le64toh(f
->header
->n_tags
) > n_objects
)
677 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
) &&
678 le64toh(f
->header
->n_entry_arrays
) > n_objects
)
681 if (JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_array_n_entries
) &&
682 le32toh(f
->header
->tail_entry_array_n_entries
) > n_entries
)
685 if (journal_file_writable(f
)) {
686 sd_id128_t machine_id
;
690 r
= sd_id128_get_machine(&machine_id
);
691 if (ERRNO_IS_NEG_MACHINE_ID_UNSET(r
)) /* Gracefully handle the machine ID not being initialized yet */
692 machine_id
= SD_ID128_NULL
;
696 if (!sd_id128_equal(machine_id
, f
->header
->machine_id
))
697 return log_debug_errno(SYNTHETIC_ERRNO(EHOSTDOWN
),
698 "Trying to open journal file from different host for writing, refusing.");
700 state
= f
->header
->state
;
702 if (state
== STATE_ARCHIVED
)
703 return -ESHUTDOWN
; /* Already archived */
704 if (state
== STATE_ONLINE
)
705 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
),
706 "Journal file %s is already online. Assuming unclean closing.",
708 if (state
!= STATE_OFFLINE
)
709 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
),
710 "Journal file %s has unknown state %i.",
713 if (f
->header
->field_hash_table_size
== 0 || f
->header
->data_hash_table_size
== 0)
720 int journal_file_fstat(JournalFile
*f
) {
726 if (fstat(f
->fd
, &f
->last_stat
) < 0)
729 f
->last_stat_usec
= now(CLOCK_MONOTONIC
);
731 /* Refuse dealing with files that aren't regular */
732 r
= stat_verify_regular(&f
->last_stat
);
736 /* Refuse appending to files that are already deleted */
737 if (f
->last_stat
.st_nlink
<= 0)
743 static int journal_file_allocate(JournalFile
*f
, uint64_t offset
, uint64_t size
) {
744 uint64_t old_size
, new_size
, old_header_size
, old_arena_size
;
750 /* We assume that this file is not sparse, and we know that for sure, since we always call
751 * posix_fallocate() ourselves */
753 if (size
> PAGE_ALIGN_DOWN_U64(UINT64_MAX
) - offset
)
756 if (mmap_cache_fd_got_sigbus(f
->cache_fd
))
759 old_header_size
= le64toh(READ_NOW(f
->header
->header_size
));
760 old_arena_size
= le64toh(READ_NOW(f
->header
->arena_size
));
761 if (old_arena_size
> PAGE_ALIGN_DOWN_U64(UINT64_MAX
) - old_header_size
)
764 old_size
= old_header_size
+ old_arena_size
;
766 new_size
= MAX(PAGE_ALIGN_U64(offset
+ size
), old_header_size
);
768 if (new_size
<= old_size
) {
770 /* We already pre-allocated enough space, but before
771 * we write to it, let's check with fstat() if the
772 * file got deleted, in order make sure we don't throw
773 * away the data immediately. Don't check fstat() for
774 * all writes though, but only once ever 10s. */
776 if (f
->last_stat_usec
+ LAST_STAT_REFRESH_USEC
> now(CLOCK_MONOTONIC
))
779 return journal_file_fstat(f
);
782 /* Allocate more space. */
784 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
787 /* Refuse to go over 4G in compact mode so offsets can be stored in 32-bit. */
788 if (JOURNAL_HEADER_COMPACT(f
->header
) && new_size
> UINT32_MAX
)
791 if (new_size
> f
->metrics
.min_size
&& f
->metrics
.keep_free
> 0) {
794 if (fstatvfs(f
->fd
, &svfs
) >= 0) {
797 available
= LESS_BY(u64_multiply_safe(svfs
.f_bfree
, svfs
.f_bsize
), f
->metrics
.keep_free
);
799 if (new_size
- old_size
> available
)
804 /* Increase by larger blocks at once */
805 new_size
= ROUND_UP(new_size
, FILE_SIZE_INCREASE
);
806 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
807 new_size
= f
->metrics
.max_size
;
809 /* Note that the glibc fallocate() fallback is very
810 inefficient, hence we try to minimize the allocation area
812 r
= posix_fallocate_loop(f
->fd
, old_size
, new_size
- old_size
);
816 f
->header
->arena_size
= htole64(new_size
- old_header_size
);
818 return journal_file_fstat(f
);
821 static int journal_file_move_to(
834 /* This function may clear, overwrite, or alter previously cached entries with the same type. After
835 * this function has been called, all previously read objects with the same type may be invalidated,
836 * hence must be re-read before use. */
841 if (size
> UINT64_MAX
- offset
)
844 /* Avoid SIGBUS on invalid accesses */
845 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
) {
846 /* Hmm, out of range? Let's refresh the fstat() data
847 * first, before we trust that check. */
849 r
= journal_file_fstat(f
);
853 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
)
854 return -EADDRNOTAVAIL
;
857 return mmap_cache_fd_get(f
->cache_fd
, type_to_category(type
), keep_always
, offset
, size
, &f
->last_stat
, ret
);
860 static uint64_t minimum_header_size(JournalFile
*f
, Object
*o
) {
862 static const uint64_t table
[] = {
863 [OBJECT_DATA
] = sizeof(DataObject
),
864 [OBJECT_FIELD
] = sizeof(FieldObject
),
865 [OBJECT_ENTRY
] = sizeof(EntryObject
),
866 [OBJECT_DATA_HASH_TABLE
] = sizeof(HashTableObject
),
867 [OBJECT_FIELD_HASH_TABLE
] = sizeof(HashTableObject
),
868 [OBJECT_ENTRY_ARRAY
] = sizeof(EntryArrayObject
),
869 [OBJECT_TAG
] = sizeof(TagObject
),
875 if (o
->object
.type
== OBJECT_DATA
)
876 return journal_file_data_payload_offset(f
);
878 if (o
->object
.type
>= ELEMENTSOF(table
) || table
[o
->object
.type
] <= 0)
879 return sizeof(ObjectHeader
);
881 return table
[o
->object
.type
];
884 static int check_object_header(JournalFile
*f
, Object
*o
, ObjectType type
, uint64_t offset
) {
890 s
= le64toh(READ_NOW(o
->object
.size
));
892 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
893 "Attempt to move to uninitialized object: %" PRIu64
,
896 if (s
< sizeof(ObjectHeader
))
897 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
898 "Attempt to move to overly short object with size %"PRIu64
": %" PRIu64
,
901 if (o
->object
.type
<= OBJECT_UNUSED
|| o
->object
.type
>= _OBJECT_TYPE_MAX
)
902 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
903 "Attempt to move to object with invalid type (%u): %" PRIu64
,
904 o
->object
.type
, offset
);
906 if (type
> OBJECT_UNUSED
&& o
->object
.type
!= type
)
907 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
908 "Found %s object while expecting %s object: %" PRIu64
,
909 journal_object_type_to_string(o
->object
.type
),
910 journal_object_type_to_string(type
),
913 if (s
< minimum_header_size(f
, o
))
914 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
915 "Size of %s object (%"PRIu64
") is smaller than the minimum object size (%"PRIu64
"): %" PRIu64
,
916 journal_object_type_to_string(o
->object
.type
),
918 minimum_header_size(f
, o
),
924 /* Lightweight object checks. We want this to be fast, so that we won't
925 * slowdown every journal_file_move_to_object() call too much. */
926 static int check_object(JournalFile
*f
, Object
*o
, uint64_t offset
) {
930 switch (o
->object
.type
) {
933 if ((le64toh(o
->data
.entry_offset
) == 0) ^ (le64toh(o
->data
.n_entries
) == 0))
934 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
935 "Bad data n_entries: %" PRIu64
": %" PRIu64
,
936 le64toh(o
->data
.n_entries
),
939 if (le64toh(o
->object
.size
) <= journal_file_data_payload_offset(f
))
940 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
941 "Bad data size (<= %zu): %" PRIu64
": %" PRIu64
,
942 journal_file_data_payload_offset(f
),
943 le64toh(o
->object
.size
),
946 if (!VALID64(le64toh(o
->data
.next_hash_offset
)) ||
947 !VALID64(le64toh(o
->data
.next_field_offset
)) ||
948 !VALID64(le64toh(o
->data
.entry_offset
)) ||
949 !VALID64(le64toh(o
->data
.entry_array_offset
)))
950 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
951 "Invalid offset, next_hash_offset=" OFSfmt
", next_field_offset=" OFSfmt
", entry_offset=" OFSfmt
", entry_array_offset=" OFSfmt
": %" PRIu64
,
952 le64toh(o
->data
.next_hash_offset
),
953 le64toh(o
->data
.next_field_offset
),
954 le64toh(o
->data
.entry_offset
),
955 le64toh(o
->data
.entry_array_offset
),
961 if (le64toh(o
->object
.size
) <= offsetof(Object
, field
.payload
))
962 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
963 "Bad field size (<= %zu): %" PRIu64
": %" PRIu64
,
964 offsetof(Object
, field
.payload
),
965 le64toh(o
->object
.size
),
968 if (!VALID64(le64toh(o
->field
.next_hash_offset
)) ||
969 !VALID64(le64toh(o
->field
.head_data_offset
)))
970 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
971 "Invalid offset, next_hash_offset=" OFSfmt
", head_data_offset=" OFSfmt
": %" PRIu64
,
972 le64toh(o
->field
.next_hash_offset
),
973 le64toh(o
->field
.head_data_offset
),
980 sz
= le64toh(READ_NOW(o
->object
.size
));
981 if (sz
< offsetof(Object
, entry
.items
) ||
982 (sz
- offsetof(Object
, entry
.items
)) % journal_file_entry_item_size(f
) != 0)
983 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
984 "Bad entry size (<= %zu): %" PRIu64
": %" PRIu64
,
985 offsetof(Object
, entry
.items
),
989 if ((sz
- offsetof(Object
, entry
.items
)) / journal_file_entry_item_size(f
) <= 0)
990 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
991 "Invalid number items in entry: %" PRIu64
": %" PRIu64
,
992 (sz
- offsetof(Object
, entry
.items
)) / journal_file_entry_item_size(f
),
995 if (le64toh(o
->entry
.seqnum
) <= 0)
996 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
997 "Invalid entry seqnum: %" PRIx64
": %" PRIu64
,
998 le64toh(o
->entry
.seqnum
),
1001 if (!VALID_REALTIME(le64toh(o
->entry
.realtime
)))
1002 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1003 "Invalid entry realtime timestamp: %" PRIu64
": %" PRIu64
,
1004 le64toh(o
->entry
.realtime
),
1007 if (!VALID_MONOTONIC(le64toh(o
->entry
.monotonic
)))
1008 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1009 "Invalid entry monotonic timestamp: %" PRIu64
": %" PRIu64
,
1010 le64toh(o
->entry
.monotonic
),
1013 if (sd_id128_is_null(o
->entry
.boot_id
))
1014 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1015 "Invalid object entry with an empty boot ID: %" PRIu64
,
1021 case OBJECT_DATA_HASH_TABLE
:
1022 case OBJECT_FIELD_HASH_TABLE
: {
1025 sz
= le64toh(READ_NOW(o
->object
.size
));
1026 if (sz
< offsetof(Object
, hash_table
.items
) ||
1027 (sz
- offsetof(Object
, hash_table
.items
)) % sizeof(HashItem
) != 0 ||
1028 (sz
- offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
) <= 0)
1029 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1030 "Invalid %s hash table size: %" PRIu64
": %" PRIu64
,
1031 journal_object_type_to_string(o
->object
.type
),
1038 case OBJECT_ENTRY_ARRAY
: {
1041 sz
= le64toh(READ_NOW(o
->object
.size
));
1042 if (sz
< offsetof(Object
, entry_array
.items
) ||
1043 (sz
- offsetof(Object
, entry_array
.items
)) % journal_file_entry_array_item_size(f
) != 0 ||
1044 (sz
- offsetof(Object
, entry_array
.items
)) / journal_file_entry_array_item_size(f
) <= 0)
1045 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1046 "Invalid object entry array size: %" PRIu64
": %" PRIu64
,
1049 /* Here, we request that the offset of each entry array object is in strictly increasing order. */
1050 next
= le64toh(o
->entry_array
.next_entry_array_offset
);
1051 if (!VALID64(next
) || (next
> 0 && next
<= offset
))
1052 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1053 "Invalid object entry array next_entry_array_offset: %" PRIu64
": %" PRIu64
,
1061 if (le64toh(o
->object
.size
) != sizeof(TagObject
))
1062 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1063 "Invalid object tag size: %" PRIu64
": %" PRIu64
,
1064 le64toh(o
->object
.size
),
1067 if (!VALID_EPOCH(le64toh(o
->tag
.epoch
)))
1068 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1069 "Invalid object tag epoch: %" PRIu64
": %" PRIu64
,
1070 le64toh(o
->tag
.epoch
), offset
);
1078 int journal_file_move_to_object(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
**ret
) {
1084 /* Even if this function fails, it may clear, overwrite, or alter previously cached entries with the
1085 * same type. After this function has been called, all previously read objects with the same type may
1086 * be invalidated, hence must be re-read before use. */
1088 /* Objects may only be located at multiple of 64 bit */
1089 if (!VALID64(offset
))
1090 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1091 "Attempt to move to %s object at non-64-bit boundary: %" PRIu64
,
1092 journal_object_type_to_string(type
),
1095 /* Object may not be located in the file header */
1096 if (offset
< le64toh(f
->header
->header_size
))
1097 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1098 "Attempt to move to %s object located in file header: %" PRIu64
,
1099 journal_object_type_to_string(type
),
1102 r
= journal_file_move_to(f
, type
, false, offset
, sizeof(ObjectHeader
), (void**) &o
);
1106 r
= check_object_header(f
, o
, type
, offset
);
1110 r
= journal_file_move_to(f
, type
, false, offset
, le64toh(READ_NOW(o
->object
.size
)), (void**) &o
);
1114 r
= check_object_header(f
, o
, type
, offset
);
1118 r
= check_object(f
, o
, offset
);
1128 int journal_file_read_object_header(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
*ret
) {
1135 /* Objects may only be located at multiple of 64 bit */
1136 if (!VALID64(offset
))
1137 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1138 "Attempt to read %s object at non-64-bit boundary: %" PRIu64
,
1139 journal_object_type_to_string(type
), offset
);
1141 /* Object may not be located in the file header */
1142 if (offset
< le64toh(f
->header
->header_size
))
1143 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1144 "Attempt to read %s object located in file header: %" PRIu64
,
1145 journal_object_type_to_string(type
), offset
);
1147 /* This will likely read too much data but it avoids having to call pread() twice. */
1148 n
= pread(f
->fd
, &o
, sizeof(o
), offset
);
1150 return log_debug_errno(errno
, "Failed to read journal %s object at offset: %" PRIu64
,
1151 journal_object_type_to_string(type
), offset
);
1153 if ((size_t) n
< sizeof(o
.object
))
1154 return log_debug_errno(SYNTHETIC_ERRNO(EIO
),
1155 "Failed to read short %s object at offset: %" PRIu64
,
1156 journal_object_type_to_string(type
), offset
);
1158 r
= check_object_header(f
, &o
, type
, offset
);
1162 if ((size_t) n
< minimum_header_size(f
, &o
))
1163 return log_debug_errno(SYNTHETIC_ERRNO(EIO
),
1164 "Short read while reading %s object: %" PRIu64
,
1165 journal_object_type_to_string(type
), offset
);
1167 r
= check_object(f
, &o
, offset
);
1177 static uint64_t inc_seqnum(uint64_t seqnum
) {
1178 if (seqnum
< UINT64_MAX
-1)
1181 return 1; /* skip over UINT64_MAX and 0 when we run out of seqnums and start again */
1184 static uint64_t journal_file_entry_seqnum(
1188 uint64_t next_seqnum
;
1193 /* Picks a new sequence number for the entry we are about to add and returns it. */
1195 next_seqnum
= inc_seqnum(le64toh(f
->header
->tail_entry_seqnum
));
1197 /* If an external seqnum counter was passed, we update both the local and the external one, and set
1198 * it to the maximum of both */
1200 *seqnum
= next_seqnum
= MAX(inc_seqnum(*seqnum
), next_seqnum
);
1202 f
->header
->tail_entry_seqnum
= htole64(next_seqnum
);
1204 if (f
->header
->head_entry_seqnum
== 0)
1205 f
->header
->head_entry_seqnum
= htole64(next_seqnum
);
1210 int journal_file_append_object(
1214 Object
**ret_object
,
1215 uint64_t *ret_offset
) {
1223 assert(type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
);
1224 assert(size
>= sizeof(ObjectHeader
));
1226 r
= journal_file_set_online(f
);
1230 r
= journal_file_tail_end_by_mmap(f
, &p
);
1234 r
= journal_file_allocate(f
, p
, size
);
1238 r
= journal_file_move_to(f
, type
, false, p
, size
, (void**) &o
);
1242 o
->object
= (ObjectHeader
) {
1244 .size
= htole64(size
),
1247 f
->header
->tail_object_offset
= htole64(p
);
1248 f
->header
->n_objects
= htole64(le64toh(f
->header
->n_objects
) + 1);
1259 static int journal_file_setup_data_hash_table(JournalFile
*f
) {
1267 /* We estimate that we need 1 hash table entry per 768 bytes
1268 of journal file and we want to make sure we never get
1269 beyond 75% fill level. Calculate the hash table size for
1270 the maximum file size based on these metrics. */
1272 s
= (f
->metrics
.max_size
* 4 / 768 / 3) * sizeof(HashItem
);
1273 if (s
< DEFAULT_DATA_HASH_TABLE_SIZE
)
1274 s
= DEFAULT_DATA_HASH_TABLE_SIZE
;
1276 log_debug("Reserving %"PRIu64
" entries in data hash table.", s
/ sizeof(HashItem
));
1278 r
= journal_file_append_object(f
,
1279 OBJECT_DATA_HASH_TABLE
,
1280 offsetof(Object
, hash_table
.items
) + s
,
1285 memzero(o
->hash_table
.items
, s
);
1287 f
->header
->data_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1288 f
->header
->data_hash_table_size
= htole64(s
);
1293 static int journal_file_setup_field_hash_table(JournalFile
*f
) {
1301 /* We use a fixed size hash table for the fields as this
1302 * number should grow very slowly only */
1304 s
= DEFAULT_FIELD_HASH_TABLE_SIZE
;
1305 log_debug("Reserving %"PRIu64
" entries in field hash table.", s
/ sizeof(HashItem
));
1307 r
= journal_file_append_object(f
,
1308 OBJECT_FIELD_HASH_TABLE
,
1309 offsetof(Object
, hash_table
.items
) + s
,
1314 memzero(o
->hash_table
.items
, s
);
1316 f
->header
->field_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1317 f
->header
->field_hash_table_size
= htole64(s
);
1322 int journal_file_map_data_hash_table(JournalFile
*f
) {
1330 if (f
->data_hash_table
)
1333 p
= le64toh(f
->header
->data_hash_table_offset
);
1334 s
= le64toh(f
->header
->data_hash_table_size
);
1336 r
= journal_file_move_to(f
,
1337 OBJECT_DATA_HASH_TABLE
,
1344 f
->data_hash_table
= t
;
1348 int journal_file_map_field_hash_table(JournalFile
*f
) {
1356 if (f
->field_hash_table
)
1359 p
= le64toh(f
->header
->field_hash_table_offset
);
1360 s
= le64toh(f
->header
->field_hash_table_size
);
1362 r
= journal_file_move_to(f
,
1363 OBJECT_FIELD_HASH_TABLE
,
1370 f
->field_hash_table
= t
;
1374 static int journal_file_link_field(
1385 assert(f
->field_hash_table
);
1389 if (o
->object
.type
!= OBJECT_FIELD
)
1392 m
= le64toh(READ_NOW(f
->header
->field_hash_table_size
)) / sizeof(HashItem
);
1396 /* This might alter the window we are looking at */
1397 o
->field
.next_hash_offset
= o
->field
.head_data_offset
= 0;
1400 p
= le64toh(f
->field_hash_table
[h
].tail_hash_offset
);
1402 f
->field_hash_table
[h
].head_hash_offset
= htole64(offset
);
1404 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1408 o
->field
.next_hash_offset
= htole64(offset
);
1411 f
->field_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1413 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
1414 f
->header
->n_fields
= htole64(le64toh(f
->header
->n_fields
) + 1);
1419 static int journal_file_link_data(
1430 assert(f
->data_hash_table
);
1434 if (o
->object
.type
!= OBJECT_DATA
)
1437 m
= le64toh(READ_NOW(f
->header
->data_hash_table_size
)) / sizeof(HashItem
);
1441 /* This might alter the window we are looking at */
1442 o
->data
.next_hash_offset
= o
->data
.next_field_offset
= 0;
1443 o
->data
.entry_offset
= o
->data
.entry_array_offset
= 0;
1444 o
->data
.n_entries
= 0;
1447 p
= le64toh(f
->data_hash_table
[h
].tail_hash_offset
);
1449 /* Only entry in the hash table is easy */
1450 f
->data_hash_table
[h
].head_hash_offset
= htole64(offset
);
1452 /* Move back to the previous data object, to patch in
1455 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1459 o
->data
.next_hash_offset
= htole64(offset
);
1462 f
->data_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1464 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
1465 f
->header
->n_data
= htole64(le64toh(f
->header
->n_data
) + 1);
1470 static int get_next_hash_offset(
1473 le64_t
*next_hash_offset
,
1475 le64_t
*header_max_depth
) {
1481 assert(next_hash_offset
);
1484 nextp
= le64toh(READ_NOW(*next_hash_offset
));
1486 if (nextp
<= *p
) /* Refuse going in loops */
1487 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1488 "Detected hash item loop in %s, refusing.", f
->path
);
1492 /* If the depth of this hash chain is larger than all others we have seen so far, record it */
1493 if (header_max_depth
&& journal_file_writable(f
))
1494 *header_max_depth
= htole64(MAX(*depth
, le64toh(*header_max_depth
)));
1501 int journal_file_find_field_object_with_hash(
1506 Object
**ret_object
,
1507 uint64_t *ret_offset
) {
1509 uint64_t p
, osize
, h
, m
, depth
= 0;
1517 /* If the field hash table is empty, we can't find anything */
1518 if (le64toh(f
->header
->field_hash_table_size
) <= 0)
1521 /* Map the field hash table, if it isn't mapped yet. */
1522 r
= journal_file_map_field_hash_table(f
);
1526 osize
= offsetof(Object
, field
.payload
) + size
;
1528 m
= le64toh(READ_NOW(f
->header
->field_hash_table_size
)) / sizeof(HashItem
);
1533 p
= le64toh(f
->field_hash_table
[h
].head_hash_offset
);
1537 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1541 if (le64toh(o
->field
.hash
) == hash
&&
1542 le64toh(o
->object
.size
) == osize
&&
1543 memcmp(o
->field
.payload
, field
, size
) == 0) {
1553 r
= get_next_hash_offset(
1556 &o
->field
.next_hash_offset
,
1558 JOURNAL_HEADER_CONTAINS(f
->header
, field_hash_chain_depth
) ? &f
->header
->field_hash_chain_depth
: NULL
);
1566 uint64_t journal_file_hash_data(
1573 assert(data
|| sz
== 0);
1575 /* We try to unify our codebase on siphash, hence new-styled journal files utilizing the keyed hash
1576 * function use siphash. Old journal files use the Jenkins hash. */
1578 if (JOURNAL_HEADER_KEYED_HASH(f
->header
))
1579 return siphash24(data
, sz
, f
->header
->file_id
.bytes
);
1581 return jenkins_hash64(data
, sz
);
1584 int journal_file_find_field_object(
1588 Object
**ret_object
,
1589 uint64_t *ret_offset
) {
1595 return journal_file_find_field_object_with_hash(
1598 journal_file_hash_data(f
, field
, size
),
1599 ret_object
, ret_offset
);
1602 int journal_file_find_data_object_with_hash(
1607 Object
**ret_object
,
1608 uint64_t *ret_offset
) {
1610 uint64_t p
, h
, m
, depth
= 0;
1615 assert(data
|| size
== 0);
1617 /* If there's no data hash table, then there's no entry. */
1618 if (le64toh(f
->header
->data_hash_table_size
) <= 0)
1621 /* Map the data hash table, if it isn't mapped yet. */
1622 r
= journal_file_map_data_hash_table(f
);
1626 m
= le64toh(READ_NOW(f
->header
->data_hash_table_size
)) / sizeof(HashItem
);
1631 p
= le64toh(f
->data_hash_table
[h
].head_hash_offset
);
1638 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1642 if (le64toh(o
->data
.hash
) != hash
)
1645 r
= journal_file_data_payload(f
, o
, p
, NULL
, 0, 0, &d
, &rsize
);
1648 assert(r
> 0); /* journal_file_data_payload() always returns > 0 if no field is provided. */
1650 if (memcmp_nn(data
, size
, d
, rsize
) == 0) {
1661 r
= get_next_hash_offset(
1664 &o
->data
.next_hash_offset
,
1666 JOURNAL_HEADER_CONTAINS(f
->header
, data_hash_chain_depth
) ? &f
->header
->data_hash_chain_depth
: NULL
);
1674 int journal_file_find_data_object(
1678 Object
**ret_object
,
1679 uint64_t *ret_offset
) {
1682 assert(data
|| size
== 0);
1684 return journal_file_find_data_object_with_hash(
1687 journal_file_hash_data(f
, data
, size
),
1688 ret_object
, ret_offset
);
1691 bool journal_field_valid(const char *p
, size_t l
, bool allow_protected
) {
1692 /* We kinda enforce POSIX syntax recommendations for
1693 environment variables here, but make a couple of additional
1696 http://pubs.opengroup.org/onlinepubs/000095399/basedefs/xbd_chap08.html */
1703 /* No empty field names */
1707 /* Don't allow names longer than 64 chars */
1711 /* Variables starting with an underscore are protected */
1712 if (!allow_protected
&& p
[0] == '_')
1715 /* Don't allow digits as first character */
1716 if (ascii_isdigit(p
[0]))
1719 /* Only allow A-Z0-9 and '_' */
1720 for (const char *a
= p
; a
< p
+ l
; a
++)
1721 if ((*a
< 'A' || *a
> 'Z') &&
1722 !ascii_isdigit(*a
) &&
1729 static int journal_file_append_field(
1733 Object
**ret_object
,
1734 uint64_t *ret_offset
) {
1745 if (!journal_field_valid(field
, size
, true))
1748 hash
= journal_file_hash_data(f
, field
, size
);
1750 r
= journal_file_find_field_object_with_hash(f
, field
, size
, hash
, ret_object
, ret_offset
);
1756 osize
= offsetof(Object
, field
.payload
) + size
;
1757 r
= journal_file_append_object(f
, OBJECT_FIELD
, osize
, &o
, &p
);
1761 o
->field
.hash
= htole64(hash
);
1762 memcpy(o
->field
.payload
, field
, size
);
1764 r
= journal_file_link_field(f
, o
, p
, hash
);
1768 /* The linking might have altered the window, so let's only pass the offset to hmac which will
1769 * move to the object again if needed. */
1772 r
= journal_file_hmac_put_object(f
, OBJECT_FIELD
, NULL
, p
);
1778 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, ret_object
);
1789 static int maybe_compress_payload(JournalFile
*f
, uint8_t *dst
, const uint8_t *src
, uint64_t size
, size_t *rsize
) {
1793 #if HAVE_COMPRESSION
1797 c
= JOURNAL_FILE_COMPRESSION(f
);
1798 if (c
== COMPRESSION_NONE
|| size
< f
->compress_threshold_bytes
)
1801 r
= compress_blob(c
, src
, size
, dst
, size
- 1, rsize
);
1803 return log_debug_errno(r
, "Failed to compress data object using %s, ignoring: %m", compression_to_string(c
));
1805 log_debug("Compressed data object %"PRIu64
" -> %zu using %s", size
, *rsize
, compression_to_string(c
));
1807 return 1; /* compressed */
1813 static int journal_file_append_data(
1817 Object
**ret_object
,
1818 uint64_t *ret_offset
) {
1820 uint64_t hash
, p
, osize
;
1828 if (!data
|| size
== 0)
1831 hash
= journal_file_hash_data(f
, data
, size
);
1833 r
= journal_file_find_data_object_with_hash(f
, data
, size
, hash
, ret_object
, ret_offset
);
1839 eq
= memchr(data
, '=', size
);
1843 osize
= journal_file_data_payload_offset(f
) + size
;
1844 r
= journal_file_append_object(f
, OBJECT_DATA
, osize
, &o
, &p
);
1848 o
->data
.hash
= htole64(hash
);
1850 r
= maybe_compress_payload(f
, journal_file_data_payload_field(f
, o
), data
, size
, &rsize
);
1852 /* We don't really care failures, let's continue without compression */
1853 memcpy_safe(journal_file_data_payload_field(f
, o
), data
, size
);
1855 Compression c
= JOURNAL_FILE_COMPRESSION(f
);
1857 assert(c
>= 0 && c
< _COMPRESSION_MAX
&& c
!= COMPRESSION_NONE
);
1859 o
->object
.size
= htole64(journal_file_data_payload_offset(f
) + rsize
);
1860 o
->object
.flags
|= COMPRESSION_TO_OBJECT_FLAG(c
);
1863 r
= journal_file_link_data(f
, o
, p
, hash
);
1867 /* The linking might have altered the window, so let's refresh our pointer. */
1868 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1873 r
= journal_file_hmac_put_object(f
, OBJECT_DATA
, o
, p
);
1878 /* Create field object ... */
1879 r
= journal_file_append_field(f
, data
, (uint8_t*) eq
- (uint8_t*) data
, &fo
, NULL
);
1883 /* ... and link it in. */
1884 o
->data
.next_field_offset
= fo
->field
.head_data_offset
;
1885 fo
->field
.head_data_offset
= le64toh(p
);
1896 static int maybe_decompress_payload(
1900 Compression compression
,
1902 size_t field_length
,
1903 size_t data_threshold
,
1909 /* We can't read objects larger than 4G on a 32-bit machine */
1910 if ((uint64_t) (size_t) size
!= size
)
1913 if (compression
!= COMPRESSION_NONE
) {
1914 #if HAVE_COMPRESSION
1919 r
= decompress_startswith(compression
, payload
, size
, &f
->compress_buffer
, field
,
1922 return log_debug_errno(r
,
1923 "Cannot decompress %s object of length %" PRIu64
": %m",
1924 compression_to_string(compression
),
1935 r
= decompress_blob(compression
, payload
, size
, &f
->compress_buffer
, &rsize
, 0);
1940 *ret_data
= f
->compress_buffer
;
1944 return -EPROTONOSUPPORT
;
1947 if (field
&& (size
< field_length
+ 1 || memcmp(payload
, field
, field_length
) != 0 || payload
[field_length
] != '=')) {
1956 *ret_data
= payload
;
1958 *ret_size
= (size_t) size
;
1964 int journal_file_data_payload(
1969 size_t field_length
,
1970 size_t data_threshold
,
1979 assert(!field
== (field_length
== 0)); /* These must be specified together. */
1982 r
= journal_file_move_to_object(f
, OBJECT_DATA
, offset
, &o
);
1987 size
= le64toh(READ_NOW(o
->object
.size
));
1988 if (size
< journal_file_data_payload_offset(f
))
1991 size
-= journal_file_data_payload_offset(f
);
1993 c
= COMPRESSION_FROM_OBJECT(o
);
1995 return -EPROTONOSUPPORT
;
1997 return maybe_decompress_payload(f
, journal_file_data_payload_field(f
, o
), size
, c
, field
,
1998 field_length
, data_threshold
, ret_data
, ret_size
);
2001 uint64_t journal_file_entry_n_items(JournalFile
*f
, Object
*o
) {
2007 if (o
->object
.type
!= OBJECT_ENTRY
)
2010 sz
= le64toh(READ_NOW(o
->object
.size
));
2011 if (sz
< offsetof(Object
, entry
.items
))
2014 return (sz
- offsetof(Object
, entry
.items
)) / journal_file_entry_item_size(f
);
2017 uint64_t journal_file_entry_array_n_items(JournalFile
*f
, Object
*o
) {
2023 if (o
->object
.type
!= OBJECT_ENTRY_ARRAY
)
2026 sz
= le64toh(READ_NOW(o
->object
.size
));
2027 if (sz
< offsetof(Object
, entry_array
.items
))
2030 return (sz
- offsetof(Object
, entry_array
.items
)) / journal_file_entry_array_item_size(f
);
2033 uint64_t journal_file_hash_table_n_items(Object
*o
) {
2038 if (!IN_SET(o
->object
.type
, OBJECT_DATA_HASH_TABLE
, OBJECT_FIELD_HASH_TABLE
))
2041 sz
= le64toh(READ_NOW(o
->object
.size
));
2042 if (sz
< offsetof(Object
, hash_table
.items
))
2045 return (sz
- offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
);
2048 static void write_entry_array_item(JournalFile
*f
, Object
*o
, uint64_t i
, uint64_t p
) {
2052 if (JOURNAL_HEADER_COMPACT(f
->header
)) {
2053 assert(p
<= UINT32_MAX
);
2054 o
->entry_array
.items
.compact
[i
] = htole32(p
);
2056 o
->entry_array
.items
.regular
[i
] = htole64(p
);
2059 static int link_entry_into_array(
2067 uint64_t n
= 0, ap
= 0, q
, i
, a
, hidx
;
2077 a
= tail
? le32toh(*tail
) : le64toh(*first
);
2078 hidx
= le64toh(READ_NOW(*idx
));
2079 i
= tidx
? le32toh(READ_NOW(*tidx
)) : hidx
;
2082 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2086 n
= journal_file_entry_array_n_items(f
, o
);
2088 write_entry_array_item(f
, o
, i
, p
);
2089 *idx
= htole64(hidx
+ 1);
2091 *tidx
= htole32(le32toh(*tidx
) + 1);
2097 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
2108 r
= journal_file_append_object(f
, OBJECT_ENTRY_ARRAY
,
2109 offsetof(Object
, entry_array
.items
) + n
* journal_file_entry_array_item_size(f
),
2115 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY_ARRAY
, o
, q
);
2120 write_entry_array_item(f
, o
, i
, p
);
2123 *first
= htole64(q
);
2125 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, ap
, &o
);
2129 o
->entry_array
.next_entry_array_offset
= htole64(q
);
2135 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
2136 f
->header
->n_entry_arrays
= htole64(le64toh(f
->header
->n_entry_arrays
) + 1);
2138 *idx
= htole64(hidx
+ 1);
2145 static int link_entry_into_array_plus_one(
2163 hidx
= le64toh(READ_NOW(*idx
));
2164 if (hidx
== UINT64_MAX
)
2167 *extra
= htole64(p
);
2171 i
= htole64(hidx
- 1);
2172 r
= link_entry_into_array(f
, first
, &i
, tail
, tidx
, p
);
2177 *idx
= htole64(hidx
+ 1);
2181 static int journal_file_link_entry_item(JournalFile
*f
, uint64_t offset
, uint64_t p
) {
2188 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
2192 return link_entry_into_array_plus_one(f
,
2193 &o
->data
.entry_offset
,
2194 &o
->data
.entry_array_offset
,
2196 JOURNAL_HEADER_COMPACT(f
->header
) ? &o
->data
.compact
.tail_entry_array_offset
: NULL
,
2197 JOURNAL_HEADER_COMPACT(f
->header
) ? &o
->data
.compact
.tail_entry_array_n_entries
: NULL
,
2201 static int journal_file_link_entry(
2205 const EntryItem items
[],
2215 if (o
->object
.type
!= OBJECT_ENTRY
)
2218 __atomic_thread_fence(__ATOMIC_SEQ_CST
);
2220 /* Link up the entry itself */
2221 r
= link_entry_into_array(f
,
2222 &f
->header
->entry_array_offset
,
2223 &f
->header
->n_entries
,
2224 JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_array_offset
) ? &f
->header
->tail_entry_array_offset
: NULL
,
2225 JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_array_n_entries
) ? &f
->header
->tail_entry_array_n_entries
: NULL
,
2230 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
2232 if (f
->header
->head_entry_realtime
== 0)
2233 f
->header
->head_entry_realtime
= o
->entry
.realtime
;
2235 f
->header
->tail_entry_realtime
= o
->entry
.realtime
;
2236 f
->header
->tail_entry_monotonic
= o
->entry
.monotonic
;
2237 if (JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_offset
))
2238 f
->header
->tail_entry_offset
= htole64(offset
);
2239 f
->newest_mtime
= 0; /* we have a new tail entry now, explicitly invalidate newest boot id/timestamp info */
2241 /* Link up the items */
2242 for (uint64_t i
= 0; i
< n_items
; i
++) {
2245 /* If we fail to link an entry item because we can't allocate a new entry array, don't fail
2246 * immediately but try to link the other entry items since it might still be possible to link
2247 * those if they don't require a new entry array to be allocated. */
2249 k
= journal_file_link_entry_item(f
, offset
, items
[i
].object_offset
);
2259 static void write_entry_item(JournalFile
*f
, Object
*o
, uint64_t i
, const EntryItem
*item
) {
2264 if (JOURNAL_HEADER_COMPACT(f
->header
)) {
2265 assert(item
->object_offset
<= UINT32_MAX
);
2266 o
->entry
.items
.compact
[i
].object_offset
= htole32(item
->object_offset
);
2268 o
->entry
.items
.regular
[i
].object_offset
= htole64(item
->object_offset
);
2269 o
->entry
.items
.regular
[i
].hash
= htole64(item
->hash
);
2273 static int journal_file_append_entry_internal(
2275 const dual_timestamp
*ts
,
2276 const sd_id128_t
*boot_id
,
2277 const sd_id128_t
*machine_id
,
2279 const EntryItem items
[],
2282 sd_id128_t
*seqnum_id
,
2283 Object
**ret_object
,
2284 uint64_t *ret_offset
) {
2295 assert(!sd_id128_is_null(*boot_id
));
2296 assert(items
|| n_items
== 0);
2298 if (f
->strict_order
) {
2299 /* If requested be stricter with ordering in this journal file, to make searching via
2300 * bisection fully deterministic. This is an optional feature, so that if desired journal
2301 * files can be written where the ordering is not strictly enforced (in which case bisection
2302 * will yield *a* result, but not the *only* result, when searching for points in
2303 * time). Strict ordering mode is enabled when journald originally writes the files, but
2304 * might not necessarily be if other tools (the remoting tools for example) write journal
2305 * files from combined sources.
2307 * Typically, if any of the errors generated here are seen journald will just rotate the
2308 * journal files and start anew. */
2310 if (ts
->realtime
< le64toh(f
->header
->tail_entry_realtime
))
2311 return log_debug_errno(SYNTHETIC_ERRNO(EREMCHG
),
2312 "Realtime timestamp %" PRIu64
" smaller than previous realtime "
2313 "timestamp %" PRIu64
", refusing entry.",
2314 ts
->realtime
, le64toh(f
->header
->tail_entry_realtime
));
2316 if (sd_id128_equal(*boot_id
, f
->header
->tail_entry_boot_id
) &&
2317 ts
->monotonic
< le64toh(f
->header
->tail_entry_monotonic
))
2318 return log_debug_errno(
2319 SYNTHETIC_ERRNO(ENOTNAM
),
2320 "Monotonic timestamp %" PRIu64
2321 " smaller than previous monotonic timestamp %" PRIu64
2322 " while having the same boot ID, refusing entry.",
2324 le64toh(f
->header
->tail_entry_monotonic
));
2328 /* Settle the passed in sequence number ID */
2330 if (sd_id128_is_null(*seqnum_id
))
2331 *seqnum_id
= f
->header
->seqnum_id
; /* Caller has none assigned, then copy the one from the file */
2332 else if (!sd_id128_equal(*seqnum_id
, f
->header
->seqnum_id
)) {
2333 /* Different seqnum IDs? We can't allow entries from multiple IDs end up in the same journal.*/
2334 if (le64toh(f
->header
->n_entries
) == 0)
2335 f
->header
->seqnum_id
= *seqnum_id
; /* Caller has one, and file so far has no entries, then copy the one from the caller */
2337 return log_debug_errno(SYNTHETIC_ERRNO(EILSEQ
),
2338 "Sequence number IDs don't match, refusing entry.");
2342 if (machine_id
&& sd_id128_is_null(f
->header
->machine_id
))
2343 /* Initialize machine ID when not set yet */
2344 f
->header
->machine_id
= *machine_id
;
2346 osize
= offsetof(Object
, entry
.items
) + (n_items
* journal_file_entry_item_size(f
));
2348 r
= journal_file_append_object(f
, OBJECT_ENTRY
, osize
, &o
, &np
);
2352 o
->entry
.seqnum
= htole64(journal_file_entry_seqnum(f
, seqnum
));
2353 o
->entry
.realtime
= htole64(ts
->realtime
);
2354 o
->entry
.monotonic
= htole64(ts
->monotonic
);
2355 o
->entry
.xor_hash
= htole64(xor_hash
);
2356 o
->entry
.boot_id
= f
->header
->tail_entry_boot_id
= *boot_id
;
2358 for (size_t i
= 0; i
< n_items
; i
++)
2359 write_entry_item(f
, o
, i
, &items
[i
]);
2362 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY
, o
, np
);
2367 r
= journal_file_link_entry(f
, o
, np
, items
, n_items
);
2380 void journal_file_post_change(JournalFile
*f
) {
2386 /* inotify() does not receive IN_MODIFY events from file
2387 * accesses done via mmap(). After each access we hence
2388 * trigger IN_MODIFY by truncating the journal file to its
2389 * current size which triggers IN_MODIFY. */
2391 __atomic_thread_fence(__ATOMIC_SEQ_CST
);
2393 if (ftruncate(f
->fd
, f
->last_stat
.st_size
) < 0)
2394 log_debug_errno(errno
, "Failed to truncate file to its own size: %m");
2397 static int post_change_thunk(sd_event_source
*timer
, uint64_t usec
, void *userdata
) {
2400 journal_file_post_change(userdata
);
2405 static void schedule_post_change(JournalFile
*f
) {
2410 assert(f
->post_change_timer
);
2412 assert_se(e
= sd_event_source_get_event(f
->post_change_timer
));
2414 /* If we are already going down, post the change immediately. */
2415 if (IN_SET(sd_event_get_state(e
), SD_EVENT_EXITING
, SD_EVENT_FINISHED
))
2418 r
= sd_event_source_get_enabled(f
->post_change_timer
, NULL
);
2420 log_debug_errno(r
, "Failed to get ftruncate timer state: %m");
2426 r
= sd_event_source_set_time_relative(f
->post_change_timer
, f
->post_change_timer_period
);
2428 log_debug_errno(r
, "Failed to set time for scheduling ftruncate: %m");
2432 r
= sd_event_source_set_enabled(f
->post_change_timer
, SD_EVENT_ONESHOT
);
2434 log_debug_errno(r
, "Failed to enable scheduled ftruncate: %m");
2441 /* On failure, let's simply post the change immediately. */
2442 journal_file_post_change(f
);
2445 /* Enable coalesced change posting in a timer on the provided sd_event instance */
2446 int journal_file_enable_post_change_timer(JournalFile
*f
, sd_event
*e
, usec_t t
) {
2447 _cleanup_(sd_event_source_unrefp
) sd_event_source
*timer
= NULL
;
2451 assert_return(!f
->post_change_timer
, -EINVAL
);
2455 r
= sd_event_add_time(e
, &timer
, CLOCK_MONOTONIC
, 0, 0, post_change_thunk
, f
);
2459 r
= sd_event_source_set_enabled(timer
, SD_EVENT_OFF
);
2463 f
->post_change_timer
= TAKE_PTR(timer
);
2464 f
->post_change_timer_period
= t
;
2469 static int entry_item_cmp(const EntryItem
*a
, const EntryItem
*b
) {
2470 return CMP(ASSERT_PTR(a
)->object_offset
, ASSERT_PTR(b
)->object_offset
);
2473 static size_t remove_duplicate_entry_items(EntryItem items
[], size_t n
) {
2476 assert(items
|| n
== 0);
2481 for (size_t i
= 1; i
< n
; i
++)
2482 if (items
[i
].object_offset
!= items
[j
- 1].object_offset
)
2483 items
[j
++] = items
[i
];
2488 int journal_file_append_entry(
2490 const dual_timestamp
*ts
,
2491 const sd_id128_t
*boot_id
,
2492 const struct iovec iovec
[],
2495 sd_id128_t
*seqnum_id
,
2496 Object
**ret_object
,
2497 uint64_t *ret_offset
) {
2499 _cleanup_free_ EntryItem
*items_alloc
= NULL
;
2501 uint64_t xor_hash
= 0;
2502 struct dual_timestamp _ts
;
2503 sd_id128_t _boot_id
, _machine_id
, *machine_id
;
2509 assert(n_iovec
> 0);
2512 if (!VALID_REALTIME(ts
->realtime
))
2513 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
2514 "Invalid realtime timestamp %" PRIu64
", refusing entry.",
2516 if (!VALID_MONOTONIC(ts
->monotonic
))
2517 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
2518 "Invalid monotomic timestamp %" PRIu64
", refusing entry.",
2521 dual_timestamp_get(&_ts
);
2526 if (sd_id128_is_null(*boot_id
))
2527 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
), "Empty boot ID, refusing entry.");
2529 r
= sd_id128_get_boot(&_boot_id
);
2533 boot_id
= &_boot_id
;
2536 r
= sd_id128_get_machine(&_machine_id
);
2537 if (ERRNO_IS_NEG_MACHINE_ID_UNSET(r
))
2538 /* Gracefully handle the machine ID not being initialized yet */
2543 machine_id
= &_machine_id
;
2546 r
= journal_file_maybe_append_tag(f
, ts
->realtime
);
2551 if (n_iovec
< ALLOCA_MAX
/ sizeof(EntryItem
) / 2)
2552 items
= newa(EntryItem
, n_iovec
);
2554 items_alloc
= new(EntryItem
, n_iovec
);
2558 items
= items_alloc
;
2561 for (size_t i
= 0; i
< n_iovec
; i
++) {
2565 r
= journal_file_append_data(f
, iovec
[i
].iov_base
, iovec
[i
].iov_len
, &o
, &p
);
2569 /* When calculating the XOR hash field, we need to take special care if the "keyed-hash"
2570 * journal file flag is on. We use the XOR hash field to quickly determine the identity of a
2571 * specific record, and give records with otherwise identical position (i.e. match in seqno,
2572 * timestamp, …) a stable ordering. But for that we can't have it that the hash of the
2573 * objects in each file is different since they are keyed. Hence let's calculate the Jenkins
2574 * hash here for that. This also has the benefit that cursors for old and new journal files
2575 * are completely identical (they include the XOR hash after all). For classic Jenkins-hash
2576 * files things are easier, we can just take the value from the stored record directly. */
2578 if (JOURNAL_HEADER_KEYED_HASH(f
->header
))
2579 xor_hash
^= jenkins_hash64(iovec
[i
].iov_base
, iovec
[i
].iov_len
);
2581 xor_hash
^= le64toh(o
->data
.hash
);
2583 items
[i
] = (EntryItem
) {
2585 .hash
= le64toh(o
->data
.hash
),
2589 /* Order by the position on disk, in order to improve seek
2590 * times for rotating media. */
2591 typesafe_qsort(items
, n_iovec
, entry_item_cmp
);
2592 n_iovec
= remove_duplicate_entry_items(items
, n_iovec
);
2594 r
= journal_file_append_entry_internal(
2607 /* If the memory mapping triggered a SIGBUS then we return an
2608 * IO error and ignore the error code passed down to us, since
2609 * it is very likely just an effect of a nullified replacement
2612 if (mmap_cache_fd_got_sigbus(f
->cache_fd
))
2615 if (f
->post_change_timer
)
2616 schedule_post_change(f
);
2618 journal_file_post_change(f
);
2623 typedef struct ChainCacheItem
{
2624 uint64_t first
; /* The offset of the entry array object at the beginning of the chain,
2625 * i.e., le64toh(f->header->entry_array_offset), or le64toh(o->data.entry_offset). */
2626 uint64_t array
; /* The offset of the cached entry array object. */
2627 uint64_t begin
; /* The offset of the first item in the cached array. */
2628 uint64_t total
; /* The total number of items in all arrays before the cached one in the chain. */
2629 uint64_t last_index
; /* The last index we looked at in the cached array, to optimize locality when bisecting. */
2632 static void chain_cache_put(
2639 uint64_t last_index
) {
2644 /* If the chain item to cache for this chain is the
2645 * first one it's not worth caching anything */
2649 if (ordered_hashmap_size(h
) >= CHAIN_CACHE_MAX
) {
2650 ci
= ordered_hashmap_steal_first(h
);
2653 ci
= new(ChainCacheItem
, 1);
2660 if (ordered_hashmap_put(h
, &ci
->first
, ci
) < 0) {
2665 assert(ci
->first
== first
);
2670 ci
->last_index
= last_index
;
2673 static int bump_array_index(uint64_t *i
, direction_t direction
, uint64_t n
) {
2676 /* Increase or decrease the specified index, in the right direction. */
2678 if (direction
== DIRECTION_DOWN
) {
2693 static int bump_entry_array(
2695 Object
*o
, /* the current entry array object. */
2696 uint64_t offset
, /* the offset of the entry array object. */
2697 uint64_t first
, /* The offset of the first entry array object in the chain. */
2698 direction_t direction
,
2706 if (direction
== DIRECTION_DOWN
) {
2708 assert(o
->object
.type
== OBJECT_ENTRY_ARRAY
);
2710 *ret
= le64toh(o
->entry_array
.next_entry_array_offset
);
2713 /* Entry array chains are a singly linked list, so to find the previous array in the chain, we have
2714 * to start iterating from the top. */
2718 uint64_t p
= first
, q
= 0;
2719 while (p
> 0 && p
!= offset
) {
2720 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, p
, &o
);
2725 p
= le64toh(o
->entry_array
.next_entry_array_offset
);
2728 /* If we can't find the previous entry array in the entry array chain, we're likely dealing with a
2729 * corrupted journal file. */
2739 static int generic_array_get(
2741 uint64_t first
, /* The offset of the first entry array object in the chain. */
2742 uint64_t i
, /* The index of the target object counted from the beginning of the entry array chain. */
2743 direction_t direction
,
2744 Object
**ret_object
, /* The found object. */
2745 uint64_t *ret_offset
) { /* The offset of the found object. */
2747 uint64_t a
, t
= 0, k
;
2754 /* FIXME: fix return value assignment on success. */
2758 /* Try the chain cache first */
2759 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2760 if (ci
&& i
> ci
->total
) {
2767 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2768 if (IN_SET(r
, -EBADMSG
, -EADDRNOTAVAIL
)) {
2769 /* If there's corruption and we're going downwards, let's pretend we reached the
2770 * final entry in the entry array chain. */
2772 if (direction
== DIRECTION_DOWN
)
2775 /* If there's corruption and we're going upwards, move back to the previous entry
2776 * array and start iterating entries from there. */
2784 k
= journal_file_entry_array_n_items(f
, o
);
2791 /* The index is larger than the number of elements in the array. Let's move to the next array. */
2794 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
2797 /* If we've found the right location, now look for the first non-corrupt entry object (in the right
2801 if (i
== UINT64_MAX
) {
2802 r
= bump_entry_array(f
, o
, a
, first
, direction
, &a
);
2806 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2810 k
= journal_file_entry_array_n_items(f
, o
);
2814 if (direction
== DIRECTION_DOWN
)
2817 /* We moved to the previous array. The total must be decreased. */
2819 return -EBADMSG
; /* chain cache is broken ? */
2829 p
= journal_file_entry_array_item(f
, o
, i
);
2831 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, ret_object
);
2833 /* Let's cache this item for the next invocation */
2834 chain_cache_put(f
->chain_cache
, ci
, first
, a
, journal_file_entry_array_item(f
, o
, 0), t
, i
);
2841 if (!IN_SET(r
, -EADDRNOTAVAIL
, -EBADMSG
))
2844 /* OK, so this entry is borked. Most likely some entry didn't get synced to
2845 * disk properly, let's see if the next one might work for us instead. */
2846 log_debug_errno(r
, "Entry item %" PRIu64
" is bad, skipping over it.", i
);
2848 } while (bump_array_index(&i
, direction
, k
) > 0);
2850 /* All entries tried in the above do-while loop are broken. Let's move to the next (or previous) array. */
2852 if (direction
== DIRECTION_DOWN
)
2853 /* We are going to the next array, the total must be incremented. */
2868 static int generic_array_bisect_one(
2870 uint64_t a
, /* offset of entry array object. */
2871 uint64_t i
, /* index of the entry item we will test. */
2873 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2874 direction_t direction
,
2877 uint64_t *ret_offset
) {
2884 assert(test_object
);
2888 assert(i
<= *right
);
2890 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2894 p
= journal_file_entry_array_item(f
, array
, i
);
2898 r
= test_object(f
, p
, needle
);
2899 if (IN_SET(r
, -EBADMSG
, -EADDRNOTAVAIL
)) {
2900 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short.");
2902 return -ENOANO
; /* recognizable error */
2907 if (r
== TEST_FOUND
)
2908 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2910 if (r
== TEST_RIGHT
)
2921 static int generic_array_bisect(
2926 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2927 direction_t direction
,
2928 Object
**ret_object
,
2929 uint64_t *ret_offset
,
2930 uint64_t *ret_idx
) {
2932 /* Given an entry array chain, this function finds the object "closest" to the given needle in the
2933 * chain, taking into account the provided direction. A function can be provided to determine how
2934 * an object is matched against the given needle.
2936 * Given a journal file, the offset of an object and the needle, the test_object() function should
2937 * return TEST_LEFT if the needle is located earlier in the entry array chain, TEST_LEFT if the
2938 * needle is located later in the entry array chain and TEST_FOUND if the object matches the needle.
2939 * If test_object() returns TEST_FOUND for a specific object, that object's information will be used
2940 * to populate the return values of this function. If test_object() never returns TEST_FOUND, the
2941 * return values are populated with the details of one of the objects closest to the needle. If the
2942 * direction is DIRECTION_UP, the earlier object is used. Otherwise, the later object is used.
2945 uint64_t a
, p
, t
= 0, i
= 0, last_p
= 0, last_index
= UINT64_MAX
;
2946 bool subtract_one
= false;
2952 assert(test_object
);
2954 /* Start with the first array in the chain */
2957 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2958 if (ci
&& n
> ci
->total
&& ci
->begin
!= 0) {
2959 /* Ah, we have iterated this bisection array chain previously! Let's see if we can skip ahead
2960 * in the chain, as far as the last time. But we can't jump backwards in the chain, so let's
2961 * check that first. */
2963 r
= test_object(f
, ci
->begin
, needle
);
2967 if (r
== TEST_LEFT
) {
2968 /* OK, what we are looking for is right of the begin of this EntryArray, so let's
2969 * jump straight to previously cached array in the chain */
2974 last_index
= ci
->last_index
;
2979 uint64_t left
= 0, right
, k
, lp
;
2981 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2985 k
= journal_file_entry_array_n_items(f
, array
);
2991 r
= generic_array_bisect_one(f
, a
, right
, needle
, test_object
, direction
, &left
, &right
, &lp
);
2999 if (r
== TEST_RIGHT
) {
3000 /* If we cached the last index we looked at, let's try to not to jump too wildly
3001 * around and see if we can limit the range to look at early to the immediate
3002 * neighbors of the last index we looked at. */
3004 if (last_index
> 0 && last_index
- 1 < right
) {
3005 r
= generic_array_bisect_one(f
, a
, last_index
- 1, needle
, test_object
, direction
, &left
, &right
, NULL
);
3006 if (r
< 0 && r
!= -ENOANO
)
3010 if (last_index
< right
) {
3011 r
= generic_array_bisect_one(f
, a
, last_index
+ 1, needle
, test_object
, direction
, &left
, &right
, NULL
);
3012 if (r
< 0 && r
!= -ENOANO
)
3017 if (left
== right
) {
3018 if (direction
== DIRECTION_UP
)
3019 subtract_one
= true;
3025 assert(left
< right
);
3026 i
= (left
+ right
) / 2;
3028 r
= generic_array_bisect_one(f
, a
, i
, needle
, test_object
, direction
, &left
, &right
, NULL
);
3029 if (r
< 0 && r
!= -ENOANO
)
3035 if (direction
== DIRECTION_UP
) {
3037 subtract_one
= true;
3048 last_index
= UINT64_MAX
;
3049 a
= le64toh(array
->entry_array
.next_entry_array_offset
);
3055 if (subtract_one
&& t
== 0 && i
== 0)
3058 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
3062 p
= journal_file_entry_array_item(f
, array
, 0);
3066 /* Let's cache this item for the next invocation */
3067 chain_cache_put(f
->chain_cache
, ci
, first
, a
, p
, t
, subtract_one
? (i
> 0 ? i
-1 : UINT64_MAX
) : i
);
3069 if (subtract_one
&& i
== 0)
3071 else if (subtract_one
)
3072 p
= journal_file_entry_array_item(f
, array
, i
- 1);
3074 p
= journal_file_entry_array_item(f
, array
, i
);
3077 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, ret_object
);
3086 *ret_idx
= t
+ i
+ (subtract_one
? -1 : 0);
3091 static int generic_array_bisect_plus_one(
3097 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
3098 direction_t direction
,
3099 Object
**ret_object
,
3100 uint64_t *ret_offset
) {
3105 assert(test_object
);
3110 /* This bisects the array in object 'first', but first checks an extra. */
3111 r
= test_object(f
, extra
, needle
);
3115 if (direction
== DIRECTION_DOWN
) {
3116 /* If we are going downwards, then we need to return the first object that passes the test.
3117 * When there is no object that passes the test, we need to return the first object that
3118 * test_object() returns TEST_RIGHT for. */
3120 TEST_FOUND
, /* The 'extra' object passes the test. Hence, this is the first
3121 * object that passes the test. */
3122 TEST_RIGHT
)) /* The 'extra' object is the first object that test_object() returns
3123 * TEST_RIGHT for, and no object exists even in the chained arrays
3124 * that passes the test. */
3125 goto use_extra
; /* The 'extra' object is exactly the one we are looking for. It is
3126 * not necessary to bisect the chained arrays. */
3128 /* Otherwise, the 'extra' object is not the one we are looking for. Search in the arrays. */
3131 /* If we are going upwards, then we need to return the last object that passes the test.
3132 * When there is no object that passes the test, we need to return the the last object that
3133 * test_object() returns TEST_LEFT for. */
3134 if (r
== TEST_RIGHT
)
3135 return 0; /* Not only the 'extra' object, but also all objects in the chained arrays
3136 * will never get TEST_FOUND or TEST_LEFT. The object we are looking for
3137 * does not exist. */
3139 /* Even if the 'extra' object passes the test, there may be multiple objects in the arrays
3140 * that also pass the test. Hence, we need to bisect the arrays for finding the last matching
3144 r
= generic_array_bisect(f
, first
, n
-1, needle
, test_object
, direction
, ret_object
, ret_offset
, NULL
);
3146 return r
; /* When > 0, the found object is the first (or last, when DIRECTION_UP) object.
3147 * Hence, return the found object now. */
3149 /* No matching object found in the chained arrays.
3150 * DIRECTION_DOWN : the 'extra' object neither matches the condition. There is no matching object.
3151 * DIRECTION_UP : the 'extra' object matches the condition. So, return it. */
3152 if (direction
== DIRECTION_DOWN
)
3157 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, ret_object
);
3163 *ret_offset
= extra
;
3168 static int test_object_offset(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
3174 else if (p
< needle
)
3180 int journal_file_move_to_entry_by_offset(
3183 direction_t direction
,
3184 Object
**ret_object
,
3185 uint64_t *ret_offset
) {
3190 return generic_array_bisect(
3192 le64toh(f
->header
->entry_array_offset
),
3193 le64toh(f
->header
->n_entries
),
3197 ret_object
, ret_offset
, NULL
);
3200 static int test_object_seqnum(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
3208 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
3212 sq
= le64toh(READ_NOW(o
->entry
.seqnum
));
3215 else if (sq
< needle
)
3221 int journal_file_move_to_entry_by_seqnum(
3224 direction_t direction
,
3225 Object
**ret_object
,
3226 uint64_t *ret_offset
) {
3231 return generic_array_bisect(
3233 le64toh(f
->header
->entry_array_offset
),
3234 le64toh(f
->header
->n_entries
),
3238 ret_object
, ret_offset
, NULL
);
3241 static int test_object_realtime(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
3249 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
3253 rt
= le64toh(READ_NOW(o
->entry
.realtime
));
3256 else if (rt
< needle
)
3262 int journal_file_move_to_entry_by_realtime(
3265 direction_t direction
,
3266 Object
**ret_object
,
3267 uint64_t *ret_offset
) {
3272 return generic_array_bisect(
3274 le64toh(f
->header
->entry_array_offset
),
3275 le64toh(f
->header
->n_entries
),
3277 test_object_realtime
,
3279 ret_object
, ret_offset
, NULL
);
3282 static int test_object_monotonic(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
3290 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
3294 m
= le64toh(READ_NOW(o
->entry
.monotonic
));
3297 else if (m
< needle
)
3303 static int find_data_object_by_boot_id(
3306 Object
**ret_object
,
3307 uint64_t *ret_offset
) {
3309 char t
[STRLEN("_BOOT_ID=") + 32 + 1] = "_BOOT_ID=";
3313 sd_id128_to_string(boot_id
, t
+ 9);
3314 return journal_file_find_data_object(f
, t
, sizeof(t
) - 1, ret_object
, ret_offset
);
3317 int journal_file_move_to_entry_by_monotonic(
3321 direction_t direction
,
3322 Object
**ret_object
,
3323 uint64_t *ret_offset
) {
3330 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, NULL
);
3334 return generic_array_bisect_plus_one(
3336 le64toh(o
->data
.entry_offset
),
3337 le64toh(o
->data
.entry_array_offset
),
3338 le64toh(o
->data
.n_entries
),
3340 test_object_monotonic
,
3342 ret_object
, ret_offset
);
3345 void journal_file_reset_location(JournalFile
*f
) {
3348 f
->location_type
= LOCATION_HEAD
;
3349 f
->current_offset
= 0;
3350 f
->current_seqnum
= 0;
3351 f
->current_realtime
= 0;
3352 f
->current_monotonic
= 0;
3353 zero(f
->current_boot_id
);
3354 f
->current_xor_hash
= 0;
3356 /* Also reset the previous reading direction. Otherwise, next_beyond_location() may wrongly handle we
3357 * already hit EOF. See issue #29216. */
3358 f
->last_direction
= _DIRECTION_INVALID
;
3361 void journal_file_save_location(JournalFile
*f
, Object
*o
, uint64_t offset
) {
3365 f
->location_type
= LOCATION_SEEK
;
3366 f
->current_offset
= offset
;
3367 f
->current_seqnum
= le64toh(o
->entry
.seqnum
);
3368 f
->current_realtime
= le64toh(o
->entry
.realtime
);
3369 f
->current_monotonic
= le64toh(o
->entry
.monotonic
);
3370 f
->current_boot_id
= o
->entry
.boot_id
;
3371 f
->current_xor_hash
= le64toh(o
->entry
.xor_hash
);
3374 static bool check_properly_ordered(uint64_t new_offset
, uint64_t old_offset
, direction_t direction
) {
3376 /* Consider it an error if any of the two offsets is uninitialized */
3377 if (old_offset
== 0 || new_offset
== 0)
3380 /* If we go down, the new offset must be larger than the old one. */
3381 return direction
== DIRECTION_DOWN
?
3382 new_offset
> old_offset
:
3383 new_offset
< old_offset
;
3386 int journal_file_next_entry(
3389 direction_t direction
,
3390 Object
**ret_object
,
3391 uint64_t *ret_offset
) {
3400 /* FIXME: fix return value assignment. */
3402 n
= le64toh(READ_NOW(f
->header
->n_entries
));
3406 /* When the input offset 'p' is zero, return the first (or last on DIRECTION_UP) entry. */
3408 return generic_array_get(f
,
3409 le64toh(f
->header
->entry_array_offset
),
3410 direction
== DIRECTION_DOWN
? 0 : n
- 1,
3412 ret_object
, ret_offset
);
3414 /* Otherwise, first find the nearest entry object. */
3415 r
= generic_array_bisect(f
,
3416 le64toh(f
->header
->entry_array_offset
),
3417 le64toh(f
->header
->n_entries
),
3421 ret_object
? &o
: NULL
, &q
, &i
);
3425 assert(direction
== DIRECTION_DOWN
? p
<= q
: q
<= p
);
3427 /* If the input offset 'p' points to an entry object, generic_array_bisect() should provides
3428 * the same offset, and the index needs to be shifted. Otherwise, use the found object as is,
3429 * as it is the nearest entry object from the input offset 'p'. */
3434 r
= bump_array_index(&i
, direction
, n
);
3438 /* And jump to it */
3439 r
= generic_array_get(f
, le64toh(f
->header
->entry_array_offset
), i
, direction
, ret_object
? &o
: NULL
, &q
);
3443 /* Ensure our array is properly ordered. */
3444 if (!check_properly_ordered(q
, p
, direction
))
3445 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
3446 "%s: entry array not properly ordered at entry index %" PRIu64
,
3457 int journal_file_move_to_entry_for_data(
3460 direction_t direction
,
3461 Object
**ret_object
,
3462 uint64_t *ret_offset
) {
3464 uint64_t extra
, first
, n
;
3469 assert(d
->object
.type
== OBJECT_DATA
);
3470 assert(IN_SET(direction
, DIRECTION_DOWN
, DIRECTION_UP
));
3472 /* FIXME: fix return value assignment. */
3474 /* This returns the first (when the direction is down, otherwise the last) entry linked to the
3475 * specified data object. */
3477 n
= le64toh(d
->data
.n_entries
);
3480 n
--; /* n_entries is the number of entries linked to the data object, including the 'extra' entry. */
3482 extra
= le64toh(d
->data
.entry_offset
);
3483 first
= le64toh(d
->data
.entry_array_offset
);
3485 if (direction
== DIRECTION_DOWN
&& extra
> 0) {
3486 /* When we are going downwards, first try to read the extra entry. */
3487 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, ret_object
);
3490 if (!IN_SET(r
, -EADDRNOTAVAIL
, -EBADMSG
))
3495 /* DIRECTION_DOWN : The extra entry is broken, falling back to the entries in the array.
3496 * DIRECTION_UP : Try to find a valid entry in the array from the tail. */
3497 r
= generic_array_get(f
,
3499 direction
== DIRECTION_DOWN
? 0 : n
- 1,
3501 ret_object
, ret_offset
);
3502 if (!IN_SET(r
, 0, -EADDRNOTAVAIL
, -EBADMSG
))
3503 return r
; /* found or critical error. */
3506 if (direction
== DIRECTION_UP
&& extra
> 0) {
3507 /* No valid entry exists in the chained array, falling back to the extra entry. */
3508 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, ret_object
);
3517 *ret_offset
= extra
;
3522 int journal_file_move_to_entry_by_offset_for_data(
3526 direction_t direction
,
3527 Object
**ret
, uint64_t *ret_offset
) {
3531 assert(d
->object
.type
== OBJECT_DATA
);
3533 return generic_array_bisect_plus_one(
3535 le64toh(d
->data
.entry_offset
),
3536 le64toh(d
->data
.entry_array_offset
),
3537 le64toh(d
->data
.n_entries
),
3544 int journal_file_move_to_entry_by_monotonic_for_data(
3549 direction_t direction
,
3550 Object
**ret_object
,
3551 uint64_t *ret_offset
) {
3553 uint64_t z
, entry_offset
, entry_array_offset
, n_entries
;
3559 assert(d
->object
.type
== OBJECT_DATA
);
3561 /* Save all the required data before the data object gets invalidated. */
3562 entry_offset
= le64toh(READ_NOW(d
->data
.entry_offset
));
3563 entry_array_offset
= le64toh(READ_NOW(d
->data
.entry_array_offset
));
3564 n_entries
= le64toh(READ_NOW(d
->data
.n_entries
));
3566 /* First, seek by time */
3567 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, NULL
);
3571 r
= generic_array_bisect_plus_one(f
,
3572 le64toh(o
->data
.entry_offset
),
3573 le64toh(o
->data
.entry_array_offset
),
3574 le64toh(o
->data
.n_entries
),
3576 test_object_monotonic
,
3582 /* And now, continue seeking until we find an entry that exists in both bisection arrays. */
3586 /* The journal entry found by the above bisect_plus_one() may not have the specified data,
3587 * that is, it may not be linked in the data object. So, we need to check that. */
3589 r
= generic_array_bisect_plus_one(f
,
3596 ret_object
? &entry
: NULL
, &p
);
3600 break; /* The journal entry has the specified data. Yay! */
3602 /* If the entry does not have the data, then move to the next (or previous, depends on the
3603 * 'direction') entry linked to the data object. But, the next entry may be in another boot.
3604 * So, we need to check that the entry has the matching boot ID. */
3606 r
= generic_array_bisect_plus_one(f
,
3607 le64toh(o
->data
.entry_offset
),
3608 le64toh(o
->data
.entry_array_offset
),
3609 le64toh(o
->data
.n_entries
),
3613 ret_object
? &entry
: NULL
, &z
);
3617 break; /* The journal entry has the specified boot ID. Yay! */
3619 /* If not, let's try to the next entry... */
3623 *ret_object
= entry
;
3629 int journal_file_move_to_entry_by_seqnum_for_data(
3633 direction_t direction
,
3634 Object
**ret_object
,
3635 uint64_t *ret_offset
) {
3639 assert(d
->object
.type
== OBJECT_DATA
);
3641 return generic_array_bisect_plus_one(
3643 le64toh(d
->data
.entry_offset
),
3644 le64toh(d
->data
.entry_array_offset
),
3645 le64toh(d
->data
.n_entries
),
3649 ret_object
, ret_offset
);
3652 int journal_file_move_to_entry_by_realtime_for_data(
3656 direction_t direction
,
3657 Object
**ret
, uint64_t *ret_offset
) {
3661 assert(d
->object
.type
== OBJECT_DATA
);
3663 return generic_array_bisect_plus_one(
3665 le64toh(d
->data
.entry_offset
),
3666 le64toh(d
->data
.entry_array_offset
),
3667 le64toh(d
->data
.n_entries
),
3669 test_object_realtime
,
3674 void journal_file_dump(JournalFile
*f
) {
3682 journal_file_print_header(f
);
3684 p
= le64toh(READ_NOW(f
->header
->header_size
));
3689 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &o
);
3693 s
= journal_object_type_to_string(o
->object
.type
);
3695 switch (o
->object
.type
) {
3700 printf("Type: %s seqnum=%"PRIu64
" monotonic=%"PRIu64
" realtime=%"PRIu64
"\n",
3702 le64toh(o
->entry
.seqnum
),
3703 le64toh(o
->entry
.monotonic
),
3704 le64toh(o
->entry
.realtime
));
3710 printf("Type: %s seqnum=%"PRIu64
" epoch=%"PRIu64
"\n",
3712 le64toh(o
->tag
.seqnum
),
3713 le64toh(o
->tag
.epoch
));
3718 printf("Type: %s \n", s
);
3720 printf("Type: unknown (%i)", o
->object
.type
);
3725 c
= COMPRESSION_FROM_OBJECT(o
);
3726 if (c
> COMPRESSION_NONE
)
3727 printf("Flags: %s\n",
3728 compression_to_string(c
));
3730 if (p
== le64toh(f
->header
->tail_object_offset
))
3733 p
+= ALIGN64(le64toh(o
->object
.size
));
3738 log_error("File corrupt");
3741 /* Note: the lifetime of the compound literal is the immediately surrounding block. */
3742 #define FORMAT_TIMESTAMP_SAFE(t) (FORMAT_TIMESTAMP(t) ?: " --- ")
3744 void journal_file_print_header(JournalFile
*f
) {
3750 printf("File path: %s\n"
3754 "Sequential number ID: %s\n"
3756 "Compatible flags:%s%s%s\n"
3757 "Incompatible flags:%s%s%s%s%s%s\n"
3758 "Header size: %"PRIu64
"\n"
3759 "Arena size: %"PRIu64
"\n"
3760 "Data hash table size: %"PRIu64
"\n"
3761 "Field hash table size: %"PRIu64
"\n"
3762 "Rotate suggested: %s\n"
3763 "Head sequential number: %"PRIu64
" (%"PRIx64
")\n"
3764 "Tail sequential number: %"PRIu64
" (%"PRIx64
")\n"
3765 "Head realtime timestamp: %s (%"PRIx64
")\n"
3766 "Tail realtime timestamp: %s (%"PRIx64
")\n"
3767 "Tail monotonic timestamp: %s (%"PRIx64
")\n"
3768 "Objects: %"PRIu64
"\n"
3769 "Entry objects: %"PRIu64
"\n",
3771 SD_ID128_TO_STRING(f
->header
->file_id
),
3772 SD_ID128_TO_STRING(f
->header
->machine_id
),
3773 SD_ID128_TO_STRING(f
->header
->tail_entry_boot_id
),
3774 SD_ID128_TO_STRING(f
->header
->seqnum_id
),
3775 f
->header
->state
== STATE_OFFLINE
? "OFFLINE" :
3776 f
->header
->state
== STATE_ONLINE
? "ONLINE" :
3777 f
->header
->state
== STATE_ARCHIVED
? "ARCHIVED" : "UNKNOWN",
3778 JOURNAL_HEADER_SEALED(f
->header
) ? " SEALED" : "",
3779 JOURNAL_HEADER_TAIL_ENTRY_BOOT_ID(f
->header
) ? " TAIL_ENTRY_BOOT_ID" : "",
3780 (le32toh(f
->header
->compatible_flags
) & ~HEADER_COMPATIBLE_ANY
) ? " ???" : "",
3781 JOURNAL_HEADER_COMPRESSED_XZ(f
->header
) ? " COMPRESSED-XZ" : "",
3782 JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
) ? " COMPRESSED-LZ4" : "",
3783 JOURNAL_HEADER_COMPRESSED_ZSTD(f
->header
) ? " COMPRESSED-ZSTD" : "",
3784 JOURNAL_HEADER_KEYED_HASH(f
->header
) ? " KEYED-HASH" : "",
3785 JOURNAL_HEADER_COMPACT(f
->header
) ? " COMPACT" : "",
3786 (le32toh(f
->header
->incompatible_flags
) & ~HEADER_INCOMPATIBLE_ANY
) ? " ???" : "",
3787 le64toh(f
->header
->header_size
),
3788 le64toh(f
->header
->arena_size
),
3789 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3790 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
),
3791 yes_no(journal_file_rotate_suggested(f
, 0, LOG_DEBUG
)),
3792 le64toh(f
->header
->head_entry_seqnum
), le64toh(f
->header
->head_entry_seqnum
),
3793 le64toh(f
->header
->tail_entry_seqnum
), le64toh(f
->header
->tail_entry_seqnum
),
3794 FORMAT_TIMESTAMP_SAFE(le64toh(f
->header
->head_entry_realtime
)), le64toh(f
->header
->head_entry_realtime
),
3795 FORMAT_TIMESTAMP_SAFE(le64toh(f
->header
->tail_entry_realtime
)), le64toh(f
->header
->tail_entry_realtime
),
3796 FORMAT_TIMESPAN(le64toh(f
->header
->tail_entry_monotonic
), USEC_PER_MSEC
), le64toh(f
->header
->tail_entry_monotonic
),
3797 le64toh(f
->header
->n_objects
),
3798 le64toh(f
->header
->n_entries
));
3800 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3801 printf("Data objects: %"PRIu64
"\n"
3802 "Data hash table fill: %.1f%%\n",
3803 le64toh(f
->header
->n_data
),
3804 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))));
3806 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3807 printf("Field objects: %"PRIu64
"\n"
3808 "Field hash table fill: %.1f%%\n",
3809 le64toh(f
->header
->n_fields
),
3810 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))));
3812 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
))
3813 printf("Tag objects: %"PRIu64
"\n",
3814 le64toh(f
->header
->n_tags
));
3815 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
3816 printf("Entry array objects: %"PRIu64
"\n",
3817 le64toh(f
->header
->n_entry_arrays
));
3819 if (JOURNAL_HEADER_CONTAINS(f
->header
, field_hash_chain_depth
))
3820 printf("Deepest field hash chain: %" PRIu64
"\n",
3821 f
->header
->field_hash_chain_depth
);
3823 if (JOURNAL_HEADER_CONTAINS(f
->header
, data_hash_chain_depth
))
3824 printf("Deepest data hash chain: %" PRIu64
"\n",
3825 f
->header
->data_hash_chain_depth
);
3827 if (fstat(f
->fd
, &st
) >= 0)
3828 printf("Disk usage: %s\n", FORMAT_BYTES((uint64_t) st
.st_blocks
* 512ULL));
3831 static int journal_file_warn_btrfs(JournalFile
*f
) {
3837 /* Before we write anything, check if the COW logic is turned
3838 * off on btrfs. Given our write pattern that is quite
3839 * unfriendly to COW file systems this should greatly improve
3840 * performance on COW file systems, such as btrfs, at the
3841 * expense of data integrity features (which shouldn't be too
3842 * bad, given that we do our own checksumming). */
3844 r
= fd_is_fs_type(f
->fd
, BTRFS_SUPER_MAGIC
);
3846 return log_ratelimit_warning_errno(r
, JOURNAL_LOG_RATELIMIT
, "Failed to determine if journal is on btrfs: %m");
3850 r
= read_attr_fd(f
->fd
, &attrs
);
3852 return log_ratelimit_warning_errno(r
, JOURNAL_LOG_RATELIMIT
, "Failed to read file attributes: %m");
3854 if (attrs
& FS_NOCOW_FL
) {
3855 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3859 log_ratelimit_notice(JOURNAL_LOG_RATELIMIT
,
3860 "Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3861 "This is likely to slow down journal access substantially, please consider turning "
3862 "off the copy-on-write file attribute on the journal directory, using chattr +C.",
3868 static void journal_default_metrics(JournalMetrics
*m
, int fd
, bool compact
) {
3870 uint64_t fs_size
= 0;
3875 if (fstatvfs(fd
, &ss
) >= 0)
3876 fs_size
= u64_multiply_safe(ss
.f_frsize
, ss
.f_blocks
);
3878 log_debug_errno(errno
, "Failed to determine disk size: %m");
3880 if (m
->max_use
== UINT64_MAX
) {
3883 m
->max_use
= CLAMP(PAGE_ALIGN_U64(fs_size
/ 10), /* 10% of file system size */
3884 MAX_USE_LOWER
, MAX_USE_UPPER
);
3886 m
->max_use
= MAX_USE_LOWER
;
3888 m
->max_use
= PAGE_ALIGN_U64(m
->max_use
);
3890 if (m
->max_use
!= 0 && m
->max_use
< JOURNAL_FILE_SIZE_MIN
*2)
3891 m
->max_use
= JOURNAL_FILE_SIZE_MIN
*2;
3894 if (m
->min_use
== UINT64_MAX
) {
3896 m
->min_use
= CLAMP(PAGE_ALIGN_U64(fs_size
/ 50), /* 2% of file system size */
3897 MIN_USE_LOW
, MIN_USE_HIGH
);
3899 m
->min_use
= MIN_USE_LOW
;
3902 if (m
->min_use
> m
->max_use
)
3903 m
->min_use
= m
->max_use
;
3905 if (m
->max_size
== UINT64_MAX
)
3906 m
->max_size
= MIN(PAGE_ALIGN_U64(m
->max_use
/ 8), /* 8 chunks */
3909 m
->max_size
= PAGE_ALIGN_U64(m
->max_size
);
3911 if (compact
&& m
->max_size
> JOURNAL_COMPACT_SIZE_MAX
)
3912 m
->max_size
= JOURNAL_COMPACT_SIZE_MAX
;
3914 if (m
->max_size
!= 0) {
3915 if (m
->max_size
< JOURNAL_FILE_SIZE_MIN
)
3916 m
->max_size
= JOURNAL_FILE_SIZE_MIN
;
3918 if (m
->max_use
!= 0 && m
->max_size
*2 > m
->max_use
)
3919 m
->max_use
= m
->max_size
*2;
3922 if (m
->min_size
== UINT64_MAX
)
3923 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3925 m
->min_size
= CLAMP(PAGE_ALIGN_U64(m
->min_size
),
3926 JOURNAL_FILE_SIZE_MIN
,
3927 m
->max_size
?: UINT64_MAX
);
3929 if (m
->keep_free
== UINT64_MAX
) {
3931 m
->keep_free
= MIN(PAGE_ALIGN_U64(fs_size
/ 20), /* 5% of file system size */
3934 m
->keep_free
= DEFAULT_KEEP_FREE
;
3937 if (m
->n_max_files
== UINT64_MAX
)
3938 m
->n_max_files
= DEFAULT_N_MAX_FILES
;
3940 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64
,
3941 FORMAT_BYTES(m
->min_use
),
3942 FORMAT_BYTES(m
->max_use
),
3943 FORMAT_BYTES(m
->max_size
),
3944 FORMAT_BYTES(m
->min_size
),
3945 FORMAT_BYTES(m
->keep_free
),
3949 int journal_file_open(
3953 JournalFileFlags file_flags
,
3955 uint64_t compress_threshold_bytes
,
3956 JournalMetrics
*metrics
,
3957 MMapCache
*mmap_cache
,
3958 JournalFile
*template,
3959 JournalFile
**ret
) {
3961 bool newly_created
= false;
3966 assert(fd
>= 0 || fname
);
3967 assert(file_flags
>= 0);
3968 assert(file_flags
<= _JOURNAL_FILE_FLAGS_MAX
);
3972 if (!IN_SET((open_flags
& O_ACCMODE
), O_RDONLY
, O_RDWR
))
3975 if ((open_flags
& O_ACCMODE
) == O_RDONLY
&& FLAGS_SET(open_flags
, O_CREAT
))
3978 if (fname
&& (open_flags
& O_CREAT
) && !endswith(fname
, ".journal"))
3981 f
= new(JournalFile
, 1);
3985 *f
= (JournalFile
) {
3988 .open_flags
= open_flags
,
3989 .compress_threshold_bytes
= compress_threshold_bytes
== UINT64_MAX
?
3990 DEFAULT_COMPRESS_THRESHOLD
:
3991 MAX(MIN_COMPRESS_THRESHOLD
, compress_threshold_bytes
),
3992 .strict_order
= FLAGS_SET(file_flags
, JOURNAL_STRICT_ORDER
),
3993 .newest_boot_id_prioq_idx
= PRIOQ_IDX_NULL
,
3994 .last_direction
= _DIRECTION_INVALID
,
3998 f
->path
= strdup(fname
);
4006 /* If we don't know the path, fill in something explanatory and vaguely useful */
4007 if (asprintf(&f
->path
, "/proc/self/%i", fd
) < 0) {
4013 f
->chain_cache
= ordered_hashmap_new(&uint64_hash_ops
);
4014 if (!f
->chain_cache
) {
4020 /* We pass O_NONBLOCK here, so that in case somebody pointed us to some character device node or FIFO
4021 * or so, we likely fail quickly than block for long. For regular files O_NONBLOCK has no effect, hence
4022 * it doesn't hurt in that case. */
4024 f
->fd
= openat_report_new(AT_FDCWD
, f
->path
, f
->open_flags
|O_CLOEXEC
|O_NONBLOCK
, f
->mode
, &newly_created
);
4030 /* fds we opened here by us should also be closed by us. */
4033 r
= fd_nonblock(f
->fd
, false);
4037 if (!newly_created
) {
4038 r
= journal_file_fstat(f
);
4043 r
= journal_file_fstat(f
);
4047 /* If we just got the fd passed in, we don't really know if we created the file anew */
4048 newly_created
= f
->last_stat
.st_size
== 0 && journal_file_writable(f
);
4051 r
= mmap_cache_add_fd(mmap_cache
, f
->fd
, mmap_prot_from_open_flags(open_flags
), &f
->cache_fd
);
4055 if (newly_created
) {
4056 (void) journal_file_warn_btrfs(f
);
4058 /* Let's attach the creation time to the journal file, so that the vacuuming code knows the age of this
4059 * file even if the file might end up corrupted one day... Ideally we'd just use the creation time many
4060 * file systems maintain for each file, but the API to query this is very new, hence let's emulate this
4061 * via extended attributes. If extended attributes are not supported we'll just skip this, and rely
4062 * solely on mtime/atime/ctime of the file. */
4063 (void) fd_setcrtime(f
->fd
, 0);
4065 r
= journal_file_init_header(f
, file_flags
, template);
4069 r
= journal_file_fstat(f
);
4074 if (f
->last_stat
.st_size
< (off_t
) HEADER_SIZE_MIN
) {
4079 r
= mmap_cache_fd_get(f
->cache_fd
, MMAP_CACHE_CATEGORY_HEADER
, true, 0, PAGE_ALIGN(sizeof(Header
)), &f
->last_stat
, &h
);
4081 /* Some file systems (jffs2 or p9fs) don't support mmap() properly (or only read-only
4082 * mmap()), and return EINVAL in that case. Let's propagate that as a more recognizable error
4092 if (!newly_created
) {
4093 r
= journal_file_verify_header(f
);
4099 if (!newly_created
&& journal_file_writable(f
) && JOURNAL_HEADER_SEALED(f
->header
)) {
4100 r
= journal_file_fss_load(f
);
4106 if (journal_file_writable(f
)) {
4108 journal_default_metrics(metrics
, f
->fd
, JOURNAL_HEADER_COMPACT(f
->header
));
4109 f
->metrics
= *metrics
;
4110 } else if (template)
4111 f
->metrics
= template->metrics
;
4113 r
= journal_file_refresh_header(f
);
4119 r
= journal_file_hmac_setup(f
);
4124 if (newly_created
) {
4125 r
= journal_file_setup_field_hash_table(f
);
4129 r
= journal_file_setup_data_hash_table(f
);
4134 r
= journal_file_append_first_tag(f
);
4140 if (mmap_cache_fd_got_sigbus(f
->cache_fd
)) {
4145 if (template && template->post_change_timer
) {
4146 r
= journal_file_enable_post_change_timer(
4148 sd_event_source_get_event(template->post_change_timer
),
4149 template->post_change_timer_period
);
4155 /* The file is opened now successfully, thus we take possession of any passed in fd. */
4158 if (DEBUG_LOGGING
) {
4159 static int last_seal
= -1, last_keyed_hash
= -1;
4160 static Compression last_compression
= _COMPRESSION_INVALID
;
4161 static uint64_t last_bytes
= UINT64_MAX
;
4163 if (last_seal
!= JOURNAL_HEADER_SEALED(f
->header
) ||
4164 last_keyed_hash
!= JOURNAL_HEADER_KEYED_HASH(f
->header
) ||
4165 last_compression
!= JOURNAL_FILE_COMPRESSION(f
) ||
4166 last_bytes
!= f
->compress_threshold_bytes
) {
4168 log_debug("Journal effective settings seal=%s keyed_hash=%s compress=%s compress_threshold_bytes=%s",
4169 yes_no(JOURNAL_HEADER_SEALED(f
->header
)), yes_no(JOURNAL_HEADER_KEYED_HASH(f
->header
)),
4170 compression_to_string(JOURNAL_FILE_COMPRESSION(f
)), FORMAT_BYTES(f
->compress_threshold_bytes
));
4171 last_seal
= JOURNAL_HEADER_SEALED(f
->header
);
4172 last_keyed_hash
= JOURNAL_HEADER_KEYED_HASH(f
->header
);
4173 last_compression
= JOURNAL_FILE_COMPRESSION(f
);
4174 last_bytes
= f
->compress_threshold_bytes
;
4182 if (f
->cache_fd
&& mmap_cache_fd_got_sigbus(f
->cache_fd
))
4185 (void) journal_file_close(f
);
4187 if (newly_created
&& fd
< 0)
4188 (void) unlink(fname
);
4193 int journal_file_parse_uid_from_filename(const char *path
, uid_t
*ret_uid
) {
4194 _cleanup_free_
char *buf
= NULL
, *p
= NULL
;
4195 const char *a
, *b
, *at
;
4198 /* This helper returns -EREMOTE when the filename doesn't match user online/offline journal
4199 * pattern. Hence it currently doesn't parse archived or disposed user journals. */
4204 r
= path_extract_filename(path
, &p
);
4207 if (r
== O_DIRECTORY
)
4210 a
= startswith(p
, "user-");
4213 b
= endswith(p
, ".journal");
4217 at
= strchr(a
, '@');
4221 buf
= strndup(a
, b
-a
);
4225 return parse_uid(buf
, ret_uid
);
4228 int journal_file_archive(JournalFile
*f
, char **ret_previous_path
) {
4229 _cleanup_free_
char *p
= NULL
;
4233 if (!journal_file_writable(f
))
4236 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
4237 * rotation, since we don't know the actual path, and couldn't rename the file hence. */
4238 if (path_startswith(f
->path
, "/proc/self/fd"))
4241 if (!endswith(f
->path
, ".journal"))
4244 if (asprintf(&p
, "%.*s@" SD_ID128_FORMAT_STR
"-%016"PRIx64
"-%016"PRIx64
".journal",
4245 (int) strlen(f
->path
) - 8, f
->path
,
4246 SD_ID128_FORMAT_VAL(f
->header
->seqnum_id
),
4247 le64toh(f
->header
->head_entry_seqnum
),
4248 le64toh(f
->header
->head_entry_realtime
)) < 0)
4251 /* Try to rename the file to the archived version. If the file already was deleted, we'll get ENOENT, let's
4252 * ignore that case. */
4253 if (rename(f
->path
, p
) < 0 && errno
!= ENOENT
)
4256 /* Sync the rename to disk */
4257 (void) fsync_directory_of_file(f
->fd
);
4259 if (ret_previous_path
)
4260 *ret_previous_path
= f
->path
;
4264 f
->path
= TAKE_PTR(p
);
4266 /* Set as archive so offlining commits w/state=STATE_ARCHIVED. Previously we would set old_file->header->state
4267 * to STATE_ARCHIVED directly here, but journal_file_set_offline() short-circuits when state != STATE_ONLINE,
4268 * which would result in the rotated journal never getting fsync() called before closing. Now we simply queue
4269 * the archive state by setting an archive bit, leaving the state as STATE_ONLINE so proper offlining
4276 int journal_file_dispose(int dir_fd
, const char *fname
) {
4277 _cleanup_free_
char *p
= NULL
;
4281 /* Renames a journal file to *.journal~, i.e. to mark it as corrupted or otherwise uncleanly shutdown. Note that
4282 * this is done without looking into the file or changing any of its contents. The idea is that this is called
4283 * whenever something is suspicious and we want to move the file away and make clear that it is not accessed
4284 * for writing anymore. */
4286 if (!endswith(fname
, ".journal"))
4289 if (asprintf(&p
, "%.*s@%016" PRIx64
"-%016" PRIx64
".journal~",
4290 (int) strlen(fname
) - 8, fname
,
4291 now(CLOCK_REALTIME
),
4295 if (renameat(dir_fd
, fname
, dir_fd
, p
) < 0)
4301 int journal_file_copy_entry(
4307 sd_id128_t
*seqnum_id
) {
4309 _cleanup_free_ EntryItem
*items_alloc
= NULL
;
4311 uint64_t n
, m
= 0, xor_hash
= 0;
4321 if (!journal_file_writable(to
))
4324 ts
= (dual_timestamp
) {
4325 .monotonic
= le64toh(o
->entry
.monotonic
),
4326 .realtime
= le64toh(o
->entry
.realtime
),
4328 boot_id
= o
->entry
.boot_id
;
4330 n
= journal_file_entry_n_items(from
, o
);
4334 if (n
< ALLOCA_MAX
/ sizeof(EntryItem
) / 2)
4335 items
= newa(EntryItem
, n
);
4337 items_alloc
= new(EntryItem
, n
);
4341 items
= items_alloc
;
4344 for (uint64_t i
= 0; i
< n
; i
++) {
4350 q
= journal_file_entry_item_object_offset(from
, o
, i
);
4351 r
= journal_file_data_payload(from
, NULL
, q
, NULL
, 0, 0, &data
, &l
);
4352 if (IN_SET(r
, -EADDRNOTAVAIL
, -EBADMSG
)) {
4353 log_debug_errno(r
, "Entry item %"PRIu64
" data object is bad, skipping over it: %m", i
);
4363 r
= journal_file_append_data(to
, data
, l
, &u
, &h
);
4367 if (JOURNAL_HEADER_KEYED_HASH(to
->header
))
4368 xor_hash
^= jenkins_hash64(data
, l
);
4370 xor_hash
^= le64toh(u
->data
.hash
);
4372 items
[m
++] = (EntryItem
) {
4374 .hash
= le64toh(u
->data
.hash
),
4381 r
= journal_file_append_entry_internal(
4385 &from
->header
->machine_id
,
4391 /* ret_object= */ NULL
,
4392 /* ret_offset= */ NULL
);
4394 if (mmap_cache_fd_got_sigbus(to
->cache_fd
))
4400 void journal_reset_metrics(JournalMetrics
*m
) {
4403 /* Set everything to "pick automatic values". */
4405 *m
= (JournalMetrics
) {
4406 .min_use
= UINT64_MAX
,
4407 .max_use
= UINT64_MAX
,
4408 .min_size
= UINT64_MAX
,
4409 .max_size
= UINT64_MAX
,
4410 .keep_free
= UINT64_MAX
,
4411 .n_max_files
= UINT64_MAX
,
4415 int journal_file_get_cutoff_realtime_usec(JournalFile
*f
, usec_t
*ret_from
, usec_t
*ret_to
) {
4418 assert(ret_from
|| ret_to
);
4421 if (f
->header
->head_entry_realtime
== 0)
4424 *ret_from
= le64toh(f
->header
->head_entry_realtime
);
4428 if (f
->header
->tail_entry_realtime
== 0)
4431 *ret_to
= le64toh(f
->header
->tail_entry_realtime
);
4437 int journal_file_get_cutoff_monotonic_usec(JournalFile
*f
, sd_id128_t boot_id
, usec_t
*ret_from
, usec_t
*ret_to
) {
4443 assert(ret_from
|| ret_to
);
4445 /* FIXME: fix return value assignment on success with 0. */
4447 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &p
);
4451 if (le64toh(o
->data
.n_entries
) <= 0)
4455 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, le64toh(o
->data
.entry_offset
), &o
);
4459 *ret_from
= le64toh(o
->entry
.monotonic
);
4463 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
4467 r
= journal_file_move_to_entry_for_data(f
, o
, DIRECTION_UP
, &o
, NULL
);
4471 *ret_to
= le64toh(o
->entry
.monotonic
);
4477 bool journal_file_rotate_suggested(JournalFile
*f
, usec_t max_file_usec
, int log_level
) {
4481 /* If we gained new header fields we gained new features,
4482 * hence suggest a rotation */
4483 if (le64toh(f
->header
->header_size
) < sizeof(Header
)) {
4484 log_ratelimit_full(log_level
, JOURNAL_LOG_RATELIMIT
,
4485 "%s uses an outdated header, suggesting rotation.", f
->path
);
4489 /* Let's check if the hash tables grew over a certain fill level (75%, borrowing this value from
4490 * Java's hash table implementation), and if so suggest a rotation. To calculate the fill level we
4491 * need the n_data field, which only exists in newer versions. */
4493 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
4494 if (le64toh(f
->header
->n_data
) * 4ULL > (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
4496 log_level
, JOURNAL_LOG_RATELIMIT
,
4497 "Data hash table of %s has a fill level at %.1f (%"PRIu64
" of %"PRIu64
" items, %"PRIu64
" file size, %"PRIu64
" bytes per hash table item), suggesting rotation.",
4499 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))),
4500 le64toh(f
->header
->n_data
),
4501 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
4502 (uint64_t) f
->last_stat
.st_size
,
4503 f
->last_stat
.st_size
/ le64toh(f
->header
->n_data
));
4507 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
4508 if (le64toh(f
->header
->n_fields
) * 4ULL > (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
4510 log_level
, JOURNAL_LOG_RATELIMIT
,
4511 "Field hash table of %s has a fill level at %.1f (%"PRIu64
" of %"PRIu64
" items), suggesting rotation.",
4513 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))),
4514 le64toh(f
->header
->n_fields
),
4515 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
));
4519 /* If there are too many hash collisions somebody is most likely playing games with us. Hence, if our
4520 * longest chain is longer than some threshold, let's suggest rotation. */
4521 if (JOURNAL_HEADER_CONTAINS(f
->header
, data_hash_chain_depth
) &&
4522 le64toh(f
->header
->data_hash_chain_depth
) > HASH_CHAIN_DEPTH_MAX
) {
4524 log_level
, JOURNAL_LOG_RATELIMIT
,
4525 "Data hash table of %s has deepest hash chain of length %" PRIu64
", suggesting rotation.",
4526 f
->path
, le64toh(f
->header
->data_hash_chain_depth
));
4530 if (JOURNAL_HEADER_CONTAINS(f
->header
, field_hash_chain_depth
) &&
4531 le64toh(f
->header
->field_hash_chain_depth
) > HASH_CHAIN_DEPTH_MAX
) {
4533 log_level
, JOURNAL_LOG_RATELIMIT
,
4534 "Field hash table of %s has deepest hash chain of length at %" PRIu64
", suggesting rotation.",
4535 f
->path
, le64toh(f
->header
->field_hash_chain_depth
));
4539 /* Are the data objects properly indexed by field objects? */
4540 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
4541 JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
4542 le64toh(f
->header
->n_data
) > 0 &&
4543 le64toh(f
->header
->n_fields
) == 0) {
4545 log_level
, JOURNAL_LOG_RATELIMIT
,
4546 "Data objects of %s are not indexed by field objects, suggesting rotation.",
4551 if (max_file_usec
> 0) {
4554 h
= le64toh(f
->header
->head_entry_realtime
);
4555 t
= now(CLOCK_REALTIME
);
4557 if (h
> 0 && t
> h
+ max_file_usec
) {
4559 log_level
, JOURNAL_LOG_RATELIMIT
,
4560 "Oldest entry in %s is older than the configured file retention duration (%s), suggesting rotation.",
4561 f
->path
, FORMAT_TIMESPAN(max_file_usec
, USEC_PER_SEC
));
4569 static const char * const journal_object_type_table
[] = {
4570 [OBJECT_UNUSED
] = "unused",
4571 [OBJECT_DATA
] = "data",
4572 [OBJECT_FIELD
] = "field",
4573 [OBJECT_ENTRY
] = "entry",
4574 [OBJECT_DATA_HASH_TABLE
] = "data hash table",
4575 [OBJECT_FIELD_HASH_TABLE
] = "field hash table",
4576 [OBJECT_ENTRY_ARRAY
] = "entry array",
4577 [OBJECT_TAG
] = "tag",
4580 DEFINE_STRING_TABLE_LOOKUP_TO_STRING(journal_object_type
, ObjectType
);