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 int journal_file_verify_header(JournalFile
*f
) {
516 uint64_t arena_size
, header_size
;
521 if (memcmp(f
->header
->signature
, HEADER_SIGNATURE
, 8))
524 /* In both read and write mode we refuse to open files with incompatible
525 * flags we don't know. */
526 if (warn_wrong_flags(f
, false))
527 return -EPROTONOSUPPORT
;
529 /* When open for writing we refuse to open files with compatible flags, too. */
530 if (journal_file_writable(f
) && warn_wrong_flags(f
, true))
531 return -EPROTONOSUPPORT
;
533 if (f
->header
->state
>= _STATE_MAX
)
536 header_size
= le64toh(READ_NOW(f
->header
->header_size
));
538 /* The first addition was n_data, so check that we are at least this large */
539 if (header_size
< HEADER_SIZE_MIN
)
542 /* When open for writing we refuse to open files with a mismatch of the header size, i.e. writing to
543 * files implementing older or new header structures. */
544 if (journal_file_writable(f
) && header_size
!= sizeof(Header
))
545 return -EPROTONOSUPPORT
;
547 if (JOURNAL_HEADER_SEALED(f
->header
) && !JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
550 arena_size
= le64toh(READ_NOW(f
->header
->arena_size
));
552 if (UINT64_MAX
- header_size
< arena_size
|| header_size
+ arena_size
> (uint64_t) f
->last_stat
.st_size
)
555 if (le64toh(f
->header
->tail_object_offset
) > header_size
+ arena_size
)
558 if (!VALID64(le64toh(f
->header
->data_hash_table_offset
)) ||
559 !VALID64(le64toh(f
->header
->field_hash_table_offset
)) ||
560 !VALID64(le64toh(f
->header
->tail_object_offset
)) ||
561 !VALID64(le64toh(f
->header
->entry_array_offset
)))
564 if (JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_offset
) &&
565 le64toh(f
->header
->tail_entry_offset
) != 0 &&
566 !VALID64(le64toh(f
->header
->tail_entry_offset
)))
569 if (journal_file_writable(f
)) {
570 sd_id128_t machine_id
;
574 r
= sd_id128_get_machine(&machine_id
);
576 if (!ERRNO_IS_MACHINE_ID_UNSET(r
)) /* handle graceful if machine ID is not initialized yet */
579 machine_id
= SD_ID128_NULL
;
582 if (!sd_id128_equal(machine_id
, f
->header
->machine_id
))
583 return log_debug_errno(SYNTHETIC_ERRNO(EHOSTDOWN
),
584 "Trying to open journal file from different host for writing, refusing.");
586 state
= f
->header
->state
;
588 if (state
== STATE_ARCHIVED
)
589 return -ESHUTDOWN
; /* Already archived */
590 if (state
== STATE_ONLINE
)
591 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
),
592 "Journal file %s is already online. Assuming unclean closing.",
594 if (state
!= STATE_OFFLINE
)
595 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
),
596 "Journal file %s has unknown state %i.",
599 if (f
->header
->field_hash_table_size
== 0 || f
->header
->data_hash_table_size
== 0)
606 int journal_file_fstat(JournalFile
*f
) {
612 if (fstat(f
->fd
, &f
->last_stat
) < 0)
615 f
->last_stat_usec
= now(CLOCK_MONOTONIC
);
617 /* Refuse dealing with files that aren't regular */
618 r
= stat_verify_regular(&f
->last_stat
);
622 /* Refuse appending to files that are already deleted */
623 if (f
->last_stat
.st_nlink
<= 0)
629 static int journal_file_allocate(JournalFile
*f
, uint64_t offset
, uint64_t size
) {
630 uint64_t old_size
, new_size
, old_header_size
, old_arena_size
;
636 /* We assume that this file is not sparse, and we know that for sure, since we always call
637 * posix_fallocate() ourselves */
639 if (size
> PAGE_ALIGN_DOWN(UINT64_MAX
) - offset
)
642 if (mmap_cache_fd_got_sigbus(f
->cache_fd
))
645 old_header_size
= le64toh(READ_NOW(f
->header
->header_size
));
646 old_arena_size
= le64toh(READ_NOW(f
->header
->arena_size
));
647 if (old_arena_size
> PAGE_ALIGN_DOWN(UINT64_MAX
) - old_header_size
)
650 old_size
= old_header_size
+ old_arena_size
;
652 new_size
= MAX(PAGE_ALIGN(offset
+ size
), old_header_size
);
654 if (new_size
<= old_size
) {
656 /* We already pre-allocated enough space, but before
657 * we write to it, let's check with fstat() if the
658 * file got deleted, in order make sure we don't throw
659 * away the data immediately. Don't check fstat() for
660 * all writes though, but only once ever 10s. */
662 if (f
->last_stat_usec
+ LAST_STAT_REFRESH_USEC
> now(CLOCK_MONOTONIC
))
665 return journal_file_fstat(f
);
668 /* Allocate more space. */
670 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
673 /* Refuse to go over 4G in compact mode so offsets can be stored in 32-bit. */
674 if (JOURNAL_HEADER_COMPACT(f
->header
) && new_size
> UINT32_MAX
)
677 if (new_size
> f
->metrics
.min_size
&& f
->metrics
.keep_free
> 0) {
680 if (fstatvfs(f
->fd
, &svfs
) >= 0) {
683 available
= LESS_BY((uint64_t) svfs
.f_bfree
* (uint64_t) svfs
.f_bsize
, f
->metrics
.keep_free
);
685 if (new_size
- old_size
> available
)
690 /* Increase by larger blocks at once */
691 new_size
= ROUND_UP(new_size
, FILE_SIZE_INCREASE
);
692 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
693 new_size
= f
->metrics
.max_size
;
695 /* Note that the glibc fallocate() fallback is very
696 inefficient, hence we try to minimize the allocation area
698 r
= posix_fallocate_loop(f
->fd
, old_size
, new_size
- old_size
);
702 f
->header
->arena_size
= htole64(new_size
- old_header_size
);
704 return journal_file_fstat(f
);
707 static unsigned type_to_context(ObjectType type
) {
708 /* One context for each type, plus one catch-all for the rest */
709 assert_cc(_OBJECT_TYPE_MAX
<= MMAP_CACHE_MAX_CONTEXTS
);
710 assert_cc(CONTEXT_HEADER
< MMAP_CACHE_MAX_CONTEXTS
);
711 return type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
? type
: 0;
714 static int journal_file_move_to(
727 /* This function may clear, overwrite, or alter previously cached entries. After this function has
728 * been called, all objects except for one obtained by this function are invalidated and must be
729 * re-read before use. */
734 if (size
> UINT64_MAX
- offset
)
737 /* Avoid SIGBUS on invalid accesses */
738 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
) {
739 /* Hmm, out of range? Let's refresh the fstat() data
740 * first, before we trust that check. */
742 r
= journal_file_fstat(f
);
746 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
)
747 return -EADDRNOTAVAIL
;
750 return mmap_cache_fd_get(f
->cache_fd
, type_to_context(type
), keep_always
, offset
, size
, &f
->last_stat
, ret
);
753 static uint64_t minimum_header_size(JournalFile
*f
, Object
*o
) {
755 static const uint64_t table
[] = {
756 [OBJECT_DATA
] = sizeof(DataObject
),
757 [OBJECT_FIELD
] = sizeof(FieldObject
),
758 [OBJECT_ENTRY
] = sizeof(EntryObject
),
759 [OBJECT_DATA_HASH_TABLE
] = sizeof(HashTableObject
),
760 [OBJECT_FIELD_HASH_TABLE
] = sizeof(HashTableObject
),
761 [OBJECT_ENTRY_ARRAY
] = sizeof(EntryArrayObject
),
762 [OBJECT_TAG
] = sizeof(TagObject
),
768 if (o
->object
.type
== OBJECT_DATA
)
769 return journal_file_data_payload_offset(f
);
771 if (o
->object
.type
>= ELEMENTSOF(table
) || table
[o
->object
.type
] <= 0)
772 return sizeof(ObjectHeader
);
774 return table
[o
->object
.type
];
777 static int check_object_header(JournalFile
*f
, Object
*o
, ObjectType type
, uint64_t offset
) {
783 s
= le64toh(READ_NOW(o
->object
.size
));
785 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
786 "Attempt to move to uninitialized object: %" PRIu64
,
789 if (s
< sizeof(ObjectHeader
))
790 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
791 "Attempt to move to overly short object with size %"PRIu64
": %" PRIu64
,
794 if (o
->object
.type
<= OBJECT_UNUSED
|| o
->object
.type
>= _OBJECT_TYPE_MAX
)
795 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
796 "Attempt to move to object with invalid type (%u): %" PRIu64
,
797 o
->object
.type
, offset
);
799 if (type
> OBJECT_UNUSED
&& o
->object
.type
!= type
)
800 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
801 "Found %s object while expecting %s object: %" PRIu64
,
802 journal_object_type_to_string(o
->object
.type
),
803 journal_object_type_to_string(type
),
806 if (s
< minimum_header_size(f
, o
))
807 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
808 "Size of %s object (%"PRIu64
") is smaller than the minimum object size (%"PRIu64
"): %" PRIu64
,
809 journal_object_type_to_string(o
->object
.type
),
811 minimum_header_size(f
, o
),
817 /* Lightweight object checks. We want this to be fast, so that we won't
818 * slowdown every journal_file_move_to_object() call too much. */
819 static int check_object(JournalFile
*f
, Object
*o
, uint64_t offset
) {
823 switch (o
->object
.type
) {
826 if ((le64toh(o
->data
.entry_offset
) == 0) ^ (le64toh(o
->data
.n_entries
) == 0))
827 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
828 "Bad data n_entries: %" PRIu64
": %" PRIu64
,
829 le64toh(o
->data
.n_entries
),
832 if (le64toh(o
->object
.size
) <= journal_file_data_payload_offset(f
))
833 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
834 "Bad data size (<= %zu): %" PRIu64
": %" PRIu64
,
835 journal_file_data_payload_offset(f
),
836 le64toh(o
->object
.size
),
839 if (!VALID64(le64toh(o
->data
.next_hash_offset
)) ||
840 !VALID64(le64toh(o
->data
.next_field_offset
)) ||
841 !VALID64(le64toh(o
->data
.entry_offset
)) ||
842 !VALID64(le64toh(o
->data
.entry_array_offset
)))
843 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
844 "Invalid offset, next_hash_offset=" OFSfmt
", next_field_offset=" OFSfmt
", entry_offset=" OFSfmt
", entry_array_offset=" OFSfmt
": %" PRIu64
,
845 le64toh(o
->data
.next_hash_offset
),
846 le64toh(o
->data
.next_field_offset
),
847 le64toh(o
->data
.entry_offset
),
848 le64toh(o
->data
.entry_array_offset
),
854 if (le64toh(o
->object
.size
) <= offsetof(Object
, field
.payload
))
855 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
856 "Bad field size (<= %zu): %" PRIu64
": %" PRIu64
,
857 offsetof(Object
, field
.payload
),
858 le64toh(o
->object
.size
),
861 if (!VALID64(le64toh(o
->field
.next_hash_offset
)) ||
862 !VALID64(le64toh(o
->field
.head_data_offset
)))
863 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
864 "Invalid offset, next_hash_offset=" OFSfmt
", head_data_offset=" OFSfmt
": %" PRIu64
,
865 le64toh(o
->field
.next_hash_offset
),
866 le64toh(o
->field
.head_data_offset
),
873 sz
= le64toh(READ_NOW(o
->object
.size
));
874 if (sz
< offsetof(Object
, entry
.items
) ||
875 (sz
- offsetof(Object
, entry
.items
)) % journal_file_entry_item_size(f
) != 0)
876 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
877 "Bad entry size (<= %zu): %" PRIu64
": %" PRIu64
,
878 offsetof(Object
, entry
.items
),
882 if ((sz
- offsetof(Object
, entry
.items
)) / journal_file_entry_item_size(f
) <= 0)
883 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
884 "Invalid number items in entry: %" PRIu64
": %" PRIu64
,
885 (sz
- offsetof(Object
, entry
.items
)) / journal_file_entry_item_size(f
),
888 if (le64toh(o
->entry
.seqnum
) <= 0)
889 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
890 "Invalid entry seqnum: %" PRIx64
": %" PRIu64
,
891 le64toh(o
->entry
.seqnum
),
894 if (!VALID_REALTIME(le64toh(o
->entry
.realtime
)))
895 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
896 "Invalid entry realtime timestamp: %" PRIu64
": %" PRIu64
,
897 le64toh(o
->entry
.realtime
),
900 if (!VALID_MONOTONIC(le64toh(o
->entry
.monotonic
)))
901 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
902 "Invalid entry monotonic timestamp: %" PRIu64
": %" PRIu64
,
903 le64toh(o
->entry
.monotonic
),
909 case OBJECT_DATA_HASH_TABLE
:
910 case OBJECT_FIELD_HASH_TABLE
: {
913 sz
= le64toh(READ_NOW(o
->object
.size
));
914 if (sz
< offsetof(Object
, hash_table
.items
) ||
915 (sz
- offsetof(Object
, hash_table
.items
)) % sizeof(HashItem
) != 0 ||
916 (sz
- offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
) <= 0)
917 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
918 "Invalid %s hash table size: %" PRIu64
": %" PRIu64
,
919 journal_object_type_to_string(o
->object
.type
),
926 case OBJECT_ENTRY_ARRAY
: {
929 sz
= le64toh(READ_NOW(o
->object
.size
));
930 if (sz
< offsetof(Object
, entry_array
.items
) ||
931 (sz
- offsetof(Object
, entry_array
.items
)) % journal_file_entry_array_item_size(f
) != 0 ||
932 (sz
- offsetof(Object
, entry_array
.items
)) / journal_file_entry_array_item_size(f
) <= 0)
933 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
934 "Invalid object entry array size: %" PRIu64
": %" PRIu64
,
937 /* Here, we request that the offset of each entry array object is in strictly increasing order. */
938 next
= le64toh(o
->entry_array
.next_entry_array_offset
);
939 if (!VALID64(next
) || (next
> 0 && next
<= offset
))
940 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
941 "Invalid object entry array next_entry_array_offset: %" PRIu64
": %" PRIu64
,
949 if (le64toh(o
->object
.size
) != sizeof(TagObject
))
950 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
951 "Invalid object tag size: %" PRIu64
": %" PRIu64
,
952 le64toh(o
->object
.size
),
955 if (!VALID_EPOCH(le64toh(o
->tag
.epoch
)))
956 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
957 "Invalid object tag epoch: %" PRIu64
": %" PRIu64
,
958 le64toh(o
->tag
.epoch
), offset
);
966 int journal_file_move_to_object(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
**ret
) {
972 /* Even if this function fails, it may clear, overwrite, or alter previously cached entries. After
973 * this function has been called, all objects except for one obtained by this function are
974 * invalidated and must be re-read before use.. */
976 /* Objects may only be located at multiple of 64 bit */
977 if (!VALID64(offset
))
978 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
979 "Attempt to move to %s object at non-64bit boundary: %" PRIu64
,
980 journal_object_type_to_string(type
),
983 /* Object may not be located in the file header */
984 if (offset
< le64toh(f
->header
->header_size
))
985 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
986 "Attempt to move to %s object located in file header: %" PRIu64
,
987 journal_object_type_to_string(type
),
990 r
= journal_file_move_to(f
, type
, false, offset
, sizeof(ObjectHeader
), (void**) &o
);
994 r
= check_object_header(f
, o
, type
, offset
);
998 r
= journal_file_move_to(f
, type
, false, offset
, le64toh(READ_NOW(o
->object
.size
)), (void**) &o
);
1002 r
= check_object_header(f
, o
, type
, offset
);
1006 r
= check_object(f
, o
, offset
);
1016 int journal_file_read_object_header(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
*ret
) {
1023 /* Objects may only be located at multiple of 64 bit */
1024 if (!VALID64(offset
))
1025 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1026 "Attempt to read %s object at non-64bit boundary: %" PRIu64
,
1027 journal_object_type_to_string(type
), offset
);
1029 /* Object may not be located in the file header */
1030 if (offset
< le64toh(f
->header
->header_size
))
1031 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1032 "Attempt to read %s object located in file header: %" PRIu64
,
1033 journal_object_type_to_string(type
), offset
);
1035 /* This will likely read too much data but it avoids having to call pread() twice. */
1036 n
= pread(f
->fd
, &o
, sizeof(o
), offset
);
1038 return log_debug_errno(errno
, "Failed to read journal %s object at offset: %" PRIu64
,
1039 journal_object_type_to_string(type
), offset
);
1041 if ((size_t) n
< sizeof(o
.object
))
1042 return log_debug_errno(SYNTHETIC_ERRNO(EIO
),
1043 "Failed to read short %s object at offset: %" PRIu64
,
1044 journal_object_type_to_string(type
), offset
);
1046 r
= check_object_header(f
, &o
, type
, offset
);
1050 if ((size_t) n
< minimum_header_size(f
, &o
))
1051 return log_debug_errno(SYNTHETIC_ERRNO(EIO
),
1052 "Short read while reading %s object: %" PRIu64
,
1053 journal_object_type_to_string(type
), offset
);
1055 r
= check_object(f
, &o
, offset
);
1065 static uint64_t inc_seqnum(uint64_t seqnum
) {
1066 if (seqnum
< UINT64_MAX
-1)
1069 return 1; /* skip over UINT64_MAX and 0 when we run out of seqnums and start again */
1072 static uint64_t journal_file_entry_seqnum(
1076 uint64_t next_seqnum
;
1081 /* Picks a new sequence number for the entry we are about to add and returns it. */
1083 next_seqnum
= inc_seqnum(le64toh(f
->header
->tail_entry_seqnum
));
1085 /* If an external seqnum counter was passed, we update both the local and the external one, and set
1086 * it to the maximum of both */
1088 *seqnum
= next_seqnum
= MAX(inc_seqnum(*seqnum
), next_seqnum
);
1090 f
->header
->tail_entry_seqnum
= htole64(next_seqnum
);
1092 if (f
->header
->head_entry_seqnum
== 0)
1093 f
->header
->head_entry_seqnum
= htole64(next_seqnum
);
1098 int journal_file_append_object(
1102 Object
**ret_object
,
1103 uint64_t *ret_offset
) {
1111 assert(type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
);
1112 assert(size
>= sizeof(ObjectHeader
));
1114 r
= journal_file_set_online(f
);
1118 r
= journal_file_tail_end_by_mmap(f
, &p
);
1122 r
= journal_file_allocate(f
, p
, size
);
1126 r
= journal_file_move_to(f
, type
, false, p
, size
, (void**) &o
);
1130 o
->object
= (ObjectHeader
) {
1132 .size
= htole64(size
),
1135 f
->header
->tail_object_offset
= htole64(p
);
1136 f
->header
->n_objects
= htole64(le64toh(f
->header
->n_objects
) + 1);
1147 static int journal_file_setup_data_hash_table(JournalFile
*f
) {
1155 /* We estimate that we need 1 hash table entry per 768 bytes
1156 of journal file and we want to make sure we never get
1157 beyond 75% fill level. Calculate the hash table size for
1158 the maximum file size based on these metrics. */
1160 s
= (f
->metrics
.max_size
* 4 / 768 / 3) * sizeof(HashItem
);
1161 if (s
< DEFAULT_DATA_HASH_TABLE_SIZE
)
1162 s
= DEFAULT_DATA_HASH_TABLE_SIZE
;
1164 log_debug("Reserving %"PRIu64
" entries in data hash table.", s
/ sizeof(HashItem
));
1166 r
= journal_file_append_object(f
,
1167 OBJECT_DATA_HASH_TABLE
,
1168 offsetof(Object
, hash_table
.items
) + s
,
1173 memzero(o
->hash_table
.items
, s
);
1175 f
->header
->data_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1176 f
->header
->data_hash_table_size
= htole64(s
);
1181 static int journal_file_setup_field_hash_table(JournalFile
*f
) {
1189 /* We use a fixed size hash table for the fields as this
1190 * number should grow very slowly only */
1192 s
= DEFAULT_FIELD_HASH_TABLE_SIZE
;
1193 log_debug("Reserving %"PRIu64
" entries in field hash table.", s
/ sizeof(HashItem
));
1195 r
= journal_file_append_object(f
,
1196 OBJECT_FIELD_HASH_TABLE
,
1197 offsetof(Object
, hash_table
.items
) + s
,
1202 memzero(o
->hash_table
.items
, s
);
1204 f
->header
->field_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1205 f
->header
->field_hash_table_size
= htole64(s
);
1210 int journal_file_map_data_hash_table(JournalFile
*f
) {
1218 if (f
->data_hash_table
)
1221 p
= le64toh(f
->header
->data_hash_table_offset
);
1222 s
= le64toh(f
->header
->data_hash_table_size
);
1224 r
= journal_file_move_to(f
,
1225 OBJECT_DATA_HASH_TABLE
,
1232 f
->data_hash_table
= t
;
1236 int journal_file_map_field_hash_table(JournalFile
*f
) {
1244 if (f
->field_hash_table
)
1247 p
= le64toh(f
->header
->field_hash_table_offset
);
1248 s
= le64toh(f
->header
->field_hash_table_size
);
1250 r
= journal_file_move_to(f
,
1251 OBJECT_FIELD_HASH_TABLE
,
1258 f
->field_hash_table
= t
;
1262 static int journal_file_link_field(
1273 assert(f
->field_hash_table
);
1277 if (o
->object
.type
!= OBJECT_FIELD
)
1280 m
= le64toh(READ_NOW(f
->header
->field_hash_table_size
)) / sizeof(HashItem
);
1284 /* This might alter the window we are looking at */
1285 o
->field
.next_hash_offset
= o
->field
.head_data_offset
= 0;
1288 p
= le64toh(f
->field_hash_table
[h
].tail_hash_offset
);
1290 f
->field_hash_table
[h
].head_hash_offset
= htole64(offset
);
1292 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1296 o
->field
.next_hash_offset
= htole64(offset
);
1299 f
->field_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1301 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
1302 f
->header
->n_fields
= htole64(le64toh(f
->header
->n_fields
) + 1);
1307 static int journal_file_link_data(
1318 assert(f
->data_hash_table
);
1322 if (o
->object
.type
!= OBJECT_DATA
)
1325 m
= le64toh(READ_NOW(f
->header
->data_hash_table_size
)) / sizeof(HashItem
);
1329 /* This might alter the window we are looking at */
1330 o
->data
.next_hash_offset
= o
->data
.next_field_offset
= 0;
1331 o
->data
.entry_offset
= o
->data
.entry_array_offset
= 0;
1332 o
->data
.n_entries
= 0;
1335 p
= le64toh(f
->data_hash_table
[h
].tail_hash_offset
);
1337 /* Only entry in the hash table is easy */
1338 f
->data_hash_table
[h
].head_hash_offset
= htole64(offset
);
1340 /* Move back to the previous data object, to patch in
1343 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1347 o
->data
.next_hash_offset
= htole64(offset
);
1350 f
->data_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1352 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
1353 f
->header
->n_data
= htole64(le64toh(f
->header
->n_data
) + 1);
1358 static int get_next_hash_offset(
1361 le64_t
*next_hash_offset
,
1363 le64_t
*header_max_depth
) {
1369 assert(next_hash_offset
);
1372 nextp
= le64toh(READ_NOW(*next_hash_offset
));
1374 if (nextp
<= *p
) /* Refuse going in loops */
1375 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1376 "Detected hash item loop in %s, refusing.", f
->path
);
1380 /* If the depth of this hash chain is larger than all others we have seen so far, record it */
1381 if (header_max_depth
&& journal_file_writable(f
))
1382 *header_max_depth
= htole64(MAX(*depth
, le64toh(*header_max_depth
)));
1389 int journal_file_find_field_object_with_hash(
1394 Object
**ret_object
,
1395 uint64_t *ret_offset
) {
1397 uint64_t p
, osize
, h
, m
, depth
= 0;
1405 /* If the field hash table is empty, we can't find anything */
1406 if (le64toh(f
->header
->field_hash_table_size
) <= 0)
1409 /* Map the field hash table, if it isn't mapped yet. */
1410 r
= journal_file_map_field_hash_table(f
);
1414 osize
= offsetof(Object
, field
.payload
) + size
;
1416 m
= le64toh(READ_NOW(f
->header
->field_hash_table_size
)) / sizeof(HashItem
);
1421 p
= le64toh(f
->field_hash_table
[h
].head_hash_offset
);
1425 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1429 if (le64toh(o
->field
.hash
) == hash
&&
1430 le64toh(o
->object
.size
) == osize
&&
1431 memcmp(o
->field
.payload
, field
, size
) == 0) {
1441 r
= get_next_hash_offset(
1444 &o
->field
.next_hash_offset
,
1446 JOURNAL_HEADER_CONTAINS(f
->header
, field_hash_chain_depth
) ? &f
->header
->field_hash_chain_depth
: NULL
);
1454 uint64_t journal_file_hash_data(
1461 assert(data
|| sz
== 0);
1463 /* We try to unify our codebase on siphash, hence new-styled journal files utilizing the keyed hash
1464 * function use siphash. Old journal files use the Jenkins hash. */
1466 if (JOURNAL_HEADER_KEYED_HASH(f
->header
))
1467 return siphash24(data
, sz
, f
->header
->file_id
.bytes
);
1469 return jenkins_hash64(data
, sz
);
1472 int journal_file_find_field_object(
1476 Object
**ret_object
,
1477 uint64_t *ret_offset
) {
1483 return journal_file_find_field_object_with_hash(
1486 journal_file_hash_data(f
, field
, size
),
1487 ret_object
, ret_offset
);
1490 int journal_file_find_data_object_with_hash(
1495 Object
**ret_object
,
1496 uint64_t *ret_offset
) {
1498 uint64_t p
, h
, m
, depth
= 0;
1503 assert(data
|| size
== 0);
1505 /* If there's no data hash table, then there's no entry. */
1506 if (le64toh(f
->header
->data_hash_table_size
) <= 0)
1509 /* Map the data hash table, if it isn't mapped yet. */
1510 r
= journal_file_map_data_hash_table(f
);
1514 m
= le64toh(READ_NOW(f
->header
->data_hash_table_size
)) / sizeof(HashItem
);
1519 p
= le64toh(f
->data_hash_table
[h
].head_hash_offset
);
1526 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1530 if (le64toh(o
->data
.hash
) != hash
)
1533 r
= journal_file_data_payload(f
, o
, p
, NULL
, 0, 0, &d
, &rsize
);
1536 assert(r
> 0); /* journal_file_data_payload() always returns > 0 if no field is provided. */
1538 if (memcmp_nn(data
, size
, d
, rsize
) == 0) {
1549 r
= get_next_hash_offset(
1552 &o
->data
.next_hash_offset
,
1554 JOURNAL_HEADER_CONTAINS(f
->header
, data_hash_chain_depth
) ? &f
->header
->data_hash_chain_depth
: NULL
);
1562 int journal_file_find_data_object(
1566 Object
**ret_object
,
1567 uint64_t *ret_offset
) {
1570 assert(data
|| size
== 0);
1572 return journal_file_find_data_object_with_hash(
1575 journal_file_hash_data(f
, data
, size
),
1576 ret_object
, ret_offset
);
1579 bool journal_field_valid(const char *p
, size_t l
, bool allow_protected
) {
1580 /* We kinda enforce POSIX syntax recommendations for
1581 environment variables here, but make a couple of additional
1584 http://pubs.opengroup.org/onlinepubs/000095399/basedefs/xbd_chap08.html */
1591 /* No empty field names */
1595 /* Don't allow names longer than 64 chars */
1599 /* Variables starting with an underscore are protected */
1600 if (!allow_protected
&& p
[0] == '_')
1603 /* Don't allow digits as first character */
1604 if (ascii_isdigit(p
[0]))
1607 /* Only allow A-Z0-9 and '_' */
1608 for (const char *a
= p
; a
< p
+ l
; a
++)
1609 if ((*a
< 'A' || *a
> 'Z') &&
1610 !ascii_isdigit(*a
) &&
1617 static int journal_file_append_field(
1621 Object
**ret_object
,
1622 uint64_t *ret_offset
) {
1633 if (!journal_field_valid(field
, size
, true))
1636 hash
= journal_file_hash_data(f
, field
, size
);
1638 r
= journal_file_find_field_object_with_hash(f
, field
, size
, hash
, ret_object
, ret_offset
);
1644 osize
= offsetof(Object
, field
.payload
) + size
;
1645 r
= journal_file_append_object(f
, OBJECT_FIELD
, osize
, &o
, &p
);
1649 o
->field
.hash
= htole64(hash
);
1650 memcpy(o
->field
.payload
, field
, size
);
1652 r
= journal_file_link_field(f
, o
, p
, hash
);
1656 /* The linking might have altered the window, so let's only pass the offset to hmac which will
1657 * move to the object again if needed. */
1660 r
= journal_file_hmac_put_object(f
, OBJECT_FIELD
, NULL
, p
);
1666 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, ret_object
);
1677 static int maybe_compress_payload(JournalFile
*f
, uint8_t *dst
, const uint8_t *src
, uint64_t size
, size_t *rsize
) {
1681 #if HAVE_COMPRESSION
1685 c
= JOURNAL_FILE_COMPRESSION(f
);
1686 if (c
== COMPRESSION_NONE
|| size
< f
->compress_threshold_bytes
)
1689 r
= compress_blob(c
, src
, size
, dst
, size
- 1, rsize
);
1691 return log_debug_errno(r
, "Failed to compress data object using %s, ignoring: %m", compression_to_string(c
));
1693 log_debug("Compressed data object %"PRIu64
" -> %zu using %s", size
, *rsize
, compression_to_string(c
));
1695 return 1; /* compressed */
1701 static int journal_file_append_data(
1705 Object
**ret_object
,
1706 uint64_t *ret_offset
) {
1708 uint64_t hash
, p
, osize
;
1716 if (!data
|| size
== 0)
1719 hash
= journal_file_hash_data(f
, data
, size
);
1721 r
= journal_file_find_data_object_with_hash(f
, data
, size
, hash
, ret_object
, ret_offset
);
1727 eq
= memchr(data
, '=', size
);
1731 osize
= journal_file_data_payload_offset(f
) + size
;
1732 r
= journal_file_append_object(f
, OBJECT_DATA
, osize
, &o
, &p
);
1736 o
->data
.hash
= htole64(hash
);
1738 r
= maybe_compress_payload(f
, journal_file_data_payload_field(f
, o
), data
, size
, &rsize
);
1740 /* We don't really care failures, let's continue without compression */
1741 memcpy_safe(journal_file_data_payload_field(f
, o
), data
, size
);
1743 Compression c
= JOURNAL_FILE_COMPRESSION(f
);
1745 assert(c
>= 0 && c
< _COMPRESSION_MAX
&& c
!= COMPRESSION_NONE
);
1747 o
->object
.size
= htole64(journal_file_data_payload_offset(f
) + rsize
);
1748 o
->object
.flags
|= COMPRESSION_TO_OBJECT_FLAG(c
);
1751 r
= journal_file_link_data(f
, o
, p
, hash
);
1755 /* The linking might have altered the window, so let's refresh our pointer. */
1756 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1761 r
= journal_file_hmac_put_object(f
, OBJECT_DATA
, o
, p
);
1766 /* Create field object ... */
1767 r
= journal_file_append_field(f
, data
, (uint8_t*) eq
- (uint8_t*) data
, &fo
, NULL
);
1771 /* ... and link it in. */
1772 o
->data
.next_field_offset
= fo
->field
.head_data_offset
;
1773 fo
->field
.head_data_offset
= le64toh(p
);
1784 static int maybe_decompress_payload(
1788 Compression compression
,
1790 size_t field_length
,
1791 size_t data_threshold
,
1797 /* We can't read objects larger than 4G on a 32bit machine */
1798 if ((uint64_t) (size_t) size
!= size
)
1801 if (compression
!= COMPRESSION_NONE
) {
1802 #if HAVE_COMPRESSION
1807 r
= decompress_startswith(compression
, payload
, size
, &f
->compress_buffer
, field
,
1810 return log_debug_errno(r
,
1811 "Cannot decompress %s object of length %" PRIu64
": %m",
1812 compression_to_string(compression
),
1823 r
= decompress_blob(compression
, payload
, size
, &f
->compress_buffer
, &rsize
, 0);
1828 *ret_data
= f
->compress_buffer
;
1832 return -EPROTONOSUPPORT
;
1835 if (field
&& (size
< field_length
+ 1 || memcmp(payload
, field
, field_length
) != 0 || payload
[field_length
] != '=')) {
1844 *ret_data
= payload
;
1846 *ret_size
= (size_t) size
;
1852 int journal_file_data_payload(
1857 size_t field_length
,
1858 size_t data_threshold
,
1867 assert(!field
== (field_length
== 0)); /* These must be specified together. */
1870 r
= journal_file_move_to_object(f
, OBJECT_DATA
, offset
, &o
);
1875 size
= le64toh(READ_NOW(o
->object
.size
));
1876 if (size
< journal_file_data_payload_offset(f
))
1879 size
-= journal_file_data_payload_offset(f
);
1881 c
= COMPRESSION_FROM_OBJECT(o
);
1883 return -EPROTONOSUPPORT
;
1885 return maybe_decompress_payload(f
, journal_file_data_payload_field(f
, o
), size
, c
, field
,
1886 field_length
, data_threshold
, ret_data
, ret_size
);
1889 uint64_t journal_file_entry_n_items(JournalFile
*f
, Object
*o
) {
1895 if (o
->object
.type
!= OBJECT_ENTRY
)
1898 sz
= le64toh(READ_NOW(o
->object
.size
));
1899 if (sz
< offsetof(Object
, entry
.items
))
1902 return (sz
- offsetof(Object
, entry
.items
)) / journal_file_entry_item_size(f
);
1905 uint64_t journal_file_entry_array_n_items(JournalFile
*f
, Object
*o
) {
1911 if (o
->object
.type
!= OBJECT_ENTRY_ARRAY
)
1914 sz
= le64toh(READ_NOW(o
->object
.size
));
1915 if (sz
< offsetof(Object
, entry_array
.items
))
1918 return (sz
- offsetof(Object
, entry_array
.items
)) / journal_file_entry_array_item_size(f
);
1921 uint64_t journal_file_hash_table_n_items(Object
*o
) {
1926 if (!IN_SET(o
->object
.type
, OBJECT_DATA_HASH_TABLE
, OBJECT_FIELD_HASH_TABLE
))
1929 sz
= le64toh(READ_NOW(o
->object
.size
));
1930 if (sz
< offsetof(Object
, hash_table
.items
))
1933 return (sz
- offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
);
1936 static void write_entry_array_item(JournalFile
*f
, Object
*o
, uint64_t i
, uint64_t p
) {
1940 if (JOURNAL_HEADER_COMPACT(f
->header
)) {
1941 assert(p
<= UINT32_MAX
);
1942 o
->entry_array
.items
.compact
[i
] = htole32(p
);
1944 o
->entry_array
.items
.regular
[i
] = htole64(p
);
1947 static int link_entry_into_array(
1955 uint64_t n
= 0, ap
= 0, q
, i
, a
, hidx
;
1965 a
= tail
? le32toh(*tail
) : le64toh(*first
);
1966 hidx
= le64toh(READ_NOW(*idx
));
1967 i
= tidx
? le32toh(READ_NOW(*tidx
)) : hidx
;
1970 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
1974 n
= journal_file_entry_array_n_items(f
, o
);
1976 write_entry_array_item(f
, o
, i
, p
);
1977 *idx
= htole64(hidx
+ 1);
1979 *tidx
= htole32(le32toh(*tidx
) + 1);
1985 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
1996 r
= journal_file_append_object(f
, OBJECT_ENTRY_ARRAY
,
1997 offsetof(Object
, entry_array
.items
) + n
* journal_file_entry_array_item_size(f
),
2003 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY_ARRAY
, o
, q
);
2008 write_entry_array_item(f
, o
, i
, p
);
2011 *first
= htole64(q
);
2013 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, ap
, &o
);
2017 o
->entry_array
.next_entry_array_offset
= htole64(q
);
2023 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
2024 f
->header
->n_entry_arrays
= htole64(le64toh(f
->header
->n_entry_arrays
) + 1);
2026 *idx
= htole64(hidx
+ 1);
2033 static int link_entry_into_array_plus_one(
2051 hidx
= le64toh(READ_NOW(*idx
));
2052 if (hidx
== UINT64_MAX
)
2055 *extra
= htole64(p
);
2059 i
= htole64(hidx
- 1);
2060 r
= link_entry_into_array(f
, first
, &i
, tail
, tidx
, p
);
2065 *idx
= htole64(hidx
+ 1);
2069 static int journal_file_link_entry_item(JournalFile
*f
, uint64_t offset
, uint64_t p
) {
2076 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
2080 return link_entry_into_array_plus_one(f
,
2081 &o
->data
.entry_offset
,
2082 &o
->data
.entry_array_offset
,
2084 JOURNAL_HEADER_COMPACT(f
->header
) ? &o
->data
.compact
.tail_entry_array_offset
: NULL
,
2085 JOURNAL_HEADER_COMPACT(f
->header
) ? &o
->data
.compact
.tail_entry_array_n_entries
: NULL
,
2089 static int journal_file_link_entry(
2093 const EntryItem items
[],
2103 if (o
->object
.type
!= OBJECT_ENTRY
)
2106 __atomic_thread_fence(__ATOMIC_SEQ_CST
);
2108 /* Link up the entry itself */
2109 r
= link_entry_into_array(f
,
2110 &f
->header
->entry_array_offset
,
2111 &f
->header
->n_entries
,
2112 JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_array_offset
) ? &f
->header
->tail_entry_array_offset
: NULL
,
2113 JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_array_n_entries
) ? &f
->header
->tail_entry_array_n_entries
: NULL
,
2118 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
2120 if (f
->header
->head_entry_realtime
== 0)
2121 f
->header
->head_entry_realtime
= o
->entry
.realtime
;
2123 f
->header
->tail_entry_realtime
= o
->entry
.realtime
;
2124 f
->header
->tail_entry_monotonic
= o
->entry
.monotonic
;
2125 if (JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_offset
))
2126 f
->header
->tail_entry_offset
= htole64(offset
);
2127 f
->newest_mtime
= 0; /* we have a new tail entry now, explicitly invalidate newest boot id/timestamp info */
2129 /* Link up the items */
2130 for (uint64_t i
= 0; i
< n_items
; i
++) {
2133 /* If we fail to link an entry item because we can't allocate a new entry array, don't fail
2134 * immediately but try to link the other entry items since it might still be possible to link
2135 * those if they don't require a new entry array to be allocated. */
2137 k
= journal_file_link_entry_item(f
, offset
, items
[i
].object_offset
);
2147 static void write_entry_item(JournalFile
*f
, Object
*o
, uint64_t i
, const EntryItem
*item
) {
2152 if (JOURNAL_HEADER_COMPACT(f
->header
)) {
2153 assert(item
->object_offset
<= UINT32_MAX
);
2154 o
->entry
.items
.compact
[i
].object_offset
= htole32(item
->object_offset
);
2156 o
->entry
.items
.regular
[i
].object_offset
= htole64(item
->object_offset
);
2157 o
->entry
.items
.regular
[i
].hash
= htole64(item
->hash
);
2161 static int journal_file_append_entry_internal(
2163 const dual_timestamp
*ts
,
2164 const sd_id128_t
*boot_id
,
2165 const sd_id128_t
*machine_id
,
2167 const EntryItem items
[],
2170 sd_id128_t
*seqnum_id
,
2171 Object
**ret_object
,
2172 uint64_t *ret_offset
) {
2182 assert(items
|| n_items
== 0);
2184 if (f
->strict_order
) {
2185 /* If requested be stricter with ordering in this journal file, to make searching via
2186 * bisection fully deterministic. This is an optional feature, so that if desired journal
2187 * files can be written where the ordering is not strictly enforced (in which case bisection
2188 * will yield *a* result, but not the *only* result, when searching for points in
2189 * time). Strict ordering mode is enabled when journald originally writes the files, but
2190 * might not necessarily be if other tools (the remoting tools for example) write journal
2191 * files from combined sources.
2193 * Typically, if any of the errors generated here are seen journald will just rotate the
2194 * journal files and start anew. */
2196 if (ts
->realtime
< le64toh(f
->header
->tail_entry_realtime
))
2197 return log_debug_errno(SYNTHETIC_ERRNO(EREMCHG
),
2198 "Realtime timestamp %" PRIu64
" smaller than previous realtime "
2199 "timestamp %" PRIu64
", refusing entry.",
2200 ts
->realtime
, le64toh(f
->header
->tail_entry_realtime
));
2202 if (!sd_id128_is_null(f
->header
->tail_entry_boot_id
) && boot_id
) {
2204 if (!sd_id128_equal(f
->header
->tail_entry_boot_id
, *boot_id
))
2205 return log_debug_errno(SYNTHETIC_ERRNO(EREMOTE
),
2206 "Boot ID to write is different from previous boot id, refusing entry.");
2208 if (ts
->monotonic
< le64toh(f
->header
->tail_entry_monotonic
))
2209 return log_debug_errno(SYNTHETIC_ERRNO(ENOTNAM
),
2210 "Monotonic timestamp %" PRIu64
" smaller than previous monotonic "
2211 "timestamp %" PRIu64
", refusing entry.",
2212 ts
->monotonic
, le64toh(f
->header
->tail_entry_monotonic
));
2217 /* Settle the passed in sequence number ID */
2219 if (sd_id128_is_null(*seqnum_id
))
2220 *seqnum_id
= f
->header
->seqnum_id
; /* Caller has none assigned, then copy the one from the file */
2221 else if (!sd_id128_equal(*seqnum_id
, f
->header
->seqnum_id
)) {
2222 /* Different seqnum IDs? We can't allow entries from multiple IDs end up in the same journal.*/
2223 if (le64toh(f
->header
->n_entries
) == 0)
2224 f
->header
->seqnum_id
= *seqnum_id
; /* Caller has one, and file so far has no entries, then copy the one from the caller */
2226 return log_debug_errno(SYNTHETIC_ERRNO(EILSEQ
),
2227 "Sequence number IDs don't match, refusing entry.");
2231 if (machine_id
&& sd_id128_is_null(f
->header
->machine_id
))
2232 /* Initialize machine ID when not set yet */
2233 f
->header
->machine_id
= *machine_id
;
2235 osize
= offsetof(Object
, entry
.items
) + (n_items
* journal_file_entry_item_size(f
));
2237 r
= journal_file_append_object(f
, OBJECT_ENTRY
, osize
, &o
, &np
);
2241 o
->entry
.seqnum
= htole64(journal_file_entry_seqnum(f
, seqnum
));
2242 o
->entry
.realtime
= htole64(ts
->realtime
);
2243 o
->entry
.monotonic
= htole64(ts
->monotonic
);
2244 o
->entry
.xor_hash
= htole64(xor_hash
);
2246 f
->header
->tail_entry_boot_id
= *boot_id
;
2247 o
->entry
.boot_id
= f
->header
->tail_entry_boot_id
;
2249 for (size_t i
= 0; i
< n_items
; i
++)
2250 write_entry_item(f
, o
, i
, &items
[i
]);
2253 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY
, o
, np
);
2258 r
= journal_file_link_entry(f
, o
, np
, items
, n_items
);
2271 void journal_file_post_change(JournalFile
*f
) {
2277 /* inotify() does not receive IN_MODIFY events from file
2278 * accesses done via mmap(). After each access we hence
2279 * trigger IN_MODIFY by truncating the journal file to its
2280 * current size which triggers IN_MODIFY. */
2282 __atomic_thread_fence(__ATOMIC_SEQ_CST
);
2284 if (ftruncate(f
->fd
, f
->last_stat
.st_size
) < 0)
2285 log_debug_errno(errno
, "Failed to truncate file to its own size: %m");
2288 static int post_change_thunk(sd_event_source
*timer
, uint64_t usec
, void *userdata
) {
2291 journal_file_post_change(userdata
);
2296 static void schedule_post_change(JournalFile
*f
) {
2301 assert(f
->post_change_timer
);
2303 assert_se(e
= sd_event_source_get_event(f
->post_change_timer
));
2305 /* If we are already going down, post the change immediately. */
2306 if (IN_SET(sd_event_get_state(e
), SD_EVENT_EXITING
, SD_EVENT_FINISHED
))
2309 r
= sd_event_source_get_enabled(f
->post_change_timer
, NULL
);
2311 log_debug_errno(r
, "Failed to get ftruncate timer state: %m");
2317 r
= sd_event_source_set_time_relative(f
->post_change_timer
, f
->post_change_timer_period
);
2319 log_debug_errno(r
, "Failed to set time for scheduling ftruncate: %m");
2323 r
= sd_event_source_set_enabled(f
->post_change_timer
, SD_EVENT_ONESHOT
);
2325 log_debug_errno(r
, "Failed to enable scheduled ftruncate: %m");
2332 /* On failure, let's simply post the change immediately. */
2333 journal_file_post_change(f
);
2336 /* Enable coalesced change posting in a timer on the provided sd_event instance */
2337 int journal_file_enable_post_change_timer(JournalFile
*f
, sd_event
*e
, usec_t t
) {
2338 _cleanup_(sd_event_source_unrefp
) sd_event_source
*timer
= NULL
;
2342 assert_return(!f
->post_change_timer
, -EINVAL
);
2346 r
= sd_event_add_time(e
, &timer
, CLOCK_MONOTONIC
, 0, 0, post_change_thunk
, f
);
2350 r
= sd_event_source_set_enabled(timer
, SD_EVENT_OFF
);
2354 f
->post_change_timer
= TAKE_PTR(timer
);
2355 f
->post_change_timer_period
= t
;
2360 static int entry_item_cmp(const EntryItem
*a
, const EntryItem
*b
) {
2361 return CMP(ASSERT_PTR(a
)->object_offset
, ASSERT_PTR(b
)->object_offset
);
2364 static size_t remove_duplicate_entry_items(EntryItem items
[], size_t n
) {
2367 assert(items
|| n
== 0);
2372 for (size_t i
= 1; i
< n
; i
++)
2373 if (items
[i
].object_offset
!= items
[j
- 1].object_offset
)
2374 items
[j
++] = items
[i
];
2379 int journal_file_append_entry(
2381 const dual_timestamp
*ts
,
2382 const sd_id128_t
*boot_id
,
2383 const struct iovec iovec
[],
2386 sd_id128_t
*seqnum_id
,
2387 Object
**ret_object
,
2388 uint64_t *ret_offset
) {
2390 _cleanup_free_ EntryItem
*items_alloc
= NULL
;
2392 uint64_t xor_hash
= 0;
2393 struct dual_timestamp _ts
;
2394 sd_id128_t _boot_id
, _machine_id
, *machine_id
;
2400 assert(n_iovec
> 0);
2403 if (!VALID_REALTIME(ts
->realtime
))
2404 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
2405 "Invalid realtime timestamp %" PRIu64
", refusing entry.",
2407 if (!VALID_MONOTONIC(ts
->monotonic
))
2408 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
2409 "Invalid monotomic timestamp %" PRIu64
", refusing entry.",
2412 dual_timestamp_get(&_ts
);
2417 r
= sd_id128_get_boot(&_boot_id
);
2421 boot_id
= &_boot_id
;
2424 r
= sd_id128_get_machine(&_machine_id
);
2426 if (!ERRNO_IS_MACHINE_ID_UNSET(r
))
2429 /* If the machine ID is not initialized yet, handle gracefully */
2432 machine_id
= &_machine_id
;
2435 r
= journal_file_maybe_append_tag(f
, ts
->realtime
);
2440 if (n_iovec
< ALLOCA_MAX
/ sizeof(EntryItem
) / 2)
2441 items
= newa(EntryItem
, n_iovec
);
2443 items_alloc
= new(EntryItem
, n_iovec
);
2447 items
= items_alloc
;
2450 for (size_t i
= 0; i
< n_iovec
; i
++) {
2454 r
= journal_file_append_data(f
, iovec
[i
].iov_base
, iovec
[i
].iov_len
, &o
, &p
);
2458 /* When calculating the XOR hash field, we need to take special care if the "keyed-hash"
2459 * journal file flag is on. We use the XOR hash field to quickly determine the identity of a
2460 * specific record, and give records with otherwise identical position (i.e. match in seqno,
2461 * timestamp, …) a stable ordering. But for that we can't have it that the hash of the
2462 * objects in each file is different since they are keyed. Hence let's calculate the Jenkins
2463 * hash here for that. This also has the benefit that cursors for old and new journal files
2464 * are completely identical (they include the XOR hash after all). For classic Jenkins-hash
2465 * files things are easier, we can just take the value from the stored record directly. */
2467 if (JOURNAL_HEADER_KEYED_HASH(f
->header
))
2468 xor_hash
^= jenkins_hash64(iovec
[i
].iov_base
, iovec
[i
].iov_len
);
2470 xor_hash
^= le64toh(o
->data
.hash
);
2472 items
[i
] = (EntryItem
) {
2474 .hash
= le64toh(o
->data
.hash
),
2478 /* Order by the position on disk, in order to improve seek
2479 * times for rotating media. */
2480 typesafe_qsort(items
, n_iovec
, entry_item_cmp
);
2481 n_iovec
= remove_duplicate_entry_items(items
, n_iovec
);
2483 r
= journal_file_append_entry_internal(
2496 /* If the memory mapping triggered a SIGBUS then we return an
2497 * IO error and ignore the error code passed down to us, since
2498 * it is very likely just an effect of a nullified replacement
2501 if (mmap_cache_fd_got_sigbus(f
->cache_fd
))
2504 if (f
->post_change_timer
)
2505 schedule_post_change(f
);
2507 journal_file_post_change(f
);
2512 typedef struct ChainCacheItem
{
2513 uint64_t first
; /* the array at the beginning of the chain */
2514 uint64_t array
; /* the cached array */
2515 uint64_t begin
; /* the first item in the cached array */
2516 uint64_t total
; /* the total number of items in all arrays before this one in the chain */
2517 uint64_t last_index
; /* the last index we looked at, to optimize locality when bisecting */
2520 static void chain_cache_put(
2527 uint64_t last_index
) {
2532 /* If the chain item to cache for this chain is the
2533 * first one it's not worth caching anything */
2537 if (ordered_hashmap_size(h
) >= CHAIN_CACHE_MAX
) {
2538 ci
= ordered_hashmap_steal_first(h
);
2541 ci
= new(ChainCacheItem
, 1);
2548 if (ordered_hashmap_put(h
, &ci
->first
, ci
) < 0) {
2553 assert(ci
->first
== first
);
2558 ci
->last_index
= last_index
;
2561 static int bump_array_index(uint64_t *i
, direction_t direction
, uint64_t n
) {
2564 /* Increase or decrease the specified index, in the right direction. */
2566 if (direction
== DIRECTION_DOWN
) {
2581 static int bump_entry_array(
2586 direction_t direction
,
2596 if (direction
== DIRECTION_DOWN
) {
2598 *ret
= le64toh(o
->entry_array
.next_entry_array_offset
);
2602 /* Entry array chains are a singly linked list, so to find the previous array in the chain, we have
2603 * to start iterating from the top. */
2607 while (p
> 0 && p
!= offset
) {
2608 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, p
, &o
);
2613 p
= le64toh(o
->entry_array
.next_entry_array_offset
);
2616 /* If we can't find the previous entry array in the entry array chain, we're likely dealing with a
2617 * corrupted journal file. */
2626 static int generic_array_get(
2630 direction_t direction
,
2631 Object
**ret_object
,
2632 uint64_t *ret_offset
) {
2634 uint64_t a
, t
= 0, k
;
2641 /* FIXME: fix return value assignment on success. */
2645 /* Try the chain cache first */
2646 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2647 if (ci
&& i
> ci
->total
) {
2654 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2655 if (IN_SET(r
, -EBADMSG
, -EADDRNOTAVAIL
)) {
2656 /* If there's corruption and we're going downwards, let's pretend we reached the
2657 * final entry in the entry array chain. */
2659 if (direction
== DIRECTION_DOWN
)
2662 /* If there's corruption and we're going upwards, move back to the previous entry
2663 * array and start iterating entries from there. */
2665 r
= bump_entry_array(f
, NULL
, a
, first
, DIRECTION_UP
, &a
);
2676 k
= journal_file_entry_array_n_items(f
, o
);
2682 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
2685 /* If we've found the right location, now look for the first non-corrupt entry object (in the right
2689 /* In the first iteration of the while loop, we reuse i, k and o from the previous while
2691 if (i
== UINT64_MAX
) {
2692 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2696 k
= journal_file_entry_array_n_items(f
, o
);
2700 i
= direction
== DIRECTION_DOWN
? 0 : k
- 1;
2706 p
= journal_file_entry_array_item(f
, o
, i
);
2708 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, ret_object
);
2710 /* Let's cache this item for the next invocation */
2711 chain_cache_put(f
->chain_cache
, ci
, first
, a
, journal_file_entry_array_item(f
, o
, 0), t
, i
);
2718 if (!IN_SET(r
, -EADDRNOTAVAIL
, -EBADMSG
))
2721 /* OK, so this entry is borked. Most likely some entry didn't get synced to
2722 * disk properly, let's see if the next one might work for us instead. */
2723 log_debug_errno(r
, "Entry item %" PRIu64
" is bad, skipping over it.", i
);
2725 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2729 } while (bump_array_index(&i
, direction
, k
) > 0);
2731 r
= bump_entry_array(f
, o
, a
, first
, direction
, &a
);
2742 static int generic_array_get_plus_one(
2747 direction_t direction
,
2748 Object
**ret_object
,
2749 uint64_t *ret_offset
) {
2755 /* FIXME: fix return value assignment on success. */
2758 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, ret_object
);
2759 if (IN_SET(r
, -EADDRNOTAVAIL
, -EBADMSG
))
2760 return generic_array_get(f
, first
, 0, direction
, ret_object
, ret_offset
);
2765 *ret_offset
= extra
;
2770 return generic_array_get(f
, first
, i
- 1, direction
, ret_object
, ret_offset
);
2779 static int generic_array_bisect_one(
2781 uint64_t a
, /* offset of entry array object. */
2782 uint64_t i
, /* index of the entry item we will test. */
2784 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2785 direction_t direction
,
2788 uint64_t *ret_offset
) {
2795 assert(test_object
);
2799 assert(i
<= *right
);
2801 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2805 p
= journal_file_entry_array_item(f
, array
, i
);
2809 r
= test_object(f
, p
, needle
);
2810 if (IN_SET(r
, -EBADMSG
, -EADDRNOTAVAIL
)) {
2811 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short.");
2813 return -ENOANO
; /* recognizable error */
2818 if (r
== TEST_FOUND
)
2819 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2821 if (r
== TEST_RIGHT
)
2832 static int generic_array_bisect(
2837 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2838 direction_t direction
,
2839 Object
**ret_object
,
2840 uint64_t *ret_offset
,
2841 uint64_t *ret_idx
) {
2843 /* Given an entry array chain, this function finds the object "closest" to the given needle in the
2844 * chain, taking into account the provided direction. A function can be provided to determine how
2845 * an object is matched against the given needle.
2847 * Given a journal file, the offset of an object and the needle, the test_object() function should
2848 * return TEST_LEFT if the needle is located earlier in the entry array chain, TEST_LEFT if the
2849 * needle is located later in the entry array chain and TEST_FOUND if the object matches the needle.
2850 * If test_object() returns TEST_FOUND for a specific object, that object's information will be used
2851 * to populate the return values of this function. If test_object() never returns TEST_FOUND, the
2852 * return values are populated with the details of one of the objects closest to the needle. If the
2853 * direction is DIRECTION_UP, the earlier object is used. Otherwise, the later object is used.
2856 uint64_t a
, p
, t
= 0, i
= 0, last_p
= 0, last_index
= UINT64_MAX
;
2857 bool subtract_one
= false;
2863 assert(test_object
);
2865 /* Start with the first array in the chain */
2868 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2869 if (ci
&& n
> ci
->total
&& ci
->begin
!= 0) {
2870 /* Ah, we have iterated this bisection array chain previously! Let's see if we can skip ahead
2871 * in the chain, as far as the last time. But we can't jump backwards in the chain, so let's
2872 * check that first. */
2874 r
= test_object(f
, ci
->begin
, needle
);
2878 if (r
== TEST_LEFT
) {
2879 /* OK, what we are looking for is right of the begin of this EntryArray, so let's
2880 * jump straight to previously cached array in the chain */
2885 last_index
= ci
->last_index
;
2890 uint64_t left
= 0, right
, k
, lp
;
2892 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2896 k
= journal_file_entry_array_n_items(f
, array
);
2902 r
= generic_array_bisect_one(f
, a
, right
, needle
, test_object
, direction
, &left
, &right
, &lp
);
2910 if (r
== TEST_RIGHT
) {
2911 /* If we cached the last index we looked at, let's try to not to jump too wildly
2912 * around and see if we can limit the range to look at early to the immediate
2913 * neighbors of the last index we looked at. */
2915 if (last_index
> 0 && last_index
- 1 < right
) {
2916 r
= generic_array_bisect_one(f
, a
, last_index
- 1, needle
, test_object
, direction
, &left
, &right
, NULL
);
2917 if (r
< 0 && r
!= -ENOANO
)
2921 if (last_index
< right
) {
2922 r
= generic_array_bisect_one(f
, a
, last_index
+ 1, needle
, test_object
, direction
, &left
, &right
, NULL
);
2923 if (r
< 0 && r
!= -ENOANO
)
2928 if (left
== right
) {
2929 if (direction
== DIRECTION_UP
)
2930 subtract_one
= true;
2936 assert(left
< right
);
2937 i
= (left
+ right
) / 2;
2939 r
= generic_array_bisect_one(f
, a
, i
, needle
, test_object
, direction
, &left
, &right
, NULL
);
2940 if (r
< 0 && r
!= -ENOANO
)
2946 if (direction
== DIRECTION_UP
) {
2948 subtract_one
= true;
2959 last_index
= UINT64_MAX
;
2960 a
= le64toh(array
->entry_array
.next_entry_array_offset
);
2966 if (subtract_one
&& t
== 0 && i
== 0)
2969 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2973 p
= journal_file_entry_array_item(f
, array
, 0);
2977 /* Let's cache this item for the next invocation */
2978 chain_cache_put(f
->chain_cache
, ci
, first
, a
, p
, t
, subtract_one
? (i
> 0 ? i
-1 : UINT64_MAX
) : i
);
2980 if (subtract_one
&& i
== 0)
2982 else if (subtract_one
)
2983 p
= journal_file_entry_array_item(f
, array
, i
- 1);
2985 p
= journal_file_entry_array_item(f
, array
, i
);
2988 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, ret_object
);
2997 *ret_idx
= t
+ i
+ (subtract_one
? -1 : 0);
3002 static int generic_array_bisect_plus_one(
3008 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
3009 direction_t direction
,
3010 Object
**ret_object
,
3011 uint64_t *ret_offset
,
3012 uint64_t *ret_idx
) {
3015 bool step_back
= false;
3018 assert(test_object
);
3023 /* This bisects the array in object 'first', but first checks
3025 r
= test_object(f
, extra
, needle
);
3029 if (r
== TEST_FOUND
)
3030 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
3032 /* if we are looking with DIRECTION_UP then we need to first
3033 see if in the actual array there is a matching entry, and
3034 return the last one of that. But if there isn't any we need
3035 to return this one. Hence remember this, and return it
3038 step_back
= direction
== DIRECTION_UP
;
3040 if (r
== TEST_RIGHT
) {
3041 if (direction
== DIRECTION_DOWN
)
3047 r
= generic_array_bisect(f
, first
, n
-1, needle
, test_object
, direction
, ret_object
, ret_offset
, ret_idx
);
3049 if (r
== 0 && step_back
)
3052 if (r
> 0 && ret_idx
)
3059 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, ret_object
);
3065 *ret_offset
= extra
;
3073 _pure_
static int test_object_offset(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
3079 else if (p
< needle
)
3085 int journal_file_move_to_entry_by_offset(
3088 direction_t direction
,
3089 Object
**ret_object
,
3090 uint64_t *ret_offset
) {
3095 return generic_array_bisect(
3097 le64toh(f
->header
->entry_array_offset
),
3098 le64toh(f
->header
->n_entries
),
3102 ret_object
, ret_offset
, NULL
);
3105 static int test_object_seqnum(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
3113 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
3117 sq
= le64toh(READ_NOW(o
->entry
.seqnum
));
3120 else if (sq
< needle
)
3126 int journal_file_move_to_entry_by_seqnum(
3129 direction_t direction
,
3130 Object
**ret_object
,
3131 uint64_t *ret_offset
) {
3136 return generic_array_bisect(
3138 le64toh(f
->header
->entry_array_offset
),
3139 le64toh(f
->header
->n_entries
),
3143 ret_object
, ret_offset
, NULL
);
3146 static int test_object_realtime(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
3154 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
3158 rt
= le64toh(READ_NOW(o
->entry
.realtime
));
3161 else if (rt
< needle
)
3167 int journal_file_move_to_entry_by_realtime(
3170 direction_t direction
,
3171 Object
**ret_object
,
3172 uint64_t *ret_offset
) {
3177 return generic_array_bisect(
3179 le64toh(f
->header
->entry_array_offset
),
3180 le64toh(f
->header
->n_entries
),
3182 test_object_realtime
,
3184 ret_object
, ret_offset
, NULL
);
3187 static int test_object_monotonic(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
3195 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
3199 m
= le64toh(READ_NOW(o
->entry
.monotonic
));
3202 else if (m
< needle
)
3208 static int find_data_object_by_boot_id(
3211 Object
**ret_object
,
3212 uint64_t *ret_offset
) {
3214 char t
[STRLEN("_BOOT_ID=") + 32 + 1] = "_BOOT_ID=";
3218 sd_id128_to_string(boot_id
, t
+ 9);
3219 return journal_file_find_data_object(f
, t
, sizeof(t
) - 1, ret_object
, ret_offset
);
3222 int journal_file_move_to_entry_by_monotonic(
3226 direction_t direction
,
3227 Object
**ret_object
,
3228 uint64_t *ret_offset
) {
3235 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, NULL
);
3241 return generic_array_bisect_plus_one(
3243 le64toh(o
->data
.entry_offset
),
3244 le64toh(o
->data
.entry_array_offset
),
3245 le64toh(o
->data
.n_entries
),
3247 test_object_monotonic
,
3249 ret_object
, ret_offset
, NULL
);
3252 void journal_file_reset_location(JournalFile
*f
) {
3255 f
->location_type
= LOCATION_HEAD
;
3256 f
->current_offset
= 0;
3257 f
->current_seqnum
= 0;
3258 f
->current_realtime
= 0;
3259 f
->current_monotonic
= 0;
3260 zero(f
->current_boot_id
);
3261 f
->current_xor_hash
= 0;
3264 void journal_file_save_location(JournalFile
*f
, Object
*o
, uint64_t offset
) {
3268 f
->location_type
= LOCATION_SEEK
;
3269 f
->current_offset
= offset
;
3270 f
->current_seqnum
= le64toh(o
->entry
.seqnum
);
3271 f
->current_realtime
= le64toh(o
->entry
.realtime
);
3272 f
->current_monotonic
= le64toh(o
->entry
.monotonic
);
3273 f
->current_boot_id
= o
->entry
.boot_id
;
3274 f
->current_xor_hash
= le64toh(o
->entry
.xor_hash
);
3277 static bool check_properly_ordered(uint64_t new_offset
, uint64_t old_offset
, direction_t direction
) {
3279 /* Consider it an error if any of the two offsets is uninitialized */
3280 if (old_offset
== 0 || new_offset
== 0)
3283 /* If we go down, the new offset must be larger than the old one. */
3284 return direction
== DIRECTION_DOWN
?
3285 new_offset
> old_offset
:
3286 new_offset
< old_offset
;
3289 int journal_file_next_entry(
3292 direction_t direction
,
3293 Object
**ret_object
,
3294 uint64_t *ret_offset
) {
3302 /* FIXME: fix return value assignment. */
3304 n
= le64toh(READ_NOW(f
->header
->n_entries
));
3309 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
3311 r
= generic_array_bisect(f
,
3312 le64toh(f
->header
->entry_array_offset
),
3313 le64toh(f
->header
->n_entries
),
3322 r
= bump_array_index(&i
, direction
, n
);
3327 /* And jump to it */
3328 r
= generic_array_get(f
, le64toh(f
->header
->entry_array_offset
), i
, direction
, ret_object
, &ofs
);
3332 /* Ensure our array is properly ordered. */
3333 if (p
> 0 && !check_properly_ordered(ofs
, p
, direction
))
3334 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
3335 "%s: entry array not properly ordered at entry %" PRIu64
,
3344 int journal_file_next_entry_for_data(
3347 direction_t direction
,
3348 Object
**ret_object
,
3349 uint64_t *ret_offset
) {
3356 assert(d
->object
.type
== OBJECT_DATA
);
3358 /* FIXME: fix return value assignment. */
3360 n
= le64toh(READ_NOW(d
->data
.n_entries
));
3364 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
3366 r
= generic_array_get_plus_one(f
,
3367 le64toh(d
->data
.entry_offset
),
3368 le64toh(d
->data
.entry_array_offset
),
3381 int journal_file_move_to_entry_by_offset_for_data(
3385 direction_t direction
,
3386 Object
**ret
, uint64_t *ret_offset
) {
3390 assert(d
->object
.type
== OBJECT_DATA
);
3392 return generic_array_bisect_plus_one(
3394 le64toh(d
->data
.entry_offset
),
3395 le64toh(d
->data
.entry_array_offset
),
3396 le64toh(d
->data
.n_entries
),
3400 ret
, ret_offset
, NULL
);
3403 int journal_file_move_to_entry_by_monotonic_for_data(
3408 direction_t direction
,
3409 Object
**ret_object
,
3410 uint64_t *ret_offset
) {
3412 uint64_t b
, z
, entry_offset
, entry_array_offset
, n_entries
;
3418 assert(d
->object
.type
== OBJECT_DATA
);
3420 /* Save all the required data before the data object gets invalidated. */
3421 entry_offset
= le64toh(READ_NOW(d
->data
.entry_offset
));
3422 entry_array_offset
= le64toh(READ_NOW(d
->data
.entry_array_offset
));
3423 n_entries
= le64toh(READ_NOW(d
->data
.n_entries
));
3425 /* First, seek by time */
3426 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &b
);
3432 r
= generic_array_bisect_plus_one(f
,
3433 le64toh(o
->data
.entry_offset
),
3434 le64toh(o
->data
.entry_array_offset
),
3435 le64toh(o
->data
.n_entries
),
3437 test_object_monotonic
,
3443 /* And now, continue seeking until we find an entry that
3444 * exists in both bisection arrays */
3446 r
= journal_file_move_to_object(f
, OBJECT_DATA
, b
, &o
);
3453 r
= generic_array_bisect_plus_one(f
,
3464 r
= generic_array_bisect_plus_one(f
,
3465 le64toh(o
->data
.entry_offset
),
3466 le64toh(o
->data
.entry_array_offset
),
3467 le64toh(o
->data
.n_entries
),
3478 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, q
, ret_object
);
3493 int journal_file_move_to_entry_by_seqnum_for_data(
3497 direction_t direction
,
3498 Object
**ret_object
,
3499 uint64_t *ret_offset
) {
3503 assert(d
->object
.type
== OBJECT_DATA
);
3505 return generic_array_bisect_plus_one(
3507 le64toh(d
->data
.entry_offset
),
3508 le64toh(d
->data
.entry_array_offset
),
3509 le64toh(d
->data
.n_entries
),
3513 ret_object
, ret_offset
, NULL
);
3516 int journal_file_move_to_entry_by_realtime_for_data(
3520 direction_t direction
,
3521 Object
**ret
, uint64_t *ret_offset
) {
3525 assert(d
->object
.type
== OBJECT_DATA
);
3527 return generic_array_bisect_plus_one(
3529 le64toh(d
->data
.entry_offset
),
3530 le64toh(d
->data
.entry_array_offset
),
3531 le64toh(d
->data
.n_entries
),
3533 test_object_realtime
,
3535 ret
, ret_offset
, NULL
);
3538 void journal_file_dump(JournalFile
*f
) {
3546 journal_file_print_header(f
);
3548 p
= le64toh(READ_NOW(f
->header
->header_size
));
3553 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &o
);
3557 s
= journal_object_type_to_string(o
->object
.type
);
3559 switch (o
->object
.type
) {
3564 printf("Type: %s seqnum=%"PRIu64
" monotonic=%"PRIu64
" realtime=%"PRIu64
"\n",
3566 le64toh(o
->entry
.seqnum
),
3567 le64toh(o
->entry
.monotonic
),
3568 le64toh(o
->entry
.realtime
));
3574 printf("Type: %s seqnum=%"PRIu64
" epoch=%"PRIu64
"\n",
3576 le64toh(o
->tag
.seqnum
),
3577 le64toh(o
->tag
.epoch
));
3582 printf("Type: %s \n", s
);
3584 printf("Type: unknown (%i)", o
->object
.type
);
3589 c
= COMPRESSION_FROM_OBJECT(o
);
3590 if (c
> COMPRESSION_NONE
)
3591 printf("Flags: %s\n",
3592 compression_to_string(c
));
3594 if (p
== le64toh(f
->header
->tail_object_offset
))
3597 p
+= ALIGN64(le64toh(o
->object
.size
));
3602 log_error("File corrupt");
3605 /* Note: the lifetime of the compound literal is the immediately surrounding block. */
3606 #define FORMAT_TIMESTAMP_SAFE(t) (FORMAT_TIMESTAMP(t) ?: " --- ")
3608 void journal_file_print_header(JournalFile
*f
) {
3614 printf("File path: %s\n"
3618 "Sequential number ID: %s\n"
3620 "Compatible flags:%s%s%s\n"
3621 "Incompatible flags:%s%s%s%s%s%s\n"
3622 "Header size: %"PRIu64
"\n"
3623 "Arena size: %"PRIu64
"\n"
3624 "Data hash table size: %"PRIu64
"\n"
3625 "Field hash table size: %"PRIu64
"\n"
3626 "Rotate suggested: %s\n"
3627 "Head sequential number: %"PRIu64
" (%"PRIx64
")\n"
3628 "Tail sequential number: %"PRIu64
" (%"PRIx64
")\n"
3629 "Head realtime timestamp: %s (%"PRIx64
")\n"
3630 "Tail realtime timestamp: %s (%"PRIx64
")\n"
3631 "Tail monotonic timestamp: %s (%"PRIx64
")\n"
3632 "Objects: %"PRIu64
"\n"
3633 "Entry objects: %"PRIu64
"\n",
3635 SD_ID128_TO_STRING(f
->header
->file_id
),
3636 SD_ID128_TO_STRING(f
->header
->machine_id
),
3637 SD_ID128_TO_STRING(f
->header
->tail_entry_boot_id
),
3638 SD_ID128_TO_STRING(f
->header
->seqnum_id
),
3639 f
->header
->state
== STATE_OFFLINE
? "OFFLINE" :
3640 f
->header
->state
== STATE_ONLINE
? "ONLINE" :
3641 f
->header
->state
== STATE_ARCHIVED
? "ARCHIVED" : "UNKNOWN",
3642 JOURNAL_HEADER_SEALED(f
->header
) ? " SEALED" : "",
3643 JOURNAL_HEADER_TAIL_ENTRY_BOOT_ID(f
->header
) ? " TAIL_ENTRY_BOOT_ID" : "",
3644 (le32toh(f
->header
->compatible_flags
) & ~HEADER_COMPATIBLE_ANY
) ? " ???" : "",
3645 JOURNAL_HEADER_COMPRESSED_XZ(f
->header
) ? " COMPRESSED-XZ" : "",
3646 JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
) ? " COMPRESSED-LZ4" : "",
3647 JOURNAL_HEADER_COMPRESSED_ZSTD(f
->header
) ? " COMPRESSED-ZSTD" : "",
3648 JOURNAL_HEADER_KEYED_HASH(f
->header
) ? " KEYED-HASH" : "",
3649 JOURNAL_HEADER_COMPACT(f
->header
) ? " COMPACT" : "",
3650 (le32toh(f
->header
->incompatible_flags
) & ~HEADER_INCOMPATIBLE_ANY
) ? " ???" : "",
3651 le64toh(f
->header
->header_size
),
3652 le64toh(f
->header
->arena_size
),
3653 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3654 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
),
3655 yes_no(journal_file_rotate_suggested(f
, 0, LOG_DEBUG
)),
3656 le64toh(f
->header
->head_entry_seqnum
), le64toh(f
->header
->head_entry_seqnum
),
3657 le64toh(f
->header
->tail_entry_seqnum
), le64toh(f
->header
->tail_entry_seqnum
),
3658 FORMAT_TIMESTAMP_SAFE(le64toh(f
->header
->head_entry_realtime
)), le64toh(f
->header
->head_entry_realtime
),
3659 FORMAT_TIMESTAMP_SAFE(le64toh(f
->header
->tail_entry_realtime
)), le64toh(f
->header
->tail_entry_realtime
),
3660 FORMAT_TIMESPAN(le64toh(f
->header
->tail_entry_monotonic
), USEC_PER_MSEC
), le64toh(f
->header
->tail_entry_monotonic
),
3661 le64toh(f
->header
->n_objects
),
3662 le64toh(f
->header
->n_entries
));
3664 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3665 printf("Data objects: %"PRIu64
"\n"
3666 "Data hash table fill: %.1f%%\n",
3667 le64toh(f
->header
->n_data
),
3668 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))));
3670 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3671 printf("Field objects: %"PRIu64
"\n"
3672 "Field hash table fill: %.1f%%\n",
3673 le64toh(f
->header
->n_fields
),
3674 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))));
3676 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
))
3677 printf("Tag objects: %"PRIu64
"\n",
3678 le64toh(f
->header
->n_tags
));
3679 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
3680 printf("Entry array objects: %"PRIu64
"\n",
3681 le64toh(f
->header
->n_entry_arrays
));
3683 if (JOURNAL_HEADER_CONTAINS(f
->header
, field_hash_chain_depth
))
3684 printf("Deepest field hash chain: %" PRIu64
"\n",
3685 f
->header
->field_hash_chain_depth
);
3687 if (JOURNAL_HEADER_CONTAINS(f
->header
, data_hash_chain_depth
))
3688 printf("Deepest data hash chain: %" PRIu64
"\n",
3689 f
->header
->data_hash_chain_depth
);
3691 if (fstat(f
->fd
, &st
) >= 0)
3692 printf("Disk usage: %s\n", FORMAT_BYTES((uint64_t) st
.st_blocks
* 512ULL));
3695 static int journal_file_warn_btrfs(JournalFile
*f
) {
3701 /* Before we write anything, check if the COW logic is turned
3702 * off on btrfs. Given our write pattern that is quite
3703 * unfriendly to COW file systems this should greatly improve
3704 * performance on COW file systems, such as btrfs, at the
3705 * expense of data integrity features (which shouldn't be too
3706 * bad, given that we do our own checksumming). */
3708 r
= fd_is_fs_type(f
->fd
, BTRFS_SUPER_MAGIC
);
3710 return log_ratelimit_warning_errno(r
, JOURNAL_LOG_RATELIMIT
, "Failed to determine if journal is on btrfs: %m");
3714 r
= read_attr_fd(f
->fd
, &attrs
);
3716 return log_ratelimit_warning_errno(r
, JOURNAL_LOG_RATELIMIT
, "Failed to read file attributes: %m");
3718 if (attrs
& FS_NOCOW_FL
) {
3719 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3723 log_ratelimit_notice(JOURNAL_LOG_RATELIMIT
,
3724 "Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3725 "This is likely to slow down journal access substantially, please consider turning "
3726 "off the copy-on-write file attribute on the journal directory, using chattr +C.",
3732 static void journal_default_metrics(JournalMetrics
*m
, int fd
, bool compact
) {
3734 uint64_t fs_size
= 0;
3739 if (fstatvfs(fd
, &ss
) >= 0)
3740 fs_size
= ss
.f_frsize
* ss
.f_blocks
;
3742 log_debug_errno(errno
, "Failed to determine disk size: %m");
3744 if (m
->max_use
== UINT64_MAX
) {
3747 m
->max_use
= CLAMP(PAGE_ALIGN(fs_size
/ 10), /* 10% of file system size */
3748 MAX_USE_LOWER
, MAX_USE_UPPER
);
3750 m
->max_use
= MAX_USE_LOWER
;
3752 m
->max_use
= PAGE_ALIGN(m
->max_use
);
3754 if (m
->max_use
!= 0 && m
->max_use
< JOURNAL_FILE_SIZE_MIN
*2)
3755 m
->max_use
= JOURNAL_FILE_SIZE_MIN
*2;
3758 if (m
->min_use
== UINT64_MAX
) {
3760 m
->min_use
= CLAMP(PAGE_ALIGN(fs_size
/ 50), /* 2% of file system size */
3761 MIN_USE_LOW
, MIN_USE_HIGH
);
3763 m
->min_use
= MIN_USE_LOW
;
3766 if (m
->min_use
> m
->max_use
)
3767 m
->min_use
= m
->max_use
;
3769 if (m
->max_size
== UINT64_MAX
)
3770 m
->max_size
= MIN(PAGE_ALIGN(m
->max_use
/ 8), /* 8 chunks */
3773 m
->max_size
= PAGE_ALIGN(m
->max_size
);
3775 if (compact
&& m
->max_size
> JOURNAL_COMPACT_SIZE_MAX
)
3776 m
->max_size
= JOURNAL_COMPACT_SIZE_MAX
;
3778 if (m
->max_size
!= 0) {
3779 if (m
->max_size
< JOURNAL_FILE_SIZE_MIN
)
3780 m
->max_size
= JOURNAL_FILE_SIZE_MIN
;
3782 if (m
->max_use
!= 0 && m
->max_size
*2 > m
->max_use
)
3783 m
->max_use
= m
->max_size
*2;
3786 if (m
->min_size
== UINT64_MAX
)
3787 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3789 m
->min_size
= CLAMP(PAGE_ALIGN(m
->min_size
),
3790 JOURNAL_FILE_SIZE_MIN
,
3791 m
->max_size
?: UINT64_MAX
);
3793 if (m
->keep_free
== UINT64_MAX
) {
3795 m
->keep_free
= MIN(PAGE_ALIGN(fs_size
/ 20), /* 5% of file system size */
3798 m
->keep_free
= DEFAULT_KEEP_FREE
;
3801 if (m
->n_max_files
== UINT64_MAX
)
3802 m
->n_max_files
= DEFAULT_N_MAX_FILES
;
3804 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64
,
3805 FORMAT_BYTES(m
->min_use
),
3806 FORMAT_BYTES(m
->max_use
),
3807 FORMAT_BYTES(m
->max_size
),
3808 FORMAT_BYTES(m
->min_size
),
3809 FORMAT_BYTES(m
->keep_free
),
3813 int journal_file_open(
3817 JournalFileFlags file_flags
,
3819 uint64_t compress_threshold_bytes
,
3820 JournalMetrics
*metrics
,
3821 MMapCache
*mmap_cache
,
3822 JournalFile
*template,
3823 JournalFile
**ret
) {
3825 bool newly_created
= false;
3830 assert(fd
>= 0 || fname
);
3831 assert(file_flags
>= 0);
3832 assert(file_flags
<= _JOURNAL_FILE_FLAGS_MAX
);
3836 if (!IN_SET((open_flags
& O_ACCMODE
), O_RDONLY
, O_RDWR
))
3839 if ((open_flags
& O_ACCMODE
) == O_RDONLY
&& FLAGS_SET(open_flags
, O_CREAT
))
3842 if (fname
&& (open_flags
& O_CREAT
) && !endswith(fname
, ".journal"))
3845 f
= new(JournalFile
, 1);
3849 *f
= (JournalFile
) {
3852 .open_flags
= open_flags
,
3853 .compress_threshold_bytes
= compress_threshold_bytes
== UINT64_MAX
?
3854 DEFAULT_COMPRESS_THRESHOLD
:
3855 MAX(MIN_COMPRESS_THRESHOLD
, compress_threshold_bytes
),
3856 .strict_order
= FLAGS_SET(file_flags
, JOURNAL_STRICT_ORDER
),
3857 .newest_boot_id_prioq_idx
= PRIOQ_IDX_NULL
,
3861 f
->path
= strdup(fname
);
3869 /* If we don't know the path, fill in something explanatory and vaguely useful */
3870 if (asprintf(&f
->path
, "/proc/self/%i", fd
) < 0) {
3876 f
->chain_cache
= ordered_hashmap_new(&uint64_hash_ops
);
3877 if (!f
->chain_cache
) {
3883 /* We pass O_NONBLOCK here, so that in case somebody pointed us to some character device node or FIFO
3884 * or so, we likely fail quickly than block for long. For regular files O_NONBLOCK has no effect, hence
3885 * it doesn't hurt in that case. */
3887 f
->fd
= openat_report_new(AT_FDCWD
, f
->path
, f
->open_flags
|O_CLOEXEC
|O_NONBLOCK
, f
->mode
, &newly_created
);
3893 /* fds we opened here by us should also be closed by us. */
3896 r
= fd_nonblock(f
->fd
, false);
3900 if (!newly_created
) {
3901 r
= journal_file_fstat(f
);
3906 r
= journal_file_fstat(f
);
3910 /* If we just got the fd passed in, we don't really know if we created the file anew */
3911 newly_created
= f
->last_stat
.st_size
== 0 && journal_file_writable(f
);
3914 f
->cache_fd
= mmap_cache_add_fd(mmap_cache
, f
->fd
, mmap_prot_from_open_flags(open_flags
));
3920 if (newly_created
) {
3921 (void) journal_file_warn_btrfs(f
);
3923 /* Let's attach the creation time to the journal file, so that the vacuuming code knows the age of this
3924 * file even if the file might end up corrupted one day... Ideally we'd just use the creation time many
3925 * file systems maintain for each file, but the API to query this is very new, hence let's emulate this
3926 * via extended attributes. If extended attributes are not supported we'll just skip this, and rely
3927 * solely on mtime/atime/ctime of the file. */
3928 (void) fd_setcrtime(f
->fd
, 0);
3930 r
= journal_file_init_header(f
, file_flags
, template);
3934 r
= journal_file_fstat(f
);
3939 if (f
->last_stat
.st_size
< (off_t
) HEADER_SIZE_MIN
) {
3944 r
= mmap_cache_fd_get(f
->cache_fd
, CONTEXT_HEADER
, true, 0, PAGE_ALIGN(sizeof(Header
)), &f
->last_stat
, &h
);
3946 /* Some file systems (jffs2 or p9fs) don't support mmap() properly (or only read-only
3947 * mmap()), and return EINVAL in that case. Let's propagate that as a more recognizable error
3957 if (!newly_created
) {
3958 r
= journal_file_verify_header(f
);
3964 if (!newly_created
&& journal_file_writable(f
) && JOURNAL_HEADER_SEALED(f
->header
)) {
3965 r
= journal_file_fss_load(f
);
3971 if (journal_file_writable(f
)) {
3973 journal_default_metrics(metrics
, f
->fd
, JOURNAL_HEADER_COMPACT(f
->header
));
3974 f
->metrics
= *metrics
;
3975 } else if (template)
3976 f
->metrics
= template->metrics
;
3978 r
= journal_file_refresh_header(f
);
3984 r
= journal_file_hmac_setup(f
);
3989 if (newly_created
) {
3990 r
= journal_file_setup_field_hash_table(f
);
3994 r
= journal_file_setup_data_hash_table(f
);
3999 r
= journal_file_append_first_tag(f
);
4005 if (mmap_cache_fd_got_sigbus(f
->cache_fd
)) {
4010 if (template && template->post_change_timer
) {
4011 r
= journal_file_enable_post_change_timer(
4013 sd_event_source_get_event(template->post_change_timer
),
4014 template->post_change_timer_period
);
4020 /* The file is opened now successfully, thus we take possession of any passed in fd. */
4023 if (DEBUG_LOGGING
) {
4024 static int last_seal
= -1, last_keyed_hash
= -1;
4025 static Compression last_compression
= _COMPRESSION_INVALID
;
4026 static uint64_t last_bytes
= UINT64_MAX
;
4028 if (last_seal
!= JOURNAL_HEADER_SEALED(f
->header
) ||
4029 last_keyed_hash
!= JOURNAL_HEADER_KEYED_HASH(f
->header
) ||
4030 last_compression
!= JOURNAL_FILE_COMPRESSION(f
) ||
4031 last_bytes
!= f
->compress_threshold_bytes
) {
4033 log_debug("Journal effective settings seal=%s keyed_hash=%s compress=%s compress_threshold_bytes=%s",
4034 yes_no(JOURNAL_HEADER_SEALED(f
->header
)), yes_no(JOURNAL_HEADER_KEYED_HASH(f
->header
)),
4035 compression_to_string(JOURNAL_FILE_COMPRESSION(f
)), FORMAT_BYTES(f
->compress_threshold_bytes
));
4036 last_seal
= JOURNAL_HEADER_SEALED(f
->header
);
4037 last_keyed_hash
= JOURNAL_HEADER_KEYED_HASH(f
->header
);
4038 last_compression
= JOURNAL_FILE_COMPRESSION(f
);
4039 last_bytes
= f
->compress_threshold_bytes
;
4047 if (f
->cache_fd
&& mmap_cache_fd_got_sigbus(f
->cache_fd
))
4050 (void) journal_file_close(f
);
4052 if (newly_created
&& fd
< 0)
4053 (void) unlink(fname
);
4058 int journal_file_parse_uid_from_filename(const char *path
, uid_t
*ret_uid
) {
4059 _cleanup_free_
char *buf
= NULL
, *p
= NULL
;
4060 const char *a
, *b
, *at
;
4063 /* This helper returns -EREMOTE when the filename doesn't match user online/offline journal
4064 * pattern. Hence it currently doesn't parse archived or disposed user journals. */
4069 r
= path_extract_filename(path
, &p
);
4072 if (r
== O_DIRECTORY
)
4075 a
= startswith(p
, "user-");
4078 b
= endswith(p
, ".journal");
4082 at
= strchr(a
, '@');
4086 buf
= strndup(a
, b
-a
);
4090 return parse_uid(buf
, ret_uid
);
4093 int journal_file_archive(JournalFile
*f
, char **ret_previous_path
) {
4094 _cleanup_free_
char *p
= NULL
;
4098 if (!journal_file_writable(f
))
4101 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
4102 * rotation, since we don't know the actual path, and couldn't rename the file hence. */
4103 if (path_startswith(f
->path
, "/proc/self/fd"))
4106 if (!endswith(f
->path
, ".journal"))
4109 if (asprintf(&p
, "%.*s@" SD_ID128_FORMAT_STR
"-%016"PRIx64
"-%016"PRIx64
".journal",
4110 (int) strlen(f
->path
) - 8, f
->path
,
4111 SD_ID128_FORMAT_VAL(f
->header
->seqnum_id
),
4112 le64toh(f
->header
->head_entry_seqnum
),
4113 le64toh(f
->header
->head_entry_realtime
)) < 0)
4116 /* Try to rename the file to the archived version. If the file already was deleted, we'll get ENOENT, let's
4117 * ignore that case. */
4118 if (rename(f
->path
, p
) < 0 && errno
!= ENOENT
)
4121 /* Sync the rename to disk */
4122 (void) fsync_directory_of_file(f
->fd
);
4124 if (ret_previous_path
)
4125 *ret_previous_path
= f
->path
;
4129 f
->path
= TAKE_PTR(p
);
4131 /* Set as archive so offlining commits w/state=STATE_ARCHIVED. Previously we would set old_file->header->state
4132 * to STATE_ARCHIVED directly here, but journal_file_set_offline() short-circuits when state != STATE_ONLINE,
4133 * which would result in the rotated journal never getting fsync() called before closing. Now we simply queue
4134 * the archive state by setting an archive bit, leaving the state as STATE_ONLINE so proper offlining
4141 int journal_file_dispose(int dir_fd
, const char *fname
) {
4142 _cleanup_free_
char *p
= NULL
;
4146 /* Renames a journal file to *.journal~, i.e. to mark it as corrupted or otherwise uncleanly shutdown. Note that
4147 * this is done without looking into the file or changing any of its contents. The idea is that this is called
4148 * whenever something is suspicious and we want to move the file away and make clear that it is not accessed
4149 * for writing anymore. */
4151 if (!endswith(fname
, ".journal"))
4154 if (asprintf(&p
, "%.*s@%016" PRIx64
"-%016" PRIx64
".journal~",
4155 (int) strlen(fname
) - 8, fname
,
4156 now(CLOCK_REALTIME
),
4160 if (renameat(dir_fd
, fname
, dir_fd
, p
) < 0)
4166 int journal_file_copy_entry(
4172 sd_id128_t
*seqnum_id
) {
4174 _cleanup_free_ EntryItem
*items_alloc
= NULL
;
4176 uint64_t n
, m
= 0, xor_hash
= 0;
4186 if (!journal_file_writable(to
))
4189 ts
= (dual_timestamp
) {
4190 .monotonic
= le64toh(o
->entry
.monotonic
),
4191 .realtime
= le64toh(o
->entry
.realtime
),
4193 boot_id
= o
->entry
.boot_id
;
4195 n
= journal_file_entry_n_items(from
, o
);
4199 if (n
< ALLOCA_MAX
/ sizeof(EntryItem
) / 2)
4200 items
= newa(EntryItem
, n
);
4202 items_alloc
= new(EntryItem
, n
);
4206 items
= items_alloc
;
4209 for (uint64_t i
= 0; i
< n
; i
++) {
4215 q
= journal_file_entry_item_object_offset(from
, o
, i
);
4216 r
= journal_file_data_payload(from
, NULL
, q
, NULL
, 0, 0, &data
, &l
);
4217 if (IN_SET(r
, -EADDRNOTAVAIL
, -EBADMSG
)) {
4218 log_debug_errno(r
, "Entry item %"PRIu64
" data object is bad, skipping over it: %m", i
);
4228 r
= journal_file_append_data(to
, data
, l
, &u
, &h
);
4232 if (JOURNAL_HEADER_KEYED_HASH(to
->header
))
4233 xor_hash
^= jenkins_hash64(data
, l
);
4235 xor_hash
^= le64toh(u
->data
.hash
);
4237 items
[m
++] = (EntryItem
) {
4239 .hash
= le64toh(u
->data
.hash
),
4243 /* The above journal_file_data_payload() may clear or overwrite cached object. Hence, we need
4244 * to re-read the object from the cache. */
4245 r
= journal_file_move_to_object(from
, OBJECT_ENTRY
, p
, &o
);
4253 r
= journal_file_append_entry_internal(
4257 &from
->header
->machine_id
,
4263 /* ret_object= */ NULL
,
4264 /* ret_offset= */ NULL
);
4266 if (mmap_cache_fd_got_sigbus(to
->cache_fd
))
4272 void journal_reset_metrics(JournalMetrics
*m
) {
4275 /* Set everything to "pick automatic values". */
4277 *m
= (JournalMetrics
) {
4278 .min_use
= UINT64_MAX
,
4279 .max_use
= UINT64_MAX
,
4280 .min_size
= UINT64_MAX
,
4281 .max_size
= UINT64_MAX
,
4282 .keep_free
= UINT64_MAX
,
4283 .n_max_files
= UINT64_MAX
,
4287 int journal_file_get_cutoff_realtime_usec(JournalFile
*f
, usec_t
*ret_from
, usec_t
*ret_to
) {
4290 assert(ret_from
|| ret_to
);
4293 if (f
->header
->head_entry_realtime
== 0)
4296 *ret_from
= le64toh(f
->header
->head_entry_realtime
);
4300 if (f
->header
->tail_entry_realtime
== 0)
4303 *ret_to
= le64toh(f
->header
->tail_entry_realtime
);
4309 int journal_file_get_cutoff_monotonic_usec(JournalFile
*f
, sd_id128_t boot_id
, usec_t
*ret_from
, usec_t
*ret_to
) {
4315 assert(ret_from
|| ret_to
);
4317 /* FIXME: fix return value assignment on success with 0. */
4319 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &p
);
4323 if (le64toh(o
->data
.n_entries
) <= 0)
4327 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, le64toh(o
->data
.entry_offset
), &o
);
4331 *ret_from
= le64toh(o
->entry
.monotonic
);
4335 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
4339 r
= generic_array_get_plus_one(f
,
4340 le64toh(o
->data
.entry_offset
),
4341 le64toh(o
->data
.entry_array_offset
),
4342 le64toh(o
->data
.n_entries
) - 1,
4348 *ret_to
= le64toh(o
->entry
.monotonic
);
4354 bool journal_file_rotate_suggested(JournalFile
*f
, usec_t max_file_usec
, int log_level
) {
4358 /* If we gained new header fields we gained new features,
4359 * hence suggest a rotation */
4360 if (le64toh(f
->header
->header_size
) < sizeof(Header
)) {
4361 log_ratelimit_full(log_level
, JOURNAL_LOG_RATELIMIT
,
4362 "%s uses an outdated header, suggesting rotation.", f
->path
);
4366 /* Let's check if the hash tables grew over a certain fill level (75%, borrowing this value from
4367 * Java's hash table implementation), and if so suggest a rotation. To calculate the fill level we
4368 * need the n_data field, which only exists in newer versions. */
4370 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
4371 if (le64toh(f
->header
->n_data
) * 4ULL > (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
4373 log_level
, JOURNAL_LOG_RATELIMIT
,
4374 "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.",
4376 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))),
4377 le64toh(f
->header
->n_data
),
4378 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
4379 (uint64_t) f
->last_stat
.st_size
,
4380 f
->last_stat
.st_size
/ le64toh(f
->header
->n_data
));
4384 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
4385 if (le64toh(f
->header
->n_fields
) * 4ULL > (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
4387 log_level
, JOURNAL_LOG_RATELIMIT
,
4388 "Field hash table of %s has a fill level at %.1f (%"PRIu64
" of %"PRIu64
" items), suggesting rotation.",
4390 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))),
4391 le64toh(f
->header
->n_fields
),
4392 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
));
4396 /* If there are too many hash collisions somebody is most likely playing games with us. Hence, if our
4397 * longest chain is longer than some threshold, let's suggest rotation. */
4398 if (JOURNAL_HEADER_CONTAINS(f
->header
, data_hash_chain_depth
) &&
4399 le64toh(f
->header
->data_hash_chain_depth
) > HASH_CHAIN_DEPTH_MAX
) {
4401 log_level
, JOURNAL_LOG_RATELIMIT
,
4402 "Data hash table of %s has deepest hash chain of length %" PRIu64
", suggesting rotation.",
4403 f
->path
, le64toh(f
->header
->data_hash_chain_depth
));
4407 if (JOURNAL_HEADER_CONTAINS(f
->header
, field_hash_chain_depth
) &&
4408 le64toh(f
->header
->field_hash_chain_depth
) > HASH_CHAIN_DEPTH_MAX
) {
4410 log_level
, JOURNAL_LOG_RATELIMIT
,
4411 "Field hash table of %s has deepest hash chain of length at %" PRIu64
", suggesting rotation.",
4412 f
->path
, le64toh(f
->header
->field_hash_chain_depth
));
4416 /* Are the data objects properly indexed by field objects? */
4417 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
4418 JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
4419 le64toh(f
->header
->n_data
) > 0 &&
4420 le64toh(f
->header
->n_fields
) == 0) {
4422 log_level
, JOURNAL_LOG_RATELIMIT
,
4423 "Data objects of %s are not indexed by field objects, suggesting rotation.",
4428 if (max_file_usec
> 0) {
4431 h
= le64toh(f
->header
->head_entry_realtime
);
4432 t
= now(CLOCK_REALTIME
);
4434 if (h
> 0 && t
> h
+ max_file_usec
) {
4436 log_level
, JOURNAL_LOG_RATELIMIT
,
4437 "Oldest entry in %s is older than the configured file retention duration (%s), suggesting rotation.",
4438 f
->path
, FORMAT_TIMESPAN(max_file_usec
, USEC_PER_SEC
));
4446 static const char * const journal_object_type_table
[] = {
4447 [OBJECT_UNUSED
] = "unused",
4448 [OBJECT_DATA
] = "data",
4449 [OBJECT_FIELD
] = "field",
4450 [OBJECT_ENTRY
] = "entry",
4451 [OBJECT_DATA_HASH_TABLE
] = "data hash table",
4452 [OBJECT_FIELD_HASH_TABLE
] = "field hash table",
4453 [OBJECT_ENTRY_ARRAY
] = "entry array",
4454 [OBJECT_TAG
] = "tag",
4457 DEFINE_STRING_TABLE_LOOKUP_TO_STRING(journal_object_type
, ObjectType
);