1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
6 #include <linux/magic.h>
10 #include <sys/statvfs.h>
16 #include "alloc-util.h"
17 #include "chattr-util.h"
21 #include "format-util.h"
23 #include "id128-util.h"
24 #include "journal-authenticate.h"
25 #include "journal-def.h"
26 #include "journal-file.h"
27 #include "journal-internal.h"
29 #include "memory-util.h"
30 #include "missing_threads.h"
31 #include "path-util.h"
33 #include "random-util.h"
35 #include "sort-util.h"
36 #include "stat-util.h"
37 #include "string-table.h"
38 #include "string-util.h"
40 #include "sync-util.h"
41 #include "user-util.h"
42 #include "xattr-util.h"
44 #define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
45 #define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
47 #define DEFAULT_COMPRESS_THRESHOLD (512ULL)
48 #define MIN_COMPRESS_THRESHOLD (8ULL)
50 /* This is the minimum journal file size */
51 #define JOURNAL_FILE_SIZE_MIN (512 * 1024ULL) /* 512 KiB */
52 #define JOURNAL_COMPACT_SIZE_MAX UINT32_MAX /* 4 GiB */
54 /* These are the lower and upper bounds if we deduce the max_use value
55 * from the file system size */
56 #define MAX_USE_LOWER (1 * 1024 * 1024ULL) /* 1 MiB */
57 #define MAX_USE_UPPER (4 * 1024 * 1024 * 1024ULL) /* 4 GiB */
59 /* Those are the lower and upper bounds for the minimal use limit,
60 * i.e. how much we'll use even if keep_free suggests otherwise. */
61 #define MIN_USE_LOW (1 * 1024 * 1024ULL) /* 1 MiB */
62 #define MIN_USE_HIGH (16 * 1024 * 1024ULL) /* 16 MiB */
64 /* This is the upper bound if we deduce max_size from max_use */
65 #define MAX_SIZE_UPPER (128 * 1024 * 1024ULL) /* 128 MiB */
67 /* This is the upper bound if we deduce the keep_free value from the
69 #define KEEP_FREE_UPPER (4 * 1024 * 1024 * 1024ULL) /* 4 GiB */
71 /* This is the keep_free value when we can't determine the system
73 #define DEFAULT_KEEP_FREE (1024 * 1024ULL) /* 1 MB */
75 /* This is the default maximum number of journal files to keep around. */
76 #define DEFAULT_N_MAX_FILES 100
78 /* n_data was the first entry we added after the initial file format design */
79 #define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
81 /* How many entries to keep in the entry array chain cache at max */
82 #define CHAIN_CACHE_MAX 20
84 /* How much to increase the journal file size at once each time we allocate something new. */
85 #define FILE_SIZE_INCREASE (8 * 1024 * 1024ULL) /* 8MB */
87 /* Reread fstat() of the file for detecting deletions at least this often */
88 #define LAST_STAT_REFRESH_USEC (5*USEC_PER_SEC)
90 /* The mmap context to use for the header we pick as one above the last defined typed */
91 #define CONTEXT_HEADER _OBJECT_TYPE_MAX
93 /* Longest hash chain to rotate after */
94 #define HASH_CHAIN_DEPTH_MAX 100
97 # pragma GCC diagnostic ignored "-Waddress-of-packed-member"
100 static int mmap_prot_from_open_flags(int flags
) {
101 switch (flags
& O_ACCMODE
) {
107 return PROT_READ
|PROT_WRITE
;
109 assert_not_reached();
113 int journal_file_tail_end_by_pread(JournalFile
*f
, uint64_t *ret_offset
) {
121 /* Same as journal_file_tail_end_by_mmap() below, but operates with pread() to avoid the mmap cache
122 * (and thus is thread safe) */
124 p
= le64toh(f
->header
->tail_object_offset
);
126 p
= le64toh(f
->header
->header_size
);
131 r
= journal_file_read_object_header(f
, OBJECT_UNUSED
, p
, &tail
);
135 sz
= le64toh(tail
.object
.size
);
136 if (sz
> UINT64_MAX
- sizeof(uint64_t) + 1)
140 if (p
> UINT64_MAX
- sz
)
151 int journal_file_tail_end_by_mmap(JournalFile
*f
, uint64_t *ret_offset
) {
159 /* Same as journal_file_tail_end_by_pread() above, but operates with the usual mmap logic */
161 p
= le64toh(f
->header
->tail_object_offset
);
163 p
= le64toh(f
->header
->header_size
);
168 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &tail
);
172 sz
= le64toh(READ_NOW(tail
->object
.size
));
173 if (sz
> UINT64_MAX
- sizeof(uint64_t) + 1)
177 if (p
> UINT64_MAX
- sz
)
188 int journal_file_set_offline_thread_join(JournalFile
*f
) {
193 if (f
->offline_state
== OFFLINE_JOINED
)
196 r
= pthread_join(f
->offline_thread
, NULL
);
200 f
->offline_state
= OFFLINE_JOINED
;
202 if (mmap_cache_fd_got_sigbus(f
->cache_fd
))
208 static int journal_file_set_online(JournalFile
*f
) {
213 if (!journal_file_writable(f
))
216 if (f
->fd
< 0 || !f
->header
)
220 switch (f
->offline_state
) {
222 /* No offline thread, no need to wait. */
226 case OFFLINE_SYNCING
: {
227 OfflineState tmp_state
= OFFLINE_SYNCING
;
228 if (!__atomic_compare_exchange_n(&f
->offline_state
, &tmp_state
, OFFLINE_CANCEL
,
229 false, __ATOMIC_SEQ_CST
, __ATOMIC_SEQ_CST
))
232 /* Canceled syncing prior to offlining, no need to wait. */
236 case OFFLINE_AGAIN_FROM_SYNCING
: {
237 OfflineState tmp_state
= OFFLINE_AGAIN_FROM_SYNCING
;
238 if (!__atomic_compare_exchange_n(&f
->offline_state
, &tmp_state
, OFFLINE_CANCEL
,
239 false, __ATOMIC_SEQ_CST
, __ATOMIC_SEQ_CST
))
242 /* Canceled restart from syncing, no need to wait. */
246 case OFFLINE_AGAIN_FROM_OFFLINING
: {
247 OfflineState tmp_state
= OFFLINE_AGAIN_FROM_OFFLINING
;
248 if (!__atomic_compare_exchange_n(&f
->offline_state
, &tmp_state
, OFFLINE_CANCEL
,
249 false, __ATOMIC_SEQ_CST
, __ATOMIC_SEQ_CST
))
252 /* Canceled restart from offlining, must wait for offlining to complete however. */
257 r
= journal_file_set_offline_thread_join(f
);
267 if (mmap_cache_fd_got_sigbus(f
->cache_fd
))
270 switch (f
->header
->state
) {
275 f
->header
->state
= STATE_ONLINE
;
284 JournalFile
* journal_file_close(JournalFile
*f
) {
288 assert(f
->newest_boot_id_prioq_idx
== PRIOQ_IDX_NULL
);
291 mmap_cache_fd_free(f
->cache_fd
);
297 ordered_hashmap_free_free(f
->chain_cache
);
300 free(f
->compress_buffer
);
305 munmap(f
->fss_file
, PAGE_ALIGN(f
->fss_file_size
));
307 free(f
->fsprg_state
);
312 gcry_md_close(f
->hmac
);
318 static bool keyed_hash_requested(void) {
319 static thread_local
int cached
= -1;
323 r
= getenv_bool("SYSTEMD_JOURNAL_KEYED_HASH");
326 log_debug_errno(r
, "Failed to parse $SYSTEMD_JOURNAL_KEYED_HASH environment variable, ignoring: %m");
335 static bool compact_mode_requested(void) {
336 static thread_local
int cached
= -1;
340 r
= getenv_bool("SYSTEMD_JOURNAL_COMPACT");
343 log_debug_errno(r
, "Failed to parse $SYSTEMD_JOURNAL_COMPACT environment variable, ignoring: %m");
353 static Compression
getenv_compression(void) {
358 e
= getenv("SYSTEMD_JOURNAL_COMPRESS");
360 return DEFAULT_COMPRESSION
;
362 r
= parse_boolean(e
);
364 return r
? DEFAULT_COMPRESSION
: COMPRESSION_NONE
;
366 c
= compression_from_string(e
);
368 log_debug_errno(c
, "Failed to parse SYSTEMD_JOURNAL_COMPRESS value, ignoring: %s", e
);
369 return DEFAULT_COMPRESSION
;
372 if (!compression_supported(c
)) {
373 log_debug("Unsupported compression algorithm specified, ignoring: %s", e
);
374 return DEFAULT_COMPRESSION
;
381 static Compression
compression_requested(void) {
383 static thread_local Compression cached
= _COMPRESSION_INVALID
;
386 cached
= getenv_compression();
390 return COMPRESSION_NONE
;
394 static int journal_file_init_header(
396 JournalFileFlags file_flags
,
397 JournalFile
*template) {
406 /* Try to load the FSPRG state, and if we can't, then just don't do sealing */
407 seal
= FLAGS_SET(file_flags
, JOURNAL_SEAL
) && journal_file_fss_load(f
) >= 0;
411 .header_size
= htole64(ALIGN64(sizeof(h
))),
412 .incompatible_flags
= htole32(
413 FLAGS_SET(file_flags
, JOURNAL_COMPRESS
) * COMPRESSION_TO_HEADER_INCOMPATIBLE_FLAG(compression_requested()) |
414 keyed_hash_requested() * HEADER_INCOMPATIBLE_KEYED_HASH
|
415 compact_mode_requested() * HEADER_INCOMPATIBLE_COMPACT
),
416 .compatible_flags
= htole32(
417 (seal
* HEADER_COMPATIBLE_SEALED
) |
418 HEADER_COMPATIBLE_TAIL_ENTRY_BOOT_ID
),
421 assert_cc(sizeof(h
.signature
) == sizeof(HEADER_SIGNATURE
));
422 memcpy(h
.signature
, HEADER_SIGNATURE
, sizeof(HEADER_SIGNATURE
));
424 r
= sd_id128_randomize(&h
.file_id
);
428 r
= sd_id128_get_machine(&h
.machine_id
);
429 if (r
< 0 && !ERRNO_IS_MACHINE_ID_UNSET(r
))
430 return r
; /* If we have no valid machine ID (test environment?), let's simply leave the
431 * machine ID field all zeroes. */
434 h
.seqnum_id
= template->header
->seqnum_id
;
435 h
.tail_entry_seqnum
= template->header
->tail_entry_seqnum
;
437 h
.seqnum_id
= h
.file_id
;
439 k
= pwrite(f
->fd
, &h
, sizeof(h
), 0);
448 static int journal_file_refresh_header(JournalFile
*f
) {
454 /* We used to update the header's boot ID field here, but we don't do that anymore, as per
455 * HEADER_COMPATIBLE_TAIL_ENTRY_BOOT_ID */
457 r
= journal_file_set_online(f
);
459 /* Sync the online state to disk; likely just created a new file, also sync the directory this file
461 (void) fsync_full(f
->fd
);
466 static bool warn_wrong_flags(const JournalFile
*f
, bool compatible
) {
467 const uint32_t any
= compatible
? HEADER_COMPATIBLE_ANY
: HEADER_INCOMPATIBLE_ANY
,
468 supported
= compatible
? HEADER_COMPATIBLE_SUPPORTED
: HEADER_INCOMPATIBLE_SUPPORTED
;
469 const char *type
= compatible
? "compatible" : "incompatible";
475 flags
= le32toh(compatible
? f
->header
->compatible_flags
: f
->header
->incompatible_flags
);
477 if (flags
& ~supported
) {
479 log_debug("Journal file %s has unknown %s flags 0x%"PRIx32
,
480 f
->path
, type
, flags
& ~any
);
481 flags
= (flags
& any
) & ~supported
;
485 _cleanup_free_
char *t
= NULL
;
488 if (flags
& HEADER_COMPATIBLE_SEALED
)
489 strv
[n
++] = "sealed";
491 if (flags
& HEADER_INCOMPATIBLE_COMPRESSED_XZ
)
492 strv
[n
++] = "xz-compressed";
493 if (flags
& HEADER_INCOMPATIBLE_COMPRESSED_LZ4
)
494 strv
[n
++] = "lz4-compressed";
495 if (flags
& HEADER_INCOMPATIBLE_COMPRESSED_ZSTD
)
496 strv
[n
++] = "zstd-compressed";
497 if (flags
& HEADER_INCOMPATIBLE_KEYED_HASH
)
498 strv
[n
++] = "keyed-hash";
499 if (flags
& HEADER_INCOMPATIBLE_COMPACT
)
500 strv
[n
++] = "compact";
503 assert(n
< ELEMENTSOF(strv
));
505 t
= strv_join((char**) strv
, ", ");
506 log_debug("Journal file %s uses %s %s %s disabled at compilation time.",
507 f
->path
, type
, n
> 1 ? "flags" : "flag", strnull(t
));
515 static bool offset_is_valid(uint64_t offset
, uint64_t header_size
, uint64_t tail_object_offset
) {
518 if (!VALID64(offset
))
520 if (offset
< header_size
)
522 if (offset
> tail_object_offset
)
527 static bool hash_table_is_valid(uint64_t offset
, uint64_t size
, uint64_t header_size
, uint64_t arena_size
, uint64_t tail_object_offset
) {
528 if ((offset
== 0) != (size
== 0))
532 if (offset
<= offsetof(Object
, hash_table
.items
))
534 offset
-= offsetof(Object
, hash_table
.items
);
535 if (!offset_is_valid(offset
, header_size
, tail_object_offset
))
537 assert(offset
<= header_size
+ arena_size
);
538 if (size
> header_size
+ arena_size
- offset
)
543 static int journal_file_verify_header(JournalFile
*f
) {
544 uint64_t arena_size
, header_size
;
549 if (memcmp(f
->header
->signature
, HEADER_SIGNATURE
, 8))
552 /* In both read and write mode we refuse to open files with incompatible
553 * flags we don't know. */
554 if (warn_wrong_flags(f
, false))
555 return -EPROTONOSUPPORT
;
557 /* When open for writing we refuse to open files with compatible flags, too. */
558 if (journal_file_writable(f
) && warn_wrong_flags(f
, true))
559 return -EPROTONOSUPPORT
;
561 if (f
->header
->state
>= _STATE_MAX
)
564 header_size
= le64toh(READ_NOW(f
->header
->header_size
));
566 /* The first addition was n_data, so check that we are at least this large */
567 if (header_size
< HEADER_SIZE_MIN
)
570 /* When open for writing we refuse to open files with a mismatch of the header size, i.e. writing to
571 * files implementing older or new header structures. */
572 if (journal_file_writable(f
) && header_size
!= sizeof(Header
))
573 return -EPROTONOSUPPORT
;
575 if (JOURNAL_HEADER_SEALED(f
->header
) && !JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
578 arena_size
= le64toh(READ_NOW(f
->header
->arena_size
));
580 if (UINT64_MAX
- header_size
< arena_size
|| header_size
+ arena_size
> (uint64_t) f
->last_stat
.st_size
)
583 uint64_t tail_object_offset
= le64toh(f
->header
->tail_object_offset
);
584 if (!offset_is_valid(tail_object_offset
, header_size
, UINT64_MAX
))
586 if (header_size
+ arena_size
< tail_object_offset
)
588 if (header_size
+ arena_size
- tail_object_offset
< sizeof(ObjectHeader
))
591 if (!hash_table_is_valid(le64toh(f
->header
->data_hash_table_offset
),
592 le64toh(f
->header
->data_hash_table_size
),
593 header_size
, arena_size
, tail_object_offset
))
596 if (!hash_table_is_valid(le64toh(f
->header
->field_hash_table_offset
),
597 le64toh(f
->header
->field_hash_table_size
),
598 header_size
, arena_size
, tail_object_offset
))
601 uint64_t entry_array_offset
= le64toh(f
->header
->entry_array_offset
);
602 if (!offset_is_valid(entry_array_offset
, header_size
, tail_object_offset
))
605 if (JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_array_offset
)) {
606 uint32_t offset
= le32toh(f
->header
->tail_entry_array_offset
);
607 uint32_t n
= le32toh(f
->header
->tail_entry_array_n_entries
);
609 if (!offset_is_valid(offset
, header_size
, tail_object_offset
))
611 if (entry_array_offset
> offset
)
613 if (entry_array_offset
== 0 && offset
!= 0)
615 if ((offset
== 0) != (n
== 0))
617 assert(offset
<= header_size
+ arena_size
);
618 if ((uint64_t) n
* journal_file_entry_array_item_size(f
) > header_size
+ arena_size
- offset
)
622 if (JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_offset
))
623 if (!offset_is_valid(le64toh(f
->header
->tail_entry_offset
), header_size
, tail_object_offset
))
626 /* Verify number of objects */
627 uint64_t n_objects
= le64toh(f
->header
->n_objects
);
628 if (n_objects
> arena_size
/ sizeof(ObjectHeader
))
631 uint64_t n_entries
= le64toh(f
->header
->n_entries
);
632 if (n_entries
> n_objects
)
635 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
636 le64toh(f
->header
->n_data
) > n_objects
)
639 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
640 le64toh(f
->header
->n_fields
) > n_objects
)
643 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
) &&
644 le64toh(f
->header
->n_tags
) > n_objects
)
647 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
) &&
648 le64toh(f
->header
->n_entry_arrays
) > n_objects
)
651 if (JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_array_n_entries
) &&
652 le32toh(f
->header
->tail_entry_array_n_entries
) > n_entries
)
655 if (journal_file_writable(f
)) {
656 sd_id128_t machine_id
;
660 r
= sd_id128_get_machine(&machine_id
);
662 if (!ERRNO_IS_MACHINE_ID_UNSET(r
)) /* handle graceful if machine ID is not initialized yet */
665 machine_id
= SD_ID128_NULL
;
668 if (!sd_id128_equal(machine_id
, f
->header
->machine_id
))
669 return log_debug_errno(SYNTHETIC_ERRNO(EHOSTDOWN
),
670 "Trying to open journal file from different host for writing, refusing.");
672 state
= f
->header
->state
;
674 if (state
== STATE_ARCHIVED
)
675 return -ESHUTDOWN
; /* Already archived */
676 if (state
== STATE_ONLINE
)
677 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
),
678 "Journal file %s is already online. Assuming unclean closing.",
680 if (state
!= STATE_OFFLINE
)
681 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
),
682 "Journal file %s has unknown state %i.",
685 if (f
->header
->field_hash_table_size
== 0 || f
->header
->data_hash_table_size
== 0)
692 int journal_file_fstat(JournalFile
*f
) {
698 if (fstat(f
->fd
, &f
->last_stat
) < 0)
701 f
->last_stat_usec
= now(CLOCK_MONOTONIC
);
703 /* Refuse dealing with files that aren't regular */
704 r
= stat_verify_regular(&f
->last_stat
);
708 /* Refuse appending to files that are already deleted */
709 if (f
->last_stat
.st_nlink
<= 0)
715 static int journal_file_allocate(JournalFile
*f
, uint64_t offset
, uint64_t size
) {
716 uint64_t old_size
, new_size
, old_header_size
, old_arena_size
;
722 /* We assume that this file is not sparse, and we know that for sure, since we always call
723 * posix_fallocate() ourselves */
725 if (size
> PAGE_ALIGN_DOWN(UINT64_MAX
) - offset
)
728 if (mmap_cache_fd_got_sigbus(f
->cache_fd
))
731 old_header_size
= le64toh(READ_NOW(f
->header
->header_size
));
732 old_arena_size
= le64toh(READ_NOW(f
->header
->arena_size
));
733 if (old_arena_size
> PAGE_ALIGN_DOWN(UINT64_MAX
) - old_header_size
)
736 old_size
= old_header_size
+ old_arena_size
;
738 new_size
= MAX(PAGE_ALIGN(offset
+ size
), old_header_size
);
740 if (new_size
<= old_size
) {
742 /* We already pre-allocated enough space, but before
743 * we write to it, let's check with fstat() if the
744 * file got deleted, in order make sure we don't throw
745 * away the data immediately. Don't check fstat() for
746 * all writes though, but only once ever 10s. */
748 if (f
->last_stat_usec
+ LAST_STAT_REFRESH_USEC
> now(CLOCK_MONOTONIC
))
751 return journal_file_fstat(f
);
754 /* Allocate more space. */
756 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
759 /* Refuse to go over 4G in compact mode so offsets can be stored in 32-bit. */
760 if (JOURNAL_HEADER_COMPACT(f
->header
) && new_size
> UINT32_MAX
)
763 if (new_size
> f
->metrics
.min_size
&& f
->metrics
.keep_free
> 0) {
766 if (fstatvfs(f
->fd
, &svfs
) >= 0) {
769 available
= LESS_BY((uint64_t) svfs
.f_bfree
* (uint64_t) svfs
.f_bsize
, f
->metrics
.keep_free
);
771 if (new_size
- old_size
> available
)
776 /* Increase by larger blocks at once */
777 new_size
= ROUND_UP(new_size
, FILE_SIZE_INCREASE
);
778 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
779 new_size
= f
->metrics
.max_size
;
781 /* Note that the glibc fallocate() fallback is very
782 inefficient, hence we try to minimize the allocation area
784 r
= posix_fallocate_loop(f
->fd
, old_size
, new_size
- old_size
);
788 f
->header
->arena_size
= htole64(new_size
- old_header_size
);
790 return journal_file_fstat(f
);
793 static unsigned type_to_context(ObjectType type
) {
794 /* One context for each type, plus one catch-all for the rest */
795 assert_cc(_OBJECT_TYPE_MAX
<= MMAP_CACHE_MAX_CONTEXTS
);
796 assert_cc(CONTEXT_HEADER
< MMAP_CACHE_MAX_CONTEXTS
);
797 return type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
? type
: 0;
800 static int journal_file_move_to(
813 /* This function may clear, overwrite, or alter previously cached entries. After this function has
814 * been called, all objects except for one obtained by this function are invalidated and must be
815 * re-read before use. */
820 if (size
> UINT64_MAX
- offset
)
823 /* Avoid SIGBUS on invalid accesses */
824 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
) {
825 /* Hmm, out of range? Let's refresh the fstat() data
826 * first, before we trust that check. */
828 r
= journal_file_fstat(f
);
832 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
)
833 return -EADDRNOTAVAIL
;
836 return mmap_cache_fd_get(f
->cache_fd
, type_to_context(type
), keep_always
, offset
, size
, &f
->last_stat
, ret
);
839 static uint64_t minimum_header_size(JournalFile
*f
, Object
*o
) {
841 static const uint64_t table
[] = {
842 [OBJECT_DATA
] = sizeof(DataObject
),
843 [OBJECT_FIELD
] = sizeof(FieldObject
),
844 [OBJECT_ENTRY
] = sizeof(EntryObject
),
845 [OBJECT_DATA_HASH_TABLE
] = sizeof(HashTableObject
),
846 [OBJECT_FIELD_HASH_TABLE
] = sizeof(HashTableObject
),
847 [OBJECT_ENTRY_ARRAY
] = sizeof(EntryArrayObject
),
848 [OBJECT_TAG
] = sizeof(TagObject
),
854 if (o
->object
.type
== OBJECT_DATA
)
855 return journal_file_data_payload_offset(f
);
857 if (o
->object
.type
>= ELEMENTSOF(table
) || table
[o
->object
.type
] <= 0)
858 return sizeof(ObjectHeader
);
860 return table
[o
->object
.type
];
863 static int check_object_header(JournalFile
*f
, Object
*o
, ObjectType type
, uint64_t offset
) {
869 s
= le64toh(READ_NOW(o
->object
.size
));
871 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
872 "Attempt to move to uninitialized object: %" PRIu64
,
875 if (s
< sizeof(ObjectHeader
))
876 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
877 "Attempt to move to overly short object with size %"PRIu64
": %" PRIu64
,
880 if (o
->object
.type
<= OBJECT_UNUSED
|| o
->object
.type
>= _OBJECT_TYPE_MAX
)
881 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
882 "Attempt to move to object with invalid type (%u): %" PRIu64
,
883 o
->object
.type
, offset
);
885 if (type
> OBJECT_UNUSED
&& o
->object
.type
!= type
)
886 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
887 "Found %s object while expecting %s object: %" PRIu64
,
888 journal_object_type_to_string(o
->object
.type
),
889 journal_object_type_to_string(type
),
892 if (s
< minimum_header_size(f
, o
))
893 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
894 "Size of %s object (%"PRIu64
") is smaller than the minimum object size (%"PRIu64
"): %" PRIu64
,
895 journal_object_type_to_string(o
->object
.type
),
897 minimum_header_size(f
, o
),
903 /* Lightweight object checks. We want this to be fast, so that we won't
904 * slowdown every journal_file_move_to_object() call too much. */
905 static int check_object(JournalFile
*f
, Object
*o
, uint64_t offset
) {
909 switch (o
->object
.type
) {
912 if ((le64toh(o
->data
.entry_offset
) == 0) ^ (le64toh(o
->data
.n_entries
) == 0))
913 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
914 "Bad data n_entries: %" PRIu64
": %" PRIu64
,
915 le64toh(o
->data
.n_entries
),
918 if (le64toh(o
->object
.size
) <= journal_file_data_payload_offset(f
))
919 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
920 "Bad data size (<= %zu): %" PRIu64
": %" PRIu64
,
921 journal_file_data_payload_offset(f
),
922 le64toh(o
->object
.size
),
925 if (!VALID64(le64toh(o
->data
.next_hash_offset
)) ||
926 !VALID64(le64toh(o
->data
.next_field_offset
)) ||
927 !VALID64(le64toh(o
->data
.entry_offset
)) ||
928 !VALID64(le64toh(o
->data
.entry_array_offset
)))
929 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
930 "Invalid offset, next_hash_offset=" OFSfmt
", next_field_offset=" OFSfmt
", entry_offset=" OFSfmt
", entry_array_offset=" OFSfmt
": %" PRIu64
,
931 le64toh(o
->data
.next_hash_offset
),
932 le64toh(o
->data
.next_field_offset
),
933 le64toh(o
->data
.entry_offset
),
934 le64toh(o
->data
.entry_array_offset
),
940 if (le64toh(o
->object
.size
) <= offsetof(Object
, field
.payload
))
941 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
942 "Bad field size (<= %zu): %" PRIu64
": %" PRIu64
,
943 offsetof(Object
, field
.payload
),
944 le64toh(o
->object
.size
),
947 if (!VALID64(le64toh(o
->field
.next_hash_offset
)) ||
948 !VALID64(le64toh(o
->field
.head_data_offset
)))
949 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
950 "Invalid offset, next_hash_offset=" OFSfmt
", head_data_offset=" OFSfmt
": %" PRIu64
,
951 le64toh(o
->field
.next_hash_offset
),
952 le64toh(o
->field
.head_data_offset
),
959 sz
= le64toh(READ_NOW(o
->object
.size
));
960 if (sz
< offsetof(Object
, entry
.items
) ||
961 (sz
- offsetof(Object
, entry
.items
)) % journal_file_entry_item_size(f
) != 0)
962 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
963 "Bad entry size (<= %zu): %" PRIu64
": %" PRIu64
,
964 offsetof(Object
, entry
.items
),
968 if ((sz
- offsetof(Object
, entry
.items
)) / journal_file_entry_item_size(f
) <= 0)
969 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
970 "Invalid number items in entry: %" PRIu64
": %" PRIu64
,
971 (sz
- offsetof(Object
, entry
.items
)) / journal_file_entry_item_size(f
),
974 if (le64toh(o
->entry
.seqnum
) <= 0)
975 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
976 "Invalid entry seqnum: %" PRIx64
": %" PRIu64
,
977 le64toh(o
->entry
.seqnum
),
980 if (!VALID_REALTIME(le64toh(o
->entry
.realtime
)))
981 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
982 "Invalid entry realtime timestamp: %" PRIu64
": %" PRIu64
,
983 le64toh(o
->entry
.realtime
),
986 if (!VALID_MONOTONIC(le64toh(o
->entry
.monotonic
)))
987 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
988 "Invalid entry monotonic timestamp: %" PRIu64
": %" PRIu64
,
989 le64toh(o
->entry
.monotonic
),
995 case OBJECT_DATA_HASH_TABLE
:
996 case OBJECT_FIELD_HASH_TABLE
: {
999 sz
= le64toh(READ_NOW(o
->object
.size
));
1000 if (sz
< offsetof(Object
, hash_table
.items
) ||
1001 (sz
- offsetof(Object
, hash_table
.items
)) % sizeof(HashItem
) != 0 ||
1002 (sz
- offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
) <= 0)
1003 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1004 "Invalid %s hash table size: %" PRIu64
": %" PRIu64
,
1005 journal_object_type_to_string(o
->object
.type
),
1012 case OBJECT_ENTRY_ARRAY
: {
1015 sz
= le64toh(READ_NOW(o
->object
.size
));
1016 if (sz
< offsetof(Object
, entry_array
.items
) ||
1017 (sz
- offsetof(Object
, entry_array
.items
)) % journal_file_entry_array_item_size(f
) != 0 ||
1018 (sz
- offsetof(Object
, entry_array
.items
)) / journal_file_entry_array_item_size(f
) <= 0)
1019 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1020 "Invalid object entry array size: %" PRIu64
": %" PRIu64
,
1023 /* Here, we request that the offset of each entry array object is in strictly increasing order. */
1024 next
= le64toh(o
->entry_array
.next_entry_array_offset
);
1025 if (!VALID64(next
) || (next
> 0 && next
<= offset
))
1026 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1027 "Invalid object entry array next_entry_array_offset: %" PRIu64
": %" PRIu64
,
1035 if (le64toh(o
->object
.size
) != sizeof(TagObject
))
1036 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1037 "Invalid object tag size: %" PRIu64
": %" PRIu64
,
1038 le64toh(o
->object
.size
),
1041 if (!VALID_EPOCH(le64toh(o
->tag
.epoch
)))
1042 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1043 "Invalid object tag epoch: %" PRIu64
": %" PRIu64
,
1044 le64toh(o
->tag
.epoch
), offset
);
1052 int journal_file_move_to_object(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
**ret
) {
1058 /* Even if this function fails, it may clear, overwrite, or alter previously cached entries. After
1059 * this function has been called, all objects except for one obtained by this function are
1060 * invalidated and must be re-read before use.. */
1062 /* Objects may only be located at multiple of 64 bit */
1063 if (!VALID64(offset
))
1064 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1065 "Attempt to move to %s object at non-64-bit boundary: %" PRIu64
,
1066 journal_object_type_to_string(type
),
1069 /* Object may not be located in the file header */
1070 if (offset
< le64toh(f
->header
->header_size
))
1071 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1072 "Attempt to move to %s object located in file header: %" PRIu64
,
1073 journal_object_type_to_string(type
),
1076 r
= journal_file_move_to(f
, type
, false, offset
, sizeof(ObjectHeader
), (void**) &o
);
1080 r
= check_object_header(f
, o
, type
, offset
);
1084 r
= journal_file_move_to(f
, type
, false, offset
, le64toh(READ_NOW(o
->object
.size
)), (void**) &o
);
1088 r
= check_object_header(f
, o
, type
, offset
);
1092 r
= check_object(f
, o
, offset
);
1102 int journal_file_read_object_header(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
*ret
) {
1109 /* Objects may only be located at multiple of 64 bit */
1110 if (!VALID64(offset
))
1111 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1112 "Attempt to read %s object at non-64-bit boundary: %" PRIu64
,
1113 journal_object_type_to_string(type
), offset
);
1115 /* Object may not be located in the file header */
1116 if (offset
< le64toh(f
->header
->header_size
))
1117 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1118 "Attempt to read %s object located in file header: %" PRIu64
,
1119 journal_object_type_to_string(type
), offset
);
1121 /* This will likely read too much data but it avoids having to call pread() twice. */
1122 n
= pread(f
->fd
, &o
, sizeof(o
), offset
);
1124 return log_debug_errno(errno
, "Failed to read journal %s object at offset: %" PRIu64
,
1125 journal_object_type_to_string(type
), offset
);
1127 if ((size_t) n
< sizeof(o
.object
))
1128 return log_debug_errno(SYNTHETIC_ERRNO(EIO
),
1129 "Failed to read short %s object at offset: %" PRIu64
,
1130 journal_object_type_to_string(type
), offset
);
1132 r
= check_object_header(f
, &o
, type
, offset
);
1136 if ((size_t) n
< minimum_header_size(f
, &o
))
1137 return log_debug_errno(SYNTHETIC_ERRNO(EIO
),
1138 "Short read while reading %s object: %" PRIu64
,
1139 journal_object_type_to_string(type
), offset
);
1141 r
= check_object(f
, &o
, offset
);
1151 static uint64_t inc_seqnum(uint64_t seqnum
) {
1152 if (seqnum
< UINT64_MAX
-1)
1155 return 1; /* skip over UINT64_MAX and 0 when we run out of seqnums and start again */
1158 static uint64_t journal_file_entry_seqnum(
1162 uint64_t next_seqnum
;
1167 /* Picks a new sequence number for the entry we are about to add and returns it. */
1169 next_seqnum
= inc_seqnum(le64toh(f
->header
->tail_entry_seqnum
));
1171 /* If an external seqnum counter was passed, we update both the local and the external one, and set
1172 * it to the maximum of both */
1174 *seqnum
= next_seqnum
= MAX(inc_seqnum(*seqnum
), next_seqnum
);
1176 f
->header
->tail_entry_seqnum
= htole64(next_seqnum
);
1178 if (f
->header
->head_entry_seqnum
== 0)
1179 f
->header
->head_entry_seqnum
= htole64(next_seqnum
);
1184 int journal_file_append_object(
1188 Object
**ret_object
,
1189 uint64_t *ret_offset
) {
1197 assert(type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
);
1198 assert(size
>= sizeof(ObjectHeader
));
1200 r
= journal_file_set_online(f
);
1204 r
= journal_file_tail_end_by_mmap(f
, &p
);
1208 r
= journal_file_allocate(f
, p
, size
);
1212 r
= journal_file_move_to(f
, type
, false, p
, size
, (void**) &o
);
1216 o
->object
= (ObjectHeader
) {
1218 .size
= htole64(size
),
1221 f
->header
->tail_object_offset
= htole64(p
);
1222 f
->header
->n_objects
= htole64(le64toh(f
->header
->n_objects
) + 1);
1233 static int journal_file_setup_data_hash_table(JournalFile
*f
) {
1241 /* We estimate that we need 1 hash table entry per 768 bytes
1242 of journal file and we want to make sure we never get
1243 beyond 75% fill level. Calculate the hash table size for
1244 the maximum file size based on these metrics. */
1246 s
= (f
->metrics
.max_size
* 4 / 768 / 3) * sizeof(HashItem
);
1247 if (s
< DEFAULT_DATA_HASH_TABLE_SIZE
)
1248 s
= DEFAULT_DATA_HASH_TABLE_SIZE
;
1250 log_debug("Reserving %"PRIu64
" entries in data hash table.", s
/ sizeof(HashItem
));
1252 r
= journal_file_append_object(f
,
1253 OBJECT_DATA_HASH_TABLE
,
1254 offsetof(Object
, hash_table
.items
) + s
,
1259 memzero(o
->hash_table
.items
, s
);
1261 f
->header
->data_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1262 f
->header
->data_hash_table_size
= htole64(s
);
1267 static int journal_file_setup_field_hash_table(JournalFile
*f
) {
1275 /* We use a fixed size hash table for the fields as this
1276 * number should grow very slowly only */
1278 s
= DEFAULT_FIELD_HASH_TABLE_SIZE
;
1279 log_debug("Reserving %"PRIu64
" entries in field hash table.", s
/ sizeof(HashItem
));
1281 r
= journal_file_append_object(f
,
1282 OBJECT_FIELD_HASH_TABLE
,
1283 offsetof(Object
, hash_table
.items
) + s
,
1288 memzero(o
->hash_table
.items
, s
);
1290 f
->header
->field_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1291 f
->header
->field_hash_table_size
= htole64(s
);
1296 int journal_file_map_data_hash_table(JournalFile
*f
) {
1304 if (f
->data_hash_table
)
1307 p
= le64toh(f
->header
->data_hash_table_offset
);
1308 s
= le64toh(f
->header
->data_hash_table_size
);
1310 r
= journal_file_move_to(f
,
1311 OBJECT_DATA_HASH_TABLE
,
1318 f
->data_hash_table
= t
;
1322 int journal_file_map_field_hash_table(JournalFile
*f
) {
1330 if (f
->field_hash_table
)
1333 p
= le64toh(f
->header
->field_hash_table_offset
);
1334 s
= le64toh(f
->header
->field_hash_table_size
);
1336 r
= journal_file_move_to(f
,
1337 OBJECT_FIELD_HASH_TABLE
,
1344 f
->field_hash_table
= t
;
1348 static int journal_file_link_field(
1359 assert(f
->field_hash_table
);
1363 if (o
->object
.type
!= OBJECT_FIELD
)
1366 m
= le64toh(READ_NOW(f
->header
->field_hash_table_size
)) / sizeof(HashItem
);
1370 /* This might alter the window we are looking at */
1371 o
->field
.next_hash_offset
= o
->field
.head_data_offset
= 0;
1374 p
= le64toh(f
->field_hash_table
[h
].tail_hash_offset
);
1376 f
->field_hash_table
[h
].head_hash_offset
= htole64(offset
);
1378 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1382 o
->field
.next_hash_offset
= htole64(offset
);
1385 f
->field_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1387 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
1388 f
->header
->n_fields
= htole64(le64toh(f
->header
->n_fields
) + 1);
1393 static int journal_file_link_data(
1404 assert(f
->data_hash_table
);
1408 if (o
->object
.type
!= OBJECT_DATA
)
1411 m
= le64toh(READ_NOW(f
->header
->data_hash_table_size
)) / sizeof(HashItem
);
1415 /* This might alter the window we are looking at */
1416 o
->data
.next_hash_offset
= o
->data
.next_field_offset
= 0;
1417 o
->data
.entry_offset
= o
->data
.entry_array_offset
= 0;
1418 o
->data
.n_entries
= 0;
1421 p
= le64toh(f
->data_hash_table
[h
].tail_hash_offset
);
1423 /* Only entry in the hash table is easy */
1424 f
->data_hash_table
[h
].head_hash_offset
= htole64(offset
);
1426 /* Move back to the previous data object, to patch in
1429 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1433 o
->data
.next_hash_offset
= htole64(offset
);
1436 f
->data_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1438 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
1439 f
->header
->n_data
= htole64(le64toh(f
->header
->n_data
) + 1);
1444 static int get_next_hash_offset(
1447 le64_t
*next_hash_offset
,
1449 le64_t
*header_max_depth
) {
1455 assert(next_hash_offset
);
1458 nextp
= le64toh(READ_NOW(*next_hash_offset
));
1460 if (nextp
<= *p
) /* Refuse going in loops */
1461 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
1462 "Detected hash item loop in %s, refusing.", f
->path
);
1466 /* If the depth of this hash chain is larger than all others we have seen so far, record it */
1467 if (header_max_depth
&& journal_file_writable(f
))
1468 *header_max_depth
= htole64(MAX(*depth
, le64toh(*header_max_depth
)));
1475 int journal_file_find_field_object_with_hash(
1480 Object
**ret_object
,
1481 uint64_t *ret_offset
) {
1483 uint64_t p
, osize
, h
, m
, depth
= 0;
1491 /* If the field hash table is empty, we can't find anything */
1492 if (le64toh(f
->header
->field_hash_table_size
) <= 0)
1495 /* Map the field hash table, if it isn't mapped yet. */
1496 r
= journal_file_map_field_hash_table(f
);
1500 osize
= offsetof(Object
, field
.payload
) + size
;
1502 m
= le64toh(READ_NOW(f
->header
->field_hash_table_size
)) / sizeof(HashItem
);
1507 p
= le64toh(f
->field_hash_table
[h
].head_hash_offset
);
1511 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1515 if (le64toh(o
->field
.hash
) == hash
&&
1516 le64toh(o
->object
.size
) == osize
&&
1517 memcmp(o
->field
.payload
, field
, size
) == 0) {
1527 r
= get_next_hash_offset(
1530 &o
->field
.next_hash_offset
,
1532 JOURNAL_HEADER_CONTAINS(f
->header
, field_hash_chain_depth
) ? &f
->header
->field_hash_chain_depth
: NULL
);
1540 uint64_t journal_file_hash_data(
1547 assert(data
|| sz
== 0);
1549 /* We try to unify our codebase on siphash, hence new-styled journal files utilizing the keyed hash
1550 * function use siphash. Old journal files use the Jenkins hash. */
1552 if (JOURNAL_HEADER_KEYED_HASH(f
->header
))
1553 return siphash24(data
, sz
, f
->header
->file_id
.bytes
);
1555 return jenkins_hash64(data
, sz
);
1558 int journal_file_find_field_object(
1562 Object
**ret_object
,
1563 uint64_t *ret_offset
) {
1569 return journal_file_find_field_object_with_hash(
1572 journal_file_hash_data(f
, field
, size
),
1573 ret_object
, ret_offset
);
1576 int journal_file_find_data_object_with_hash(
1581 Object
**ret_object
,
1582 uint64_t *ret_offset
) {
1584 uint64_t p
, h
, m
, depth
= 0;
1589 assert(data
|| size
== 0);
1591 /* If there's no data hash table, then there's no entry. */
1592 if (le64toh(f
->header
->data_hash_table_size
) <= 0)
1595 /* Map the data hash table, if it isn't mapped yet. */
1596 r
= journal_file_map_data_hash_table(f
);
1600 m
= le64toh(READ_NOW(f
->header
->data_hash_table_size
)) / sizeof(HashItem
);
1605 p
= le64toh(f
->data_hash_table
[h
].head_hash_offset
);
1612 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1616 if (le64toh(o
->data
.hash
) != hash
)
1619 r
= journal_file_data_payload(f
, o
, p
, NULL
, 0, 0, &d
, &rsize
);
1622 assert(r
> 0); /* journal_file_data_payload() always returns > 0 if no field is provided. */
1624 if (memcmp_nn(data
, size
, d
, rsize
) == 0) {
1635 r
= get_next_hash_offset(
1638 &o
->data
.next_hash_offset
,
1640 JOURNAL_HEADER_CONTAINS(f
->header
, data_hash_chain_depth
) ? &f
->header
->data_hash_chain_depth
: NULL
);
1648 int journal_file_find_data_object(
1652 Object
**ret_object
,
1653 uint64_t *ret_offset
) {
1656 assert(data
|| size
== 0);
1658 return journal_file_find_data_object_with_hash(
1661 journal_file_hash_data(f
, data
, size
),
1662 ret_object
, ret_offset
);
1665 bool journal_field_valid(const char *p
, size_t l
, bool allow_protected
) {
1666 /* We kinda enforce POSIX syntax recommendations for
1667 environment variables here, but make a couple of additional
1670 http://pubs.opengroup.org/onlinepubs/000095399/basedefs/xbd_chap08.html */
1677 /* No empty field names */
1681 /* Don't allow names longer than 64 chars */
1685 /* Variables starting with an underscore are protected */
1686 if (!allow_protected
&& p
[0] == '_')
1689 /* Don't allow digits as first character */
1690 if (ascii_isdigit(p
[0]))
1693 /* Only allow A-Z0-9 and '_' */
1694 for (const char *a
= p
; a
< p
+ l
; a
++)
1695 if ((*a
< 'A' || *a
> 'Z') &&
1696 !ascii_isdigit(*a
) &&
1703 static int journal_file_append_field(
1707 Object
**ret_object
,
1708 uint64_t *ret_offset
) {
1719 if (!journal_field_valid(field
, size
, true))
1722 hash
= journal_file_hash_data(f
, field
, size
);
1724 r
= journal_file_find_field_object_with_hash(f
, field
, size
, hash
, ret_object
, ret_offset
);
1730 osize
= offsetof(Object
, field
.payload
) + size
;
1731 r
= journal_file_append_object(f
, OBJECT_FIELD
, osize
, &o
, &p
);
1735 o
->field
.hash
= htole64(hash
);
1736 memcpy(o
->field
.payload
, field
, size
);
1738 r
= journal_file_link_field(f
, o
, p
, hash
);
1742 /* The linking might have altered the window, so let's only pass the offset to hmac which will
1743 * move to the object again if needed. */
1746 r
= journal_file_hmac_put_object(f
, OBJECT_FIELD
, NULL
, p
);
1752 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, ret_object
);
1763 static int maybe_compress_payload(JournalFile
*f
, uint8_t *dst
, const uint8_t *src
, uint64_t size
, size_t *rsize
) {
1767 #if HAVE_COMPRESSION
1771 c
= JOURNAL_FILE_COMPRESSION(f
);
1772 if (c
== COMPRESSION_NONE
|| size
< f
->compress_threshold_bytes
)
1775 r
= compress_blob(c
, src
, size
, dst
, size
- 1, rsize
);
1777 return log_debug_errno(r
, "Failed to compress data object using %s, ignoring: %m", compression_to_string(c
));
1779 log_debug("Compressed data object %"PRIu64
" -> %zu using %s", size
, *rsize
, compression_to_string(c
));
1781 return 1; /* compressed */
1787 static int journal_file_append_data(
1791 Object
**ret_object
,
1792 uint64_t *ret_offset
) {
1794 uint64_t hash
, p
, osize
;
1802 if (!data
|| size
== 0)
1805 hash
= journal_file_hash_data(f
, data
, size
);
1807 r
= journal_file_find_data_object_with_hash(f
, data
, size
, hash
, ret_object
, ret_offset
);
1813 eq
= memchr(data
, '=', size
);
1817 osize
= journal_file_data_payload_offset(f
) + size
;
1818 r
= journal_file_append_object(f
, OBJECT_DATA
, osize
, &o
, &p
);
1822 o
->data
.hash
= htole64(hash
);
1824 r
= maybe_compress_payload(f
, journal_file_data_payload_field(f
, o
), data
, size
, &rsize
);
1826 /* We don't really care failures, let's continue without compression */
1827 memcpy_safe(journal_file_data_payload_field(f
, o
), data
, size
);
1829 Compression c
= JOURNAL_FILE_COMPRESSION(f
);
1831 assert(c
>= 0 && c
< _COMPRESSION_MAX
&& c
!= COMPRESSION_NONE
);
1833 o
->object
.size
= htole64(journal_file_data_payload_offset(f
) + rsize
);
1834 o
->object
.flags
|= COMPRESSION_TO_OBJECT_FLAG(c
);
1837 r
= journal_file_link_data(f
, o
, p
, hash
);
1841 /* The linking might have altered the window, so let's refresh our pointer. */
1842 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1847 r
= journal_file_hmac_put_object(f
, OBJECT_DATA
, o
, p
);
1852 /* Create field object ... */
1853 r
= journal_file_append_field(f
, data
, (uint8_t*) eq
- (uint8_t*) data
, &fo
, NULL
);
1857 /* ... and link it in. */
1858 o
->data
.next_field_offset
= fo
->field
.head_data_offset
;
1859 fo
->field
.head_data_offset
= le64toh(p
);
1870 static int maybe_decompress_payload(
1874 Compression compression
,
1876 size_t field_length
,
1877 size_t data_threshold
,
1883 /* We can't read objects larger than 4G on a 32-bit machine */
1884 if ((uint64_t) (size_t) size
!= size
)
1887 if (compression
!= COMPRESSION_NONE
) {
1888 #if HAVE_COMPRESSION
1893 r
= decompress_startswith(compression
, payload
, size
, &f
->compress_buffer
, field
,
1896 return log_debug_errno(r
,
1897 "Cannot decompress %s object of length %" PRIu64
": %m",
1898 compression_to_string(compression
),
1909 r
= decompress_blob(compression
, payload
, size
, &f
->compress_buffer
, &rsize
, 0);
1914 *ret_data
= f
->compress_buffer
;
1918 return -EPROTONOSUPPORT
;
1921 if (field
&& (size
< field_length
+ 1 || memcmp(payload
, field
, field_length
) != 0 || payload
[field_length
] != '=')) {
1930 *ret_data
= payload
;
1932 *ret_size
= (size_t) size
;
1938 int journal_file_data_payload(
1943 size_t field_length
,
1944 size_t data_threshold
,
1953 assert(!field
== (field_length
== 0)); /* These must be specified together. */
1956 r
= journal_file_move_to_object(f
, OBJECT_DATA
, offset
, &o
);
1961 size
= le64toh(READ_NOW(o
->object
.size
));
1962 if (size
< journal_file_data_payload_offset(f
))
1965 size
-= journal_file_data_payload_offset(f
);
1967 c
= COMPRESSION_FROM_OBJECT(o
);
1969 return -EPROTONOSUPPORT
;
1971 return maybe_decompress_payload(f
, journal_file_data_payload_field(f
, o
), size
, c
, field
,
1972 field_length
, data_threshold
, ret_data
, ret_size
);
1975 uint64_t journal_file_entry_n_items(JournalFile
*f
, Object
*o
) {
1981 if (o
->object
.type
!= OBJECT_ENTRY
)
1984 sz
= le64toh(READ_NOW(o
->object
.size
));
1985 if (sz
< offsetof(Object
, entry
.items
))
1988 return (sz
- offsetof(Object
, entry
.items
)) / journal_file_entry_item_size(f
);
1991 uint64_t journal_file_entry_array_n_items(JournalFile
*f
, Object
*o
) {
1997 if (o
->object
.type
!= OBJECT_ENTRY_ARRAY
)
2000 sz
= le64toh(READ_NOW(o
->object
.size
));
2001 if (sz
< offsetof(Object
, entry_array
.items
))
2004 return (sz
- offsetof(Object
, entry_array
.items
)) / journal_file_entry_array_item_size(f
);
2007 uint64_t journal_file_hash_table_n_items(Object
*o
) {
2012 if (!IN_SET(o
->object
.type
, OBJECT_DATA_HASH_TABLE
, OBJECT_FIELD_HASH_TABLE
))
2015 sz
= le64toh(READ_NOW(o
->object
.size
));
2016 if (sz
< offsetof(Object
, hash_table
.items
))
2019 return (sz
- offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
);
2022 static void write_entry_array_item(JournalFile
*f
, Object
*o
, uint64_t i
, uint64_t p
) {
2026 if (JOURNAL_HEADER_COMPACT(f
->header
)) {
2027 assert(p
<= UINT32_MAX
);
2028 o
->entry_array
.items
.compact
[i
] = htole32(p
);
2030 o
->entry_array
.items
.regular
[i
] = htole64(p
);
2033 static int link_entry_into_array(
2041 uint64_t n
= 0, ap
= 0, q
, i
, a
, hidx
;
2051 a
= tail
? le32toh(*tail
) : le64toh(*first
);
2052 hidx
= le64toh(READ_NOW(*idx
));
2053 i
= tidx
? le32toh(READ_NOW(*tidx
)) : hidx
;
2056 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2060 n
= journal_file_entry_array_n_items(f
, o
);
2062 write_entry_array_item(f
, o
, i
, p
);
2063 *idx
= htole64(hidx
+ 1);
2065 *tidx
= htole32(le32toh(*tidx
) + 1);
2071 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
2082 r
= journal_file_append_object(f
, OBJECT_ENTRY_ARRAY
,
2083 offsetof(Object
, entry_array
.items
) + n
* journal_file_entry_array_item_size(f
),
2089 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY_ARRAY
, o
, q
);
2094 write_entry_array_item(f
, o
, i
, p
);
2097 *first
= htole64(q
);
2099 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, ap
, &o
);
2103 o
->entry_array
.next_entry_array_offset
= htole64(q
);
2109 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
2110 f
->header
->n_entry_arrays
= htole64(le64toh(f
->header
->n_entry_arrays
) + 1);
2112 *idx
= htole64(hidx
+ 1);
2119 static int link_entry_into_array_plus_one(
2137 hidx
= le64toh(READ_NOW(*idx
));
2138 if (hidx
== UINT64_MAX
)
2141 *extra
= htole64(p
);
2145 i
= htole64(hidx
- 1);
2146 r
= link_entry_into_array(f
, first
, &i
, tail
, tidx
, p
);
2151 *idx
= htole64(hidx
+ 1);
2155 static int journal_file_link_entry_item(JournalFile
*f
, uint64_t offset
, uint64_t p
) {
2162 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
2166 return link_entry_into_array_plus_one(f
,
2167 &o
->data
.entry_offset
,
2168 &o
->data
.entry_array_offset
,
2170 JOURNAL_HEADER_COMPACT(f
->header
) ? &o
->data
.compact
.tail_entry_array_offset
: NULL
,
2171 JOURNAL_HEADER_COMPACT(f
->header
) ? &o
->data
.compact
.tail_entry_array_n_entries
: NULL
,
2175 static int journal_file_link_entry(
2179 const EntryItem items
[],
2189 if (o
->object
.type
!= OBJECT_ENTRY
)
2192 __atomic_thread_fence(__ATOMIC_SEQ_CST
);
2194 /* Link up the entry itself */
2195 r
= link_entry_into_array(f
,
2196 &f
->header
->entry_array_offset
,
2197 &f
->header
->n_entries
,
2198 JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_array_offset
) ? &f
->header
->tail_entry_array_offset
: NULL
,
2199 JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_array_n_entries
) ? &f
->header
->tail_entry_array_n_entries
: NULL
,
2204 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
2206 if (f
->header
->head_entry_realtime
== 0)
2207 f
->header
->head_entry_realtime
= o
->entry
.realtime
;
2209 f
->header
->tail_entry_realtime
= o
->entry
.realtime
;
2210 f
->header
->tail_entry_monotonic
= o
->entry
.monotonic
;
2211 if (JOURNAL_HEADER_CONTAINS(f
->header
, tail_entry_offset
))
2212 f
->header
->tail_entry_offset
= htole64(offset
);
2213 f
->newest_mtime
= 0; /* we have a new tail entry now, explicitly invalidate newest boot id/timestamp info */
2215 /* Link up the items */
2216 for (uint64_t i
= 0; i
< n_items
; i
++) {
2219 /* If we fail to link an entry item because we can't allocate a new entry array, don't fail
2220 * immediately but try to link the other entry items since it might still be possible to link
2221 * those if they don't require a new entry array to be allocated. */
2223 k
= journal_file_link_entry_item(f
, offset
, items
[i
].object_offset
);
2233 static void write_entry_item(JournalFile
*f
, Object
*o
, uint64_t i
, const EntryItem
*item
) {
2238 if (JOURNAL_HEADER_COMPACT(f
->header
)) {
2239 assert(item
->object_offset
<= UINT32_MAX
);
2240 o
->entry
.items
.compact
[i
].object_offset
= htole32(item
->object_offset
);
2242 o
->entry
.items
.regular
[i
].object_offset
= htole64(item
->object_offset
);
2243 o
->entry
.items
.regular
[i
].hash
= htole64(item
->hash
);
2247 static int journal_file_append_entry_internal(
2249 const dual_timestamp
*ts
,
2250 const sd_id128_t
*boot_id
,
2251 const sd_id128_t
*machine_id
,
2253 const EntryItem items
[],
2256 sd_id128_t
*seqnum_id
,
2257 Object
**ret_object
,
2258 uint64_t *ret_offset
) {
2268 assert(items
|| n_items
== 0);
2270 if (f
->strict_order
) {
2271 /* If requested be stricter with ordering in this journal file, to make searching via
2272 * bisection fully deterministic. This is an optional feature, so that if desired journal
2273 * files can be written where the ordering is not strictly enforced (in which case bisection
2274 * will yield *a* result, but not the *only* result, when searching for points in
2275 * time). Strict ordering mode is enabled when journald originally writes the files, but
2276 * might not necessarily be if other tools (the remoting tools for example) write journal
2277 * files from combined sources.
2279 * Typically, if any of the errors generated here are seen journald will just rotate the
2280 * journal files and start anew. */
2282 if (ts
->realtime
< le64toh(f
->header
->tail_entry_realtime
))
2283 return log_debug_errno(SYNTHETIC_ERRNO(EREMCHG
),
2284 "Realtime timestamp %" PRIu64
" smaller than previous realtime "
2285 "timestamp %" PRIu64
", refusing entry.",
2286 ts
->realtime
, le64toh(f
->header
->tail_entry_realtime
));
2288 if (!sd_id128_is_null(f
->header
->tail_entry_boot_id
) && boot_id
) {
2290 if (!sd_id128_equal(f
->header
->tail_entry_boot_id
, *boot_id
))
2291 return log_debug_errno(SYNTHETIC_ERRNO(EREMOTE
),
2292 "Boot ID to write is different from previous boot id, refusing entry.");
2294 if (ts
->monotonic
< le64toh(f
->header
->tail_entry_monotonic
))
2295 return log_debug_errno(SYNTHETIC_ERRNO(ENOTNAM
),
2296 "Monotonic timestamp %" PRIu64
" smaller than previous monotonic "
2297 "timestamp %" PRIu64
", refusing entry.",
2298 ts
->monotonic
, le64toh(f
->header
->tail_entry_monotonic
));
2303 /* Settle the passed in sequence number ID */
2305 if (sd_id128_is_null(*seqnum_id
))
2306 *seqnum_id
= f
->header
->seqnum_id
; /* Caller has none assigned, then copy the one from the file */
2307 else if (!sd_id128_equal(*seqnum_id
, f
->header
->seqnum_id
)) {
2308 /* Different seqnum IDs? We can't allow entries from multiple IDs end up in the same journal.*/
2309 if (le64toh(f
->header
->n_entries
) == 0)
2310 f
->header
->seqnum_id
= *seqnum_id
; /* Caller has one, and file so far has no entries, then copy the one from the caller */
2312 return log_debug_errno(SYNTHETIC_ERRNO(EILSEQ
),
2313 "Sequence number IDs don't match, refusing entry.");
2317 if (machine_id
&& sd_id128_is_null(f
->header
->machine_id
))
2318 /* Initialize machine ID when not set yet */
2319 f
->header
->machine_id
= *machine_id
;
2321 osize
= offsetof(Object
, entry
.items
) + (n_items
* journal_file_entry_item_size(f
));
2323 r
= journal_file_append_object(f
, OBJECT_ENTRY
, osize
, &o
, &np
);
2327 o
->entry
.seqnum
= htole64(journal_file_entry_seqnum(f
, seqnum
));
2328 o
->entry
.realtime
= htole64(ts
->realtime
);
2329 o
->entry
.monotonic
= htole64(ts
->monotonic
);
2330 o
->entry
.xor_hash
= htole64(xor_hash
);
2332 f
->header
->tail_entry_boot_id
= *boot_id
;
2333 o
->entry
.boot_id
= f
->header
->tail_entry_boot_id
;
2335 for (size_t i
= 0; i
< n_items
; i
++)
2336 write_entry_item(f
, o
, i
, &items
[i
]);
2339 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY
, o
, np
);
2344 r
= journal_file_link_entry(f
, o
, np
, items
, n_items
);
2357 void journal_file_post_change(JournalFile
*f
) {
2363 /* inotify() does not receive IN_MODIFY events from file
2364 * accesses done via mmap(). After each access we hence
2365 * trigger IN_MODIFY by truncating the journal file to its
2366 * current size which triggers IN_MODIFY. */
2368 __atomic_thread_fence(__ATOMIC_SEQ_CST
);
2370 if (ftruncate(f
->fd
, f
->last_stat
.st_size
) < 0)
2371 log_debug_errno(errno
, "Failed to truncate file to its own size: %m");
2374 static int post_change_thunk(sd_event_source
*timer
, uint64_t usec
, void *userdata
) {
2377 journal_file_post_change(userdata
);
2382 static void schedule_post_change(JournalFile
*f
) {
2387 assert(f
->post_change_timer
);
2389 assert_se(e
= sd_event_source_get_event(f
->post_change_timer
));
2391 /* If we are already going down, post the change immediately. */
2392 if (IN_SET(sd_event_get_state(e
), SD_EVENT_EXITING
, SD_EVENT_FINISHED
))
2395 r
= sd_event_source_get_enabled(f
->post_change_timer
, NULL
);
2397 log_debug_errno(r
, "Failed to get ftruncate timer state: %m");
2403 r
= sd_event_source_set_time_relative(f
->post_change_timer
, f
->post_change_timer_period
);
2405 log_debug_errno(r
, "Failed to set time for scheduling ftruncate: %m");
2409 r
= sd_event_source_set_enabled(f
->post_change_timer
, SD_EVENT_ONESHOT
);
2411 log_debug_errno(r
, "Failed to enable scheduled ftruncate: %m");
2418 /* On failure, let's simply post the change immediately. */
2419 journal_file_post_change(f
);
2422 /* Enable coalesced change posting in a timer on the provided sd_event instance */
2423 int journal_file_enable_post_change_timer(JournalFile
*f
, sd_event
*e
, usec_t t
) {
2424 _cleanup_(sd_event_source_unrefp
) sd_event_source
*timer
= NULL
;
2428 assert_return(!f
->post_change_timer
, -EINVAL
);
2432 r
= sd_event_add_time(e
, &timer
, CLOCK_MONOTONIC
, 0, 0, post_change_thunk
, f
);
2436 r
= sd_event_source_set_enabled(timer
, SD_EVENT_OFF
);
2440 f
->post_change_timer
= TAKE_PTR(timer
);
2441 f
->post_change_timer_period
= t
;
2446 static int entry_item_cmp(const EntryItem
*a
, const EntryItem
*b
) {
2447 return CMP(ASSERT_PTR(a
)->object_offset
, ASSERT_PTR(b
)->object_offset
);
2450 static size_t remove_duplicate_entry_items(EntryItem items
[], size_t n
) {
2453 assert(items
|| n
== 0);
2458 for (size_t i
= 1; i
< n
; i
++)
2459 if (items
[i
].object_offset
!= items
[j
- 1].object_offset
)
2460 items
[j
++] = items
[i
];
2465 int journal_file_append_entry(
2467 const dual_timestamp
*ts
,
2468 const sd_id128_t
*boot_id
,
2469 const struct iovec iovec
[],
2472 sd_id128_t
*seqnum_id
,
2473 Object
**ret_object
,
2474 uint64_t *ret_offset
) {
2476 _cleanup_free_ EntryItem
*items_alloc
= NULL
;
2478 uint64_t xor_hash
= 0;
2479 struct dual_timestamp _ts
;
2480 sd_id128_t _boot_id
, _machine_id
, *machine_id
;
2486 assert(n_iovec
> 0);
2489 if (!VALID_REALTIME(ts
->realtime
))
2490 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
2491 "Invalid realtime timestamp %" PRIu64
", refusing entry.",
2493 if (!VALID_MONOTONIC(ts
->monotonic
))
2494 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
2495 "Invalid monotomic timestamp %" PRIu64
", refusing entry.",
2498 dual_timestamp_get(&_ts
);
2503 r
= sd_id128_get_boot(&_boot_id
);
2507 boot_id
= &_boot_id
;
2510 r
= sd_id128_get_machine(&_machine_id
);
2512 if (!ERRNO_IS_MACHINE_ID_UNSET(r
))
2515 /* If the machine ID is not initialized yet, handle gracefully */
2518 machine_id
= &_machine_id
;
2521 r
= journal_file_maybe_append_tag(f
, ts
->realtime
);
2526 if (n_iovec
< ALLOCA_MAX
/ sizeof(EntryItem
) / 2)
2527 items
= newa(EntryItem
, n_iovec
);
2529 items_alloc
= new(EntryItem
, n_iovec
);
2533 items
= items_alloc
;
2536 for (size_t i
= 0; i
< n_iovec
; i
++) {
2540 r
= journal_file_append_data(f
, iovec
[i
].iov_base
, iovec
[i
].iov_len
, &o
, &p
);
2544 /* When calculating the XOR hash field, we need to take special care if the "keyed-hash"
2545 * journal file flag is on. We use the XOR hash field to quickly determine the identity of a
2546 * specific record, and give records with otherwise identical position (i.e. match in seqno,
2547 * timestamp, …) a stable ordering. But for that we can't have it that the hash of the
2548 * objects in each file is different since they are keyed. Hence let's calculate the Jenkins
2549 * hash here for that. This also has the benefit that cursors for old and new journal files
2550 * are completely identical (they include the XOR hash after all). For classic Jenkins-hash
2551 * files things are easier, we can just take the value from the stored record directly. */
2553 if (JOURNAL_HEADER_KEYED_HASH(f
->header
))
2554 xor_hash
^= jenkins_hash64(iovec
[i
].iov_base
, iovec
[i
].iov_len
);
2556 xor_hash
^= le64toh(o
->data
.hash
);
2558 items
[i
] = (EntryItem
) {
2560 .hash
= le64toh(o
->data
.hash
),
2564 /* Order by the position on disk, in order to improve seek
2565 * times for rotating media. */
2566 typesafe_qsort(items
, n_iovec
, entry_item_cmp
);
2567 n_iovec
= remove_duplicate_entry_items(items
, n_iovec
);
2569 r
= journal_file_append_entry_internal(
2582 /* If the memory mapping triggered a SIGBUS then we return an
2583 * IO error and ignore the error code passed down to us, since
2584 * it is very likely just an effect of a nullified replacement
2587 if (mmap_cache_fd_got_sigbus(f
->cache_fd
))
2590 if (f
->post_change_timer
)
2591 schedule_post_change(f
);
2593 journal_file_post_change(f
);
2598 typedef struct ChainCacheItem
{
2599 uint64_t first
; /* the array at the beginning of the chain */
2600 uint64_t array
; /* the cached array */
2601 uint64_t begin
; /* the first item in the cached array */
2602 uint64_t total
; /* the total number of items in all arrays before this one in the chain */
2603 uint64_t last_index
; /* the last index we looked at, to optimize locality when bisecting */
2606 static void chain_cache_put(
2613 uint64_t last_index
) {
2618 /* If the chain item to cache for this chain is the
2619 * first one it's not worth caching anything */
2623 if (ordered_hashmap_size(h
) >= CHAIN_CACHE_MAX
) {
2624 ci
= ordered_hashmap_steal_first(h
);
2627 ci
= new(ChainCacheItem
, 1);
2634 if (ordered_hashmap_put(h
, &ci
->first
, ci
) < 0) {
2639 assert(ci
->first
== first
);
2644 ci
->last_index
= last_index
;
2647 static int bump_array_index(uint64_t *i
, direction_t direction
, uint64_t n
) {
2650 /* Increase or decrease the specified index, in the right direction. */
2652 if (direction
== DIRECTION_DOWN
) {
2667 static int bump_entry_array(
2672 direction_t direction
,
2682 if (direction
== DIRECTION_DOWN
) {
2684 *ret
= le64toh(o
->entry_array
.next_entry_array_offset
);
2688 /* Entry array chains are a singly linked list, so to find the previous array in the chain, we have
2689 * to start iterating from the top. */
2693 while (p
> 0 && p
!= offset
) {
2694 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, p
, &o
);
2699 p
= le64toh(o
->entry_array
.next_entry_array_offset
);
2702 /* If we can't find the previous entry array in the entry array chain, we're likely dealing with a
2703 * corrupted journal file. */
2712 static int generic_array_get(
2716 direction_t direction
,
2717 Object
**ret_object
,
2718 uint64_t *ret_offset
) {
2720 uint64_t a
, t
= 0, k
;
2727 /* FIXME: fix return value assignment on success. */
2731 /* Try the chain cache first */
2732 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2733 if (ci
&& i
> ci
->total
) {
2740 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2741 if (IN_SET(r
, -EBADMSG
, -EADDRNOTAVAIL
)) {
2742 /* If there's corruption and we're going downwards, let's pretend we reached the
2743 * final entry in the entry array chain. */
2745 if (direction
== DIRECTION_DOWN
)
2748 /* If there's corruption and we're going upwards, move back to the previous entry
2749 * array and start iterating entries from there. */
2751 r
= bump_entry_array(f
, NULL
, a
, first
, DIRECTION_UP
, &a
);
2762 k
= journal_file_entry_array_n_items(f
, o
);
2768 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
2771 /* If we've found the right location, now look for the first non-corrupt entry object (in the right
2775 /* In the first iteration of the while loop, we reuse i, k and o from the previous while
2777 if (i
== UINT64_MAX
) {
2778 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2782 k
= journal_file_entry_array_n_items(f
, o
);
2786 i
= direction
== DIRECTION_DOWN
? 0 : k
- 1;
2792 p
= journal_file_entry_array_item(f
, o
, i
);
2794 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, ret_object
);
2796 /* Let's cache this item for the next invocation */
2797 chain_cache_put(f
->chain_cache
, ci
, first
, a
, journal_file_entry_array_item(f
, o
, 0), t
, i
);
2804 if (!IN_SET(r
, -EADDRNOTAVAIL
, -EBADMSG
))
2807 /* OK, so this entry is borked. Most likely some entry didn't get synced to
2808 * disk properly, let's see if the next one might work for us instead. */
2809 log_debug_errno(r
, "Entry item %" PRIu64
" is bad, skipping over it.", i
);
2811 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2815 } while (bump_array_index(&i
, direction
, k
) > 0);
2817 r
= bump_entry_array(f
, o
, a
, first
, direction
, &a
);
2828 static int generic_array_get_plus_one(
2833 direction_t direction
,
2834 Object
**ret_object
,
2835 uint64_t *ret_offset
) {
2841 /* FIXME: fix return value assignment on success. */
2844 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, ret_object
);
2845 if (IN_SET(r
, -EADDRNOTAVAIL
, -EBADMSG
))
2846 return generic_array_get(f
, first
, 0, direction
, ret_object
, ret_offset
);
2851 *ret_offset
= extra
;
2856 return generic_array_get(f
, first
, i
- 1, direction
, ret_object
, ret_offset
);
2865 static int generic_array_bisect_one(
2867 uint64_t a
, /* offset of entry array object. */
2868 uint64_t i
, /* index of the entry item we will test. */
2870 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2871 direction_t direction
,
2874 uint64_t *ret_offset
) {
2881 assert(test_object
);
2885 assert(i
<= *right
);
2887 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2891 p
= journal_file_entry_array_item(f
, array
, i
);
2895 r
= test_object(f
, p
, needle
);
2896 if (IN_SET(r
, -EBADMSG
, -EADDRNOTAVAIL
)) {
2897 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short.");
2899 return -ENOANO
; /* recognizable error */
2904 if (r
== TEST_FOUND
)
2905 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2907 if (r
== TEST_RIGHT
)
2918 static int generic_array_bisect(
2923 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2924 direction_t direction
,
2925 Object
**ret_object
,
2926 uint64_t *ret_offset
,
2927 uint64_t *ret_idx
) {
2929 /* Given an entry array chain, this function finds the object "closest" to the given needle in the
2930 * chain, taking into account the provided direction. A function can be provided to determine how
2931 * an object is matched against the given needle.
2933 * Given a journal file, the offset of an object and the needle, the test_object() function should
2934 * return TEST_LEFT if the needle is located earlier in the entry array chain, TEST_LEFT if the
2935 * needle is located later in the entry array chain and TEST_FOUND if the object matches the needle.
2936 * If test_object() returns TEST_FOUND for a specific object, that object's information will be used
2937 * to populate the return values of this function. If test_object() never returns TEST_FOUND, the
2938 * return values are populated with the details of one of the objects closest to the needle. If the
2939 * direction is DIRECTION_UP, the earlier object is used. Otherwise, the later object is used.
2942 uint64_t a
, p
, t
= 0, i
= 0, last_p
= 0, last_index
= UINT64_MAX
;
2943 bool subtract_one
= false;
2949 assert(test_object
);
2951 /* Start with the first array in the chain */
2954 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2955 if (ci
&& n
> ci
->total
&& ci
->begin
!= 0) {
2956 /* Ah, we have iterated this bisection array chain previously! Let's see if we can skip ahead
2957 * in the chain, as far as the last time. But we can't jump backwards in the chain, so let's
2958 * check that first. */
2960 r
= test_object(f
, ci
->begin
, needle
);
2964 if (r
== TEST_LEFT
) {
2965 /* OK, what we are looking for is right of the begin of this EntryArray, so let's
2966 * jump straight to previously cached array in the chain */
2971 last_index
= ci
->last_index
;
2976 uint64_t left
= 0, right
, k
, lp
;
2978 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2982 k
= journal_file_entry_array_n_items(f
, array
);
2988 r
= generic_array_bisect_one(f
, a
, right
, needle
, test_object
, direction
, &left
, &right
, &lp
);
2996 if (r
== TEST_RIGHT
) {
2997 /* If we cached the last index we looked at, let's try to not to jump too wildly
2998 * around and see if we can limit the range to look at early to the immediate
2999 * neighbors of the last index we looked at. */
3001 if (last_index
> 0 && last_index
- 1 < right
) {
3002 r
= generic_array_bisect_one(f
, a
, last_index
- 1, needle
, test_object
, direction
, &left
, &right
, NULL
);
3003 if (r
< 0 && r
!= -ENOANO
)
3007 if (last_index
< right
) {
3008 r
= generic_array_bisect_one(f
, a
, last_index
+ 1, needle
, test_object
, direction
, &left
, &right
, NULL
);
3009 if (r
< 0 && r
!= -ENOANO
)
3014 if (left
== right
) {
3015 if (direction
== DIRECTION_UP
)
3016 subtract_one
= true;
3022 assert(left
< right
);
3023 i
= (left
+ right
) / 2;
3025 r
= generic_array_bisect_one(f
, a
, i
, needle
, test_object
, direction
, &left
, &right
, NULL
);
3026 if (r
< 0 && r
!= -ENOANO
)
3032 if (direction
== DIRECTION_UP
) {
3034 subtract_one
= true;
3045 last_index
= UINT64_MAX
;
3046 a
= le64toh(array
->entry_array
.next_entry_array_offset
);
3052 if (subtract_one
&& t
== 0 && i
== 0)
3055 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
3059 p
= journal_file_entry_array_item(f
, array
, 0);
3063 /* Let's cache this item for the next invocation */
3064 chain_cache_put(f
->chain_cache
, ci
, first
, a
, p
, t
, subtract_one
? (i
> 0 ? i
-1 : UINT64_MAX
) : i
);
3066 if (subtract_one
&& i
== 0)
3068 else if (subtract_one
)
3069 p
= journal_file_entry_array_item(f
, array
, i
- 1);
3071 p
= journal_file_entry_array_item(f
, array
, i
);
3074 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, ret_object
);
3083 *ret_idx
= t
+ i
+ (subtract_one
? -1 : 0);
3088 static int generic_array_bisect_plus_one(
3094 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
3095 direction_t direction
,
3096 Object
**ret_object
,
3097 uint64_t *ret_offset
,
3098 uint64_t *ret_idx
) {
3101 bool step_back
= false;
3104 assert(test_object
);
3109 /* This bisects the array in object 'first', but first checks
3111 r
= test_object(f
, extra
, needle
);
3115 if (r
== TEST_FOUND
)
3116 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
3118 /* if we are looking with DIRECTION_UP then we need to first
3119 see if in the actual array there is a matching entry, and
3120 return the last one of that. But if there isn't any we need
3121 to return this one. Hence remember this, and return it
3124 step_back
= direction
== DIRECTION_UP
;
3126 if (r
== TEST_RIGHT
) {
3127 if (direction
== DIRECTION_DOWN
)
3133 r
= generic_array_bisect(f
, first
, n
-1, needle
, test_object
, direction
, ret_object
, ret_offset
, ret_idx
);
3135 if (r
== 0 && step_back
)
3138 if (r
> 0 && ret_idx
)
3145 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, ret_object
);
3151 *ret_offset
= extra
;
3159 _pure_
static int test_object_offset(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
3165 else if (p
< needle
)
3171 int journal_file_move_to_entry_by_offset(
3174 direction_t direction
,
3175 Object
**ret_object
,
3176 uint64_t *ret_offset
) {
3181 return generic_array_bisect(
3183 le64toh(f
->header
->entry_array_offset
),
3184 le64toh(f
->header
->n_entries
),
3188 ret_object
, ret_offset
, NULL
);
3191 static int test_object_seqnum(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
3199 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
3203 sq
= le64toh(READ_NOW(o
->entry
.seqnum
));
3206 else if (sq
< needle
)
3212 int journal_file_move_to_entry_by_seqnum(
3215 direction_t direction
,
3216 Object
**ret_object
,
3217 uint64_t *ret_offset
) {
3222 return generic_array_bisect(
3224 le64toh(f
->header
->entry_array_offset
),
3225 le64toh(f
->header
->n_entries
),
3229 ret_object
, ret_offset
, NULL
);
3232 static int test_object_realtime(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
3240 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
3244 rt
= le64toh(READ_NOW(o
->entry
.realtime
));
3247 else if (rt
< needle
)
3253 int journal_file_move_to_entry_by_realtime(
3256 direction_t direction
,
3257 Object
**ret_object
,
3258 uint64_t *ret_offset
) {
3263 return generic_array_bisect(
3265 le64toh(f
->header
->entry_array_offset
),
3266 le64toh(f
->header
->n_entries
),
3268 test_object_realtime
,
3270 ret_object
, ret_offset
, NULL
);
3273 static int test_object_monotonic(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
3281 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
3285 m
= le64toh(READ_NOW(o
->entry
.monotonic
));
3288 else if (m
< needle
)
3294 static int find_data_object_by_boot_id(
3297 Object
**ret_object
,
3298 uint64_t *ret_offset
) {
3300 char t
[STRLEN("_BOOT_ID=") + 32 + 1] = "_BOOT_ID=";
3304 sd_id128_to_string(boot_id
, t
+ 9);
3305 return journal_file_find_data_object(f
, t
, sizeof(t
) - 1, ret_object
, ret_offset
);
3308 int journal_file_move_to_entry_by_monotonic(
3312 direction_t direction
,
3313 Object
**ret_object
,
3314 uint64_t *ret_offset
) {
3321 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, NULL
);
3327 return generic_array_bisect_plus_one(
3329 le64toh(o
->data
.entry_offset
),
3330 le64toh(o
->data
.entry_array_offset
),
3331 le64toh(o
->data
.n_entries
),
3333 test_object_monotonic
,
3335 ret_object
, ret_offset
, NULL
);
3338 void journal_file_reset_location(JournalFile
*f
) {
3341 f
->location_type
= LOCATION_HEAD
;
3342 f
->current_offset
= 0;
3343 f
->current_seqnum
= 0;
3344 f
->current_realtime
= 0;
3345 f
->current_monotonic
= 0;
3346 zero(f
->current_boot_id
);
3347 f
->current_xor_hash
= 0;
3350 void journal_file_save_location(JournalFile
*f
, Object
*o
, uint64_t offset
) {
3354 f
->location_type
= LOCATION_SEEK
;
3355 f
->current_offset
= offset
;
3356 f
->current_seqnum
= le64toh(o
->entry
.seqnum
);
3357 f
->current_realtime
= le64toh(o
->entry
.realtime
);
3358 f
->current_monotonic
= le64toh(o
->entry
.monotonic
);
3359 f
->current_boot_id
= o
->entry
.boot_id
;
3360 f
->current_xor_hash
= le64toh(o
->entry
.xor_hash
);
3363 static bool check_properly_ordered(uint64_t new_offset
, uint64_t old_offset
, direction_t direction
) {
3365 /* Consider it an error if any of the two offsets is uninitialized */
3366 if (old_offset
== 0 || new_offset
== 0)
3369 /* If we go down, the new offset must be larger than the old one. */
3370 return direction
== DIRECTION_DOWN
?
3371 new_offset
> old_offset
:
3372 new_offset
< old_offset
;
3375 int journal_file_next_entry(
3378 direction_t direction
,
3379 Object
**ret_object
,
3380 uint64_t *ret_offset
) {
3388 /* FIXME: fix return value assignment. */
3390 n
= le64toh(READ_NOW(f
->header
->n_entries
));
3395 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
3397 r
= generic_array_bisect(f
,
3398 le64toh(f
->header
->entry_array_offset
),
3399 le64toh(f
->header
->n_entries
),
3408 r
= bump_array_index(&i
, direction
, n
);
3413 /* And jump to it */
3414 r
= generic_array_get(f
, le64toh(f
->header
->entry_array_offset
), i
, direction
, ret_object
, &ofs
);
3418 /* Ensure our array is properly ordered. */
3419 if (p
> 0 && !check_properly_ordered(ofs
, p
, direction
))
3420 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
3421 "%s: entry array not properly ordered at entry %" PRIu64
,
3430 int journal_file_next_entry_for_data(
3433 direction_t direction
,
3434 Object
**ret_object
,
3435 uint64_t *ret_offset
) {
3442 assert(d
->object
.type
== OBJECT_DATA
);
3444 /* FIXME: fix return value assignment. */
3446 n
= le64toh(READ_NOW(d
->data
.n_entries
));
3450 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
3452 r
= generic_array_get_plus_one(f
,
3453 le64toh(d
->data
.entry_offset
),
3454 le64toh(d
->data
.entry_array_offset
),
3467 int journal_file_move_to_entry_by_offset_for_data(
3471 direction_t direction
,
3472 Object
**ret
, uint64_t *ret_offset
) {
3476 assert(d
->object
.type
== OBJECT_DATA
);
3478 return generic_array_bisect_plus_one(
3480 le64toh(d
->data
.entry_offset
),
3481 le64toh(d
->data
.entry_array_offset
),
3482 le64toh(d
->data
.n_entries
),
3486 ret
, ret_offset
, NULL
);
3489 int journal_file_move_to_entry_by_monotonic_for_data(
3494 direction_t direction
,
3495 Object
**ret_object
,
3496 uint64_t *ret_offset
) {
3498 uint64_t b
, z
, entry_offset
, entry_array_offset
, n_entries
;
3504 assert(d
->object
.type
== OBJECT_DATA
);
3506 /* Save all the required data before the data object gets invalidated. */
3507 entry_offset
= le64toh(READ_NOW(d
->data
.entry_offset
));
3508 entry_array_offset
= le64toh(READ_NOW(d
->data
.entry_array_offset
));
3509 n_entries
= le64toh(READ_NOW(d
->data
.n_entries
));
3511 /* First, seek by time */
3512 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &b
);
3518 r
= generic_array_bisect_plus_one(f
,
3519 le64toh(o
->data
.entry_offset
),
3520 le64toh(o
->data
.entry_array_offset
),
3521 le64toh(o
->data
.n_entries
),
3523 test_object_monotonic
,
3529 /* And now, continue seeking until we find an entry that
3530 * exists in both bisection arrays */
3532 r
= journal_file_move_to_object(f
, OBJECT_DATA
, b
, &o
);
3539 r
= generic_array_bisect_plus_one(f
,
3550 r
= generic_array_bisect_plus_one(f
,
3551 le64toh(o
->data
.entry_offset
),
3552 le64toh(o
->data
.entry_array_offset
),
3553 le64toh(o
->data
.n_entries
),
3564 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, q
, ret_object
);
3579 int journal_file_move_to_entry_by_seqnum_for_data(
3583 direction_t direction
,
3584 Object
**ret_object
,
3585 uint64_t *ret_offset
) {
3589 assert(d
->object
.type
== OBJECT_DATA
);
3591 return generic_array_bisect_plus_one(
3593 le64toh(d
->data
.entry_offset
),
3594 le64toh(d
->data
.entry_array_offset
),
3595 le64toh(d
->data
.n_entries
),
3599 ret_object
, ret_offset
, NULL
);
3602 int journal_file_move_to_entry_by_realtime_for_data(
3606 direction_t direction
,
3607 Object
**ret
, uint64_t *ret_offset
) {
3611 assert(d
->object
.type
== OBJECT_DATA
);
3613 return generic_array_bisect_plus_one(
3615 le64toh(d
->data
.entry_offset
),
3616 le64toh(d
->data
.entry_array_offset
),
3617 le64toh(d
->data
.n_entries
),
3619 test_object_realtime
,
3621 ret
, ret_offset
, NULL
);
3624 void journal_file_dump(JournalFile
*f
) {
3632 journal_file_print_header(f
);
3634 p
= le64toh(READ_NOW(f
->header
->header_size
));
3639 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &o
);
3643 s
= journal_object_type_to_string(o
->object
.type
);
3645 switch (o
->object
.type
) {
3650 printf("Type: %s seqnum=%"PRIu64
" monotonic=%"PRIu64
" realtime=%"PRIu64
"\n",
3652 le64toh(o
->entry
.seqnum
),
3653 le64toh(o
->entry
.monotonic
),
3654 le64toh(o
->entry
.realtime
));
3660 printf("Type: %s seqnum=%"PRIu64
" epoch=%"PRIu64
"\n",
3662 le64toh(o
->tag
.seqnum
),
3663 le64toh(o
->tag
.epoch
));
3668 printf("Type: %s \n", s
);
3670 printf("Type: unknown (%i)", o
->object
.type
);
3675 c
= COMPRESSION_FROM_OBJECT(o
);
3676 if (c
> COMPRESSION_NONE
)
3677 printf("Flags: %s\n",
3678 compression_to_string(c
));
3680 if (p
== le64toh(f
->header
->tail_object_offset
))
3683 p
+= ALIGN64(le64toh(o
->object
.size
));
3688 log_error("File corrupt");
3691 /* Note: the lifetime of the compound literal is the immediately surrounding block. */
3692 #define FORMAT_TIMESTAMP_SAFE(t) (FORMAT_TIMESTAMP(t) ?: " --- ")
3694 void journal_file_print_header(JournalFile
*f
) {
3700 printf("File path: %s\n"
3704 "Sequential number ID: %s\n"
3706 "Compatible flags:%s%s%s\n"
3707 "Incompatible flags:%s%s%s%s%s%s\n"
3708 "Header size: %"PRIu64
"\n"
3709 "Arena size: %"PRIu64
"\n"
3710 "Data hash table size: %"PRIu64
"\n"
3711 "Field hash table size: %"PRIu64
"\n"
3712 "Rotate suggested: %s\n"
3713 "Head sequential number: %"PRIu64
" (%"PRIx64
")\n"
3714 "Tail sequential number: %"PRIu64
" (%"PRIx64
")\n"
3715 "Head realtime timestamp: %s (%"PRIx64
")\n"
3716 "Tail realtime timestamp: %s (%"PRIx64
")\n"
3717 "Tail monotonic timestamp: %s (%"PRIx64
")\n"
3718 "Objects: %"PRIu64
"\n"
3719 "Entry objects: %"PRIu64
"\n",
3721 SD_ID128_TO_STRING(f
->header
->file_id
),
3722 SD_ID128_TO_STRING(f
->header
->machine_id
),
3723 SD_ID128_TO_STRING(f
->header
->tail_entry_boot_id
),
3724 SD_ID128_TO_STRING(f
->header
->seqnum_id
),
3725 f
->header
->state
== STATE_OFFLINE
? "OFFLINE" :
3726 f
->header
->state
== STATE_ONLINE
? "ONLINE" :
3727 f
->header
->state
== STATE_ARCHIVED
? "ARCHIVED" : "UNKNOWN",
3728 JOURNAL_HEADER_SEALED(f
->header
) ? " SEALED" : "",
3729 JOURNAL_HEADER_TAIL_ENTRY_BOOT_ID(f
->header
) ? " TAIL_ENTRY_BOOT_ID" : "",
3730 (le32toh(f
->header
->compatible_flags
) & ~HEADER_COMPATIBLE_ANY
) ? " ???" : "",
3731 JOURNAL_HEADER_COMPRESSED_XZ(f
->header
) ? " COMPRESSED-XZ" : "",
3732 JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
) ? " COMPRESSED-LZ4" : "",
3733 JOURNAL_HEADER_COMPRESSED_ZSTD(f
->header
) ? " COMPRESSED-ZSTD" : "",
3734 JOURNAL_HEADER_KEYED_HASH(f
->header
) ? " KEYED-HASH" : "",
3735 JOURNAL_HEADER_COMPACT(f
->header
) ? " COMPACT" : "",
3736 (le32toh(f
->header
->incompatible_flags
) & ~HEADER_INCOMPATIBLE_ANY
) ? " ???" : "",
3737 le64toh(f
->header
->header_size
),
3738 le64toh(f
->header
->arena_size
),
3739 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3740 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
),
3741 yes_no(journal_file_rotate_suggested(f
, 0, LOG_DEBUG
)),
3742 le64toh(f
->header
->head_entry_seqnum
), le64toh(f
->header
->head_entry_seqnum
),
3743 le64toh(f
->header
->tail_entry_seqnum
), le64toh(f
->header
->tail_entry_seqnum
),
3744 FORMAT_TIMESTAMP_SAFE(le64toh(f
->header
->head_entry_realtime
)), le64toh(f
->header
->head_entry_realtime
),
3745 FORMAT_TIMESTAMP_SAFE(le64toh(f
->header
->tail_entry_realtime
)), le64toh(f
->header
->tail_entry_realtime
),
3746 FORMAT_TIMESPAN(le64toh(f
->header
->tail_entry_monotonic
), USEC_PER_MSEC
), le64toh(f
->header
->tail_entry_monotonic
),
3747 le64toh(f
->header
->n_objects
),
3748 le64toh(f
->header
->n_entries
));
3750 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3751 printf("Data objects: %"PRIu64
"\n"
3752 "Data hash table fill: %.1f%%\n",
3753 le64toh(f
->header
->n_data
),
3754 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))));
3756 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3757 printf("Field objects: %"PRIu64
"\n"
3758 "Field hash table fill: %.1f%%\n",
3759 le64toh(f
->header
->n_fields
),
3760 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))));
3762 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
))
3763 printf("Tag objects: %"PRIu64
"\n",
3764 le64toh(f
->header
->n_tags
));
3765 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
3766 printf("Entry array objects: %"PRIu64
"\n",
3767 le64toh(f
->header
->n_entry_arrays
));
3769 if (JOURNAL_HEADER_CONTAINS(f
->header
, field_hash_chain_depth
))
3770 printf("Deepest field hash chain: %" PRIu64
"\n",
3771 f
->header
->field_hash_chain_depth
);
3773 if (JOURNAL_HEADER_CONTAINS(f
->header
, data_hash_chain_depth
))
3774 printf("Deepest data hash chain: %" PRIu64
"\n",
3775 f
->header
->data_hash_chain_depth
);
3777 if (fstat(f
->fd
, &st
) >= 0)
3778 printf("Disk usage: %s\n", FORMAT_BYTES((uint64_t) st
.st_blocks
* 512ULL));
3781 static int journal_file_warn_btrfs(JournalFile
*f
) {
3787 /* Before we write anything, check if the COW logic is turned
3788 * off on btrfs. Given our write pattern that is quite
3789 * unfriendly to COW file systems this should greatly improve
3790 * performance on COW file systems, such as btrfs, at the
3791 * expense of data integrity features (which shouldn't be too
3792 * bad, given that we do our own checksumming). */
3794 r
= fd_is_fs_type(f
->fd
, BTRFS_SUPER_MAGIC
);
3796 return log_ratelimit_warning_errno(r
, JOURNAL_LOG_RATELIMIT
, "Failed to determine if journal is on btrfs: %m");
3800 r
= read_attr_fd(f
->fd
, &attrs
);
3802 return log_ratelimit_warning_errno(r
, JOURNAL_LOG_RATELIMIT
, "Failed to read file attributes: %m");
3804 if (attrs
& FS_NOCOW_FL
) {
3805 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3809 log_ratelimit_notice(JOURNAL_LOG_RATELIMIT
,
3810 "Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3811 "This is likely to slow down journal access substantially, please consider turning "
3812 "off the copy-on-write file attribute on the journal directory, using chattr +C.",
3818 static void journal_default_metrics(JournalMetrics
*m
, int fd
, bool compact
) {
3820 uint64_t fs_size
= 0;
3825 if (fstatvfs(fd
, &ss
) >= 0)
3826 fs_size
= ss
.f_frsize
* ss
.f_blocks
;
3828 log_debug_errno(errno
, "Failed to determine disk size: %m");
3830 if (m
->max_use
== UINT64_MAX
) {
3833 m
->max_use
= CLAMP(PAGE_ALIGN(fs_size
/ 10), /* 10% of file system size */
3834 MAX_USE_LOWER
, MAX_USE_UPPER
);
3836 m
->max_use
= MAX_USE_LOWER
;
3838 m
->max_use
= PAGE_ALIGN(m
->max_use
);
3840 if (m
->max_use
!= 0 && m
->max_use
< JOURNAL_FILE_SIZE_MIN
*2)
3841 m
->max_use
= JOURNAL_FILE_SIZE_MIN
*2;
3844 if (m
->min_use
== UINT64_MAX
) {
3846 m
->min_use
= CLAMP(PAGE_ALIGN(fs_size
/ 50), /* 2% of file system size */
3847 MIN_USE_LOW
, MIN_USE_HIGH
);
3849 m
->min_use
= MIN_USE_LOW
;
3852 if (m
->min_use
> m
->max_use
)
3853 m
->min_use
= m
->max_use
;
3855 if (m
->max_size
== UINT64_MAX
)
3856 m
->max_size
= MIN(PAGE_ALIGN(m
->max_use
/ 8), /* 8 chunks */
3859 m
->max_size
= PAGE_ALIGN(m
->max_size
);
3861 if (compact
&& m
->max_size
> JOURNAL_COMPACT_SIZE_MAX
)
3862 m
->max_size
= JOURNAL_COMPACT_SIZE_MAX
;
3864 if (m
->max_size
!= 0) {
3865 if (m
->max_size
< JOURNAL_FILE_SIZE_MIN
)
3866 m
->max_size
= JOURNAL_FILE_SIZE_MIN
;
3868 if (m
->max_use
!= 0 && m
->max_size
*2 > m
->max_use
)
3869 m
->max_use
= m
->max_size
*2;
3872 if (m
->min_size
== UINT64_MAX
)
3873 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3875 m
->min_size
= CLAMP(PAGE_ALIGN(m
->min_size
),
3876 JOURNAL_FILE_SIZE_MIN
,
3877 m
->max_size
?: UINT64_MAX
);
3879 if (m
->keep_free
== UINT64_MAX
) {
3881 m
->keep_free
= MIN(PAGE_ALIGN(fs_size
/ 20), /* 5% of file system size */
3884 m
->keep_free
= DEFAULT_KEEP_FREE
;
3887 if (m
->n_max_files
== UINT64_MAX
)
3888 m
->n_max_files
= DEFAULT_N_MAX_FILES
;
3890 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64
,
3891 FORMAT_BYTES(m
->min_use
),
3892 FORMAT_BYTES(m
->max_use
),
3893 FORMAT_BYTES(m
->max_size
),
3894 FORMAT_BYTES(m
->min_size
),
3895 FORMAT_BYTES(m
->keep_free
),
3899 int journal_file_open(
3903 JournalFileFlags file_flags
,
3905 uint64_t compress_threshold_bytes
,
3906 JournalMetrics
*metrics
,
3907 MMapCache
*mmap_cache
,
3908 JournalFile
*template,
3909 JournalFile
**ret
) {
3911 bool newly_created
= false;
3916 assert(fd
>= 0 || fname
);
3917 assert(file_flags
>= 0);
3918 assert(file_flags
<= _JOURNAL_FILE_FLAGS_MAX
);
3922 if (!IN_SET((open_flags
& O_ACCMODE
), O_RDONLY
, O_RDWR
))
3925 if ((open_flags
& O_ACCMODE
) == O_RDONLY
&& FLAGS_SET(open_flags
, O_CREAT
))
3928 if (fname
&& (open_flags
& O_CREAT
) && !endswith(fname
, ".journal"))
3931 f
= new(JournalFile
, 1);
3935 *f
= (JournalFile
) {
3938 .open_flags
= open_flags
,
3939 .compress_threshold_bytes
= compress_threshold_bytes
== UINT64_MAX
?
3940 DEFAULT_COMPRESS_THRESHOLD
:
3941 MAX(MIN_COMPRESS_THRESHOLD
, compress_threshold_bytes
),
3942 .strict_order
= FLAGS_SET(file_flags
, JOURNAL_STRICT_ORDER
),
3943 .newest_boot_id_prioq_idx
= PRIOQ_IDX_NULL
,
3947 f
->path
= strdup(fname
);
3955 /* If we don't know the path, fill in something explanatory and vaguely useful */
3956 if (asprintf(&f
->path
, "/proc/self/%i", fd
) < 0) {
3962 f
->chain_cache
= ordered_hashmap_new(&uint64_hash_ops
);
3963 if (!f
->chain_cache
) {
3969 /* We pass O_NONBLOCK here, so that in case somebody pointed us to some character device node or FIFO
3970 * or so, we likely fail quickly than block for long. For regular files O_NONBLOCK has no effect, hence
3971 * it doesn't hurt in that case. */
3973 f
->fd
= openat_report_new(AT_FDCWD
, f
->path
, f
->open_flags
|O_CLOEXEC
|O_NONBLOCK
, f
->mode
, &newly_created
);
3979 /* fds we opened here by us should also be closed by us. */
3982 r
= fd_nonblock(f
->fd
, false);
3986 if (!newly_created
) {
3987 r
= journal_file_fstat(f
);
3992 r
= journal_file_fstat(f
);
3996 /* If we just got the fd passed in, we don't really know if we created the file anew */
3997 newly_created
= f
->last_stat
.st_size
== 0 && journal_file_writable(f
);
4000 f
->cache_fd
= mmap_cache_add_fd(mmap_cache
, f
->fd
, mmap_prot_from_open_flags(open_flags
));
4006 if (newly_created
) {
4007 (void) journal_file_warn_btrfs(f
);
4009 /* Let's attach the creation time to the journal file, so that the vacuuming code knows the age of this
4010 * file even if the file might end up corrupted one day... Ideally we'd just use the creation time many
4011 * file systems maintain for each file, but the API to query this is very new, hence let's emulate this
4012 * via extended attributes. If extended attributes are not supported we'll just skip this, and rely
4013 * solely on mtime/atime/ctime of the file. */
4014 (void) fd_setcrtime(f
->fd
, 0);
4016 r
= journal_file_init_header(f
, file_flags
, template);
4020 r
= journal_file_fstat(f
);
4025 if (f
->last_stat
.st_size
< (off_t
) HEADER_SIZE_MIN
) {
4030 r
= mmap_cache_fd_get(f
->cache_fd
, CONTEXT_HEADER
, true, 0, PAGE_ALIGN(sizeof(Header
)), &f
->last_stat
, &h
);
4032 /* Some file systems (jffs2 or p9fs) don't support mmap() properly (or only read-only
4033 * mmap()), and return EINVAL in that case. Let's propagate that as a more recognizable error
4043 if (!newly_created
) {
4044 r
= journal_file_verify_header(f
);
4050 if (!newly_created
&& journal_file_writable(f
) && JOURNAL_HEADER_SEALED(f
->header
)) {
4051 r
= journal_file_fss_load(f
);
4057 if (journal_file_writable(f
)) {
4059 journal_default_metrics(metrics
, f
->fd
, JOURNAL_HEADER_COMPACT(f
->header
));
4060 f
->metrics
= *metrics
;
4061 } else if (template)
4062 f
->metrics
= template->metrics
;
4064 r
= journal_file_refresh_header(f
);
4070 r
= journal_file_hmac_setup(f
);
4075 if (newly_created
) {
4076 r
= journal_file_setup_field_hash_table(f
);
4080 r
= journal_file_setup_data_hash_table(f
);
4085 r
= journal_file_append_first_tag(f
);
4091 if (mmap_cache_fd_got_sigbus(f
->cache_fd
)) {
4096 if (template && template->post_change_timer
) {
4097 r
= journal_file_enable_post_change_timer(
4099 sd_event_source_get_event(template->post_change_timer
),
4100 template->post_change_timer_period
);
4106 /* The file is opened now successfully, thus we take possession of any passed in fd. */
4109 if (DEBUG_LOGGING
) {
4110 static int last_seal
= -1, last_keyed_hash
= -1;
4111 static Compression last_compression
= _COMPRESSION_INVALID
;
4112 static uint64_t last_bytes
= UINT64_MAX
;
4114 if (last_seal
!= JOURNAL_HEADER_SEALED(f
->header
) ||
4115 last_keyed_hash
!= JOURNAL_HEADER_KEYED_HASH(f
->header
) ||
4116 last_compression
!= JOURNAL_FILE_COMPRESSION(f
) ||
4117 last_bytes
!= f
->compress_threshold_bytes
) {
4119 log_debug("Journal effective settings seal=%s keyed_hash=%s compress=%s compress_threshold_bytes=%s",
4120 yes_no(JOURNAL_HEADER_SEALED(f
->header
)), yes_no(JOURNAL_HEADER_KEYED_HASH(f
->header
)),
4121 compression_to_string(JOURNAL_FILE_COMPRESSION(f
)), FORMAT_BYTES(f
->compress_threshold_bytes
));
4122 last_seal
= JOURNAL_HEADER_SEALED(f
->header
);
4123 last_keyed_hash
= JOURNAL_HEADER_KEYED_HASH(f
->header
);
4124 last_compression
= JOURNAL_FILE_COMPRESSION(f
);
4125 last_bytes
= f
->compress_threshold_bytes
;
4133 if (f
->cache_fd
&& mmap_cache_fd_got_sigbus(f
->cache_fd
))
4136 (void) journal_file_close(f
);
4138 if (newly_created
&& fd
< 0)
4139 (void) unlink(fname
);
4144 int journal_file_parse_uid_from_filename(const char *path
, uid_t
*ret_uid
) {
4145 _cleanup_free_
char *buf
= NULL
, *p
= NULL
;
4146 const char *a
, *b
, *at
;
4149 /* This helper returns -EREMOTE when the filename doesn't match user online/offline journal
4150 * pattern. Hence it currently doesn't parse archived or disposed user journals. */
4155 r
= path_extract_filename(path
, &p
);
4158 if (r
== O_DIRECTORY
)
4161 a
= startswith(p
, "user-");
4164 b
= endswith(p
, ".journal");
4168 at
= strchr(a
, '@');
4172 buf
= strndup(a
, b
-a
);
4176 return parse_uid(buf
, ret_uid
);
4179 int journal_file_archive(JournalFile
*f
, char **ret_previous_path
) {
4180 _cleanup_free_
char *p
= NULL
;
4184 if (!journal_file_writable(f
))
4187 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
4188 * rotation, since we don't know the actual path, and couldn't rename the file hence. */
4189 if (path_startswith(f
->path
, "/proc/self/fd"))
4192 if (!endswith(f
->path
, ".journal"))
4195 if (asprintf(&p
, "%.*s@" SD_ID128_FORMAT_STR
"-%016"PRIx64
"-%016"PRIx64
".journal",
4196 (int) strlen(f
->path
) - 8, f
->path
,
4197 SD_ID128_FORMAT_VAL(f
->header
->seqnum_id
),
4198 le64toh(f
->header
->head_entry_seqnum
),
4199 le64toh(f
->header
->head_entry_realtime
)) < 0)
4202 /* Try to rename the file to the archived version. If the file already was deleted, we'll get ENOENT, let's
4203 * ignore that case. */
4204 if (rename(f
->path
, p
) < 0 && errno
!= ENOENT
)
4207 /* Sync the rename to disk */
4208 (void) fsync_directory_of_file(f
->fd
);
4210 if (ret_previous_path
)
4211 *ret_previous_path
= f
->path
;
4215 f
->path
= TAKE_PTR(p
);
4217 /* Set as archive so offlining commits w/state=STATE_ARCHIVED. Previously we would set old_file->header->state
4218 * to STATE_ARCHIVED directly here, but journal_file_set_offline() short-circuits when state != STATE_ONLINE,
4219 * which would result in the rotated journal never getting fsync() called before closing. Now we simply queue
4220 * the archive state by setting an archive bit, leaving the state as STATE_ONLINE so proper offlining
4227 int journal_file_dispose(int dir_fd
, const char *fname
) {
4228 _cleanup_free_
char *p
= NULL
;
4232 /* Renames a journal file to *.journal~, i.e. to mark it as corrupted or otherwise uncleanly shutdown. Note that
4233 * this is done without looking into the file or changing any of its contents. The idea is that this is called
4234 * whenever something is suspicious and we want to move the file away and make clear that it is not accessed
4235 * for writing anymore. */
4237 if (!endswith(fname
, ".journal"))
4240 if (asprintf(&p
, "%.*s@%016" PRIx64
"-%016" PRIx64
".journal~",
4241 (int) strlen(fname
) - 8, fname
,
4242 now(CLOCK_REALTIME
),
4246 if (renameat(dir_fd
, fname
, dir_fd
, p
) < 0)
4252 int journal_file_copy_entry(
4258 sd_id128_t
*seqnum_id
) {
4260 _cleanup_free_ EntryItem
*items_alloc
= NULL
;
4262 uint64_t n
, m
= 0, xor_hash
= 0;
4272 if (!journal_file_writable(to
))
4275 ts
= (dual_timestamp
) {
4276 .monotonic
= le64toh(o
->entry
.monotonic
),
4277 .realtime
= le64toh(o
->entry
.realtime
),
4279 boot_id
= o
->entry
.boot_id
;
4281 n
= journal_file_entry_n_items(from
, o
);
4285 if (n
< ALLOCA_MAX
/ sizeof(EntryItem
) / 2)
4286 items
= newa(EntryItem
, n
);
4288 items_alloc
= new(EntryItem
, n
);
4292 items
= items_alloc
;
4295 for (uint64_t i
= 0; i
< n
; i
++) {
4301 q
= journal_file_entry_item_object_offset(from
, o
, i
);
4302 r
= journal_file_data_payload(from
, NULL
, q
, NULL
, 0, 0, &data
, &l
);
4303 if (IN_SET(r
, -EADDRNOTAVAIL
, -EBADMSG
)) {
4304 log_debug_errno(r
, "Entry item %"PRIu64
" data object is bad, skipping over it: %m", i
);
4314 r
= journal_file_append_data(to
, data
, l
, &u
, &h
);
4318 if (JOURNAL_HEADER_KEYED_HASH(to
->header
))
4319 xor_hash
^= jenkins_hash64(data
, l
);
4321 xor_hash
^= le64toh(u
->data
.hash
);
4323 items
[m
++] = (EntryItem
) {
4325 .hash
= le64toh(u
->data
.hash
),
4329 /* The above journal_file_data_payload() may clear or overwrite cached object. Hence, we need
4330 * to re-read the object from the cache. */
4331 r
= journal_file_move_to_object(from
, OBJECT_ENTRY
, p
, &o
);
4339 r
= journal_file_append_entry_internal(
4343 &from
->header
->machine_id
,
4349 /* ret_object= */ NULL
,
4350 /* ret_offset= */ NULL
);
4352 if (mmap_cache_fd_got_sigbus(to
->cache_fd
))
4358 void journal_reset_metrics(JournalMetrics
*m
) {
4361 /* Set everything to "pick automatic values". */
4363 *m
= (JournalMetrics
) {
4364 .min_use
= UINT64_MAX
,
4365 .max_use
= UINT64_MAX
,
4366 .min_size
= UINT64_MAX
,
4367 .max_size
= UINT64_MAX
,
4368 .keep_free
= UINT64_MAX
,
4369 .n_max_files
= UINT64_MAX
,
4373 int journal_file_get_cutoff_realtime_usec(JournalFile
*f
, usec_t
*ret_from
, usec_t
*ret_to
) {
4376 assert(ret_from
|| ret_to
);
4379 if (f
->header
->head_entry_realtime
== 0)
4382 *ret_from
= le64toh(f
->header
->head_entry_realtime
);
4386 if (f
->header
->tail_entry_realtime
== 0)
4389 *ret_to
= le64toh(f
->header
->tail_entry_realtime
);
4395 int journal_file_get_cutoff_monotonic_usec(JournalFile
*f
, sd_id128_t boot_id
, usec_t
*ret_from
, usec_t
*ret_to
) {
4401 assert(ret_from
|| ret_to
);
4403 /* FIXME: fix return value assignment on success with 0. */
4405 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &p
);
4409 if (le64toh(o
->data
.n_entries
) <= 0)
4413 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, le64toh(o
->data
.entry_offset
), &o
);
4417 *ret_from
= le64toh(o
->entry
.monotonic
);
4421 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
4425 r
= generic_array_get_plus_one(f
,
4426 le64toh(o
->data
.entry_offset
),
4427 le64toh(o
->data
.entry_array_offset
),
4428 le64toh(o
->data
.n_entries
) - 1,
4434 *ret_to
= le64toh(o
->entry
.monotonic
);
4440 bool journal_file_rotate_suggested(JournalFile
*f
, usec_t max_file_usec
, int log_level
) {
4444 /* If we gained new header fields we gained new features,
4445 * hence suggest a rotation */
4446 if (le64toh(f
->header
->header_size
) < sizeof(Header
)) {
4447 log_ratelimit_full(log_level
, JOURNAL_LOG_RATELIMIT
,
4448 "%s uses an outdated header, suggesting rotation.", f
->path
);
4452 /* Let's check if the hash tables grew over a certain fill level (75%, borrowing this value from
4453 * Java's hash table implementation), and if so suggest a rotation. To calculate the fill level we
4454 * need the n_data field, which only exists in newer versions. */
4456 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
4457 if (le64toh(f
->header
->n_data
) * 4ULL > (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
4459 log_level
, JOURNAL_LOG_RATELIMIT
,
4460 "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.",
4462 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))),
4463 le64toh(f
->header
->n_data
),
4464 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
4465 (uint64_t) f
->last_stat
.st_size
,
4466 f
->last_stat
.st_size
/ le64toh(f
->header
->n_data
));
4470 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
4471 if (le64toh(f
->header
->n_fields
) * 4ULL > (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
4473 log_level
, JOURNAL_LOG_RATELIMIT
,
4474 "Field hash table of %s has a fill level at %.1f (%"PRIu64
" of %"PRIu64
" items), suggesting rotation.",
4476 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))),
4477 le64toh(f
->header
->n_fields
),
4478 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
));
4482 /* If there are too many hash collisions somebody is most likely playing games with us. Hence, if our
4483 * longest chain is longer than some threshold, let's suggest rotation. */
4484 if (JOURNAL_HEADER_CONTAINS(f
->header
, data_hash_chain_depth
) &&
4485 le64toh(f
->header
->data_hash_chain_depth
) > HASH_CHAIN_DEPTH_MAX
) {
4487 log_level
, JOURNAL_LOG_RATELIMIT
,
4488 "Data hash table of %s has deepest hash chain of length %" PRIu64
", suggesting rotation.",
4489 f
->path
, le64toh(f
->header
->data_hash_chain_depth
));
4493 if (JOURNAL_HEADER_CONTAINS(f
->header
, field_hash_chain_depth
) &&
4494 le64toh(f
->header
->field_hash_chain_depth
) > HASH_CHAIN_DEPTH_MAX
) {
4496 log_level
, JOURNAL_LOG_RATELIMIT
,
4497 "Field hash table of %s has deepest hash chain of length at %" PRIu64
", suggesting rotation.",
4498 f
->path
, le64toh(f
->header
->field_hash_chain_depth
));
4502 /* Are the data objects properly indexed by field objects? */
4503 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
4504 JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
4505 le64toh(f
->header
->n_data
) > 0 &&
4506 le64toh(f
->header
->n_fields
) == 0) {
4508 log_level
, JOURNAL_LOG_RATELIMIT
,
4509 "Data objects of %s are not indexed by field objects, suggesting rotation.",
4514 if (max_file_usec
> 0) {
4517 h
= le64toh(f
->header
->head_entry_realtime
);
4518 t
= now(CLOCK_REALTIME
);
4520 if (h
> 0 && t
> h
+ max_file_usec
) {
4522 log_level
, JOURNAL_LOG_RATELIMIT
,
4523 "Oldest entry in %s is older than the configured file retention duration (%s), suggesting rotation.",
4524 f
->path
, FORMAT_TIMESPAN(max_file_usec
, USEC_PER_SEC
));
4532 static const char * const journal_object_type_table
[] = {
4533 [OBJECT_UNUSED
] = "unused",
4534 [OBJECT_DATA
] = "data",
4535 [OBJECT_FIELD
] = "field",
4536 [OBJECT_ENTRY
] = "entry",
4537 [OBJECT_DATA_HASH_TABLE
] = "data hash table",
4538 [OBJECT_FIELD_HASH_TABLE
] = "field hash table",
4539 [OBJECT_ENTRY_ARRAY
] = "entry array",
4540 [OBJECT_TAG
] = "tag",
4543 DEFINE_STRING_TABLE_LOOKUP_TO_STRING(journal_object_type
, ObjectType
);