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 /* This is the minimum journal file size */
51 #define JOURNAL_FILE_SIZE_MIN (512 * 1024ULL) /* 512 KiB */
52 #define JOURNAL_COMPACT_SIZE_MAX UINT32_MAX /* 4 GiB */
54 /* These are the lower and upper bounds if we deduce the max_use value
55 * from the file system size */
56 #define MAX_USE_LOWER (1 * 1024 * 1024ULL) /* 1 MiB */
57 #define MAX_USE_UPPER (4 * 1024 * 1024 * 1024ULL) /* 4 GiB */
59 /* Those are the lower and upper bounds for the minimal use limit,
60 * i.e. how much we'll use even if keep_free suggests otherwise. */
61 #define MIN_USE_LOW (1 * 1024 * 1024ULL) /* 1 MiB */
62 #define MIN_USE_HIGH (16 * 1024 * 1024ULL) /* 16 MiB */
64 /* This is the upper bound if we deduce max_size from max_use */
65 #define MAX_SIZE_UPPER (128 * 1024 * 1024ULL) /* 128 MiB */
67 /* This is the upper bound if we deduce the keep_free value from the
69 #define KEEP_FREE_UPPER (4 * 1024 * 1024 * 1024ULL) /* 4 GiB */
71 /* This is the keep_free value when we can't determine the system
73 #define DEFAULT_KEEP_FREE (1024 * 1024ULL) /* 1 MB */
75 /* This is the default maximum number of journal files to keep around. */
76 #define DEFAULT_N_MAX_FILES 100
78 /* n_data was the first entry we added after the initial file format design */
79 #define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
81 /* How many entries to keep in the entry array chain cache at max */
82 #define CHAIN_CACHE_MAX 20
84 /* How much to increase the journal file size at once each time we allocate something new. */
85 #define FILE_SIZE_INCREASE (8 * 1024 * 1024ULL) /* 8MB */
87 /* Reread fstat() of the file for detecting deletions at least this often */
88 #define LAST_STAT_REFRESH_USEC (5*USEC_PER_SEC)
90 /* The mmap context to use for the header we pick as one above the last defined typed */
91 #define CONTEXT_HEADER _OBJECT_TYPE_MAX
93 /* Longest hash chain to rotate after */
94 #define HASH_CHAIN_DEPTH_MAX 100
97 # pragma GCC diagnostic ignored "-Waddress-of-packed-member"
100 static int mmap_prot_from_open_flags(int flags
) {
101 switch (flags
& O_ACCMODE
) {
107 return PROT_READ
|PROT_WRITE
;
109 assert_not_reached();
113 int journal_file_tail_end_by_pread(JournalFile
*f
, uint64_t *ret_offset
) {
121 /* Same as journal_file_tail_end_by_mmap() below, but operates with pread() to avoid the mmap cache
122 * (and thus is thread safe) */
124 p
= le64toh(f
->header
->tail_object_offset
);
126 p
= le64toh(f
->header
->header_size
);
131 r
= journal_file_read_object_header(f
, OBJECT_UNUSED
, p
, &tail
);
135 sz
= le64toh(tail
.object
.size
);
136 if (sz
> UINT64_MAX
- sizeof(uint64_t) + 1)
140 if (p
> UINT64_MAX
- sz
)
151 int journal_file_tail_end_by_mmap(JournalFile
*f
, uint64_t *ret_offset
) {
159 /* Same as journal_file_tail_end_by_pread() above, but operates with the usual mmap logic */
161 p
= le64toh(f
->header
->tail_object_offset
);
163 p
= le64toh(f
->header
->header_size
);
168 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &tail
);
172 sz
= le64toh(READ_NOW(tail
->object
.size
));
173 if (sz
> UINT64_MAX
- sizeof(uint64_t) + 1)
177 if (p
> UINT64_MAX
- sz
)
188 int journal_file_set_offline_thread_join(JournalFile
*f
) {
193 if (f
->offline_state
== OFFLINE_JOINED
)
196 r
= pthread_join(f
->offline_thread
, NULL
);
200 f
->offline_state
= OFFLINE_JOINED
;
202 if (mmap_cache_fd_got_sigbus(f
->cache_fd
))
208 static int journal_file_set_online(JournalFile
*f
) {
213 if (!journal_file_writable(f
))
216 if (f
->fd
< 0 || !f
->header
)
220 switch (f
->offline_state
) {
222 /* No offline thread, no need to wait. */
226 case OFFLINE_SYNCING
: {
227 OfflineState tmp_state
= OFFLINE_SYNCING
;
228 if (!__atomic_compare_exchange_n(&f
->offline_state
, &tmp_state
, OFFLINE_CANCEL
,
229 false, __ATOMIC_SEQ_CST
, __ATOMIC_SEQ_CST
))
232 /* Canceled syncing prior to offlining, no need to wait. */
236 case OFFLINE_AGAIN_FROM_SYNCING
: {
237 OfflineState tmp_state
= OFFLINE_AGAIN_FROM_SYNCING
;
238 if (!__atomic_compare_exchange_n(&f
->offline_state
, &tmp_state
, OFFLINE_CANCEL
,
239 false, __ATOMIC_SEQ_CST
, __ATOMIC_SEQ_CST
))
242 /* Canceled restart from syncing, no need to wait. */
246 case OFFLINE_AGAIN_FROM_OFFLINING
: {
247 OfflineState tmp_state
= OFFLINE_AGAIN_FROM_OFFLINING
;
248 if (!__atomic_compare_exchange_n(&f
->offline_state
, &tmp_state
, OFFLINE_CANCEL
,
249 false, __ATOMIC_SEQ_CST
, __ATOMIC_SEQ_CST
))
252 /* Canceled restart from offlining, must wait for offlining to complete however. */
257 r
= journal_file_set_offline_thread_join(f
);
267 if (mmap_cache_fd_got_sigbus(f
->cache_fd
))
270 switch (f
->header
->state
) {
275 f
->header
->state
= STATE_ONLINE
;
284 JournalFile
* journal_file_close(JournalFile
*f
) {
288 assert(f
->newest_boot_id_prioq_idx
== PRIOQ_IDX_NULL
);
291 mmap_cache_fd_free(f
->cache_fd
);
297 ordered_hashmap_free_free(f
->chain_cache
);
300 free(f
->compress_buffer
);
305 munmap(f
->fss_file
, PAGE_ALIGN(f
->fss_file_size
));
307 free(f
->fsprg_state
);
312 gcry_md_close(f
->hmac
);
318 static bool keyed_hash_requested(void) {
319 static thread_local
int cached
= -1;
323 r
= getenv_bool("SYSTEMD_JOURNAL_KEYED_HASH");
326 log_debug_errno(r
, "Failed to parse $SYSTEMD_JOURNAL_KEYED_HASH environment variable, ignoring: %m");
335 static bool compact_mode_requested(void) {
336 static thread_local
int cached
= -1;
340 r
= getenv_bool("SYSTEMD_JOURNAL_COMPACT");
343 log_debug_errno(r
, "Failed to parse $SYSTEMD_JOURNAL_COMPACT environment variable, ignoring: %m");
353 static Compression
getenv_compression(void) {
358 e
= getenv("SYSTEMD_JOURNAL_COMPRESS");
360 return DEFAULT_COMPRESSION
;
362 r
= parse_boolean(e
);
364 return r
? DEFAULT_COMPRESSION
: COMPRESSION_NONE
;
366 c
= compression_from_string(e
);
368 log_debug_errno(c
, "Failed to parse SYSTEMD_JOURNAL_COMPRESS value, ignoring: %s", e
);
369 return DEFAULT_COMPRESSION
;
372 if (!compression_supported(c
)) {
373 log_debug("Unsupported compression algorithm specified, ignoring: %s", e
);
374 return DEFAULT_COMPRESSION
;
381 static Compression
compression_requested(void) {
383 static thread_local Compression cached
= _COMPRESSION_INVALID
;
386 cached
= getenv_compression();
390 return COMPRESSION_NONE
;
394 static int journal_file_init_header(
396 JournalFileFlags file_flags
,
397 JournalFile
*template) {
406 /* Try to load the FSPRG state, and if we can't, then just don't do sealing */
407 seal
= FLAGS_SET(file_flags
, JOURNAL_SEAL
) && journal_file_fss_load(f
) >= 0;
411 .header_size
= htole64(ALIGN64(sizeof(h
))),
412 .incompatible_flags
= htole32(
413 FLAGS_SET(file_flags
, JOURNAL_COMPRESS
) * COMPRESSION_TO_HEADER_INCOMPATIBLE_FLAG(compression_requested()) |
414 keyed_hash_requested() * HEADER_INCOMPATIBLE_KEYED_HASH
|
415 compact_mode_requested() * HEADER_INCOMPATIBLE_COMPACT
),
416 .compatible_flags
= htole32(
417 (seal
* HEADER_COMPATIBLE_SEALED
) |
418 HEADER_COMPATIBLE_TAIL_ENTRY_BOOT_ID
),
421 assert_cc(sizeof(h
.signature
) == sizeof(HEADER_SIGNATURE
));
422 memcpy(h
.signature
, HEADER_SIGNATURE
, sizeof(HEADER_SIGNATURE
));
424 r
= sd_id128_randomize(&h
.file_id
);
428 r
= sd_id128_get_machine(&h
.machine_id
);
429 if (r
< 0 && !ERRNO_IS_MACHINE_ID_UNSET(r
))
430 return r
; /* If we have no valid machine ID (test environment?), let's simply leave the
431 * machine ID field all zeroes. */
434 h
.seqnum_id
= template->header
->seqnum_id
;
435 h
.tail_entry_seqnum
= template->header
->tail_entry_seqnum
;
437 h
.seqnum_id
= h
.file_id
;
439 k
= pwrite(f
->fd
, &h
, sizeof(h
), 0);
448 static int journal_file_refresh_header(JournalFile
*f
) {
454 /* We used to update the header's boot ID field here, but we don't do that anymore, as per
455 * HEADER_COMPATIBLE_TAIL_ENTRY_BOOT_ID */
457 r
= journal_file_set_online(f
);
459 /* Sync the online state to disk; likely just created a new file, also sync the directory this file
461 (void) fsync_full(f
->fd
);
466 static bool warn_wrong_flags(const JournalFile
*f
, bool compatible
) {
467 const uint32_t any
= compatible
? HEADER_COMPATIBLE_ANY
: HEADER_INCOMPATIBLE_ANY
,
468 supported
= compatible
? HEADER_COMPATIBLE_SUPPORTED
: HEADER_INCOMPATIBLE_SUPPORTED
;
469 const char *type
= compatible
? "compatible" : "incompatible";
475 flags
= le32toh(compatible
? f
->header
->compatible_flags
: f
->header
->incompatible_flags
);
477 if (flags
& ~supported
) {
479 log_debug("Journal file %s has unknown %s flags 0x%"PRIx32
,
480 f
->path
, type
, flags
& ~any
);
481 flags
= (flags
& any
) & ~supported
;
485 _cleanup_free_
char *t
= NULL
;
488 if (flags
& HEADER_COMPATIBLE_SEALED
)
489 strv
[n
++] = "sealed";
491 if (flags
& HEADER_INCOMPATIBLE_COMPRESSED_XZ
)
492 strv
[n
++] = "xz-compressed";
493 if (flags
& HEADER_INCOMPATIBLE_COMPRESSED_LZ4
)
494 strv
[n
++] = "lz4-compressed";
495 if (flags
& HEADER_INCOMPATIBLE_COMPRESSED_ZSTD
)
496 strv
[n
++] = "zstd-compressed";
497 if (flags
& HEADER_INCOMPATIBLE_KEYED_HASH
)
498 strv
[n
++] = "keyed-hash";
499 if (flags
& HEADER_INCOMPATIBLE_COMPACT
)
500 strv
[n
++] = "compact";
503 assert(n
< ELEMENTSOF(strv
));
505 t
= strv_join((char**) strv
, ", ");
506 log_debug("Journal file %s uses %s %s %s disabled at compilation time.",
507 f
->path
, type
, n
> 1 ? "flags" : "flag", strnull(t
));
515 static bool offset_is_valid(uint64_t offset
, uint64_t header_size
, uint64_t tail_object_offset
) {
518 if (!VALID64(offset
))
520 if (offset
< header_size
)
522 if (offset
> tail_object_offset
)
527 static bool hash_table_is_valid(uint64_t offset
, uint64_t size
, uint64_t header_size
, uint64_t arena_size
, uint64_t tail_object_offset
) {
528 if ((offset
== 0) != (size
== 0))
532 if (offset
<= offsetof(Object
, hash_table
.items
))
534 offset
-= offsetof(Object
, hash_table
.items
);
535 if (!offset_is_valid(offset
, header_size
, tail_object_offset
))
537 assert(offset
<= header_size
+ arena_size
);
538 if (size
> header_size
+ arena_size
- offset
)
543 static int journal_file_verify_header(JournalFile
*f
) {
544 uint64_t arena_size
, header_size
;
549 if (memcmp(f
->header
->signature
, HEADER_SIGNATURE
, 8))
552 /* In both read and write mode we refuse to open files with incompatible
553 * flags we don't know. */
554 if (warn_wrong_flags(f
, false))
555 return -EPROTONOSUPPORT
;
557 /* When open for writing we refuse to open files with compatible flags, too. */
558 if (journal_file_writable(f
) && warn_wrong_flags(f
, true))
559 return -EPROTONOSUPPORT
;
561 if (f
->header
->state
>= _STATE_MAX
)
564 header_size
= le64toh(READ_NOW(f
->header
->header_size
));
566 /* The first addition was n_data, so check that we are at least this large */
567 if (header_size
< HEADER_SIZE_MIN
)
570 /* When open for writing we refuse to open files with a mismatch of the header size, i.e. writing to
571 * files implementing older or new header structures. */
572 if (journal_file_writable(f
) && header_size
!= sizeof(Header
))
573 return -EPROTONOSUPPORT
;
575 /* Don't write to journal files without the new boot ID update behavior guarantee. */
576 if (journal_file_writable(f
) && !JOURNAL_HEADER_TAIL_ENTRY_BOOT_ID(f
->header
))
577 return -EPROTONOSUPPORT
;
579 if (JOURNAL_HEADER_SEALED(f
->header
) && !JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
582 arena_size
= le64toh(READ_NOW(f
->header
->arena_size
));
584 if (UINT64_MAX
- header_size
< arena_size
|| header_size
+ arena_size
> (uint64_t) f
->last_stat
.st_size
)
587 uint64_t tail_object_offset
= le64toh(f
->header
->tail_object_offset
);
588 if (!offset_is_valid(tail_object_offset
, header_size
, UINT64_MAX
))
590 if (header_size
+ arena_size
< tail_object_offset
)
592 if (header_size
+ arena_size
- tail_object_offset
< sizeof(ObjectHeader
))
595 if (!hash_table_is_valid(le64toh(f
->header
->data_hash_table_offset
),
596 le64toh(f
->header
->data_hash_table_size
),
597 header_size
, arena_size
, tail_object_offset
))
600 if (!hash_table_is_valid(le64toh(f
->header
->field_hash_table_offset
),
601 le64toh(f
->header
->field_hash_table_size
),
602 header_size
, arena_size
, tail_object_offset
))
605 uint64_t entry_array_offset
= le64toh(f
->header
->entry_array_offset
);
606 if (!offset_is_valid(entry_array_offset
, header_size
, tail_object_offset
))
609 if (JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_array_offset
)) {
610 uint32_t offset
= le32toh(f
->header
->tail_entry_array_offset
);
611 uint32_t n
= le32toh(f
->header
->tail_entry_array_n_entries
);
613 if (!offset_is_valid(offset
, header_size
, tail_object_offset
))
615 if (entry_array_offset
> offset
)
617 if (entry_array_offset
== 0 && offset
!= 0)
619 if ((offset
== 0) != (n
== 0))
621 assert(offset
<= header_size
+ arena_size
);
622 if ((uint64_t) n
* journal_file_entry_array_item_size(f
) > header_size
+ arena_size
- offset
)
626 if (JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_offset
))
627 if (!offset_is_valid(le64toh(f
->header
->tail_entry_offset
), header_size
, tail_object_offset
))
630 /* Verify number of objects */
631 uint64_t n_objects
= le64toh(f
->header
->n_objects
);
632 if (n_objects
> arena_size
/ sizeof(ObjectHeader
))
635 uint64_t n_entries
= le64toh(f
->header
->n_entries
);
636 if (n_entries
> n_objects
)
639 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
640 le64toh(f
->header
->n_data
) > n_objects
)
643 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
644 le64toh(f
->header
->n_fields
) > n_objects
)
647 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
) &&
648 le64toh(f
->header
->n_tags
) > n_objects
)
651 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
) &&
652 le64toh(f
->header
->n_entry_arrays
) > n_objects
)
655 if (JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_array_n_entries
) &&
656 le32toh(f
->header
->tail_entry_array_n_entries
) > n_entries
)
659 if (journal_file_writable(f
)) {
660 sd_id128_t machine_id
;
664 r
= sd_id128_get_machine(&machine_id
);
665 if (ERRNO_IS_NEG_MACHINE_ID_UNSET(r
)) /* Gracefully handle the machine ID not being initialized yet */
666 machine_id
= SD_ID128_NULL
;
670 if (!sd_id128_equal(machine_id
, f
->header
->machine_id
))
671 return log_debug_errno(SYNTHETIC_ERRNO(EHOSTDOWN
),
672 "Trying to open journal file from different host for writing, refusing.");
674 state
= f
->header
->state
;
676 if (state
== STATE_ARCHIVED
)
677 return -ESHUTDOWN
; /* Already archived */
678 if (state
== STATE_ONLINE
)
679 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
),
680 "Journal file %s is already online. Assuming unclean closing.",
682 if (state
!= STATE_OFFLINE
)
683 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
),
684 "Journal file %s has unknown state %i.",
687 if (f
->header
->field_hash_table_size
== 0 || f
->header
->data_hash_table_size
== 0)
694 int journal_file_fstat(JournalFile
*f
) {
700 if (fstat(f
->fd
, &f
->last_stat
) < 0)
703 f
->last_stat_usec
= now(CLOCK_MONOTONIC
);
705 /* Refuse dealing with files that aren't regular */
706 r
= stat_verify_regular(&f
->last_stat
);
710 /* Refuse appending to files that are already deleted */
711 if (f
->last_stat
.st_nlink
<= 0)
717 static int journal_file_allocate(JournalFile
*f
, uint64_t offset
, uint64_t size
) {
718 uint64_t old_size
, new_size
, old_header_size
, old_arena_size
;
724 /* We assume that this file is not sparse, and we know that for sure, since we always call
725 * posix_fallocate() ourselves */
727 if (size
> PAGE_ALIGN_DOWN(UINT64_MAX
) - offset
)
730 if (mmap_cache_fd_got_sigbus(f
->cache_fd
))
733 old_header_size
= le64toh(READ_NOW(f
->header
->header_size
));
734 old_arena_size
= le64toh(READ_NOW(f
->header
->arena_size
));
735 if (old_arena_size
> PAGE_ALIGN_DOWN(UINT64_MAX
) - old_header_size
)
738 old_size
= old_header_size
+ old_arena_size
;
740 new_size
= MAX(PAGE_ALIGN(offset
+ size
), old_header_size
);
742 if (new_size
<= old_size
) {
744 /* We already pre-allocated enough space, but before
745 * we write to it, let's check with fstat() if the
746 * file got deleted, in order make sure we don't throw
747 * away the data immediately. Don't check fstat() for
748 * all writes though, but only once ever 10s. */
750 if (f
->last_stat_usec
+ LAST_STAT_REFRESH_USEC
> now(CLOCK_MONOTONIC
))
753 return journal_file_fstat(f
);
756 /* Allocate more space. */
758 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
761 /* Refuse to go over 4G in compact mode so offsets can be stored in 32-bit. */
762 if (JOURNAL_HEADER_COMPACT(f
->header
) && new_size
> UINT32_MAX
)
765 if (new_size
> f
->metrics
.min_size
&& f
->metrics
.keep_free
> 0) {
768 if (fstatvfs(f
->fd
, &svfs
) >= 0) {
771 available
= LESS_BY((uint64_t) svfs
.f_bfree
* (uint64_t) svfs
.f_bsize
, f
->metrics
.keep_free
);
773 if (new_size
- old_size
> available
)
778 /* Increase by larger blocks at once */
779 new_size
= ROUND_UP(new_size
, FILE_SIZE_INCREASE
);
780 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
781 new_size
= f
->metrics
.max_size
;
783 /* Note that the glibc fallocate() fallback is very
784 inefficient, hence we try to minimize the allocation area
786 r
= posix_fallocate_loop(f
->fd
, old_size
, new_size
- old_size
);
790 f
->header
->arena_size
= htole64(new_size
- old_header_size
);
792 return journal_file_fstat(f
);
795 static unsigned type_to_context(ObjectType type
) {
796 /* One context for each type, plus one catch-all for the rest */
797 assert_cc(_OBJECT_TYPE_MAX
<= MMAP_CACHE_MAX_CONTEXTS
);
798 assert_cc(CONTEXT_HEADER
< MMAP_CACHE_MAX_CONTEXTS
);
799 return type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
? type
: 0;
802 static int journal_file_move_to(
815 /* This function may clear, overwrite, or alter previously cached entries. After this function has
816 * been called, all objects except for one obtained by this function are invalidated and must be
817 * re-read before use. */
822 if (size
> UINT64_MAX
- offset
)
825 /* Avoid SIGBUS on invalid accesses */
826 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
) {
827 /* Hmm, out of range? Let's refresh the fstat() data
828 * first, before we trust that check. */
830 r
= journal_file_fstat(f
);
834 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
)
835 return -EADDRNOTAVAIL
;
838 return mmap_cache_fd_get(f
->cache_fd
, type_to_context(type
), keep_always
, offset
, size
, &f
->last_stat
, ret
);
841 static uint64_t minimum_header_size(JournalFile
*f
, Object
*o
) {
843 static const uint64_t table
[] = {
844 [OBJECT_DATA
] = sizeof(DataObject
),
845 [OBJECT_FIELD
] = sizeof(FieldObject
),
846 [OBJECT_ENTRY
] = sizeof(EntryObject
),
847 [OBJECT_DATA_HASH_TABLE
] = sizeof(HashTableObject
),
848 [OBJECT_FIELD_HASH_TABLE
] = sizeof(HashTableObject
),
849 [OBJECT_ENTRY_ARRAY
] = sizeof(EntryArrayObject
),
850 [OBJECT_TAG
] = sizeof(TagObject
),
856 if (o
->object
.type
== OBJECT_DATA
)
857 return journal_file_data_payload_offset(f
);
859 if (o
->object
.type
>= ELEMENTSOF(table
) || table
[o
->object
.type
] <= 0)
860 return sizeof(ObjectHeader
);
862 return table
[o
->object
.type
];
865 static int check_object_header(JournalFile
*f
, Object
*o
, ObjectType type
, uint64_t offset
) {
871 s
= le64toh(READ_NOW(o
->object
.size
));
873 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
874 "Attempt to move to uninitialized object: %" PRIu64
,
877 if (s
< sizeof(ObjectHeader
))
878 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
879 "Attempt to move to overly short object with size %"PRIu64
": %" PRIu64
,
882 if (o
->object
.type
<= OBJECT_UNUSED
|| o
->object
.type
>= _OBJECT_TYPE_MAX
)
883 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
884 "Attempt to move to object with invalid type (%u): %" PRIu64
,
885 o
->object
.type
, offset
);
887 if (type
> OBJECT_UNUSED
&& o
->object
.type
!= type
)
888 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
889 "Found %s object while expecting %s object: %" PRIu64
,
890 journal_object_type_to_string(o
->object
.type
),
891 journal_object_type_to_string(type
),
894 if (s
< minimum_header_size(f
, o
))
895 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
896 "Size of %s object (%"PRIu64
") is smaller than the minimum object size (%"PRIu64
"): %" PRIu64
,
897 journal_object_type_to_string(o
->object
.type
),
899 minimum_header_size(f
, o
),
905 /* Lightweight object checks. We want this to be fast, so that we won't
906 * slowdown every journal_file_move_to_object() call too much. */
907 static int check_object(JournalFile
*f
, Object
*o
, uint64_t offset
) {
911 switch (o
->object
.type
) {
914 if ((le64toh(o
->data
.entry_offset
) == 0) ^ (le64toh(o
->data
.n_entries
) == 0))
915 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
916 "Bad data n_entries: %" PRIu64
": %" PRIu64
,
917 le64toh(o
->data
.n_entries
),
920 if (le64toh(o
->object
.size
) <= journal_file_data_payload_offset(f
))
921 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
922 "Bad data size (<= %zu): %" PRIu64
": %" PRIu64
,
923 journal_file_data_payload_offset(f
),
924 le64toh(o
->object
.size
),
927 if (!VALID64(le64toh(o
->data
.next_hash_offset
)) ||
928 !VALID64(le64toh(o
->data
.next_field_offset
)) ||
929 !VALID64(le64toh(o
->data
.entry_offset
)) ||
930 !VALID64(le64toh(o
->data
.entry_array_offset
)))
931 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
932 "Invalid offset, next_hash_offset=" OFSfmt
", next_field_offset=" OFSfmt
", entry_offset=" OFSfmt
", entry_array_offset=" OFSfmt
": %" PRIu64
,
933 le64toh(o
->data
.next_hash_offset
),
934 le64toh(o
->data
.next_field_offset
),
935 le64toh(o
->data
.entry_offset
),
936 le64toh(o
->data
.entry_array_offset
),
942 if (le64toh(o
->object
.size
) <= offsetof(Object
, field
.payload
))
943 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
944 "Bad field size (<= %zu): %" PRIu64
": %" PRIu64
,
945 offsetof(Object
, field
.payload
),
946 le64toh(o
->object
.size
),
949 if (!VALID64(le64toh(o
->field
.next_hash_offset
)) ||
950 !VALID64(le64toh(o
->field
.head_data_offset
)))
951 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
952 "Invalid offset, next_hash_offset=" OFSfmt
", head_data_offset=" OFSfmt
": %" PRIu64
,
953 le64toh(o
->field
.next_hash_offset
),
954 le64toh(o
->field
.head_data_offset
),
961 sz
= le64toh(READ_NOW(o
->object
.size
));
962 if (sz
< offsetof(Object
, entry
.items
) ||
963 (sz
- offsetof(Object
, entry
.items
)) % journal_file_entry_item_size(f
) != 0)
964 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
965 "Bad entry size (<= %zu): %" PRIu64
": %" PRIu64
,
966 offsetof(Object
, entry
.items
),
970 if ((sz
- offsetof(Object
, entry
.items
)) / journal_file_entry_item_size(f
) <= 0)
971 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
972 "Invalid number items in entry: %" PRIu64
": %" PRIu64
,
973 (sz
- offsetof(Object
, entry
.items
)) / journal_file_entry_item_size(f
),
976 if (le64toh(o
->entry
.seqnum
) <= 0)
977 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
978 "Invalid entry seqnum: %" PRIx64
": %" PRIu64
,
979 le64toh(o
->entry
.seqnum
),
982 if (!VALID_REALTIME(le64toh(o
->entry
.realtime
)))
983 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
984 "Invalid entry realtime timestamp: %" PRIu64
": %" PRIu64
,
985 le64toh(o
->entry
.realtime
),
988 if (!VALID_MONOTONIC(le64toh(o
->entry
.monotonic
)))
989 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
990 "Invalid entry monotonic timestamp: %" PRIu64
": %" PRIu64
,
991 le64toh(o
->entry
.monotonic
),
997 case OBJECT_DATA_HASH_TABLE
:
998 case OBJECT_FIELD_HASH_TABLE
: {
1001 sz
= le64toh(READ_NOW(o
->object
.size
));
1002 if (sz
< offsetof(Object
, hash_table
.items
) ||
1003 (sz
- offsetof(Object
, hash_table
.items
)) % sizeof(HashItem
) != 0 ||
1004 (sz
- offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
) <= 0)
1005 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1006 "Invalid %s hash table size: %" PRIu64
": %" PRIu64
,
1007 journal_object_type_to_string(o
->object
.type
),
1014 case OBJECT_ENTRY_ARRAY
: {
1017 sz
= le64toh(READ_NOW(o
->object
.size
));
1018 if (sz
< offsetof(Object
, entry_array
.items
) ||
1019 (sz
- offsetof(Object
, entry_array
.items
)) % journal_file_entry_array_item_size(f
) != 0 ||
1020 (sz
- offsetof(Object
, entry_array
.items
)) / journal_file_entry_array_item_size(f
) <= 0)
1021 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1022 "Invalid object entry array size: %" PRIu64
": %" PRIu64
,
1025 /* Here, we request that the offset of each entry array object is in strictly increasing order. */
1026 next
= le64toh(o
->entry_array
.next_entry_array_offset
);
1027 if (!VALID64(next
) || (next
> 0 && next
<= offset
))
1028 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1029 "Invalid object entry array next_entry_array_offset: %" PRIu64
": %" PRIu64
,
1037 if (le64toh(o
->object
.size
) != sizeof(TagObject
))
1038 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1039 "Invalid object tag size: %" PRIu64
": %" PRIu64
,
1040 le64toh(o
->object
.size
),
1043 if (!VALID_EPOCH(le64toh(o
->tag
.epoch
)))
1044 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1045 "Invalid object tag epoch: %" PRIu64
": %" PRIu64
,
1046 le64toh(o
->tag
.epoch
), offset
);
1054 int journal_file_move_to_object(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
**ret
) {
1060 /* Even if this function fails, it may clear, overwrite, or alter previously cached entries. After
1061 * this function has been called, all objects except for one obtained by this function are
1062 * invalidated and must be re-read before use.. */
1064 /* Objects may only be located at multiple of 64 bit */
1065 if (!VALID64(offset
))
1066 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1067 "Attempt to move to %s object at non-64-bit boundary: %" PRIu64
,
1068 journal_object_type_to_string(type
),
1071 /* Object may not be located in the file header */
1072 if (offset
< le64toh(f
->header
->header_size
))
1073 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1074 "Attempt to move to %s object located in file header: %" PRIu64
,
1075 journal_object_type_to_string(type
),
1078 r
= journal_file_move_to(f
, type
, false, offset
, sizeof(ObjectHeader
), (void**) &o
);
1082 r
= check_object_header(f
, o
, type
, offset
);
1086 r
= journal_file_move_to(f
, type
, false, offset
, le64toh(READ_NOW(o
->object
.size
)), (void**) &o
);
1090 r
= check_object_header(f
, o
, type
, offset
);
1094 r
= check_object(f
, o
, offset
);
1104 int journal_file_read_object_header(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
*ret
) {
1111 /* Objects may only be located at multiple of 64 bit */
1112 if (!VALID64(offset
))
1113 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1114 "Attempt to read %s object at non-64-bit boundary: %" PRIu64
,
1115 journal_object_type_to_string(type
), offset
);
1117 /* Object may not be located in the file header */
1118 if (offset
< le64toh(f
->header
->header_size
))
1119 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1120 "Attempt to read %s object located in file header: %" PRIu64
,
1121 journal_object_type_to_string(type
), offset
);
1123 /* This will likely read too much data but it avoids having to call pread() twice. */
1124 n
= pread(f
->fd
, &o
, sizeof(o
), offset
);
1126 return log_debug_errno(errno
, "Failed to read journal %s object at offset: %" PRIu64
,
1127 journal_object_type_to_string(type
), offset
);
1129 if ((size_t) n
< sizeof(o
.object
))
1130 return log_debug_errno(SYNTHETIC_ERRNO(EIO
),
1131 "Failed to read short %s object at offset: %" PRIu64
,
1132 journal_object_type_to_string(type
), offset
);
1134 r
= check_object_header(f
, &o
, type
, offset
);
1138 if ((size_t) n
< minimum_header_size(f
, &o
))
1139 return log_debug_errno(SYNTHETIC_ERRNO(EIO
),
1140 "Short read while reading %s object: %" PRIu64
,
1141 journal_object_type_to_string(type
), offset
);
1143 r
= check_object(f
, &o
, offset
);
1153 static uint64_t inc_seqnum(uint64_t seqnum
) {
1154 if (seqnum
< UINT64_MAX
-1)
1157 return 1; /* skip over UINT64_MAX and 0 when we run out of seqnums and start again */
1160 static uint64_t journal_file_entry_seqnum(
1164 uint64_t next_seqnum
;
1169 /* Picks a new sequence number for the entry we are about to add and returns it. */
1171 next_seqnum
= inc_seqnum(le64toh(f
->header
->tail_entry_seqnum
));
1173 /* If an external seqnum counter was passed, we update both the local and the external one, and set
1174 * it to the maximum of both */
1176 *seqnum
= next_seqnum
= MAX(inc_seqnum(*seqnum
), next_seqnum
);
1178 f
->header
->tail_entry_seqnum
= htole64(next_seqnum
);
1180 if (f
->header
->head_entry_seqnum
== 0)
1181 f
->header
->head_entry_seqnum
= htole64(next_seqnum
);
1186 int journal_file_append_object(
1190 Object
**ret_object
,
1191 uint64_t *ret_offset
) {
1199 assert(type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
);
1200 assert(size
>= sizeof(ObjectHeader
));
1202 r
= journal_file_set_online(f
);
1206 r
= journal_file_tail_end_by_mmap(f
, &p
);
1210 r
= journal_file_allocate(f
, p
, size
);
1214 r
= journal_file_move_to(f
, type
, false, p
, size
, (void**) &o
);
1218 o
->object
= (ObjectHeader
) {
1220 .size
= htole64(size
),
1223 f
->header
->tail_object_offset
= htole64(p
);
1224 f
->header
->n_objects
= htole64(le64toh(f
->header
->n_objects
) + 1);
1235 static int journal_file_setup_data_hash_table(JournalFile
*f
) {
1243 /* We estimate that we need 1 hash table entry per 768 bytes
1244 of journal file and we want to make sure we never get
1245 beyond 75% fill level. Calculate the hash table size for
1246 the maximum file size based on these metrics. */
1248 s
= (f
->metrics
.max_size
* 4 / 768 / 3) * sizeof(HashItem
);
1249 if (s
< DEFAULT_DATA_HASH_TABLE_SIZE
)
1250 s
= DEFAULT_DATA_HASH_TABLE_SIZE
;
1252 log_debug("Reserving %"PRIu64
" entries in data hash table.", s
/ sizeof(HashItem
));
1254 r
= journal_file_append_object(f
,
1255 OBJECT_DATA_HASH_TABLE
,
1256 offsetof(Object
, hash_table
.items
) + s
,
1261 memzero(o
->hash_table
.items
, s
);
1263 f
->header
->data_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1264 f
->header
->data_hash_table_size
= htole64(s
);
1269 static int journal_file_setup_field_hash_table(JournalFile
*f
) {
1277 /* We use a fixed size hash table for the fields as this
1278 * number should grow very slowly only */
1280 s
= DEFAULT_FIELD_HASH_TABLE_SIZE
;
1281 log_debug("Reserving %"PRIu64
" entries in field hash table.", s
/ sizeof(HashItem
));
1283 r
= journal_file_append_object(f
,
1284 OBJECT_FIELD_HASH_TABLE
,
1285 offsetof(Object
, hash_table
.items
) + s
,
1290 memzero(o
->hash_table
.items
, s
);
1292 f
->header
->field_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1293 f
->header
->field_hash_table_size
= htole64(s
);
1298 int journal_file_map_data_hash_table(JournalFile
*f
) {
1306 if (f
->data_hash_table
)
1309 p
= le64toh(f
->header
->data_hash_table_offset
);
1310 s
= le64toh(f
->header
->data_hash_table_size
);
1312 r
= journal_file_move_to(f
,
1313 OBJECT_DATA_HASH_TABLE
,
1320 f
->data_hash_table
= t
;
1324 int journal_file_map_field_hash_table(JournalFile
*f
) {
1332 if (f
->field_hash_table
)
1335 p
= le64toh(f
->header
->field_hash_table_offset
);
1336 s
= le64toh(f
->header
->field_hash_table_size
);
1338 r
= journal_file_move_to(f
,
1339 OBJECT_FIELD_HASH_TABLE
,
1346 f
->field_hash_table
= t
;
1350 static int journal_file_link_field(
1361 assert(f
->field_hash_table
);
1365 if (o
->object
.type
!= OBJECT_FIELD
)
1368 m
= le64toh(READ_NOW(f
->header
->field_hash_table_size
)) / sizeof(HashItem
);
1372 /* This might alter the window we are looking at */
1373 o
->field
.next_hash_offset
= o
->field
.head_data_offset
= 0;
1376 p
= le64toh(f
->field_hash_table
[h
].tail_hash_offset
);
1378 f
->field_hash_table
[h
].head_hash_offset
= htole64(offset
);
1380 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1384 o
->field
.next_hash_offset
= htole64(offset
);
1387 f
->field_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1389 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
1390 f
->header
->n_fields
= htole64(le64toh(f
->header
->n_fields
) + 1);
1395 static int journal_file_link_data(
1406 assert(f
->data_hash_table
);
1410 if (o
->object
.type
!= OBJECT_DATA
)
1413 m
= le64toh(READ_NOW(f
->header
->data_hash_table_size
)) / sizeof(HashItem
);
1417 /* This might alter the window we are looking at */
1418 o
->data
.next_hash_offset
= o
->data
.next_field_offset
= 0;
1419 o
->data
.entry_offset
= o
->data
.entry_array_offset
= 0;
1420 o
->data
.n_entries
= 0;
1423 p
= le64toh(f
->data_hash_table
[h
].tail_hash_offset
);
1425 /* Only entry in the hash table is easy */
1426 f
->data_hash_table
[h
].head_hash_offset
= htole64(offset
);
1428 /* Move back to the previous data object, to patch in
1431 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1435 o
->data
.next_hash_offset
= htole64(offset
);
1438 f
->data_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1440 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
1441 f
->header
->n_data
= htole64(le64toh(f
->header
->n_data
) + 1);
1446 static int get_next_hash_offset(
1449 le64_t
*next_hash_offset
,
1451 le64_t
*header_max_depth
) {
1457 assert(next_hash_offset
);
1460 nextp
= le64toh(READ_NOW(*next_hash_offset
));
1462 if (nextp
<= *p
) /* Refuse going in loops */
1463 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1464 "Detected hash item loop in %s, refusing.", f
->path
);
1468 /* If the depth of this hash chain is larger than all others we have seen so far, record it */
1469 if (header_max_depth
&& journal_file_writable(f
))
1470 *header_max_depth
= htole64(MAX(*depth
, le64toh(*header_max_depth
)));
1477 int journal_file_find_field_object_with_hash(
1482 Object
**ret_object
,
1483 uint64_t *ret_offset
) {
1485 uint64_t p
, osize
, h
, m
, depth
= 0;
1493 /* If the field hash table is empty, we can't find anything */
1494 if (le64toh(f
->header
->field_hash_table_size
) <= 0)
1497 /* Map the field hash table, if it isn't mapped yet. */
1498 r
= journal_file_map_field_hash_table(f
);
1502 osize
= offsetof(Object
, field
.payload
) + size
;
1504 m
= le64toh(READ_NOW(f
->header
->field_hash_table_size
)) / sizeof(HashItem
);
1509 p
= le64toh(f
->field_hash_table
[h
].head_hash_offset
);
1513 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1517 if (le64toh(o
->field
.hash
) == hash
&&
1518 le64toh(o
->object
.size
) == osize
&&
1519 memcmp(o
->field
.payload
, field
, size
) == 0) {
1529 r
= get_next_hash_offset(
1532 &o
->field
.next_hash_offset
,
1534 JOURNAL_HEADER_CONTAINS(f
->header
, field_hash_chain_depth
) ? &f
->header
->field_hash_chain_depth
: NULL
);
1542 uint64_t journal_file_hash_data(
1549 assert(data
|| sz
== 0);
1551 /* We try to unify our codebase on siphash, hence new-styled journal files utilizing the keyed hash
1552 * function use siphash. Old journal files use the Jenkins hash. */
1554 if (JOURNAL_HEADER_KEYED_HASH(f
->header
))
1555 return siphash24(data
, sz
, f
->header
->file_id
.bytes
);
1557 return jenkins_hash64(data
, sz
);
1560 int journal_file_find_field_object(
1564 Object
**ret_object
,
1565 uint64_t *ret_offset
) {
1571 return journal_file_find_field_object_with_hash(
1574 journal_file_hash_data(f
, field
, size
),
1575 ret_object
, ret_offset
);
1578 int journal_file_find_data_object_with_hash(
1583 Object
**ret_object
,
1584 uint64_t *ret_offset
) {
1586 uint64_t p
, h
, m
, depth
= 0;
1591 assert(data
|| size
== 0);
1593 /* If there's no data hash table, then there's no entry. */
1594 if (le64toh(f
->header
->data_hash_table_size
) <= 0)
1597 /* Map the data hash table, if it isn't mapped yet. */
1598 r
= journal_file_map_data_hash_table(f
);
1602 m
= le64toh(READ_NOW(f
->header
->data_hash_table_size
)) / sizeof(HashItem
);
1607 p
= le64toh(f
->data_hash_table
[h
].head_hash_offset
);
1614 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1618 if (le64toh(o
->data
.hash
) != hash
)
1621 r
= journal_file_data_payload(f
, o
, p
, NULL
, 0, 0, &d
, &rsize
);
1624 assert(r
> 0); /* journal_file_data_payload() always returns > 0 if no field is provided. */
1626 if (memcmp_nn(data
, size
, d
, rsize
) == 0) {
1637 r
= get_next_hash_offset(
1640 &o
->data
.next_hash_offset
,
1642 JOURNAL_HEADER_CONTAINS(f
->header
, data_hash_chain_depth
) ? &f
->header
->data_hash_chain_depth
: NULL
);
1650 int journal_file_find_data_object(
1654 Object
**ret_object
,
1655 uint64_t *ret_offset
) {
1658 assert(data
|| size
== 0);
1660 return journal_file_find_data_object_with_hash(
1663 journal_file_hash_data(f
, data
, size
),
1664 ret_object
, ret_offset
);
1667 bool journal_field_valid(const char *p
, size_t l
, bool allow_protected
) {
1668 /* We kinda enforce POSIX syntax recommendations for
1669 environment variables here, but make a couple of additional
1672 http://pubs.opengroup.org/onlinepubs/000095399/basedefs/xbd_chap08.html */
1679 /* No empty field names */
1683 /* Don't allow names longer than 64 chars */
1687 /* Variables starting with an underscore are protected */
1688 if (!allow_protected
&& p
[0] == '_')
1691 /* Don't allow digits as first character */
1692 if (ascii_isdigit(p
[0]))
1695 /* Only allow A-Z0-9 and '_' */
1696 for (const char *a
= p
; a
< p
+ l
; a
++)
1697 if ((*a
< 'A' || *a
> 'Z') &&
1698 !ascii_isdigit(*a
) &&
1705 static int journal_file_append_field(
1709 Object
**ret_object
,
1710 uint64_t *ret_offset
) {
1721 if (!journal_field_valid(field
, size
, true))
1724 hash
= journal_file_hash_data(f
, field
, size
);
1726 r
= journal_file_find_field_object_with_hash(f
, field
, size
, hash
, ret_object
, ret_offset
);
1732 osize
= offsetof(Object
, field
.payload
) + size
;
1733 r
= journal_file_append_object(f
, OBJECT_FIELD
, osize
, &o
, &p
);
1737 o
->field
.hash
= htole64(hash
);
1738 memcpy(o
->field
.payload
, field
, size
);
1740 r
= journal_file_link_field(f
, o
, p
, hash
);
1744 /* The linking might have altered the window, so let's only pass the offset to hmac which will
1745 * move to the object again if needed. */
1748 r
= journal_file_hmac_put_object(f
, OBJECT_FIELD
, NULL
, p
);
1754 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, ret_object
);
1765 static int maybe_compress_payload(JournalFile
*f
, uint8_t *dst
, const uint8_t *src
, uint64_t size
, size_t *rsize
) {
1769 #if HAVE_COMPRESSION
1773 c
= JOURNAL_FILE_COMPRESSION(f
);
1774 if (c
== COMPRESSION_NONE
|| size
< f
->compress_threshold_bytes
)
1777 r
= compress_blob(c
, src
, size
, dst
, size
- 1, rsize
);
1779 return log_debug_errno(r
, "Failed to compress data object using %s, ignoring: %m", compression_to_string(c
));
1781 log_debug("Compressed data object %"PRIu64
" -> %zu using %s", size
, *rsize
, compression_to_string(c
));
1783 return 1; /* compressed */
1789 static int journal_file_append_data(
1793 Object
**ret_object
,
1794 uint64_t *ret_offset
) {
1796 uint64_t hash
, p
, osize
;
1804 if (!data
|| size
== 0)
1807 hash
= journal_file_hash_data(f
, data
, size
);
1809 r
= journal_file_find_data_object_with_hash(f
, data
, size
, hash
, ret_object
, ret_offset
);
1815 eq
= memchr(data
, '=', size
);
1819 osize
= journal_file_data_payload_offset(f
) + size
;
1820 r
= journal_file_append_object(f
, OBJECT_DATA
, osize
, &o
, &p
);
1824 o
->data
.hash
= htole64(hash
);
1826 r
= maybe_compress_payload(f
, journal_file_data_payload_field(f
, o
), data
, size
, &rsize
);
1828 /* We don't really care failures, let's continue without compression */
1829 memcpy_safe(journal_file_data_payload_field(f
, o
), data
, size
);
1831 Compression c
= JOURNAL_FILE_COMPRESSION(f
);
1833 assert(c
>= 0 && c
< _COMPRESSION_MAX
&& c
!= COMPRESSION_NONE
);
1835 o
->object
.size
= htole64(journal_file_data_payload_offset(f
) + rsize
);
1836 o
->object
.flags
|= COMPRESSION_TO_OBJECT_FLAG(c
);
1839 r
= journal_file_link_data(f
, o
, p
, hash
);
1843 /* The linking might have altered the window, so let's refresh our pointer. */
1844 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1849 r
= journal_file_hmac_put_object(f
, OBJECT_DATA
, o
, p
);
1854 /* Create field object ... */
1855 r
= journal_file_append_field(f
, data
, (uint8_t*) eq
- (uint8_t*) data
, &fo
, NULL
);
1859 /* ... and link it in. */
1860 o
->data
.next_field_offset
= fo
->field
.head_data_offset
;
1861 fo
->field
.head_data_offset
= le64toh(p
);
1872 static int maybe_decompress_payload(
1876 Compression compression
,
1878 size_t field_length
,
1879 size_t data_threshold
,
1885 /* We can't read objects larger than 4G on a 32-bit machine */
1886 if ((uint64_t) (size_t) size
!= size
)
1889 if (compression
!= COMPRESSION_NONE
) {
1890 #if HAVE_COMPRESSION
1895 r
= decompress_startswith(compression
, payload
, size
, &f
->compress_buffer
, field
,
1898 return log_debug_errno(r
,
1899 "Cannot decompress %s object of length %" PRIu64
": %m",
1900 compression_to_string(compression
),
1911 r
= decompress_blob(compression
, payload
, size
, &f
->compress_buffer
, &rsize
, 0);
1916 *ret_data
= f
->compress_buffer
;
1920 return -EPROTONOSUPPORT
;
1923 if (field
&& (size
< field_length
+ 1 || memcmp(payload
, field
, field_length
) != 0 || payload
[field_length
] != '=')) {
1932 *ret_data
= payload
;
1934 *ret_size
= (size_t) size
;
1940 int journal_file_data_payload(
1945 size_t field_length
,
1946 size_t data_threshold
,
1955 assert(!field
== (field_length
== 0)); /* These must be specified together. */
1958 r
= journal_file_move_to_object(f
, OBJECT_DATA
, offset
, &o
);
1963 size
= le64toh(READ_NOW(o
->object
.size
));
1964 if (size
< journal_file_data_payload_offset(f
))
1967 size
-= journal_file_data_payload_offset(f
);
1969 c
= COMPRESSION_FROM_OBJECT(o
);
1971 return -EPROTONOSUPPORT
;
1973 return maybe_decompress_payload(f
, journal_file_data_payload_field(f
, o
), size
, c
, field
,
1974 field_length
, data_threshold
, ret_data
, ret_size
);
1977 uint64_t journal_file_entry_n_items(JournalFile
*f
, Object
*o
) {
1983 if (o
->object
.type
!= OBJECT_ENTRY
)
1986 sz
= le64toh(READ_NOW(o
->object
.size
));
1987 if (sz
< offsetof(Object
, entry
.items
))
1990 return (sz
- offsetof(Object
, entry
.items
)) / journal_file_entry_item_size(f
);
1993 uint64_t journal_file_entry_array_n_items(JournalFile
*f
, Object
*o
) {
1999 if (o
->object
.type
!= OBJECT_ENTRY_ARRAY
)
2002 sz
= le64toh(READ_NOW(o
->object
.size
));
2003 if (sz
< offsetof(Object
, entry_array
.items
))
2006 return (sz
- offsetof(Object
, entry_array
.items
)) / journal_file_entry_array_item_size(f
);
2009 uint64_t journal_file_hash_table_n_items(Object
*o
) {
2014 if (!IN_SET(o
->object
.type
, OBJECT_DATA_HASH_TABLE
, OBJECT_FIELD_HASH_TABLE
))
2017 sz
= le64toh(READ_NOW(o
->object
.size
));
2018 if (sz
< offsetof(Object
, hash_table
.items
))
2021 return (sz
- offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
);
2024 static void write_entry_array_item(JournalFile
*f
, Object
*o
, uint64_t i
, uint64_t p
) {
2028 if (JOURNAL_HEADER_COMPACT(f
->header
)) {
2029 assert(p
<= UINT32_MAX
);
2030 o
->entry_array
.items
.compact
[i
] = htole32(p
);
2032 o
->entry_array
.items
.regular
[i
] = htole64(p
);
2035 static int link_entry_into_array(
2043 uint64_t n
= 0, ap
= 0, q
, i
, a
, hidx
;
2053 a
= tail
? le32toh(*tail
) : le64toh(*first
);
2054 hidx
= le64toh(READ_NOW(*idx
));
2055 i
= tidx
? le32toh(READ_NOW(*tidx
)) : hidx
;
2058 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2062 n
= journal_file_entry_array_n_items(f
, o
);
2064 write_entry_array_item(f
, o
, i
, p
);
2065 *idx
= htole64(hidx
+ 1);
2067 *tidx
= htole32(le32toh(*tidx
) + 1);
2073 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
2084 r
= journal_file_append_object(f
, OBJECT_ENTRY_ARRAY
,
2085 offsetof(Object
, entry_array
.items
) + n
* journal_file_entry_array_item_size(f
),
2091 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY_ARRAY
, o
, q
);
2096 write_entry_array_item(f
, o
, i
, p
);
2099 *first
= htole64(q
);
2101 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, ap
, &o
);
2105 o
->entry_array
.next_entry_array_offset
= htole64(q
);
2111 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
2112 f
->header
->n_entry_arrays
= htole64(le64toh(f
->header
->n_entry_arrays
) + 1);
2114 *idx
= htole64(hidx
+ 1);
2121 static int link_entry_into_array_plus_one(
2139 hidx
= le64toh(READ_NOW(*idx
));
2140 if (hidx
== UINT64_MAX
)
2143 *extra
= htole64(p
);
2147 i
= htole64(hidx
- 1);
2148 r
= link_entry_into_array(f
, first
, &i
, tail
, tidx
, p
);
2153 *idx
= htole64(hidx
+ 1);
2157 static int journal_file_link_entry_item(JournalFile
*f
, uint64_t offset
, uint64_t p
) {
2164 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
2168 return link_entry_into_array_plus_one(f
,
2169 &o
->data
.entry_offset
,
2170 &o
->data
.entry_array_offset
,
2172 JOURNAL_HEADER_COMPACT(f
->header
) ? &o
->data
.compact
.tail_entry_array_offset
: NULL
,
2173 JOURNAL_HEADER_COMPACT(f
->header
) ? &o
->data
.compact
.tail_entry_array_n_entries
: NULL
,
2177 static int journal_file_link_entry(
2181 const EntryItem items
[],
2191 if (o
->object
.type
!= OBJECT_ENTRY
)
2194 __atomic_thread_fence(__ATOMIC_SEQ_CST
);
2196 /* Link up the entry itself */
2197 r
= link_entry_into_array(f
,
2198 &f
->header
->entry_array_offset
,
2199 &f
->header
->n_entries
,
2200 JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_array_offset
) ? &f
->header
->tail_entry_array_offset
: NULL
,
2201 JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_array_n_entries
) ? &f
->header
->tail_entry_array_n_entries
: NULL
,
2206 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
2208 if (f
->header
->head_entry_realtime
== 0)
2209 f
->header
->head_entry_realtime
= o
->entry
.realtime
;
2211 f
->header
->tail_entry_realtime
= o
->entry
.realtime
;
2212 f
->header
->tail_entry_monotonic
= o
->entry
.monotonic
;
2213 if (JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_offset
))
2214 f
->header
->tail_entry_offset
= htole64(offset
);
2215 f
->newest_mtime
= 0; /* we have a new tail entry now, explicitly invalidate newest boot id/timestamp info */
2217 /* Link up the items */
2218 for (uint64_t i
= 0; i
< n_items
; i
++) {
2221 /* If we fail to link an entry item because we can't allocate a new entry array, don't fail
2222 * immediately but try to link the other entry items since it might still be possible to link
2223 * those if they don't require a new entry array to be allocated. */
2225 k
= journal_file_link_entry_item(f
, offset
, items
[i
].object_offset
);
2235 static void write_entry_item(JournalFile
*f
, Object
*o
, uint64_t i
, const EntryItem
*item
) {
2240 if (JOURNAL_HEADER_COMPACT(f
->header
)) {
2241 assert(item
->object_offset
<= UINT32_MAX
);
2242 o
->entry
.items
.compact
[i
].object_offset
= htole32(item
->object_offset
);
2244 o
->entry
.items
.regular
[i
].object_offset
= htole64(item
->object_offset
);
2245 o
->entry
.items
.regular
[i
].hash
= htole64(item
->hash
);
2249 static int journal_file_append_entry_internal(
2251 const dual_timestamp
*ts
,
2252 const sd_id128_t
*boot_id
,
2253 const sd_id128_t
*machine_id
,
2255 const EntryItem items
[],
2258 sd_id128_t
*seqnum_id
,
2259 Object
**ret_object
,
2260 uint64_t *ret_offset
) {
2270 assert(items
|| n_items
== 0);
2272 if (f
->strict_order
) {
2273 /* If requested be stricter with ordering in this journal file, to make searching via
2274 * bisection fully deterministic. This is an optional feature, so that if desired journal
2275 * files can be written where the ordering is not strictly enforced (in which case bisection
2276 * will yield *a* result, but not the *only* result, when searching for points in
2277 * time). Strict ordering mode is enabled when journald originally writes the files, but
2278 * might not necessarily be if other tools (the remoting tools for example) write journal
2279 * files from combined sources.
2281 * Typically, if any of the errors generated here are seen journald will just rotate the
2282 * journal files and start anew. */
2284 if (ts
->realtime
< le64toh(f
->header
->tail_entry_realtime
))
2285 return log_debug_errno(SYNTHETIC_ERRNO(EREMCHG
),
2286 "Realtime timestamp %" PRIu64
" smaller than previous realtime "
2287 "timestamp %" PRIu64
", refusing entry.",
2288 ts
->realtime
, le64toh(f
->header
->tail_entry_realtime
));
2290 if ((!boot_id
|| sd_id128_equal(*boot_id
, f
->header
->tail_entry_boot_id
)) &&
2291 ts
->monotonic
< le64toh(f
->header
->tail_entry_monotonic
))
2292 return log_debug_errno(
2293 SYNTHETIC_ERRNO(ENOTNAM
),
2294 "Monotonic timestamp %" PRIu64
2295 " smaller than previous monotonic timestamp %" PRIu64
2296 " while having the same boot ID, refusing entry.",
2298 le64toh(f
->header
->tail_entry_monotonic
));
2302 /* Settle the passed in sequence number ID */
2304 if (sd_id128_is_null(*seqnum_id
))
2305 *seqnum_id
= f
->header
->seqnum_id
; /* Caller has none assigned, then copy the one from the file */
2306 else if (!sd_id128_equal(*seqnum_id
, f
->header
->seqnum_id
)) {
2307 /* Different seqnum IDs? We can't allow entries from multiple IDs end up in the same journal.*/
2308 if (le64toh(f
->header
->n_entries
) == 0)
2309 f
->header
->seqnum_id
= *seqnum_id
; /* Caller has one, and file so far has no entries, then copy the one from the caller */
2311 return log_debug_errno(SYNTHETIC_ERRNO(EILSEQ
),
2312 "Sequence number IDs don't match, refusing entry.");
2316 if (machine_id
&& sd_id128_is_null(f
->header
->machine_id
))
2317 /* Initialize machine ID when not set yet */
2318 f
->header
->machine_id
= *machine_id
;
2320 osize
= offsetof(Object
, entry
.items
) + (n_items
* journal_file_entry_item_size(f
));
2322 r
= journal_file_append_object(f
, OBJECT_ENTRY
, osize
, &o
, &np
);
2326 o
->entry
.seqnum
= htole64(journal_file_entry_seqnum(f
, seqnum
));
2327 o
->entry
.realtime
= htole64(ts
->realtime
);
2328 o
->entry
.monotonic
= htole64(ts
->monotonic
);
2329 o
->entry
.xor_hash
= htole64(xor_hash
);
2331 f
->header
->tail_entry_boot_id
= *boot_id
;
2332 o
->entry
.boot_id
= f
->header
->tail_entry_boot_id
;
2334 for (size_t i
= 0; i
< n_items
; i
++)
2335 write_entry_item(f
, o
, i
, &items
[i
]);
2338 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY
, o
, np
);
2343 r
= journal_file_link_entry(f
, o
, np
, items
, n_items
);
2356 void journal_file_post_change(JournalFile
*f
) {
2362 /* inotify() does not receive IN_MODIFY events from file
2363 * accesses done via mmap(). After each access we hence
2364 * trigger IN_MODIFY by truncating the journal file to its
2365 * current size which triggers IN_MODIFY. */
2367 __atomic_thread_fence(__ATOMIC_SEQ_CST
);
2369 if (ftruncate(f
->fd
, f
->last_stat
.st_size
) < 0)
2370 log_debug_errno(errno
, "Failed to truncate file to its own size: %m");
2373 static int post_change_thunk(sd_event_source
*timer
, uint64_t usec
, void *userdata
) {
2376 journal_file_post_change(userdata
);
2381 static void schedule_post_change(JournalFile
*f
) {
2386 assert(f
->post_change_timer
);
2388 assert_se(e
= sd_event_source_get_event(f
->post_change_timer
));
2390 /* If we are already going down, post the change immediately. */
2391 if (IN_SET(sd_event_get_state(e
), SD_EVENT_EXITING
, SD_EVENT_FINISHED
))
2394 r
= sd_event_source_get_enabled(f
->post_change_timer
, NULL
);
2396 log_debug_errno(r
, "Failed to get ftruncate timer state: %m");
2402 r
= sd_event_source_set_time_relative(f
->post_change_timer
, f
->post_change_timer_period
);
2404 log_debug_errno(r
, "Failed to set time for scheduling ftruncate: %m");
2408 r
= sd_event_source_set_enabled(f
->post_change_timer
, SD_EVENT_ONESHOT
);
2410 log_debug_errno(r
, "Failed to enable scheduled ftruncate: %m");
2417 /* On failure, let's simply post the change immediately. */
2418 journal_file_post_change(f
);
2421 /* Enable coalesced change posting in a timer on the provided sd_event instance */
2422 int journal_file_enable_post_change_timer(JournalFile
*f
, sd_event
*e
, usec_t t
) {
2423 _cleanup_(sd_event_source_unrefp
) sd_event_source
*timer
= NULL
;
2427 assert_return(!f
->post_change_timer
, -EINVAL
);
2431 r
= sd_event_add_time(e
, &timer
, CLOCK_MONOTONIC
, 0, 0, post_change_thunk
, f
);
2435 r
= sd_event_source_set_enabled(timer
, SD_EVENT_OFF
);
2439 f
->post_change_timer
= TAKE_PTR(timer
);
2440 f
->post_change_timer_period
= t
;
2445 static int entry_item_cmp(const EntryItem
*a
, const EntryItem
*b
) {
2446 return CMP(ASSERT_PTR(a
)->object_offset
, ASSERT_PTR(b
)->object_offset
);
2449 static size_t remove_duplicate_entry_items(EntryItem items
[], size_t n
) {
2452 assert(items
|| n
== 0);
2457 for (size_t i
= 1; i
< n
; i
++)
2458 if (items
[i
].object_offset
!= items
[j
- 1].object_offset
)
2459 items
[j
++] = items
[i
];
2464 int journal_file_append_entry(
2466 const dual_timestamp
*ts
,
2467 const sd_id128_t
*boot_id
,
2468 const struct iovec iovec
[],
2471 sd_id128_t
*seqnum_id
,
2472 Object
**ret_object
,
2473 uint64_t *ret_offset
) {
2475 _cleanup_free_ EntryItem
*items_alloc
= NULL
;
2477 uint64_t xor_hash
= 0;
2478 struct dual_timestamp _ts
;
2479 sd_id128_t _boot_id
, _machine_id
, *machine_id
;
2485 assert(n_iovec
> 0);
2488 if (!VALID_REALTIME(ts
->realtime
))
2489 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
2490 "Invalid realtime timestamp %" PRIu64
", refusing entry.",
2492 if (!VALID_MONOTONIC(ts
->monotonic
))
2493 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
2494 "Invalid monotomic timestamp %" PRIu64
", refusing entry.",
2497 dual_timestamp_get(&_ts
);
2502 r
= sd_id128_get_boot(&_boot_id
);
2506 boot_id
= &_boot_id
;
2509 r
= sd_id128_get_machine(&_machine_id
);
2510 if (ERRNO_IS_NEG_MACHINE_ID_UNSET(r
))
2511 /* Gracefully handle the machine ID not being initialized yet */
2516 machine_id
= &_machine_id
;
2519 r
= journal_file_maybe_append_tag(f
, ts
->realtime
);
2524 if (n_iovec
< ALLOCA_MAX
/ sizeof(EntryItem
) / 2)
2525 items
= newa(EntryItem
, n_iovec
);
2527 items_alloc
= new(EntryItem
, n_iovec
);
2531 items
= items_alloc
;
2534 for (size_t i
= 0; i
< n_iovec
; i
++) {
2538 r
= journal_file_append_data(f
, iovec
[i
].iov_base
, iovec
[i
].iov_len
, &o
, &p
);
2542 /* When calculating the XOR hash field, we need to take special care if the "keyed-hash"
2543 * journal file flag is on. We use the XOR hash field to quickly determine the identity of a
2544 * specific record, and give records with otherwise identical position (i.e. match in seqno,
2545 * timestamp, …) a stable ordering. But for that we can't have it that the hash of the
2546 * objects in each file is different since they are keyed. Hence let's calculate the Jenkins
2547 * hash here for that. This also has the benefit that cursors for old and new journal files
2548 * are completely identical (they include the XOR hash after all). For classic Jenkins-hash
2549 * files things are easier, we can just take the value from the stored record directly. */
2551 if (JOURNAL_HEADER_KEYED_HASH(f
->header
))
2552 xor_hash
^= jenkins_hash64(iovec
[i
].iov_base
, iovec
[i
].iov_len
);
2554 xor_hash
^= le64toh(o
->data
.hash
);
2556 items
[i
] = (EntryItem
) {
2558 .hash
= le64toh(o
->data
.hash
),
2562 /* Order by the position on disk, in order to improve seek
2563 * times for rotating media. */
2564 typesafe_qsort(items
, n_iovec
, entry_item_cmp
);
2565 n_iovec
= remove_duplicate_entry_items(items
, n_iovec
);
2567 r
= journal_file_append_entry_internal(
2580 /* If the memory mapping triggered a SIGBUS then we return an
2581 * IO error and ignore the error code passed down to us, since
2582 * it is very likely just an effect of a nullified replacement
2585 if (mmap_cache_fd_got_sigbus(f
->cache_fd
))
2588 if (f
->post_change_timer
)
2589 schedule_post_change(f
);
2591 journal_file_post_change(f
);
2596 typedef struct ChainCacheItem
{
2597 uint64_t first
; /* the array at the beginning of the chain */
2598 uint64_t array
; /* the cached array */
2599 uint64_t begin
; /* the first item in the cached array */
2600 uint64_t total
; /* the total number of items in all arrays before this one in the chain */
2601 uint64_t last_index
; /* the last index we looked at, to optimize locality when bisecting */
2604 static void chain_cache_put(
2611 uint64_t last_index
) {
2616 /* If the chain item to cache for this chain is the
2617 * first one it's not worth caching anything */
2621 if (ordered_hashmap_size(h
) >= CHAIN_CACHE_MAX
) {
2622 ci
= ordered_hashmap_steal_first(h
);
2625 ci
= new(ChainCacheItem
, 1);
2632 if (ordered_hashmap_put(h
, &ci
->first
, ci
) < 0) {
2637 assert(ci
->first
== first
);
2642 ci
->last_index
= last_index
;
2645 static int bump_array_index(uint64_t *i
, direction_t direction
, uint64_t n
) {
2648 /* Increase or decrease the specified index, in the right direction. */
2650 if (direction
== DIRECTION_DOWN
) {
2665 static int bump_entry_array(
2670 direction_t direction
,
2680 if (direction
== DIRECTION_DOWN
) {
2682 *ret
= le64toh(o
->entry_array
.next_entry_array_offset
);
2686 /* Entry array chains are a singly linked list, so to find the previous array in the chain, we have
2687 * to start iterating from the top. */
2691 while (p
> 0 && p
!= offset
) {
2692 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, p
, &o
);
2697 p
= le64toh(o
->entry_array
.next_entry_array_offset
);
2700 /* If we can't find the previous entry array in the entry array chain, we're likely dealing with a
2701 * corrupted journal file. */
2710 static int generic_array_get(
2714 direction_t direction
,
2715 Object
**ret_object
,
2716 uint64_t *ret_offset
) {
2718 uint64_t a
, t
= 0, k
;
2725 /* FIXME: fix return value assignment on success. */
2729 /* Try the chain cache first */
2730 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2731 if (ci
&& i
> ci
->total
) {
2738 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2739 if (IN_SET(r
, -EBADMSG
, -EADDRNOTAVAIL
)) {
2740 /* If there's corruption and we're going downwards, let's pretend we reached the
2741 * final entry in the entry array chain. */
2743 if (direction
== DIRECTION_DOWN
)
2746 /* If there's corruption and we're going upwards, move back to the previous entry
2747 * array and start iterating entries from there. */
2749 r
= bump_entry_array(f
, NULL
, a
, first
, DIRECTION_UP
, &a
);
2760 k
= journal_file_entry_array_n_items(f
, o
);
2766 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
2769 /* If we've found the right location, now look for the first non-corrupt entry object (in the right
2773 /* In the first iteration of the while loop, we reuse i, k and o from the previous while
2775 if (i
== UINT64_MAX
) {
2776 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2780 k
= journal_file_entry_array_n_items(f
, o
);
2784 i
= direction
== DIRECTION_DOWN
? 0 : k
- 1;
2790 p
= journal_file_entry_array_item(f
, o
, i
);
2792 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, ret_object
);
2794 /* Let's cache this item for the next invocation */
2795 chain_cache_put(f
->chain_cache
, ci
, first
, a
, journal_file_entry_array_item(f
, o
, 0), t
, i
);
2802 if (!IN_SET(r
, -EADDRNOTAVAIL
, -EBADMSG
))
2805 /* OK, so this entry is borked. Most likely some entry didn't get synced to
2806 * disk properly, let's see if the next one might work for us instead. */
2807 log_debug_errno(r
, "Entry item %" PRIu64
" is bad, skipping over it.", i
);
2809 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2813 } while (bump_array_index(&i
, direction
, k
) > 0);
2815 r
= bump_entry_array(f
, o
, a
, first
, direction
, &a
);
2826 static int generic_array_get_plus_one(
2831 direction_t direction
,
2832 Object
**ret_object
,
2833 uint64_t *ret_offset
) {
2839 /* FIXME: fix return value assignment on success. */
2842 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, ret_object
);
2843 if (IN_SET(r
, -EADDRNOTAVAIL
, -EBADMSG
))
2844 return generic_array_get(f
, first
, 0, direction
, ret_object
, ret_offset
);
2849 *ret_offset
= extra
;
2854 return generic_array_get(f
, first
, i
- 1, direction
, ret_object
, ret_offset
);
2863 static int generic_array_bisect_one(
2865 uint64_t a
, /* offset of entry array object. */
2866 uint64_t i
, /* index of the entry item we will test. */
2868 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2869 direction_t direction
,
2872 uint64_t *ret_offset
) {
2879 assert(test_object
);
2883 assert(i
<= *right
);
2885 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2889 p
= journal_file_entry_array_item(f
, array
, i
);
2893 r
= test_object(f
, p
, needle
);
2894 if (IN_SET(r
, -EBADMSG
, -EADDRNOTAVAIL
)) {
2895 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short.");
2897 return -ENOANO
; /* recognizable error */
2902 if (r
== TEST_FOUND
)
2903 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2905 if (r
== TEST_RIGHT
)
2916 static int generic_array_bisect(
2921 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2922 direction_t direction
,
2923 Object
**ret_object
,
2924 uint64_t *ret_offset
,
2925 uint64_t *ret_idx
) {
2927 /* Given an entry array chain, this function finds the object "closest" to the given needle in the
2928 * chain, taking into account the provided direction. A function can be provided to determine how
2929 * an object is matched against the given needle.
2931 * Given a journal file, the offset of an object and the needle, the test_object() function should
2932 * return TEST_LEFT if the needle is located earlier in the entry array chain, TEST_LEFT if the
2933 * needle is located later in the entry array chain and TEST_FOUND if the object matches the needle.
2934 * If test_object() returns TEST_FOUND for a specific object, that object's information will be used
2935 * to populate the return values of this function. If test_object() never returns TEST_FOUND, the
2936 * return values are populated with the details of one of the objects closest to the needle. If the
2937 * direction is DIRECTION_UP, the earlier object is used. Otherwise, the later object is used.
2940 uint64_t a
, p
, t
= 0, i
= 0, last_p
= 0, last_index
= UINT64_MAX
;
2941 bool subtract_one
= false;
2947 assert(test_object
);
2949 /* Start with the first array in the chain */
2952 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2953 if (ci
&& n
> ci
->total
&& ci
->begin
!= 0) {
2954 /* Ah, we have iterated this bisection array chain previously! Let's see if we can skip ahead
2955 * in the chain, as far as the last time. But we can't jump backwards in the chain, so let's
2956 * check that first. */
2958 r
= test_object(f
, ci
->begin
, needle
);
2962 if (r
== TEST_LEFT
) {
2963 /* OK, what we are looking for is right of the begin of this EntryArray, so let's
2964 * jump straight to previously cached array in the chain */
2969 last_index
= ci
->last_index
;
2974 uint64_t left
= 0, right
, k
, lp
;
2976 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2980 k
= journal_file_entry_array_n_items(f
, array
);
2986 r
= generic_array_bisect_one(f
, a
, right
, needle
, test_object
, direction
, &left
, &right
, &lp
);
2994 if (r
== TEST_RIGHT
) {
2995 /* If we cached the last index we looked at, let's try to not to jump too wildly
2996 * around and see if we can limit the range to look at early to the immediate
2997 * neighbors of the last index we looked at. */
2999 if (last_index
> 0 && last_index
- 1 < right
) {
3000 r
= generic_array_bisect_one(f
, a
, last_index
- 1, needle
, test_object
, direction
, &left
, &right
, NULL
);
3001 if (r
< 0 && r
!= -ENOANO
)
3005 if (last_index
< right
) {
3006 r
= generic_array_bisect_one(f
, a
, last_index
+ 1, needle
, test_object
, direction
, &left
, &right
, NULL
);
3007 if (r
< 0 && r
!= -ENOANO
)
3012 if (left
== right
) {
3013 if (direction
== DIRECTION_UP
)
3014 subtract_one
= true;
3020 assert(left
< right
);
3021 i
= (left
+ right
) / 2;
3023 r
= generic_array_bisect_one(f
, a
, i
, needle
, test_object
, direction
, &left
, &right
, NULL
);
3024 if (r
< 0 && r
!= -ENOANO
)
3030 if (direction
== DIRECTION_UP
) {
3032 subtract_one
= true;
3043 last_index
= UINT64_MAX
;
3044 a
= le64toh(array
->entry_array
.next_entry_array_offset
);
3050 if (subtract_one
&& t
== 0 && i
== 0)
3053 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
3057 p
= journal_file_entry_array_item(f
, array
, 0);
3061 /* Let's cache this item for the next invocation */
3062 chain_cache_put(f
->chain_cache
, ci
, first
, a
, p
, t
, subtract_one
? (i
> 0 ? i
-1 : UINT64_MAX
) : i
);
3064 if (subtract_one
&& i
== 0)
3066 else if (subtract_one
)
3067 p
= journal_file_entry_array_item(f
, array
, i
- 1);
3069 p
= journal_file_entry_array_item(f
, array
, i
);
3072 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, ret_object
);
3081 *ret_idx
= t
+ i
+ (subtract_one
? -1 : 0);
3086 static int generic_array_bisect_plus_one(
3092 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
3093 direction_t direction
,
3094 Object
**ret_object
,
3095 uint64_t *ret_offset
,
3096 uint64_t *ret_idx
) {
3099 bool step_back
= false;
3102 assert(test_object
);
3107 /* This bisects the array in object 'first', but first checks
3109 r
= test_object(f
, extra
, needle
);
3113 if (r
== TEST_FOUND
)
3114 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
3116 /* if we are looking with DIRECTION_UP then we need to first
3117 see if in the actual array there is a matching entry, and
3118 return the last one of that. But if there isn't any we need
3119 to return this one. Hence remember this, and return it
3122 step_back
= direction
== DIRECTION_UP
;
3124 if (r
== TEST_RIGHT
) {
3125 if (direction
== DIRECTION_DOWN
)
3131 r
= generic_array_bisect(f
, first
, n
-1, needle
, test_object
, direction
, ret_object
, ret_offset
, ret_idx
);
3133 if (r
== 0 && step_back
)
3136 if (r
> 0 && ret_idx
)
3143 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, ret_object
);
3149 *ret_offset
= extra
;
3157 static int test_object_offset(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
3163 else if (p
< needle
)
3169 int journal_file_move_to_entry_by_offset(
3172 direction_t direction
,
3173 Object
**ret_object
,
3174 uint64_t *ret_offset
) {
3179 return generic_array_bisect(
3181 le64toh(f
->header
->entry_array_offset
),
3182 le64toh(f
->header
->n_entries
),
3186 ret_object
, ret_offset
, NULL
);
3189 static int test_object_seqnum(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
3197 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
3201 sq
= le64toh(READ_NOW(o
->entry
.seqnum
));
3204 else if (sq
< needle
)
3210 int journal_file_move_to_entry_by_seqnum(
3213 direction_t direction
,
3214 Object
**ret_object
,
3215 uint64_t *ret_offset
) {
3220 return generic_array_bisect(
3222 le64toh(f
->header
->entry_array_offset
),
3223 le64toh(f
->header
->n_entries
),
3227 ret_object
, ret_offset
, NULL
);
3230 static int test_object_realtime(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
3238 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
3242 rt
= le64toh(READ_NOW(o
->entry
.realtime
));
3245 else if (rt
< needle
)
3251 int journal_file_move_to_entry_by_realtime(
3254 direction_t direction
,
3255 Object
**ret_object
,
3256 uint64_t *ret_offset
) {
3261 return generic_array_bisect(
3263 le64toh(f
->header
->entry_array_offset
),
3264 le64toh(f
->header
->n_entries
),
3266 test_object_realtime
,
3268 ret_object
, ret_offset
, NULL
);
3271 static int test_object_monotonic(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
3279 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
3283 m
= le64toh(READ_NOW(o
->entry
.monotonic
));
3286 else if (m
< needle
)
3292 static int find_data_object_by_boot_id(
3295 Object
**ret_object
,
3296 uint64_t *ret_offset
) {
3298 char t
[STRLEN("_BOOT_ID=") + 32 + 1] = "_BOOT_ID=";
3302 sd_id128_to_string(boot_id
, t
+ 9);
3303 return journal_file_find_data_object(f
, t
, sizeof(t
) - 1, ret_object
, ret_offset
);
3306 int journal_file_move_to_entry_by_monotonic(
3310 direction_t direction
,
3311 Object
**ret_object
,
3312 uint64_t *ret_offset
) {
3319 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, NULL
);
3325 return generic_array_bisect_plus_one(
3327 le64toh(o
->data
.entry_offset
),
3328 le64toh(o
->data
.entry_array_offset
),
3329 le64toh(o
->data
.n_entries
),
3331 test_object_monotonic
,
3333 ret_object
, ret_offset
, NULL
);
3336 void journal_file_reset_location(JournalFile
*f
) {
3339 f
->location_type
= LOCATION_HEAD
;
3340 f
->current_offset
= 0;
3341 f
->current_seqnum
= 0;
3342 f
->current_realtime
= 0;
3343 f
->current_monotonic
= 0;
3344 zero(f
->current_boot_id
);
3345 f
->current_xor_hash
= 0;
3348 void journal_file_save_location(JournalFile
*f
, Object
*o
, uint64_t offset
) {
3352 f
->location_type
= LOCATION_SEEK
;
3353 f
->current_offset
= offset
;
3354 f
->current_seqnum
= le64toh(o
->entry
.seqnum
);
3355 f
->current_realtime
= le64toh(o
->entry
.realtime
);
3356 f
->current_monotonic
= le64toh(o
->entry
.monotonic
);
3357 f
->current_boot_id
= o
->entry
.boot_id
;
3358 f
->current_xor_hash
= le64toh(o
->entry
.xor_hash
);
3361 static bool check_properly_ordered(uint64_t new_offset
, uint64_t old_offset
, direction_t direction
) {
3363 /* Consider it an error if any of the two offsets is uninitialized */
3364 if (old_offset
== 0 || new_offset
== 0)
3367 /* If we go down, the new offset must be larger than the old one. */
3368 return direction
== DIRECTION_DOWN
?
3369 new_offset
> old_offset
:
3370 new_offset
< old_offset
;
3373 int journal_file_next_entry(
3376 direction_t direction
,
3377 Object
**ret_object
,
3378 uint64_t *ret_offset
) {
3386 /* FIXME: fix return value assignment. */
3388 n
= le64toh(READ_NOW(f
->header
->n_entries
));
3393 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
3395 r
= generic_array_bisect(f
,
3396 le64toh(f
->header
->entry_array_offset
),
3397 le64toh(f
->header
->n_entries
),
3406 r
= bump_array_index(&i
, direction
, n
);
3411 /* And jump to it */
3412 r
= generic_array_get(f
, le64toh(f
->header
->entry_array_offset
), i
, direction
, ret_object
, &ofs
);
3416 /* Ensure our array is properly ordered. */
3417 if (p
> 0 && !check_properly_ordered(ofs
, p
, direction
))
3418 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
3419 "%s: entry array not properly ordered at entry %" PRIu64
,
3428 int journal_file_next_entry_for_data(
3431 direction_t direction
,
3432 Object
**ret_object
,
3433 uint64_t *ret_offset
) {
3440 assert(d
->object
.type
== OBJECT_DATA
);
3442 /* FIXME: fix return value assignment. */
3444 n
= le64toh(READ_NOW(d
->data
.n_entries
));
3448 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
3450 r
= generic_array_get_plus_one(f
,
3451 le64toh(d
->data
.entry_offset
),
3452 le64toh(d
->data
.entry_array_offset
),
3465 int journal_file_move_to_entry_by_offset_for_data(
3469 direction_t direction
,
3470 Object
**ret
, uint64_t *ret_offset
) {
3474 assert(d
->object
.type
== OBJECT_DATA
);
3476 return generic_array_bisect_plus_one(
3478 le64toh(d
->data
.entry_offset
),
3479 le64toh(d
->data
.entry_array_offset
),
3480 le64toh(d
->data
.n_entries
),
3484 ret
, ret_offset
, NULL
);
3487 int journal_file_move_to_entry_by_monotonic_for_data(
3492 direction_t direction
,
3493 Object
**ret_object
,
3494 uint64_t *ret_offset
) {
3496 uint64_t b
, z
, entry_offset
, entry_array_offset
, n_entries
;
3502 assert(d
->object
.type
== OBJECT_DATA
);
3504 /* Save all the required data before the data object gets invalidated. */
3505 entry_offset
= le64toh(READ_NOW(d
->data
.entry_offset
));
3506 entry_array_offset
= le64toh(READ_NOW(d
->data
.entry_array_offset
));
3507 n_entries
= le64toh(READ_NOW(d
->data
.n_entries
));
3509 /* First, seek by time */
3510 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &b
);
3516 r
= generic_array_bisect_plus_one(f
,
3517 le64toh(o
->data
.entry_offset
),
3518 le64toh(o
->data
.entry_array_offset
),
3519 le64toh(o
->data
.n_entries
),
3521 test_object_monotonic
,
3527 /* And now, continue seeking until we find an entry that
3528 * exists in both bisection arrays */
3530 r
= journal_file_move_to_object(f
, OBJECT_DATA
, b
, &o
);
3537 r
= generic_array_bisect_plus_one(f
,
3548 r
= generic_array_bisect_plus_one(f
,
3549 le64toh(o
->data
.entry_offset
),
3550 le64toh(o
->data
.entry_array_offset
),
3551 le64toh(o
->data
.n_entries
),
3562 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, q
, ret_object
);
3577 int journal_file_move_to_entry_by_seqnum_for_data(
3581 direction_t direction
,
3582 Object
**ret_object
,
3583 uint64_t *ret_offset
) {
3587 assert(d
->object
.type
== OBJECT_DATA
);
3589 return generic_array_bisect_plus_one(
3591 le64toh(d
->data
.entry_offset
),
3592 le64toh(d
->data
.entry_array_offset
),
3593 le64toh(d
->data
.n_entries
),
3597 ret_object
, ret_offset
, NULL
);
3600 int journal_file_move_to_entry_by_realtime_for_data(
3604 direction_t direction
,
3605 Object
**ret
, uint64_t *ret_offset
) {
3609 assert(d
->object
.type
== OBJECT_DATA
);
3611 return generic_array_bisect_plus_one(
3613 le64toh(d
->data
.entry_offset
),
3614 le64toh(d
->data
.entry_array_offset
),
3615 le64toh(d
->data
.n_entries
),
3617 test_object_realtime
,
3619 ret
, ret_offset
, NULL
);
3622 void journal_file_dump(JournalFile
*f
) {
3630 journal_file_print_header(f
);
3632 p
= le64toh(READ_NOW(f
->header
->header_size
));
3637 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &o
);
3641 s
= journal_object_type_to_string(o
->object
.type
);
3643 switch (o
->object
.type
) {
3648 printf("Type: %s seqnum=%"PRIu64
" monotonic=%"PRIu64
" realtime=%"PRIu64
"\n",
3650 le64toh(o
->entry
.seqnum
),
3651 le64toh(o
->entry
.monotonic
),
3652 le64toh(o
->entry
.realtime
));
3658 printf("Type: %s seqnum=%"PRIu64
" epoch=%"PRIu64
"\n",
3660 le64toh(o
->tag
.seqnum
),
3661 le64toh(o
->tag
.epoch
));
3666 printf("Type: %s \n", s
);
3668 printf("Type: unknown (%i)", o
->object
.type
);
3673 c
= COMPRESSION_FROM_OBJECT(o
);
3674 if (c
> COMPRESSION_NONE
)
3675 printf("Flags: %s\n",
3676 compression_to_string(c
));
3678 if (p
== le64toh(f
->header
->tail_object_offset
))
3681 p
+= ALIGN64(le64toh(o
->object
.size
));
3686 log_error("File corrupt");
3689 /* Note: the lifetime of the compound literal is the immediately surrounding block. */
3690 #define FORMAT_TIMESTAMP_SAFE(t) (FORMAT_TIMESTAMP(t) ?: " --- ")
3692 void journal_file_print_header(JournalFile
*f
) {
3698 printf("File path: %s\n"
3702 "Sequential number ID: %s\n"
3704 "Compatible flags:%s%s%s\n"
3705 "Incompatible flags:%s%s%s%s%s%s\n"
3706 "Header size: %"PRIu64
"\n"
3707 "Arena size: %"PRIu64
"\n"
3708 "Data hash table size: %"PRIu64
"\n"
3709 "Field hash table size: %"PRIu64
"\n"
3710 "Rotate suggested: %s\n"
3711 "Head sequential number: %"PRIu64
" (%"PRIx64
")\n"
3712 "Tail sequential number: %"PRIu64
" (%"PRIx64
")\n"
3713 "Head realtime timestamp: %s (%"PRIx64
")\n"
3714 "Tail realtime timestamp: %s (%"PRIx64
")\n"
3715 "Tail monotonic timestamp: %s (%"PRIx64
")\n"
3716 "Objects: %"PRIu64
"\n"
3717 "Entry objects: %"PRIu64
"\n",
3719 SD_ID128_TO_STRING(f
->header
->file_id
),
3720 SD_ID128_TO_STRING(f
->header
->machine_id
),
3721 SD_ID128_TO_STRING(f
->header
->tail_entry_boot_id
),
3722 SD_ID128_TO_STRING(f
->header
->seqnum_id
),
3723 f
->header
->state
== STATE_OFFLINE
? "OFFLINE" :
3724 f
->header
->state
== STATE_ONLINE
? "ONLINE" :
3725 f
->header
->state
== STATE_ARCHIVED
? "ARCHIVED" : "UNKNOWN",
3726 JOURNAL_HEADER_SEALED(f
->header
) ? " SEALED" : "",
3727 JOURNAL_HEADER_TAIL_ENTRY_BOOT_ID(f
->header
) ? " TAIL_ENTRY_BOOT_ID" : "",
3728 (le32toh(f
->header
->compatible_flags
) & ~HEADER_COMPATIBLE_ANY
) ? " ???" : "",
3729 JOURNAL_HEADER_COMPRESSED_XZ(f
->header
) ? " COMPRESSED-XZ" : "",
3730 JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
) ? " COMPRESSED-LZ4" : "",
3731 JOURNAL_HEADER_COMPRESSED_ZSTD(f
->header
) ? " COMPRESSED-ZSTD" : "",
3732 JOURNAL_HEADER_KEYED_HASH(f
->header
) ? " KEYED-HASH" : "",
3733 JOURNAL_HEADER_COMPACT(f
->header
) ? " COMPACT" : "",
3734 (le32toh(f
->header
->incompatible_flags
) & ~HEADER_INCOMPATIBLE_ANY
) ? " ???" : "",
3735 le64toh(f
->header
->header_size
),
3736 le64toh(f
->header
->arena_size
),
3737 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3738 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
),
3739 yes_no(journal_file_rotate_suggested(f
, 0, LOG_DEBUG
)),
3740 le64toh(f
->header
->head_entry_seqnum
), le64toh(f
->header
->head_entry_seqnum
),
3741 le64toh(f
->header
->tail_entry_seqnum
), le64toh(f
->header
->tail_entry_seqnum
),
3742 FORMAT_TIMESTAMP_SAFE(le64toh(f
->header
->head_entry_realtime
)), le64toh(f
->header
->head_entry_realtime
),
3743 FORMAT_TIMESTAMP_SAFE(le64toh(f
->header
->tail_entry_realtime
)), le64toh(f
->header
->tail_entry_realtime
),
3744 FORMAT_TIMESPAN(le64toh(f
->header
->tail_entry_monotonic
), USEC_PER_MSEC
), le64toh(f
->header
->tail_entry_monotonic
),
3745 le64toh(f
->header
->n_objects
),
3746 le64toh(f
->header
->n_entries
));
3748 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3749 printf("Data objects: %"PRIu64
"\n"
3750 "Data hash table fill: %.1f%%\n",
3751 le64toh(f
->header
->n_data
),
3752 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))));
3754 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3755 printf("Field objects: %"PRIu64
"\n"
3756 "Field hash table fill: %.1f%%\n",
3757 le64toh(f
->header
->n_fields
),
3758 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))));
3760 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
))
3761 printf("Tag objects: %"PRIu64
"\n",
3762 le64toh(f
->header
->n_tags
));
3763 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
3764 printf("Entry array objects: %"PRIu64
"\n",
3765 le64toh(f
->header
->n_entry_arrays
));
3767 if (JOURNAL_HEADER_CONTAINS(f
->header
, field_hash_chain_depth
))
3768 printf("Deepest field hash chain: %" PRIu64
"\n",
3769 f
->header
->field_hash_chain_depth
);
3771 if (JOURNAL_HEADER_CONTAINS(f
->header
, data_hash_chain_depth
))
3772 printf("Deepest data hash chain: %" PRIu64
"\n",
3773 f
->header
->data_hash_chain_depth
);
3775 if (fstat(f
->fd
, &st
) >= 0)
3776 printf("Disk usage: %s\n", FORMAT_BYTES((uint64_t) st
.st_blocks
* 512ULL));
3779 static int journal_file_warn_btrfs(JournalFile
*f
) {
3785 /* Before we write anything, check if the COW logic is turned
3786 * off on btrfs. Given our write pattern that is quite
3787 * unfriendly to COW file systems this should greatly improve
3788 * performance on COW file systems, such as btrfs, at the
3789 * expense of data integrity features (which shouldn't be too
3790 * bad, given that we do our own checksumming). */
3792 r
= fd_is_fs_type(f
->fd
, BTRFS_SUPER_MAGIC
);
3794 return log_ratelimit_warning_errno(r
, JOURNAL_LOG_RATELIMIT
, "Failed to determine if journal is on btrfs: %m");
3798 r
= read_attr_fd(f
->fd
, &attrs
);
3800 return log_ratelimit_warning_errno(r
, JOURNAL_LOG_RATELIMIT
, "Failed to read file attributes: %m");
3802 if (attrs
& FS_NOCOW_FL
) {
3803 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3807 log_ratelimit_notice(JOURNAL_LOG_RATELIMIT
,
3808 "Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3809 "This is likely to slow down journal access substantially, please consider turning "
3810 "off the copy-on-write file attribute on the journal directory, using chattr +C.",
3816 static void journal_default_metrics(JournalMetrics
*m
, int fd
, bool compact
) {
3818 uint64_t fs_size
= 0;
3823 if (fstatvfs(fd
, &ss
) >= 0)
3824 fs_size
= ss
.f_frsize
* ss
.f_blocks
;
3826 log_debug_errno(errno
, "Failed to determine disk size: %m");
3828 if (m
->max_use
== UINT64_MAX
) {
3831 m
->max_use
= CLAMP(PAGE_ALIGN(fs_size
/ 10), /* 10% of file system size */
3832 MAX_USE_LOWER
, MAX_USE_UPPER
);
3834 m
->max_use
= MAX_USE_LOWER
;
3836 m
->max_use
= PAGE_ALIGN(m
->max_use
);
3838 if (m
->max_use
!= 0 && m
->max_use
< JOURNAL_FILE_SIZE_MIN
*2)
3839 m
->max_use
= JOURNAL_FILE_SIZE_MIN
*2;
3842 if (m
->min_use
== UINT64_MAX
) {
3844 m
->min_use
= CLAMP(PAGE_ALIGN(fs_size
/ 50), /* 2% of file system size */
3845 MIN_USE_LOW
, MIN_USE_HIGH
);
3847 m
->min_use
= MIN_USE_LOW
;
3850 if (m
->min_use
> m
->max_use
)
3851 m
->min_use
= m
->max_use
;
3853 if (m
->max_size
== UINT64_MAX
)
3854 m
->max_size
= MIN(PAGE_ALIGN(m
->max_use
/ 8), /* 8 chunks */
3857 m
->max_size
= PAGE_ALIGN(m
->max_size
);
3859 if (compact
&& m
->max_size
> JOURNAL_COMPACT_SIZE_MAX
)
3860 m
->max_size
= JOURNAL_COMPACT_SIZE_MAX
;
3862 if (m
->max_size
!= 0) {
3863 if (m
->max_size
< JOURNAL_FILE_SIZE_MIN
)
3864 m
->max_size
= JOURNAL_FILE_SIZE_MIN
;
3866 if (m
->max_use
!= 0 && m
->max_size
*2 > m
->max_use
)
3867 m
->max_use
= m
->max_size
*2;
3870 if (m
->min_size
== UINT64_MAX
)
3871 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3873 m
->min_size
= CLAMP(PAGE_ALIGN(m
->min_size
),
3874 JOURNAL_FILE_SIZE_MIN
,
3875 m
->max_size
?: UINT64_MAX
);
3877 if (m
->keep_free
== UINT64_MAX
) {
3879 m
->keep_free
= MIN(PAGE_ALIGN(fs_size
/ 20), /* 5% of file system size */
3882 m
->keep_free
= DEFAULT_KEEP_FREE
;
3885 if (m
->n_max_files
== UINT64_MAX
)
3886 m
->n_max_files
= DEFAULT_N_MAX_FILES
;
3888 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64
,
3889 FORMAT_BYTES(m
->min_use
),
3890 FORMAT_BYTES(m
->max_use
),
3891 FORMAT_BYTES(m
->max_size
),
3892 FORMAT_BYTES(m
->min_size
),
3893 FORMAT_BYTES(m
->keep_free
),
3897 int journal_file_open(
3901 JournalFileFlags file_flags
,
3903 uint64_t compress_threshold_bytes
,
3904 JournalMetrics
*metrics
,
3905 MMapCache
*mmap_cache
,
3906 JournalFile
*template,
3907 JournalFile
**ret
) {
3909 bool newly_created
= false;
3914 assert(fd
>= 0 || fname
);
3915 assert(file_flags
>= 0);
3916 assert(file_flags
<= _JOURNAL_FILE_FLAGS_MAX
);
3920 if (!IN_SET((open_flags
& O_ACCMODE
), O_RDONLY
, O_RDWR
))
3923 if ((open_flags
& O_ACCMODE
) == O_RDONLY
&& FLAGS_SET(open_flags
, O_CREAT
))
3926 if (fname
&& (open_flags
& O_CREAT
) && !endswith(fname
, ".journal"))
3929 f
= new(JournalFile
, 1);
3933 *f
= (JournalFile
) {
3936 .open_flags
= open_flags
,
3937 .compress_threshold_bytes
= compress_threshold_bytes
== UINT64_MAX
?
3938 DEFAULT_COMPRESS_THRESHOLD
:
3939 MAX(MIN_COMPRESS_THRESHOLD
, compress_threshold_bytes
),
3940 .strict_order
= FLAGS_SET(file_flags
, JOURNAL_STRICT_ORDER
),
3941 .newest_boot_id_prioq_idx
= PRIOQ_IDX_NULL
,
3945 f
->path
= strdup(fname
);
3953 /* If we don't know the path, fill in something explanatory and vaguely useful */
3954 if (asprintf(&f
->path
, "/proc/self/%i", fd
) < 0) {
3960 f
->chain_cache
= ordered_hashmap_new(&uint64_hash_ops
);
3961 if (!f
->chain_cache
) {
3967 /* We pass O_NONBLOCK here, so that in case somebody pointed us to some character device node or FIFO
3968 * or so, we likely fail quickly than block for long. For regular files O_NONBLOCK has no effect, hence
3969 * it doesn't hurt in that case. */
3971 f
->fd
= openat_report_new(AT_FDCWD
, f
->path
, f
->open_flags
|O_CLOEXEC
|O_NONBLOCK
, f
->mode
, &newly_created
);
3977 /* fds we opened here by us should also be closed by us. */
3980 r
= fd_nonblock(f
->fd
, false);
3984 if (!newly_created
) {
3985 r
= journal_file_fstat(f
);
3990 r
= journal_file_fstat(f
);
3994 /* If we just got the fd passed in, we don't really know if we created the file anew */
3995 newly_created
= f
->last_stat
.st_size
== 0 && journal_file_writable(f
);
3998 f
->cache_fd
= mmap_cache_add_fd(mmap_cache
, f
->fd
, mmap_prot_from_open_flags(open_flags
));
4004 if (newly_created
) {
4005 (void) journal_file_warn_btrfs(f
);
4007 /* Let's attach the creation time to the journal file, so that the vacuuming code knows the age of this
4008 * file even if the file might end up corrupted one day... Ideally we'd just use the creation time many
4009 * file systems maintain for each file, but the API to query this is very new, hence let's emulate this
4010 * via extended attributes. If extended attributes are not supported we'll just skip this, and rely
4011 * solely on mtime/atime/ctime of the file. */
4012 (void) fd_setcrtime(f
->fd
, 0);
4014 r
= journal_file_init_header(f
, file_flags
, template);
4018 r
= journal_file_fstat(f
);
4023 if (f
->last_stat
.st_size
< (off_t
) HEADER_SIZE_MIN
) {
4028 r
= mmap_cache_fd_get(f
->cache_fd
, CONTEXT_HEADER
, true, 0, PAGE_ALIGN(sizeof(Header
)), &f
->last_stat
, &h
);
4030 /* Some file systems (jffs2 or p9fs) don't support mmap() properly (or only read-only
4031 * mmap()), and return EINVAL in that case. Let's propagate that as a more recognizable error
4041 if (!newly_created
) {
4042 r
= journal_file_verify_header(f
);
4048 if (!newly_created
&& journal_file_writable(f
) && JOURNAL_HEADER_SEALED(f
->header
)) {
4049 r
= journal_file_fss_load(f
);
4055 if (journal_file_writable(f
)) {
4057 journal_default_metrics(metrics
, f
->fd
, JOURNAL_HEADER_COMPACT(f
->header
));
4058 f
->metrics
= *metrics
;
4059 } else if (template)
4060 f
->metrics
= template->metrics
;
4062 r
= journal_file_refresh_header(f
);
4068 r
= journal_file_hmac_setup(f
);
4073 if (newly_created
) {
4074 r
= journal_file_setup_field_hash_table(f
);
4078 r
= journal_file_setup_data_hash_table(f
);
4083 r
= journal_file_append_first_tag(f
);
4089 if (mmap_cache_fd_got_sigbus(f
->cache_fd
)) {
4094 if (template && template->post_change_timer
) {
4095 r
= journal_file_enable_post_change_timer(
4097 sd_event_source_get_event(template->post_change_timer
),
4098 template->post_change_timer_period
);
4104 /* The file is opened now successfully, thus we take possession of any passed in fd. */
4107 if (DEBUG_LOGGING
) {
4108 static int last_seal
= -1, last_keyed_hash
= -1;
4109 static Compression last_compression
= _COMPRESSION_INVALID
;
4110 static uint64_t last_bytes
= UINT64_MAX
;
4112 if (last_seal
!= JOURNAL_HEADER_SEALED(f
->header
) ||
4113 last_keyed_hash
!= JOURNAL_HEADER_KEYED_HASH(f
->header
) ||
4114 last_compression
!= JOURNAL_FILE_COMPRESSION(f
) ||
4115 last_bytes
!= f
->compress_threshold_bytes
) {
4117 log_debug("Journal effective settings seal=%s keyed_hash=%s compress=%s compress_threshold_bytes=%s",
4118 yes_no(JOURNAL_HEADER_SEALED(f
->header
)), yes_no(JOURNAL_HEADER_KEYED_HASH(f
->header
)),
4119 compression_to_string(JOURNAL_FILE_COMPRESSION(f
)), FORMAT_BYTES(f
->compress_threshold_bytes
));
4120 last_seal
= JOURNAL_HEADER_SEALED(f
->header
);
4121 last_keyed_hash
= JOURNAL_HEADER_KEYED_HASH(f
->header
);
4122 last_compression
= JOURNAL_FILE_COMPRESSION(f
);
4123 last_bytes
= f
->compress_threshold_bytes
;
4131 if (f
->cache_fd
&& mmap_cache_fd_got_sigbus(f
->cache_fd
))
4134 (void) journal_file_close(f
);
4136 if (newly_created
&& fd
< 0)
4137 (void) unlink(fname
);
4142 int journal_file_parse_uid_from_filename(const char *path
, uid_t
*ret_uid
) {
4143 _cleanup_free_
char *buf
= NULL
, *p
= NULL
;
4144 const char *a
, *b
, *at
;
4147 /* This helper returns -EREMOTE when the filename doesn't match user online/offline journal
4148 * pattern. Hence it currently doesn't parse archived or disposed user journals. */
4153 r
= path_extract_filename(path
, &p
);
4156 if (r
== O_DIRECTORY
)
4159 a
= startswith(p
, "user-");
4162 b
= endswith(p
, ".journal");
4166 at
= strchr(a
, '@');
4170 buf
= strndup(a
, b
-a
);
4174 return parse_uid(buf
, ret_uid
);
4177 int journal_file_archive(JournalFile
*f
, char **ret_previous_path
) {
4178 _cleanup_free_
char *p
= NULL
;
4182 if (!journal_file_writable(f
))
4185 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
4186 * rotation, since we don't know the actual path, and couldn't rename the file hence. */
4187 if (path_startswith(f
->path
, "/proc/self/fd"))
4190 if (!endswith(f
->path
, ".journal"))
4193 if (asprintf(&p
, "%.*s@" SD_ID128_FORMAT_STR
"-%016"PRIx64
"-%016"PRIx64
".journal",
4194 (int) strlen(f
->path
) - 8, f
->path
,
4195 SD_ID128_FORMAT_VAL(f
->header
->seqnum_id
),
4196 le64toh(f
->header
->head_entry_seqnum
),
4197 le64toh(f
->header
->head_entry_realtime
)) < 0)
4200 /* Try to rename the file to the archived version. If the file already was deleted, we'll get ENOENT, let's
4201 * ignore that case. */
4202 if (rename(f
->path
, p
) < 0 && errno
!= ENOENT
)
4205 /* Sync the rename to disk */
4206 (void) fsync_directory_of_file(f
->fd
);
4208 if (ret_previous_path
)
4209 *ret_previous_path
= f
->path
;
4213 f
->path
= TAKE_PTR(p
);
4215 /* Set as archive so offlining commits w/state=STATE_ARCHIVED. Previously we would set old_file->header->state
4216 * to STATE_ARCHIVED directly here, but journal_file_set_offline() short-circuits when state != STATE_ONLINE,
4217 * which would result in the rotated journal never getting fsync() called before closing. Now we simply queue
4218 * the archive state by setting an archive bit, leaving the state as STATE_ONLINE so proper offlining
4225 int journal_file_dispose(int dir_fd
, const char *fname
) {
4226 _cleanup_free_
char *p
= NULL
;
4230 /* Renames a journal file to *.journal~, i.e. to mark it as corrupted or otherwise uncleanly shutdown. Note that
4231 * this is done without looking into the file or changing any of its contents. The idea is that this is called
4232 * whenever something is suspicious and we want to move the file away and make clear that it is not accessed
4233 * for writing anymore. */
4235 if (!endswith(fname
, ".journal"))
4238 if (asprintf(&p
, "%.*s@%016" PRIx64
"-%016" PRIx64
".journal~",
4239 (int) strlen(fname
) - 8, fname
,
4240 now(CLOCK_REALTIME
),
4244 if (renameat(dir_fd
, fname
, dir_fd
, p
) < 0)
4250 int journal_file_copy_entry(
4256 sd_id128_t
*seqnum_id
) {
4258 _cleanup_free_ EntryItem
*items_alloc
= NULL
;
4260 uint64_t n
, m
= 0, xor_hash
= 0;
4270 if (!journal_file_writable(to
))
4273 ts
= (dual_timestamp
) {
4274 .monotonic
= le64toh(o
->entry
.monotonic
),
4275 .realtime
= le64toh(o
->entry
.realtime
),
4277 boot_id
= o
->entry
.boot_id
;
4279 n
= journal_file_entry_n_items(from
, o
);
4283 if (n
< ALLOCA_MAX
/ sizeof(EntryItem
) / 2)
4284 items
= newa(EntryItem
, n
);
4286 items_alloc
= new(EntryItem
, n
);
4290 items
= items_alloc
;
4293 for (uint64_t i
= 0; i
< n
; i
++) {
4299 q
= journal_file_entry_item_object_offset(from
, o
, i
);
4300 r
= journal_file_data_payload(from
, NULL
, q
, NULL
, 0, 0, &data
, &l
);
4301 if (IN_SET(r
, -EADDRNOTAVAIL
, -EBADMSG
)) {
4302 log_debug_errno(r
, "Entry item %"PRIu64
" data object is bad, skipping over it: %m", i
);
4312 r
= journal_file_append_data(to
, data
, l
, &u
, &h
);
4316 if (JOURNAL_HEADER_KEYED_HASH(to
->header
))
4317 xor_hash
^= jenkins_hash64(data
, l
);
4319 xor_hash
^= le64toh(u
->data
.hash
);
4321 items
[m
++] = (EntryItem
) {
4323 .hash
= le64toh(u
->data
.hash
),
4327 /* The above journal_file_data_payload() may clear or overwrite cached object. Hence, we need
4328 * to re-read the object from the cache. */
4329 r
= journal_file_move_to_object(from
, OBJECT_ENTRY
, p
, &o
);
4337 r
= journal_file_append_entry_internal(
4341 &from
->header
->machine_id
,
4347 /* ret_object= */ NULL
,
4348 /* ret_offset= */ NULL
);
4350 if (mmap_cache_fd_got_sigbus(to
->cache_fd
))
4356 void journal_reset_metrics(JournalMetrics
*m
) {
4359 /* Set everything to "pick automatic values". */
4361 *m
= (JournalMetrics
) {
4362 .min_use
= UINT64_MAX
,
4363 .max_use
= UINT64_MAX
,
4364 .min_size
= UINT64_MAX
,
4365 .max_size
= UINT64_MAX
,
4366 .keep_free
= UINT64_MAX
,
4367 .n_max_files
= UINT64_MAX
,
4371 int journal_file_get_cutoff_realtime_usec(JournalFile
*f
, usec_t
*ret_from
, usec_t
*ret_to
) {
4374 assert(ret_from
|| ret_to
);
4377 if (f
->header
->head_entry_realtime
== 0)
4380 *ret_from
= le64toh(f
->header
->head_entry_realtime
);
4384 if (f
->header
->tail_entry_realtime
== 0)
4387 *ret_to
= le64toh(f
->header
->tail_entry_realtime
);
4393 int journal_file_get_cutoff_monotonic_usec(JournalFile
*f
, sd_id128_t boot_id
, usec_t
*ret_from
, usec_t
*ret_to
) {
4399 assert(ret_from
|| ret_to
);
4401 /* FIXME: fix return value assignment on success with 0. */
4403 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &p
);
4407 if (le64toh(o
->data
.n_entries
) <= 0)
4411 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, le64toh(o
->data
.entry_offset
), &o
);
4415 *ret_from
= le64toh(o
->entry
.monotonic
);
4419 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
4423 r
= generic_array_get_plus_one(f
,
4424 le64toh(o
->data
.entry_offset
),
4425 le64toh(o
->data
.entry_array_offset
),
4426 le64toh(o
->data
.n_entries
) - 1,
4432 *ret_to
= le64toh(o
->entry
.monotonic
);
4438 bool journal_file_rotate_suggested(JournalFile
*f
, usec_t max_file_usec
, int log_level
) {
4442 /* If we gained new header fields we gained new features,
4443 * hence suggest a rotation */
4444 if (le64toh(f
->header
->header_size
) < sizeof(Header
)) {
4445 log_ratelimit_full(log_level
, JOURNAL_LOG_RATELIMIT
,
4446 "%s uses an outdated header, suggesting rotation.", f
->path
);
4450 /* Let's check if the hash tables grew over a certain fill level (75%, borrowing this value from
4451 * Java's hash table implementation), and if so suggest a rotation. To calculate the fill level we
4452 * need the n_data field, which only exists in newer versions. */
4454 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
4455 if (le64toh(f
->header
->n_data
) * 4ULL > (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
4457 log_level
, JOURNAL_LOG_RATELIMIT
,
4458 "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.",
4460 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))),
4461 le64toh(f
->header
->n_data
),
4462 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
4463 (uint64_t) f
->last_stat
.st_size
,
4464 f
->last_stat
.st_size
/ le64toh(f
->header
->n_data
));
4468 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
4469 if (le64toh(f
->header
->n_fields
) * 4ULL > (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
4471 log_level
, JOURNAL_LOG_RATELIMIT
,
4472 "Field hash table of %s has a fill level at %.1f (%"PRIu64
" of %"PRIu64
" items), suggesting rotation.",
4474 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))),
4475 le64toh(f
->header
->n_fields
),
4476 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
));
4480 /* If there are too many hash collisions somebody is most likely playing games with us. Hence, if our
4481 * longest chain is longer than some threshold, let's suggest rotation. */
4482 if (JOURNAL_HEADER_CONTAINS(f
->header
, data_hash_chain_depth
) &&
4483 le64toh(f
->header
->data_hash_chain_depth
) > HASH_CHAIN_DEPTH_MAX
) {
4485 log_level
, JOURNAL_LOG_RATELIMIT
,
4486 "Data hash table of %s has deepest hash chain of length %" PRIu64
", suggesting rotation.",
4487 f
->path
, le64toh(f
->header
->data_hash_chain_depth
));
4491 if (JOURNAL_HEADER_CONTAINS(f
->header
, field_hash_chain_depth
) &&
4492 le64toh(f
->header
->field_hash_chain_depth
) > HASH_CHAIN_DEPTH_MAX
) {
4494 log_level
, JOURNAL_LOG_RATELIMIT
,
4495 "Field hash table of %s has deepest hash chain of length at %" PRIu64
", suggesting rotation.",
4496 f
->path
, le64toh(f
->header
->field_hash_chain_depth
));
4500 /* Are the data objects properly indexed by field objects? */
4501 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
4502 JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
4503 le64toh(f
->header
->n_data
) > 0 &&
4504 le64toh(f
->header
->n_fields
) == 0) {
4506 log_level
, JOURNAL_LOG_RATELIMIT
,
4507 "Data objects of %s are not indexed by field objects, suggesting rotation.",
4512 if (max_file_usec
> 0) {
4515 h
= le64toh(f
->header
->head_entry_realtime
);
4516 t
= now(CLOCK_REALTIME
);
4518 if (h
> 0 && t
> h
+ max_file_usec
) {
4520 log_level
, JOURNAL_LOG_RATELIMIT
,
4521 "Oldest entry in %s is older than the configured file retention duration (%s), suggesting rotation.",
4522 f
->path
, FORMAT_TIMESPAN(max_file_usec
, USEC_PER_SEC
));
4530 static const char * const journal_object_type_table
[] = {
4531 [OBJECT_UNUSED
] = "unused",
4532 [OBJECT_DATA
] = "data",
4533 [OBJECT_FIELD
] = "field",
4534 [OBJECT_ENTRY
] = "entry",
4535 [OBJECT_DATA_HASH_TABLE
] = "data hash table",
4536 [OBJECT_FIELD_HASH_TABLE
] = "field hash table",
4537 [OBJECT_ENTRY_ARRAY
] = "entry array",
4538 [OBJECT_TAG
] = "tag",
4541 DEFINE_STRING_TABLE_LOOKUP_TO_STRING(journal_object_type
, ObjectType
);