1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
6 #include <linux/magic.h>
10 #include <sys/statvfs.h>
16 #include "alloc-util.h"
17 #include "chattr-util.h"
21 #include "format-util.h"
23 #include "id128-util.h"
24 #include "journal-authenticate.h"
25 #include "journal-def.h"
26 #include "journal-file.h"
27 #include "journal-internal.h"
29 #include "memory-util.h"
30 #include "missing_threads.h"
31 #include "path-util.h"
33 #include "random-util.h"
35 #include "sort-util.h"
36 #include "stat-util.h"
37 #include "string-table.h"
38 #include "string-util.h"
40 #include "sync-util.h"
41 #include "user-util.h"
42 #include "xattr-util.h"
44 #define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
45 #define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
47 #define DEFAULT_COMPRESS_THRESHOLD (512ULL)
48 #define MIN_COMPRESS_THRESHOLD (8ULL)
50 #define U64_KB UINT64_C(1024)
51 #define U64_MB (UINT64_C(1024) * U64_KB)
52 #define U64_GB (UINT64_C(1024) * U64_MB)
54 /* This is the minimum journal file size */
55 #define JOURNAL_FILE_SIZE_MIN (512 * U64_KB) /* 512 KiB */
56 #define JOURNAL_COMPACT_SIZE_MAX ((uint64_t) UINT32_MAX) /* 4 GiB */
58 /* These are the lower and upper bounds if we deduce the max_use value from the file system size */
59 #define MAX_USE_LOWER (1 * U64_MB) /* 1 MiB */
60 #define MAX_USE_UPPER (4 * U64_GB) /* 4 GiB */
62 /* Those are the lower and upper bounds for the minimal use limit,
63 * i.e. how much we'll use even if keep_free suggests otherwise. */
64 #define MIN_USE_LOW (1 * U64_MB) /* 1 MiB */
65 #define MIN_USE_HIGH (16 * U64_MB) /* 16 MiB */
67 /* This is the upper bound if we deduce max_size from max_use */
68 #define MAX_SIZE_UPPER (128 * U64_MB) /* 128 MiB */
70 /* This is the upper bound if we deduce the keep_free value from the file system size */
71 #define KEEP_FREE_UPPER (4 * U64_GB) /* 4 GiB */
73 /* This is the keep_free value when we can't determine the system size */
74 #define DEFAULT_KEEP_FREE (1 * U64_MB) /* 1 MB */
76 /* This is the default maximum number of journal files to keep around. */
77 #define DEFAULT_N_MAX_FILES 100
79 /* n_data was the first entry we added after the initial file format design */
80 #define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
82 /* How many entries to keep in the entry array chain cache at max */
83 #define CHAIN_CACHE_MAX 20
85 /* How much to increase the journal file size at once each time we allocate something new. */
86 #define FILE_SIZE_INCREASE (8 * U64_MB) /* 8MB */
88 /* Reread fstat() of the file for detecting deletions at least this often */
89 #define LAST_STAT_REFRESH_USEC (5*USEC_PER_SEC)
91 /* Longest hash chain to rotate after */
92 #define HASH_CHAIN_DEPTH_MAX 100
95 # pragma GCC diagnostic ignored "-Waddress-of-packed-member"
98 static int mmap_prot_from_open_flags(int flags
) {
99 switch (flags
& O_ACCMODE
) {
105 return PROT_READ
|PROT_WRITE
;
107 assert_not_reached();
111 int journal_file_tail_end_by_pread(JournalFile
*f
, uint64_t *ret_offset
) {
119 /* Same as journal_file_tail_end_by_mmap() below, but operates with pread() to avoid the mmap cache
120 * (and thus is thread safe) */
122 p
= le64toh(f
->header
->tail_object_offset
);
124 p
= le64toh(f
->header
->header_size
);
129 r
= journal_file_read_object_header(f
, OBJECT_UNUSED
, p
, &tail
);
133 sz
= le64toh(tail
.object
.size
);
134 if (sz
> UINT64_MAX
- sizeof(uint64_t) + 1)
138 if (p
> UINT64_MAX
- sz
)
149 int journal_file_tail_end_by_mmap(JournalFile
*f
, uint64_t *ret_offset
) {
157 /* Same as journal_file_tail_end_by_pread() above, but operates with the usual mmap logic */
159 p
= le64toh(f
->header
->tail_object_offset
);
161 p
= le64toh(f
->header
->header_size
);
166 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &tail
);
170 sz
= le64toh(READ_NOW(tail
->object
.size
));
171 if (sz
> UINT64_MAX
- sizeof(uint64_t) + 1)
175 if (p
> UINT64_MAX
- sz
)
186 int journal_file_set_offline_thread_join(JournalFile
*f
) {
191 if (f
->offline_state
== OFFLINE_JOINED
)
194 r
= pthread_join(f
->offline_thread
, NULL
);
198 f
->offline_state
= OFFLINE_JOINED
;
200 if (mmap_cache_fd_got_sigbus(f
->cache_fd
))
206 static int journal_file_set_online(JournalFile
*f
) {
211 if (!journal_file_writable(f
))
214 if (f
->fd
< 0 || !f
->header
)
218 switch (f
->offline_state
) {
220 /* No offline thread, no need to wait. */
224 case OFFLINE_SYNCING
: {
225 OfflineState tmp_state
= OFFLINE_SYNCING
;
226 if (!__atomic_compare_exchange_n(&f
->offline_state
, &tmp_state
, OFFLINE_CANCEL
,
227 false, __ATOMIC_SEQ_CST
, __ATOMIC_SEQ_CST
))
230 /* Canceled syncing prior to offlining, no need to wait. */
234 case OFFLINE_AGAIN_FROM_SYNCING
: {
235 OfflineState tmp_state
= OFFLINE_AGAIN_FROM_SYNCING
;
236 if (!__atomic_compare_exchange_n(&f
->offline_state
, &tmp_state
, OFFLINE_CANCEL
,
237 false, __ATOMIC_SEQ_CST
, __ATOMIC_SEQ_CST
))
240 /* Canceled restart from syncing, no need to wait. */
244 case OFFLINE_AGAIN_FROM_OFFLINING
: {
245 OfflineState tmp_state
= OFFLINE_AGAIN_FROM_OFFLINING
;
246 if (!__atomic_compare_exchange_n(&f
->offline_state
, &tmp_state
, OFFLINE_CANCEL
,
247 false, __ATOMIC_SEQ_CST
, __ATOMIC_SEQ_CST
))
250 /* Canceled restart from offlining, must wait for offlining to complete however. */
255 r
= journal_file_set_offline_thread_join(f
);
265 if (mmap_cache_fd_got_sigbus(f
->cache_fd
))
268 switch (f
->header
->state
) {
273 f
->header
->state
= STATE_ONLINE
;
282 JournalFile
* journal_file_close(JournalFile
*f
) {
286 assert(f
->newest_boot_id_prioq_idx
== PRIOQ_IDX_NULL
);
289 mmap_cache_fd_free(f
->cache_fd
);
295 ordered_hashmap_free_free(f
->chain_cache
);
298 free(f
->compress_buffer
);
303 size_t sz
= PAGE_ALIGN(f
->fss_file_size
);
304 assert(sz
< SIZE_MAX
);
305 munmap(f
->fss_file
, sz
);
307 free(f
->fsprg_state
);
312 gcry_md_close(f
->hmac
);
318 static bool keyed_hash_requested(void) {
319 static thread_local
int cached
= -1;
323 r
= getenv_bool("SYSTEMD_JOURNAL_KEYED_HASH");
326 log_debug_errno(r
, "Failed to parse $SYSTEMD_JOURNAL_KEYED_HASH environment variable, ignoring: %m");
335 static bool compact_mode_requested(void) {
336 static thread_local
int cached
= -1;
340 r
= getenv_bool("SYSTEMD_JOURNAL_COMPACT");
343 log_debug_errno(r
, "Failed to parse $SYSTEMD_JOURNAL_COMPACT environment variable, ignoring: %m");
353 static Compression
getenv_compression(void) {
358 e
= getenv("SYSTEMD_JOURNAL_COMPRESS");
360 return DEFAULT_COMPRESSION
;
362 r
= parse_boolean(e
);
364 return r
? DEFAULT_COMPRESSION
: COMPRESSION_NONE
;
366 c
= compression_from_string(e
);
368 log_debug_errno(c
, "Failed to parse SYSTEMD_JOURNAL_COMPRESS value, ignoring: %s", e
);
369 return DEFAULT_COMPRESSION
;
372 if (!compression_supported(c
)) {
373 log_debug("Unsupported compression algorithm specified, ignoring: %s", e
);
374 return DEFAULT_COMPRESSION
;
381 static Compression
compression_requested(void) {
383 static thread_local Compression cached
= _COMPRESSION_INVALID
;
386 cached
= getenv_compression();
390 return COMPRESSION_NONE
;
394 static int journal_file_init_header(
396 JournalFileFlags file_flags
,
397 JournalFile
*template) {
406 /* Try to load the FSPRG state, and if we can't, then just don't do sealing */
407 seal
= FLAGS_SET(file_flags
, JOURNAL_SEAL
) && journal_file_fss_load(f
) >= 0;
411 .header_size
= htole64(ALIGN64(sizeof(h
))),
412 .incompatible_flags
= htole32(
413 FLAGS_SET(file_flags
, JOURNAL_COMPRESS
) * COMPRESSION_TO_HEADER_INCOMPATIBLE_FLAG(compression_requested()) |
414 keyed_hash_requested() * HEADER_INCOMPATIBLE_KEYED_HASH
|
415 compact_mode_requested() * HEADER_INCOMPATIBLE_COMPACT
),
416 .compatible_flags
= htole32(
417 (seal
* (HEADER_COMPATIBLE_SEALED
| HEADER_COMPATIBLE_SEALED_CONTINUOUS
) ) |
418 HEADER_COMPATIBLE_TAIL_ENTRY_BOOT_ID
),
421 assert_cc(sizeof(h
.signature
) == sizeof(HEADER_SIGNATURE
));
422 memcpy(h
.signature
, HEADER_SIGNATURE
, sizeof(HEADER_SIGNATURE
));
424 r
= sd_id128_randomize(&h
.file_id
);
428 r
= sd_id128_get_machine(&h
.machine_id
);
429 if (r
< 0 && !ERRNO_IS_MACHINE_ID_UNSET(r
))
430 return r
; /* If we have no valid machine ID (test environment?), let's simply leave the
431 * machine ID field all zeroes. */
434 h
.seqnum_id
= template->header
->seqnum_id
;
435 h
.tail_entry_seqnum
= template->header
->tail_entry_seqnum
;
437 h
.seqnum_id
= h
.file_id
;
439 k
= pwrite(f
->fd
, &h
, sizeof(h
), 0);
448 static int journal_file_refresh_header(JournalFile
*f
) {
454 /* We used to update the header's boot ID field here, but we don't do that anymore, as per
455 * HEADER_COMPATIBLE_TAIL_ENTRY_BOOT_ID */
457 r
= journal_file_set_online(f
);
459 /* Sync the online state to disk; likely just created a new file, also sync the directory this file
461 (void) fsync_full(f
->fd
);
466 static bool warn_wrong_flags(const JournalFile
*f
, bool compatible
) {
467 const uint32_t any
= compatible
? HEADER_COMPATIBLE_ANY
: HEADER_INCOMPATIBLE_ANY
,
468 supported
= compatible
? HEADER_COMPATIBLE_SUPPORTED
: HEADER_INCOMPATIBLE_SUPPORTED
;
469 const char *type
= compatible
? "compatible" : "incompatible";
475 flags
= le32toh(compatible
? f
->header
->compatible_flags
: f
->header
->incompatible_flags
);
477 if (flags
& ~supported
) {
479 log_debug("Journal file %s has unknown %s flags 0x%"PRIx32
,
480 f
->path
, type
, flags
& ~any
);
481 flags
= (flags
& any
) & ~supported
;
485 _cleanup_free_
char *t
= NULL
;
488 if (flags
& HEADER_COMPATIBLE_SEALED
)
489 strv
[n
++] = "sealed";
490 if (flags
& HEADER_COMPATIBLE_SEALED_CONTINUOUS
)
491 strv
[n
++] = "sealed-continuous";
493 if (flags
& HEADER_INCOMPATIBLE_COMPRESSED_XZ
)
494 strv
[n
++] = "xz-compressed";
495 if (flags
& HEADER_INCOMPATIBLE_COMPRESSED_LZ4
)
496 strv
[n
++] = "lz4-compressed";
497 if (flags
& HEADER_INCOMPATIBLE_COMPRESSED_ZSTD
)
498 strv
[n
++] = "zstd-compressed";
499 if (flags
& HEADER_INCOMPATIBLE_KEYED_HASH
)
500 strv
[n
++] = "keyed-hash";
501 if (flags
& HEADER_INCOMPATIBLE_COMPACT
)
502 strv
[n
++] = "compact";
505 assert(n
< ELEMENTSOF(strv
));
507 t
= strv_join((char**) strv
, ", ");
508 log_debug("Journal file %s uses %s %s %s disabled at compilation time.",
509 f
->path
, type
, n
> 1 ? "flags" : "flag", strnull(t
));
517 static bool offset_is_valid(uint64_t offset
, uint64_t header_size
, uint64_t tail_object_offset
) {
520 if (!VALID64(offset
))
522 if (offset
< header_size
)
524 if (offset
> tail_object_offset
)
529 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
) {
530 if ((offset
== 0) != (size
== 0))
534 if (offset
<= offsetof(Object
, hash_table
.items
))
536 offset
-= offsetof(Object
, hash_table
.items
);
537 if (!offset_is_valid(offset
, header_size
, tail_object_offset
))
539 assert(offset
<= header_size
+ arena_size
);
540 if (size
> header_size
+ arena_size
- offset
)
545 static int journal_file_verify_header(JournalFile
*f
) {
546 uint64_t arena_size
, header_size
;
551 if (memcmp(f
->header
->signature
, HEADER_SIGNATURE
, 8))
554 /* In both read and write mode we refuse to open files with incompatible
555 * flags we don't know. */
556 if (warn_wrong_flags(f
, false))
557 return -EPROTONOSUPPORT
;
559 /* When open for writing we refuse to open files with compatible flags, too. */
560 if (journal_file_writable(f
) && warn_wrong_flags(f
, true))
561 return -EPROTONOSUPPORT
;
563 if (f
->header
->state
>= _STATE_MAX
)
566 header_size
= le64toh(READ_NOW(f
->header
->header_size
));
568 /* The first addition was n_data, so check that we are at least this large */
569 if (header_size
< HEADER_SIZE_MIN
)
572 /* When open for writing we refuse to open files with a mismatch of the header size, i.e. writing to
573 * files implementing older or new header structures. */
574 if (journal_file_writable(f
) && header_size
!= sizeof(Header
))
575 return -EPROTONOSUPPORT
;
577 /* Don't write to journal files without the new boot ID update behavior guarantee. */
578 if (journal_file_writable(f
) && !JOURNAL_HEADER_TAIL_ENTRY_BOOT_ID(f
->header
))
579 return -EPROTONOSUPPORT
;
581 if (JOURNAL_HEADER_SEALED(f
->header
) && !JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
584 arena_size
= le64toh(READ_NOW(f
->header
->arena_size
));
586 if (UINT64_MAX
- header_size
< arena_size
|| header_size
+ arena_size
> (uint64_t) f
->last_stat
.st_size
)
589 uint64_t tail_object_offset
= le64toh(f
->header
->tail_object_offset
);
590 if (!offset_is_valid(tail_object_offset
, header_size
, UINT64_MAX
))
592 if (header_size
+ arena_size
< tail_object_offset
)
594 if (header_size
+ arena_size
- tail_object_offset
< sizeof(ObjectHeader
))
597 if (!hash_table_is_valid(le64toh(f
->header
->data_hash_table_offset
),
598 le64toh(f
->header
->data_hash_table_size
),
599 header_size
, arena_size
, tail_object_offset
))
602 if (!hash_table_is_valid(le64toh(f
->header
->field_hash_table_offset
),
603 le64toh(f
->header
->field_hash_table_size
),
604 header_size
, arena_size
, tail_object_offset
))
607 uint64_t entry_array_offset
= le64toh(f
->header
->entry_array_offset
);
608 if (!offset_is_valid(entry_array_offset
, header_size
, tail_object_offset
))
611 if (JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_array_offset
)) {
612 uint32_t offset
= le32toh(f
->header
->tail_entry_array_offset
);
613 uint32_t n
= le32toh(f
->header
->tail_entry_array_n_entries
);
615 if (!offset_is_valid(offset
, header_size
, tail_object_offset
))
617 if (entry_array_offset
> offset
)
619 if (entry_array_offset
== 0 && offset
!= 0)
621 if ((offset
== 0) != (n
== 0))
623 assert(offset
<= header_size
+ arena_size
);
624 if ((uint64_t) n
* journal_file_entry_array_item_size(f
) > header_size
+ arena_size
- offset
)
628 if (JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_offset
)) {
629 uint64_t offset
= le64toh(f
->header
->tail_entry_offset
);
631 if (!offset_is_valid(offset
, header_size
, tail_object_offset
))
635 /* When there is an entry object, then these fields must be filled. */
636 if (sd_id128_is_null(f
->header
->tail_entry_boot_id
))
638 if (!VALID_REALTIME(le64toh(f
->header
->head_entry_realtime
)))
640 if (!VALID_REALTIME(le64toh(f
->header
->tail_entry_realtime
)))
642 if (!VALID_MONOTONIC(le64toh(f
->header
->tail_entry_realtime
)))
645 /* Otherwise, the fields must be zero. */
646 if (JOURNAL_HEADER_TAIL_ENTRY_BOOT_ID(f
->header
) &&
647 !sd_id128_is_null(f
->header
->tail_entry_boot_id
))
649 if (f
->header
->head_entry_realtime
!= 0)
651 if (f
->header
->tail_entry_realtime
!= 0)
653 if (f
->header
->tail_entry_realtime
!= 0)
658 /* Verify number of objects */
659 uint64_t n_objects
= le64toh(f
->header
->n_objects
);
660 if (n_objects
> arena_size
/ sizeof(ObjectHeader
))
663 uint64_t n_entries
= le64toh(f
->header
->n_entries
);
664 if (n_entries
> n_objects
)
667 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
668 le64toh(f
->header
->n_data
) > n_objects
)
671 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
672 le64toh(f
->header
->n_fields
) > n_objects
)
675 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
) &&
676 le64toh(f
->header
->n_tags
) > n_objects
)
679 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
) &&
680 le64toh(f
->header
->n_entry_arrays
) > n_objects
)
683 if (JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_array_n_entries
) &&
684 le32toh(f
->header
->tail_entry_array_n_entries
) > n_entries
)
687 if (journal_file_writable(f
)) {
688 sd_id128_t machine_id
;
692 r
= sd_id128_get_machine(&machine_id
);
693 if (ERRNO_IS_NEG_MACHINE_ID_UNSET(r
)) /* Gracefully handle the machine ID not being initialized yet */
694 machine_id
= SD_ID128_NULL
;
698 if (!sd_id128_equal(machine_id
, f
->header
->machine_id
))
699 return log_debug_errno(SYNTHETIC_ERRNO(EHOSTDOWN
),
700 "Trying to open journal file from different host for writing, refusing.");
702 state
= f
->header
->state
;
704 if (state
== STATE_ARCHIVED
)
705 return -ESHUTDOWN
; /* Already archived */
706 if (state
== STATE_ONLINE
)
707 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
),
708 "Journal file %s is already online. Assuming unclean closing.",
710 if (state
!= STATE_OFFLINE
)
711 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
),
712 "Journal file %s has unknown state %i.",
715 if (f
->header
->field_hash_table_size
== 0 || f
->header
->data_hash_table_size
== 0)
722 int journal_file_fstat(JournalFile
*f
) {
728 if (fstat(f
->fd
, &f
->last_stat
) < 0)
731 f
->last_stat_usec
= now(CLOCK_MONOTONIC
);
733 /* Refuse dealing with files that aren't regular */
734 r
= stat_verify_regular(&f
->last_stat
);
738 /* Refuse appending to files that are already deleted */
739 if (f
->last_stat
.st_nlink
<= 0)
745 static int journal_file_allocate(JournalFile
*f
, uint64_t offset
, uint64_t size
) {
746 uint64_t old_size
, new_size
, old_header_size
, old_arena_size
;
752 /* We assume that this file is not sparse, and we know that for sure, since we always call
753 * posix_fallocate() ourselves */
755 if (size
> PAGE_ALIGN_DOWN_U64(UINT64_MAX
) - offset
)
758 if (mmap_cache_fd_got_sigbus(f
->cache_fd
))
761 old_header_size
= le64toh(READ_NOW(f
->header
->header_size
));
762 old_arena_size
= le64toh(READ_NOW(f
->header
->arena_size
));
763 if (old_arena_size
> PAGE_ALIGN_DOWN_U64(UINT64_MAX
) - old_header_size
)
766 old_size
= old_header_size
+ old_arena_size
;
768 new_size
= MAX(PAGE_ALIGN_U64(offset
+ size
), old_header_size
);
770 if (new_size
<= old_size
) {
772 /* We already pre-allocated enough space, but before
773 * we write to it, let's check with fstat() if the
774 * file got deleted, in order make sure we don't throw
775 * away the data immediately. Don't check fstat() for
776 * all writes though, but only once ever 10s. */
778 if (f
->last_stat_usec
+ LAST_STAT_REFRESH_USEC
> now(CLOCK_MONOTONIC
))
781 return journal_file_fstat(f
);
784 /* Allocate more space. */
786 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
789 /* Refuse to go over 4G in compact mode so offsets can be stored in 32-bit. */
790 if (JOURNAL_HEADER_COMPACT(f
->header
) && new_size
> UINT32_MAX
)
793 if (new_size
> f
->metrics
.min_size
&& f
->metrics
.keep_free
> 0) {
796 if (fstatvfs(f
->fd
, &svfs
) >= 0) {
799 available
= LESS_BY(u64_multiply_safe(svfs
.f_bfree
, svfs
.f_bsize
), f
->metrics
.keep_free
);
801 if (new_size
- old_size
> available
)
806 /* Increase by larger blocks at once */
807 new_size
= ROUND_UP(new_size
, FILE_SIZE_INCREASE
);
808 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
809 new_size
= f
->metrics
.max_size
;
811 /* Note that the glibc fallocate() fallback is very
812 inefficient, hence we try to minimize the allocation area
814 r
= posix_fallocate_loop(f
->fd
, old_size
, new_size
- old_size
);
818 f
->header
->arena_size
= htole64(new_size
- old_header_size
);
820 return journal_file_fstat(f
);
823 static int journal_file_move_to(
836 /* This function may clear, overwrite, or alter previously cached entries with the same type. After
837 * this function has been called, all previously read objects with the same type may be invalidated,
838 * hence must be re-read before use. */
843 if (size
> UINT64_MAX
- offset
)
846 /* Avoid SIGBUS on invalid accesses */
847 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
) {
848 /* Hmm, out of range? Let's refresh the fstat() data
849 * first, before we trust that check. */
851 r
= journal_file_fstat(f
);
855 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
)
856 return -EADDRNOTAVAIL
;
859 return mmap_cache_fd_get(f
->cache_fd
, type_to_category(type
), keep_always
, offset
, size
, &f
->last_stat
, ret
);
862 static uint64_t minimum_header_size(JournalFile
*f
, Object
*o
) {
864 static const uint64_t table
[] = {
865 [OBJECT_DATA
] = sizeof(DataObject
),
866 [OBJECT_FIELD
] = sizeof(FieldObject
),
867 [OBJECT_ENTRY
] = sizeof(EntryObject
),
868 [OBJECT_DATA_HASH_TABLE
] = sizeof(HashTableObject
),
869 [OBJECT_FIELD_HASH_TABLE
] = sizeof(HashTableObject
),
870 [OBJECT_ENTRY_ARRAY
] = sizeof(EntryArrayObject
),
871 [OBJECT_TAG
] = sizeof(TagObject
),
877 if (o
->object
.type
== OBJECT_DATA
)
878 return journal_file_data_payload_offset(f
);
880 if (o
->object
.type
>= ELEMENTSOF(table
) || table
[o
->object
.type
] <= 0)
881 return sizeof(ObjectHeader
);
883 return table
[o
->object
.type
];
886 static int check_object_header(JournalFile
*f
, Object
*o
, ObjectType type
, uint64_t offset
) {
892 s
= le64toh(READ_NOW(o
->object
.size
));
894 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
895 "Attempt to move to uninitialized object: %" PRIu64
,
898 if (s
< sizeof(ObjectHeader
))
899 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
900 "Attempt to move to overly short object with size %"PRIu64
": %" PRIu64
,
903 if (o
->object
.type
<= OBJECT_UNUSED
|| o
->object
.type
>= _OBJECT_TYPE_MAX
)
904 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
905 "Attempt to move to object with invalid type (%u): %" PRIu64
,
906 o
->object
.type
, offset
);
908 if (type
> OBJECT_UNUSED
&& o
->object
.type
!= type
)
909 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
910 "Found %s object while expecting %s object: %" PRIu64
,
911 journal_object_type_to_string(o
->object
.type
),
912 journal_object_type_to_string(type
),
915 if (s
< minimum_header_size(f
, o
))
916 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
917 "Size of %s object (%"PRIu64
") is smaller than the minimum object size (%"PRIu64
"): %" PRIu64
,
918 journal_object_type_to_string(o
->object
.type
),
920 minimum_header_size(f
, o
),
926 /* Lightweight object checks. We want this to be fast, so that we won't
927 * slowdown every journal_file_move_to_object() call too much. */
928 static int check_object(JournalFile
*f
, Object
*o
, uint64_t offset
) {
932 switch (o
->object
.type
) {
935 if ((le64toh(o
->data
.entry_offset
) == 0) ^ (le64toh(o
->data
.n_entries
) == 0))
936 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
937 "Bad data n_entries: %" PRIu64
": %" PRIu64
,
938 le64toh(o
->data
.n_entries
),
941 if (le64toh(o
->object
.size
) <= journal_file_data_payload_offset(f
))
942 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
943 "Bad data size (<= %zu): %" PRIu64
": %" PRIu64
,
944 journal_file_data_payload_offset(f
),
945 le64toh(o
->object
.size
),
948 if (!VALID64(le64toh(o
->data
.next_hash_offset
)) ||
949 !VALID64(le64toh(o
->data
.next_field_offset
)) ||
950 !VALID64(le64toh(o
->data
.entry_offset
)) ||
951 !VALID64(le64toh(o
->data
.entry_array_offset
)))
952 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
953 "Invalid offset, next_hash_offset=" OFSfmt
", next_field_offset=" OFSfmt
", entry_offset=" OFSfmt
", entry_array_offset=" OFSfmt
": %" PRIu64
,
954 le64toh(o
->data
.next_hash_offset
),
955 le64toh(o
->data
.next_field_offset
),
956 le64toh(o
->data
.entry_offset
),
957 le64toh(o
->data
.entry_array_offset
),
963 if (le64toh(o
->object
.size
) <= offsetof(Object
, field
.payload
))
964 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
965 "Bad field size (<= %zu): %" PRIu64
": %" PRIu64
,
966 offsetof(Object
, field
.payload
),
967 le64toh(o
->object
.size
),
970 if (!VALID64(le64toh(o
->field
.next_hash_offset
)) ||
971 !VALID64(le64toh(o
->field
.head_data_offset
)))
972 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
973 "Invalid offset, next_hash_offset=" OFSfmt
", head_data_offset=" OFSfmt
": %" PRIu64
,
974 le64toh(o
->field
.next_hash_offset
),
975 le64toh(o
->field
.head_data_offset
),
982 sz
= le64toh(READ_NOW(o
->object
.size
));
983 if (sz
< offsetof(Object
, entry
.items
) ||
984 (sz
- offsetof(Object
, entry
.items
)) % journal_file_entry_item_size(f
) != 0)
985 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
986 "Bad entry size (<= %zu): %" PRIu64
": %" PRIu64
,
987 offsetof(Object
, entry
.items
),
991 if ((sz
- offsetof(Object
, entry
.items
)) / journal_file_entry_item_size(f
) <= 0)
992 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
993 "Invalid number items in entry: %" PRIu64
": %" PRIu64
,
994 (sz
- offsetof(Object
, entry
.items
)) / journal_file_entry_item_size(f
),
997 if (le64toh(o
->entry
.seqnum
) <= 0)
998 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
999 "Invalid entry seqnum: %" PRIx64
": %" PRIu64
,
1000 le64toh(o
->entry
.seqnum
),
1003 if (!VALID_REALTIME(le64toh(o
->entry
.realtime
)))
1004 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1005 "Invalid entry realtime timestamp: %" PRIu64
": %" PRIu64
,
1006 le64toh(o
->entry
.realtime
),
1009 if (!VALID_MONOTONIC(le64toh(o
->entry
.monotonic
)))
1010 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1011 "Invalid entry monotonic timestamp: %" PRIu64
": %" PRIu64
,
1012 le64toh(o
->entry
.monotonic
),
1015 if (sd_id128_is_null(o
->entry
.boot_id
))
1016 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1017 "Invalid object entry with an empty boot ID: %" PRIu64
,
1023 case OBJECT_DATA_HASH_TABLE
:
1024 case OBJECT_FIELD_HASH_TABLE
: {
1027 sz
= le64toh(READ_NOW(o
->object
.size
));
1028 if (sz
< offsetof(Object
, hash_table
.items
) ||
1029 (sz
- offsetof(Object
, hash_table
.items
)) % sizeof(HashItem
) != 0 ||
1030 (sz
- offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
) <= 0)
1031 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1032 "Invalid %s hash table size: %" PRIu64
": %" PRIu64
,
1033 journal_object_type_to_string(o
->object
.type
),
1040 case OBJECT_ENTRY_ARRAY
: {
1043 sz
= le64toh(READ_NOW(o
->object
.size
));
1044 if (sz
< offsetof(Object
, entry_array
.items
) ||
1045 (sz
- offsetof(Object
, entry_array
.items
)) % journal_file_entry_array_item_size(f
) != 0 ||
1046 (sz
- offsetof(Object
, entry_array
.items
)) / journal_file_entry_array_item_size(f
) <= 0)
1047 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1048 "Invalid object entry array size: %" PRIu64
": %" PRIu64
,
1051 /* Here, we request that the offset of each entry array object is in strictly increasing order. */
1052 next
= le64toh(o
->entry_array
.next_entry_array_offset
);
1053 if (!VALID64(next
) || (next
> 0 && next
<= offset
))
1054 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1055 "Invalid object entry array next_entry_array_offset: %" PRIu64
": %" PRIu64
,
1063 if (le64toh(o
->object
.size
) != sizeof(TagObject
))
1064 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1065 "Invalid object tag size: %" PRIu64
": %" PRIu64
,
1066 le64toh(o
->object
.size
),
1069 if (!VALID_EPOCH(le64toh(o
->tag
.epoch
)))
1070 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1071 "Invalid object tag epoch: %" PRIu64
": %" PRIu64
,
1072 le64toh(o
->tag
.epoch
), offset
);
1080 int journal_file_move_to_object(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
**ret
) {
1086 /* Even if this function fails, it may clear, overwrite, or alter previously cached entries with the
1087 * same type. After this function has been called, all previously read objects with the same type may
1088 * be invalidated, hence must be re-read before use. */
1090 /* Objects may only be located at multiple of 64 bit */
1091 if (!VALID64(offset
))
1092 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1093 "Attempt to move to %s object at non-64-bit boundary: %" PRIu64
,
1094 journal_object_type_to_string(type
),
1097 /* Object may not be located in the file header */
1098 if (offset
< le64toh(f
->header
->header_size
))
1099 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1100 "Attempt to move to %s object located in file header: %" PRIu64
,
1101 journal_object_type_to_string(type
),
1104 r
= journal_file_move_to(f
, type
, false, offset
, sizeof(ObjectHeader
), (void**) &o
);
1108 r
= check_object_header(f
, o
, type
, offset
);
1112 r
= journal_file_move_to(f
, type
, false, offset
, le64toh(READ_NOW(o
->object
.size
)), (void**) &o
);
1116 r
= check_object_header(f
, o
, type
, offset
);
1120 r
= check_object(f
, o
, offset
);
1130 int journal_file_pin_object(JournalFile
*f
, Object
*o
) {
1134 /* This attaches the mmap window that provides the object to the 'pinning' category. So, reading
1135 * another object with the same type will not invalidate the object, until this function is called
1136 * for another object. */
1137 return mmap_cache_fd_pin(f
->cache_fd
, type_to_category(o
->object
.type
), o
, le64toh(o
->object
.size
));
1140 int journal_file_read_object_header(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
*ret
) {
1147 /* Objects may only be located at multiple of 64 bit */
1148 if (!VALID64(offset
))
1149 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1150 "Attempt to read %s object at non-64-bit boundary: %" PRIu64
,
1151 journal_object_type_to_string(type
), offset
);
1153 /* Object may not be located in the file header */
1154 if (offset
< le64toh(f
->header
->header_size
))
1155 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1156 "Attempt to read %s object located in file header: %" PRIu64
,
1157 journal_object_type_to_string(type
), offset
);
1159 /* This will likely read too much data but it avoids having to call pread() twice. */
1160 n
= pread(f
->fd
, &o
, sizeof(o
), offset
);
1162 return log_debug_errno(errno
, "Failed to read journal %s object at offset: %" PRIu64
,
1163 journal_object_type_to_string(type
), offset
);
1165 if ((size_t) n
< sizeof(o
.object
))
1166 return log_debug_errno(SYNTHETIC_ERRNO(EIO
),
1167 "Failed to read short %s object at offset: %" PRIu64
,
1168 journal_object_type_to_string(type
), offset
);
1170 r
= check_object_header(f
, &o
, type
, offset
);
1174 if ((size_t) n
< minimum_header_size(f
, &o
))
1175 return log_debug_errno(SYNTHETIC_ERRNO(EIO
),
1176 "Short read while reading %s object: %" PRIu64
,
1177 journal_object_type_to_string(type
), offset
);
1179 r
= check_object(f
, &o
, offset
);
1189 static uint64_t inc_seqnum(uint64_t seqnum
) {
1190 if (seqnum
< UINT64_MAX
-1)
1193 return 1; /* skip over UINT64_MAX and 0 when we run out of seqnums and start again */
1196 static uint64_t journal_file_entry_seqnum(
1200 uint64_t next_seqnum
;
1205 /* Picks a new sequence number for the entry we are about to add and returns it. */
1207 next_seqnum
= inc_seqnum(le64toh(f
->header
->tail_entry_seqnum
));
1209 /* If an external seqnum counter was passed, we update both the local and the external one, and set
1210 * it to the maximum of both */
1212 *seqnum
= next_seqnum
= MAX(inc_seqnum(*seqnum
), next_seqnum
);
1214 f
->header
->tail_entry_seqnum
= htole64(next_seqnum
);
1216 if (f
->header
->head_entry_seqnum
== 0)
1217 f
->header
->head_entry_seqnum
= htole64(next_seqnum
);
1222 int journal_file_append_object(
1226 Object
**ret_object
,
1227 uint64_t *ret_offset
) {
1235 assert(type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
);
1236 assert(size
>= sizeof(ObjectHeader
));
1238 r
= journal_file_set_online(f
);
1242 r
= journal_file_tail_end_by_mmap(f
, &p
);
1246 r
= journal_file_allocate(f
, p
, size
);
1250 r
= journal_file_move_to(f
, type
, false, p
, size
, (void**) &o
);
1254 o
->object
= (ObjectHeader
) {
1256 .size
= htole64(size
),
1259 f
->header
->tail_object_offset
= htole64(p
);
1260 f
->header
->n_objects
= htole64(le64toh(f
->header
->n_objects
) + 1);
1271 static int journal_file_setup_data_hash_table(JournalFile
*f
) {
1279 /* We estimate that we need 1 hash table entry per 768 bytes
1280 of journal file and we want to make sure we never get
1281 beyond 75% fill level. Calculate the hash table size for
1282 the maximum file size based on these metrics. */
1284 s
= (f
->metrics
.max_size
* 4 / 768 / 3) * sizeof(HashItem
);
1285 if (s
< DEFAULT_DATA_HASH_TABLE_SIZE
)
1286 s
= DEFAULT_DATA_HASH_TABLE_SIZE
;
1288 log_debug("Reserving %"PRIu64
" entries in data hash table.", s
/ sizeof(HashItem
));
1290 r
= journal_file_append_object(f
,
1291 OBJECT_DATA_HASH_TABLE
,
1292 offsetof(Object
, hash_table
.items
) + s
,
1297 memzero(o
->hash_table
.items
, s
);
1299 f
->header
->data_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1300 f
->header
->data_hash_table_size
= htole64(s
);
1305 static int journal_file_setup_field_hash_table(JournalFile
*f
) {
1313 /* We use a fixed size hash table for the fields as this
1314 * number should grow very slowly only */
1316 s
= DEFAULT_FIELD_HASH_TABLE_SIZE
;
1317 log_debug("Reserving %"PRIu64
" entries in field hash table.", s
/ sizeof(HashItem
));
1319 r
= journal_file_append_object(f
,
1320 OBJECT_FIELD_HASH_TABLE
,
1321 offsetof(Object
, hash_table
.items
) + s
,
1326 memzero(o
->hash_table
.items
, s
);
1328 f
->header
->field_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1329 f
->header
->field_hash_table_size
= htole64(s
);
1334 int journal_file_map_data_hash_table(JournalFile
*f
) {
1342 if (f
->data_hash_table
)
1345 p
= le64toh(f
->header
->data_hash_table_offset
);
1346 s
= le64toh(f
->header
->data_hash_table_size
);
1348 r
= journal_file_move_to(f
,
1349 OBJECT_DATA_HASH_TABLE
,
1356 f
->data_hash_table
= t
;
1360 int journal_file_map_field_hash_table(JournalFile
*f
) {
1368 if (f
->field_hash_table
)
1371 p
= le64toh(f
->header
->field_hash_table_offset
);
1372 s
= le64toh(f
->header
->field_hash_table_size
);
1374 r
= journal_file_move_to(f
,
1375 OBJECT_FIELD_HASH_TABLE
,
1382 f
->field_hash_table
= t
;
1386 static int journal_file_link_field(
1397 assert(f
->field_hash_table
);
1401 if (o
->object
.type
!= OBJECT_FIELD
)
1404 m
= le64toh(READ_NOW(f
->header
->field_hash_table_size
)) / sizeof(HashItem
);
1408 /* This might alter the window we are looking at */
1409 o
->field
.next_hash_offset
= o
->field
.head_data_offset
= 0;
1412 p
= le64toh(f
->field_hash_table
[h
].tail_hash_offset
);
1414 f
->field_hash_table
[h
].head_hash_offset
= htole64(offset
);
1416 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1420 o
->field
.next_hash_offset
= htole64(offset
);
1423 f
->field_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1425 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
1426 f
->header
->n_fields
= htole64(le64toh(f
->header
->n_fields
) + 1);
1431 static int journal_file_link_data(
1442 assert(f
->data_hash_table
);
1446 if (o
->object
.type
!= OBJECT_DATA
)
1449 m
= le64toh(READ_NOW(f
->header
->data_hash_table_size
)) / sizeof(HashItem
);
1453 /* This might alter the window we are looking at */
1454 o
->data
.next_hash_offset
= o
->data
.next_field_offset
= 0;
1455 o
->data
.entry_offset
= o
->data
.entry_array_offset
= 0;
1456 o
->data
.n_entries
= 0;
1459 p
= le64toh(f
->data_hash_table
[h
].tail_hash_offset
);
1461 /* Only entry in the hash table is easy */
1462 f
->data_hash_table
[h
].head_hash_offset
= htole64(offset
);
1464 /* Move back to the previous data object, to patch in
1467 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1471 o
->data
.next_hash_offset
= htole64(offset
);
1474 f
->data_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1476 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
1477 f
->header
->n_data
= htole64(le64toh(f
->header
->n_data
) + 1);
1482 static int get_next_hash_offset(
1485 le64_t
*next_hash_offset
,
1487 le64_t
*header_max_depth
) {
1493 assert(next_hash_offset
);
1496 nextp
= le64toh(READ_NOW(*next_hash_offset
));
1498 if (nextp
<= *p
) /* Refuse going in loops */
1499 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1500 "Detected hash item loop in %s, refusing.", f
->path
);
1504 /* If the depth of this hash chain is larger than all others we have seen so far, record it */
1505 if (header_max_depth
&& journal_file_writable(f
))
1506 *header_max_depth
= htole64(MAX(*depth
, le64toh(*header_max_depth
)));
1513 int journal_file_find_field_object_with_hash(
1518 Object
**ret_object
,
1519 uint64_t *ret_offset
) {
1521 uint64_t p
, osize
, h
, m
, depth
= 0;
1529 /* If the field hash table is empty, we can't find anything */
1530 if (le64toh(f
->header
->field_hash_table_size
) <= 0)
1533 /* Map the field hash table, if it isn't mapped yet. */
1534 r
= journal_file_map_field_hash_table(f
);
1538 osize
= offsetof(Object
, field
.payload
) + size
;
1540 m
= le64toh(READ_NOW(f
->header
->field_hash_table_size
)) / sizeof(HashItem
);
1545 p
= le64toh(f
->field_hash_table
[h
].head_hash_offset
);
1549 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1553 if (le64toh(o
->field
.hash
) == hash
&&
1554 le64toh(o
->object
.size
) == osize
&&
1555 memcmp(o
->field
.payload
, field
, size
) == 0) {
1565 r
= get_next_hash_offset(
1568 &o
->field
.next_hash_offset
,
1570 JOURNAL_HEADER_CONTAINS(f
->header
, field_hash_chain_depth
) ? &f
->header
->field_hash_chain_depth
: NULL
);
1578 uint64_t journal_file_hash_data(
1585 assert(data
|| sz
== 0);
1587 /* We try to unify our codebase on siphash, hence new-styled journal files utilizing the keyed hash
1588 * function use siphash. Old journal files use the Jenkins hash. */
1590 if (JOURNAL_HEADER_KEYED_HASH(f
->header
))
1591 return siphash24(data
, sz
, f
->header
->file_id
.bytes
);
1593 return jenkins_hash64(data
, sz
);
1596 int journal_file_find_field_object(
1600 Object
**ret_object
,
1601 uint64_t *ret_offset
) {
1607 return journal_file_find_field_object_with_hash(
1610 journal_file_hash_data(f
, field
, size
),
1611 ret_object
, ret_offset
);
1614 int journal_file_find_data_object_with_hash(
1619 Object
**ret_object
,
1620 uint64_t *ret_offset
) {
1622 uint64_t p
, h
, m
, depth
= 0;
1627 assert(data
|| size
== 0);
1629 /* If there's no data hash table, then there's no entry. */
1630 if (le64toh(f
->header
->data_hash_table_size
) <= 0)
1633 /* Map the data hash table, if it isn't mapped yet. */
1634 r
= journal_file_map_data_hash_table(f
);
1638 m
= le64toh(READ_NOW(f
->header
->data_hash_table_size
)) / sizeof(HashItem
);
1643 p
= le64toh(f
->data_hash_table
[h
].head_hash_offset
);
1650 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1654 if (le64toh(o
->data
.hash
) != hash
)
1657 r
= journal_file_data_payload(f
, o
, p
, NULL
, 0, 0, &d
, &rsize
);
1660 assert(r
> 0); /* journal_file_data_payload() always returns > 0 if no field is provided. */
1662 if (memcmp_nn(data
, size
, d
, rsize
) == 0) {
1673 r
= get_next_hash_offset(
1676 &o
->data
.next_hash_offset
,
1678 JOURNAL_HEADER_CONTAINS(f
->header
, data_hash_chain_depth
) ? &f
->header
->data_hash_chain_depth
: NULL
);
1686 int journal_file_find_data_object(
1690 Object
**ret_object
,
1691 uint64_t *ret_offset
) {
1694 assert(data
|| size
== 0);
1696 return journal_file_find_data_object_with_hash(
1699 journal_file_hash_data(f
, data
, size
),
1700 ret_object
, ret_offset
);
1703 bool journal_field_valid(const char *p
, size_t l
, bool allow_protected
) {
1704 /* We kinda enforce POSIX syntax recommendations for
1705 environment variables here, but make a couple of additional
1708 http://pubs.opengroup.org/onlinepubs/000095399/basedefs/xbd_chap08.html */
1715 /* No empty field names */
1719 /* Don't allow names longer than 64 chars */
1723 /* Variables starting with an underscore are protected */
1724 if (!allow_protected
&& p
[0] == '_')
1727 /* Don't allow digits as first character */
1728 if (ascii_isdigit(p
[0]))
1731 /* Only allow A-Z0-9 and '_' */
1732 for (const char *a
= p
; a
< p
+ l
; a
++)
1733 if ((*a
< 'A' || *a
> 'Z') &&
1734 !ascii_isdigit(*a
) &&
1741 static int journal_file_append_field(
1745 Object
**ret_object
,
1746 uint64_t *ret_offset
) {
1757 if (!journal_field_valid(field
, size
, true))
1760 hash
= journal_file_hash_data(f
, field
, size
);
1762 r
= journal_file_find_field_object_with_hash(f
, field
, size
, hash
, ret_object
, ret_offset
);
1768 osize
= offsetof(Object
, field
.payload
) + size
;
1769 r
= journal_file_append_object(f
, OBJECT_FIELD
, osize
, &o
, &p
);
1773 o
->field
.hash
= htole64(hash
);
1774 memcpy(o
->field
.payload
, field
, size
);
1776 r
= journal_file_link_field(f
, o
, p
, hash
);
1780 /* The linking might have altered the window, so let's only pass the offset to hmac which will
1781 * move to the object again if needed. */
1784 r
= journal_file_hmac_put_object(f
, OBJECT_FIELD
, NULL
, p
);
1790 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, ret_object
);
1801 static int maybe_compress_payload(JournalFile
*f
, uint8_t *dst
, const uint8_t *src
, uint64_t size
, size_t *rsize
) {
1805 #if HAVE_COMPRESSION
1809 c
= JOURNAL_FILE_COMPRESSION(f
);
1810 if (c
== COMPRESSION_NONE
|| size
< f
->compress_threshold_bytes
)
1813 r
= compress_blob(c
, src
, size
, dst
, size
- 1, rsize
);
1815 return log_debug_errno(r
, "Failed to compress data object using %s, ignoring: %m", compression_to_string(c
));
1817 log_debug("Compressed data object %"PRIu64
" -> %zu using %s", size
, *rsize
, compression_to_string(c
));
1819 return 1; /* compressed */
1825 static int journal_file_append_data(
1829 Object
**ret_object
,
1830 uint64_t *ret_offset
) {
1832 uint64_t hash
, p
, osize
;
1840 if (!data
|| size
== 0)
1843 hash
= journal_file_hash_data(f
, data
, size
);
1845 r
= journal_file_find_data_object_with_hash(f
, data
, size
, hash
, ret_object
, ret_offset
);
1851 eq
= memchr(data
, '=', size
);
1855 osize
= journal_file_data_payload_offset(f
) + size
;
1856 r
= journal_file_append_object(f
, OBJECT_DATA
, osize
, &o
, &p
);
1860 o
->data
.hash
= htole64(hash
);
1862 r
= maybe_compress_payload(f
, journal_file_data_payload_field(f
, o
), data
, size
, &rsize
);
1864 /* We don't really care failures, let's continue without compression */
1865 memcpy_safe(journal_file_data_payload_field(f
, o
), data
, size
);
1867 Compression c
= JOURNAL_FILE_COMPRESSION(f
);
1869 assert(c
>= 0 && c
< _COMPRESSION_MAX
&& c
!= COMPRESSION_NONE
);
1871 o
->object
.size
= htole64(journal_file_data_payload_offset(f
) + rsize
);
1872 o
->object
.flags
|= COMPRESSION_TO_OBJECT_FLAG(c
);
1875 r
= journal_file_link_data(f
, o
, p
, hash
);
1879 /* The linking might have altered the window, so let's refresh our pointer. */
1880 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1885 r
= journal_file_hmac_put_object(f
, OBJECT_DATA
, o
, p
);
1890 /* Create field object ... */
1891 r
= journal_file_append_field(f
, data
, (uint8_t*) eq
- (uint8_t*) data
, &fo
, NULL
);
1895 /* ... and link it in. */
1896 o
->data
.next_field_offset
= fo
->field
.head_data_offset
;
1897 fo
->field
.head_data_offset
= le64toh(p
);
1908 static int maybe_decompress_payload(
1912 Compression compression
,
1914 size_t field_length
,
1915 size_t data_threshold
,
1921 /* We can't read objects larger than 4G on a 32-bit machine */
1922 if ((uint64_t) (size_t) size
!= size
)
1925 if (compression
!= COMPRESSION_NONE
) {
1926 #if HAVE_COMPRESSION
1931 r
= decompress_startswith(compression
, payload
, size
, &f
->compress_buffer
, field
,
1934 return log_debug_errno(r
,
1935 "Cannot decompress %s object of length %" PRIu64
": %m",
1936 compression_to_string(compression
),
1947 r
= decompress_blob(compression
, payload
, size
, &f
->compress_buffer
, &rsize
, 0);
1952 *ret_data
= f
->compress_buffer
;
1956 return -EPROTONOSUPPORT
;
1959 if (field
&& (size
< field_length
+ 1 || memcmp(payload
, field
, field_length
) != 0 || payload
[field_length
] != '=')) {
1968 *ret_data
= payload
;
1970 *ret_size
= (size_t) size
;
1976 int journal_file_data_payload(
1981 size_t field_length
,
1982 size_t data_threshold
,
1991 assert(!field
== (field_length
== 0)); /* These must be specified together. */
1994 r
= journal_file_move_to_object(f
, OBJECT_DATA
, offset
, &o
);
1999 size
= le64toh(READ_NOW(o
->object
.size
));
2000 if (size
< journal_file_data_payload_offset(f
))
2003 size
-= journal_file_data_payload_offset(f
);
2005 c
= COMPRESSION_FROM_OBJECT(o
);
2007 return -EPROTONOSUPPORT
;
2009 return maybe_decompress_payload(f
, journal_file_data_payload_field(f
, o
), size
, c
, field
,
2010 field_length
, data_threshold
, ret_data
, ret_size
);
2013 uint64_t journal_file_entry_n_items(JournalFile
*f
, Object
*o
) {
2019 if (o
->object
.type
!= OBJECT_ENTRY
)
2022 sz
= le64toh(READ_NOW(o
->object
.size
));
2023 if (sz
< offsetof(Object
, entry
.items
))
2026 return (sz
- offsetof(Object
, entry
.items
)) / journal_file_entry_item_size(f
);
2029 uint64_t journal_file_entry_array_n_items(JournalFile
*f
, Object
*o
) {
2035 if (o
->object
.type
!= OBJECT_ENTRY_ARRAY
)
2038 sz
= le64toh(READ_NOW(o
->object
.size
));
2039 if (sz
< offsetof(Object
, entry_array
.items
))
2042 return (sz
- offsetof(Object
, entry_array
.items
)) / journal_file_entry_array_item_size(f
);
2045 uint64_t journal_file_hash_table_n_items(Object
*o
) {
2050 if (!IN_SET(o
->object
.type
, OBJECT_DATA_HASH_TABLE
, OBJECT_FIELD_HASH_TABLE
))
2053 sz
= le64toh(READ_NOW(o
->object
.size
));
2054 if (sz
< offsetof(Object
, hash_table
.items
))
2057 return (sz
- offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
);
2060 static void write_entry_array_item(JournalFile
*f
, Object
*o
, uint64_t i
, uint64_t p
) {
2064 if (JOURNAL_HEADER_COMPACT(f
->header
)) {
2065 assert(p
<= UINT32_MAX
);
2066 o
->entry_array
.items
.compact
[i
] = htole32(p
);
2068 o
->entry_array
.items
.regular
[i
] = htole64(p
);
2071 static int link_entry_into_array(
2079 uint64_t n
= 0, ap
= 0, q
, i
, a
, hidx
;
2089 a
= tail
? le32toh(*tail
) : le64toh(*first
);
2090 hidx
= le64toh(READ_NOW(*idx
));
2091 i
= tidx
? le32toh(READ_NOW(*tidx
)) : hidx
;
2094 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2098 n
= journal_file_entry_array_n_items(f
, o
);
2100 write_entry_array_item(f
, o
, i
, p
);
2101 *idx
= htole64(hidx
+ 1);
2103 *tidx
= htole32(le32toh(*tidx
) + 1);
2109 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
2120 r
= journal_file_append_object(f
, OBJECT_ENTRY_ARRAY
,
2121 offsetof(Object
, entry_array
.items
) + n
* journal_file_entry_array_item_size(f
),
2127 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY_ARRAY
, o
, q
);
2132 write_entry_array_item(f
, o
, i
, p
);
2135 *first
= htole64(q
);
2137 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, ap
, &o
);
2141 o
->entry_array
.next_entry_array_offset
= htole64(q
);
2147 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
2148 f
->header
->n_entry_arrays
= htole64(le64toh(f
->header
->n_entry_arrays
) + 1);
2150 *idx
= htole64(hidx
+ 1);
2157 static int link_entry_into_array_plus_one(
2175 hidx
= le64toh(READ_NOW(*idx
));
2176 if (hidx
== UINT64_MAX
)
2179 *extra
= htole64(p
);
2183 i
= htole64(hidx
- 1);
2184 r
= link_entry_into_array(f
, first
, &i
, tail
, tidx
, p
);
2189 *idx
= htole64(hidx
+ 1);
2193 static int journal_file_link_entry_item(JournalFile
*f
, uint64_t offset
, uint64_t p
) {
2200 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
2204 return link_entry_into_array_plus_one(f
,
2205 &o
->data
.entry_offset
,
2206 &o
->data
.entry_array_offset
,
2208 JOURNAL_HEADER_COMPACT(f
->header
) ? &o
->data
.compact
.tail_entry_array_offset
: NULL
,
2209 JOURNAL_HEADER_COMPACT(f
->header
) ? &o
->data
.compact
.tail_entry_array_n_entries
: NULL
,
2213 static int journal_file_link_entry(
2217 const EntryItem items
[],
2227 if (o
->object
.type
!= OBJECT_ENTRY
)
2230 __atomic_thread_fence(__ATOMIC_SEQ_CST
);
2232 /* Link up the entry itself */
2233 r
= link_entry_into_array(f
,
2234 &f
->header
->entry_array_offset
,
2235 &f
->header
->n_entries
,
2236 JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_array_offset
) ? &f
->header
->tail_entry_array_offset
: NULL
,
2237 JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_array_n_entries
) ? &f
->header
->tail_entry_array_n_entries
: NULL
,
2242 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
2244 if (f
->header
->head_entry_realtime
== 0)
2245 f
->header
->head_entry_realtime
= o
->entry
.realtime
;
2247 f
->header
->tail_entry_realtime
= o
->entry
.realtime
;
2248 f
->header
->tail_entry_monotonic
= o
->entry
.monotonic
;
2249 if (JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_offset
))
2250 f
->header
->tail_entry_offset
= htole64(offset
);
2251 f
->newest_mtime
= 0; /* we have a new tail entry now, explicitly invalidate newest boot id/timestamp info */
2253 /* Link up the items */
2254 for (uint64_t i
= 0; i
< n_items
; i
++) {
2257 /* If we fail to link an entry item because we can't allocate a new entry array, don't fail
2258 * immediately but try to link the other entry items since it might still be possible to link
2259 * those if they don't require a new entry array to be allocated. */
2261 k
= journal_file_link_entry_item(f
, offset
, items
[i
].object_offset
);
2271 static void write_entry_item(JournalFile
*f
, Object
*o
, uint64_t i
, const EntryItem
*item
) {
2276 if (JOURNAL_HEADER_COMPACT(f
->header
)) {
2277 assert(item
->object_offset
<= UINT32_MAX
);
2278 o
->entry
.items
.compact
[i
].object_offset
= htole32(item
->object_offset
);
2280 o
->entry
.items
.regular
[i
].object_offset
= htole64(item
->object_offset
);
2281 o
->entry
.items
.regular
[i
].hash
= htole64(item
->hash
);
2285 static int journal_file_append_entry_internal(
2287 const dual_timestamp
*ts
,
2288 const sd_id128_t
*boot_id
,
2289 const sd_id128_t
*machine_id
,
2291 const EntryItem items
[],
2294 sd_id128_t
*seqnum_id
,
2295 Object
**ret_object
,
2296 uint64_t *ret_offset
) {
2307 assert(!sd_id128_is_null(*boot_id
));
2308 assert(items
|| n_items
== 0);
2310 if (f
->strict_order
) {
2311 /* If requested be stricter with ordering in this journal file, to make searching via
2312 * bisection fully deterministic. This is an optional feature, so that if desired journal
2313 * files can be written where the ordering is not strictly enforced (in which case bisection
2314 * will yield *a* result, but not the *only* result, when searching for points in
2315 * time). Strict ordering mode is enabled when journald originally writes the files, but
2316 * might not necessarily be if other tools (the remoting tools for example) write journal
2317 * files from combined sources.
2319 * Typically, if any of the errors generated here are seen journald will just rotate the
2320 * journal files and start anew. */
2322 if (ts
->realtime
< le64toh(f
->header
->tail_entry_realtime
))
2323 return log_debug_errno(SYNTHETIC_ERRNO(EREMCHG
),
2324 "Realtime timestamp %" PRIu64
" smaller than previous realtime "
2325 "timestamp %" PRIu64
", refusing entry.",
2326 ts
->realtime
, le64toh(f
->header
->tail_entry_realtime
));
2328 if (sd_id128_equal(*boot_id
, f
->header
->tail_entry_boot_id
) &&
2329 ts
->monotonic
< le64toh(f
->header
->tail_entry_monotonic
))
2330 return log_debug_errno(
2331 SYNTHETIC_ERRNO(ENOTNAM
),
2332 "Monotonic timestamp %" PRIu64
2333 " smaller than previous monotonic timestamp %" PRIu64
2334 " while having the same boot ID, refusing entry.",
2336 le64toh(f
->header
->tail_entry_monotonic
));
2340 /* Settle the passed in sequence number ID */
2342 if (sd_id128_is_null(*seqnum_id
))
2343 *seqnum_id
= f
->header
->seqnum_id
; /* Caller has none assigned, then copy the one from the file */
2344 else if (!sd_id128_equal(*seqnum_id
, f
->header
->seqnum_id
)) {
2345 /* Different seqnum IDs? We can't allow entries from multiple IDs end up in the same journal.*/
2346 if (le64toh(f
->header
->n_entries
) == 0)
2347 f
->header
->seqnum_id
= *seqnum_id
; /* Caller has one, and file so far has no entries, then copy the one from the caller */
2349 return log_debug_errno(SYNTHETIC_ERRNO(EILSEQ
),
2350 "Sequence number IDs don't match, refusing entry.");
2354 if (machine_id
&& sd_id128_is_null(f
->header
->machine_id
))
2355 /* Initialize machine ID when not set yet */
2356 f
->header
->machine_id
= *machine_id
;
2358 osize
= offsetof(Object
, entry
.items
) + (n_items
* journal_file_entry_item_size(f
));
2360 r
= journal_file_append_object(f
, OBJECT_ENTRY
, osize
, &o
, &np
);
2364 o
->entry
.seqnum
= htole64(journal_file_entry_seqnum(f
, seqnum
));
2365 o
->entry
.realtime
= htole64(ts
->realtime
);
2366 o
->entry
.monotonic
= htole64(ts
->monotonic
);
2367 o
->entry
.xor_hash
= htole64(xor_hash
);
2368 o
->entry
.boot_id
= f
->header
->tail_entry_boot_id
= *boot_id
;
2370 for (size_t i
= 0; i
< n_items
; i
++)
2371 write_entry_item(f
, o
, i
, &items
[i
]);
2374 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY
, o
, np
);
2379 r
= journal_file_link_entry(f
, o
, np
, items
, n_items
);
2392 void journal_file_post_change(JournalFile
*f
) {
2398 /* inotify() does not receive IN_MODIFY events from file
2399 * accesses done via mmap(). After each access we hence
2400 * trigger IN_MODIFY by truncating the journal file to its
2401 * current size which triggers IN_MODIFY. */
2403 __atomic_thread_fence(__ATOMIC_SEQ_CST
);
2405 if (ftruncate(f
->fd
, f
->last_stat
.st_size
) < 0)
2406 log_debug_errno(errno
, "Failed to truncate file to its own size: %m");
2409 static int post_change_thunk(sd_event_source
*timer
, uint64_t usec
, void *userdata
) {
2412 journal_file_post_change(userdata
);
2417 static void schedule_post_change(JournalFile
*f
) {
2422 assert(f
->post_change_timer
);
2424 assert_se(e
= sd_event_source_get_event(f
->post_change_timer
));
2426 /* If we are already going down, post the change immediately. */
2427 if (IN_SET(sd_event_get_state(e
), SD_EVENT_EXITING
, SD_EVENT_FINISHED
))
2430 r
= sd_event_source_get_enabled(f
->post_change_timer
, NULL
);
2432 log_debug_errno(r
, "Failed to get ftruncate timer state: %m");
2438 r
= sd_event_source_set_time_relative(f
->post_change_timer
, f
->post_change_timer_period
);
2440 log_debug_errno(r
, "Failed to set time for scheduling ftruncate: %m");
2444 r
= sd_event_source_set_enabled(f
->post_change_timer
, SD_EVENT_ONESHOT
);
2446 log_debug_errno(r
, "Failed to enable scheduled ftruncate: %m");
2453 /* On failure, let's simply post the change immediately. */
2454 journal_file_post_change(f
);
2457 /* Enable coalesced change posting in a timer on the provided sd_event instance */
2458 int journal_file_enable_post_change_timer(JournalFile
*f
, sd_event
*e
, usec_t t
) {
2459 _cleanup_(sd_event_source_unrefp
) sd_event_source
*timer
= NULL
;
2463 assert_return(!f
->post_change_timer
, -EINVAL
);
2467 r
= sd_event_add_time(e
, &timer
, CLOCK_MONOTONIC
, 0, 0, post_change_thunk
, f
);
2471 r
= sd_event_source_set_enabled(timer
, SD_EVENT_OFF
);
2475 f
->post_change_timer
= TAKE_PTR(timer
);
2476 f
->post_change_timer_period
= t
;
2481 static int entry_item_cmp(const EntryItem
*a
, const EntryItem
*b
) {
2482 return CMP(ASSERT_PTR(a
)->object_offset
, ASSERT_PTR(b
)->object_offset
);
2485 static size_t remove_duplicate_entry_items(EntryItem items
[], size_t n
) {
2488 assert(items
|| n
== 0);
2493 for (size_t i
= 1; i
< n
; i
++)
2494 if (items
[i
].object_offset
!= items
[j
- 1].object_offset
)
2495 items
[j
++] = items
[i
];
2500 int journal_file_append_entry(
2502 const dual_timestamp
*ts
,
2503 const sd_id128_t
*boot_id
,
2504 const struct iovec iovec
[],
2507 sd_id128_t
*seqnum_id
,
2508 Object
**ret_object
,
2509 uint64_t *ret_offset
) {
2511 _cleanup_free_ EntryItem
*items_alloc
= NULL
;
2513 uint64_t xor_hash
= 0;
2514 struct dual_timestamp _ts
;
2515 sd_id128_t _boot_id
, _machine_id
, *machine_id
;
2521 assert(n_iovec
> 0);
2524 if (!VALID_REALTIME(ts
->realtime
))
2525 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
2526 "Invalid realtime timestamp %" PRIu64
", refusing entry.",
2528 if (!VALID_MONOTONIC(ts
->monotonic
))
2529 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
2530 "Invalid monotomic timestamp %" PRIu64
", refusing entry.",
2533 dual_timestamp_get(&_ts
);
2538 if (sd_id128_is_null(*boot_id
))
2539 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
), "Empty boot ID, refusing entry.");
2541 r
= sd_id128_get_boot(&_boot_id
);
2545 boot_id
= &_boot_id
;
2548 r
= sd_id128_get_machine(&_machine_id
);
2549 if (ERRNO_IS_NEG_MACHINE_ID_UNSET(r
))
2550 /* Gracefully handle the machine ID not being initialized yet */
2555 machine_id
= &_machine_id
;
2558 r
= journal_file_maybe_append_tag(f
, ts
->realtime
);
2563 if (n_iovec
< ALLOCA_MAX
/ sizeof(EntryItem
) / 2)
2564 items
= newa(EntryItem
, n_iovec
);
2566 items_alloc
= new(EntryItem
, n_iovec
);
2570 items
= items_alloc
;
2573 for (size_t i
= 0; i
< n_iovec
; i
++) {
2577 r
= journal_file_append_data(f
, iovec
[i
].iov_base
, iovec
[i
].iov_len
, &o
, &p
);
2581 /* When calculating the XOR hash field, we need to take special care if the "keyed-hash"
2582 * journal file flag is on. We use the XOR hash field to quickly determine the identity of a
2583 * specific record, and give records with otherwise identical position (i.e. match in seqno,
2584 * timestamp, …) a stable ordering. But for that we can't have it that the hash of the
2585 * objects in each file is different since they are keyed. Hence let's calculate the Jenkins
2586 * hash here for that. This also has the benefit that cursors for old and new journal files
2587 * are completely identical (they include the XOR hash after all). For classic Jenkins-hash
2588 * files things are easier, we can just take the value from the stored record directly. */
2590 if (JOURNAL_HEADER_KEYED_HASH(f
->header
))
2591 xor_hash
^= jenkins_hash64(iovec
[i
].iov_base
, iovec
[i
].iov_len
);
2593 xor_hash
^= le64toh(o
->data
.hash
);
2595 items
[i
] = (EntryItem
) {
2597 .hash
= le64toh(o
->data
.hash
),
2601 /* Order by the position on disk, in order to improve seek
2602 * times for rotating media. */
2603 typesafe_qsort(items
, n_iovec
, entry_item_cmp
);
2604 n_iovec
= remove_duplicate_entry_items(items
, n_iovec
);
2606 r
= journal_file_append_entry_internal(
2619 /* If the memory mapping triggered a SIGBUS then we return an
2620 * IO error and ignore the error code passed down to us, since
2621 * it is very likely just an effect of a nullified replacement
2624 if (mmap_cache_fd_got_sigbus(f
->cache_fd
))
2627 if (f
->post_change_timer
)
2628 schedule_post_change(f
);
2630 journal_file_post_change(f
);
2635 typedef struct ChainCacheItem
{
2636 uint64_t first
; /* The offset of the entry array object at the beginning of the chain,
2637 * i.e., le64toh(f->header->entry_array_offset), or le64toh(o->data.entry_offset). */
2638 uint64_t array
; /* The offset of the cached entry array object. */
2639 uint64_t begin
; /* The offset of the first item in the cached array. */
2640 uint64_t total
; /* The total number of items in all arrays before the cached one in the chain. */
2641 uint64_t last_index
; /* The last index we looked at in the cached array, to optimize locality when bisecting. */
2644 static void chain_cache_put(
2651 uint64_t last_index
) {
2656 /* If the chain item to cache for this chain is the
2657 * first one it's not worth caching anything */
2661 if (ordered_hashmap_size(h
) >= CHAIN_CACHE_MAX
) {
2662 ci
= ordered_hashmap_steal_first(h
);
2665 ci
= new(ChainCacheItem
, 1);
2672 if (ordered_hashmap_put(h
, &ci
->first
, ci
) < 0) {
2677 assert(ci
->first
== first
);
2682 ci
->last_index
= last_index
;
2685 static int bump_array_index(uint64_t *i
, direction_t direction
, uint64_t n
) {
2688 /* Increase or decrease the specified index, in the right direction. */
2690 if (direction
== DIRECTION_DOWN
) {
2705 static int bump_entry_array(
2707 Object
*o
, /* the current entry array object. */
2708 uint64_t offset
, /* the offset of the entry array object. */
2709 uint64_t first
, /* The offset of the first entry array object in the chain. */
2710 direction_t direction
,
2718 if (direction
== DIRECTION_DOWN
) {
2720 assert(o
->object
.type
== OBJECT_ENTRY_ARRAY
);
2722 *ret
= le64toh(o
->entry_array
.next_entry_array_offset
);
2725 /* Entry array chains are a singly linked list, so to find the previous array in the chain, we have
2726 * to start iterating from the top. */
2730 uint64_t p
= first
, q
= 0;
2731 while (p
> 0 && p
!= offset
) {
2732 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, p
, &o
);
2737 p
= le64toh(o
->entry_array
.next_entry_array_offset
);
2740 /* If we can't find the previous entry array in the entry array chain, we're likely dealing with a
2741 * corrupted journal file. */
2751 static int generic_array_get(
2753 uint64_t first
, /* The offset of the first entry array object in the chain. */
2754 uint64_t i
, /* The index of the target object counted from the beginning of the entry array chain. */
2755 direction_t direction
,
2756 Object
**ret_object
, /* The found object. */
2757 uint64_t *ret_offset
) { /* The offset of the found object. */
2759 uint64_t a
, t
= 0, k
;
2766 /* FIXME: fix return value assignment on success. */
2770 /* Try the chain cache first */
2771 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2772 if (ci
&& i
> ci
->total
) {
2779 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2780 if (IN_SET(r
, -EBADMSG
, -EADDRNOTAVAIL
)) {
2781 /* If there's corruption and we're going downwards, let's pretend we reached the
2782 * final entry in the entry array chain. */
2784 if (direction
== DIRECTION_DOWN
)
2787 /* If there's corruption and we're going upwards, move back to the previous entry
2788 * array and start iterating entries from there. */
2796 k
= journal_file_entry_array_n_items(f
, o
);
2803 /* The index is larger than the number of elements in the array. Let's move to the next array. */
2806 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
2809 /* If we've found the right location, now look for the first non-corrupt entry object (in the right
2813 if (i
== UINT64_MAX
) {
2814 r
= bump_entry_array(f
, o
, a
, first
, direction
, &a
);
2818 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2822 k
= journal_file_entry_array_n_items(f
, o
);
2826 if (direction
== DIRECTION_DOWN
)
2829 /* We moved to the previous array. The total must be decreased. */
2831 return -EBADMSG
; /* chain cache is broken ? */
2841 p
= journal_file_entry_array_item(f
, o
, i
);
2843 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, ret_object
);
2845 /* Let's cache this item for the next invocation */
2846 chain_cache_put(f
->chain_cache
, ci
, first
, a
, journal_file_entry_array_item(f
, o
, 0), t
, i
);
2853 if (!IN_SET(r
, -EADDRNOTAVAIL
, -EBADMSG
))
2856 /* OK, so this entry is borked. Most likely some entry didn't get synced to
2857 * disk properly, let's see if the next one might work for us instead. */
2858 log_debug_errno(r
, "Entry item %" PRIu64
" is bad, skipping over it.", i
);
2860 } while (bump_array_index(&i
, direction
, k
) > 0);
2862 /* All entries tried in the above do-while loop are broken. Let's move to the next (or previous) array. */
2864 if (direction
== DIRECTION_DOWN
)
2865 /* We are going to the next array, the total must be incremented. */
2875 TEST_FOUND
, /* The current object passes the test. */
2876 TEST_LEFT
, /* The current object is in an earlier position, and the object we are looking
2877 * for should exist in a later position. */
2878 TEST_RIGHT
, /* The current object is in a later position, and the object we are looking for
2879 * should exist in an earlier position. */
2880 TEST_GOTO_NEXT
, /* No matching object exists in this array and earlier arrays, go to the next array. */
2881 TEST_GOTO_PREVIOUS
, /* No matching object exists in this array and later arrays, go to the previous array. */
2884 static int generic_array_bisect_step(
2886 Object
*array
, /* entry array object */
2887 uint64_t i
, /* index of the entry item in the array we will test. */
2889 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2890 direction_t direction
,
2891 uint64_t *m
, /* The maximum number of the entries we will check in the array. */
2892 uint64_t *left
, /* The index of the left boundary in the array. */
2893 uint64_t *right
) { /* The index of the right boundary in the array. */
2900 assert(test_object
);
2905 assert(i
<= *right
);
2906 assert(*right
< *m
);
2908 p
= journal_file_entry_array_item(f
, array
, i
);
2912 r
= test_object(f
, p
, needle
);
2913 if (IN_SET(r
, -EBADMSG
, -EADDRNOTAVAIL
)) {
2914 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short.");
2916 /* The first entry in the array is corrupted, let's go back to the previous array. */
2918 return TEST_GOTO_PREVIOUS
;
2920 /* Otherwise, cutting the array short. So, here we limit the number of elements we will see
2921 * in this array, and set the right boundary to the last possibly non-corrupted object. */
2929 if (r
== TEST_FOUND
)
2930 /* There may be multiple entries that match with the needle. When the direction is down, we
2931 * need to find the first matching entry, hence the right boundary can be moved, but the left
2932 * one cannot. Similarly, when the direction is up, we need to find the last matching entry,
2933 * hence the left boundary can be moved, but the right one cannot. */
2934 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2936 if (r
== TEST_RIGHT
) {
2937 /* Currently, left --- needle --- i --- right, hence we can move the right boundary to i. */
2938 if (direction
== DIRECTION_DOWN
)
2942 return TEST_GOTO_PREVIOUS
;
2946 /* Currently, left --- i --- needle --- right, hence we can move the left boundary to i. */
2947 if (direction
== DIRECTION_DOWN
) {
2948 /* Note, here *m is always positive, as by the assertions at the beginning, we have
2949 * 0 <= *left <= i <= *right < m */
2951 return TEST_GOTO_NEXT
;
2961 static int generic_array_bisect(
2963 uint64_t first
, /* The offset of the first entry array object in the chain. */
2964 uint64_t n
, /* The total number of elements in the chain of the entry array. */
2965 uint64_t needle
, /* The target value (e.g. seqnum, monotonic, realtime, ...). */
2966 int (*test_object
)(JournalFile
*f
,
2967 uint64_t p
, /* the offset of the (data or entry) object that will be tested. */
2969 direction_t direction
,
2970 Object
**ret_object
, /* The found object. */
2971 uint64_t *ret_offset
, /* The offset of the found object. */
2972 uint64_t *ret_idx
) { /* The index of the found object counted from the beginning of the entry array chain. */
2974 /* Given an entry array chain, this function finds the object "closest" to the given needle in the
2975 * chain, taking into account the provided direction. A function can be provided to determine how
2976 * an object is matched against the given needle.
2978 * Given a journal file, the offset of an object and the needle, the test_object() function should
2979 * return TEST_RIGHT if the needle is located earlier in the entry array chain, TEST_LEFT if the
2980 * needle is located later in the entry array chain, and TEST_FOUND if the object matches the needle.
2981 * If test_object() returns TEST_FOUND for a specific object, that object's information will be used
2982 * to populate the return values of this function. If test_object() never returns TEST_FOUND, the
2983 * return values are populated with the details of one of the objects closest to the needle. If the
2984 * direction is DIRECTION_UP, the earlier object is used. Otherwise, the later object is used.
2985 * If there are multiple objects that test_object() return TEST_FOUND for, then the first matching
2986 * object returned when direction is DIRECTION_DOWN. Otherwise the last object is returned. */
2988 uint64_t a
, p
, t
= 0, i
, last_index
= UINT64_MAX
;
2994 assert(test_object
);
2999 /* Start with the first array in the chain */
3002 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
3003 if (ci
&& n
> ci
->total
&& ci
->begin
!= 0) {
3004 /* Ah, we have iterated this bisection array chain previously! Let's see if we can skip ahead
3005 * in the chain, as far as the last time. But we can't jump backwards in the chain, so let's
3006 * check that first. */
3008 r
= test_object(f
, ci
->begin
, needle
);
3012 if (r
== TEST_LEFT
) {
3013 /* OK, what we are looking for is right of the begin of this EntryArray, so let's
3014 * jump straight to previously cached array in the chain */
3019 last_index
= ci
->last_index
;
3024 uint64_t left
, right
, k
, m
;
3026 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
3030 k
= journal_file_entry_array_n_items(f
, array
);
3038 if (direction
== DIRECTION_UP
) {
3039 /* If we're going upwards, the last entry of the previous array may pass the test,
3040 * and the first entry of the current array may not pass. In that case, the last
3041 * entry of the previous array must be returned. Hence, we need to test the first
3042 * entry of the current array. */
3043 r
= generic_array_bisect_step(f
, array
, 0, needle
, test_object
, direction
, &m
, &left
, &right
);
3046 if (r
== TEST_GOTO_PREVIOUS
)
3050 /* Test the last entry of this array, to determine if we should go to the next array. */
3051 r
= generic_array_bisect_step(f
, array
, right
, needle
, test_object
, direction
, &m
, &left
, &right
);
3054 if (r
== TEST_GOTO_PREVIOUS
)
3057 /* The expected entry should be in this array, (or the last entry of the previous array). */
3058 if (r
== TEST_RIGHT
) {
3060 /* If we cached the last index we looked at, let's try to not to jump too wildly
3061 * around and see if we can limit the range to look at early to the immediate
3062 * neighbors of the last index we looked at. */
3064 if (last_index
> 0 && left
< last_index
- 1 && last_index
- 1 < right
) {
3065 r
= generic_array_bisect_step(f
, array
, last_index
- 1, needle
, test_object
, direction
, &m
, &left
, &right
);
3068 if (r
== TEST_GOTO_PREVIOUS
)
3072 if (last_index
< UINT64_MAX
&& left
< last_index
+ 1 && last_index
+ 1 < right
) {
3073 r
= generic_array_bisect_step(f
, array
, last_index
+ 1, needle
, test_object
, direction
, &m
, &left
, &right
);
3076 if (r
== TEST_GOTO_PREVIOUS
)
3081 if (left
== right
) {
3086 assert(left
< right
);
3087 i
= (left
+ right
+ (direction
== DIRECTION_UP
)) / 2;
3089 r
= generic_array_bisect_step(f
, array
, i
, needle
, test_object
, direction
, &m
, &left
, &right
);
3092 if (r
== TEST_GOTO_PREVIOUS
)
3097 /* Not found in this array (or the last entry of this array should be returned), go to the next array. */
3098 assert(r
== (direction
== DIRECTION_DOWN
? TEST_GOTO_NEXT
: TEST_LEFT
));
3101 if (direction
== DIRECTION_UP
) {
3112 last_index
= UINT64_MAX
;
3113 a
= le64toh(array
->entry_array
.next_entry_array_offset
);
3119 /* Not found in the current array, return the last entry of the previous array. */
3120 assert(r
== TEST_GOTO_PREVIOUS
);
3122 /* The current array is the first in the chain. no previous array. */
3126 /* When we are going downwards, there is no matching entries in the previous array. */
3127 if (direction
== DIRECTION_DOWN
)
3130 /* Indicate to go to the previous array later. Note, do not move to the previous array here,
3131 * as that may invalidate the current array object in the mmap cache and
3132 * journal_file_entry_array_item() below may read invalid address. */
3136 p
= journal_file_entry_array_item(f
, array
, 0);
3140 /* Let's cache this item for the next invocation */
3141 chain_cache_put(f
->chain_cache
, ci
, first
, a
, p
, t
, i
);
3143 if (i
== UINT64_MAX
) {
3146 /* Get the last entry of the previous array. */
3148 r
= bump_entry_array(f
, NULL
, a
, first
, DIRECTION_UP
, &a
);
3152 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
3156 m
= journal_file_entry_array_n_items(f
, array
);
3157 if (m
== 0 || t
< m
)
3164 p
= journal_file_entry_array_item(f
, array
, i
);
3169 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, ret_object
);
3183 static int generic_array_bisect_for_data(
3187 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
3188 direction_t direction
,
3189 Object
**ret_object
,
3190 uint64_t *ret_offset
) {
3192 uint64_t extra
, first
, n
;
3197 assert(d
->object
.type
== OBJECT_DATA
);
3198 assert(test_object
);
3200 n
= le64toh(d
->data
.n_entries
);
3203 n
--; /* n_entries is the number of entries linked to the data object, including the 'extra' entry. */
3205 extra
= le64toh(d
->data
.entry_offset
);
3206 first
= le64toh(d
->data
.entry_array_offset
);
3208 /* This bisects the array in object 'first', but first checks an extra. */
3209 r
= test_object(f
, extra
, needle
);
3213 if (direction
== DIRECTION_DOWN
) {
3214 /* If we are going downwards, then we need to return the first object that passes the test.
3215 * When there is no object that passes the test, we need to return the first object that
3216 * test_object() returns TEST_RIGHT for. */
3218 TEST_FOUND
, /* The 'extra' object passes the test. Hence, this is the first
3219 * object that passes the test. */
3220 TEST_RIGHT
)) /* The 'extra' object is the first object that test_object() returns
3221 * TEST_RIGHT for, and no object exists even in the chained arrays
3222 * that passes the test. */
3223 goto use_extra
; /* The 'extra' object is exactly the one we are looking for. It is
3224 * not necessary to bisect the chained arrays. */
3226 /* Otherwise, the 'extra' object is not the one we are looking for. Search in the arrays. */
3229 /* If we are going upwards, then we need to return the last object that passes the test.
3230 * When there is no object that passes the test, we need to return the the last object that
3231 * test_object() returns TEST_LEFT for. */
3232 if (r
== TEST_RIGHT
)
3233 return 0; /* Not only the 'extra' object, but also all objects in the chained arrays
3234 * will never get TEST_FOUND or TEST_LEFT. The object we are looking for
3235 * does not exist. */
3237 /* Even if the 'extra' object passes the test, there may be multiple objects in the arrays
3238 * that also pass the test. Hence, we need to bisect the arrays for finding the last matching
3242 r
= generic_array_bisect(f
, first
, n
, needle
, test_object
, direction
, ret_object
, ret_offset
, NULL
);
3244 return r
; /* When > 0, the found object is the first (or last, when DIRECTION_UP) object.
3245 * Hence, return the found object now. */
3247 /* No matching object found in the chained arrays.
3248 * DIRECTION_DOWN : the 'extra' object neither matches the condition. There is no matching object.
3249 * DIRECTION_UP : the 'extra' object matches the condition. So, return it. */
3250 if (direction
== DIRECTION_DOWN
)
3255 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, ret_object
);
3261 *ret_offset
= extra
;
3266 static int test_object_offset(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
3272 else if (p
< needle
)
3278 int journal_file_move_to_entry_by_offset(
3281 direction_t direction
,
3282 Object
**ret_object
,
3283 uint64_t *ret_offset
) {
3288 return generic_array_bisect(
3290 le64toh(f
->header
->entry_array_offset
),
3291 le64toh(f
->header
->n_entries
),
3295 ret_object
, ret_offset
, NULL
);
3298 static int test_object_seqnum(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
3306 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
3310 sq
= le64toh(READ_NOW(o
->entry
.seqnum
));
3313 else if (sq
< needle
)
3319 int journal_file_move_to_entry_by_seqnum(
3322 direction_t direction
,
3323 Object
**ret_object
,
3324 uint64_t *ret_offset
) {
3329 return generic_array_bisect(
3331 le64toh(f
->header
->entry_array_offset
),
3332 le64toh(f
->header
->n_entries
),
3336 ret_object
, ret_offset
, NULL
);
3339 static int test_object_realtime(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
3347 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
3351 rt
= le64toh(READ_NOW(o
->entry
.realtime
));
3354 else if (rt
< needle
)
3360 int journal_file_move_to_entry_by_realtime(
3363 direction_t direction
,
3364 Object
**ret_object
,
3365 uint64_t *ret_offset
) {
3370 return generic_array_bisect(
3372 le64toh(f
->header
->entry_array_offset
),
3373 le64toh(f
->header
->n_entries
),
3375 test_object_realtime
,
3377 ret_object
, ret_offset
, NULL
);
3380 static int test_object_monotonic(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
3388 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
3392 m
= le64toh(READ_NOW(o
->entry
.monotonic
));
3395 else if (m
< needle
)
3401 static int find_data_object_by_boot_id(
3404 Object
**ret_object
,
3405 uint64_t *ret_offset
) {
3407 char t
[STRLEN("_BOOT_ID=") + 32 + 1] = "_BOOT_ID=";
3411 sd_id128_to_string(boot_id
, t
+ 9);
3412 return journal_file_find_data_object(f
, t
, sizeof(t
) - 1, ret_object
, ret_offset
);
3415 int journal_file_move_to_entry_by_monotonic(
3419 direction_t direction
,
3420 Object
**ret_object
,
3421 uint64_t *ret_offset
) {
3428 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, NULL
);
3432 return generic_array_bisect_for_data(
3436 test_object_monotonic
,
3438 ret_object
, ret_offset
);
3441 void journal_file_reset_location(JournalFile
*f
) {
3444 f
->location_type
= LOCATION_HEAD
;
3445 f
->current_offset
= 0;
3446 f
->current_seqnum
= 0;
3447 f
->current_realtime
= 0;
3448 f
->current_monotonic
= 0;
3449 zero(f
->current_boot_id
);
3450 f
->current_xor_hash
= 0;
3452 /* Also reset the previous reading direction. Otherwise, next_beyond_location() may wrongly handle we
3453 * already hit EOF. See issue #29216. */
3454 f
->last_direction
= _DIRECTION_INVALID
;
3457 void journal_file_save_location(JournalFile
*f
, Object
*o
, uint64_t offset
) {
3461 f
->location_type
= LOCATION_SEEK
;
3462 f
->current_offset
= offset
;
3463 f
->current_seqnum
= le64toh(o
->entry
.seqnum
);
3464 f
->current_realtime
= le64toh(o
->entry
.realtime
);
3465 f
->current_monotonic
= le64toh(o
->entry
.monotonic
);
3466 f
->current_boot_id
= o
->entry
.boot_id
;
3467 f
->current_xor_hash
= le64toh(o
->entry
.xor_hash
);
3470 static bool check_properly_ordered(uint64_t new_offset
, uint64_t old_offset
, direction_t direction
) {
3472 /* Consider it an error if any of the two offsets is uninitialized */
3473 if (old_offset
== 0 || new_offset
== 0)
3476 /* If we go down, the new offset must be larger than the old one. */
3477 return direction
== DIRECTION_DOWN
?
3478 new_offset
> old_offset
:
3479 new_offset
< old_offset
;
3482 int journal_file_next_entry(
3485 direction_t direction
,
3486 Object
**ret_object
,
3487 uint64_t *ret_offset
) {
3496 /* FIXME: fix return value assignment. */
3498 n
= le64toh(READ_NOW(f
->header
->n_entries
));
3502 /* When the input offset 'p' is zero, return the first (or last on DIRECTION_UP) entry. */
3504 return generic_array_get(f
,
3505 le64toh(f
->header
->entry_array_offset
),
3506 direction
== DIRECTION_DOWN
? 0 : n
- 1,
3508 ret_object
, ret_offset
);
3510 /* Otherwise, first find the nearest entry object. */
3511 r
= generic_array_bisect(f
,
3512 le64toh(f
->header
->entry_array_offset
),
3513 le64toh(f
->header
->n_entries
),
3517 ret_object
? &o
: NULL
, &q
, &i
);
3521 assert(direction
== DIRECTION_DOWN
? p
<= q
: q
<= p
);
3523 /* If the input offset 'p' points to an entry object, generic_array_bisect() should provides
3524 * the same offset, and the index needs to be shifted. Otherwise, use the found object as is,
3525 * as it is the nearest entry object from the input offset 'p'. */
3530 r
= bump_array_index(&i
, direction
, n
);
3534 /* And jump to it */
3535 r
= generic_array_get(f
, le64toh(f
->header
->entry_array_offset
), i
, direction
, ret_object
? &o
: NULL
, &q
);
3539 /* Ensure our array is properly ordered. */
3540 if (!check_properly_ordered(q
, p
, direction
))
3541 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
3542 "%s: entry array not properly ordered at entry index %" PRIu64
,
3553 int journal_file_move_to_entry_for_data(
3556 direction_t direction
,
3557 Object
**ret_object
,
3558 uint64_t *ret_offset
) {
3560 uint64_t extra
, first
, n
;
3565 assert(d
->object
.type
== OBJECT_DATA
);
3566 assert(IN_SET(direction
, DIRECTION_DOWN
, DIRECTION_UP
));
3568 /* FIXME: fix return value assignment. */
3570 /* This returns the first (when the direction is down, otherwise the last) entry linked to the
3571 * specified data object. */
3573 n
= le64toh(d
->data
.n_entries
);
3576 n
--; /* n_entries is the number of entries linked to the data object, including the 'extra' entry. */
3578 extra
= le64toh(d
->data
.entry_offset
);
3579 first
= le64toh(d
->data
.entry_array_offset
);
3581 if (direction
== DIRECTION_DOWN
&& extra
> 0) {
3582 /* When we are going downwards, first try to read the extra entry. */
3583 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, ret_object
);
3586 if (!IN_SET(r
, -EADDRNOTAVAIL
, -EBADMSG
))
3591 /* DIRECTION_DOWN : The extra entry is broken, falling back to the entries in the array.
3592 * DIRECTION_UP : Try to find a valid entry in the array from the tail. */
3593 r
= generic_array_get(f
,
3595 direction
== DIRECTION_DOWN
? 0 : n
- 1,
3597 ret_object
, ret_offset
);
3598 if (!IN_SET(r
, 0, -EADDRNOTAVAIL
, -EBADMSG
))
3599 return r
; /* found or critical error. */
3602 if (direction
== DIRECTION_UP
&& extra
> 0) {
3603 /* No valid entry exists in the chained array, falling back to the extra entry. */
3604 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, ret_object
);
3613 *ret_offset
= extra
;
3618 int journal_file_move_to_entry_by_offset_for_data(
3622 direction_t direction
,
3623 Object
**ret
, uint64_t *ret_offset
) {
3627 assert(d
->object
.type
== OBJECT_DATA
);
3629 return generic_array_bisect_for_data(
3638 int journal_file_move_to_entry_by_monotonic_for_data(
3643 direction_t direction
,
3644 Object
**ret_object
,
3645 uint64_t *ret_offset
) {
3653 assert(d
->object
.type
== OBJECT_DATA
);
3655 /* First, pin the given data object, before reading the _BOOT_ID= data object below. */
3656 r
= journal_file_pin_object(f
, d
);
3660 /* Then, read a data object for _BOOT_ID= and seek by time. */
3661 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, NULL
);
3665 r
= generic_array_bisect_for_data(f
,
3668 test_object_monotonic
,
3674 /* And now, continue seeking until we find an entry that exists in both bisection arrays. */
3678 /* The journal entry found by the above bisect_plus_one() may not have the specified data,
3679 * that is, it may not be linked in the data object. So, we need to check that. */
3681 r
= journal_file_move_to_entry_by_offset_for_data(
3682 f
, d
, z
, direction
, ret_object
? &entry
: NULL
, &p
);
3686 break; /* The journal entry has the specified data. Yay! */
3688 /* If the entry does not have the data, then move to the next (or previous, depends on the
3689 * 'direction') entry linked to the data object. But, the next entry may be in another boot.
3690 * So, we need to check that the entry has the matching boot ID. */
3692 r
= journal_file_move_to_entry_by_offset_for_data(
3693 f
, o
, p
, direction
, ret_object
? &entry
: NULL
, &z
);
3697 break; /* The journal entry has the specified boot ID. Yay! */
3699 /* If not, let's try to the next entry... */
3703 *ret_object
= entry
;
3709 int journal_file_move_to_entry_by_seqnum_for_data(
3713 direction_t direction
,
3714 Object
**ret_object
,
3715 uint64_t *ret_offset
) {
3719 assert(d
->object
.type
== OBJECT_DATA
);
3721 return generic_array_bisect_for_data(
3727 ret_object
, ret_offset
);
3730 int journal_file_move_to_entry_by_realtime_for_data(
3734 direction_t direction
,
3735 Object
**ret
, uint64_t *ret_offset
) {
3739 assert(d
->object
.type
== OBJECT_DATA
);
3741 return generic_array_bisect_for_data(
3745 test_object_realtime
,
3750 void journal_file_dump(JournalFile
*f
) {
3758 journal_file_print_header(f
);
3760 p
= le64toh(READ_NOW(f
->header
->header_size
));
3765 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &o
);
3769 s
= journal_object_type_to_string(o
->object
.type
);
3771 switch (o
->object
.type
) {
3776 printf("Type: %s seqnum=%"PRIu64
" monotonic=%"PRIu64
" realtime=%"PRIu64
"\n",
3778 le64toh(o
->entry
.seqnum
),
3779 le64toh(o
->entry
.monotonic
),
3780 le64toh(o
->entry
.realtime
));
3786 printf("Type: %s seqnum=%"PRIu64
" epoch=%"PRIu64
"\n",
3788 le64toh(o
->tag
.seqnum
),
3789 le64toh(o
->tag
.epoch
));
3794 printf("Type: %s \n", s
);
3796 printf("Type: unknown (%i)", o
->object
.type
);
3801 c
= COMPRESSION_FROM_OBJECT(o
);
3802 if (c
> COMPRESSION_NONE
)
3803 printf("Flags: %s\n",
3804 compression_to_string(c
));
3806 if (p
== le64toh(f
->header
->tail_object_offset
))
3809 p
+= ALIGN64(le64toh(o
->object
.size
));
3814 log_error("File corrupt");
3817 /* Note: the lifetime of the compound literal is the immediately surrounding block. */
3818 #define FORMAT_TIMESTAMP_SAFE(t) (FORMAT_TIMESTAMP(t) ?: " --- ")
3820 void journal_file_print_header(JournalFile
*f
) {
3826 printf("File path: %s\n"
3830 "Sequential number ID: %s\n"
3832 "Compatible flags:%s%s%s%s\n"
3833 "Incompatible flags:%s%s%s%s%s%s\n"
3834 "Header size: %"PRIu64
"\n"
3835 "Arena size: %"PRIu64
"\n"
3836 "Data hash table size: %"PRIu64
"\n"
3837 "Field hash table size: %"PRIu64
"\n"
3838 "Rotate suggested: %s\n"
3839 "Head sequential number: %"PRIu64
" (%"PRIx64
")\n"
3840 "Tail sequential number: %"PRIu64
" (%"PRIx64
")\n"
3841 "Head realtime timestamp: %s (%"PRIx64
")\n"
3842 "Tail realtime timestamp: %s (%"PRIx64
")\n"
3843 "Tail monotonic timestamp: %s (%"PRIx64
")\n"
3844 "Objects: %"PRIu64
"\n"
3845 "Entry objects: %"PRIu64
"\n",
3847 SD_ID128_TO_STRING(f
->header
->file_id
),
3848 SD_ID128_TO_STRING(f
->header
->machine_id
),
3849 SD_ID128_TO_STRING(f
->header
->tail_entry_boot_id
),
3850 SD_ID128_TO_STRING(f
->header
->seqnum_id
),
3851 f
->header
->state
== STATE_OFFLINE
? "OFFLINE" :
3852 f
->header
->state
== STATE_ONLINE
? "ONLINE" :
3853 f
->header
->state
== STATE_ARCHIVED
? "ARCHIVED" : "UNKNOWN",
3854 JOURNAL_HEADER_SEALED(f
->header
) ? " SEALED" : "",
3855 JOURNAL_HEADER_SEALED_CONTINUOUS(f
->header
) ? " SEALED_CONTINUOUS" : "",
3856 JOURNAL_HEADER_TAIL_ENTRY_BOOT_ID(f
->header
) ? " TAIL_ENTRY_BOOT_ID" : "",
3857 (le32toh(f
->header
->compatible_flags
) & ~HEADER_COMPATIBLE_ANY
) ? " ???" : "",
3858 JOURNAL_HEADER_COMPRESSED_XZ(f
->header
) ? " COMPRESSED-XZ" : "",
3859 JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
) ? " COMPRESSED-LZ4" : "",
3860 JOURNAL_HEADER_COMPRESSED_ZSTD(f
->header
) ? " COMPRESSED-ZSTD" : "",
3861 JOURNAL_HEADER_KEYED_HASH(f
->header
) ? " KEYED-HASH" : "",
3862 JOURNAL_HEADER_COMPACT(f
->header
) ? " COMPACT" : "",
3863 (le32toh(f
->header
->incompatible_flags
) & ~HEADER_INCOMPATIBLE_ANY
) ? " ???" : "",
3864 le64toh(f
->header
->header_size
),
3865 le64toh(f
->header
->arena_size
),
3866 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3867 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
),
3868 yes_no(journal_file_rotate_suggested(f
, 0, LOG_DEBUG
)),
3869 le64toh(f
->header
->head_entry_seqnum
), le64toh(f
->header
->head_entry_seqnum
),
3870 le64toh(f
->header
->tail_entry_seqnum
), le64toh(f
->header
->tail_entry_seqnum
),
3871 FORMAT_TIMESTAMP_SAFE(le64toh(f
->header
->head_entry_realtime
)), le64toh(f
->header
->head_entry_realtime
),
3872 FORMAT_TIMESTAMP_SAFE(le64toh(f
->header
->tail_entry_realtime
)), le64toh(f
->header
->tail_entry_realtime
),
3873 FORMAT_TIMESPAN(le64toh(f
->header
->tail_entry_monotonic
), USEC_PER_MSEC
), le64toh(f
->header
->tail_entry_monotonic
),
3874 le64toh(f
->header
->n_objects
),
3875 le64toh(f
->header
->n_entries
));
3877 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3878 printf("Data objects: %"PRIu64
"\n"
3879 "Data hash table fill: %.1f%%\n",
3880 le64toh(f
->header
->n_data
),
3881 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))));
3883 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3884 printf("Field objects: %"PRIu64
"\n"
3885 "Field hash table fill: %.1f%%\n",
3886 le64toh(f
->header
->n_fields
),
3887 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))));
3889 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
))
3890 printf("Tag objects: %"PRIu64
"\n",
3891 le64toh(f
->header
->n_tags
));
3892 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
3893 printf("Entry array objects: %"PRIu64
"\n",
3894 le64toh(f
->header
->n_entry_arrays
));
3896 if (JOURNAL_HEADER_CONTAINS(f
->header
, field_hash_chain_depth
))
3897 printf("Deepest field hash chain: %" PRIu64
"\n",
3898 f
->header
->field_hash_chain_depth
);
3900 if (JOURNAL_HEADER_CONTAINS(f
->header
, data_hash_chain_depth
))
3901 printf("Deepest data hash chain: %" PRIu64
"\n",
3902 f
->header
->data_hash_chain_depth
);
3904 if (fstat(f
->fd
, &st
) >= 0)
3905 printf("Disk usage: %s\n", FORMAT_BYTES((uint64_t) st
.st_blocks
* 512ULL));
3908 static int journal_file_warn_btrfs(JournalFile
*f
) {
3914 /* Before we write anything, check if the COW logic is turned
3915 * off on btrfs. Given our write pattern that is quite
3916 * unfriendly to COW file systems this should greatly improve
3917 * performance on COW file systems, such as btrfs, at the
3918 * expense of data integrity features (which shouldn't be too
3919 * bad, given that we do our own checksumming). */
3921 r
= fd_is_fs_type(f
->fd
, BTRFS_SUPER_MAGIC
);
3923 return log_ratelimit_warning_errno(r
, JOURNAL_LOG_RATELIMIT
, "Failed to determine if journal is on btrfs: %m");
3927 r
= read_attr_fd(f
->fd
, &attrs
);
3929 return log_ratelimit_warning_errno(r
, JOURNAL_LOG_RATELIMIT
, "Failed to read file attributes: %m");
3931 if (attrs
& FS_NOCOW_FL
) {
3932 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3936 log_ratelimit_notice(JOURNAL_LOG_RATELIMIT
,
3937 "Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3938 "This is likely to slow down journal access substantially, please consider turning "
3939 "off the copy-on-write file attribute on the journal directory, using chattr +C.",
3945 static void journal_default_metrics(JournalMetrics
*m
, int fd
, bool compact
) {
3947 uint64_t fs_size
= 0;
3952 if (fstatvfs(fd
, &ss
) >= 0)
3953 fs_size
= u64_multiply_safe(ss
.f_frsize
, ss
.f_blocks
);
3955 log_debug_errno(errno
, "Failed to determine disk size: %m");
3957 if (m
->max_use
== UINT64_MAX
) {
3960 m
->max_use
= CLAMP(PAGE_ALIGN_U64(fs_size
/ 10), /* 10% of file system size */
3961 MAX_USE_LOWER
, MAX_USE_UPPER
);
3963 m
->max_use
= MAX_USE_LOWER
;
3965 m
->max_use
= PAGE_ALIGN_U64(m
->max_use
);
3967 if (m
->max_use
!= 0 && m
->max_use
< JOURNAL_FILE_SIZE_MIN
*2)
3968 m
->max_use
= JOURNAL_FILE_SIZE_MIN
*2;
3971 if (m
->min_use
== UINT64_MAX
) {
3973 m
->min_use
= CLAMP(PAGE_ALIGN_U64(fs_size
/ 50), /* 2% of file system size */
3974 MIN_USE_LOW
, MIN_USE_HIGH
);
3976 m
->min_use
= MIN_USE_LOW
;
3979 if (m
->min_use
> m
->max_use
)
3980 m
->min_use
= m
->max_use
;
3982 if (m
->max_size
== UINT64_MAX
)
3983 m
->max_size
= MIN(PAGE_ALIGN_U64(m
->max_use
/ 8), /* 8 chunks */
3986 m
->max_size
= PAGE_ALIGN_U64(m
->max_size
);
3988 if (compact
&& m
->max_size
> JOURNAL_COMPACT_SIZE_MAX
)
3989 m
->max_size
= JOURNAL_COMPACT_SIZE_MAX
;
3991 if (m
->max_size
!= 0) {
3992 if (m
->max_size
< JOURNAL_FILE_SIZE_MIN
)
3993 m
->max_size
= JOURNAL_FILE_SIZE_MIN
;
3995 if (m
->max_use
!= 0 && m
->max_size
*2 > m
->max_use
)
3996 m
->max_use
= m
->max_size
*2;
3999 if (m
->min_size
== UINT64_MAX
)
4000 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
4002 m
->min_size
= CLAMP(PAGE_ALIGN_U64(m
->min_size
),
4003 JOURNAL_FILE_SIZE_MIN
,
4004 m
->max_size
?: UINT64_MAX
);
4006 if (m
->keep_free
== UINT64_MAX
) {
4008 m
->keep_free
= MIN(PAGE_ALIGN_U64(fs_size
/ 20), /* 5% of file system size */
4011 m
->keep_free
= DEFAULT_KEEP_FREE
;
4014 if (m
->n_max_files
== UINT64_MAX
)
4015 m
->n_max_files
= DEFAULT_N_MAX_FILES
;
4017 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64
,
4018 FORMAT_BYTES(m
->min_use
),
4019 FORMAT_BYTES(m
->max_use
),
4020 FORMAT_BYTES(m
->max_size
),
4021 FORMAT_BYTES(m
->min_size
),
4022 FORMAT_BYTES(m
->keep_free
),
4026 int journal_file_open(
4030 JournalFileFlags file_flags
,
4032 uint64_t compress_threshold_bytes
,
4033 JournalMetrics
*metrics
,
4034 MMapCache
*mmap_cache
,
4035 JournalFile
*template,
4036 JournalFile
**ret
) {
4038 bool newly_created
= false;
4043 assert(fd
>= 0 || fname
);
4044 assert(file_flags
>= 0);
4045 assert(file_flags
<= _JOURNAL_FILE_FLAGS_MAX
);
4049 if (!IN_SET((open_flags
& O_ACCMODE
), O_RDONLY
, O_RDWR
))
4052 if ((open_flags
& O_ACCMODE
) == O_RDONLY
&& FLAGS_SET(open_flags
, O_CREAT
))
4055 if (fname
&& (open_flags
& O_CREAT
) && !endswith(fname
, ".journal"))
4058 f
= new(JournalFile
, 1);
4062 *f
= (JournalFile
) {
4065 .open_flags
= open_flags
,
4066 .compress_threshold_bytes
= compress_threshold_bytes
== UINT64_MAX
?
4067 DEFAULT_COMPRESS_THRESHOLD
:
4068 MAX(MIN_COMPRESS_THRESHOLD
, compress_threshold_bytes
),
4069 .strict_order
= FLAGS_SET(file_flags
, JOURNAL_STRICT_ORDER
),
4070 .newest_boot_id_prioq_idx
= PRIOQ_IDX_NULL
,
4071 .last_direction
= _DIRECTION_INVALID
,
4075 f
->path
= strdup(fname
);
4083 /* If we don't know the path, fill in something explanatory and vaguely useful */
4084 if (asprintf(&f
->path
, "/proc/self/%i", fd
) < 0) {
4090 f
->chain_cache
= ordered_hashmap_new(&uint64_hash_ops
);
4091 if (!f
->chain_cache
) {
4097 /* We pass O_NONBLOCK here, so that in case somebody pointed us to some character device node or FIFO
4098 * or so, we likely fail quickly than block for long. For regular files O_NONBLOCK has no effect, hence
4099 * it doesn't hurt in that case. */
4101 f
->fd
= openat_report_new(AT_FDCWD
, f
->path
, f
->open_flags
|O_CLOEXEC
|O_NONBLOCK
, f
->mode
, &newly_created
);
4107 /* fds we opened here by us should also be closed by us. */
4110 r
= fd_nonblock(f
->fd
, false);
4114 if (!newly_created
) {
4115 r
= journal_file_fstat(f
);
4120 r
= journal_file_fstat(f
);
4124 /* If we just got the fd passed in, we don't really know if we created the file anew */
4125 newly_created
= f
->last_stat
.st_size
== 0 && journal_file_writable(f
);
4128 r
= mmap_cache_add_fd(mmap_cache
, f
->fd
, mmap_prot_from_open_flags(open_flags
), &f
->cache_fd
);
4132 if (newly_created
) {
4133 (void) journal_file_warn_btrfs(f
);
4135 /* Let's attach the creation time to the journal file, so that the vacuuming code knows the age of this
4136 * file even if the file might end up corrupted one day... Ideally we'd just use the creation time many
4137 * file systems maintain for each file, but the API to query this is very new, hence let's emulate this
4138 * via extended attributes. If extended attributes are not supported we'll just skip this, and rely
4139 * solely on mtime/atime/ctime of the file. */
4140 (void) fd_setcrtime(f
->fd
, 0);
4142 r
= journal_file_init_header(f
, file_flags
, template);
4146 r
= journal_file_fstat(f
);
4151 if (f
->last_stat
.st_size
< (off_t
) HEADER_SIZE_MIN
) {
4156 r
= mmap_cache_fd_get(f
->cache_fd
, MMAP_CACHE_CATEGORY_HEADER
, true, 0, PAGE_ALIGN(sizeof(Header
)), &f
->last_stat
, &h
);
4158 /* Some file systems (jffs2 or p9fs) don't support mmap() properly (or only read-only
4159 * mmap()), and return EINVAL in that case. Let's propagate that as a more recognizable error
4169 if (!newly_created
) {
4170 r
= journal_file_verify_header(f
);
4176 if (!newly_created
&& journal_file_writable(f
) && JOURNAL_HEADER_SEALED(f
->header
)) {
4177 r
= journal_file_fss_load(f
);
4183 if (journal_file_writable(f
)) {
4185 journal_default_metrics(metrics
, f
->fd
, JOURNAL_HEADER_COMPACT(f
->header
));
4186 f
->metrics
= *metrics
;
4187 } else if (template)
4188 f
->metrics
= template->metrics
;
4190 r
= journal_file_refresh_header(f
);
4196 r
= journal_file_hmac_setup(f
);
4201 if (newly_created
) {
4202 r
= journal_file_setup_field_hash_table(f
);
4206 r
= journal_file_setup_data_hash_table(f
);
4211 r
= journal_file_append_first_tag(f
);
4217 if (mmap_cache_fd_got_sigbus(f
->cache_fd
)) {
4222 if (template && template->post_change_timer
) {
4223 r
= journal_file_enable_post_change_timer(
4225 sd_event_source_get_event(template->post_change_timer
),
4226 template->post_change_timer_period
);
4232 /* The file is opened now successfully, thus we take possession of any passed in fd. */
4235 if (DEBUG_LOGGING
) {
4236 static int last_seal
= -1, last_keyed_hash
= -1;
4237 static Compression last_compression
= _COMPRESSION_INVALID
;
4238 static uint64_t last_bytes
= UINT64_MAX
;
4240 if (last_seal
!= JOURNAL_HEADER_SEALED(f
->header
) ||
4241 last_keyed_hash
!= JOURNAL_HEADER_KEYED_HASH(f
->header
) ||
4242 last_compression
!= JOURNAL_FILE_COMPRESSION(f
) ||
4243 last_bytes
!= f
->compress_threshold_bytes
) {
4245 log_debug("Journal effective settings seal=%s keyed_hash=%s compress=%s compress_threshold_bytes=%s",
4246 yes_no(JOURNAL_HEADER_SEALED(f
->header
)), yes_no(JOURNAL_HEADER_KEYED_HASH(f
->header
)),
4247 compression_to_string(JOURNAL_FILE_COMPRESSION(f
)), FORMAT_BYTES(f
->compress_threshold_bytes
));
4248 last_seal
= JOURNAL_HEADER_SEALED(f
->header
);
4249 last_keyed_hash
= JOURNAL_HEADER_KEYED_HASH(f
->header
);
4250 last_compression
= JOURNAL_FILE_COMPRESSION(f
);
4251 last_bytes
= f
->compress_threshold_bytes
;
4259 if (f
->cache_fd
&& mmap_cache_fd_got_sigbus(f
->cache_fd
))
4262 (void) journal_file_close(f
);
4264 if (newly_created
&& fd
< 0)
4265 (void) unlink(fname
);
4270 int journal_file_parse_uid_from_filename(const char *path
, uid_t
*ret_uid
) {
4271 _cleanup_free_
char *buf
= NULL
, *p
= NULL
;
4272 const char *a
, *b
, *at
;
4275 /* This helper returns -EREMOTE when the filename doesn't match user online/offline journal
4276 * pattern. Hence it currently doesn't parse archived or disposed user journals. */
4281 r
= path_extract_filename(path
, &p
);
4284 if (r
== O_DIRECTORY
)
4287 a
= startswith(p
, "user-");
4290 b
= endswith(p
, ".journal");
4294 at
= strchr(a
, '@');
4298 buf
= strndup(a
, b
-a
);
4302 return parse_uid(buf
, ret_uid
);
4305 int journal_file_archive(JournalFile
*f
, char **ret_previous_path
) {
4306 _cleanup_free_
char *p
= NULL
;
4310 if (!journal_file_writable(f
))
4313 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
4314 * rotation, since we don't know the actual path, and couldn't rename the file hence. */
4315 if (path_startswith(f
->path
, "/proc/self/fd"))
4318 if (!endswith(f
->path
, ".journal"))
4321 if (asprintf(&p
, "%.*s@" SD_ID128_FORMAT_STR
"-%016"PRIx64
"-%016"PRIx64
".journal",
4322 (int) strlen(f
->path
) - 8, f
->path
,
4323 SD_ID128_FORMAT_VAL(f
->header
->seqnum_id
),
4324 le64toh(f
->header
->head_entry_seqnum
),
4325 le64toh(f
->header
->head_entry_realtime
)) < 0)
4328 /* Try to rename the file to the archived version. If the file already was deleted, we'll get ENOENT, let's
4329 * ignore that case. */
4330 if (rename(f
->path
, p
) < 0 && errno
!= ENOENT
)
4333 /* Sync the rename to disk */
4334 (void) fsync_directory_of_file(f
->fd
);
4336 if (ret_previous_path
)
4337 *ret_previous_path
= f
->path
;
4341 f
->path
= TAKE_PTR(p
);
4343 /* Set as archive so offlining commits w/state=STATE_ARCHIVED. Previously we would set old_file->header->state
4344 * to STATE_ARCHIVED directly here, but journal_file_set_offline() short-circuits when state != STATE_ONLINE,
4345 * which would result in the rotated journal never getting fsync() called before closing. Now we simply queue
4346 * the archive state by setting an archive bit, leaving the state as STATE_ONLINE so proper offlining
4353 int journal_file_dispose(int dir_fd
, const char *fname
) {
4354 _cleanup_free_
char *p
= NULL
;
4358 /* Renames a journal file to *.journal~, i.e. to mark it as corrupted or otherwise uncleanly shutdown. Note that
4359 * this is done without looking into the file or changing any of its contents. The idea is that this is called
4360 * whenever something is suspicious and we want to move the file away and make clear that it is not accessed
4361 * for writing anymore. */
4363 if (!endswith(fname
, ".journal"))
4366 if (asprintf(&p
, "%.*s@%016" PRIx64
"-%016" PRIx64
".journal~",
4367 (int) strlen(fname
) - 8, fname
,
4368 now(CLOCK_REALTIME
),
4372 if (renameat(dir_fd
, fname
, dir_fd
, p
) < 0)
4378 int journal_file_copy_entry(
4384 sd_id128_t
*seqnum_id
) {
4386 _cleanup_free_ EntryItem
*items_alloc
= NULL
;
4388 uint64_t n
, m
= 0, xor_hash
= 0;
4398 if (!journal_file_writable(to
))
4401 ts
= (dual_timestamp
) {
4402 .monotonic
= le64toh(o
->entry
.monotonic
),
4403 .realtime
= le64toh(o
->entry
.realtime
),
4405 boot_id
= o
->entry
.boot_id
;
4407 n
= journal_file_entry_n_items(from
, o
);
4411 if (n
< ALLOCA_MAX
/ sizeof(EntryItem
) / 2)
4412 items
= newa(EntryItem
, n
);
4414 items_alloc
= new(EntryItem
, n
);
4418 items
= items_alloc
;
4421 for (uint64_t i
= 0; i
< n
; i
++) {
4427 q
= journal_file_entry_item_object_offset(from
, o
, i
);
4428 r
= journal_file_data_payload(from
, NULL
, q
, NULL
, 0, 0, &data
, &l
);
4429 if (IN_SET(r
, -EADDRNOTAVAIL
, -EBADMSG
)) {
4430 log_debug_errno(r
, "Entry item %"PRIu64
" data object is bad, skipping over it: %m", i
);
4440 r
= journal_file_append_data(to
, data
, l
, &u
, &h
);
4444 if (JOURNAL_HEADER_KEYED_HASH(to
->header
))
4445 xor_hash
^= jenkins_hash64(data
, l
);
4447 xor_hash
^= le64toh(u
->data
.hash
);
4449 items
[m
++] = (EntryItem
) {
4451 .hash
= le64toh(u
->data
.hash
),
4458 r
= journal_file_append_entry_internal(
4462 &from
->header
->machine_id
,
4468 /* ret_object= */ NULL
,
4469 /* ret_offset= */ NULL
);
4471 if (mmap_cache_fd_got_sigbus(to
->cache_fd
))
4477 void journal_reset_metrics(JournalMetrics
*m
) {
4480 /* Set everything to "pick automatic values". */
4482 *m
= (JournalMetrics
) {
4483 .min_use
= UINT64_MAX
,
4484 .max_use
= UINT64_MAX
,
4485 .min_size
= UINT64_MAX
,
4486 .max_size
= UINT64_MAX
,
4487 .keep_free
= UINT64_MAX
,
4488 .n_max_files
= UINT64_MAX
,
4492 int journal_file_get_cutoff_realtime_usec(JournalFile
*f
, usec_t
*ret_from
, usec_t
*ret_to
) {
4495 assert(ret_from
|| ret_to
);
4498 if (f
->header
->head_entry_realtime
== 0)
4501 *ret_from
= le64toh(f
->header
->head_entry_realtime
);
4505 if (f
->header
->tail_entry_realtime
== 0)
4508 *ret_to
= le64toh(f
->header
->tail_entry_realtime
);
4514 int journal_file_get_cutoff_monotonic_usec(JournalFile
*f
, sd_id128_t boot_id
, usec_t
*ret_from
, usec_t
*ret_to
) {
4520 assert(ret_from
|| ret_to
);
4522 /* FIXME: fix return value assignment on success with 0. */
4524 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &p
);
4528 if (le64toh(o
->data
.n_entries
) <= 0)
4532 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, le64toh(o
->data
.entry_offset
), &o
);
4536 *ret_from
= le64toh(o
->entry
.monotonic
);
4540 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
4544 r
= journal_file_move_to_entry_for_data(f
, o
, DIRECTION_UP
, &o
, NULL
);
4548 *ret_to
= le64toh(o
->entry
.monotonic
);
4554 bool journal_file_rotate_suggested(JournalFile
*f
, usec_t max_file_usec
, int log_level
) {
4558 /* If we gained new header fields we gained new features,
4559 * hence suggest a rotation */
4560 if (le64toh(f
->header
->header_size
) < sizeof(Header
)) {
4561 log_ratelimit_full(log_level
, JOURNAL_LOG_RATELIMIT
,
4562 "%s uses an outdated header, suggesting rotation.", f
->path
);
4566 /* Let's check if the hash tables grew over a certain fill level (75%, borrowing this value from
4567 * Java's hash table implementation), and if so suggest a rotation. To calculate the fill level we
4568 * need the n_data field, which only exists in newer versions. */
4570 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
4571 if (le64toh(f
->header
->n_data
) * 4ULL > (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
4573 log_level
, JOURNAL_LOG_RATELIMIT
,
4574 "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.",
4576 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))),
4577 le64toh(f
->header
->n_data
),
4578 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
4579 (uint64_t) f
->last_stat
.st_size
,
4580 f
->last_stat
.st_size
/ le64toh(f
->header
->n_data
));
4584 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
4585 if (le64toh(f
->header
->n_fields
) * 4ULL > (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
4587 log_level
, JOURNAL_LOG_RATELIMIT
,
4588 "Field hash table of %s has a fill level at %.1f (%"PRIu64
" of %"PRIu64
" items), suggesting rotation.",
4590 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))),
4591 le64toh(f
->header
->n_fields
),
4592 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
));
4596 /* If there are too many hash collisions somebody is most likely playing games with us. Hence, if our
4597 * longest chain is longer than some threshold, let's suggest rotation. */
4598 if (JOURNAL_HEADER_CONTAINS(f
->header
, data_hash_chain_depth
) &&
4599 le64toh(f
->header
->data_hash_chain_depth
) > HASH_CHAIN_DEPTH_MAX
) {
4601 log_level
, JOURNAL_LOG_RATELIMIT
,
4602 "Data hash table of %s has deepest hash chain of length %" PRIu64
", suggesting rotation.",
4603 f
->path
, le64toh(f
->header
->data_hash_chain_depth
));
4607 if (JOURNAL_HEADER_CONTAINS(f
->header
, field_hash_chain_depth
) &&
4608 le64toh(f
->header
->field_hash_chain_depth
) > HASH_CHAIN_DEPTH_MAX
) {
4610 log_level
, JOURNAL_LOG_RATELIMIT
,
4611 "Field hash table of %s has deepest hash chain of length at %" PRIu64
", suggesting rotation.",
4612 f
->path
, le64toh(f
->header
->field_hash_chain_depth
));
4616 /* Are the data objects properly indexed by field objects? */
4617 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
4618 JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
4619 le64toh(f
->header
->n_data
) > 0 &&
4620 le64toh(f
->header
->n_fields
) == 0) {
4622 log_level
, JOURNAL_LOG_RATELIMIT
,
4623 "Data objects of %s are not indexed by field objects, suggesting rotation.",
4628 if (max_file_usec
> 0) {
4631 h
= le64toh(f
->header
->head_entry_realtime
);
4632 t
= now(CLOCK_REALTIME
);
4634 if (h
> 0 && t
> h
+ max_file_usec
) {
4636 log_level
, JOURNAL_LOG_RATELIMIT
,
4637 "Oldest entry in %s is older than the configured file retention duration (%s), suggesting rotation.",
4638 f
->path
, FORMAT_TIMESPAN(max_file_usec
, USEC_PER_SEC
));
4646 static const char * const journal_object_type_table
[] = {
4647 [OBJECT_UNUSED
] = "unused",
4648 [OBJECT_DATA
] = "data",
4649 [OBJECT_FIELD
] = "field",
4650 [OBJECT_ENTRY
] = "entry",
4651 [OBJECT_DATA_HASH_TABLE
] = "data hash table",
4652 [OBJECT_FIELD_HASH_TABLE
] = "field hash table",
4653 [OBJECT_ENTRY_ARRAY
] = "entry array",
4654 [OBJECT_TAG
] = "tag",
4657 DEFINE_STRING_TABLE_LOOKUP_TO_STRING(journal_object_type
, ObjectType
);