1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
6 #include <linux/magic.h>
10 #include <sys/statvfs.h>
16 #include "alloc-util.h"
17 #include "chattr-util.h"
21 #include "format-util.h"
23 #include "id128-util.h"
24 #include "journal-authenticate.h"
25 #include "journal-def.h"
26 #include "journal-file.h"
27 #include "journal-internal.h"
29 #include "memory-util.h"
30 #include "missing_threads.h"
31 #include "path-util.h"
33 #include "random-util.h"
35 #include "sort-util.h"
36 #include "stat-util.h"
37 #include "string-table.h"
38 #include "string-util.h"
40 #include "sync-util.h"
41 #include "user-util.h"
42 #include "xattr-util.h"
44 #define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
45 #define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
47 #define DEFAULT_COMPRESS_THRESHOLD (512ULL)
48 #define MIN_COMPRESS_THRESHOLD (8ULL)
50 #define U64_KB UINT64_C(1024)
51 #define U64_MB (UINT64_C(1024) * U64_KB)
52 #define U64_GB (UINT64_C(1024) * U64_MB)
54 /* This is the minimum journal file size */
55 #define JOURNAL_FILE_SIZE_MIN (512 * U64_KB) /* 512 KiB */
56 #define JOURNAL_COMPACT_SIZE_MAX ((uint64_t) UINT32_MAX) /* 4 GiB */
58 /* These are the lower and upper bounds if we deduce the max_use value from the file system size */
59 #define MAX_USE_LOWER (1 * U64_MB) /* 1 MiB */
60 #define MAX_USE_UPPER (4 * U64_GB) /* 4 GiB */
62 /* Those are the lower and upper bounds for the minimal use limit,
63 * i.e. how much we'll use even if keep_free suggests otherwise. */
64 #define MIN_USE_LOW (1 * U64_MB) /* 1 MiB */
65 #define MIN_USE_HIGH (16 * U64_MB) /* 16 MiB */
67 /* This is the upper bound if we deduce max_size from max_use */
68 #define MAX_SIZE_UPPER (128 * U64_MB) /* 128 MiB */
70 /* This is the upper bound if we deduce the keep_free value from the file system size */
71 #define KEEP_FREE_UPPER (4 * U64_GB) /* 4 GiB */
73 /* This is the keep_free value when we can't determine the system size */
74 #define DEFAULT_KEEP_FREE (1 * U64_MB) /* 1 MB */
76 /* This is the default maximum number of journal files to keep around. */
77 #define DEFAULT_N_MAX_FILES 100
79 /* n_data was the first entry we added after the initial file format design */
80 #define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
82 /* How many entries to keep in the entry array chain cache at max */
83 #define CHAIN_CACHE_MAX 20
85 /* How much to increase the journal file size at once each time we allocate something new. */
86 #define FILE_SIZE_INCREASE (8 * U64_MB) /* 8MB */
88 /* Reread fstat() of the file for detecting deletions at least this often */
89 #define LAST_STAT_REFRESH_USEC (5*USEC_PER_SEC)
91 /* The mmap context to use for the header we pick as one above the last defined typed */
92 #define CONTEXT_HEADER _OBJECT_TYPE_MAX
94 /* Longest hash chain to rotate after */
95 #define HASH_CHAIN_DEPTH_MAX 100
98 # pragma GCC diagnostic ignored "-Waddress-of-packed-member"
101 static int mmap_prot_from_open_flags(int flags
) {
102 switch (flags
& O_ACCMODE
) {
108 return PROT_READ
|PROT_WRITE
;
110 assert_not_reached();
114 int journal_file_tail_end_by_pread(JournalFile
*f
, uint64_t *ret_offset
) {
122 /* Same as journal_file_tail_end_by_mmap() below, but operates with pread() to avoid the mmap cache
123 * (and thus is thread safe) */
125 p
= le64toh(f
->header
->tail_object_offset
);
127 p
= le64toh(f
->header
->header_size
);
132 r
= journal_file_read_object_header(f
, OBJECT_UNUSED
, p
, &tail
);
136 sz
= le64toh(tail
.object
.size
);
137 if (sz
> UINT64_MAX
- sizeof(uint64_t) + 1)
141 if (p
> UINT64_MAX
- sz
)
152 int journal_file_tail_end_by_mmap(JournalFile
*f
, uint64_t *ret_offset
) {
160 /* Same as journal_file_tail_end_by_pread() above, but operates with the usual mmap logic */
162 p
= le64toh(f
->header
->tail_object_offset
);
164 p
= le64toh(f
->header
->header_size
);
169 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &tail
);
173 sz
= le64toh(READ_NOW(tail
->object
.size
));
174 if (sz
> UINT64_MAX
- sizeof(uint64_t) + 1)
178 if (p
> UINT64_MAX
- sz
)
189 int journal_file_set_offline_thread_join(JournalFile
*f
) {
194 if (f
->offline_state
== OFFLINE_JOINED
)
197 r
= pthread_join(f
->offline_thread
, NULL
);
201 f
->offline_state
= OFFLINE_JOINED
;
203 if (mmap_cache_fd_got_sigbus(f
->cache_fd
))
209 static int journal_file_set_online(JournalFile
*f
) {
214 if (!journal_file_writable(f
))
217 if (f
->fd
< 0 || !f
->header
)
221 switch (f
->offline_state
) {
223 /* No offline thread, no need to wait. */
227 case OFFLINE_SYNCING
: {
228 OfflineState tmp_state
= OFFLINE_SYNCING
;
229 if (!__atomic_compare_exchange_n(&f
->offline_state
, &tmp_state
, OFFLINE_CANCEL
,
230 false, __ATOMIC_SEQ_CST
, __ATOMIC_SEQ_CST
))
233 /* Canceled syncing prior to offlining, no need to wait. */
237 case OFFLINE_AGAIN_FROM_SYNCING
: {
238 OfflineState tmp_state
= OFFLINE_AGAIN_FROM_SYNCING
;
239 if (!__atomic_compare_exchange_n(&f
->offline_state
, &tmp_state
, OFFLINE_CANCEL
,
240 false, __ATOMIC_SEQ_CST
, __ATOMIC_SEQ_CST
))
243 /* Canceled restart from syncing, no need to wait. */
247 case OFFLINE_AGAIN_FROM_OFFLINING
: {
248 OfflineState tmp_state
= OFFLINE_AGAIN_FROM_OFFLINING
;
249 if (!__atomic_compare_exchange_n(&f
->offline_state
, &tmp_state
, OFFLINE_CANCEL
,
250 false, __ATOMIC_SEQ_CST
, __ATOMIC_SEQ_CST
))
253 /* Canceled restart from offlining, must wait for offlining to complete however. */
258 r
= journal_file_set_offline_thread_join(f
);
268 if (mmap_cache_fd_got_sigbus(f
->cache_fd
))
271 switch (f
->header
->state
) {
276 f
->header
->state
= STATE_ONLINE
;
285 JournalFile
* journal_file_close(JournalFile
*f
) {
289 assert(f
->newest_boot_id_prioq_idx
== PRIOQ_IDX_NULL
);
292 mmap_cache_fd_free(f
->cache_fd
);
298 ordered_hashmap_free_free(f
->chain_cache
);
301 free(f
->compress_buffer
);
306 size_t sz
= PAGE_ALIGN(f
->fss_file_size
);
307 assert(sz
< SIZE_MAX
);
308 munmap(f
->fss_file
, sz
);
310 free(f
->fsprg_state
);
315 gcry_md_close(f
->hmac
);
321 static bool keyed_hash_requested(void) {
322 static thread_local
int cached
= -1;
326 r
= getenv_bool("SYSTEMD_JOURNAL_KEYED_HASH");
329 log_debug_errno(r
, "Failed to parse $SYSTEMD_JOURNAL_KEYED_HASH environment variable, ignoring: %m");
338 static bool compact_mode_requested(void) {
339 static thread_local
int cached
= -1;
343 r
= getenv_bool("SYSTEMD_JOURNAL_COMPACT");
346 log_debug_errno(r
, "Failed to parse $SYSTEMD_JOURNAL_COMPACT environment variable, ignoring: %m");
356 static Compression
getenv_compression(void) {
361 e
= getenv("SYSTEMD_JOURNAL_COMPRESS");
363 return DEFAULT_COMPRESSION
;
365 r
= parse_boolean(e
);
367 return r
? DEFAULT_COMPRESSION
: COMPRESSION_NONE
;
369 c
= compression_from_string(e
);
371 log_debug_errno(c
, "Failed to parse SYSTEMD_JOURNAL_COMPRESS value, ignoring: %s", e
);
372 return DEFAULT_COMPRESSION
;
375 if (!compression_supported(c
)) {
376 log_debug("Unsupported compression algorithm specified, ignoring: %s", e
);
377 return DEFAULT_COMPRESSION
;
384 static Compression
compression_requested(void) {
386 static thread_local Compression cached
= _COMPRESSION_INVALID
;
389 cached
= getenv_compression();
393 return COMPRESSION_NONE
;
397 static int journal_file_init_header(
399 JournalFileFlags file_flags
,
400 JournalFile
*template) {
409 /* Try to load the FSPRG state, and if we can't, then just don't do sealing */
410 seal
= FLAGS_SET(file_flags
, JOURNAL_SEAL
) && journal_file_fss_load(f
) >= 0;
414 .header_size
= htole64(ALIGN64(sizeof(h
))),
415 .incompatible_flags
= htole32(
416 FLAGS_SET(file_flags
, JOURNAL_COMPRESS
) * COMPRESSION_TO_HEADER_INCOMPATIBLE_FLAG(compression_requested()) |
417 keyed_hash_requested() * HEADER_INCOMPATIBLE_KEYED_HASH
|
418 compact_mode_requested() * HEADER_INCOMPATIBLE_COMPACT
),
419 .compatible_flags
= htole32(
420 (seal
* HEADER_COMPATIBLE_SEALED
) |
421 HEADER_COMPATIBLE_TAIL_ENTRY_BOOT_ID
),
424 assert_cc(sizeof(h
.signature
) == sizeof(HEADER_SIGNATURE
));
425 memcpy(h
.signature
, HEADER_SIGNATURE
, sizeof(HEADER_SIGNATURE
));
427 r
= sd_id128_randomize(&h
.file_id
);
431 r
= sd_id128_get_machine(&h
.machine_id
);
432 if (r
< 0 && !ERRNO_IS_MACHINE_ID_UNSET(r
))
433 return r
; /* If we have no valid machine ID (test environment?), let's simply leave the
434 * machine ID field all zeroes. */
437 h
.seqnum_id
= template->header
->seqnum_id
;
438 h
.tail_entry_seqnum
= template->header
->tail_entry_seqnum
;
440 h
.seqnum_id
= h
.file_id
;
442 k
= pwrite(f
->fd
, &h
, sizeof(h
), 0);
451 static int journal_file_refresh_header(JournalFile
*f
) {
457 /* We used to update the header's boot ID field here, but we don't do that anymore, as per
458 * HEADER_COMPATIBLE_TAIL_ENTRY_BOOT_ID */
460 r
= journal_file_set_online(f
);
462 /* Sync the online state to disk; likely just created a new file, also sync the directory this file
464 (void) fsync_full(f
->fd
);
469 static bool warn_wrong_flags(const JournalFile
*f
, bool compatible
) {
470 const uint32_t any
= compatible
? HEADER_COMPATIBLE_ANY
: HEADER_INCOMPATIBLE_ANY
,
471 supported
= compatible
? HEADER_COMPATIBLE_SUPPORTED
: HEADER_INCOMPATIBLE_SUPPORTED
;
472 const char *type
= compatible
? "compatible" : "incompatible";
478 flags
= le32toh(compatible
? f
->header
->compatible_flags
: f
->header
->incompatible_flags
);
480 if (flags
& ~supported
) {
482 log_debug("Journal file %s has unknown %s flags 0x%"PRIx32
,
483 f
->path
, type
, flags
& ~any
);
484 flags
= (flags
& any
) & ~supported
;
488 _cleanup_free_
char *t
= NULL
;
491 if (flags
& HEADER_COMPATIBLE_SEALED
)
492 strv
[n
++] = "sealed";
494 if (flags
& HEADER_INCOMPATIBLE_COMPRESSED_XZ
)
495 strv
[n
++] = "xz-compressed";
496 if (flags
& HEADER_INCOMPATIBLE_COMPRESSED_LZ4
)
497 strv
[n
++] = "lz4-compressed";
498 if (flags
& HEADER_INCOMPATIBLE_COMPRESSED_ZSTD
)
499 strv
[n
++] = "zstd-compressed";
500 if (flags
& HEADER_INCOMPATIBLE_KEYED_HASH
)
501 strv
[n
++] = "keyed-hash";
502 if (flags
& HEADER_INCOMPATIBLE_COMPACT
)
503 strv
[n
++] = "compact";
506 assert(n
< ELEMENTSOF(strv
));
508 t
= strv_join((char**) strv
, ", ");
509 log_debug("Journal file %s uses %s %s %s disabled at compilation time.",
510 f
->path
, type
, n
> 1 ? "flags" : "flag", strnull(t
));
518 static bool offset_is_valid(uint64_t offset
, uint64_t header_size
, uint64_t tail_object_offset
) {
521 if (!VALID64(offset
))
523 if (offset
< header_size
)
525 if (offset
> tail_object_offset
)
530 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
) {
531 if ((offset
== 0) != (size
== 0))
535 if (offset
<= offsetof(Object
, hash_table
.items
))
537 offset
-= offsetof(Object
, hash_table
.items
);
538 if (!offset_is_valid(offset
, header_size
, tail_object_offset
))
540 assert(offset
<= header_size
+ arena_size
);
541 if (size
> header_size
+ arena_size
- offset
)
546 static int journal_file_verify_header(JournalFile
*f
) {
547 uint64_t arena_size
, header_size
;
552 if (memcmp(f
->header
->signature
, HEADER_SIGNATURE
, 8))
555 /* In both read and write mode we refuse to open files with incompatible
556 * flags we don't know. */
557 if (warn_wrong_flags(f
, false))
558 return -EPROTONOSUPPORT
;
560 /* When open for writing we refuse to open files with compatible flags, too. */
561 if (journal_file_writable(f
) && warn_wrong_flags(f
, true))
562 return -EPROTONOSUPPORT
;
564 if (f
->header
->state
>= _STATE_MAX
)
567 header_size
= le64toh(READ_NOW(f
->header
->header_size
));
569 /* The first addition was n_data, so check that we are at least this large */
570 if (header_size
< HEADER_SIZE_MIN
)
573 /* When open for writing we refuse to open files with a mismatch of the header size, i.e. writing to
574 * files implementing older or new header structures. */
575 if (journal_file_writable(f
) && header_size
!= sizeof(Header
))
576 return -EPROTONOSUPPORT
;
578 /* Don't write to journal files without the new boot ID update behavior guarantee. */
579 if (journal_file_writable(f
) && !JOURNAL_HEADER_TAIL_ENTRY_BOOT_ID(f
->header
))
580 return -EPROTONOSUPPORT
;
582 if (JOURNAL_HEADER_SEALED(f
->header
) && !JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
585 arena_size
= le64toh(READ_NOW(f
->header
->arena_size
));
587 if (UINT64_MAX
- header_size
< arena_size
|| header_size
+ arena_size
> (uint64_t) f
->last_stat
.st_size
)
590 uint64_t tail_object_offset
= le64toh(f
->header
->tail_object_offset
);
591 if (!offset_is_valid(tail_object_offset
, header_size
, UINT64_MAX
))
593 if (header_size
+ arena_size
< tail_object_offset
)
595 if (header_size
+ arena_size
- tail_object_offset
< sizeof(ObjectHeader
))
598 if (!hash_table_is_valid(le64toh(f
->header
->data_hash_table_offset
),
599 le64toh(f
->header
->data_hash_table_size
),
600 header_size
, arena_size
, tail_object_offset
))
603 if (!hash_table_is_valid(le64toh(f
->header
->field_hash_table_offset
),
604 le64toh(f
->header
->field_hash_table_size
),
605 header_size
, arena_size
, tail_object_offset
))
608 uint64_t entry_array_offset
= le64toh(f
->header
->entry_array_offset
);
609 if (!offset_is_valid(entry_array_offset
, header_size
, tail_object_offset
))
612 if (JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_array_offset
)) {
613 uint32_t offset
= le32toh(f
->header
->tail_entry_array_offset
);
614 uint32_t n
= le32toh(f
->header
->tail_entry_array_n_entries
);
616 if (!offset_is_valid(offset
, header_size
, tail_object_offset
))
618 if (entry_array_offset
> offset
)
620 if (entry_array_offset
== 0 && offset
!= 0)
622 if ((offset
== 0) != (n
== 0))
624 assert(offset
<= header_size
+ arena_size
);
625 if ((uint64_t) n
* journal_file_entry_array_item_size(f
) > header_size
+ arena_size
- offset
)
629 if (JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_offset
)) {
630 uint64_t offset
= le64toh(f
->header
->tail_entry_offset
);
632 if (!offset_is_valid(offset
, header_size
, tail_object_offset
))
636 /* When there is an entry object, then these fields must be filled. */
637 if (sd_id128_is_null(f
->header
->tail_entry_boot_id
))
639 if (!VALID_REALTIME(le64toh(f
->header
->head_entry_realtime
)))
641 if (!VALID_REALTIME(le64toh(f
->header
->tail_entry_realtime
)))
643 if (!VALID_MONOTONIC(le64toh(f
->header
->tail_entry_realtime
)))
646 /* Otherwise, the fields must be zero. */
647 if (JOURNAL_HEADER_TAIL_ENTRY_BOOT_ID(f
->header
) &&
648 !sd_id128_is_null(f
->header
->tail_entry_boot_id
))
650 if (f
->header
->head_entry_realtime
!= 0)
652 if (f
->header
->tail_entry_realtime
!= 0)
654 if (f
->header
->tail_entry_realtime
!= 0)
659 /* Verify number of objects */
660 uint64_t n_objects
= le64toh(f
->header
->n_objects
);
661 if (n_objects
> arena_size
/ sizeof(ObjectHeader
))
664 uint64_t n_entries
= le64toh(f
->header
->n_entries
);
665 if (n_entries
> n_objects
)
668 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
669 le64toh(f
->header
->n_data
) > n_objects
)
672 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
673 le64toh(f
->header
->n_fields
) > n_objects
)
676 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
) &&
677 le64toh(f
->header
->n_tags
) > n_objects
)
680 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
) &&
681 le64toh(f
->header
->n_entry_arrays
) > n_objects
)
684 if (JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_array_n_entries
) &&
685 le32toh(f
->header
->tail_entry_array_n_entries
) > n_entries
)
688 if (journal_file_writable(f
)) {
689 sd_id128_t machine_id
;
693 r
= sd_id128_get_machine(&machine_id
);
694 if (ERRNO_IS_NEG_MACHINE_ID_UNSET(r
)) /* Gracefully handle the machine ID not being initialized yet */
695 machine_id
= SD_ID128_NULL
;
699 if (!sd_id128_equal(machine_id
, f
->header
->machine_id
))
700 return log_debug_errno(SYNTHETIC_ERRNO(EHOSTDOWN
),
701 "Trying to open journal file from different host for writing, refusing.");
703 state
= f
->header
->state
;
705 if (state
== STATE_ARCHIVED
)
706 return -ESHUTDOWN
; /* Already archived */
707 if (state
== STATE_ONLINE
)
708 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
),
709 "Journal file %s is already online. Assuming unclean closing.",
711 if (state
!= STATE_OFFLINE
)
712 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
),
713 "Journal file %s has unknown state %i.",
716 if (f
->header
->field_hash_table_size
== 0 || f
->header
->data_hash_table_size
== 0)
723 int journal_file_fstat(JournalFile
*f
) {
729 if (fstat(f
->fd
, &f
->last_stat
) < 0)
732 f
->last_stat_usec
= now(CLOCK_MONOTONIC
);
734 /* Refuse dealing with files that aren't regular */
735 r
= stat_verify_regular(&f
->last_stat
);
739 /* Refuse appending to files that are already deleted */
740 if (f
->last_stat
.st_nlink
<= 0)
746 static int journal_file_allocate(JournalFile
*f
, uint64_t offset
, uint64_t size
) {
747 uint64_t old_size
, new_size
, old_header_size
, old_arena_size
;
753 /* We assume that this file is not sparse, and we know that for sure, since we always call
754 * posix_fallocate() ourselves */
756 if (size
> PAGE_ALIGN_DOWN_U64(UINT64_MAX
) - offset
)
759 if (mmap_cache_fd_got_sigbus(f
->cache_fd
))
762 old_header_size
= le64toh(READ_NOW(f
->header
->header_size
));
763 old_arena_size
= le64toh(READ_NOW(f
->header
->arena_size
));
764 if (old_arena_size
> PAGE_ALIGN_DOWN_U64(UINT64_MAX
) - old_header_size
)
767 old_size
= old_header_size
+ old_arena_size
;
769 new_size
= MAX(PAGE_ALIGN_U64(offset
+ size
), old_header_size
);
771 if (new_size
<= old_size
) {
773 /* We already pre-allocated enough space, but before
774 * we write to it, let's check with fstat() if the
775 * file got deleted, in order make sure we don't throw
776 * away the data immediately. Don't check fstat() for
777 * all writes though, but only once ever 10s. */
779 if (f
->last_stat_usec
+ LAST_STAT_REFRESH_USEC
> now(CLOCK_MONOTONIC
))
782 return journal_file_fstat(f
);
785 /* Allocate more space. */
787 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
790 /* Refuse to go over 4G in compact mode so offsets can be stored in 32-bit. */
791 if (JOURNAL_HEADER_COMPACT(f
->header
) && new_size
> UINT32_MAX
)
794 if (new_size
> f
->metrics
.min_size
&& f
->metrics
.keep_free
> 0) {
797 if (fstatvfs(f
->fd
, &svfs
) >= 0) {
800 available
= LESS_BY(u64_multiply_safe(svfs
.f_bfree
, svfs
.f_bsize
), f
->metrics
.keep_free
);
802 if (new_size
- old_size
> available
)
807 /* Increase by larger blocks at once */
808 new_size
= ROUND_UP(new_size
, FILE_SIZE_INCREASE
);
809 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
810 new_size
= f
->metrics
.max_size
;
812 /* Note that the glibc fallocate() fallback is very
813 inefficient, hence we try to minimize the allocation area
815 r
= posix_fallocate_loop(f
->fd
, old_size
, new_size
- old_size
);
819 f
->header
->arena_size
= htole64(new_size
- old_header_size
);
821 return journal_file_fstat(f
);
824 static unsigned type_to_context(ObjectType type
) {
825 /* One context for each type, plus one catch-all for the rest */
826 assert_cc(_OBJECT_TYPE_MAX
<= MMAP_CACHE_MAX_CONTEXTS
);
827 assert_cc(CONTEXT_HEADER
< MMAP_CACHE_MAX_CONTEXTS
);
828 return type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
? type
: 0;
831 static int journal_file_move_to(
844 /* This function may clear, overwrite, or alter previously cached entries with the same type. After
845 * this function has been called, all previously read objects with the same type may be invalidated,
846 * hence must be re-read before use. */
851 if (size
> UINT64_MAX
- offset
)
854 /* Avoid SIGBUS on invalid accesses */
855 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
) {
856 /* Hmm, out of range? Let's refresh the fstat() data
857 * first, before we trust that check. */
859 r
= journal_file_fstat(f
);
863 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
)
864 return -EADDRNOTAVAIL
;
867 return mmap_cache_fd_get(f
->cache_fd
, type_to_context(type
), keep_always
, offset
, size
, &f
->last_stat
, ret
);
870 static uint64_t minimum_header_size(JournalFile
*f
, Object
*o
) {
872 static const uint64_t table
[] = {
873 [OBJECT_DATA
] = sizeof(DataObject
),
874 [OBJECT_FIELD
] = sizeof(FieldObject
),
875 [OBJECT_ENTRY
] = sizeof(EntryObject
),
876 [OBJECT_DATA_HASH_TABLE
] = sizeof(HashTableObject
),
877 [OBJECT_FIELD_HASH_TABLE
] = sizeof(HashTableObject
),
878 [OBJECT_ENTRY_ARRAY
] = sizeof(EntryArrayObject
),
879 [OBJECT_TAG
] = sizeof(TagObject
),
885 if (o
->object
.type
== OBJECT_DATA
)
886 return journal_file_data_payload_offset(f
);
888 if (o
->object
.type
>= ELEMENTSOF(table
) || table
[o
->object
.type
] <= 0)
889 return sizeof(ObjectHeader
);
891 return table
[o
->object
.type
];
894 static int check_object_header(JournalFile
*f
, Object
*o
, ObjectType type
, uint64_t offset
) {
900 s
= le64toh(READ_NOW(o
->object
.size
));
902 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
903 "Attempt to move to uninitialized object: %" PRIu64
,
906 if (s
< sizeof(ObjectHeader
))
907 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
908 "Attempt to move to overly short object with size %"PRIu64
": %" PRIu64
,
911 if (o
->object
.type
<= OBJECT_UNUSED
|| o
->object
.type
>= _OBJECT_TYPE_MAX
)
912 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
913 "Attempt to move to object with invalid type (%u): %" PRIu64
,
914 o
->object
.type
, offset
);
916 if (type
> OBJECT_UNUSED
&& o
->object
.type
!= type
)
917 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
918 "Found %s object while expecting %s object: %" PRIu64
,
919 journal_object_type_to_string(o
->object
.type
),
920 journal_object_type_to_string(type
),
923 if (s
< minimum_header_size(f
, o
))
924 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
925 "Size of %s object (%"PRIu64
") is smaller than the minimum object size (%"PRIu64
"): %" PRIu64
,
926 journal_object_type_to_string(o
->object
.type
),
928 minimum_header_size(f
, o
),
934 /* Lightweight object checks. We want this to be fast, so that we won't
935 * slowdown every journal_file_move_to_object() call too much. */
936 static int check_object(JournalFile
*f
, Object
*o
, uint64_t offset
) {
940 switch (o
->object
.type
) {
943 if ((le64toh(o
->data
.entry_offset
) == 0) ^ (le64toh(o
->data
.n_entries
) == 0))
944 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
945 "Bad data n_entries: %" PRIu64
": %" PRIu64
,
946 le64toh(o
->data
.n_entries
),
949 if (le64toh(o
->object
.size
) <= journal_file_data_payload_offset(f
))
950 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
951 "Bad data size (<= %zu): %" PRIu64
": %" PRIu64
,
952 journal_file_data_payload_offset(f
),
953 le64toh(o
->object
.size
),
956 if (!VALID64(le64toh(o
->data
.next_hash_offset
)) ||
957 !VALID64(le64toh(o
->data
.next_field_offset
)) ||
958 !VALID64(le64toh(o
->data
.entry_offset
)) ||
959 !VALID64(le64toh(o
->data
.entry_array_offset
)))
960 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
961 "Invalid offset, next_hash_offset=" OFSfmt
", next_field_offset=" OFSfmt
", entry_offset=" OFSfmt
", entry_array_offset=" OFSfmt
": %" PRIu64
,
962 le64toh(o
->data
.next_hash_offset
),
963 le64toh(o
->data
.next_field_offset
),
964 le64toh(o
->data
.entry_offset
),
965 le64toh(o
->data
.entry_array_offset
),
971 if (le64toh(o
->object
.size
) <= offsetof(Object
, field
.payload
))
972 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
973 "Bad field size (<= %zu): %" PRIu64
": %" PRIu64
,
974 offsetof(Object
, field
.payload
),
975 le64toh(o
->object
.size
),
978 if (!VALID64(le64toh(o
->field
.next_hash_offset
)) ||
979 !VALID64(le64toh(o
->field
.head_data_offset
)))
980 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
981 "Invalid offset, next_hash_offset=" OFSfmt
", head_data_offset=" OFSfmt
": %" PRIu64
,
982 le64toh(o
->field
.next_hash_offset
),
983 le64toh(o
->field
.head_data_offset
),
990 sz
= le64toh(READ_NOW(o
->object
.size
));
991 if (sz
< offsetof(Object
, entry
.items
) ||
992 (sz
- offsetof(Object
, entry
.items
)) % journal_file_entry_item_size(f
) != 0)
993 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
994 "Bad entry size (<= %zu): %" PRIu64
": %" PRIu64
,
995 offsetof(Object
, entry
.items
),
999 if ((sz
- offsetof(Object
, entry
.items
)) / journal_file_entry_item_size(f
) <= 0)
1000 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1001 "Invalid number items in entry: %" PRIu64
": %" PRIu64
,
1002 (sz
- offsetof(Object
, entry
.items
)) / journal_file_entry_item_size(f
),
1005 if (le64toh(o
->entry
.seqnum
) <= 0)
1006 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1007 "Invalid entry seqnum: %" PRIx64
": %" PRIu64
,
1008 le64toh(o
->entry
.seqnum
),
1011 if (!VALID_REALTIME(le64toh(o
->entry
.realtime
)))
1012 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1013 "Invalid entry realtime timestamp: %" PRIu64
": %" PRIu64
,
1014 le64toh(o
->entry
.realtime
),
1017 if (!VALID_MONOTONIC(le64toh(o
->entry
.monotonic
)))
1018 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1019 "Invalid entry monotonic timestamp: %" PRIu64
": %" PRIu64
,
1020 le64toh(o
->entry
.monotonic
),
1023 if (sd_id128_is_null(o
->entry
.boot_id
))
1024 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1025 "Invalid object entry with an empty boot ID: %" PRIu64
,
1031 case OBJECT_DATA_HASH_TABLE
:
1032 case OBJECT_FIELD_HASH_TABLE
: {
1035 sz
= le64toh(READ_NOW(o
->object
.size
));
1036 if (sz
< offsetof(Object
, hash_table
.items
) ||
1037 (sz
- offsetof(Object
, hash_table
.items
)) % sizeof(HashItem
) != 0 ||
1038 (sz
- offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
) <= 0)
1039 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1040 "Invalid %s hash table size: %" PRIu64
": %" PRIu64
,
1041 journal_object_type_to_string(o
->object
.type
),
1048 case OBJECT_ENTRY_ARRAY
: {
1051 sz
= le64toh(READ_NOW(o
->object
.size
));
1052 if (sz
< offsetof(Object
, entry_array
.items
) ||
1053 (sz
- offsetof(Object
, entry_array
.items
)) % journal_file_entry_array_item_size(f
) != 0 ||
1054 (sz
- offsetof(Object
, entry_array
.items
)) / journal_file_entry_array_item_size(f
) <= 0)
1055 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1056 "Invalid object entry array size: %" PRIu64
": %" PRIu64
,
1059 /* Here, we request that the offset of each entry array object is in strictly increasing order. */
1060 next
= le64toh(o
->entry_array
.next_entry_array_offset
);
1061 if (!VALID64(next
) || (next
> 0 && next
<= offset
))
1062 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1063 "Invalid object entry array next_entry_array_offset: %" PRIu64
": %" PRIu64
,
1071 if (le64toh(o
->object
.size
) != sizeof(TagObject
))
1072 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1073 "Invalid object tag size: %" PRIu64
": %" PRIu64
,
1074 le64toh(o
->object
.size
),
1077 if (!VALID_EPOCH(le64toh(o
->tag
.epoch
)))
1078 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1079 "Invalid object tag epoch: %" PRIu64
": %" PRIu64
,
1080 le64toh(o
->tag
.epoch
), offset
);
1088 int journal_file_move_to_object(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
**ret
) {
1094 /* Even if this function fails, it may clear, overwrite, or alter previously cached entries with the
1095 * same type. After this function has been called, all previously read objects with the same type may
1096 * be invalidated, hence must be re-read before use. */
1098 /* Objects may only be located at multiple of 64 bit */
1099 if (!VALID64(offset
))
1100 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1101 "Attempt to move to %s object at non-64-bit boundary: %" PRIu64
,
1102 journal_object_type_to_string(type
),
1105 /* Object may not be located in the file header */
1106 if (offset
< le64toh(f
->header
->header_size
))
1107 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1108 "Attempt to move to %s object located in file header: %" PRIu64
,
1109 journal_object_type_to_string(type
),
1112 r
= journal_file_move_to(f
, type
, false, offset
, sizeof(ObjectHeader
), (void**) &o
);
1116 r
= check_object_header(f
, o
, type
, offset
);
1120 r
= journal_file_move_to(f
, type
, false, offset
, le64toh(READ_NOW(o
->object
.size
)), (void**) &o
);
1124 r
= check_object_header(f
, o
, type
, offset
);
1128 r
= check_object(f
, o
, offset
);
1138 int journal_file_read_object_header(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
*ret
) {
1145 /* Objects may only be located at multiple of 64 bit */
1146 if (!VALID64(offset
))
1147 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1148 "Attempt to read %s object at non-64-bit boundary: %" PRIu64
,
1149 journal_object_type_to_string(type
), offset
);
1151 /* Object may not be located in the file header */
1152 if (offset
< le64toh(f
->header
->header_size
))
1153 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1154 "Attempt to read %s object located in file header: %" PRIu64
,
1155 journal_object_type_to_string(type
), offset
);
1157 /* This will likely read too much data but it avoids having to call pread() twice. */
1158 n
= pread(f
->fd
, &o
, sizeof(o
), offset
);
1160 return log_debug_errno(errno
, "Failed to read journal %s object at offset: %" PRIu64
,
1161 journal_object_type_to_string(type
), offset
);
1163 if ((size_t) n
< sizeof(o
.object
))
1164 return log_debug_errno(SYNTHETIC_ERRNO(EIO
),
1165 "Failed to read short %s object at offset: %" PRIu64
,
1166 journal_object_type_to_string(type
), offset
);
1168 r
= check_object_header(f
, &o
, type
, offset
);
1172 if ((size_t) n
< minimum_header_size(f
, &o
))
1173 return log_debug_errno(SYNTHETIC_ERRNO(EIO
),
1174 "Short read while reading %s object: %" PRIu64
,
1175 journal_object_type_to_string(type
), offset
);
1177 r
= check_object(f
, &o
, offset
);
1187 static uint64_t inc_seqnum(uint64_t seqnum
) {
1188 if (seqnum
< UINT64_MAX
-1)
1191 return 1; /* skip over UINT64_MAX and 0 when we run out of seqnums and start again */
1194 static uint64_t journal_file_entry_seqnum(
1198 uint64_t next_seqnum
;
1203 /* Picks a new sequence number for the entry we are about to add and returns it. */
1205 next_seqnum
= inc_seqnum(le64toh(f
->header
->tail_entry_seqnum
));
1207 /* If an external seqnum counter was passed, we update both the local and the external one, and set
1208 * it to the maximum of both */
1210 *seqnum
= next_seqnum
= MAX(inc_seqnum(*seqnum
), next_seqnum
);
1212 f
->header
->tail_entry_seqnum
= htole64(next_seqnum
);
1214 if (f
->header
->head_entry_seqnum
== 0)
1215 f
->header
->head_entry_seqnum
= htole64(next_seqnum
);
1220 int journal_file_append_object(
1224 Object
**ret_object
,
1225 uint64_t *ret_offset
) {
1233 assert(type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
);
1234 assert(size
>= sizeof(ObjectHeader
));
1236 r
= journal_file_set_online(f
);
1240 r
= journal_file_tail_end_by_mmap(f
, &p
);
1244 r
= journal_file_allocate(f
, p
, size
);
1248 r
= journal_file_move_to(f
, type
, false, p
, size
, (void**) &o
);
1252 o
->object
= (ObjectHeader
) {
1254 .size
= htole64(size
),
1257 f
->header
->tail_object_offset
= htole64(p
);
1258 f
->header
->n_objects
= htole64(le64toh(f
->header
->n_objects
) + 1);
1269 static int journal_file_setup_data_hash_table(JournalFile
*f
) {
1277 /* We estimate that we need 1 hash table entry per 768 bytes
1278 of journal file and we want to make sure we never get
1279 beyond 75% fill level. Calculate the hash table size for
1280 the maximum file size based on these metrics. */
1282 s
= (f
->metrics
.max_size
* 4 / 768 / 3) * sizeof(HashItem
);
1283 if (s
< DEFAULT_DATA_HASH_TABLE_SIZE
)
1284 s
= DEFAULT_DATA_HASH_TABLE_SIZE
;
1286 log_debug("Reserving %"PRIu64
" entries in data hash table.", s
/ sizeof(HashItem
));
1288 r
= journal_file_append_object(f
,
1289 OBJECT_DATA_HASH_TABLE
,
1290 offsetof(Object
, hash_table
.items
) + s
,
1295 memzero(o
->hash_table
.items
, s
);
1297 f
->header
->data_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1298 f
->header
->data_hash_table_size
= htole64(s
);
1303 static int journal_file_setup_field_hash_table(JournalFile
*f
) {
1311 /* We use a fixed size hash table for the fields as this
1312 * number should grow very slowly only */
1314 s
= DEFAULT_FIELD_HASH_TABLE_SIZE
;
1315 log_debug("Reserving %"PRIu64
" entries in field hash table.", s
/ sizeof(HashItem
));
1317 r
= journal_file_append_object(f
,
1318 OBJECT_FIELD_HASH_TABLE
,
1319 offsetof(Object
, hash_table
.items
) + s
,
1324 memzero(o
->hash_table
.items
, s
);
1326 f
->header
->field_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1327 f
->header
->field_hash_table_size
= htole64(s
);
1332 int journal_file_map_data_hash_table(JournalFile
*f
) {
1340 if (f
->data_hash_table
)
1343 p
= le64toh(f
->header
->data_hash_table_offset
);
1344 s
= le64toh(f
->header
->data_hash_table_size
);
1346 r
= journal_file_move_to(f
,
1347 OBJECT_DATA_HASH_TABLE
,
1354 f
->data_hash_table
= t
;
1358 int journal_file_map_field_hash_table(JournalFile
*f
) {
1366 if (f
->field_hash_table
)
1369 p
= le64toh(f
->header
->field_hash_table_offset
);
1370 s
= le64toh(f
->header
->field_hash_table_size
);
1372 r
= journal_file_move_to(f
,
1373 OBJECT_FIELD_HASH_TABLE
,
1380 f
->field_hash_table
= t
;
1384 static int journal_file_link_field(
1395 assert(f
->field_hash_table
);
1399 if (o
->object
.type
!= OBJECT_FIELD
)
1402 m
= le64toh(READ_NOW(f
->header
->field_hash_table_size
)) / sizeof(HashItem
);
1406 /* This might alter the window we are looking at */
1407 o
->field
.next_hash_offset
= o
->field
.head_data_offset
= 0;
1410 p
= le64toh(f
->field_hash_table
[h
].tail_hash_offset
);
1412 f
->field_hash_table
[h
].head_hash_offset
= htole64(offset
);
1414 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1418 o
->field
.next_hash_offset
= htole64(offset
);
1421 f
->field_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1423 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
1424 f
->header
->n_fields
= htole64(le64toh(f
->header
->n_fields
) + 1);
1429 static int journal_file_link_data(
1440 assert(f
->data_hash_table
);
1444 if (o
->object
.type
!= OBJECT_DATA
)
1447 m
= le64toh(READ_NOW(f
->header
->data_hash_table_size
)) / sizeof(HashItem
);
1451 /* This might alter the window we are looking at */
1452 o
->data
.next_hash_offset
= o
->data
.next_field_offset
= 0;
1453 o
->data
.entry_offset
= o
->data
.entry_array_offset
= 0;
1454 o
->data
.n_entries
= 0;
1457 p
= le64toh(f
->data_hash_table
[h
].tail_hash_offset
);
1459 /* Only entry in the hash table is easy */
1460 f
->data_hash_table
[h
].head_hash_offset
= htole64(offset
);
1462 /* Move back to the previous data object, to patch in
1465 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1469 o
->data
.next_hash_offset
= htole64(offset
);
1472 f
->data_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1474 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
1475 f
->header
->n_data
= htole64(le64toh(f
->header
->n_data
) + 1);
1480 static int get_next_hash_offset(
1483 le64_t
*next_hash_offset
,
1485 le64_t
*header_max_depth
) {
1491 assert(next_hash_offset
);
1494 nextp
= le64toh(READ_NOW(*next_hash_offset
));
1496 if (nextp
<= *p
) /* Refuse going in loops */
1497 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1498 "Detected hash item loop in %s, refusing.", f
->path
);
1502 /* If the depth of this hash chain is larger than all others we have seen so far, record it */
1503 if (header_max_depth
&& journal_file_writable(f
))
1504 *header_max_depth
= htole64(MAX(*depth
, le64toh(*header_max_depth
)));
1511 int journal_file_find_field_object_with_hash(
1516 Object
**ret_object
,
1517 uint64_t *ret_offset
) {
1519 uint64_t p
, osize
, h
, m
, depth
= 0;
1527 /* If the field hash table is empty, we can't find anything */
1528 if (le64toh(f
->header
->field_hash_table_size
) <= 0)
1531 /* Map the field hash table, if it isn't mapped yet. */
1532 r
= journal_file_map_field_hash_table(f
);
1536 osize
= offsetof(Object
, field
.payload
) + size
;
1538 m
= le64toh(READ_NOW(f
->header
->field_hash_table_size
)) / sizeof(HashItem
);
1543 p
= le64toh(f
->field_hash_table
[h
].head_hash_offset
);
1547 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1551 if (le64toh(o
->field
.hash
) == hash
&&
1552 le64toh(o
->object
.size
) == osize
&&
1553 memcmp(o
->field
.payload
, field
, size
) == 0) {
1563 r
= get_next_hash_offset(
1566 &o
->field
.next_hash_offset
,
1568 JOURNAL_HEADER_CONTAINS(f
->header
, field_hash_chain_depth
) ? &f
->header
->field_hash_chain_depth
: NULL
);
1576 uint64_t journal_file_hash_data(
1583 assert(data
|| sz
== 0);
1585 /* We try to unify our codebase on siphash, hence new-styled journal files utilizing the keyed hash
1586 * function use siphash. Old journal files use the Jenkins hash. */
1588 if (JOURNAL_HEADER_KEYED_HASH(f
->header
))
1589 return siphash24(data
, sz
, f
->header
->file_id
.bytes
);
1591 return jenkins_hash64(data
, sz
);
1594 int journal_file_find_field_object(
1598 Object
**ret_object
,
1599 uint64_t *ret_offset
) {
1605 return journal_file_find_field_object_with_hash(
1608 journal_file_hash_data(f
, field
, size
),
1609 ret_object
, ret_offset
);
1612 int journal_file_find_data_object_with_hash(
1617 Object
**ret_object
,
1618 uint64_t *ret_offset
) {
1620 uint64_t p
, h
, m
, depth
= 0;
1625 assert(data
|| size
== 0);
1627 /* If there's no data hash table, then there's no entry. */
1628 if (le64toh(f
->header
->data_hash_table_size
) <= 0)
1631 /* Map the data hash table, if it isn't mapped yet. */
1632 r
= journal_file_map_data_hash_table(f
);
1636 m
= le64toh(READ_NOW(f
->header
->data_hash_table_size
)) / sizeof(HashItem
);
1641 p
= le64toh(f
->data_hash_table
[h
].head_hash_offset
);
1648 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1652 if (le64toh(o
->data
.hash
) != hash
)
1655 r
= journal_file_data_payload(f
, o
, p
, NULL
, 0, 0, &d
, &rsize
);
1658 assert(r
> 0); /* journal_file_data_payload() always returns > 0 if no field is provided. */
1660 if (memcmp_nn(data
, size
, d
, rsize
) == 0) {
1671 r
= get_next_hash_offset(
1674 &o
->data
.next_hash_offset
,
1676 JOURNAL_HEADER_CONTAINS(f
->header
, data_hash_chain_depth
) ? &f
->header
->data_hash_chain_depth
: NULL
);
1684 int journal_file_find_data_object(
1688 Object
**ret_object
,
1689 uint64_t *ret_offset
) {
1692 assert(data
|| size
== 0);
1694 return journal_file_find_data_object_with_hash(
1697 journal_file_hash_data(f
, data
, size
),
1698 ret_object
, ret_offset
);
1701 bool journal_field_valid(const char *p
, size_t l
, bool allow_protected
) {
1702 /* We kinda enforce POSIX syntax recommendations for
1703 environment variables here, but make a couple of additional
1706 http://pubs.opengroup.org/onlinepubs/000095399/basedefs/xbd_chap08.html */
1713 /* No empty field names */
1717 /* Don't allow names longer than 64 chars */
1721 /* Variables starting with an underscore are protected */
1722 if (!allow_protected
&& p
[0] == '_')
1725 /* Don't allow digits as first character */
1726 if (ascii_isdigit(p
[0]))
1729 /* Only allow A-Z0-9 and '_' */
1730 for (const char *a
= p
; a
< p
+ l
; a
++)
1731 if ((*a
< 'A' || *a
> 'Z') &&
1732 !ascii_isdigit(*a
) &&
1739 static int journal_file_append_field(
1743 Object
**ret_object
,
1744 uint64_t *ret_offset
) {
1755 if (!journal_field_valid(field
, size
, true))
1758 hash
= journal_file_hash_data(f
, field
, size
);
1760 r
= journal_file_find_field_object_with_hash(f
, field
, size
, hash
, ret_object
, ret_offset
);
1766 osize
= offsetof(Object
, field
.payload
) + size
;
1767 r
= journal_file_append_object(f
, OBJECT_FIELD
, osize
, &o
, &p
);
1771 o
->field
.hash
= htole64(hash
);
1772 memcpy(o
->field
.payload
, field
, size
);
1774 r
= journal_file_link_field(f
, o
, p
, hash
);
1778 /* The linking might have altered the window, so let's only pass the offset to hmac which will
1779 * move to the object again if needed. */
1782 r
= journal_file_hmac_put_object(f
, OBJECT_FIELD
, NULL
, p
);
1788 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, ret_object
);
1799 static int maybe_compress_payload(JournalFile
*f
, uint8_t *dst
, const uint8_t *src
, uint64_t size
, size_t *rsize
) {
1803 #if HAVE_COMPRESSION
1807 c
= JOURNAL_FILE_COMPRESSION(f
);
1808 if (c
== COMPRESSION_NONE
|| size
< f
->compress_threshold_bytes
)
1811 r
= compress_blob(c
, src
, size
, dst
, size
- 1, rsize
);
1813 return log_debug_errno(r
, "Failed to compress data object using %s, ignoring: %m", compression_to_string(c
));
1815 log_debug("Compressed data object %"PRIu64
" -> %zu using %s", size
, *rsize
, compression_to_string(c
));
1817 return 1; /* compressed */
1823 static int journal_file_append_data(
1827 Object
**ret_object
,
1828 uint64_t *ret_offset
) {
1830 uint64_t hash
, p
, osize
;
1838 if (!data
|| size
== 0)
1841 hash
= journal_file_hash_data(f
, data
, size
);
1843 r
= journal_file_find_data_object_with_hash(f
, data
, size
, hash
, ret_object
, ret_offset
);
1849 eq
= memchr(data
, '=', size
);
1853 osize
= journal_file_data_payload_offset(f
) + size
;
1854 r
= journal_file_append_object(f
, OBJECT_DATA
, osize
, &o
, &p
);
1858 o
->data
.hash
= htole64(hash
);
1860 r
= maybe_compress_payload(f
, journal_file_data_payload_field(f
, o
), data
, size
, &rsize
);
1862 /* We don't really care failures, let's continue without compression */
1863 memcpy_safe(journal_file_data_payload_field(f
, o
), data
, size
);
1865 Compression c
= JOURNAL_FILE_COMPRESSION(f
);
1867 assert(c
>= 0 && c
< _COMPRESSION_MAX
&& c
!= COMPRESSION_NONE
);
1869 o
->object
.size
= htole64(journal_file_data_payload_offset(f
) + rsize
);
1870 o
->object
.flags
|= COMPRESSION_TO_OBJECT_FLAG(c
);
1873 r
= journal_file_link_data(f
, o
, p
, hash
);
1877 /* The linking might have altered the window, so let's refresh our pointer. */
1878 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1883 r
= journal_file_hmac_put_object(f
, OBJECT_DATA
, o
, p
);
1888 /* Create field object ... */
1889 r
= journal_file_append_field(f
, data
, (uint8_t*) eq
- (uint8_t*) data
, &fo
, NULL
);
1893 /* ... and link it in. */
1894 o
->data
.next_field_offset
= fo
->field
.head_data_offset
;
1895 fo
->field
.head_data_offset
= le64toh(p
);
1906 static int maybe_decompress_payload(
1910 Compression compression
,
1912 size_t field_length
,
1913 size_t data_threshold
,
1919 /* We can't read objects larger than 4G on a 32-bit machine */
1920 if ((uint64_t) (size_t) size
!= size
)
1923 if (compression
!= COMPRESSION_NONE
) {
1924 #if HAVE_COMPRESSION
1929 r
= decompress_startswith(compression
, payload
, size
, &f
->compress_buffer
, field
,
1932 return log_debug_errno(r
,
1933 "Cannot decompress %s object of length %" PRIu64
": %m",
1934 compression_to_string(compression
),
1945 r
= decompress_blob(compression
, payload
, size
, &f
->compress_buffer
, &rsize
, 0);
1950 *ret_data
= f
->compress_buffer
;
1954 return -EPROTONOSUPPORT
;
1957 if (field
&& (size
< field_length
+ 1 || memcmp(payload
, field
, field_length
) != 0 || payload
[field_length
] != '=')) {
1966 *ret_data
= payload
;
1968 *ret_size
= (size_t) size
;
1974 int journal_file_data_payload(
1979 size_t field_length
,
1980 size_t data_threshold
,
1989 assert(!field
== (field_length
== 0)); /* These must be specified together. */
1992 r
= journal_file_move_to_object(f
, OBJECT_DATA
, offset
, &o
);
1997 size
= le64toh(READ_NOW(o
->object
.size
));
1998 if (size
< journal_file_data_payload_offset(f
))
2001 size
-= journal_file_data_payload_offset(f
);
2003 c
= COMPRESSION_FROM_OBJECT(o
);
2005 return -EPROTONOSUPPORT
;
2007 return maybe_decompress_payload(f
, journal_file_data_payload_field(f
, o
), size
, c
, field
,
2008 field_length
, data_threshold
, ret_data
, ret_size
);
2011 uint64_t journal_file_entry_n_items(JournalFile
*f
, Object
*o
) {
2017 if (o
->object
.type
!= OBJECT_ENTRY
)
2020 sz
= le64toh(READ_NOW(o
->object
.size
));
2021 if (sz
< offsetof(Object
, entry
.items
))
2024 return (sz
- offsetof(Object
, entry
.items
)) / journal_file_entry_item_size(f
);
2027 uint64_t journal_file_entry_array_n_items(JournalFile
*f
, Object
*o
) {
2033 if (o
->object
.type
!= OBJECT_ENTRY_ARRAY
)
2036 sz
= le64toh(READ_NOW(o
->object
.size
));
2037 if (sz
< offsetof(Object
, entry_array
.items
))
2040 return (sz
- offsetof(Object
, entry_array
.items
)) / journal_file_entry_array_item_size(f
);
2043 uint64_t journal_file_hash_table_n_items(Object
*o
) {
2048 if (!IN_SET(o
->object
.type
, OBJECT_DATA_HASH_TABLE
, OBJECT_FIELD_HASH_TABLE
))
2051 sz
= le64toh(READ_NOW(o
->object
.size
));
2052 if (sz
< offsetof(Object
, hash_table
.items
))
2055 return (sz
- offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
);
2058 static void write_entry_array_item(JournalFile
*f
, Object
*o
, uint64_t i
, uint64_t p
) {
2062 if (JOURNAL_HEADER_COMPACT(f
->header
)) {
2063 assert(p
<= UINT32_MAX
);
2064 o
->entry_array
.items
.compact
[i
] = htole32(p
);
2066 o
->entry_array
.items
.regular
[i
] = htole64(p
);
2069 static int link_entry_into_array(
2077 uint64_t n
= 0, ap
= 0, q
, i
, a
, hidx
;
2087 a
= tail
? le32toh(*tail
) : le64toh(*first
);
2088 hidx
= le64toh(READ_NOW(*idx
));
2089 i
= tidx
? le32toh(READ_NOW(*tidx
)) : hidx
;
2092 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2096 n
= journal_file_entry_array_n_items(f
, o
);
2098 write_entry_array_item(f
, o
, i
, p
);
2099 *idx
= htole64(hidx
+ 1);
2101 *tidx
= htole32(le32toh(*tidx
) + 1);
2107 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
2118 r
= journal_file_append_object(f
, OBJECT_ENTRY_ARRAY
,
2119 offsetof(Object
, entry_array
.items
) + n
* journal_file_entry_array_item_size(f
),
2125 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY_ARRAY
, o
, q
);
2130 write_entry_array_item(f
, o
, i
, p
);
2133 *first
= htole64(q
);
2135 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, ap
, &o
);
2139 o
->entry_array
.next_entry_array_offset
= htole64(q
);
2145 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
2146 f
->header
->n_entry_arrays
= htole64(le64toh(f
->header
->n_entry_arrays
) + 1);
2148 *idx
= htole64(hidx
+ 1);
2155 static int link_entry_into_array_plus_one(
2173 hidx
= le64toh(READ_NOW(*idx
));
2174 if (hidx
== UINT64_MAX
)
2177 *extra
= htole64(p
);
2181 i
= htole64(hidx
- 1);
2182 r
= link_entry_into_array(f
, first
, &i
, tail
, tidx
, p
);
2187 *idx
= htole64(hidx
+ 1);
2191 static int journal_file_link_entry_item(JournalFile
*f
, uint64_t offset
, uint64_t p
) {
2198 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
2202 return link_entry_into_array_plus_one(f
,
2203 &o
->data
.entry_offset
,
2204 &o
->data
.entry_array_offset
,
2206 JOURNAL_HEADER_COMPACT(f
->header
) ? &o
->data
.compact
.tail_entry_array_offset
: NULL
,
2207 JOURNAL_HEADER_COMPACT(f
->header
) ? &o
->data
.compact
.tail_entry_array_n_entries
: NULL
,
2211 static int journal_file_link_entry(
2215 const EntryItem items
[],
2225 if (o
->object
.type
!= OBJECT_ENTRY
)
2228 __atomic_thread_fence(__ATOMIC_SEQ_CST
);
2230 /* Link up the entry itself */
2231 r
= link_entry_into_array(f
,
2232 &f
->header
->entry_array_offset
,
2233 &f
->header
->n_entries
,
2234 JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_array_offset
) ? &f
->header
->tail_entry_array_offset
: NULL
,
2235 JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_array_n_entries
) ? &f
->header
->tail_entry_array_n_entries
: NULL
,
2240 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
2242 if (f
->header
->head_entry_realtime
== 0)
2243 f
->header
->head_entry_realtime
= o
->entry
.realtime
;
2245 f
->header
->tail_entry_realtime
= o
->entry
.realtime
;
2246 f
->header
->tail_entry_monotonic
= o
->entry
.monotonic
;
2247 if (JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_offset
))
2248 f
->header
->tail_entry_offset
= htole64(offset
);
2249 f
->newest_mtime
= 0; /* we have a new tail entry now, explicitly invalidate newest boot id/timestamp info */
2251 /* Link up the items */
2252 for (uint64_t i
= 0; i
< n_items
; i
++) {
2255 /* If we fail to link an entry item because we can't allocate a new entry array, don't fail
2256 * immediately but try to link the other entry items since it might still be possible to link
2257 * those if they don't require a new entry array to be allocated. */
2259 k
= journal_file_link_entry_item(f
, offset
, items
[i
].object_offset
);
2269 static void write_entry_item(JournalFile
*f
, Object
*o
, uint64_t i
, const EntryItem
*item
) {
2274 if (JOURNAL_HEADER_COMPACT(f
->header
)) {
2275 assert(item
->object_offset
<= UINT32_MAX
);
2276 o
->entry
.items
.compact
[i
].object_offset
= htole32(item
->object_offset
);
2278 o
->entry
.items
.regular
[i
].object_offset
= htole64(item
->object_offset
);
2279 o
->entry
.items
.regular
[i
].hash
= htole64(item
->hash
);
2283 static int journal_file_append_entry_internal(
2285 const dual_timestamp
*ts
,
2286 const sd_id128_t
*boot_id
,
2287 const sd_id128_t
*machine_id
,
2289 const EntryItem items
[],
2292 sd_id128_t
*seqnum_id
,
2293 Object
**ret_object
,
2294 uint64_t *ret_offset
) {
2305 assert(!sd_id128_is_null(*boot_id
));
2306 assert(items
|| n_items
== 0);
2308 if (f
->strict_order
) {
2309 /* If requested be stricter with ordering in this journal file, to make searching via
2310 * bisection fully deterministic. This is an optional feature, so that if desired journal
2311 * files can be written where the ordering is not strictly enforced (in which case bisection
2312 * will yield *a* result, but not the *only* result, when searching for points in
2313 * time). Strict ordering mode is enabled when journald originally writes the files, but
2314 * might not necessarily be if other tools (the remoting tools for example) write journal
2315 * files from combined sources.
2317 * Typically, if any of the errors generated here are seen journald will just rotate the
2318 * journal files and start anew. */
2320 if (ts
->realtime
< le64toh(f
->header
->tail_entry_realtime
))
2321 return log_debug_errno(SYNTHETIC_ERRNO(EREMCHG
),
2322 "Realtime timestamp %" PRIu64
" smaller than previous realtime "
2323 "timestamp %" PRIu64
", refusing entry.",
2324 ts
->realtime
, le64toh(f
->header
->tail_entry_realtime
));
2326 if (sd_id128_equal(*boot_id
, f
->header
->tail_entry_boot_id
) &&
2327 ts
->monotonic
< le64toh(f
->header
->tail_entry_monotonic
))
2328 return log_debug_errno(
2329 SYNTHETIC_ERRNO(ENOTNAM
),
2330 "Monotonic timestamp %" PRIu64
2331 " smaller than previous monotonic timestamp %" PRIu64
2332 " while having the same boot ID, refusing entry.",
2334 le64toh(f
->header
->tail_entry_monotonic
));
2338 /* Settle the passed in sequence number ID */
2340 if (sd_id128_is_null(*seqnum_id
))
2341 *seqnum_id
= f
->header
->seqnum_id
; /* Caller has none assigned, then copy the one from the file */
2342 else if (!sd_id128_equal(*seqnum_id
, f
->header
->seqnum_id
)) {
2343 /* Different seqnum IDs? We can't allow entries from multiple IDs end up in the same journal.*/
2344 if (le64toh(f
->header
->n_entries
) == 0)
2345 f
->header
->seqnum_id
= *seqnum_id
; /* Caller has one, and file so far has no entries, then copy the one from the caller */
2347 return log_debug_errno(SYNTHETIC_ERRNO(EILSEQ
),
2348 "Sequence number IDs don't match, refusing entry.");
2352 if (machine_id
&& sd_id128_is_null(f
->header
->machine_id
))
2353 /* Initialize machine ID when not set yet */
2354 f
->header
->machine_id
= *machine_id
;
2356 osize
= offsetof(Object
, entry
.items
) + (n_items
* journal_file_entry_item_size(f
));
2358 r
= journal_file_append_object(f
, OBJECT_ENTRY
, osize
, &o
, &np
);
2362 o
->entry
.seqnum
= htole64(journal_file_entry_seqnum(f
, seqnum
));
2363 o
->entry
.realtime
= htole64(ts
->realtime
);
2364 o
->entry
.monotonic
= htole64(ts
->monotonic
);
2365 o
->entry
.xor_hash
= htole64(xor_hash
);
2366 o
->entry
.boot_id
= f
->header
->tail_entry_boot_id
= *boot_id
;
2368 for (size_t i
= 0; i
< n_items
; i
++)
2369 write_entry_item(f
, o
, i
, &items
[i
]);
2372 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY
, o
, np
);
2377 r
= journal_file_link_entry(f
, o
, np
, items
, n_items
);
2390 void journal_file_post_change(JournalFile
*f
) {
2396 /* inotify() does not receive IN_MODIFY events from file
2397 * accesses done via mmap(). After each access we hence
2398 * trigger IN_MODIFY by truncating the journal file to its
2399 * current size which triggers IN_MODIFY. */
2401 __atomic_thread_fence(__ATOMIC_SEQ_CST
);
2403 if (ftruncate(f
->fd
, f
->last_stat
.st_size
) < 0)
2404 log_debug_errno(errno
, "Failed to truncate file to its own size: %m");
2407 static int post_change_thunk(sd_event_source
*timer
, uint64_t usec
, void *userdata
) {
2410 journal_file_post_change(userdata
);
2415 static void schedule_post_change(JournalFile
*f
) {
2420 assert(f
->post_change_timer
);
2422 assert_se(e
= sd_event_source_get_event(f
->post_change_timer
));
2424 /* If we are already going down, post the change immediately. */
2425 if (IN_SET(sd_event_get_state(e
), SD_EVENT_EXITING
, SD_EVENT_FINISHED
))
2428 r
= sd_event_source_get_enabled(f
->post_change_timer
, NULL
);
2430 log_debug_errno(r
, "Failed to get ftruncate timer state: %m");
2436 r
= sd_event_source_set_time_relative(f
->post_change_timer
, f
->post_change_timer_period
);
2438 log_debug_errno(r
, "Failed to set time for scheduling ftruncate: %m");
2442 r
= sd_event_source_set_enabled(f
->post_change_timer
, SD_EVENT_ONESHOT
);
2444 log_debug_errno(r
, "Failed to enable scheduled ftruncate: %m");
2451 /* On failure, let's simply post the change immediately. */
2452 journal_file_post_change(f
);
2455 /* Enable coalesced change posting in a timer on the provided sd_event instance */
2456 int journal_file_enable_post_change_timer(JournalFile
*f
, sd_event
*e
, usec_t t
) {
2457 _cleanup_(sd_event_source_unrefp
) sd_event_source
*timer
= NULL
;
2461 assert_return(!f
->post_change_timer
, -EINVAL
);
2465 r
= sd_event_add_time(e
, &timer
, CLOCK_MONOTONIC
, 0, 0, post_change_thunk
, f
);
2469 r
= sd_event_source_set_enabled(timer
, SD_EVENT_OFF
);
2473 f
->post_change_timer
= TAKE_PTR(timer
);
2474 f
->post_change_timer_period
= t
;
2479 static int entry_item_cmp(const EntryItem
*a
, const EntryItem
*b
) {
2480 return CMP(ASSERT_PTR(a
)->object_offset
, ASSERT_PTR(b
)->object_offset
);
2483 static size_t remove_duplicate_entry_items(EntryItem items
[], size_t n
) {
2486 assert(items
|| n
== 0);
2491 for (size_t i
= 1; i
< n
; i
++)
2492 if (items
[i
].object_offset
!= items
[j
- 1].object_offset
)
2493 items
[j
++] = items
[i
];
2498 int journal_file_append_entry(
2500 const dual_timestamp
*ts
,
2501 const sd_id128_t
*boot_id
,
2502 const struct iovec iovec
[],
2505 sd_id128_t
*seqnum_id
,
2506 Object
**ret_object
,
2507 uint64_t *ret_offset
) {
2509 _cleanup_free_ EntryItem
*items_alloc
= NULL
;
2511 uint64_t xor_hash
= 0;
2512 struct dual_timestamp _ts
;
2513 sd_id128_t _boot_id
, _machine_id
, *machine_id
;
2519 assert(n_iovec
> 0);
2522 if (!VALID_REALTIME(ts
->realtime
))
2523 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
2524 "Invalid realtime timestamp %" PRIu64
", refusing entry.",
2526 if (!VALID_MONOTONIC(ts
->monotonic
))
2527 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
2528 "Invalid monotomic timestamp %" PRIu64
", refusing entry.",
2531 dual_timestamp_get(&_ts
);
2536 if (sd_id128_is_null(*boot_id
))
2537 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
), "Empty boot ID, refusing entry.");
2539 r
= sd_id128_get_boot(&_boot_id
);
2543 boot_id
= &_boot_id
;
2546 r
= sd_id128_get_machine(&_machine_id
);
2547 if (ERRNO_IS_NEG_MACHINE_ID_UNSET(r
))
2548 /* Gracefully handle the machine ID not being initialized yet */
2553 machine_id
= &_machine_id
;
2556 r
= journal_file_maybe_append_tag(f
, ts
->realtime
);
2561 if (n_iovec
< ALLOCA_MAX
/ sizeof(EntryItem
) / 2)
2562 items
= newa(EntryItem
, n_iovec
);
2564 items_alloc
= new(EntryItem
, n_iovec
);
2568 items
= items_alloc
;
2571 for (size_t i
= 0; i
< n_iovec
; i
++) {
2575 r
= journal_file_append_data(f
, iovec
[i
].iov_base
, iovec
[i
].iov_len
, &o
, &p
);
2579 /* When calculating the XOR hash field, we need to take special care if the "keyed-hash"
2580 * journal file flag is on. We use the XOR hash field to quickly determine the identity of a
2581 * specific record, and give records with otherwise identical position (i.e. match in seqno,
2582 * timestamp, …) a stable ordering. But for that we can't have it that the hash of the
2583 * objects in each file is different since they are keyed. Hence let's calculate the Jenkins
2584 * hash here for that. This also has the benefit that cursors for old and new journal files
2585 * are completely identical (they include the XOR hash after all). For classic Jenkins-hash
2586 * files things are easier, we can just take the value from the stored record directly. */
2588 if (JOURNAL_HEADER_KEYED_HASH(f
->header
))
2589 xor_hash
^= jenkins_hash64(iovec
[i
].iov_base
, iovec
[i
].iov_len
);
2591 xor_hash
^= le64toh(o
->data
.hash
);
2593 items
[i
] = (EntryItem
) {
2595 .hash
= le64toh(o
->data
.hash
),
2599 /* Order by the position on disk, in order to improve seek
2600 * times for rotating media. */
2601 typesafe_qsort(items
, n_iovec
, entry_item_cmp
);
2602 n_iovec
= remove_duplicate_entry_items(items
, n_iovec
);
2604 r
= journal_file_append_entry_internal(
2617 /* If the memory mapping triggered a SIGBUS then we return an
2618 * IO error and ignore the error code passed down to us, since
2619 * it is very likely just an effect of a nullified replacement
2622 if (mmap_cache_fd_got_sigbus(f
->cache_fd
))
2625 if (f
->post_change_timer
)
2626 schedule_post_change(f
);
2628 journal_file_post_change(f
);
2633 typedef struct ChainCacheItem
{
2634 uint64_t first
; /* The offset of the entry array object at the beginning of the chain,
2635 * i.e., le64toh(f->header->entry_array_offset), or le64toh(o->data.entry_offset). */
2636 uint64_t array
; /* The offset of the cached entry array object. */
2637 uint64_t begin
; /* The offset of the first item in the cached array. */
2638 uint64_t total
; /* The total number of items in all arrays before the cached one in the chain. */
2639 uint64_t last_index
; /* The last index we looked at in the cached array, to optimize locality when bisecting. */
2642 static void chain_cache_put(
2649 uint64_t last_index
) {
2654 /* If the chain item to cache for this chain is the
2655 * first one it's not worth caching anything */
2659 if (ordered_hashmap_size(h
) >= CHAIN_CACHE_MAX
) {
2660 ci
= ordered_hashmap_steal_first(h
);
2663 ci
= new(ChainCacheItem
, 1);
2670 if (ordered_hashmap_put(h
, &ci
->first
, ci
) < 0) {
2675 assert(ci
->first
== first
);
2680 ci
->last_index
= last_index
;
2683 static int bump_array_index(uint64_t *i
, direction_t direction
, uint64_t n
) {
2686 /* Increase or decrease the specified index, in the right direction. */
2688 if (direction
== DIRECTION_DOWN
) {
2703 static int bump_entry_array(
2705 Object
*o
, /* the current entry array object. */
2706 uint64_t offset
, /* the offset of the entry array object. */
2707 uint64_t first
, /* The offset of the first entry array object in the chain. */
2708 direction_t direction
,
2716 if (direction
== DIRECTION_DOWN
) {
2718 assert(o
->object
.type
== OBJECT_ENTRY_ARRAY
);
2720 *ret
= le64toh(o
->entry_array
.next_entry_array_offset
);
2723 /* Entry array chains are a singly linked list, so to find the previous array in the chain, we have
2724 * to start iterating from the top. */
2728 uint64_t p
= first
, q
= 0;
2729 while (p
> 0 && p
!= offset
) {
2730 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, p
, &o
);
2735 p
= le64toh(o
->entry_array
.next_entry_array_offset
);
2738 /* If we can't find the previous entry array in the entry array chain, we're likely dealing with a
2739 * corrupted journal file. */
2749 static int generic_array_get(
2751 uint64_t first
, /* The offset of the first entry array object in the chain. */
2752 uint64_t i
, /* The index of the target object counted from the beginning of the entry array chain. */
2753 direction_t direction
,
2754 Object
**ret_object
, /* The found object. */
2755 uint64_t *ret_offset
) { /* The offset of the found object. */
2757 uint64_t a
, t
= 0, k
;
2764 /* FIXME: fix return value assignment on success. */
2768 /* Try the chain cache first */
2769 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2770 if (ci
&& i
> ci
->total
) {
2777 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2778 if (IN_SET(r
, -EBADMSG
, -EADDRNOTAVAIL
)) {
2779 /* If there's corruption and we're going downwards, let's pretend we reached the
2780 * final entry in the entry array chain. */
2782 if (direction
== DIRECTION_DOWN
)
2785 /* If there's corruption and we're going upwards, move back to the previous entry
2786 * array and start iterating entries from there. */
2794 k
= journal_file_entry_array_n_items(f
, o
);
2801 /* The index is larger than the number of elements in the array. Let's move to the next array. */
2804 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
2807 /* If we've found the right location, now look for the first non-corrupt entry object (in the right
2811 if (i
== UINT64_MAX
) {
2812 r
= bump_entry_array(f
, o
, a
, first
, direction
, &a
);
2816 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2820 k
= journal_file_entry_array_n_items(f
, o
);
2824 if (direction
== DIRECTION_DOWN
)
2827 /* We moved to the previous array. The total must be decreased. */
2829 return -EBADMSG
; /* chain cache is broken ? */
2839 p
= journal_file_entry_array_item(f
, o
, i
);
2841 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, ret_object
);
2843 /* Let's cache this item for the next invocation */
2844 chain_cache_put(f
->chain_cache
, ci
, first
, a
, journal_file_entry_array_item(f
, o
, 0), t
, i
);
2851 if (!IN_SET(r
, -EADDRNOTAVAIL
, -EBADMSG
))
2854 /* OK, so this entry is borked. Most likely some entry didn't get synced to
2855 * disk properly, let's see if the next one might work for us instead. */
2856 log_debug_errno(r
, "Entry item %" PRIu64
" is bad, skipping over it.", i
);
2858 } while (bump_array_index(&i
, direction
, k
) > 0);
2860 /* All entries tried in the above do-while loop are broken. Let's move to the next (or previous) array. */
2862 if (direction
== DIRECTION_DOWN
)
2863 /* We are going to the next array, the total must be incremented. */
2878 static int generic_array_bisect_one(
2880 uint64_t a
, /* offset of entry array object. */
2881 uint64_t i
, /* index of the entry item we will test. */
2883 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2884 direction_t direction
,
2887 uint64_t *ret_offset
) {
2894 assert(test_object
);
2898 assert(i
<= *right
);
2900 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2904 p
= journal_file_entry_array_item(f
, array
, i
);
2908 r
= test_object(f
, p
, needle
);
2909 if (IN_SET(r
, -EBADMSG
, -EADDRNOTAVAIL
)) {
2910 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short.");
2912 return -ENOANO
; /* recognizable error */
2917 if (r
== TEST_FOUND
)
2918 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2920 if (r
== TEST_RIGHT
)
2931 static int generic_array_bisect(
2936 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2937 direction_t direction
,
2938 Object
**ret_object
,
2939 uint64_t *ret_offset
,
2940 uint64_t *ret_idx
) {
2942 /* Given an entry array chain, this function finds the object "closest" to the given needle in the
2943 * chain, taking into account the provided direction. A function can be provided to determine how
2944 * an object is matched against the given needle.
2946 * Given a journal file, the offset of an object and the needle, the test_object() function should
2947 * return TEST_LEFT if the needle is located earlier in the entry array chain, TEST_LEFT if the
2948 * needle is located later in the entry array chain and TEST_FOUND if the object matches the needle.
2949 * If test_object() returns TEST_FOUND for a specific object, that object's information will be used
2950 * to populate the return values of this function. If test_object() never returns TEST_FOUND, the
2951 * return values are populated with the details of one of the objects closest to the needle. If the
2952 * direction is DIRECTION_UP, the earlier object is used. Otherwise, the later object is used.
2955 uint64_t a
, p
, t
= 0, i
= 0, last_p
= 0, last_index
= UINT64_MAX
;
2956 bool subtract_one
= false;
2962 assert(test_object
);
2964 /* Start with the first array in the chain */
2967 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2968 if (ci
&& n
> ci
->total
&& ci
->begin
!= 0) {
2969 /* Ah, we have iterated this bisection array chain previously! Let's see if we can skip ahead
2970 * in the chain, as far as the last time. But we can't jump backwards in the chain, so let's
2971 * check that first. */
2973 r
= test_object(f
, ci
->begin
, needle
);
2977 if (r
== TEST_LEFT
) {
2978 /* OK, what we are looking for is right of the begin of this EntryArray, so let's
2979 * jump straight to previously cached array in the chain */
2984 last_index
= ci
->last_index
;
2989 uint64_t left
= 0, right
, k
, lp
;
2991 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2995 k
= journal_file_entry_array_n_items(f
, array
);
3001 r
= generic_array_bisect_one(f
, a
, right
, needle
, test_object
, direction
, &left
, &right
, &lp
);
3009 if (r
== TEST_RIGHT
) {
3010 /* If we cached the last index we looked at, let's try to not to jump too wildly
3011 * around and see if we can limit the range to look at early to the immediate
3012 * neighbors of the last index we looked at. */
3014 if (last_index
> 0 && last_index
- 1 < right
) {
3015 r
= generic_array_bisect_one(f
, a
, last_index
- 1, needle
, test_object
, direction
, &left
, &right
, NULL
);
3016 if (r
< 0 && r
!= -ENOANO
)
3020 if (last_index
< right
) {
3021 r
= generic_array_bisect_one(f
, a
, last_index
+ 1, needle
, test_object
, direction
, &left
, &right
, NULL
);
3022 if (r
< 0 && r
!= -ENOANO
)
3027 if (left
== right
) {
3028 if (direction
== DIRECTION_UP
)
3029 subtract_one
= true;
3035 assert(left
< right
);
3036 i
= (left
+ right
) / 2;
3038 r
= generic_array_bisect_one(f
, a
, i
, needle
, test_object
, direction
, &left
, &right
, NULL
);
3039 if (r
< 0 && r
!= -ENOANO
)
3045 if (direction
== DIRECTION_UP
) {
3047 subtract_one
= true;
3058 last_index
= UINT64_MAX
;
3059 a
= le64toh(array
->entry_array
.next_entry_array_offset
);
3065 if (subtract_one
&& t
== 0 && i
== 0)
3068 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
3072 p
= journal_file_entry_array_item(f
, array
, 0);
3076 /* Let's cache this item for the next invocation */
3077 chain_cache_put(f
->chain_cache
, ci
, first
, a
, p
, t
, subtract_one
? (i
> 0 ? i
-1 : UINT64_MAX
) : i
);
3079 if (subtract_one
&& i
== 0)
3081 else if (subtract_one
)
3082 p
= journal_file_entry_array_item(f
, array
, i
- 1);
3084 p
= journal_file_entry_array_item(f
, array
, i
);
3087 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, ret_object
);
3096 *ret_idx
= t
+ i
+ (subtract_one
? -1 : 0);
3101 static int generic_array_bisect_plus_one(
3107 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
3108 direction_t direction
,
3109 Object
**ret_object
,
3110 uint64_t *ret_offset
) {
3115 assert(test_object
);
3120 /* This bisects the array in object 'first', but first checks an extra. */
3121 r
= test_object(f
, extra
, needle
);
3125 if (direction
== DIRECTION_DOWN
) {
3126 /* If we are going downwards, then we need to return the first object that passes the test.
3127 * When there is no object that passes the test, we need to return the first object that
3128 * test_object() returns TEST_RIGHT for. */
3130 TEST_FOUND
, /* The 'extra' object passes the test. Hence, this is the first
3131 * object that passes the test. */
3132 TEST_RIGHT
)) /* The 'extra' object is the first object that test_object() returns
3133 * TEST_RIGHT for, and no object exists even in the chained arrays
3134 * that passes the test. */
3135 goto use_extra
; /* The 'extra' object is exactly the one we are looking for. It is
3136 * not necessary to bisect the chained arrays. */
3138 /* Otherwise, the 'extra' object is not the one we are looking for. Search in the arrays. */
3141 /* If we are going upwards, then we need to return the last object that passes the test.
3142 * When there is no object that passes the test, we need to return the the last object that
3143 * test_object() returns TEST_LEFT for. */
3144 if (r
== TEST_RIGHT
)
3145 return 0; /* Not only the 'extra' object, but also all objects in the chained arrays
3146 * will never get TEST_FOUND or TEST_LEFT. The object we are looking for
3147 * does not exist. */
3149 /* Even if the 'extra' object passes the test, there may be multiple objects in the arrays
3150 * that also pass the test. Hence, we need to bisect the arrays for finding the last matching
3154 r
= generic_array_bisect(f
, first
, n
-1, needle
, test_object
, direction
, ret_object
, ret_offset
, NULL
);
3156 return r
; /* When > 0, the found object is the first (or last, when DIRECTION_UP) object.
3157 * Hence, return the found object now. */
3159 /* No matching object found in the chained arrays.
3160 * DIRECTION_DOWN : the 'extra' object neither matches the condition. There is no matching object.
3161 * DIRECTION_UP : the 'extra' object matches the condition. So, return it. */
3162 if (direction
== DIRECTION_DOWN
)
3167 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, ret_object
);
3173 *ret_offset
= extra
;
3178 static int test_object_offset(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
3184 else if (p
< needle
)
3190 int journal_file_move_to_entry_by_offset(
3193 direction_t direction
,
3194 Object
**ret_object
,
3195 uint64_t *ret_offset
) {
3200 return generic_array_bisect(
3202 le64toh(f
->header
->entry_array_offset
),
3203 le64toh(f
->header
->n_entries
),
3207 ret_object
, ret_offset
, NULL
);
3210 static int test_object_seqnum(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
3218 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
3222 sq
= le64toh(READ_NOW(o
->entry
.seqnum
));
3225 else if (sq
< needle
)
3231 int journal_file_move_to_entry_by_seqnum(
3234 direction_t direction
,
3235 Object
**ret_object
,
3236 uint64_t *ret_offset
) {
3241 return generic_array_bisect(
3243 le64toh(f
->header
->entry_array_offset
),
3244 le64toh(f
->header
->n_entries
),
3248 ret_object
, ret_offset
, NULL
);
3251 static int test_object_realtime(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
3259 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
3263 rt
= le64toh(READ_NOW(o
->entry
.realtime
));
3266 else if (rt
< needle
)
3272 int journal_file_move_to_entry_by_realtime(
3275 direction_t direction
,
3276 Object
**ret_object
,
3277 uint64_t *ret_offset
) {
3282 return generic_array_bisect(
3284 le64toh(f
->header
->entry_array_offset
),
3285 le64toh(f
->header
->n_entries
),
3287 test_object_realtime
,
3289 ret_object
, ret_offset
, NULL
);
3292 static int test_object_monotonic(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
3300 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
3304 m
= le64toh(READ_NOW(o
->entry
.monotonic
));
3307 else if (m
< needle
)
3313 static int find_data_object_by_boot_id(
3316 Object
**ret_object
,
3317 uint64_t *ret_offset
) {
3319 char t
[STRLEN("_BOOT_ID=") + 32 + 1] = "_BOOT_ID=";
3323 sd_id128_to_string(boot_id
, t
+ 9);
3324 return journal_file_find_data_object(f
, t
, sizeof(t
) - 1, ret_object
, ret_offset
);
3327 int journal_file_move_to_entry_by_monotonic(
3331 direction_t direction
,
3332 Object
**ret_object
,
3333 uint64_t *ret_offset
) {
3340 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, NULL
);
3344 return generic_array_bisect_plus_one(
3346 le64toh(o
->data
.entry_offset
),
3347 le64toh(o
->data
.entry_array_offset
),
3348 le64toh(o
->data
.n_entries
),
3350 test_object_monotonic
,
3352 ret_object
, ret_offset
);
3355 void journal_file_reset_location(JournalFile
*f
) {
3358 f
->location_type
= LOCATION_HEAD
;
3359 f
->current_offset
= 0;
3360 f
->current_seqnum
= 0;
3361 f
->current_realtime
= 0;
3362 f
->current_monotonic
= 0;
3363 zero(f
->current_boot_id
);
3364 f
->current_xor_hash
= 0;
3366 /* Also reset the previous reading direction. Otherwise, next_beyond_location() may wrongly handle we
3367 * already hit EOF. See issue #29216. */
3368 f
->last_direction
= _DIRECTION_INVALID
;
3371 void journal_file_save_location(JournalFile
*f
, Object
*o
, uint64_t offset
) {
3375 f
->location_type
= LOCATION_SEEK
;
3376 f
->current_offset
= offset
;
3377 f
->current_seqnum
= le64toh(o
->entry
.seqnum
);
3378 f
->current_realtime
= le64toh(o
->entry
.realtime
);
3379 f
->current_monotonic
= le64toh(o
->entry
.monotonic
);
3380 f
->current_boot_id
= o
->entry
.boot_id
;
3381 f
->current_xor_hash
= le64toh(o
->entry
.xor_hash
);
3384 static bool check_properly_ordered(uint64_t new_offset
, uint64_t old_offset
, direction_t direction
) {
3386 /* Consider it an error if any of the two offsets is uninitialized */
3387 if (old_offset
== 0 || new_offset
== 0)
3390 /* If we go down, the new offset must be larger than the old one. */
3391 return direction
== DIRECTION_DOWN
?
3392 new_offset
> old_offset
:
3393 new_offset
< old_offset
;
3396 int journal_file_next_entry(
3399 direction_t direction
,
3400 Object
**ret_object
,
3401 uint64_t *ret_offset
) {
3410 /* FIXME: fix return value assignment. */
3412 n
= le64toh(READ_NOW(f
->header
->n_entries
));
3416 /* When the input offset 'p' is zero, return the first (or last on DIRECTION_UP) entry. */
3418 return generic_array_get(f
,
3419 le64toh(f
->header
->entry_array_offset
),
3420 direction
== DIRECTION_DOWN
? 0 : n
- 1,
3422 ret_object
, ret_offset
);
3424 /* Otherwise, first find the nearest entry object. */
3425 r
= generic_array_bisect(f
,
3426 le64toh(f
->header
->entry_array_offset
),
3427 le64toh(f
->header
->n_entries
),
3431 ret_object
? &o
: NULL
, &q
, &i
);
3435 assert(direction
== DIRECTION_DOWN
? p
<= q
: q
<= p
);
3437 /* If the input offset 'p' points to an entry object, generic_array_bisect() should provides
3438 * the same offset, and the index needs to be shifted. Otherwise, use the found object as is,
3439 * as it is the nearest entry object from the input offset 'p'. */
3444 r
= bump_array_index(&i
, direction
, n
);
3448 /* And jump to it */
3449 r
= generic_array_get(f
, le64toh(f
->header
->entry_array_offset
), i
, direction
, ret_object
? &o
: NULL
, &q
);
3453 /* Ensure our array is properly ordered. */
3454 if (!check_properly_ordered(q
, p
, direction
))
3455 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
3456 "%s: entry array not properly ordered at entry index %" PRIu64
,
3467 int journal_file_move_to_entry_for_data(
3470 direction_t direction
,
3471 Object
**ret_object
,
3472 uint64_t *ret_offset
) {
3474 uint64_t extra
, first
, n
;
3479 assert(d
->object
.type
== OBJECT_DATA
);
3480 assert(IN_SET(direction
, DIRECTION_DOWN
, DIRECTION_UP
));
3482 /* FIXME: fix return value assignment. */
3484 /* This returns the first (when the direction is down, otherwise the last) entry linked to the
3485 * specified data object. */
3487 n
= le64toh(d
->data
.n_entries
);
3490 n
--; /* n_entries is the number of entries linked to the data object, including the 'extra' entry. */
3492 extra
= le64toh(d
->data
.entry_offset
);
3493 first
= le64toh(d
->data
.entry_array_offset
);
3495 if (direction
== DIRECTION_DOWN
&& extra
> 0) {
3496 /* When we are going downwards, first try to read the extra entry. */
3497 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, ret_object
);
3500 if (!IN_SET(r
, -EADDRNOTAVAIL
, -EBADMSG
))
3505 /* DIRECTION_DOWN : The extra entry is broken, falling back to the entries in the array.
3506 * DIRECTION_UP : Try to find a valid entry in the array from the tail. */
3507 r
= generic_array_get(f
,
3509 direction
== DIRECTION_DOWN
? 0 : n
- 1,
3511 ret_object
, ret_offset
);
3512 if (!IN_SET(r
, 0, -EADDRNOTAVAIL
, -EBADMSG
))
3513 return r
; /* found or critical error. */
3516 if (direction
== DIRECTION_UP
&& extra
> 0) {
3517 /* No valid entry exists in the chained array, falling back to the extra entry. */
3518 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, ret_object
);
3527 *ret_offset
= extra
;
3532 int journal_file_move_to_entry_by_offset_for_data(
3536 direction_t direction
,
3537 Object
**ret
, uint64_t *ret_offset
) {
3541 assert(d
->object
.type
== OBJECT_DATA
);
3543 return generic_array_bisect_plus_one(
3545 le64toh(d
->data
.entry_offset
),
3546 le64toh(d
->data
.entry_array_offset
),
3547 le64toh(d
->data
.n_entries
),
3554 int journal_file_move_to_entry_by_monotonic_for_data(
3559 direction_t direction
,
3560 Object
**ret_object
,
3561 uint64_t *ret_offset
) {
3563 uint64_t z
, entry_offset
, entry_array_offset
, n_entries
;
3569 assert(d
->object
.type
== OBJECT_DATA
);
3571 /* Save all the required data before the data object gets invalidated. */
3572 entry_offset
= le64toh(READ_NOW(d
->data
.entry_offset
));
3573 entry_array_offset
= le64toh(READ_NOW(d
->data
.entry_array_offset
));
3574 n_entries
= le64toh(READ_NOW(d
->data
.n_entries
));
3576 /* First, seek by time */
3577 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, NULL
);
3581 r
= generic_array_bisect_plus_one(f
,
3582 le64toh(o
->data
.entry_offset
),
3583 le64toh(o
->data
.entry_array_offset
),
3584 le64toh(o
->data
.n_entries
),
3586 test_object_monotonic
,
3592 /* And now, continue seeking until we find an entry that exists in both bisection arrays. */
3596 /* The journal entry found by the above bisect_plus_one() may not have the specified data,
3597 * that is, it may not be linked in the data object. So, we need to check that. */
3599 r
= generic_array_bisect_plus_one(f
,
3606 ret_object
? &entry
: NULL
, &p
);
3610 break; /* The journal entry has the specified data. Yay! */
3612 /* If the entry does not have the data, then move to the next (or previous, depends on the
3613 * 'direction') entry linked to the data object. But, the next entry may be in another boot.
3614 * So, we need to check that the entry has the matching boot ID. */
3616 r
= generic_array_bisect_plus_one(f
,
3617 le64toh(o
->data
.entry_offset
),
3618 le64toh(o
->data
.entry_array_offset
),
3619 le64toh(o
->data
.n_entries
),
3623 ret_object
? &entry
: NULL
, &z
);
3627 break; /* The journal entry has the specified boot ID. Yay! */
3629 /* If not, let's try to the next entry... */
3633 *ret_object
= entry
;
3639 int journal_file_move_to_entry_by_seqnum_for_data(
3643 direction_t direction
,
3644 Object
**ret_object
,
3645 uint64_t *ret_offset
) {
3649 assert(d
->object
.type
== OBJECT_DATA
);
3651 return generic_array_bisect_plus_one(
3653 le64toh(d
->data
.entry_offset
),
3654 le64toh(d
->data
.entry_array_offset
),
3655 le64toh(d
->data
.n_entries
),
3659 ret_object
, ret_offset
);
3662 int journal_file_move_to_entry_by_realtime_for_data(
3666 direction_t direction
,
3667 Object
**ret
, uint64_t *ret_offset
) {
3671 assert(d
->object
.type
== OBJECT_DATA
);
3673 return generic_array_bisect_plus_one(
3675 le64toh(d
->data
.entry_offset
),
3676 le64toh(d
->data
.entry_array_offset
),
3677 le64toh(d
->data
.n_entries
),
3679 test_object_realtime
,
3684 void journal_file_dump(JournalFile
*f
) {
3692 journal_file_print_header(f
);
3694 p
= le64toh(READ_NOW(f
->header
->header_size
));
3699 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &o
);
3703 s
= journal_object_type_to_string(o
->object
.type
);
3705 switch (o
->object
.type
) {
3710 printf("Type: %s seqnum=%"PRIu64
" monotonic=%"PRIu64
" realtime=%"PRIu64
"\n",
3712 le64toh(o
->entry
.seqnum
),
3713 le64toh(o
->entry
.monotonic
),
3714 le64toh(o
->entry
.realtime
));
3720 printf("Type: %s seqnum=%"PRIu64
" epoch=%"PRIu64
"\n",
3722 le64toh(o
->tag
.seqnum
),
3723 le64toh(o
->tag
.epoch
));
3728 printf("Type: %s \n", s
);
3730 printf("Type: unknown (%i)", o
->object
.type
);
3735 c
= COMPRESSION_FROM_OBJECT(o
);
3736 if (c
> COMPRESSION_NONE
)
3737 printf("Flags: %s\n",
3738 compression_to_string(c
));
3740 if (p
== le64toh(f
->header
->tail_object_offset
))
3743 p
+= ALIGN64(le64toh(o
->object
.size
));
3748 log_error("File corrupt");
3751 /* Note: the lifetime of the compound literal is the immediately surrounding block. */
3752 #define FORMAT_TIMESTAMP_SAFE(t) (FORMAT_TIMESTAMP(t) ?: " --- ")
3754 void journal_file_print_header(JournalFile
*f
) {
3760 printf("File path: %s\n"
3764 "Sequential number ID: %s\n"
3766 "Compatible flags:%s%s%s\n"
3767 "Incompatible flags:%s%s%s%s%s%s\n"
3768 "Header size: %"PRIu64
"\n"
3769 "Arena size: %"PRIu64
"\n"
3770 "Data hash table size: %"PRIu64
"\n"
3771 "Field hash table size: %"PRIu64
"\n"
3772 "Rotate suggested: %s\n"
3773 "Head sequential number: %"PRIu64
" (%"PRIx64
")\n"
3774 "Tail sequential number: %"PRIu64
" (%"PRIx64
")\n"
3775 "Head realtime timestamp: %s (%"PRIx64
")\n"
3776 "Tail realtime timestamp: %s (%"PRIx64
")\n"
3777 "Tail monotonic timestamp: %s (%"PRIx64
")\n"
3778 "Objects: %"PRIu64
"\n"
3779 "Entry objects: %"PRIu64
"\n",
3781 SD_ID128_TO_STRING(f
->header
->file_id
),
3782 SD_ID128_TO_STRING(f
->header
->machine_id
),
3783 SD_ID128_TO_STRING(f
->header
->tail_entry_boot_id
),
3784 SD_ID128_TO_STRING(f
->header
->seqnum_id
),
3785 f
->header
->state
== STATE_OFFLINE
? "OFFLINE" :
3786 f
->header
->state
== STATE_ONLINE
? "ONLINE" :
3787 f
->header
->state
== STATE_ARCHIVED
? "ARCHIVED" : "UNKNOWN",
3788 JOURNAL_HEADER_SEALED(f
->header
) ? " SEALED" : "",
3789 JOURNAL_HEADER_TAIL_ENTRY_BOOT_ID(f
->header
) ? " TAIL_ENTRY_BOOT_ID" : "",
3790 (le32toh(f
->header
->compatible_flags
) & ~HEADER_COMPATIBLE_ANY
) ? " ???" : "",
3791 JOURNAL_HEADER_COMPRESSED_XZ(f
->header
) ? " COMPRESSED-XZ" : "",
3792 JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
) ? " COMPRESSED-LZ4" : "",
3793 JOURNAL_HEADER_COMPRESSED_ZSTD(f
->header
) ? " COMPRESSED-ZSTD" : "",
3794 JOURNAL_HEADER_KEYED_HASH(f
->header
) ? " KEYED-HASH" : "",
3795 JOURNAL_HEADER_COMPACT(f
->header
) ? " COMPACT" : "",
3796 (le32toh(f
->header
->incompatible_flags
) & ~HEADER_INCOMPATIBLE_ANY
) ? " ???" : "",
3797 le64toh(f
->header
->header_size
),
3798 le64toh(f
->header
->arena_size
),
3799 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3800 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
),
3801 yes_no(journal_file_rotate_suggested(f
, 0, LOG_DEBUG
)),
3802 le64toh(f
->header
->head_entry_seqnum
), le64toh(f
->header
->head_entry_seqnum
),
3803 le64toh(f
->header
->tail_entry_seqnum
), le64toh(f
->header
->tail_entry_seqnum
),
3804 FORMAT_TIMESTAMP_SAFE(le64toh(f
->header
->head_entry_realtime
)), le64toh(f
->header
->head_entry_realtime
),
3805 FORMAT_TIMESTAMP_SAFE(le64toh(f
->header
->tail_entry_realtime
)), le64toh(f
->header
->tail_entry_realtime
),
3806 FORMAT_TIMESPAN(le64toh(f
->header
->tail_entry_monotonic
), USEC_PER_MSEC
), le64toh(f
->header
->tail_entry_monotonic
),
3807 le64toh(f
->header
->n_objects
),
3808 le64toh(f
->header
->n_entries
));
3810 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3811 printf("Data objects: %"PRIu64
"\n"
3812 "Data hash table fill: %.1f%%\n",
3813 le64toh(f
->header
->n_data
),
3814 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))));
3816 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3817 printf("Field objects: %"PRIu64
"\n"
3818 "Field hash table fill: %.1f%%\n",
3819 le64toh(f
->header
->n_fields
),
3820 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))));
3822 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
))
3823 printf("Tag objects: %"PRIu64
"\n",
3824 le64toh(f
->header
->n_tags
));
3825 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
3826 printf("Entry array objects: %"PRIu64
"\n",
3827 le64toh(f
->header
->n_entry_arrays
));
3829 if (JOURNAL_HEADER_CONTAINS(f
->header
, field_hash_chain_depth
))
3830 printf("Deepest field hash chain: %" PRIu64
"\n",
3831 f
->header
->field_hash_chain_depth
);
3833 if (JOURNAL_HEADER_CONTAINS(f
->header
, data_hash_chain_depth
))
3834 printf("Deepest data hash chain: %" PRIu64
"\n",
3835 f
->header
->data_hash_chain_depth
);
3837 if (fstat(f
->fd
, &st
) >= 0)
3838 printf("Disk usage: %s\n", FORMAT_BYTES((uint64_t) st
.st_blocks
* 512ULL));
3841 static int journal_file_warn_btrfs(JournalFile
*f
) {
3847 /* Before we write anything, check if the COW logic is turned
3848 * off on btrfs. Given our write pattern that is quite
3849 * unfriendly to COW file systems this should greatly improve
3850 * performance on COW file systems, such as btrfs, at the
3851 * expense of data integrity features (which shouldn't be too
3852 * bad, given that we do our own checksumming). */
3854 r
= fd_is_fs_type(f
->fd
, BTRFS_SUPER_MAGIC
);
3856 return log_ratelimit_warning_errno(r
, JOURNAL_LOG_RATELIMIT
, "Failed to determine if journal is on btrfs: %m");
3860 r
= read_attr_fd(f
->fd
, &attrs
);
3862 return log_ratelimit_warning_errno(r
, JOURNAL_LOG_RATELIMIT
, "Failed to read file attributes: %m");
3864 if (attrs
& FS_NOCOW_FL
) {
3865 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3869 log_ratelimit_notice(JOURNAL_LOG_RATELIMIT
,
3870 "Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3871 "This is likely to slow down journal access substantially, please consider turning "
3872 "off the copy-on-write file attribute on the journal directory, using chattr +C.",
3878 static void journal_default_metrics(JournalMetrics
*m
, int fd
, bool compact
) {
3880 uint64_t fs_size
= 0;
3885 if (fstatvfs(fd
, &ss
) >= 0)
3886 fs_size
= u64_multiply_safe(ss
.f_frsize
, ss
.f_blocks
);
3888 log_debug_errno(errno
, "Failed to determine disk size: %m");
3890 if (m
->max_use
== UINT64_MAX
) {
3893 m
->max_use
= CLAMP(PAGE_ALIGN_U64(fs_size
/ 10), /* 10% of file system size */
3894 MAX_USE_LOWER
, MAX_USE_UPPER
);
3896 m
->max_use
= MAX_USE_LOWER
;
3898 m
->max_use
= PAGE_ALIGN_U64(m
->max_use
);
3900 if (m
->max_use
!= 0 && m
->max_use
< JOURNAL_FILE_SIZE_MIN
*2)
3901 m
->max_use
= JOURNAL_FILE_SIZE_MIN
*2;
3904 if (m
->min_use
== UINT64_MAX
) {
3906 m
->min_use
= CLAMP(PAGE_ALIGN_U64(fs_size
/ 50), /* 2% of file system size */
3907 MIN_USE_LOW
, MIN_USE_HIGH
);
3909 m
->min_use
= MIN_USE_LOW
;
3912 if (m
->min_use
> m
->max_use
)
3913 m
->min_use
= m
->max_use
;
3915 if (m
->max_size
== UINT64_MAX
)
3916 m
->max_size
= MIN(PAGE_ALIGN_U64(m
->max_use
/ 8), /* 8 chunks */
3919 m
->max_size
= PAGE_ALIGN_U64(m
->max_size
);
3921 if (compact
&& m
->max_size
> JOURNAL_COMPACT_SIZE_MAX
)
3922 m
->max_size
= JOURNAL_COMPACT_SIZE_MAX
;
3924 if (m
->max_size
!= 0) {
3925 if (m
->max_size
< JOURNAL_FILE_SIZE_MIN
)
3926 m
->max_size
= JOURNAL_FILE_SIZE_MIN
;
3928 if (m
->max_use
!= 0 && m
->max_size
*2 > m
->max_use
)
3929 m
->max_use
= m
->max_size
*2;
3932 if (m
->min_size
== UINT64_MAX
)
3933 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3935 m
->min_size
= CLAMP(PAGE_ALIGN_U64(m
->min_size
),
3936 JOURNAL_FILE_SIZE_MIN
,
3937 m
->max_size
?: UINT64_MAX
);
3939 if (m
->keep_free
== UINT64_MAX
) {
3941 m
->keep_free
= MIN(PAGE_ALIGN_U64(fs_size
/ 20), /* 5% of file system size */
3944 m
->keep_free
= DEFAULT_KEEP_FREE
;
3947 if (m
->n_max_files
== UINT64_MAX
)
3948 m
->n_max_files
= DEFAULT_N_MAX_FILES
;
3950 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64
,
3951 FORMAT_BYTES(m
->min_use
),
3952 FORMAT_BYTES(m
->max_use
),
3953 FORMAT_BYTES(m
->max_size
),
3954 FORMAT_BYTES(m
->min_size
),
3955 FORMAT_BYTES(m
->keep_free
),
3959 int journal_file_open(
3963 JournalFileFlags file_flags
,
3965 uint64_t compress_threshold_bytes
,
3966 JournalMetrics
*metrics
,
3967 MMapCache
*mmap_cache
,
3968 JournalFile
*template,
3969 JournalFile
**ret
) {
3971 bool newly_created
= false;
3976 assert(fd
>= 0 || fname
);
3977 assert(file_flags
>= 0);
3978 assert(file_flags
<= _JOURNAL_FILE_FLAGS_MAX
);
3982 if (!IN_SET((open_flags
& O_ACCMODE
), O_RDONLY
, O_RDWR
))
3985 if ((open_flags
& O_ACCMODE
) == O_RDONLY
&& FLAGS_SET(open_flags
, O_CREAT
))
3988 if (fname
&& (open_flags
& O_CREAT
) && !endswith(fname
, ".journal"))
3991 f
= new(JournalFile
, 1);
3995 *f
= (JournalFile
) {
3998 .open_flags
= open_flags
,
3999 .compress_threshold_bytes
= compress_threshold_bytes
== UINT64_MAX
?
4000 DEFAULT_COMPRESS_THRESHOLD
:
4001 MAX(MIN_COMPRESS_THRESHOLD
, compress_threshold_bytes
),
4002 .strict_order
= FLAGS_SET(file_flags
, JOURNAL_STRICT_ORDER
),
4003 .newest_boot_id_prioq_idx
= PRIOQ_IDX_NULL
,
4004 .last_direction
= _DIRECTION_INVALID
,
4008 f
->path
= strdup(fname
);
4016 /* If we don't know the path, fill in something explanatory and vaguely useful */
4017 if (asprintf(&f
->path
, "/proc/self/%i", fd
) < 0) {
4023 f
->chain_cache
= ordered_hashmap_new(&uint64_hash_ops
);
4024 if (!f
->chain_cache
) {
4030 /* We pass O_NONBLOCK here, so that in case somebody pointed us to some character device node or FIFO
4031 * or so, we likely fail quickly than block for long. For regular files O_NONBLOCK has no effect, hence
4032 * it doesn't hurt in that case. */
4034 f
->fd
= openat_report_new(AT_FDCWD
, f
->path
, f
->open_flags
|O_CLOEXEC
|O_NONBLOCK
, f
->mode
, &newly_created
);
4040 /* fds we opened here by us should also be closed by us. */
4043 r
= fd_nonblock(f
->fd
, false);
4047 if (!newly_created
) {
4048 r
= journal_file_fstat(f
);
4053 r
= journal_file_fstat(f
);
4057 /* If we just got the fd passed in, we don't really know if we created the file anew */
4058 newly_created
= f
->last_stat
.st_size
== 0 && journal_file_writable(f
);
4061 r
= mmap_cache_add_fd(mmap_cache
, f
->fd
, mmap_prot_from_open_flags(open_flags
), &f
->cache_fd
);
4065 if (newly_created
) {
4066 (void) journal_file_warn_btrfs(f
);
4068 /* Let's attach the creation time to the journal file, so that the vacuuming code knows the age of this
4069 * file even if the file might end up corrupted one day... Ideally we'd just use the creation time many
4070 * file systems maintain for each file, but the API to query this is very new, hence let's emulate this
4071 * via extended attributes. If extended attributes are not supported we'll just skip this, and rely
4072 * solely on mtime/atime/ctime of the file. */
4073 (void) fd_setcrtime(f
->fd
, 0);
4075 r
= journal_file_init_header(f
, file_flags
, template);
4079 r
= journal_file_fstat(f
);
4084 if (f
->last_stat
.st_size
< (off_t
) HEADER_SIZE_MIN
) {
4089 r
= mmap_cache_fd_get(f
->cache_fd
, CONTEXT_HEADER
, true, 0, PAGE_ALIGN(sizeof(Header
)), &f
->last_stat
, &h
);
4091 /* Some file systems (jffs2 or p9fs) don't support mmap() properly (or only read-only
4092 * mmap()), and return EINVAL in that case. Let's propagate that as a more recognizable error
4102 if (!newly_created
) {
4103 r
= journal_file_verify_header(f
);
4109 if (!newly_created
&& journal_file_writable(f
) && JOURNAL_HEADER_SEALED(f
->header
)) {
4110 r
= journal_file_fss_load(f
);
4116 if (journal_file_writable(f
)) {
4118 journal_default_metrics(metrics
, f
->fd
, JOURNAL_HEADER_COMPACT(f
->header
));
4119 f
->metrics
= *metrics
;
4120 } else if (template)
4121 f
->metrics
= template->metrics
;
4123 r
= journal_file_refresh_header(f
);
4129 r
= journal_file_hmac_setup(f
);
4134 if (newly_created
) {
4135 r
= journal_file_setup_field_hash_table(f
);
4139 r
= journal_file_setup_data_hash_table(f
);
4144 r
= journal_file_append_first_tag(f
);
4150 if (mmap_cache_fd_got_sigbus(f
->cache_fd
)) {
4155 if (template && template->post_change_timer
) {
4156 r
= journal_file_enable_post_change_timer(
4158 sd_event_source_get_event(template->post_change_timer
),
4159 template->post_change_timer_period
);
4165 /* The file is opened now successfully, thus we take possession of any passed in fd. */
4168 if (DEBUG_LOGGING
) {
4169 static int last_seal
= -1, last_keyed_hash
= -1;
4170 static Compression last_compression
= _COMPRESSION_INVALID
;
4171 static uint64_t last_bytes
= UINT64_MAX
;
4173 if (last_seal
!= JOURNAL_HEADER_SEALED(f
->header
) ||
4174 last_keyed_hash
!= JOURNAL_HEADER_KEYED_HASH(f
->header
) ||
4175 last_compression
!= JOURNAL_FILE_COMPRESSION(f
) ||
4176 last_bytes
!= f
->compress_threshold_bytes
) {
4178 log_debug("Journal effective settings seal=%s keyed_hash=%s compress=%s compress_threshold_bytes=%s",
4179 yes_no(JOURNAL_HEADER_SEALED(f
->header
)), yes_no(JOURNAL_HEADER_KEYED_HASH(f
->header
)),
4180 compression_to_string(JOURNAL_FILE_COMPRESSION(f
)), FORMAT_BYTES(f
->compress_threshold_bytes
));
4181 last_seal
= JOURNAL_HEADER_SEALED(f
->header
);
4182 last_keyed_hash
= JOURNAL_HEADER_KEYED_HASH(f
->header
);
4183 last_compression
= JOURNAL_FILE_COMPRESSION(f
);
4184 last_bytes
= f
->compress_threshold_bytes
;
4192 if (f
->cache_fd
&& mmap_cache_fd_got_sigbus(f
->cache_fd
))
4195 (void) journal_file_close(f
);
4197 if (newly_created
&& fd
< 0)
4198 (void) unlink(fname
);
4203 int journal_file_parse_uid_from_filename(const char *path
, uid_t
*ret_uid
) {
4204 _cleanup_free_
char *buf
= NULL
, *p
= NULL
;
4205 const char *a
, *b
, *at
;
4208 /* This helper returns -EREMOTE when the filename doesn't match user online/offline journal
4209 * pattern. Hence it currently doesn't parse archived or disposed user journals. */
4214 r
= path_extract_filename(path
, &p
);
4217 if (r
== O_DIRECTORY
)
4220 a
= startswith(p
, "user-");
4223 b
= endswith(p
, ".journal");
4227 at
= strchr(a
, '@');
4231 buf
= strndup(a
, b
-a
);
4235 return parse_uid(buf
, ret_uid
);
4238 int journal_file_archive(JournalFile
*f
, char **ret_previous_path
) {
4239 _cleanup_free_
char *p
= NULL
;
4243 if (!journal_file_writable(f
))
4246 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
4247 * rotation, since we don't know the actual path, and couldn't rename the file hence. */
4248 if (path_startswith(f
->path
, "/proc/self/fd"))
4251 if (!endswith(f
->path
, ".journal"))
4254 if (asprintf(&p
, "%.*s@" SD_ID128_FORMAT_STR
"-%016"PRIx64
"-%016"PRIx64
".journal",
4255 (int) strlen(f
->path
) - 8, f
->path
,
4256 SD_ID128_FORMAT_VAL(f
->header
->seqnum_id
),
4257 le64toh(f
->header
->head_entry_seqnum
),
4258 le64toh(f
->header
->head_entry_realtime
)) < 0)
4261 /* Try to rename the file to the archived version. If the file already was deleted, we'll get ENOENT, let's
4262 * ignore that case. */
4263 if (rename(f
->path
, p
) < 0 && errno
!= ENOENT
)
4266 /* Sync the rename to disk */
4267 (void) fsync_directory_of_file(f
->fd
);
4269 if (ret_previous_path
)
4270 *ret_previous_path
= f
->path
;
4274 f
->path
= TAKE_PTR(p
);
4276 /* Set as archive so offlining commits w/state=STATE_ARCHIVED. Previously we would set old_file->header->state
4277 * to STATE_ARCHIVED directly here, but journal_file_set_offline() short-circuits when state != STATE_ONLINE,
4278 * which would result in the rotated journal never getting fsync() called before closing. Now we simply queue
4279 * the archive state by setting an archive bit, leaving the state as STATE_ONLINE so proper offlining
4286 int journal_file_dispose(int dir_fd
, const char *fname
) {
4287 _cleanup_free_
char *p
= NULL
;
4291 /* Renames a journal file to *.journal~, i.e. to mark it as corrupted or otherwise uncleanly shutdown. Note that
4292 * this is done without looking into the file or changing any of its contents. The idea is that this is called
4293 * whenever something is suspicious and we want to move the file away and make clear that it is not accessed
4294 * for writing anymore. */
4296 if (!endswith(fname
, ".journal"))
4299 if (asprintf(&p
, "%.*s@%016" PRIx64
"-%016" PRIx64
".journal~",
4300 (int) strlen(fname
) - 8, fname
,
4301 now(CLOCK_REALTIME
),
4305 if (renameat(dir_fd
, fname
, dir_fd
, p
) < 0)
4311 int journal_file_copy_entry(
4317 sd_id128_t
*seqnum_id
) {
4319 _cleanup_free_ EntryItem
*items_alloc
= NULL
;
4321 uint64_t n
, m
= 0, xor_hash
= 0;
4331 if (!journal_file_writable(to
))
4334 ts
= (dual_timestamp
) {
4335 .monotonic
= le64toh(o
->entry
.monotonic
),
4336 .realtime
= le64toh(o
->entry
.realtime
),
4338 boot_id
= o
->entry
.boot_id
;
4340 n
= journal_file_entry_n_items(from
, o
);
4344 if (n
< ALLOCA_MAX
/ sizeof(EntryItem
) / 2)
4345 items
= newa(EntryItem
, n
);
4347 items_alloc
= new(EntryItem
, n
);
4351 items
= items_alloc
;
4354 for (uint64_t i
= 0; i
< n
; i
++) {
4360 q
= journal_file_entry_item_object_offset(from
, o
, i
);
4361 r
= journal_file_data_payload(from
, NULL
, q
, NULL
, 0, 0, &data
, &l
);
4362 if (IN_SET(r
, -EADDRNOTAVAIL
, -EBADMSG
)) {
4363 log_debug_errno(r
, "Entry item %"PRIu64
" data object is bad, skipping over it: %m", i
);
4373 r
= journal_file_append_data(to
, data
, l
, &u
, &h
);
4377 if (JOURNAL_HEADER_KEYED_HASH(to
->header
))
4378 xor_hash
^= jenkins_hash64(data
, l
);
4380 xor_hash
^= le64toh(u
->data
.hash
);
4382 items
[m
++] = (EntryItem
) {
4384 .hash
= le64toh(u
->data
.hash
),
4391 r
= journal_file_append_entry_internal(
4395 &from
->header
->machine_id
,
4401 /* ret_object= */ NULL
,
4402 /* ret_offset= */ NULL
);
4404 if (mmap_cache_fd_got_sigbus(to
->cache_fd
))
4410 void journal_reset_metrics(JournalMetrics
*m
) {
4413 /* Set everything to "pick automatic values". */
4415 *m
= (JournalMetrics
) {
4416 .min_use
= UINT64_MAX
,
4417 .max_use
= UINT64_MAX
,
4418 .min_size
= UINT64_MAX
,
4419 .max_size
= UINT64_MAX
,
4420 .keep_free
= UINT64_MAX
,
4421 .n_max_files
= UINT64_MAX
,
4425 int journal_file_get_cutoff_realtime_usec(JournalFile
*f
, usec_t
*ret_from
, usec_t
*ret_to
) {
4428 assert(ret_from
|| ret_to
);
4431 if (f
->header
->head_entry_realtime
== 0)
4434 *ret_from
= le64toh(f
->header
->head_entry_realtime
);
4438 if (f
->header
->tail_entry_realtime
== 0)
4441 *ret_to
= le64toh(f
->header
->tail_entry_realtime
);
4447 int journal_file_get_cutoff_monotonic_usec(JournalFile
*f
, sd_id128_t boot_id
, usec_t
*ret_from
, usec_t
*ret_to
) {
4453 assert(ret_from
|| ret_to
);
4455 /* FIXME: fix return value assignment on success with 0. */
4457 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &p
);
4461 if (le64toh(o
->data
.n_entries
) <= 0)
4465 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, le64toh(o
->data
.entry_offset
), &o
);
4469 *ret_from
= le64toh(o
->entry
.monotonic
);
4473 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
4477 r
= journal_file_move_to_entry_for_data(f
, o
, DIRECTION_UP
, &o
, NULL
);
4481 *ret_to
= le64toh(o
->entry
.monotonic
);
4487 bool journal_file_rotate_suggested(JournalFile
*f
, usec_t max_file_usec
, int log_level
) {
4491 /* If we gained new header fields we gained new features,
4492 * hence suggest a rotation */
4493 if (le64toh(f
->header
->header_size
) < sizeof(Header
)) {
4494 log_ratelimit_full(log_level
, JOURNAL_LOG_RATELIMIT
,
4495 "%s uses an outdated header, suggesting rotation.", f
->path
);
4499 /* Let's check if the hash tables grew over a certain fill level (75%, borrowing this value from
4500 * Java's hash table implementation), and if so suggest a rotation. To calculate the fill level we
4501 * need the n_data field, which only exists in newer versions. */
4503 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
4504 if (le64toh(f
->header
->n_data
) * 4ULL > (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
4506 log_level
, JOURNAL_LOG_RATELIMIT
,
4507 "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.",
4509 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))),
4510 le64toh(f
->header
->n_data
),
4511 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
4512 (uint64_t) f
->last_stat
.st_size
,
4513 f
->last_stat
.st_size
/ le64toh(f
->header
->n_data
));
4517 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
4518 if (le64toh(f
->header
->n_fields
) * 4ULL > (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
4520 log_level
, JOURNAL_LOG_RATELIMIT
,
4521 "Field hash table of %s has a fill level at %.1f (%"PRIu64
" of %"PRIu64
" items), suggesting rotation.",
4523 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))),
4524 le64toh(f
->header
->n_fields
),
4525 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
));
4529 /* If there are too many hash collisions somebody is most likely playing games with us. Hence, if our
4530 * longest chain is longer than some threshold, let's suggest rotation. */
4531 if (JOURNAL_HEADER_CONTAINS(f
->header
, data_hash_chain_depth
) &&
4532 le64toh(f
->header
->data_hash_chain_depth
) > HASH_CHAIN_DEPTH_MAX
) {
4534 log_level
, JOURNAL_LOG_RATELIMIT
,
4535 "Data hash table of %s has deepest hash chain of length %" PRIu64
", suggesting rotation.",
4536 f
->path
, le64toh(f
->header
->data_hash_chain_depth
));
4540 if (JOURNAL_HEADER_CONTAINS(f
->header
, field_hash_chain_depth
) &&
4541 le64toh(f
->header
->field_hash_chain_depth
) > HASH_CHAIN_DEPTH_MAX
) {
4543 log_level
, JOURNAL_LOG_RATELIMIT
,
4544 "Field hash table of %s has deepest hash chain of length at %" PRIu64
", suggesting rotation.",
4545 f
->path
, le64toh(f
->header
->field_hash_chain_depth
));
4549 /* Are the data objects properly indexed by field objects? */
4550 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
4551 JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
4552 le64toh(f
->header
->n_data
) > 0 &&
4553 le64toh(f
->header
->n_fields
) == 0) {
4555 log_level
, JOURNAL_LOG_RATELIMIT
,
4556 "Data objects of %s are not indexed by field objects, suggesting rotation.",
4561 if (max_file_usec
> 0) {
4564 h
= le64toh(f
->header
->head_entry_realtime
);
4565 t
= now(CLOCK_REALTIME
);
4567 if (h
> 0 && t
> h
+ max_file_usec
) {
4569 log_level
, JOURNAL_LOG_RATELIMIT
,
4570 "Oldest entry in %s is older than the configured file retention duration (%s), suggesting rotation.",
4571 f
->path
, FORMAT_TIMESPAN(max_file_usec
, USEC_PER_SEC
));
4579 static const char * const journal_object_type_table
[] = {
4580 [OBJECT_UNUSED
] = "unused",
4581 [OBJECT_DATA
] = "data",
4582 [OBJECT_FIELD
] = "field",
4583 [OBJECT_ENTRY
] = "entry",
4584 [OBJECT_DATA_HASH_TABLE
] = "data hash table",
4585 [OBJECT_FIELD_HASH_TABLE
] = "field hash table",
4586 [OBJECT_ENTRY_ARRAY
] = "entry array",
4587 [OBJECT_TAG
] = "tag",
4590 DEFINE_STRING_TABLE_LOOKUP_TO_STRING(journal_object_type
, ObjectType
);