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 /* The mmap context to use for the header we pick as one above the last defined typed */
92 #define CONTEXT_HEADER _OBJECT_TYPE_MAX
94 /* Longest hash chain to rotate after */
95 #define HASH_CHAIN_DEPTH_MAX 100
98 # pragma GCC diagnostic ignored "-Waddress-of-packed-member"
101 static int mmap_prot_from_open_flags(int flags
) {
102 switch (flags
& O_ACCMODE
) {
108 return PROT_READ
|PROT_WRITE
;
110 assert_not_reached();
114 int journal_file_tail_end_by_pread(JournalFile
*f
, uint64_t *ret_offset
) {
122 /* Same as journal_file_tail_end_by_mmap() below, but operates with pread() to avoid the mmap cache
123 * (and thus is thread safe) */
125 p
= le64toh(f
->header
->tail_object_offset
);
127 p
= le64toh(f
->header
->header_size
);
132 r
= journal_file_read_object_header(f
, OBJECT_UNUSED
, p
, &tail
);
136 sz
= le64toh(tail
.object
.size
);
137 if (sz
> UINT64_MAX
- sizeof(uint64_t) + 1)
141 if (p
> UINT64_MAX
- sz
)
152 int journal_file_tail_end_by_mmap(JournalFile
*f
, uint64_t *ret_offset
) {
160 /* Same as journal_file_tail_end_by_pread() above, but operates with the usual mmap logic */
162 p
= le64toh(f
->header
->tail_object_offset
);
164 p
= le64toh(f
->header
->header_size
);
169 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &tail
);
173 sz
= le64toh(READ_NOW(tail
->object
.size
));
174 if (sz
> UINT64_MAX
- sizeof(uint64_t) + 1)
178 if (p
> UINT64_MAX
- sz
)
189 int journal_file_set_offline_thread_join(JournalFile
*f
) {
194 if (f
->offline_state
== OFFLINE_JOINED
)
197 r
= pthread_join(f
->offline_thread
, NULL
);
201 f
->offline_state
= OFFLINE_JOINED
;
203 if (mmap_cache_fd_got_sigbus(f
->cache_fd
))
209 static int journal_file_set_online(JournalFile
*f
) {
214 if (!journal_file_writable(f
))
217 if (f
->fd
< 0 || !f
->header
)
221 switch (f
->offline_state
) {
223 /* No offline thread, no need to wait. */
227 case OFFLINE_SYNCING
: {
228 OfflineState tmp_state
= OFFLINE_SYNCING
;
229 if (!__atomic_compare_exchange_n(&f
->offline_state
, &tmp_state
, OFFLINE_CANCEL
,
230 false, __ATOMIC_SEQ_CST
, __ATOMIC_SEQ_CST
))
233 /* Canceled syncing prior to offlining, no need to wait. */
237 case OFFLINE_AGAIN_FROM_SYNCING
: {
238 OfflineState tmp_state
= OFFLINE_AGAIN_FROM_SYNCING
;
239 if (!__atomic_compare_exchange_n(&f
->offline_state
, &tmp_state
, OFFLINE_CANCEL
,
240 false, __ATOMIC_SEQ_CST
, __ATOMIC_SEQ_CST
))
243 /* Canceled restart from syncing, no need to wait. */
247 case OFFLINE_AGAIN_FROM_OFFLINING
: {
248 OfflineState tmp_state
= OFFLINE_AGAIN_FROM_OFFLINING
;
249 if (!__atomic_compare_exchange_n(&f
->offline_state
, &tmp_state
, OFFLINE_CANCEL
,
250 false, __ATOMIC_SEQ_CST
, __ATOMIC_SEQ_CST
))
253 /* Canceled restart from offlining, must wait for offlining to complete however. */
258 r
= journal_file_set_offline_thread_join(f
);
268 if (mmap_cache_fd_got_sigbus(f
->cache_fd
))
271 switch (f
->header
->state
) {
276 f
->header
->state
= STATE_ONLINE
;
285 JournalFile
* journal_file_close(JournalFile
*f
) {
289 assert(f
->newest_boot_id_prioq_idx
== PRIOQ_IDX_NULL
);
292 mmap_cache_fd_free(f
->cache_fd
);
298 ordered_hashmap_free_free(f
->chain_cache
);
301 free(f
->compress_buffer
);
306 munmap(f
->fss_file
, PAGE_ALIGN(f
->fss_file_size
));
308 free(f
->fsprg_state
);
313 gcry_md_close(f
->hmac
);
319 static bool keyed_hash_requested(void) {
320 static thread_local
int cached
= -1;
324 r
= getenv_bool("SYSTEMD_JOURNAL_KEYED_HASH");
327 log_debug_errno(r
, "Failed to parse $SYSTEMD_JOURNAL_KEYED_HASH environment variable, ignoring: %m");
336 static bool compact_mode_requested(void) {
337 static thread_local
int cached
= -1;
341 r
= getenv_bool("SYSTEMD_JOURNAL_COMPACT");
344 log_debug_errno(r
, "Failed to parse $SYSTEMD_JOURNAL_COMPACT environment variable, ignoring: %m");
354 static Compression
getenv_compression(void) {
359 e
= getenv("SYSTEMD_JOURNAL_COMPRESS");
361 return DEFAULT_COMPRESSION
;
363 r
= parse_boolean(e
);
365 return r
? DEFAULT_COMPRESSION
: COMPRESSION_NONE
;
367 c
= compression_from_string(e
);
369 log_debug_errno(c
, "Failed to parse SYSTEMD_JOURNAL_COMPRESS value, ignoring: %s", e
);
370 return DEFAULT_COMPRESSION
;
373 if (!compression_supported(c
)) {
374 log_debug("Unsupported compression algorithm specified, ignoring: %s", e
);
375 return DEFAULT_COMPRESSION
;
382 static Compression
compression_requested(void) {
384 static thread_local Compression cached
= _COMPRESSION_INVALID
;
387 cached
= getenv_compression();
391 return COMPRESSION_NONE
;
395 static int journal_file_init_header(
397 JournalFileFlags file_flags
,
398 JournalFile
*template) {
407 /* Try to load the FSPRG state, and if we can't, then just don't do sealing */
408 seal
= FLAGS_SET(file_flags
, JOURNAL_SEAL
) && journal_file_fss_load(f
) >= 0;
412 .header_size
= htole64(ALIGN64(sizeof(h
))),
413 .incompatible_flags
= htole32(
414 FLAGS_SET(file_flags
, JOURNAL_COMPRESS
) * COMPRESSION_TO_HEADER_INCOMPATIBLE_FLAG(compression_requested()) |
415 keyed_hash_requested() * HEADER_INCOMPATIBLE_KEYED_HASH
|
416 compact_mode_requested() * HEADER_INCOMPATIBLE_COMPACT
),
417 .compatible_flags
= htole32(
418 (seal
* HEADER_COMPATIBLE_SEALED
) |
419 HEADER_COMPATIBLE_TAIL_ENTRY_BOOT_ID
),
422 assert_cc(sizeof(h
.signature
) == sizeof(HEADER_SIGNATURE
));
423 memcpy(h
.signature
, HEADER_SIGNATURE
, sizeof(HEADER_SIGNATURE
));
425 r
= sd_id128_randomize(&h
.file_id
);
429 r
= sd_id128_get_machine(&h
.machine_id
);
430 if (r
< 0 && !ERRNO_IS_MACHINE_ID_UNSET(r
))
431 return r
; /* If we have no valid machine ID (test environment?), let's simply leave the
432 * machine ID field all zeroes. */
435 h
.seqnum_id
= template->header
->seqnum_id
;
436 h
.tail_entry_seqnum
= template->header
->tail_entry_seqnum
;
438 h
.seqnum_id
= h
.file_id
;
440 k
= pwrite(f
->fd
, &h
, sizeof(h
), 0);
449 static int journal_file_refresh_header(JournalFile
*f
) {
455 /* We used to update the header's boot ID field here, but we don't do that anymore, as per
456 * HEADER_COMPATIBLE_TAIL_ENTRY_BOOT_ID */
458 r
= journal_file_set_online(f
);
460 /* Sync the online state to disk; likely just created a new file, also sync the directory this file
462 (void) fsync_full(f
->fd
);
467 static bool warn_wrong_flags(const JournalFile
*f
, bool compatible
) {
468 const uint32_t any
= compatible
? HEADER_COMPATIBLE_ANY
: HEADER_INCOMPATIBLE_ANY
,
469 supported
= compatible
? HEADER_COMPATIBLE_SUPPORTED
: HEADER_INCOMPATIBLE_SUPPORTED
;
470 const char *type
= compatible
? "compatible" : "incompatible";
476 flags
= le32toh(compatible
? f
->header
->compatible_flags
: f
->header
->incompatible_flags
);
478 if (flags
& ~supported
) {
480 log_debug("Journal file %s has unknown %s flags 0x%"PRIx32
,
481 f
->path
, type
, flags
& ~any
);
482 flags
= (flags
& any
) & ~supported
;
486 _cleanup_free_
char *t
= NULL
;
489 if (flags
& HEADER_COMPATIBLE_SEALED
)
490 strv
[n
++] = "sealed";
492 if (flags
& HEADER_INCOMPATIBLE_COMPRESSED_XZ
)
493 strv
[n
++] = "xz-compressed";
494 if (flags
& HEADER_INCOMPATIBLE_COMPRESSED_LZ4
)
495 strv
[n
++] = "lz4-compressed";
496 if (flags
& HEADER_INCOMPATIBLE_COMPRESSED_ZSTD
)
497 strv
[n
++] = "zstd-compressed";
498 if (flags
& HEADER_INCOMPATIBLE_KEYED_HASH
)
499 strv
[n
++] = "keyed-hash";
500 if (flags
& HEADER_INCOMPATIBLE_COMPACT
)
501 strv
[n
++] = "compact";
504 assert(n
< ELEMENTSOF(strv
));
506 t
= strv_join((char**) strv
, ", ");
507 log_debug("Journal file %s uses %s %s %s disabled at compilation time.",
508 f
->path
, type
, n
> 1 ? "flags" : "flag", strnull(t
));
516 static bool offset_is_valid(uint64_t offset
, uint64_t header_size
, uint64_t tail_object_offset
) {
519 if (!VALID64(offset
))
521 if (offset
< header_size
)
523 if (offset
> tail_object_offset
)
528 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
) {
529 if ((offset
== 0) != (size
== 0))
533 if (offset
<= offsetof(Object
, hash_table
.items
))
535 offset
-= offsetof(Object
, hash_table
.items
);
536 if (!offset_is_valid(offset
, header_size
, tail_object_offset
))
538 assert(offset
<= header_size
+ arena_size
);
539 if (size
> header_size
+ arena_size
- offset
)
544 static int journal_file_verify_header(JournalFile
*f
) {
545 uint64_t arena_size
, header_size
;
550 if (memcmp(f
->header
->signature
, HEADER_SIGNATURE
, 8))
553 /* In both read and write mode we refuse to open files with incompatible
554 * flags we don't know. */
555 if (warn_wrong_flags(f
, false))
556 return -EPROTONOSUPPORT
;
558 /* When open for writing we refuse to open files with compatible flags, too. */
559 if (journal_file_writable(f
) && warn_wrong_flags(f
, true))
560 return -EPROTONOSUPPORT
;
562 if (f
->header
->state
>= _STATE_MAX
)
565 header_size
= le64toh(READ_NOW(f
->header
->header_size
));
567 /* The first addition was n_data, so check that we are at least this large */
568 if (header_size
< HEADER_SIZE_MIN
)
571 /* When open for writing we refuse to open files with a mismatch of the header size, i.e. writing to
572 * files implementing older or new header structures. */
573 if (journal_file_writable(f
) && header_size
!= sizeof(Header
))
574 return -EPROTONOSUPPORT
;
576 /* Don't write to journal files without the new boot ID update behavior guarantee. */
577 if (journal_file_writable(f
) && !JOURNAL_HEADER_TAIL_ENTRY_BOOT_ID(f
->header
))
578 return -EPROTONOSUPPORT
;
580 if (JOURNAL_HEADER_SEALED(f
->header
) && !JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
583 arena_size
= le64toh(READ_NOW(f
->header
->arena_size
));
585 if (UINT64_MAX
- header_size
< arena_size
|| header_size
+ arena_size
> (uint64_t) f
->last_stat
.st_size
)
588 uint64_t tail_object_offset
= le64toh(f
->header
->tail_object_offset
);
589 if (!offset_is_valid(tail_object_offset
, header_size
, UINT64_MAX
))
591 if (header_size
+ arena_size
< tail_object_offset
)
593 if (header_size
+ arena_size
- tail_object_offset
< sizeof(ObjectHeader
))
596 if (!hash_table_is_valid(le64toh(f
->header
->data_hash_table_offset
),
597 le64toh(f
->header
->data_hash_table_size
),
598 header_size
, arena_size
, tail_object_offset
))
601 if (!hash_table_is_valid(le64toh(f
->header
->field_hash_table_offset
),
602 le64toh(f
->header
->field_hash_table_size
),
603 header_size
, arena_size
, tail_object_offset
))
606 uint64_t entry_array_offset
= le64toh(f
->header
->entry_array_offset
);
607 if (!offset_is_valid(entry_array_offset
, header_size
, tail_object_offset
))
610 if (JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_array_offset
)) {
611 uint32_t offset
= le32toh(f
->header
->tail_entry_array_offset
);
612 uint32_t n
= le32toh(f
->header
->tail_entry_array_n_entries
);
614 if (!offset_is_valid(offset
, header_size
, tail_object_offset
))
616 if (entry_array_offset
> offset
)
618 if (entry_array_offset
== 0 && offset
!= 0)
620 if ((offset
== 0) != (n
== 0))
622 assert(offset
<= header_size
+ arena_size
);
623 if ((uint64_t) n
* journal_file_entry_array_item_size(f
) > header_size
+ arena_size
- offset
)
627 if (JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_offset
)) {
628 uint64_t offset
= le64toh(f
->header
->tail_entry_offset
);
630 if (!offset_is_valid(offset
, header_size
, tail_object_offset
))
634 /* When there is an entry object, then these fields must be filled. */
635 if (sd_id128_is_null(f
->header
->tail_entry_boot_id
))
637 if (!VALID_REALTIME(le64toh(f
->header
->head_entry_realtime
)))
639 if (!VALID_REALTIME(le64toh(f
->header
->tail_entry_realtime
)))
641 if (!VALID_MONOTONIC(le64toh(f
->header
->tail_entry_realtime
)))
644 /* Otherwise, the fields must be zero. */
645 if (JOURNAL_HEADER_TAIL_ENTRY_BOOT_ID(f
->header
) &&
646 !sd_id128_is_null(f
->header
->tail_entry_boot_id
))
648 if (f
->header
->head_entry_realtime
!= 0)
650 if (f
->header
->tail_entry_realtime
!= 0)
652 if (f
->header
->tail_entry_realtime
!= 0)
657 /* Verify number of objects */
658 uint64_t n_objects
= le64toh(f
->header
->n_objects
);
659 if (n_objects
> arena_size
/ sizeof(ObjectHeader
))
662 uint64_t n_entries
= le64toh(f
->header
->n_entries
);
663 if (n_entries
> n_objects
)
666 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
667 le64toh(f
->header
->n_data
) > n_objects
)
670 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
671 le64toh(f
->header
->n_fields
) > n_objects
)
674 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
) &&
675 le64toh(f
->header
->n_tags
) > n_objects
)
678 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
) &&
679 le64toh(f
->header
->n_entry_arrays
) > n_objects
)
682 if (JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_array_n_entries
) &&
683 le32toh(f
->header
->tail_entry_array_n_entries
) > n_entries
)
686 if (journal_file_writable(f
)) {
687 sd_id128_t machine_id
;
691 r
= sd_id128_get_machine(&machine_id
);
692 if (ERRNO_IS_NEG_MACHINE_ID_UNSET(r
)) /* Gracefully handle the machine ID not being initialized yet */
693 machine_id
= SD_ID128_NULL
;
697 if (!sd_id128_equal(machine_id
, f
->header
->machine_id
))
698 return log_debug_errno(SYNTHETIC_ERRNO(EHOSTDOWN
),
699 "Trying to open journal file from different host for writing, refusing.");
701 state
= f
->header
->state
;
703 if (state
== STATE_ARCHIVED
)
704 return -ESHUTDOWN
; /* Already archived */
705 if (state
== STATE_ONLINE
)
706 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
),
707 "Journal file %s is already online. Assuming unclean closing.",
709 if (state
!= STATE_OFFLINE
)
710 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
),
711 "Journal file %s has unknown state %i.",
714 if (f
->header
->field_hash_table_size
== 0 || f
->header
->data_hash_table_size
== 0)
721 int journal_file_fstat(JournalFile
*f
) {
727 if (fstat(f
->fd
, &f
->last_stat
) < 0)
730 f
->last_stat_usec
= now(CLOCK_MONOTONIC
);
732 /* Refuse dealing with files that aren't regular */
733 r
= stat_verify_regular(&f
->last_stat
);
737 /* Refuse appending to files that are already deleted */
738 if (f
->last_stat
.st_nlink
<= 0)
744 static int journal_file_allocate(JournalFile
*f
, uint64_t offset
, uint64_t size
) {
745 uint64_t old_size
, new_size
, old_header_size
, old_arena_size
;
751 /* We assume that this file is not sparse, and we know that for sure, since we always call
752 * posix_fallocate() ourselves */
754 if (size
> PAGE_ALIGN_DOWN_U64(UINT64_MAX
) - offset
)
757 if (mmap_cache_fd_got_sigbus(f
->cache_fd
))
760 old_header_size
= le64toh(READ_NOW(f
->header
->header_size
));
761 old_arena_size
= le64toh(READ_NOW(f
->header
->arena_size
));
762 if (old_arena_size
> PAGE_ALIGN_DOWN_U64(UINT64_MAX
) - old_header_size
)
765 old_size
= old_header_size
+ old_arena_size
;
767 new_size
= MAX(PAGE_ALIGN_U64(offset
+ size
), old_header_size
);
769 if (new_size
<= old_size
) {
771 /* We already pre-allocated enough space, but before
772 * we write to it, let's check with fstat() if the
773 * file got deleted, in order make sure we don't throw
774 * away the data immediately. Don't check fstat() for
775 * all writes though, but only once ever 10s. */
777 if (f
->last_stat_usec
+ LAST_STAT_REFRESH_USEC
> now(CLOCK_MONOTONIC
))
780 return journal_file_fstat(f
);
783 /* Allocate more space. */
785 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
788 /* Refuse to go over 4G in compact mode so offsets can be stored in 32-bit. */
789 if (JOURNAL_HEADER_COMPACT(f
->header
) && new_size
> UINT32_MAX
)
792 if (new_size
> f
->metrics
.min_size
&& f
->metrics
.keep_free
> 0) {
795 if (fstatvfs(f
->fd
, &svfs
) >= 0) {
798 available
= LESS_BY((uint64_t) svfs
.f_bfree
* (uint64_t) svfs
.f_bsize
, f
->metrics
.keep_free
);
800 if (new_size
- old_size
> available
)
805 /* Increase by larger blocks at once */
806 new_size
= ROUND_UP(new_size
, FILE_SIZE_INCREASE
);
807 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
808 new_size
= f
->metrics
.max_size
;
810 /* Note that the glibc fallocate() fallback is very
811 inefficient, hence we try to minimize the allocation area
813 r
= posix_fallocate_loop(f
->fd
, old_size
, new_size
- old_size
);
817 f
->header
->arena_size
= htole64(new_size
- old_header_size
);
819 return journal_file_fstat(f
);
822 static unsigned type_to_context(ObjectType type
) {
823 /* One context for each type, plus one catch-all for the rest */
824 assert_cc(_OBJECT_TYPE_MAX
<= MMAP_CACHE_MAX_CONTEXTS
);
825 assert_cc(CONTEXT_HEADER
< MMAP_CACHE_MAX_CONTEXTS
);
826 return type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
? type
: 0;
829 static int journal_file_move_to(
842 /* This function may clear, overwrite, or alter previously cached entries with the same type. After
843 * this function has been called, all previously read objects with the same type may be invalidated,
844 * hence must be re-read before use. */
849 if (size
> UINT64_MAX
- offset
)
852 /* Avoid SIGBUS on invalid accesses */
853 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
) {
854 /* Hmm, out of range? Let's refresh the fstat() data
855 * first, before we trust that check. */
857 r
= journal_file_fstat(f
);
861 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
)
862 return -EADDRNOTAVAIL
;
865 return mmap_cache_fd_get(f
->cache_fd
, type_to_context(type
), keep_always
, offset
, size
, &f
->last_stat
, ret
);
868 static uint64_t minimum_header_size(JournalFile
*f
, Object
*o
) {
870 static const uint64_t table
[] = {
871 [OBJECT_DATA
] = sizeof(DataObject
),
872 [OBJECT_FIELD
] = sizeof(FieldObject
),
873 [OBJECT_ENTRY
] = sizeof(EntryObject
),
874 [OBJECT_DATA_HASH_TABLE
] = sizeof(HashTableObject
),
875 [OBJECT_FIELD_HASH_TABLE
] = sizeof(HashTableObject
),
876 [OBJECT_ENTRY_ARRAY
] = sizeof(EntryArrayObject
),
877 [OBJECT_TAG
] = sizeof(TagObject
),
883 if (o
->object
.type
== OBJECT_DATA
)
884 return journal_file_data_payload_offset(f
);
886 if (o
->object
.type
>= ELEMENTSOF(table
) || table
[o
->object
.type
] <= 0)
887 return sizeof(ObjectHeader
);
889 return table
[o
->object
.type
];
892 static int check_object_header(JournalFile
*f
, Object
*o
, ObjectType type
, uint64_t offset
) {
898 s
= le64toh(READ_NOW(o
->object
.size
));
900 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
901 "Attempt to move to uninitialized object: %" PRIu64
,
904 if (s
< sizeof(ObjectHeader
))
905 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
906 "Attempt to move to overly short object with size %"PRIu64
": %" PRIu64
,
909 if (o
->object
.type
<= OBJECT_UNUSED
|| o
->object
.type
>= _OBJECT_TYPE_MAX
)
910 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
911 "Attempt to move to object with invalid type (%u): %" PRIu64
,
912 o
->object
.type
, offset
);
914 if (type
> OBJECT_UNUSED
&& o
->object
.type
!= type
)
915 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
916 "Found %s object while expecting %s object: %" PRIu64
,
917 journal_object_type_to_string(o
->object
.type
),
918 journal_object_type_to_string(type
),
921 if (s
< minimum_header_size(f
, o
))
922 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
923 "Size of %s object (%"PRIu64
") is smaller than the minimum object size (%"PRIu64
"): %" PRIu64
,
924 journal_object_type_to_string(o
->object
.type
),
926 minimum_header_size(f
, o
),
932 /* Lightweight object checks. We want this to be fast, so that we won't
933 * slowdown every journal_file_move_to_object() call too much. */
934 static int check_object(JournalFile
*f
, Object
*o
, uint64_t offset
) {
938 switch (o
->object
.type
) {
941 if ((le64toh(o
->data
.entry_offset
) == 0) ^ (le64toh(o
->data
.n_entries
) == 0))
942 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
943 "Bad data n_entries: %" PRIu64
": %" PRIu64
,
944 le64toh(o
->data
.n_entries
),
947 if (le64toh(o
->object
.size
) <= journal_file_data_payload_offset(f
))
948 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
949 "Bad data size (<= %zu): %" PRIu64
": %" PRIu64
,
950 journal_file_data_payload_offset(f
),
951 le64toh(o
->object
.size
),
954 if (!VALID64(le64toh(o
->data
.next_hash_offset
)) ||
955 !VALID64(le64toh(o
->data
.next_field_offset
)) ||
956 !VALID64(le64toh(o
->data
.entry_offset
)) ||
957 !VALID64(le64toh(o
->data
.entry_array_offset
)))
958 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
959 "Invalid offset, next_hash_offset=" OFSfmt
", next_field_offset=" OFSfmt
", entry_offset=" OFSfmt
", entry_array_offset=" OFSfmt
": %" PRIu64
,
960 le64toh(o
->data
.next_hash_offset
),
961 le64toh(o
->data
.next_field_offset
),
962 le64toh(o
->data
.entry_offset
),
963 le64toh(o
->data
.entry_array_offset
),
969 if (le64toh(o
->object
.size
) <= offsetof(Object
, field
.payload
))
970 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
971 "Bad field size (<= %zu): %" PRIu64
": %" PRIu64
,
972 offsetof(Object
, field
.payload
),
973 le64toh(o
->object
.size
),
976 if (!VALID64(le64toh(o
->field
.next_hash_offset
)) ||
977 !VALID64(le64toh(o
->field
.head_data_offset
)))
978 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
979 "Invalid offset, next_hash_offset=" OFSfmt
", head_data_offset=" OFSfmt
": %" PRIu64
,
980 le64toh(o
->field
.next_hash_offset
),
981 le64toh(o
->field
.head_data_offset
),
988 sz
= le64toh(READ_NOW(o
->object
.size
));
989 if (sz
< offsetof(Object
, entry
.items
) ||
990 (sz
- offsetof(Object
, entry
.items
)) % journal_file_entry_item_size(f
) != 0)
991 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
992 "Bad entry size (<= %zu): %" PRIu64
": %" PRIu64
,
993 offsetof(Object
, entry
.items
),
997 if ((sz
- offsetof(Object
, entry
.items
)) / journal_file_entry_item_size(f
) <= 0)
998 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
999 "Invalid number items in entry: %" PRIu64
": %" PRIu64
,
1000 (sz
- offsetof(Object
, entry
.items
)) / journal_file_entry_item_size(f
),
1003 if (le64toh(o
->entry
.seqnum
) <= 0)
1004 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1005 "Invalid entry seqnum: %" PRIx64
": %" PRIu64
,
1006 le64toh(o
->entry
.seqnum
),
1009 if (!VALID_REALTIME(le64toh(o
->entry
.realtime
)))
1010 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1011 "Invalid entry realtime timestamp: %" PRIu64
": %" PRIu64
,
1012 le64toh(o
->entry
.realtime
),
1015 if (!VALID_MONOTONIC(le64toh(o
->entry
.monotonic
)))
1016 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1017 "Invalid entry monotonic timestamp: %" PRIu64
": %" PRIu64
,
1018 le64toh(o
->entry
.monotonic
),
1021 if (sd_id128_is_null(o
->entry
.boot_id
))
1022 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1023 "Invalid object entry with an empty boot ID: %" PRIu64
,
1029 case OBJECT_DATA_HASH_TABLE
:
1030 case OBJECT_FIELD_HASH_TABLE
: {
1033 sz
= le64toh(READ_NOW(o
->object
.size
));
1034 if (sz
< offsetof(Object
, hash_table
.items
) ||
1035 (sz
- offsetof(Object
, hash_table
.items
)) % sizeof(HashItem
) != 0 ||
1036 (sz
- offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
) <= 0)
1037 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1038 "Invalid %s hash table size: %" PRIu64
": %" PRIu64
,
1039 journal_object_type_to_string(o
->object
.type
),
1046 case OBJECT_ENTRY_ARRAY
: {
1049 sz
= le64toh(READ_NOW(o
->object
.size
));
1050 if (sz
< offsetof(Object
, entry_array
.items
) ||
1051 (sz
- offsetof(Object
, entry_array
.items
)) % journal_file_entry_array_item_size(f
) != 0 ||
1052 (sz
- offsetof(Object
, entry_array
.items
)) / journal_file_entry_array_item_size(f
) <= 0)
1053 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1054 "Invalid object entry array size: %" PRIu64
": %" PRIu64
,
1057 /* Here, we request that the offset of each entry array object is in strictly increasing order. */
1058 next
= le64toh(o
->entry_array
.next_entry_array_offset
);
1059 if (!VALID64(next
) || (next
> 0 && next
<= offset
))
1060 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1061 "Invalid object entry array next_entry_array_offset: %" PRIu64
": %" PRIu64
,
1069 if (le64toh(o
->object
.size
) != sizeof(TagObject
))
1070 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1071 "Invalid object tag size: %" PRIu64
": %" PRIu64
,
1072 le64toh(o
->object
.size
),
1075 if (!VALID_EPOCH(le64toh(o
->tag
.epoch
)))
1076 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1077 "Invalid object tag epoch: %" PRIu64
": %" PRIu64
,
1078 le64toh(o
->tag
.epoch
), offset
);
1086 int journal_file_move_to_object(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
**ret
) {
1092 /* Even if this function fails, it may clear, overwrite, or alter previously cached entries with the
1093 * same type. After this function has been called, all previously read objects with the same type may
1094 * be invalidated, hence must be re-read before use. */
1096 /* Objects may only be located at multiple of 64 bit */
1097 if (!VALID64(offset
))
1098 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1099 "Attempt to move to %s object at non-64-bit boundary: %" PRIu64
,
1100 journal_object_type_to_string(type
),
1103 /* Object may not be located in the file header */
1104 if (offset
< le64toh(f
->header
->header_size
))
1105 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1106 "Attempt to move to %s object located in file header: %" PRIu64
,
1107 journal_object_type_to_string(type
),
1110 r
= journal_file_move_to(f
, type
, false, offset
, sizeof(ObjectHeader
), (void**) &o
);
1114 r
= check_object_header(f
, o
, type
, offset
);
1118 r
= journal_file_move_to(f
, type
, false, offset
, le64toh(READ_NOW(o
->object
.size
)), (void**) &o
);
1122 r
= check_object_header(f
, o
, type
, offset
);
1126 r
= check_object(f
, o
, offset
);
1136 int journal_file_read_object_header(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
*ret
) {
1143 /* Objects may only be located at multiple of 64 bit */
1144 if (!VALID64(offset
))
1145 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1146 "Attempt to read %s object at non-64-bit boundary: %" PRIu64
,
1147 journal_object_type_to_string(type
), offset
);
1149 /* Object may not be located in the file header */
1150 if (offset
< le64toh(f
->header
->header_size
))
1151 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1152 "Attempt to read %s object located in file header: %" PRIu64
,
1153 journal_object_type_to_string(type
), offset
);
1155 /* This will likely read too much data but it avoids having to call pread() twice. */
1156 n
= pread(f
->fd
, &o
, sizeof(o
), offset
);
1158 return log_debug_errno(errno
, "Failed to read journal %s object at offset: %" PRIu64
,
1159 journal_object_type_to_string(type
), offset
);
1161 if ((size_t) n
< sizeof(o
.object
))
1162 return log_debug_errno(SYNTHETIC_ERRNO(EIO
),
1163 "Failed to read short %s object at offset: %" PRIu64
,
1164 journal_object_type_to_string(type
), offset
);
1166 r
= check_object_header(f
, &o
, type
, offset
);
1170 if ((size_t) n
< minimum_header_size(f
, &o
))
1171 return log_debug_errno(SYNTHETIC_ERRNO(EIO
),
1172 "Short read while reading %s object: %" PRIu64
,
1173 journal_object_type_to_string(type
), offset
);
1175 r
= check_object(f
, &o
, offset
);
1185 static uint64_t inc_seqnum(uint64_t seqnum
) {
1186 if (seqnum
< UINT64_MAX
-1)
1189 return 1; /* skip over UINT64_MAX and 0 when we run out of seqnums and start again */
1192 static uint64_t journal_file_entry_seqnum(
1196 uint64_t next_seqnum
;
1201 /* Picks a new sequence number for the entry we are about to add and returns it. */
1203 next_seqnum
= inc_seqnum(le64toh(f
->header
->tail_entry_seqnum
));
1205 /* If an external seqnum counter was passed, we update both the local and the external one, and set
1206 * it to the maximum of both */
1208 *seqnum
= next_seqnum
= MAX(inc_seqnum(*seqnum
), next_seqnum
);
1210 f
->header
->tail_entry_seqnum
= htole64(next_seqnum
);
1212 if (f
->header
->head_entry_seqnum
== 0)
1213 f
->header
->head_entry_seqnum
= htole64(next_seqnum
);
1218 int journal_file_append_object(
1222 Object
**ret_object
,
1223 uint64_t *ret_offset
) {
1231 assert(type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
);
1232 assert(size
>= sizeof(ObjectHeader
));
1234 r
= journal_file_set_online(f
);
1238 r
= journal_file_tail_end_by_mmap(f
, &p
);
1242 r
= journal_file_allocate(f
, p
, size
);
1246 r
= journal_file_move_to(f
, type
, false, p
, size
, (void**) &o
);
1250 o
->object
= (ObjectHeader
) {
1252 .size
= htole64(size
),
1255 f
->header
->tail_object_offset
= htole64(p
);
1256 f
->header
->n_objects
= htole64(le64toh(f
->header
->n_objects
) + 1);
1267 static int journal_file_setup_data_hash_table(JournalFile
*f
) {
1275 /* We estimate that we need 1 hash table entry per 768 bytes
1276 of journal file and we want to make sure we never get
1277 beyond 75% fill level. Calculate the hash table size for
1278 the maximum file size based on these metrics. */
1280 s
= (f
->metrics
.max_size
* 4 / 768 / 3) * sizeof(HashItem
);
1281 if (s
< DEFAULT_DATA_HASH_TABLE_SIZE
)
1282 s
= DEFAULT_DATA_HASH_TABLE_SIZE
;
1284 log_debug("Reserving %"PRIu64
" entries in data hash table.", s
/ sizeof(HashItem
));
1286 r
= journal_file_append_object(f
,
1287 OBJECT_DATA_HASH_TABLE
,
1288 offsetof(Object
, hash_table
.items
) + s
,
1293 memzero(o
->hash_table
.items
, s
);
1295 f
->header
->data_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1296 f
->header
->data_hash_table_size
= htole64(s
);
1301 static int journal_file_setup_field_hash_table(JournalFile
*f
) {
1309 /* We use a fixed size hash table for the fields as this
1310 * number should grow very slowly only */
1312 s
= DEFAULT_FIELD_HASH_TABLE_SIZE
;
1313 log_debug("Reserving %"PRIu64
" entries in field hash table.", s
/ sizeof(HashItem
));
1315 r
= journal_file_append_object(f
,
1316 OBJECT_FIELD_HASH_TABLE
,
1317 offsetof(Object
, hash_table
.items
) + s
,
1322 memzero(o
->hash_table
.items
, s
);
1324 f
->header
->field_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1325 f
->header
->field_hash_table_size
= htole64(s
);
1330 int journal_file_map_data_hash_table(JournalFile
*f
) {
1338 if (f
->data_hash_table
)
1341 p
= le64toh(f
->header
->data_hash_table_offset
);
1342 s
= le64toh(f
->header
->data_hash_table_size
);
1344 r
= journal_file_move_to(f
,
1345 OBJECT_DATA_HASH_TABLE
,
1352 f
->data_hash_table
= t
;
1356 int journal_file_map_field_hash_table(JournalFile
*f
) {
1364 if (f
->field_hash_table
)
1367 p
= le64toh(f
->header
->field_hash_table_offset
);
1368 s
= le64toh(f
->header
->field_hash_table_size
);
1370 r
= journal_file_move_to(f
,
1371 OBJECT_FIELD_HASH_TABLE
,
1378 f
->field_hash_table
= t
;
1382 static int journal_file_link_field(
1393 assert(f
->field_hash_table
);
1397 if (o
->object
.type
!= OBJECT_FIELD
)
1400 m
= le64toh(READ_NOW(f
->header
->field_hash_table_size
)) / sizeof(HashItem
);
1404 /* This might alter the window we are looking at */
1405 o
->field
.next_hash_offset
= o
->field
.head_data_offset
= 0;
1408 p
= le64toh(f
->field_hash_table
[h
].tail_hash_offset
);
1410 f
->field_hash_table
[h
].head_hash_offset
= htole64(offset
);
1412 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1416 o
->field
.next_hash_offset
= htole64(offset
);
1419 f
->field_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1421 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
1422 f
->header
->n_fields
= htole64(le64toh(f
->header
->n_fields
) + 1);
1427 static int journal_file_link_data(
1438 assert(f
->data_hash_table
);
1442 if (o
->object
.type
!= OBJECT_DATA
)
1445 m
= le64toh(READ_NOW(f
->header
->data_hash_table_size
)) / sizeof(HashItem
);
1449 /* This might alter the window we are looking at */
1450 o
->data
.next_hash_offset
= o
->data
.next_field_offset
= 0;
1451 o
->data
.entry_offset
= o
->data
.entry_array_offset
= 0;
1452 o
->data
.n_entries
= 0;
1455 p
= le64toh(f
->data_hash_table
[h
].tail_hash_offset
);
1457 /* Only entry in the hash table is easy */
1458 f
->data_hash_table
[h
].head_hash_offset
= htole64(offset
);
1460 /* Move back to the previous data object, to patch in
1463 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1467 o
->data
.next_hash_offset
= htole64(offset
);
1470 f
->data_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1472 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
1473 f
->header
->n_data
= htole64(le64toh(f
->header
->n_data
) + 1);
1478 static int get_next_hash_offset(
1481 le64_t
*next_hash_offset
,
1483 le64_t
*header_max_depth
) {
1489 assert(next_hash_offset
);
1492 nextp
= le64toh(READ_NOW(*next_hash_offset
));
1494 if (nextp
<= *p
) /* Refuse going in loops */
1495 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1496 "Detected hash item loop in %s, refusing.", f
->path
);
1500 /* If the depth of this hash chain is larger than all others we have seen so far, record it */
1501 if (header_max_depth
&& journal_file_writable(f
))
1502 *header_max_depth
= htole64(MAX(*depth
, le64toh(*header_max_depth
)));
1509 int journal_file_find_field_object_with_hash(
1514 Object
**ret_object
,
1515 uint64_t *ret_offset
) {
1517 uint64_t p
, osize
, h
, m
, depth
= 0;
1525 /* If the field hash table is empty, we can't find anything */
1526 if (le64toh(f
->header
->field_hash_table_size
) <= 0)
1529 /* Map the field hash table, if it isn't mapped yet. */
1530 r
= journal_file_map_field_hash_table(f
);
1534 osize
= offsetof(Object
, field
.payload
) + size
;
1536 m
= le64toh(READ_NOW(f
->header
->field_hash_table_size
)) / sizeof(HashItem
);
1541 p
= le64toh(f
->field_hash_table
[h
].head_hash_offset
);
1545 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1549 if (le64toh(o
->field
.hash
) == hash
&&
1550 le64toh(o
->object
.size
) == osize
&&
1551 memcmp(o
->field
.payload
, field
, size
) == 0) {
1561 r
= get_next_hash_offset(
1564 &o
->field
.next_hash_offset
,
1566 JOURNAL_HEADER_CONTAINS(f
->header
, field_hash_chain_depth
) ? &f
->header
->field_hash_chain_depth
: NULL
);
1574 uint64_t journal_file_hash_data(
1581 assert(data
|| sz
== 0);
1583 /* We try to unify our codebase on siphash, hence new-styled journal files utilizing the keyed hash
1584 * function use siphash. Old journal files use the Jenkins hash. */
1586 if (JOURNAL_HEADER_KEYED_HASH(f
->header
))
1587 return siphash24(data
, sz
, f
->header
->file_id
.bytes
);
1589 return jenkins_hash64(data
, sz
);
1592 int journal_file_find_field_object(
1596 Object
**ret_object
,
1597 uint64_t *ret_offset
) {
1603 return journal_file_find_field_object_with_hash(
1606 journal_file_hash_data(f
, field
, size
),
1607 ret_object
, ret_offset
);
1610 int journal_file_find_data_object_with_hash(
1615 Object
**ret_object
,
1616 uint64_t *ret_offset
) {
1618 uint64_t p
, h
, m
, depth
= 0;
1623 assert(data
|| size
== 0);
1625 /* If there's no data hash table, then there's no entry. */
1626 if (le64toh(f
->header
->data_hash_table_size
) <= 0)
1629 /* Map the data hash table, if it isn't mapped yet. */
1630 r
= journal_file_map_data_hash_table(f
);
1634 m
= le64toh(READ_NOW(f
->header
->data_hash_table_size
)) / sizeof(HashItem
);
1639 p
= le64toh(f
->data_hash_table
[h
].head_hash_offset
);
1646 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1650 if (le64toh(o
->data
.hash
) != hash
)
1653 r
= journal_file_data_payload(f
, o
, p
, NULL
, 0, 0, &d
, &rsize
);
1656 assert(r
> 0); /* journal_file_data_payload() always returns > 0 if no field is provided. */
1658 if (memcmp_nn(data
, size
, d
, rsize
) == 0) {
1669 r
= get_next_hash_offset(
1672 &o
->data
.next_hash_offset
,
1674 JOURNAL_HEADER_CONTAINS(f
->header
, data_hash_chain_depth
) ? &f
->header
->data_hash_chain_depth
: NULL
);
1682 int journal_file_find_data_object(
1686 Object
**ret_object
,
1687 uint64_t *ret_offset
) {
1690 assert(data
|| size
== 0);
1692 return journal_file_find_data_object_with_hash(
1695 journal_file_hash_data(f
, data
, size
),
1696 ret_object
, ret_offset
);
1699 bool journal_field_valid(const char *p
, size_t l
, bool allow_protected
) {
1700 /* We kinda enforce POSIX syntax recommendations for
1701 environment variables here, but make a couple of additional
1704 http://pubs.opengroup.org/onlinepubs/000095399/basedefs/xbd_chap08.html */
1711 /* No empty field names */
1715 /* Don't allow names longer than 64 chars */
1719 /* Variables starting with an underscore are protected */
1720 if (!allow_protected
&& p
[0] == '_')
1723 /* Don't allow digits as first character */
1724 if (ascii_isdigit(p
[0]))
1727 /* Only allow A-Z0-9 and '_' */
1728 for (const char *a
= p
; a
< p
+ l
; a
++)
1729 if ((*a
< 'A' || *a
> 'Z') &&
1730 !ascii_isdigit(*a
) &&
1737 static int journal_file_append_field(
1741 Object
**ret_object
,
1742 uint64_t *ret_offset
) {
1753 if (!journal_field_valid(field
, size
, true))
1756 hash
= journal_file_hash_data(f
, field
, size
);
1758 r
= journal_file_find_field_object_with_hash(f
, field
, size
, hash
, ret_object
, ret_offset
);
1764 osize
= offsetof(Object
, field
.payload
) + size
;
1765 r
= journal_file_append_object(f
, OBJECT_FIELD
, osize
, &o
, &p
);
1769 o
->field
.hash
= htole64(hash
);
1770 memcpy(o
->field
.payload
, field
, size
);
1772 r
= journal_file_link_field(f
, o
, p
, hash
);
1776 /* The linking might have altered the window, so let's only pass the offset to hmac which will
1777 * move to the object again if needed. */
1780 r
= journal_file_hmac_put_object(f
, OBJECT_FIELD
, NULL
, p
);
1786 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, ret_object
);
1797 static int maybe_compress_payload(JournalFile
*f
, uint8_t *dst
, const uint8_t *src
, uint64_t size
, size_t *rsize
) {
1801 #if HAVE_COMPRESSION
1805 c
= JOURNAL_FILE_COMPRESSION(f
);
1806 if (c
== COMPRESSION_NONE
|| size
< f
->compress_threshold_bytes
)
1809 r
= compress_blob(c
, src
, size
, dst
, size
- 1, rsize
);
1811 return log_debug_errno(r
, "Failed to compress data object using %s, ignoring: %m", compression_to_string(c
));
1813 log_debug("Compressed data object %"PRIu64
" -> %zu using %s", size
, *rsize
, compression_to_string(c
));
1815 return 1; /* compressed */
1821 static int journal_file_append_data(
1825 Object
**ret_object
,
1826 uint64_t *ret_offset
) {
1828 uint64_t hash
, p
, osize
;
1836 if (!data
|| size
== 0)
1839 hash
= journal_file_hash_data(f
, data
, size
);
1841 r
= journal_file_find_data_object_with_hash(f
, data
, size
, hash
, ret_object
, ret_offset
);
1847 eq
= memchr(data
, '=', size
);
1851 osize
= journal_file_data_payload_offset(f
) + size
;
1852 r
= journal_file_append_object(f
, OBJECT_DATA
, osize
, &o
, &p
);
1856 o
->data
.hash
= htole64(hash
);
1858 r
= maybe_compress_payload(f
, journal_file_data_payload_field(f
, o
), data
, size
, &rsize
);
1860 /* We don't really care failures, let's continue without compression */
1861 memcpy_safe(journal_file_data_payload_field(f
, o
), data
, size
);
1863 Compression c
= JOURNAL_FILE_COMPRESSION(f
);
1865 assert(c
>= 0 && c
< _COMPRESSION_MAX
&& c
!= COMPRESSION_NONE
);
1867 o
->object
.size
= htole64(journal_file_data_payload_offset(f
) + rsize
);
1868 o
->object
.flags
|= COMPRESSION_TO_OBJECT_FLAG(c
);
1871 r
= journal_file_link_data(f
, o
, p
, hash
);
1875 /* The linking might have altered the window, so let's refresh our pointer. */
1876 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1881 r
= journal_file_hmac_put_object(f
, OBJECT_DATA
, o
, p
);
1886 /* Create field object ... */
1887 r
= journal_file_append_field(f
, data
, (uint8_t*) eq
- (uint8_t*) data
, &fo
, NULL
);
1891 /* ... and link it in. */
1892 o
->data
.next_field_offset
= fo
->field
.head_data_offset
;
1893 fo
->field
.head_data_offset
= le64toh(p
);
1904 static int maybe_decompress_payload(
1908 Compression compression
,
1910 size_t field_length
,
1911 size_t data_threshold
,
1917 /* We can't read objects larger than 4G on a 32-bit machine */
1918 if ((uint64_t) (size_t) size
!= size
)
1921 if (compression
!= COMPRESSION_NONE
) {
1922 #if HAVE_COMPRESSION
1927 r
= decompress_startswith(compression
, payload
, size
, &f
->compress_buffer
, field
,
1930 return log_debug_errno(r
,
1931 "Cannot decompress %s object of length %" PRIu64
": %m",
1932 compression_to_string(compression
),
1943 r
= decompress_blob(compression
, payload
, size
, &f
->compress_buffer
, &rsize
, 0);
1948 *ret_data
= f
->compress_buffer
;
1952 return -EPROTONOSUPPORT
;
1955 if (field
&& (size
< field_length
+ 1 || memcmp(payload
, field
, field_length
) != 0 || payload
[field_length
] != '=')) {
1964 *ret_data
= payload
;
1966 *ret_size
= (size_t) size
;
1972 int journal_file_data_payload(
1977 size_t field_length
,
1978 size_t data_threshold
,
1987 assert(!field
== (field_length
== 0)); /* These must be specified together. */
1990 r
= journal_file_move_to_object(f
, OBJECT_DATA
, offset
, &o
);
1995 size
= le64toh(READ_NOW(o
->object
.size
));
1996 if (size
< journal_file_data_payload_offset(f
))
1999 size
-= journal_file_data_payload_offset(f
);
2001 c
= COMPRESSION_FROM_OBJECT(o
);
2003 return -EPROTONOSUPPORT
;
2005 return maybe_decompress_payload(f
, journal_file_data_payload_field(f
, o
), size
, c
, field
,
2006 field_length
, data_threshold
, ret_data
, ret_size
);
2009 uint64_t journal_file_entry_n_items(JournalFile
*f
, Object
*o
) {
2015 if (o
->object
.type
!= OBJECT_ENTRY
)
2018 sz
= le64toh(READ_NOW(o
->object
.size
));
2019 if (sz
< offsetof(Object
, entry
.items
))
2022 return (sz
- offsetof(Object
, entry
.items
)) / journal_file_entry_item_size(f
);
2025 uint64_t journal_file_entry_array_n_items(JournalFile
*f
, Object
*o
) {
2031 if (o
->object
.type
!= OBJECT_ENTRY_ARRAY
)
2034 sz
= le64toh(READ_NOW(o
->object
.size
));
2035 if (sz
< offsetof(Object
, entry_array
.items
))
2038 return (sz
- offsetof(Object
, entry_array
.items
)) / journal_file_entry_array_item_size(f
);
2041 uint64_t journal_file_hash_table_n_items(Object
*o
) {
2046 if (!IN_SET(o
->object
.type
, OBJECT_DATA_HASH_TABLE
, OBJECT_FIELD_HASH_TABLE
))
2049 sz
= le64toh(READ_NOW(o
->object
.size
));
2050 if (sz
< offsetof(Object
, hash_table
.items
))
2053 return (sz
- offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
);
2056 static void write_entry_array_item(JournalFile
*f
, Object
*o
, uint64_t i
, uint64_t p
) {
2060 if (JOURNAL_HEADER_COMPACT(f
->header
)) {
2061 assert(p
<= UINT32_MAX
);
2062 o
->entry_array
.items
.compact
[i
] = htole32(p
);
2064 o
->entry_array
.items
.regular
[i
] = htole64(p
);
2067 static int link_entry_into_array(
2075 uint64_t n
= 0, ap
= 0, q
, i
, a
, hidx
;
2085 a
= tail
? le32toh(*tail
) : le64toh(*first
);
2086 hidx
= le64toh(READ_NOW(*idx
));
2087 i
= tidx
? le32toh(READ_NOW(*tidx
)) : hidx
;
2090 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2094 n
= journal_file_entry_array_n_items(f
, o
);
2096 write_entry_array_item(f
, o
, i
, p
);
2097 *idx
= htole64(hidx
+ 1);
2099 *tidx
= htole32(le32toh(*tidx
) + 1);
2105 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
2116 r
= journal_file_append_object(f
, OBJECT_ENTRY_ARRAY
,
2117 offsetof(Object
, entry_array
.items
) + n
* journal_file_entry_array_item_size(f
),
2123 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY_ARRAY
, o
, q
);
2128 write_entry_array_item(f
, o
, i
, p
);
2131 *first
= htole64(q
);
2133 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, ap
, &o
);
2137 o
->entry_array
.next_entry_array_offset
= htole64(q
);
2143 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
2144 f
->header
->n_entry_arrays
= htole64(le64toh(f
->header
->n_entry_arrays
) + 1);
2146 *idx
= htole64(hidx
+ 1);
2153 static int link_entry_into_array_plus_one(
2171 hidx
= le64toh(READ_NOW(*idx
));
2172 if (hidx
== UINT64_MAX
)
2175 *extra
= htole64(p
);
2179 i
= htole64(hidx
- 1);
2180 r
= link_entry_into_array(f
, first
, &i
, tail
, tidx
, p
);
2185 *idx
= htole64(hidx
+ 1);
2189 static int journal_file_link_entry_item(JournalFile
*f
, uint64_t offset
, uint64_t p
) {
2196 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
2200 return link_entry_into_array_plus_one(f
,
2201 &o
->data
.entry_offset
,
2202 &o
->data
.entry_array_offset
,
2204 JOURNAL_HEADER_COMPACT(f
->header
) ? &o
->data
.compact
.tail_entry_array_offset
: NULL
,
2205 JOURNAL_HEADER_COMPACT(f
->header
) ? &o
->data
.compact
.tail_entry_array_n_entries
: NULL
,
2209 static int journal_file_link_entry(
2213 const EntryItem items
[],
2223 if (o
->object
.type
!= OBJECT_ENTRY
)
2226 __atomic_thread_fence(__ATOMIC_SEQ_CST
);
2228 /* Link up the entry itself */
2229 r
= link_entry_into_array(f
,
2230 &f
->header
->entry_array_offset
,
2231 &f
->header
->n_entries
,
2232 JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_array_offset
) ? &f
->header
->tail_entry_array_offset
: NULL
,
2233 JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_array_n_entries
) ? &f
->header
->tail_entry_array_n_entries
: NULL
,
2238 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
2240 if (f
->header
->head_entry_realtime
== 0)
2241 f
->header
->head_entry_realtime
= o
->entry
.realtime
;
2243 f
->header
->tail_entry_realtime
= o
->entry
.realtime
;
2244 f
->header
->tail_entry_monotonic
= o
->entry
.monotonic
;
2245 if (JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_offset
))
2246 f
->header
->tail_entry_offset
= htole64(offset
);
2247 f
->newest_mtime
= 0; /* we have a new tail entry now, explicitly invalidate newest boot id/timestamp info */
2249 /* Link up the items */
2250 for (uint64_t i
= 0; i
< n_items
; i
++) {
2253 /* If we fail to link an entry item because we can't allocate a new entry array, don't fail
2254 * immediately but try to link the other entry items since it might still be possible to link
2255 * those if they don't require a new entry array to be allocated. */
2257 k
= journal_file_link_entry_item(f
, offset
, items
[i
].object_offset
);
2267 static void write_entry_item(JournalFile
*f
, Object
*o
, uint64_t i
, const EntryItem
*item
) {
2272 if (JOURNAL_HEADER_COMPACT(f
->header
)) {
2273 assert(item
->object_offset
<= UINT32_MAX
);
2274 o
->entry
.items
.compact
[i
].object_offset
= htole32(item
->object_offset
);
2276 o
->entry
.items
.regular
[i
].object_offset
= htole64(item
->object_offset
);
2277 o
->entry
.items
.regular
[i
].hash
= htole64(item
->hash
);
2281 static int journal_file_append_entry_internal(
2283 const dual_timestamp
*ts
,
2284 const sd_id128_t
*boot_id
,
2285 const sd_id128_t
*machine_id
,
2287 const EntryItem items
[],
2290 sd_id128_t
*seqnum_id
,
2291 Object
**ret_object
,
2292 uint64_t *ret_offset
) {
2303 assert(!sd_id128_is_null(*boot_id
));
2304 assert(items
|| n_items
== 0);
2306 if (f
->strict_order
) {
2307 /* If requested be stricter with ordering in this journal file, to make searching via
2308 * bisection fully deterministic. This is an optional feature, so that if desired journal
2309 * files can be written where the ordering is not strictly enforced (in which case bisection
2310 * will yield *a* result, but not the *only* result, when searching for points in
2311 * time). Strict ordering mode is enabled when journald originally writes the files, but
2312 * might not necessarily be if other tools (the remoting tools for example) write journal
2313 * files from combined sources.
2315 * Typically, if any of the errors generated here are seen journald will just rotate the
2316 * journal files and start anew. */
2318 if (ts
->realtime
< le64toh(f
->header
->tail_entry_realtime
))
2319 return log_debug_errno(SYNTHETIC_ERRNO(EREMCHG
),
2320 "Realtime timestamp %" PRIu64
" smaller than previous realtime "
2321 "timestamp %" PRIu64
", refusing entry.",
2322 ts
->realtime
, le64toh(f
->header
->tail_entry_realtime
));
2324 if (sd_id128_equal(*boot_id
, f
->header
->tail_entry_boot_id
) &&
2325 ts
->monotonic
< le64toh(f
->header
->tail_entry_monotonic
))
2326 return log_debug_errno(
2327 SYNTHETIC_ERRNO(ENOTNAM
),
2328 "Monotonic timestamp %" PRIu64
2329 " smaller than previous monotonic timestamp %" PRIu64
2330 " while having the same boot ID, refusing entry.",
2332 le64toh(f
->header
->tail_entry_monotonic
));
2336 /* Settle the passed in sequence number ID */
2338 if (sd_id128_is_null(*seqnum_id
))
2339 *seqnum_id
= f
->header
->seqnum_id
; /* Caller has none assigned, then copy the one from the file */
2340 else if (!sd_id128_equal(*seqnum_id
, f
->header
->seqnum_id
)) {
2341 /* Different seqnum IDs? We can't allow entries from multiple IDs end up in the same journal.*/
2342 if (le64toh(f
->header
->n_entries
) == 0)
2343 f
->header
->seqnum_id
= *seqnum_id
; /* Caller has one, and file so far has no entries, then copy the one from the caller */
2345 return log_debug_errno(SYNTHETIC_ERRNO(EILSEQ
),
2346 "Sequence number IDs don't match, refusing entry.");
2350 if (machine_id
&& sd_id128_is_null(f
->header
->machine_id
))
2351 /* Initialize machine ID when not set yet */
2352 f
->header
->machine_id
= *machine_id
;
2354 osize
= offsetof(Object
, entry
.items
) + (n_items
* journal_file_entry_item_size(f
));
2356 r
= journal_file_append_object(f
, OBJECT_ENTRY
, osize
, &o
, &np
);
2360 o
->entry
.seqnum
= htole64(journal_file_entry_seqnum(f
, seqnum
));
2361 o
->entry
.realtime
= htole64(ts
->realtime
);
2362 o
->entry
.monotonic
= htole64(ts
->monotonic
);
2363 o
->entry
.xor_hash
= htole64(xor_hash
);
2364 o
->entry
.boot_id
= f
->header
->tail_entry_boot_id
= *boot_id
;
2366 for (size_t i
= 0; i
< n_items
; i
++)
2367 write_entry_item(f
, o
, i
, &items
[i
]);
2370 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY
, o
, np
);
2375 r
= journal_file_link_entry(f
, o
, np
, items
, n_items
);
2388 void journal_file_post_change(JournalFile
*f
) {
2394 /* inotify() does not receive IN_MODIFY events from file
2395 * accesses done via mmap(). After each access we hence
2396 * trigger IN_MODIFY by truncating the journal file to its
2397 * current size which triggers IN_MODIFY. */
2399 __atomic_thread_fence(__ATOMIC_SEQ_CST
);
2401 if (ftruncate(f
->fd
, f
->last_stat
.st_size
) < 0)
2402 log_debug_errno(errno
, "Failed to truncate file to its own size: %m");
2405 static int post_change_thunk(sd_event_source
*timer
, uint64_t usec
, void *userdata
) {
2408 journal_file_post_change(userdata
);
2413 static void schedule_post_change(JournalFile
*f
) {
2418 assert(f
->post_change_timer
);
2420 assert_se(e
= sd_event_source_get_event(f
->post_change_timer
));
2422 /* If we are already going down, post the change immediately. */
2423 if (IN_SET(sd_event_get_state(e
), SD_EVENT_EXITING
, SD_EVENT_FINISHED
))
2426 r
= sd_event_source_get_enabled(f
->post_change_timer
, NULL
);
2428 log_debug_errno(r
, "Failed to get ftruncate timer state: %m");
2434 r
= sd_event_source_set_time_relative(f
->post_change_timer
, f
->post_change_timer_period
);
2436 log_debug_errno(r
, "Failed to set time for scheduling ftruncate: %m");
2440 r
= sd_event_source_set_enabled(f
->post_change_timer
, SD_EVENT_ONESHOT
);
2442 log_debug_errno(r
, "Failed to enable scheduled ftruncate: %m");
2449 /* On failure, let's simply post the change immediately. */
2450 journal_file_post_change(f
);
2453 /* Enable coalesced change posting in a timer on the provided sd_event instance */
2454 int journal_file_enable_post_change_timer(JournalFile
*f
, sd_event
*e
, usec_t t
) {
2455 _cleanup_(sd_event_source_unrefp
) sd_event_source
*timer
= NULL
;
2459 assert_return(!f
->post_change_timer
, -EINVAL
);
2463 r
= sd_event_add_time(e
, &timer
, CLOCK_MONOTONIC
, 0, 0, post_change_thunk
, f
);
2467 r
= sd_event_source_set_enabled(timer
, SD_EVENT_OFF
);
2471 f
->post_change_timer
= TAKE_PTR(timer
);
2472 f
->post_change_timer_period
= t
;
2477 static int entry_item_cmp(const EntryItem
*a
, const EntryItem
*b
) {
2478 return CMP(ASSERT_PTR(a
)->object_offset
, ASSERT_PTR(b
)->object_offset
);
2481 static size_t remove_duplicate_entry_items(EntryItem items
[], size_t n
) {
2484 assert(items
|| n
== 0);
2489 for (size_t i
= 1; i
< n
; i
++)
2490 if (items
[i
].object_offset
!= items
[j
- 1].object_offset
)
2491 items
[j
++] = items
[i
];
2496 int journal_file_append_entry(
2498 const dual_timestamp
*ts
,
2499 const sd_id128_t
*boot_id
,
2500 const struct iovec iovec
[],
2503 sd_id128_t
*seqnum_id
,
2504 Object
**ret_object
,
2505 uint64_t *ret_offset
) {
2507 _cleanup_free_ EntryItem
*items_alloc
= NULL
;
2509 uint64_t xor_hash
= 0;
2510 struct dual_timestamp _ts
;
2511 sd_id128_t _boot_id
, _machine_id
, *machine_id
;
2517 assert(n_iovec
> 0);
2520 if (!VALID_REALTIME(ts
->realtime
))
2521 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
2522 "Invalid realtime timestamp %" PRIu64
", refusing entry.",
2524 if (!VALID_MONOTONIC(ts
->monotonic
))
2525 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
2526 "Invalid monotomic timestamp %" PRIu64
", refusing entry.",
2529 dual_timestamp_get(&_ts
);
2534 if (sd_id128_is_null(*boot_id
))
2535 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
), "Empty boot ID, refusing entry.");
2537 r
= sd_id128_get_boot(&_boot_id
);
2541 boot_id
= &_boot_id
;
2544 r
= sd_id128_get_machine(&_machine_id
);
2545 if (ERRNO_IS_NEG_MACHINE_ID_UNSET(r
))
2546 /* Gracefully handle the machine ID not being initialized yet */
2551 machine_id
= &_machine_id
;
2554 r
= journal_file_maybe_append_tag(f
, ts
->realtime
);
2559 if (n_iovec
< ALLOCA_MAX
/ sizeof(EntryItem
) / 2)
2560 items
= newa(EntryItem
, n_iovec
);
2562 items_alloc
= new(EntryItem
, n_iovec
);
2566 items
= items_alloc
;
2569 for (size_t i
= 0; i
< n_iovec
; i
++) {
2573 r
= journal_file_append_data(f
, iovec
[i
].iov_base
, iovec
[i
].iov_len
, &o
, &p
);
2577 /* When calculating the XOR hash field, we need to take special care if the "keyed-hash"
2578 * journal file flag is on. We use the XOR hash field to quickly determine the identity of a
2579 * specific record, and give records with otherwise identical position (i.e. match in seqno,
2580 * timestamp, …) a stable ordering. But for that we can't have it that the hash of the
2581 * objects in each file is different since they are keyed. Hence let's calculate the Jenkins
2582 * hash here for that. This also has the benefit that cursors for old and new journal files
2583 * are completely identical (they include the XOR hash after all). For classic Jenkins-hash
2584 * files things are easier, we can just take the value from the stored record directly. */
2586 if (JOURNAL_HEADER_KEYED_HASH(f
->header
))
2587 xor_hash
^= jenkins_hash64(iovec
[i
].iov_base
, iovec
[i
].iov_len
);
2589 xor_hash
^= le64toh(o
->data
.hash
);
2591 items
[i
] = (EntryItem
) {
2593 .hash
= le64toh(o
->data
.hash
),
2597 /* Order by the position on disk, in order to improve seek
2598 * times for rotating media. */
2599 typesafe_qsort(items
, n_iovec
, entry_item_cmp
);
2600 n_iovec
= remove_duplicate_entry_items(items
, n_iovec
);
2602 r
= journal_file_append_entry_internal(
2615 /* If the memory mapping triggered a SIGBUS then we return an
2616 * IO error and ignore the error code passed down to us, since
2617 * it is very likely just an effect of a nullified replacement
2620 if (mmap_cache_fd_got_sigbus(f
->cache_fd
))
2623 if (f
->post_change_timer
)
2624 schedule_post_change(f
);
2626 journal_file_post_change(f
);
2631 typedef struct ChainCacheItem
{
2632 uint64_t first
; /* The offset of the entry array object at the beginning of the chain,
2633 * i.e., le64toh(f->header->entry_array_offset), or le64toh(o->data.entry_offset). */
2634 uint64_t array
; /* The offset of the cached entry array object. */
2635 uint64_t begin
; /* The offset of the first item in the cached array. */
2636 uint64_t total
; /* The total number of items in all arrays before the cached one in the chain. */
2637 uint64_t last_index
; /* The last index we looked at in the cached array, to optimize locality when bisecting. */
2640 static void chain_cache_put(
2647 uint64_t last_index
) {
2652 /* If the chain item to cache for this chain is the
2653 * first one it's not worth caching anything */
2657 if (ordered_hashmap_size(h
) >= CHAIN_CACHE_MAX
) {
2658 ci
= ordered_hashmap_steal_first(h
);
2661 ci
= new(ChainCacheItem
, 1);
2668 if (ordered_hashmap_put(h
, &ci
->first
, ci
) < 0) {
2673 assert(ci
->first
== first
);
2678 ci
->last_index
= last_index
;
2681 static int bump_array_index(uint64_t *i
, direction_t direction
, uint64_t n
) {
2684 /* Increase or decrease the specified index, in the right direction. */
2686 if (direction
== DIRECTION_DOWN
) {
2701 static int bump_entry_array(
2703 Object
*o
, /* the current entry array object. */
2704 uint64_t offset
, /* the offset of the entry array object. */
2705 uint64_t first
, /* The offset of the first entry array object in the chain. */
2706 direction_t direction
,
2714 if (direction
== DIRECTION_DOWN
) {
2716 assert(o
->object
.type
== OBJECT_ENTRY_ARRAY
);
2718 *ret
= le64toh(o
->entry_array
.next_entry_array_offset
);
2721 /* Entry array chains are a singly linked list, so to find the previous array in the chain, we have
2722 * to start iterating from the top. */
2726 uint64_t p
= first
, q
= 0;
2727 while (p
> 0 && p
!= offset
) {
2728 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, p
, &o
);
2733 p
= le64toh(o
->entry_array
.next_entry_array_offset
);
2736 /* If we can't find the previous entry array in the entry array chain, we're likely dealing with a
2737 * corrupted journal file. */
2747 static int generic_array_get(
2749 uint64_t first
, /* The offset of the first entry array object in the chain. */
2750 uint64_t i
, /* The index of the target object counted from the beginning of the entry array chain. */
2751 direction_t direction
,
2752 Object
**ret_object
, /* The found object. */
2753 uint64_t *ret_offset
) { /* The offset of the found object. */
2755 uint64_t a
, t
= 0, k
;
2762 /* FIXME: fix return value assignment on success. */
2766 /* Try the chain cache first */
2767 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2768 if (ci
&& i
> ci
->total
) {
2775 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2776 if (IN_SET(r
, -EBADMSG
, -EADDRNOTAVAIL
)) {
2777 /* If there's corruption and we're going downwards, let's pretend we reached the
2778 * final entry in the entry array chain. */
2780 if (direction
== DIRECTION_DOWN
)
2783 /* If there's corruption and we're going upwards, move back to the previous entry
2784 * array and start iterating entries from there. */
2792 k
= journal_file_entry_array_n_items(f
, o
);
2799 /* The index is larger than the number of elements in the array. Let's move to the next array. */
2802 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
2805 /* If we've found the right location, now look for the first non-corrupt entry object (in the right
2809 if (i
== UINT64_MAX
) {
2810 r
= bump_entry_array(f
, o
, a
, first
, direction
, &a
);
2814 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2818 k
= journal_file_entry_array_n_items(f
, o
);
2822 if (direction
== DIRECTION_DOWN
)
2825 /* We moved to the previous array. The total must be decreased. */
2827 return -EBADMSG
; /* chain cache is broken ? */
2837 p
= journal_file_entry_array_item(f
, o
, i
);
2839 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, ret_object
);
2841 /* Let's cache this item for the next invocation */
2842 chain_cache_put(f
->chain_cache
, ci
, first
, a
, journal_file_entry_array_item(f
, o
, 0), t
, i
);
2849 if (!IN_SET(r
, -EADDRNOTAVAIL
, -EBADMSG
))
2852 /* OK, so this entry is borked. Most likely some entry didn't get synced to
2853 * disk properly, let's see if the next one might work for us instead. */
2854 log_debug_errno(r
, "Entry item %" PRIu64
" is bad, skipping over it.", i
);
2856 } while (bump_array_index(&i
, direction
, k
) > 0);
2858 /* All entries tried in the above do-while loop are broken. Let's move to the next (or previous) array. */
2860 if (direction
== DIRECTION_DOWN
)
2861 /* We are going to the next array, the total must be incremented. */
2876 static int generic_array_bisect_one(
2878 uint64_t a
, /* offset of entry array object. */
2879 uint64_t i
, /* index of the entry item we will test. */
2881 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2882 direction_t direction
,
2885 uint64_t *ret_offset
) {
2892 assert(test_object
);
2896 assert(i
<= *right
);
2898 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2902 p
= journal_file_entry_array_item(f
, array
, i
);
2906 r
= test_object(f
, p
, needle
);
2907 if (IN_SET(r
, -EBADMSG
, -EADDRNOTAVAIL
)) {
2908 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short.");
2910 return -ENOANO
; /* recognizable error */
2915 if (r
== TEST_FOUND
)
2916 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2918 if (r
== TEST_RIGHT
)
2929 static int generic_array_bisect(
2934 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2935 direction_t direction
,
2936 Object
**ret_object
,
2937 uint64_t *ret_offset
,
2938 uint64_t *ret_idx
) {
2940 /* Given an entry array chain, this function finds the object "closest" to the given needle in the
2941 * chain, taking into account the provided direction. A function can be provided to determine how
2942 * an object is matched against the given needle.
2944 * Given a journal file, the offset of an object and the needle, the test_object() function should
2945 * return TEST_LEFT if the needle is located earlier in the entry array chain, TEST_LEFT if the
2946 * needle is located later in the entry array chain and TEST_FOUND if the object matches the needle.
2947 * If test_object() returns TEST_FOUND for a specific object, that object's information will be used
2948 * to populate the return values of this function. If test_object() never returns TEST_FOUND, the
2949 * return values are populated with the details of one of the objects closest to the needle. If the
2950 * direction is DIRECTION_UP, the earlier object is used. Otherwise, the later object is used.
2953 uint64_t a
, p
, t
= 0, i
= 0, last_p
= 0, last_index
= UINT64_MAX
;
2954 bool subtract_one
= false;
2960 assert(test_object
);
2962 /* Start with the first array in the chain */
2965 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2966 if (ci
&& n
> ci
->total
&& ci
->begin
!= 0) {
2967 /* Ah, we have iterated this bisection array chain previously! Let's see if we can skip ahead
2968 * in the chain, as far as the last time. But we can't jump backwards in the chain, so let's
2969 * check that first. */
2971 r
= test_object(f
, ci
->begin
, needle
);
2975 if (r
== TEST_LEFT
) {
2976 /* OK, what we are looking for is right of the begin of this EntryArray, so let's
2977 * jump straight to previously cached array in the chain */
2982 last_index
= ci
->last_index
;
2987 uint64_t left
= 0, right
, k
, lp
;
2989 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2993 k
= journal_file_entry_array_n_items(f
, array
);
2999 r
= generic_array_bisect_one(f
, a
, right
, needle
, test_object
, direction
, &left
, &right
, &lp
);
3007 if (r
== TEST_RIGHT
) {
3008 /* If we cached the last index we looked at, let's try to not to jump too wildly
3009 * around and see if we can limit the range to look at early to the immediate
3010 * neighbors of the last index we looked at. */
3012 if (last_index
> 0 && last_index
- 1 < right
) {
3013 r
= generic_array_bisect_one(f
, a
, last_index
- 1, needle
, test_object
, direction
, &left
, &right
, NULL
);
3014 if (r
< 0 && r
!= -ENOANO
)
3018 if (last_index
< right
) {
3019 r
= generic_array_bisect_one(f
, a
, last_index
+ 1, needle
, test_object
, direction
, &left
, &right
, NULL
);
3020 if (r
< 0 && r
!= -ENOANO
)
3025 if (left
== right
) {
3026 if (direction
== DIRECTION_UP
)
3027 subtract_one
= true;
3033 assert(left
< right
);
3034 i
= (left
+ right
) / 2;
3036 r
= generic_array_bisect_one(f
, a
, i
, needle
, test_object
, direction
, &left
, &right
, NULL
);
3037 if (r
< 0 && r
!= -ENOANO
)
3043 if (direction
== DIRECTION_UP
) {
3045 subtract_one
= true;
3056 last_index
= UINT64_MAX
;
3057 a
= le64toh(array
->entry_array
.next_entry_array_offset
);
3063 if (subtract_one
&& t
== 0 && i
== 0)
3066 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
3070 p
= journal_file_entry_array_item(f
, array
, 0);
3074 /* Let's cache this item for the next invocation */
3075 chain_cache_put(f
->chain_cache
, ci
, first
, a
, p
, t
, subtract_one
? (i
> 0 ? i
-1 : UINT64_MAX
) : i
);
3077 if (subtract_one
&& i
== 0)
3079 else if (subtract_one
)
3080 p
= journal_file_entry_array_item(f
, array
, i
- 1);
3082 p
= journal_file_entry_array_item(f
, array
, i
);
3085 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, ret_object
);
3094 *ret_idx
= t
+ i
+ (subtract_one
? -1 : 0);
3099 static int generic_array_bisect_plus_one(
3105 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
3106 direction_t direction
,
3107 Object
**ret_object
,
3108 uint64_t *ret_offset
) {
3113 assert(test_object
);
3118 /* This bisects the array in object 'first', but first checks an extra. */
3119 r
= test_object(f
, extra
, needle
);
3123 if (direction
== DIRECTION_DOWN
) {
3124 /* If we are going downwards, then we need to return the first object that passes the test.
3125 * When there is no object that passes the test, we need to return the first object that
3126 * test_object() returns TEST_RIGHT for. */
3128 TEST_FOUND
, /* The 'extra' object passes the test. Hence, this is the first
3129 * object that passes the test. */
3130 TEST_RIGHT
)) /* The 'extra' object is the first object that test_object() returns
3131 * TEST_RIGHT for, and no object exists even in the chained arrays
3132 * that passes the test. */
3133 goto use_extra
; /* The 'extra' object is exactly the one we are looking for. It is
3134 * not necessary to bisect the chained arrays. */
3136 /* Otherwise, the 'extra' object is not the one we are looking for. Search in the arrays. */
3139 /* If we are going upwards, then we need to return the last object that passes the test.
3140 * When there is no object that passes the test, we need to return the the last object that
3141 * test_object() returns TEST_LEFT for. */
3142 if (r
== TEST_RIGHT
)
3143 return 0; /* Not only the 'extra' object, but also all objects in the chained arrays
3144 * will never get TEST_FOUND or TEST_LEFT. The object we are looking for
3145 * does not exist. */
3147 /* Even if the 'extra' object passes the test, there may be multiple objects in the arrays
3148 * that also pass the test. Hence, we need to bisect the arrays for finding the last matching
3152 r
= generic_array_bisect(f
, first
, n
-1, needle
, test_object
, direction
, ret_object
, ret_offset
, NULL
);
3154 return r
; /* When > 0, the found object is the first (or last, when DIRECTION_UP) object.
3155 * Hence, return the found object now. */
3157 /* No matching object found in the chained arrays.
3158 * DIRECTION_DOWN : the 'extra' object neither matches the condition. There is no matching object.
3159 * DIRECTION_UP : the 'extra' object matches the condition. So, return it. */
3160 if (direction
== DIRECTION_DOWN
)
3165 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, ret_object
);
3171 *ret_offset
= extra
;
3176 static int test_object_offset(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
3182 else if (p
< needle
)
3188 int journal_file_move_to_entry_by_offset(
3191 direction_t direction
,
3192 Object
**ret_object
,
3193 uint64_t *ret_offset
) {
3198 return generic_array_bisect(
3200 le64toh(f
->header
->entry_array_offset
),
3201 le64toh(f
->header
->n_entries
),
3205 ret_object
, ret_offset
, NULL
);
3208 static int test_object_seqnum(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
3216 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
3220 sq
= le64toh(READ_NOW(o
->entry
.seqnum
));
3223 else if (sq
< needle
)
3229 int journal_file_move_to_entry_by_seqnum(
3232 direction_t direction
,
3233 Object
**ret_object
,
3234 uint64_t *ret_offset
) {
3239 return generic_array_bisect(
3241 le64toh(f
->header
->entry_array_offset
),
3242 le64toh(f
->header
->n_entries
),
3246 ret_object
, ret_offset
, NULL
);
3249 static int test_object_realtime(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
3257 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
3261 rt
= le64toh(READ_NOW(o
->entry
.realtime
));
3264 else if (rt
< needle
)
3270 int journal_file_move_to_entry_by_realtime(
3273 direction_t direction
,
3274 Object
**ret_object
,
3275 uint64_t *ret_offset
) {
3280 return generic_array_bisect(
3282 le64toh(f
->header
->entry_array_offset
),
3283 le64toh(f
->header
->n_entries
),
3285 test_object_realtime
,
3287 ret_object
, ret_offset
, NULL
);
3290 static int test_object_monotonic(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
3298 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
3302 m
= le64toh(READ_NOW(o
->entry
.monotonic
));
3305 else if (m
< needle
)
3311 static int find_data_object_by_boot_id(
3314 Object
**ret_object
,
3315 uint64_t *ret_offset
) {
3317 char t
[STRLEN("_BOOT_ID=") + 32 + 1] = "_BOOT_ID=";
3321 sd_id128_to_string(boot_id
, t
+ 9);
3322 return journal_file_find_data_object(f
, t
, sizeof(t
) - 1, ret_object
, ret_offset
);
3325 int journal_file_move_to_entry_by_monotonic(
3329 direction_t direction
,
3330 Object
**ret_object
,
3331 uint64_t *ret_offset
) {
3338 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, NULL
);
3342 return generic_array_bisect_plus_one(
3344 le64toh(o
->data
.entry_offset
),
3345 le64toh(o
->data
.entry_array_offset
),
3346 le64toh(o
->data
.n_entries
),
3348 test_object_monotonic
,
3350 ret_object
, ret_offset
);
3353 void journal_file_reset_location(JournalFile
*f
) {
3356 f
->location_type
= LOCATION_HEAD
;
3357 f
->current_offset
= 0;
3358 f
->current_seqnum
= 0;
3359 f
->current_realtime
= 0;
3360 f
->current_monotonic
= 0;
3361 zero(f
->current_boot_id
);
3362 f
->current_xor_hash
= 0;
3364 /* Also reset the previous reading direction. Otherwise, next_beyond_location() may wrongly handle we
3365 * already hit EOF. See issue #29216. */
3366 f
->last_direction
= _DIRECTION_INVALID
;
3369 void journal_file_save_location(JournalFile
*f
, Object
*o
, uint64_t offset
) {
3373 f
->location_type
= LOCATION_SEEK
;
3374 f
->current_offset
= offset
;
3375 f
->current_seqnum
= le64toh(o
->entry
.seqnum
);
3376 f
->current_realtime
= le64toh(o
->entry
.realtime
);
3377 f
->current_monotonic
= le64toh(o
->entry
.monotonic
);
3378 f
->current_boot_id
= o
->entry
.boot_id
;
3379 f
->current_xor_hash
= le64toh(o
->entry
.xor_hash
);
3382 static bool check_properly_ordered(uint64_t new_offset
, uint64_t old_offset
, direction_t direction
) {
3384 /* Consider it an error if any of the two offsets is uninitialized */
3385 if (old_offset
== 0 || new_offset
== 0)
3388 /* If we go down, the new offset must be larger than the old one. */
3389 return direction
== DIRECTION_DOWN
?
3390 new_offset
> old_offset
:
3391 new_offset
< old_offset
;
3394 int journal_file_next_entry(
3397 direction_t direction
,
3398 Object
**ret_object
,
3399 uint64_t *ret_offset
) {
3408 /* FIXME: fix return value assignment. */
3410 n
= le64toh(READ_NOW(f
->header
->n_entries
));
3414 /* When the input offset 'p' is zero, return the first (or last on DIRECTION_UP) entry. */
3416 return generic_array_get(f
,
3417 le64toh(f
->header
->entry_array_offset
),
3418 direction
== DIRECTION_DOWN
? 0 : n
- 1,
3420 ret_object
, ret_offset
);
3422 /* Otherwise, first find the nearest entry object. */
3423 r
= generic_array_bisect(f
,
3424 le64toh(f
->header
->entry_array_offset
),
3425 le64toh(f
->header
->n_entries
),
3429 ret_object
? &o
: NULL
, &q
, &i
);
3433 assert(direction
== DIRECTION_DOWN
? p
<= q
: q
<= p
);
3435 /* If the input offset 'p' points to an entry object, generic_array_bisect() should provides
3436 * the same offset, and the index needs to be shifted. Otherwise, use the found object as is,
3437 * as it is the nearest entry object from the input offset 'p'. */
3442 r
= bump_array_index(&i
, direction
, n
);
3446 /* And jump to it */
3447 r
= generic_array_get(f
, le64toh(f
->header
->entry_array_offset
), i
, direction
, ret_object
? &o
: NULL
, &q
);
3451 /* Ensure our array is properly ordered. */
3452 if (!check_properly_ordered(q
, p
, direction
))
3453 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
3454 "%s: entry array not properly ordered at entry index %" PRIu64
,
3465 int journal_file_move_to_entry_for_data(
3468 direction_t direction
,
3469 Object
**ret_object
,
3470 uint64_t *ret_offset
) {
3472 uint64_t extra
, first
, n
;
3477 assert(d
->object
.type
== OBJECT_DATA
);
3478 assert(IN_SET(direction
, DIRECTION_DOWN
, DIRECTION_UP
));
3480 /* FIXME: fix return value assignment. */
3482 /* This returns the first (when the direction is down, otherwise the last) entry linked to the
3483 * specified data object. */
3485 n
= le64toh(d
->data
.n_entries
);
3488 n
--; /* n_entries is the number of entries linked to the data object, including the 'extra' entry. */
3490 extra
= le64toh(d
->data
.entry_offset
);
3491 first
= le64toh(d
->data
.entry_array_offset
);
3493 if (direction
== DIRECTION_DOWN
&& extra
> 0) {
3494 /* When we are going downwards, first try to read the extra entry. */
3495 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, ret_object
);
3498 if (!IN_SET(r
, -EADDRNOTAVAIL
, -EBADMSG
))
3503 /* DIRECTION_DOWN : The extra entry is broken, falling back to the entries in the array.
3504 * DIRECTION_UP : Try to find a valid entry in the array from the tail. */
3505 r
= generic_array_get(f
,
3507 direction
== DIRECTION_DOWN
? 0 : n
- 1,
3509 ret_object
, ret_offset
);
3510 if (!IN_SET(r
, 0, -EADDRNOTAVAIL
, -EBADMSG
))
3511 return r
; /* found or critical error. */
3514 if (direction
== DIRECTION_UP
&& extra
> 0) {
3515 /* No valid entry exists in the chained array, falling back to the extra entry. */
3516 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, ret_object
);
3525 *ret_offset
= extra
;
3530 int journal_file_move_to_entry_by_offset_for_data(
3534 direction_t direction
,
3535 Object
**ret
, uint64_t *ret_offset
) {
3539 assert(d
->object
.type
== OBJECT_DATA
);
3541 return generic_array_bisect_plus_one(
3543 le64toh(d
->data
.entry_offset
),
3544 le64toh(d
->data
.entry_array_offset
),
3545 le64toh(d
->data
.n_entries
),
3552 int journal_file_move_to_entry_by_monotonic_for_data(
3557 direction_t direction
,
3558 Object
**ret_object
,
3559 uint64_t *ret_offset
) {
3561 uint64_t z
, entry_offset
, entry_array_offset
, n_entries
;
3567 assert(d
->object
.type
== OBJECT_DATA
);
3569 /* Save all the required data before the data object gets invalidated. */
3570 entry_offset
= le64toh(READ_NOW(d
->data
.entry_offset
));
3571 entry_array_offset
= le64toh(READ_NOW(d
->data
.entry_array_offset
));
3572 n_entries
= le64toh(READ_NOW(d
->data
.n_entries
));
3574 /* First, seek by time */
3575 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, NULL
);
3579 r
= generic_array_bisect_plus_one(f
,
3580 le64toh(o
->data
.entry_offset
),
3581 le64toh(o
->data
.entry_array_offset
),
3582 le64toh(o
->data
.n_entries
),
3584 test_object_monotonic
,
3590 /* And now, continue seeking until we find an entry that exists in both bisection arrays. */
3594 /* The journal entry found by the above bisect_plus_one() may not have the specified data,
3595 * that is, it may not be linked in the data object. So, we need to check that. */
3597 r
= generic_array_bisect_plus_one(f
,
3604 ret_object
? &entry
: NULL
, &p
);
3608 break; /* The journal entry has the specified data. Yay! */
3610 /* If the entry does not have the data, then move to the next (or previous, depends on the
3611 * 'direction') entry linked to the data object. But, the next entry may be in another boot.
3612 * So, we need to check that the entry has the matching boot ID. */
3614 r
= generic_array_bisect_plus_one(f
,
3615 le64toh(o
->data
.entry_offset
),
3616 le64toh(o
->data
.entry_array_offset
),
3617 le64toh(o
->data
.n_entries
),
3621 ret_object
? &entry
: NULL
, &z
);
3625 break; /* The journal entry has the specified boot ID. Yay! */
3627 /* If not, let's try to the next entry... */
3631 *ret_object
= entry
;
3637 int journal_file_move_to_entry_by_seqnum_for_data(
3641 direction_t direction
,
3642 Object
**ret_object
,
3643 uint64_t *ret_offset
) {
3647 assert(d
->object
.type
== OBJECT_DATA
);
3649 return generic_array_bisect_plus_one(
3651 le64toh(d
->data
.entry_offset
),
3652 le64toh(d
->data
.entry_array_offset
),
3653 le64toh(d
->data
.n_entries
),
3657 ret_object
, ret_offset
);
3660 int journal_file_move_to_entry_by_realtime_for_data(
3664 direction_t direction
,
3665 Object
**ret
, uint64_t *ret_offset
) {
3669 assert(d
->object
.type
== OBJECT_DATA
);
3671 return generic_array_bisect_plus_one(
3673 le64toh(d
->data
.entry_offset
),
3674 le64toh(d
->data
.entry_array_offset
),
3675 le64toh(d
->data
.n_entries
),
3677 test_object_realtime
,
3682 void journal_file_dump(JournalFile
*f
) {
3690 journal_file_print_header(f
);
3692 p
= le64toh(READ_NOW(f
->header
->header_size
));
3697 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &o
);
3701 s
= journal_object_type_to_string(o
->object
.type
);
3703 switch (o
->object
.type
) {
3708 printf("Type: %s seqnum=%"PRIu64
" monotonic=%"PRIu64
" realtime=%"PRIu64
"\n",
3710 le64toh(o
->entry
.seqnum
),
3711 le64toh(o
->entry
.monotonic
),
3712 le64toh(o
->entry
.realtime
));
3718 printf("Type: %s seqnum=%"PRIu64
" epoch=%"PRIu64
"\n",
3720 le64toh(o
->tag
.seqnum
),
3721 le64toh(o
->tag
.epoch
));
3726 printf("Type: %s \n", s
);
3728 printf("Type: unknown (%i)", o
->object
.type
);
3733 c
= COMPRESSION_FROM_OBJECT(o
);
3734 if (c
> COMPRESSION_NONE
)
3735 printf("Flags: %s\n",
3736 compression_to_string(c
));
3738 if (p
== le64toh(f
->header
->tail_object_offset
))
3741 p
+= ALIGN64(le64toh(o
->object
.size
));
3746 log_error("File corrupt");
3749 /* Note: the lifetime of the compound literal is the immediately surrounding block. */
3750 #define FORMAT_TIMESTAMP_SAFE(t) (FORMAT_TIMESTAMP(t) ?: " --- ")
3752 void journal_file_print_header(JournalFile
*f
) {
3758 printf("File path: %s\n"
3762 "Sequential number ID: %s\n"
3764 "Compatible flags:%s%s%s\n"
3765 "Incompatible flags:%s%s%s%s%s%s\n"
3766 "Header size: %"PRIu64
"\n"
3767 "Arena size: %"PRIu64
"\n"
3768 "Data hash table size: %"PRIu64
"\n"
3769 "Field hash table size: %"PRIu64
"\n"
3770 "Rotate suggested: %s\n"
3771 "Head sequential number: %"PRIu64
" (%"PRIx64
")\n"
3772 "Tail sequential number: %"PRIu64
" (%"PRIx64
")\n"
3773 "Head realtime timestamp: %s (%"PRIx64
")\n"
3774 "Tail realtime timestamp: %s (%"PRIx64
")\n"
3775 "Tail monotonic timestamp: %s (%"PRIx64
")\n"
3776 "Objects: %"PRIu64
"\n"
3777 "Entry objects: %"PRIu64
"\n",
3779 SD_ID128_TO_STRING(f
->header
->file_id
),
3780 SD_ID128_TO_STRING(f
->header
->machine_id
),
3781 SD_ID128_TO_STRING(f
->header
->tail_entry_boot_id
),
3782 SD_ID128_TO_STRING(f
->header
->seqnum_id
),
3783 f
->header
->state
== STATE_OFFLINE
? "OFFLINE" :
3784 f
->header
->state
== STATE_ONLINE
? "ONLINE" :
3785 f
->header
->state
== STATE_ARCHIVED
? "ARCHIVED" : "UNKNOWN",
3786 JOURNAL_HEADER_SEALED(f
->header
) ? " SEALED" : "",
3787 JOURNAL_HEADER_TAIL_ENTRY_BOOT_ID(f
->header
) ? " TAIL_ENTRY_BOOT_ID" : "",
3788 (le32toh(f
->header
->compatible_flags
) & ~HEADER_COMPATIBLE_ANY
) ? " ???" : "",
3789 JOURNAL_HEADER_COMPRESSED_XZ(f
->header
) ? " COMPRESSED-XZ" : "",
3790 JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
) ? " COMPRESSED-LZ4" : "",
3791 JOURNAL_HEADER_COMPRESSED_ZSTD(f
->header
) ? " COMPRESSED-ZSTD" : "",
3792 JOURNAL_HEADER_KEYED_HASH(f
->header
) ? " KEYED-HASH" : "",
3793 JOURNAL_HEADER_COMPACT(f
->header
) ? " COMPACT" : "",
3794 (le32toh(f
->header
->incompatible_flags
) & ~HEADER_INCOMPATIBLE_ANY
) ? " ???" : "",
3795 le64toh(f
->header
->header_size
),
3796 le64toh(f
->header
->arena_size
),
3797 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3798 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
),
3799 yes_no(journal_file_rotate_suggested(f
, 0, LOG_DEBUG
)),
3800 le64toh(f
->header
->head_entry_seqnum
), le64toh(f
->header
->head_entry_seqnum
),
3801 le64toh(f
->header
->tail_entry_seqnum
), le64toh(f
->header
->tail_entry_seqnum
),
3802 FORMAT_TIMESTAMP_SAFE(le64toh(f
->header
->head_entry_realtime
)), le64toh(f
->header
->head_entry_realtime
),
3803 FORMAT_TIMESTAMP_SAFE(le64toh(f
->header
->tail_entry_realtime
)), le64toh(f
->header
->tail_entry_realtime
),
3804 FORMAT_TIMESPAN(le64toh(f
->header
->tail_entry_monotonic
), USEC_PER_MSEC
), le64toh(f
->header
->tail_entry_monotonic
),
3805 le64toh(f
->header
->n_objects
),
3806 le64toh(f
->header
->n_entries
));
3808 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3809 printf("Data objects: %"PRIu64
"\n"
3810 "Data hash table fill: %.1f%%\n",
3811 le64toh(f
->header
->n_data
),
3812 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))));
3814 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3815 printf("Field objects: %"PRIu64
"\n"
3816 "Field hash table fill: %.1f%%\n",
3817 le64toh(f
->header
->n_fields
),
3818 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))));
3820 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
))
3821 printf("Tag objects: %"PRIu64
"\n",
3822 le64toh(f
->header
->n_tags
));
3823 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
3824 printf("Entry array objects: %"PRIu64
"\n",
3825 le64toh(f
->header
->n_entry_arrays
));
3827 if (JOURNAL_HEADER_CONTAINS(f
->header
, field_hash_chain_depth
))
3828 printf("Deepest field hash chain: %" PRIu64
"\n",
3829 f
->header
->field_hash_chain_depth
);
3831 if (JOURNAL_HEADER_CONTAINS(f
->header
, data_hash_chain_depth
))
3832 printf("Deepest data hash chain: %" PRIu64
"\n",
3833 f
->header
->data_hash_chain_depth
);
3835 if (fstat(f
->fd
, &st
) >= 0)
3836 printf("Disk usage: %s\n", FORMAT_BYTES((uint64_t) st
.st_blocks
* 512ULL));
3839 static int journal_file_warn_btrfs(JournalFile
*f
) {
3845 /* Before we write anything, check if the COW logic is turned
3846 * off on btrfs. Given our write pattern that is quite
3847 * unfriendly to COW file systems this should greatly improve
3848 * performance on COW file systems, such as btrfs, at the
3849 * expense of data integrity features (which shouldn't be too
3850 * bad, given that we do our own checksumming). */
3852 r
= fd_is_fs_type(f
->fd
, BTRFS_SUPER_MAGIC
);
3854 return log_ratelimit_warning_errno(r
, JOURNAL_LOG_RATELIMIT
, "Failed to determine if journal is on btrfs: %m");
3858 r
= read_attr_fd(f
->fd
, &attrs
);
3860 return log_ratelimit_warning_errno(r
, JOURNAL_LOG_RATELIMIT
, "Failed to read file attributes: %m");
3862 if (attrs
& FS_NOCOW_FL
) {
3863 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3867 log_ratelimit_notice(JOURNAL_LOG_RATELIMIT
,
3868 "Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3869 "This is likely to slow down journal access substantially, please consider turning "
3870 "off the copy-on-write file attribute on the journal directory, using chattr +C.",
3876 static void journal_default_metrics(JournalMetrics
*m
, int fd
, bool compact
) {
3878 uint64_t fs_size
= 0;
3883 if (fstatvfs(fd
, &ss
) >= 0)
3884 fs_size
= ss
.f_frsize
* ss
.f_blocks
;
3886 log_debug_errno(errno
, "Failed to determine disk size: %m");
3888 if (m
->max_use
== UINT64_MAX
) {
3891 m
->max_use
= CLAMP(PAGE_ALIGN_U64(fs_size
/ 10), /* 10% of file system size */
3892 MAX_USE_LOWER
, MAX_USE_UPPER
);
3894 m
->max_use
= MAX_USE_LOWER
;
3896 m
->max_use
= PAGE_ALIGN_U64(m
->max_use
);
3898 if (m
->max_use
!= 0 && m
->max_use
< JOURNAL_FILE_SIZE_MIN
*2)
3899 m
->max_use
= JOURNAL_FILE_SIZE_MIN
*2;
3902 if (m
->min_use
== UINT64_MAX
) {
3904 m
->min_use
= CLAMP(PAGE_ALIGN_U64(fs_size
/ 50), /* 2% of file system size */
3905 MIN_USE_LOW
, MIN_USE_HIGH
);
3907 m
->min_use
= MIN_USE_LOW
;
3910 if (m
->min_use
> m
->max_use
)
3911 m
->min_use
= m
->max_use
;
3913 if (m
->max_size
== UINT64_MAX
)
3914 m
->max_size
= MIN(PAGE_ALIGN_U64(m
->max_use
/ 8), /* 8 chunks */
3917 m
->max_size
= PAGE_ALIGN_U64(m
->max_size
);
3919 if (compact
&& m
->max_size
> JOURNAL_COMPACT_SIZE_MAX
)
3920 m
->max_size
= JOURNAL_COMPACT_SIZE_MAX
;
3922 if (m
->max_size
!= 0) {
3923 if (m
->max_size
< JOURNAL_FILE_SIZE_MIN
)
3924 m
->max_size
= JOURNAL_FILE_SIZE_MIN
;
3926 if (m
->max_use
!= 0 && m
->max_size
*2 > m
->max_use
)
3927 m
->max_use
= m
->max_size
*2;
3930 if (m
->min_size
== UINT64_MAX
)
3931 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3933 m
->min_size
= CLAMP(PAGE_ALIGN_U64(m
->min_size
),
3934 JOURNAL_FILE_SIZE_MIN
,
3935 m
->max_size
?: UINT64_MAX
);
3937 if (m
->keep_free
== UINT64_MAX
) {
3939 m
->keep_free
= MIN(PAGE_ALIGN_U64(fs_size
/ 20), /* 5% of file system size */
3942 m
->keep_free
= DEFAULT_KEEP_FREE
;
3945 if (m
->n_max_files
== UINT64_MAX
)
3946 m
->n_max_files
= DEFAULT_N_MAX_FILES
;
3948 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64
,
3949 FORMAT_BYTES(m
->min_use
),
3950 FORMAT_BYTES(m
->max_use
),
3951 FORMAT_BYTES(m
->max_size
),
3952 FORMAT_BYTES(m
->min_size
),
3953 FORMAT_BYTES(m
->keep_free
),
3957 int journal_file_open(
3961 JournalFileFlags file_flags
,
3963 uint64_t compress_threshold_bytes
,
3964 JournalMetrics
*metrics
,
3965 MMapCache
*mmap_cache
,
3966 JournalFile
*template,
3967 JournalFile
**ret
) {
3969 bool newly_created
= false;
3974 assert(fd
>= 0 || fname
);
3975 assert(file_flags
>= 0);
3976 assert(file_flags
<= _JOURNAL_FILE_FLAGS_MAX
);
3980 if (!IN_SET((open_flags
& O_ACCMODE
), O_RDONLY
, O_RDWR
))
3983 if ((open_flags
& O_ACCMODE
) == O_RDONLY
&& FLAGS_SET(open_flags
, O_CREAT
))
3986 if (fname
&& (open_flags
& O_CREAT
) && !endswith(fname
, ".journal"))
3989 f
= new(JournalFile
, 1);
3993 *f
= (JournalFile
) {
3996 .open_flags
= open_flags
,
3997 .compress_threshold_bytes
= compress_threshold_bytes
== UINT64_MAX
?
3998 DEFAULT_COMPRESS_THRESHOLD
:
3999 MAX(MIN_COMPRESS_THRESHOLD
, compress_threshold_bytes
),
4000 .strict_order
= FLAGS_SET(file_flags
, JOURNAL_STRICT_ORDER
),
4001 .newest_boot_id_prioq_idx
= PRIOQ_IDX_NULL
,
4002 .last_direction
= _DIRECTION_INVALID
,
4006 f
->path
= strdup(fname
);
4014 /* If we don't know the path, fill in something explanatory and vaguely useful */
4015 if (asprintf(&f
->path
, "/proc/self/%i", fd
) < 0) {
4021 f
->chain_cache
= ordered_hashmap_new(&uint64_hash_ops
);
4022 if (!f
->chain_cache
) {
4028 /* We pass O_NONBLOCK here, so that in case somebody pointed us to some character device node or FIFO
4029 * or so, we likely fail quickly than block for long. For regular files O_NONBLOCK has no effect, hence
4030 * it doesn't hurt in that case. */
4032 f
->fd
= openat_report_new(AT_FDCWD
, f
->path
, f
->open_flags
|O_CLOEXEC
|O_NONBLOCK
, f
->mode
, &newly_created
);
4038 /* fds we opened here by us should also be closed by us. */
4041 r
= fd_nonblock(f
->fd
, false);
4045 if (!newly_created
) {
4046 r
= journal_file_fstat(f
);
4051 r
= journal_file_fstat(f
);
4055 /* If we just got the fd passed in, we don't really know if we created the file anew */
4056 newly_created
= f
->last_stat
.st_size
== 0 && journal_file_writable(f
);
4059 r
= mmap_cache_add_fd(mmap_cache
, f
->fd
, mmap_prot_from_open_flags(open_flags
), &f
->cache_fd
);
4063 if (newly_created
) {
4064 (void) journal_file_warn_btrfs(f
);
4066 /* Let's attach the creation time to the journal file, so that the vacuuming code knows the age of this
4067 * file even if the file might end up corrupted one day... Ideally we'd just use the creation time many
4068 * file systems maintain for each file, but the API to query this is very new, hence let's emulate this
4069 * via extended attributes. If extended attributes are not supported we'll just skip this, and rely
4070 * solely on mtime/atime/ctime of the file. */
4071 (void) fd_setcrtime(f
->fd
, 0);
4073 r
= journal_file_init_header(f
, file_flags
, template);
4077 r
= journal_file_fstat(f
);
4082 if (f
->last_stat
.st_size
< (off_t
) HEADER_SIZE_MIN
) {
4087 r
= mmap_cache_fd_get(f
->cache_fd
, CONTEXT_HEADER
, true, 0, PAGE_ALIGN(sizeof(Header
)), &f
->last_stat
, &h
);
4089 /* Some file systems (jffs2 or p9fs) don't support mmap() properly (or only read-only
4090 * mmap()), and return EINVAL in that case. Let's propagate that as a more recognizable error
4100 if (!newly_created
) {
4101 r
= journal_file_verify_header(f
);
4107 if (!newly_created
&& journal_file_writable(f
) && JOURNAL_HEADER_SEALED(f
->header
)) {
4108 r
= journal_file_fss_load(f
);
4114 if (journal_file_writable(f
)) {
4116 journal_default_metrics(metrics
, f
->fd
, JOURNAL_HEADER_COMPACT(f
->header
));
4117 f
->metrics
= *metrics
;
4118 } else if (template)
4119 f
->metrics
= template->metrics
;
4121 r
= journal_file_refresh_header(f
);
4127 r
= journal_file_hmac_setup(f
);
4132 if (newly_created
) {
4133 r
= journal_file_setup_field_hash_table(f
);
4137 r
= journal_file_setup_data_hash_table(f
);
4142 r
= journal_file_append_first_tag(f
);
4148 if (mmap_cache_fd_got_sigbus(f
->cache_fd
)) {
4153 if (template && template->post_change_timer
) {
4154 r
= journal_file_enable_post_change_timer(
4156 sd_event_source_get_event(template->post_change_timer
),
4157 template->post_change_timer_period
);
4163 /* The file is opened now successfully, thus we take possession of any passed in fd. */
4166 if (DEBUG_LOGGING
) {
4167 static int last_seal
= -1, last_keyed_hash
= -1;
4168 static Compression last_compression
= _COMPRESSION_INVALID
;
4169 static uint64_t last_bytes
= UINT64_MAX
;
4171 if (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
) {
4176 log_debug("Journal effective settings seal=%s keyed_hash=%s compress=%s compress_threshold_bytes=%s",
4177 yes_no(JOURNAL_HEADER_SEALED(f
->header
)), yes_no(JOURNAL_HEADER_KEYED_HASH(f
->header
)),
4178 compression_to_string(JOURNAL_FILE_COMPRESSION(f
)), FORMAT_BYTES(f
->compress_threshold_bytes
));
4179 last_seal
= JOURNAL_HEADER_SEALED(f
->header
);
4180 last_keyed_hash
= JOURNAL_HEADER_KEYED_HASH(f
->header
);
4181 last_compression
= JOURNAL_FILE_COMPRESSION(f
);
4182 last_bytes
= f
->compress_threshold_bytes
;
4190 if (f
->cache_fd
&& mmap_cache_fd_got_sigbus(f
->cache_fd
))
4193 (void) journal_file_close(f
);
4195 if (newly_created
&& fd
< 0)
4196 (void) unlink(fname
);
4201 int journal_file_parse_uid_from_filename(const char *path
, uid_t
*ret_uid
) {
4202 _cleanup_free_
char *buf
= NULL
, *p
= NULL
;
4203 const char *a
, *b
, *at
;
4206 /* This helper returns -EREMOTE when the filename doesn't match user online/offline journal
4207 * pattern. Hence it currently doesn't parse archived or disposed user journals. */
4212 r
= path_extract_filename(path
, &p
);
4215 if (r
== O_DIRECTORY
)
4218 a
= startswith(p
, "user-");
4221 b
= endswith(p
, ".journal");
4225 at
= strchr(a
, '@');
4229 buf
= strndup(a
, b
-a
);
4233 return parse_uid(buf
, ret_uid
);
4236 int journal_file_archive(JournalFile
*f
, char **ret_previous_path
) {
4237 _cleanup_free_
char *p
= NULL
;
4241 if (!journal_file_writable(f
))
4244 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
4245 * rotation, since we don't know the actual path, and couldn't rename the file hence. */
4246 if (path_startswith(f
->path
, "/proc/self/fd"))
4249 if (!endswith(f
->path
, ".journal"))
4252 if (asprintf(&p
, "%.*s@" SD_ID128_FORMAT_STR
"-%016"PRIx64
"-%016"PRIx64
".journal",
4253 (int) strlen(f
->path
) - 8, f
->path
,
4254 SD_ID128_FORMAT_VAL(f
->header
->seqnum_id
),
4255 le64toh(f
->header
->head_entry_seqnum
),
4256 le64toh(f
->header
->head_entry_realtime
)) < 0)
4259 /* Try to rename the file to the archived version. If the file already was deleted, we'll get ENOENT, let's
4260 * ignore that case. */
4261 if (rename(f
->path
, p
) < 0 && errno
!= ENOENT
)
4264 /* Sync the rename to disk */
4265 (void) fsync_directory_of_file(f
->fd
);
4267 if (ret_previous_path
)
4268 *ret_previous_path
= f
->path
;
4272 f
->path
= TAKE_PTR(p
);
4274 /* Set as archive so offlining commits w/state=STATE_ARCHIVED. Previously we would set old_file->header->state
4275 * to STATE_ARCHIVED directly here, but journal_file_set_offline() short-circuits when state != STATE_ONLINE,
4276 * which would result in the rotated journal never getting fsync() called before closing. Now we simply queue
4277 * the archive state by setting an archive bit, leaving the state as STATE_ONLINE so proper offlining
4284 int journal_file_dispose(int dir_fd
, const char *fname
) {
4285 _cleanup_free_
char *p
= NULL
;
4289 /* Renames a journal file to *.journal~, i.e. to mark it as corrupted or otherwise uncleanly shutdown. Note that
4290 * this is done without looking into the file or changing any of its contents. The idea is that this is called
4291 * whenever something is suspicious and we want to move the file away and make clear that it is not accessed
4292 * for writing anymore. */
4294 if (!endswith(fname
, ".journal"))
4297 if (asprintf(&p
, "%.*s@%016" PRIx64
"-%016" PRIx64
".journal~",
4298 (int) strlen(fname
) - 8, fname
,
4299 now(CLOCK_REALTIME
),
4303 if (renameat(dir_fd
, fname
, dir_fd
, p
) < 0)
4309 int journal_file_copy_entry(
4315 sd_id128_t
*seqnum_id
) {
4317 _cleanup_free_ EntryItem
*items_alloc
= NULL
;
4319 uint64_t n
, m
= 0, xor_hash
= 0;
4329 if (!journal_file_writable(to
))
4332 ts
= (dual_timestamp
) {
4333 .monotonic
= le64toh(o
->entry
.monotonic
),
4334 .realtime
= le64toh(o
->entry
.realtime
),
4336 boot_id
= o
->entry
.boot_id
;
4338 n
= journal_file_entry_n_items(from
, o
);
4342 if (n
< ALLOCA_MAX
/ sizeof(EntryItem
) / 2)
4343 items
= newa(EntryItem
, n
);
4345 items_alloc
= new(EntryItem
, n
);
4349 items
= items_alloc
;
4352 for (uint64_t i
= 0; i
< n
; i
++) {
4358 q
= journal_file_entry_item_object_offset(from
, o
, i
);
4359 r
= journal_file_data_payload(from
, NULL
, q
, NULL
, 0, 0, &data
, &l
);
4360 if (IN_SET(r
, -EADDRNOTAVAIL
, -EBADMSG
)) {
4361 log_debug_errno(r
, "Entry item %"PRIu64
" data object is bad, skipping over it: %m", i
);
4371 r
= journal_file_append_data(to
, data
, l
, &u
, &h
);
4375 if (JOURNAL_HEADER_KEYED_HASH(to
->header
))
4376 xor_hash
^= jenkins_hash64(data
, l
);
4378 xor_hash
^= le64toh(u
->data
.hash
);
4380 items
[m
++] = (EntryItem
) {
4382 .hash
= le64toh(u
->data
.hash
),
4389 r
= journal_file_append_entry_internal(
4393 &from
->header
->machine_id
,
4399 /* ret_object= */ NULL
,
4400 /* ret_offset= */ NULL
);
4402 if (mmap_cache_fd_got_sigbus(to
->cache_fd
))
4408 void journal_reset_metrics(JournalMetrics
*m
) {
4411 /* Set everything to "pick automatic values". */
4413 *m
= (JournalMetrics
) {
4414 .min_use
= UINT64_MAX
,
4415 .max_use
= UINT64_MAX
,
4416 .min_size
= UINT64_MAX
,
4417 .max_size
= UINT64_MAX
,
4418 .keep_free
= UINT64_MAX
,
4419 .n_max_files
= UINT64_MAX
,
4423 int journal_file_get_cutoff_realtime_usec(JournalFile
*f
, usec_t
*ret_from
, usec_t
*ret_to
) {
4426 assert(ret_from
|| ret_to
);
4429 if (f
->header
->head_entry_realtime
== 0)
4432 *ret_from
= le64toh(f
->header
->head_entry_realtime
);
4436 if (f
->header
->tail_entry_realtime
== 0)
4439 *ret_to
= le64toh(f
->header
->tail_entry_realtime
);
4445 int journal_file_get_cutoff_monotonic_usec(JournalFile
*f
, sd_id128_t boot_id
, usec_t
*ret_from
, usec_t
*ret_to
) {
4451 assert(ret_from
|| ret_to
);
4453 /* FIXME: fix return value assignment on success with 0. */
4455 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &p
);
4459 if (le64toh(o
->data
.n_entries
) <= 0)
4463 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, le64toh(o
->data
.entry_offset
), &o
);
4467 *ret_from
= le64toh(o
->entry
.monotonic
);
4471 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
4475 r
= journal_file_move_to_entry_for_data(f
, o
, DIRECTION_UP
, &o
, NULL
);
4479 *ret_to
= le64toh(o
->entry
.monotonic
);
4485 bool journal_file_rotate_suggested(JournalFile
*f
, usec_t max_file_usec
, int log_level
) {
4489 /* If we gained new header fields we gained new features,
4490 * hence suggest a rotation */
4491 if (le64toh(f
->header
->header_size
) < sizeof(Header
)) {
4492 log_ratelimit_full(log_level
, JOURNAL_LOG_RATELIMIT
,
4493 "%s uses an outdated header, suggesting rotation.", f
->path
);
4497 /* Let's check if the hash tables grew over a certain fill level (75%, borrowing this value from
4498 * Java's hash table implementation), and if so suggest a rotation. To calculate the fill level we
4499 * need the n_data field, which only exists in newer versions. */
4501 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
4502 if (le64toh(f
->header
->n_data
) * 4ULL > (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
4504 log_level
, JOURNAL_LOG_RATELIMIT
,
4505 "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.",
4507 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))),
4508 le64toh(f
->header
->n_data
),
4509 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
4510 (uint64_t) f
->last_stat
.st_size
,
4511 f
->last_stat
.st_size
/ le64toh(f
->header
->n_data
));
4515 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
4516 if (le64toh(f
->header
->n_fields
) * 4ULL > (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
4518 log_level
, JOURNAL_LOG_RATELIMIT
,
4519 "Field hash table of %s has a fill level at %.1f (%"PRIu64
" of %"PRIu64
" items), suggesting rotation.",
4521 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))),
4522 le64toh(f
->header
->n_fields
),
4523 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
));
4527 /* If there are too many hash collisions somebody is most likely playing games with us. Hence, if our
4528 * longest chain is longer than some threshold, let's suggest rotation. */
4529 if (JOURNAL_HEADER_CONTAINS(f
->header
, data_hash_chain_depth
) &&
4530 le64toh(f
->header
->data_hash_chain_depth
) > HASH_CHAIN_DEPTH_MAX
) {
4532 log_level
, JOURNAL_LOG_RATELIMIT
,
4533 "Data hash table of %s has deepest hash chain of length %" PRIu64
", suggesting rotation.",
4534 f
->path
, le64toh(f
->header
->data_hash_chain_depth
));
4538 if (JOURNAL_HEADER_CONTAINS(f
->header
, field_hash_chain_depth
) &&
4539 le64toh(f
->header
->field_hash_chain_depth
) > HASH_CHAIN_DEPTH_MAX
) {
4541 log_level
, JOURNAL_LOG_RATELIMIT
,
4542 "Field hash table of %s has deepest hash chain of length at %" PRIu64
", suggesting rotation.",
4543 f
->path
, le64toh(f
->header
->field_hash_chain_depth
));
4547 /* Are the data objects properly indexed by field objects? */
4548 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
4549 JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
4550 le64toh(f
->header
->n_data
) > 0 &&
4551 le64toh(f
->header
->n_fields
) == 0) {
4553 log_level
, JOURNAL_LOG_RATELIMIT
,
4554 "Data objects of %s are not indexed by field objects, suggesting rotation.",
4559 if (max_file_usec
> 0) {
4562 h
= le64toh(f
->header
->head_entry_realtime
);
4563 t
= now(CLOCK_REALTIME
);
4565 if (h
> 0 && t
> h
+ max_file_usec
) {
4567 log_level
, JOURNAL_LOG_RATELIMIT
,
4568 "Oldest entry in %s is older than the configured file retention duration (%s), suggesting rotation.",
4569 f
->path
, FORMAT_TIMESPAN(max_file_usec
, USEC_PER_SEC
));
4577 static const char * const journal_object_type_table
[] = {
4578 [OBJECT_UNUSED
] = "unused",
4579 [OBJECT_DATA
] = "data",
4580 [OBJECT_FIELD
] = "field",
4581 [OBJECT_ENTRY
] = "entry",
4582 [OBJECT_DATA_HASH_TABLE
] = "data hash table",
4583 [OBJECT_FIELD_HASH_TABLE
] = "field hash table",
4584 [OBJECT_ENTRY_ARRAY
] = "entry array",
4585 [OBJECT_TAG
] = "tag",
4588 DEFINE_STRING_TABLE_LOOKUP_TO_STRING(journal_object_type
, ObjectType
);